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
= 0;
52 unsigned long num_physpages
= 0;
53 void * high_memory
= NULL
;
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
)) {
67 copy_highpage(to
, from
);
70 mem_map_t
* mem_map
= NULL
;
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 unsigned long page_nr
;
218 goto cont_copy_pte_range
;
219 if (!pte_present(pte
)) {
220 swap_duplicate(pte_to_swp_entry(pte
));
221 set_pte(dst_pte
, pte
);
222 goto cont_copy_pte_range
;
224 page_nr
= pte_pagenr(pte
);
225 if (page_nr
>= max_mapnr
||
226 PageReserved(mem_map
+page_nr
)) {
227 set_pte(dst_pte
, pte
);
228 goto cont_copy_pte_range
;
230 /* If it's a COW mapping, write protect it both in the parent and the child */
232 pte
= pte_wrprotect(pte
);
233 set_pte(src_pte
, pte
);
235 /* If it's a shared mapping, mark it clean in the child */
236 if (vma
->vm_flags
& VM_SHARED
)
237 pte
= pte_mkclean(pte
);
238 set_pte(dst_pte
, pte_mkold(pte
));
239 get_page(mem_map
+ page_nr
);
241 cont_copy_pte_range
: 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 unsigned long nr
= pte_pagenr(page
);
266 if (nr
>= max_mapnr
|| PageReserved(mem_map
+nr
))
269 * free_page() used to be able to clear swap cache
270 * entries. We may now have to do it manually.
272 free_page_and_swap_cache(mem_map
+nr
);
275 swap_free(pte_to_swp_entry(page
));
279 static inline void forget_pte(pte_t page
)
281 if (!pte_none(page
)) {
282 printk("forget_pte: old mapping existed!\n");
287 static inline int zap_pte_range(struct mm_struct
*mm
, pmd_t
* pmd
, unsigned long address
, unsigned long size
)
299 pte
= pte_offset(pmd
, address
);
300 address
&= ~PMD_MASK
;
301 if (address
+ size
> PMD_SIZE
)
302 size
= PMD_SIZE
- address
;
315 freed
+= free_pte(page
);
320 static inline int zap_pmd_range(struct mm_struct
*mm
, pgd_t
* dir
, unsigned long address
, unsigned long size
)
333 pmd
= pmd_offset(dir
, address
);
334 address
&= ~PGDIR_MASK
;
335 end
= address
+ size
;
336 if (end
> PGDIR_SIZE
)
340 freed
+= zap_pte_range(mm
, pmd
, address
, end
- address
);
341 address
= (address
+ PMD_SIZE
) & PMD_MASK
;
343 } while (address
< end
);
348 * remove user pages in a given range.
350 void zap_page_range(struct mm_struct
*mm
, unsigned long address
, unsigned long size
)
353 unsigned long end
= address
+ size
;
356 dir
= pgd_offset(mm
, address
);
359 * This is a long-lived spinlock. That's fine.
360 * There's no contention, because the page table
361 * lock only protects against kswapd anyway, and
362 * even if kswapd happened to be looking at this
363 * process we _want_ it to get stuck.
367 spin_lock(&mm
->page_table_lock
);
369 freed
+= zap_pmd_range(mm
, dir
, address
, end
- address
);
370 address
= (address
+ PGDIR_SIZE
) & PGDIR_MASK
;
372 } while (address
&& (address
< end
));
373 spin_unlock(&mm
->page_table_lock
);
375 * Update rss for the mm_struct (not necessarily current->mm)
386 * Do a quick page-table lookup for a single page.
388 static struct page
* follow_page(unsigned long address
)
393 pgd
= pgd_offset(current
->mm
, address
);
394 pmd
= pmd_offset(pgd
, address
);
396 pte_t
* pte
= pte_offset(pmd
, address
);
397 if (pte
&& pte_present(*pte
))
398 return pte_page(*pte
);
401 printk(KERN_ERR
"Missing page in follow_page\n");
406 * Given a physical address, is there a useful struct page pointing to it?
409 struct page
* get_page_map(struct page
*page
, unsigned long vaddr
)
411 if (MAP_NR(vaddr
) >= max_mapnr
)
413 if (page
== ZERO_PAGE(vaddr
))
415 if (PageReserved(page
))
421 * Force in an entire range of pages from the current process's user VA,
422 * and pin and lock the pages for IO.
425 #define dprintk(x...)
426 int map_user_kiobuf(int rw
, struct kiobuf
*iobuf
, unsigned long va
, size_t len
)
428 unsigned long ptr
, end
;
430 struct mm_struct
* mm
;
431 struct vm_area_struct
* vma
= 0;
437 /* Make sure the iobuf is not already mapped somewhere. */
442 dprintk ("map_user_kiobuf: begin\n");
444 ptr
= va
& PAGE_MASK
;
445 end
= (va
+ len
+ PAGE_SIZE
- 1) & PAGE_MASK
;
446 err
= expand_kiobuf(iobuf
, (end
- ptr
) >> PAGE_SHIFT
);
455 iobuf
->offset
= va
& ~PAGE_MASK
;
461 * First of all, try to fault in all of the necessary pages
464 if (!vma
|| ptr
>= vma
->vm_end
) {
465 vma
= find_vma(current
->mm
, ptr
);
469 if (handle_mm_fault(current
, vma
, ptr
, (rw
==READ
)) <= 0)
471 spin_lock(&mm
->page_table_lock
);
472 map
= follow_page(ptr
);
474 dprintk (KERN_ERR
"Missing page in map_user_kiobuf\n");
477 map
= get_page_map(map
, ptr
);
479 if (TryLockPage(map
)) {
482 atomic_inc(&map
->count
);
484 spin_unlock(&mm
->page_table_lock
);
485 iobuf
->maplist
[i
] = map
;
486 iobuf
->nr_pages
= ++i
;
492 dprintk ("map_user_kiobuf: end OK\n");
498 dprintk ("map_user_kiobuf: end %d\n", err
);
504 * Undo the locking so far, wait on the page we got to, and try again.
506 spin_unlock(&mm
->page_table_lock
);
511 * Did the release also unlock the page we got stuck on?
514 if (!PageLocked(map
)) {
515 /* If so, we may well have the page mapped twice
516 * in the IO address range. Bad news. Of
517 * course, it _might_ * just be a coincidence,
518 * but if it happens more than * once, chances
519 * are we have a double-mapped page. */
520 if (++doublepage
>= 3) {
532 ptr
= va
& PAGE_MASK
;
540 * Unmap all of the pages referenced by a kiobuf. We release the pages,
541 * and unlock them if they were locked.
544 void unmap_kiobuf (struct kiobuf
*iobuf
)
549 for (i
= 0; i
< iobuf
->nr_pages
; i
++) {
550 map
= iobuf
->maplist
[i
];
552 if (map
&& iobuf
->locked
) {
562 static inline void zeromap_pte_range(pte_t
* pte
, unsigned long address
,
563 unsigned long size
, pgprot_t prot
)
567 address
&= ~PMD_MASK
;
568 end
= address
+ size
;
572 pte_t zero_pte
= pte_wrprotect(mk_pte(ZERO_PAGE(address
), prot
));
573 pte_t oldpage
= *pte
;
574 set_pte(pte
, zero_pte
);
576 address
+= PAGE_SIZE
;
578 } while (address
&& (address
< end
));
581 static inline int zeromap_pmd_range(pmd_t
* pmd
, unsigned long address
,
582 unsigned long size
, pgprot_t prot
)
586 address
&= ~PGDIR_MASK
;
587 end
= address
+ size
;
588 if (end
> PGDIR_SIZE
)
591 pte_t
* pte
= pte_alloc(pmd
, address
);
594 zeromap_pte_range(pte
, address
, end
- address
, prot
);
595 address
= (address
+ PMD_SIZE
) & PMD_MASK
;
597 } while (address
&& (address
< end
));
601 int zeromap_page_range(unsigned long address
, unsigned long size
, pgprot_t prot
)
605 unsigned long beg
= address
;
606 unsigned long end
= address
+ size
;
608 dir
= pgd_offset(current
->mm
, address
);
609 flush_cache_range(current
->mm
, beg
, end
);
613 pmd_t
*pmd
= pmd_alloc(dir
, address
);
617 error
= zeromap_pmd_range(pmd
, address
, end
- address
, prot
);
620 address
= (address
+ PGDIR_SIZE
) & PGDIR_MASK
;
622 } while (address
&& (address
< end
));
623 flush_tlb_range(current
->mm
, beg
, end
);
628 * maps a range of physical memory into the requested pages. the old
629 * mappings are removed. any references to nonexistent pages results
630 * in null mappings (currently treated as "copy-on-access")
632 static inline void remap_pte_range(pte_t
* pte
, unsigned long address
, unsigned long size
,
633 unsigned long phys_addr
, pgprot_t prot
)
637 address
&= ~PMD_MASK
;
638 end
= address
+ size
;
643 pte_t oldpage
= *pte
;
646 mapnr
= MAP_NR(__va(phys_addr
));
647 if (mapnr
>= max_mapnr
|| PageReserved(mem_map
+mapnr
))
648 set_pte(pte
, mk_pte_phys(phys_addr
, prot
));
650 address
+= PAGE_SIZE
;
651 phys_addr
+= PAGE_SIZE
;
653 } while (address
&& (address
< end
));
656 static inline int remap_pmd_range(pmd_t
* pmd
, unsigned long address
, unsigned long size
,
657 unsigned long phys_addr
, pgprot_t prot
)
661 address
&= ~PGDIR_MASK
;
662 end
= address
+ size
;
663 if (end
> PGDIR_SIZE
)
665 phys_addr
-= address
;
667 pte_t
* pte
= pte_alloc(pmd
, address
);
670 remap_pte_range(pte
, address
, end
- address
, address
+ phys_addr
, prot
);
671 address
= (address
+ PMD_SIZE
) & PMD_MASK
;
673 } while (address
&& (address
< end
));
677 int remap_page_range(unsigned long from
, unsigned long phys_addr
, unsigned long size
, pgprot_t prot
)
681 unsigned long beg
= from
;
682 unsigned long end
= from
+ size
;
685 dir
= pgd_offset(current
->mm
, from
);
686 flush_cache_range(current
->mm
, beg
, end
);
690 pmd_t
*pmd
= pmd_alloc(dir
, from
);
694 error
= remap_pmd_range(pmd
, from
, end
- from
, phys_addr
+ from
, prot
);
697 from
= (from
+ PGDIR_SIZE
) & PGDIR_MASK
;
699 } while (from
&& (from
< end
));
700 flush_tlb_range(current
->mm
, beg
, end
);
705 * Establish a new mapping:
706 * - flush the old one
707 * - update the page tables
708 * - inform the TLB about the new one
710 static inline void establish_pte(struct vm_area_struct
* vma
, unsigned long address
, pte_t
*page_table
, pte_t entry
)
712 flush_tlb_page(vma
, address
);
713 set_pte(page_table
, entry
);
714 update_mmu_cache(vma
, address
, entry
);
718 * This routine handles present pages, when users try to write
719 * to a shared page. It is done by copying the page to a new address
720 * and decrementing the shared-page counter for the old page.
722 * Goto-purists beware: the only reason for goto's here is that it results
723 * in better assembly code.. The "default" path will see no jumps at all.
725 * Note that this routine assumes that the protection checks have been
726 * done by the caller (the low-level page fault routine in most cases).
727 * Thus we can safely just mark it writable once we've done any necessary
730 * We also mark the page dirty at this point even though the page will
731 * change only once the write actually happens. This avoids a few races,
732 * and potentially makes it more efficient.
734 * We enter with the page table read-lock held, and need to exit without
737 static int do_wp_page(struct task_struct
* tsk
, struct vm_area_struct
* vma
,
738 unsigned long address
, pte_t
*page_table
, pte_t pte
)
740 unsigned long map_nr
;
741 struct page
*old_page
, *new_page
;
743 map_nr
= pte_pagenr(pte
);
744 if (map_nr
>= max_mapnr
)
747 old_page
= mem_map
+ map_nr
;
750 * We can avoid the copy if:
751 * - we're the only user (count == 1)
752 * - the only other user is the swap cache,
753 * and the only swap cache user is itself,
754 * in which case we can remove the page
755 * from the swap cache.
757 switch (page_count(old_page
)) {
760 * Lock the page so that no one can look it up from
761 * the swap cache, grab a reference and start using it.
762 * Can not do lock_page, holding page_table_lock.
764 if (!PageSwapCache(old_page
) || TryLockPage(old_page
))
766 if (is_page_shared(old_page
)) {
767 UnlockPage(old_page
);
770 delete_from_swap_cache_nolock(old_page
);
771 UnlockPage(old_page
);
774 flush_cache_page(vma
, address
);
775 establish_pte(vma
, address
, page_table
, pte_mkyoung(pte_mkdirty(pte_mkwrite(pte
))));
776 spin_unlock(&tsk
->mm
->page_table_lock
);
781 * Ok, we need to copy. Oh, well..
783 spin_unlock(&tsk
->mm
->page_table_lock
);
784 new_page
= alloc_page(GFP_HIGHUSER
);
787 spin_lock(&tsk
->mm
->page_table_lock
);
790 * Re-check the pte - we dropped the lock
792 if (pte_val(*page_table
) == pte_val(pte
)) {
793 if (PageReserved(old_page
))
795 copy_cow_page(old_page
, new_page
, address
);
796 flush_page_to_ram(new_page
);
797 flush_cache_page(vma
, address
);
798 establish_pte(vma
, address
, page_table
, pte_mkwrite(pte_mkdirty(mk_pte(new_page
, vma
->vm_page_prot
))));
800 /* Free the old page.. */
803 spin_unlock(&tsk
->mm
->page_table_lock
);
804 __free_page(new_page
);
808 spin_unlock(&tsk
->mm
->page_table_lock
);
809 printk("do_wp_page: bogus page at address %08lx (nr %ld)\n",address
,map_nr
);
814 * This function zeroes out partial mmap'ed pages at truncation time..
816 static void partial_clear(struct vm_area_struct
*vma
, unsigned long address
)
822 pte_t
*page_table
, pte
;
824 page_dir
= pgd_offset(vma
->vm_mm
, address
);
825 if (pgd_none(*page_dir
))
827 if (pgd_bad(*page_dir
)) {
828 pgd_ERROR(*page_dir
);
832 page_middle
= pmd_offset(page_dir
, address
);
833 if (pmd_none(*page_middle
))
835 if (pmd_bad(*page_middle
)) {
836 pmd_ERROR(*page_middle
);
837 pmd_clear(page_middle
);
840 page_table
= pte_offset(page_middle
, address
);
842 if (!pte_present(pte
))
844 flush_cache_page(vma
, address
);
845 page
= pte_page(pte
);
846 if (page
-mem_map
>= max_mapnr
)
848 offset
= address
& ~PAGE_MASK
;
849 memclear_highpage_flush(page
, offset
, PAGE_SIZE
- offset
);
853 * Handle all mappings that got truncated by a "truncate()"
856 * NOTE! We have to be ready to update the memory sharing
857 * between the file and the memory map for a potential last
858 * incomplete page. Ugly, but necessary.
860 void vmtruncate(struct inode
* inode
, loff_t offset
)
862 unsigned long partial
, pgoff
;
863 struct vm_area_struct
* mpnt
;
864 struct address_space
*mapping
= inode
->i_mapping
;
866 if (inode
->i_size
< offset
)
868 inode
->i_size
= offset
;
869 truncate_inode_pages(mapping
, offset
);
870 spin_lock(&mapping
->i_shared_lock
);
871 if (!mapping
->i_mmap
)
874 pgoff
= (offset
+ PAGE_CACHE_SIZE
- 1) >> PAGE_CACHE_SHIFT
;
875 partial
= (unsigned long)offset
& (PAGE_CACHE_SIZE
- 1);
877 mpnt
= mapping
->i_mmap
;
879 struct mm_struct
*mm
= mpnt
->vm_mm
;
880 unsigned long start
= mpnt
->vm_start
;
881 unsigned long end
= mpnt
->vm_end
;
882 unsigned long len
= end
- start
;
885 /* mapping wholly truncated? */
886 if (mpnt
->vm_pgoff
>= pgoff
) {
887 flush_cache_range(mm
, start
, end
);
888 zap_page_range(mm
, start
, len
);
889 flush_tlb_range(mm
, start
, end
);
893 /* mapping wholly unaffected? */
894 len
= len
>> PAGE_SHIFT
;
895 diff
= pgoff
- mpnt
->vm_pgoff
;
899 /* Ok, partially affected.. */
900 start
+= diff
<< PAGE_SHIFT
;
901 len
= (len
- diff
) << PAGE_SHIFT
;
902 if (start
& ~PAGE_MASK
) {
903 partial_clear(mpnt
, start
);
904 start
= (start
+ ~PAGE_MASK
) & PAGE_MASK
;
906 flush_cache_range(mm
, start
, end
);
907 zap_page_range(mm
, start
, len
);
908 flush_tlb_range(mm
, start
, end
);
909 } while ((mpnt
= mpnt
->vm_next_share
) != NULL
);
911 spin_unlock(&mapping
->i_shared_lock
);
913 /* this should go into ->truncate */
914 inode
->i_size
= offset
;
915 if (inode
->i_op
&& inode
->i_op
->truncate
)
916 inode
->i_op
->truncate(inode
);
922 * Primitive swap readahead code. We simply read an aligned block of
923 * (1 << page_cluster) entries in the swap area. This method is chosen
924 * because it doesn't cost us any seek time. We also make sure to queue
925 * the 'original' request together with the readahead ones...
927 void swapin_readahead(swp_entry_t entry
)
930 struct page
*new_page
;
931 unsigned long offset
;
934 * Get the number of handles we should do readahead io to. Also,
935 * grab temporary references on them, releasing them as io completes.
937 num
= valid_swaphandles(entry
, &offset
);
938 for (i
= 0; i
< num
; offset
++, i
++) {
939 /* Don't block on I/O for read-ahead */
940 if (atomic_read(&nr_async_pages
) >= pager_daemon
.swap_cluster
) {
942 swap_free(SWP_ENTRY(SWP_TYPE(entry
), offset
++));
945 /* Ok, do the async read-ahead now */
946 new_page
= read_swap_cache_async(SWP_ENTRY(SWP_TYPE(entry
), offset
), 0);
947 if (new_page
!= NULL
)
948 __free_page(new_page
);
949 swap_free(SWP_ENTRY(SWP_TYPE(entry
), offset
));
954 static int do_swap_page(struct task_struct
* tsk
,
955 struct vm_area_struct
* vma
, unsigned long address
,
956 pte_t
* page_table
, swp_entry_t entry
, int write_access
)
958 struct page
*page
= lookup_swap_cache(entry
);
963 swapin_readahead(entry
);
964 page
= read_swap_cache(entry
);
969 flush_page_to_ram(page
);
970 flush_icache_page(vma
, page
);
976 pte
= mk_pte(page
, vma
->vm_page_prot
);
978 set_bit(PG_swap_entry
, &page
->flags
);
981 * Freeze the "shared"ness of the page, ie page_count + swap_count.
982 * Must lock page before transferring our swap count to already
983 * obtained page count.
987 if (write_access
&& !is_page_shared(page
)) {
988 delete_from_swap_cache_nolock(page
);
990 page
= replace_with_highmem(page
);
991 pte
= mk_pte(page
, vma
->vm_page_prot
);
992 pte
= pte_mkwrite(pte_mkdirty(pte
));
996 set_pte(page_table
, pte
);
997 /* No need to invalidate - it was non-present before */
998 update_mmu_cache(vma
, address
, pte
);
1003 * This only needs the MM semaphore
1005 static int do_anonymous_page(struct task_struct
* tsk
, struct vm_area_struct
* vma
, pte_t
*page_table
, int write_access
, unsigned long addr
)
1008 struct page
*page
= NULL
;
1009 pte_t entry
= pte_wrprotect(mk_pte(ZERO_PAGE(addr
), vma
->vm_page_prot
));
1011 page
= alloc_page(GFP_HIGHUSER
);
1014 if (PageHighMem(page
))
1016 clear_highpage(page
);
1017 entry
= pte_mkwrite(pte_mkdirty(mk_pte(page
, vma
->vm_page_prot
)));
1020 flush_page_to_ram(page
);
1022 set_pte(page_table
, entry
);
1023 /* No need to invalidate - it was non-present before */
1024 update_mmu_cache(vma
, addr
, entry
);
1029 * do_no_page() tries to create a new page mapping. It aggressively
1030 * tries to share with existing pages, but makes a separate copy if
1031 * the "write_access" parameter is true in order to avoid the next
1034 * As this is called only for pages that do not currently exist, we
1035 * do not need to flush old virtual caches or the TLB.
1037 * This is called with the MM semaphore held.
1039 static int do_no_page(struct task_struct
* tsk
, struct vm_area_struct
* vma
,
1040 unsigned long address
, int write_access
, pte_t
*page_table
)
1042 struct page
* new_page
;
1045 if (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
)
1046 return do_anonymous_page(tsk
, vma
, page_table
, write_access
, address
);
1049 * The third argument is "no_share", which tells the low-level code
1050 * to copy, not share the page even if sharing is possible. It's
1051 * essentially an early COW detection.
1053 new_page
= vma
->vm_ops
->nopage(vma
, address
& PAGE_MASK
, (vma
->vm_flags
& VM_SHARED
)?0:write_access
);
1054 if (new_page
== NULL
) /* no page was available -- SIGBUS */
1056 if (new_page
== NOPAGE_OOM
)
1061 * This silly early PAGE_DIRTY setting removes a race
1062 * due to the bad i386 page protection. But it's valid
1063 * for other architectures too.
1065 * Note that if write_access is true, we either now have
1066 * an exclusive copy of the page, or this is a shared mapping,
1067 * so we can make it writable and dirty to avoid having to
1068 * handle that later.
1070 flush_page_to_ram(new_page
);
1071 flush_icache_page(vma
, new_page
);
1072 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1074 entry
= pte_mkwrite(pte_mkdirty(entry
));
1075 } else if (page_count(new_page
) > 1 &&
1076 !(vma
->vm_flags
& VM_SHARED
))
1077 entry
= pte_wrprotect(entry
);
1078 set_pte(page_table
, entry
);
1079 /* no need to invalidate: a not-present page shouldn't be cached */
1080 update_mmu_cache(vma
, address
, entry
);
1085 * These routines also need to handle stuff like marking pages dirty
1086 * and/or accessed for architectures that don't do it in hardware (most
1087 * RISC architectures). The early dirtying is also good on the i386.
1089 * There is also a hook called "update_mmu_cache()" that architectures
1090 * with external mmu caches can use to update those (ie the Sparc or
1091 * PowerPC hashed page tables that act as extended TLBs).
1093 * Note the "page_table_lock". It is to protect against kswapd removing
1094 * pages from under us. Note that kswapd only ever _removes_ pages, never
1095 * adds them. As such, once we have noticed that the page is not present,
1096 * we can drop the lock early.
1098 * The adding of pages is protected by the MM semaphore (which we hold),
1099 * so we don't need to worry about a page being suddenly been added into
1102 static inline int handle_pte_fault(struct task_struct
*tsk
,
1103 struct vm_area_struct
* vma
, unsigned long address
,
1104 int write_access
, pte_t
* pte
)
1109 if (!pte_present(entry
)) {
1110 if (pte_none(entry
))
1111 return do_no_page(tsk
, vma
, address
, write_access
, pte
);
1112 return do_swap_page(tsk
, vma
, address
, pte
, pte_to_swp_entry(entry
), write_access
);
1116 * Ok, the entry was present, we need to get the page table
1117 * lock to synchronize with kswapd, and verify that the entry
1118 * didn't change from under us..
1120 spin_lock(&tsk
->mm
->page_table_lock
);
1121 if (pte_val(entry
) == pte_val(*pte
)) {
1123 if (!pte_write(entry
))
1124 return do_wp_page(tsk
, vma
, address
, pte
, entry
);
1126 entry
= pte_mkdirty(entry
);
1128 entry
= pte_mkyoung(entry
);
1129 establish_pte(vma
, address
, pte
, entry
);
1131 spin_unlock(&tsk
->mm
->page_table_lock
);
1136 * By the time we get here, we already hold the mm semaphore
1138 int handle_mm_fault(struct task_struct
*tsk
, struct vm_area_struct
* vma
,
1139 unsigned long address
, int write_access
)
1145 pgd
= pgd_offset(vma
->vm_mm
, address
);
1146 pmd
= pmd_alloc(pgd
, address
);
1149 pte_t
* pte
= pte_alloc(pmd
, address
);
1151 ret
= handle_pte_fault(tsk
, vma
, address
, write_access
, pte
);
1157 * Simplistic page force-in..
1159 int make_pages_present(unsigned long addr
, unsigned long end
)
1162 struct task_struct
*tsk
= current
;
1163 struct vm_area_struct
* vma
;
1165 vma
= find_vma(tsk
->mm
, addr
);
1166 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
1170 if (handle_mm_fault(tsk
, vma
, addr
, write
) < 0)
1173 } while (addr
< end
);