4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
41 #include <linux/kernel_stat.h>
43 #include <linux/hugetlb.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/highmem.h>
47 #include <linux/pagemap.h>
48 #include <linux/rmap.h>
49 #include <linux/module.h>
50 #include <linux/init.h>
52 #include <asm/pgalloc.h>
53 #include <asm/uaccess.h>
55 #include <asm/tlbflush.h>
56 #include <asm/pgtable.h>
58 #include <linux/swapops.h>
59 #include <linux/elf.h>
61 #ifndef CONFIG_NEED_MULTIPLE_NODES
62 /* use the per-pgdat data instead for discontigmem - mbligh */
63 unsigned long max_mapnr
;
66 EXPORT_SYMBOL(max_mapnr
);
67 EXPORT_SYMBOL(mem_map
);
70 unsigned long num_physpages
;
72 * A number of key systems in x86 including ioremap() rely on the assumption
73 * that high_memory defines the upper bound on direct map memory, then end
74 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
75 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
79 unsigned long vmalloc_earlyreserve
;
81 EXPORT_SYMBOL(num_physpages
);
82 EXPORT_SYMBOL(high_memory
);
83 EXPORT_SYMBOL(vmalloc_earlyreserve
);
86 * If a p?d_bad entry is found while walking page tables, report
87 * the error, before resetting entry to p?d_none. Usually (but
88 * very seldom) called out from the p?d_none_or_clear_bad macros.
91 void pgd_clear_bad(pgd_t
*pgd
)
97 void pud_clear_bad(pud_t
*pud
)
103 void pmd_clear_bad(pmd_t
*pmd
)
110 * Note: this doesn't free the actual pages themselves. That
111 * has been handled earlier when unmapping all the memory regions.
113 static void free_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
)
115 struct page
*page
= pmd_page(*pmd
);
117 pte_free_tlb(tlb
, page
);
118 dec_page_state(nr_page_table_pages
);
122 static inline void free_pmd_range(struct mmu_gather
*tlb
, pud_t
*pud
,
123 unsigned long addr
, unsigned long end
,
124 unsigned long floor
, unsigned long ceiling
)
131 pmd
= pmd_offset(pud
, addr
);
133 next
= pmd_addr_end(addr
, end
);
134 if (pmd_none_or_clear_bad(pmd
))
136 free_pte_range(tlb
, pmd
);
137 } while (pmd
++, addr
= next
, addr
!= end
);
147 if (end
- 1 > ceiling
- 1)
150 pmd
= pmd_offset(pud
, start
);
152 pmd_free_tlb(tlb
, pmd
);
155 static inline void free_pud_range(struct mmu_gather
*tlb
, pgd_t
*pgd
,
156 unsigned long addr
, unsigned long end
,
157 unsigned long floor
, unsigned long ceiling
)
164 pud
= pud_offset(pgd
, addr
);
166 next
= pud_addr_end(addr
, end
);
167 if (pud_none_or_clear_bad(pud
))
169 free_pmd_range(tlb
, pud
, addr
, next
, floor
, ceiling
);
170 } while (pud
++, addr
= next
, addr
!= end
);
176 ceiling
&= PGDIR_MASK
;
180 if (end
- 1 > ceiling
- 1)
183 pud
= pud_offset(pgd
, start
);
185 pud_free_tlb(tlb
, pud
);
189 * This function frees user-level page tables of a process.
191 * Must be called with pagetable lock held.
193 void free_pgd_range(struct mmu_gather
**tlb
,
194 unsigned long addr
, unsigned long end
,
195 unsigned long floor
, unsigned long ceiling
)
202 * The next few lines have given us lots of grief...
204 * Why are we testing PMD* at this top level? Because often
205 * there will be no work to do at all, and we'd prefer not to
206 * go all the way down to the bottom just to discover that.
208 * Why all these "- 1"s? Because 0 represents both the bottom
209 * of the address space and the top of it (using -1 for the
210 * top wouldn't help much: the masks would do the wrong thing).
211 * The rule is that addr 0 and floor 0 refer to the bottom of
212 * the address space, but end 0 and ceiling 0 refer to the top
213 * Comparisons need to use "end - 1" and "ceiling - 1" (though
214 * that end 0 case should be mythical).
216 * Wherever addr is brought up or ceiling brought down, we must
217 * be careful to reject "the opposite 0" before it confuses the
218 * subsequent tests. But what about where end is brought down
219 * by PMD_SIZE below? no, end can't go down to 0 there.
221 * Whereas we round start (addr) and ceiling down, by different
222 * masks at different levels, in order to test whether a table
223 * now has no other vmas using it, so can be freed, we don't
224 * bother to round floor or end up - the tests don't need that.
238 if (end
- 1 > ceiling
- 1)
244 pgd
= pgd_offset((*tlb
)->mm
, addr
);
246 next
= pgd_addr_end(addr
, end
);
247 if (pgd_none_or_clear_bad(pgd
))
249 free_pud_range(*tlb
, pgd
, addr
, next
, floor
, ceiling
);
250 } while (pgd
++, addr
= next
, addr
!= end
);
253 flush_tlb_pgtables((*tlb
)->mm
, start
, end
);
256 void free_pgtables(struct mmu_gather
**tlb
, struct vm_area_struct
*vma
,
257 unsigned long floor
, unsigned long ceiling
)
260 struct vm_area_struct
*next
= vma
->vm_next
;
261 unsigned long addr
= vma
->vm_start
;
263 if (is_hugepage_only_range(vma
->vm_mm
, addr
, HPAGE_SIZE
)) {
264 hugetlb_free_pgd_range(tlb
, addr
, vma
->vm_end
,
265 floor
, next
? next
->vm_start
: ceiling
);
268 * Optimization: gather nearby vmas into one call down
270 while (next
&& next
->vm_start
<= vma
->vm_end
+ PMD_SIZE
271 && !is_hugepage_only_range(vma
->vm_mm
, next
->vm_start
,
276 free_pgd_range(tlb
, addr
, vma
->vm_end
,
277 floor
, next
? next
->vm_start
: ceiling
);
283 pte_t fastcall
*pte_alloc_map(struct mm_struct
*mm
, pmd_t
*pmd
,
284 unsigned long address
)
286 if (!pmd_present(*pmd
)) {
289 spin_unlock(&mm
->page_table_lock
);
290 new = pte_alloc_one(mm
, address
);
291 spin_lock(&mm
->page_table_lock
);
295 * Because we dropped the lock, we should re-check the
296 * entry, as somebody else could have populated it..
298 if (pmd_present(*pmd
)) {
303 inc_page_state(nr_page_table_pages
);
304 pmd_populate(mm
, pmd
, new);
307 return pte_offset_map(pmd
, address
);
310 pte_t fastcall
* pte_alloc_kernel(struct mm_struct
*mm
, pmd_t
*pmd
, unsigned long address
)
312 if (!pmd_present(*pmd
)) {
315 spin_unlock(&mm
->page_table_lock
);
316 new = pte_alloc_one_kernel(mm
, address
);
317 spin_lock(&mm
->page_table_lock
);
322 * Because we dropped the lock, we should re-check the
323 * entry, as somebody else could have populated it..
325 if (pmd_present(*pmd
)) {
326 pte_free_kernel(new);
329 pmd_populate_kernel(mm
, pmd
, new);
332 return pte_offset_kernel(pmd
, address
);
335 static inline void add_mm_rss(struct mm_struct
*mm
, int file_rss
, int anon_rss
)
338 add_mm_counter(mm
, file_rss
, file_rss
);
340 add_mm_counter(mm
, anon_rss
, anon_rss
);
343 #define NO_RSS 2 /* Increment neither file_rss nor anon_rss */
346 * copy one vm_area from one task to the other. Assumes the page tables
347 * already present in the new task to be cleared in the whole range
348 * covered by this vma.
350 * dst->page_table_lock is held on entry and exit,
351 * but may be dropped within p[mg]d_alloc() and pte_alloc_map().
355 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
356 pte_t
*dst_pte
, pte_t
*src_pte
, unsigned long vm_flags
,
359 pte_t pte
= *src_pte
;
364 /* pte contains position in swap or file, so copy. */
365 if (unlikely(!pte_present(pte
))) {
366 if (!pte_file(pte
)) {
367 swap_duplicate(pte_to_swp_entry(pte
));
368 /* make sure dst_mm is on swapoff's mmlist. */
369 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
370 spin_lock(&mmlist_lock
);
371 list_add(&dst_mm
->mmlist
, &src_mm
->mmlist
);
372 spin_unlock(&mmlist_lock
);
379 /* the pte points outside of valid memory, the
380 * mapping is assumed to be good, meaningful
381 * and not mapped via rmap - duplicate the
386 page
= pfn_to_page(pfn
);
388 if (!page
|| PageReserved(page
))
392 * If it's a COW mapping, write protect it both
393 * in the parent and the child
395 if ((vm_flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
) {
396 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
401 * If it's a shared mapping, mark it clean in
404 if (vm_flags
& VM_SHARED
)
405 pte
= pte_mkclean(pte
);
406 pte
= pte_mkold(pte
);
409 anon
= !!PageAnon(page
);
412 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
416 static int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
417 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
418 unsigned long addr
, unsigned long end
)
420 pte_t
*src_pte
, *dst_pte
;
421 unsigned long vm_flags
= vma
->vm_flags
;
423 int rss
[NO_RSS
+1], anon
;
427 dst_pte
= pte_alloc_map(dst_mm
, dst_pmd
, addr
);
430 src_pte
= pte_offset_map_nested(src_pmd
, addr
);
432 spin_lock(&src_mm
->page_table_lock
);
435 * We are holding two locks at this point - either of them
436 * could generate latencies in another task on another CPU.
438 if (progress
>= 32) {
440 if (need_resched() ||
441 need_lockbreak(&src_mm
->page_table_lock
) ||
442 need_lockbreak(&dst_mm
->page_table_lock
))
445 if (pte_none(*src_pte
)) {
449 anon
= copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
,
453 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
454 spin_unlock(&src_mm
->page_table_lock
);
456 pte_unmap_nested(src_pte
- 1);
457 pte_unmap(dst_pte
- 1);
458 add_mm_rss(dst_mm
, rss
[0], rss
[1]);
459 cond_resched_lock(&dst_mm
->page_table_lock
);
465 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
466 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
467 unsigned long addr
, unsigned long end
)
469 pmd_t
*src_pmd
, *dst_pmd
;
472 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
475 src_pmd
= pmd_offset(src_pud
, addr
);
477 next
= pmd_addr_end(addr
, end
);
478 if (pmd_none_or_clear_bad(src_pmd
))
480 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
483 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
487 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
488 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
489 unsigned long addr
, unsigned long end
)
491 pud_t
*src_pud
, *dst_pud
;
494 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
497 src_pud
= pud_offset(src_pgd
, addr
);
499 next
= pud_addr_end(addr
, end
);
500 if (pud_none_or_clear_bad(src_pud
))
502 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
505 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
509 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
510 struct vm_area_struct
*vma
)
512 pgd_t
*src_pgd
, *dst_pgd
;
514 unsigned long addr
= vma
->vm_start
;
515 unsigned long end
= vma
->vm_end
;
518 * Don't copy ptes where a page fault will fill them correctly.
519 * Fork becomes much lighter when there are big shared or private
520 * readonly mappings. The tradeoff is that copy_page_range is more
521 * efficient than faulting.
523 if (!(vma
->vm_flags
& (VM_HUGETLB
|VM_NONLINEAR
|VM_RESERVED
))) {
528 if (is_vm_hugetlb_page(vma
))
529 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
531 dst_pgd
= pgd_offset(dst_mm
, addr
);
532 src_pgd
= pgd_offset(src_mm
, addr
);
534 next
= pgd_addr_end(addr
, end
);
535 if (pgd_none_or_clear_bad(src_pgd
))
537 if (copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
540 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
544 static void zap_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
,
545 unsigned long addr
, unsigned long end
,
546 struct zap_details
*details
)
552 pte
= pte_offset_map(pmd
, addr
);
557 if (pte_present(ptent
)) {
558 struct page
*page
= NULL
;
559 unsigned long pfn
= pte_pfn(ptent
);
560 if (pfn_valid(pfn
)) {
561 page
= pfn_to_page(pfn
);
562 if (PageReserved(page
))
565 if (unlikely(details
) && page
) {
567 * unmap_shared_mapping_pages() wants to
568 * invalidate cache without truncating:
569 * unmap shared but keep private pages.
571 if (details
->check_mapping
&&
572 details
->check_mapping
!= page
->mapping
)
575 * Each page->index must be checked when
576 * invalidating or truncating nonlinear.
578 if (details
->nonlinear_vma
&&
579 (page
->index
< details
->first_index
||
580 page
->index
> details
->last_index
))
583 ptent
= ptep_get_and_clear_full(tlb
->mm
, addr
, pte
,
585 tlb_remove_tlb_entry(tlb
, pte
, addr
);
588 if (unlikely(details
) && details
->nonlinear_vma
589 && linear_page_index(details
->nonlinear_vma
,
590 addr
) != page
->index
)
591 set_pte_at(tlb
->mm
, addr
, pte
,
592 pgoff_to_pte(page
->index
));
596 if (pte_dirty(ptent
))
597 set_page_dirty(page
);
598 if (pte_young(ptent
))
599 mark_page_accessed(page
);
602 page_remove_rmap(page
);
603 tlb_remove_page(tlb
, page
);
607 * If details->check_mapping, we leave swap entries;
608 * if details->nonlinear_vma, we leave file entries.
610 if (unlikely(details
))
612 if (!pte_file(ptent
))
613 free_swap_and_cache(pte_to_swp_entry(ptent
));
614 pte_clear_full(tlb
->mm
, addr
, pte
, tlb
->fullmm
);
615 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
617 add_mm_rss(tlb
->mm
, -file_rss
, -anon_rss
);
621 static inline void zap_pmd_range(struct mmu_gather
*tlb
, pud_t
*pud
,
622 unsigned long addr
, unsigned long end
,
623 struct zap_details
*details
)
628 pmd
= pmd_offset(pud
, addr
);
630 next
= pmd_addr_end(addr
, end
);
631 if (pmd_none_or_clear_bad(pmd
))
633 zap_pte_range(tlb
, pmd
, addr
, next
, details
);
634 } while (pmd
++, addr
= next
, addr
!= end
);
637 static inline void zap_pud_range(struct mmu_gather
*tlb
, pgd_t
*pgd
,
638 unsigned long addr
, unsigned long end
,
639 struct zap_details
*details
)
644 pud
= pud_offset(pgd
, addr
);
646 next
= pud_addr_end(addr
, end
);
647 if (pud_none_or_clear_bad(pud
))
649 zap_pmd_range(tlb
, pud
, addr
, next
, details
);
650 } while (pud
++, addr
= next
, addr
!= end
);
653 static void unmap_page_range(struct mmu_gather
*tlb
, struct vm_area_struct
*vma
,
654 unsigned long addr
, unsigned long end
,
655 struct zap_details
*details
)
660 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
664 tlb_start_vma(tlb
, vma
);
665 pgd
= pgd_offset(vma
->vm_mm
, addr
);
667 next
= pgd_addr_end(addr
, end
);
668 if (pgd_none_or_clear_bad(pgd
))
670 zap_pud_range(tlb
, pgd
, addr
, next
, details
);
671 } while (pgd
++, addr
= next
, addr
!= end
);
672 tlb_end_vma(tlb
, vma
);
675 #ifdef CONFIG_PREEMPT
676 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
678 /* No preempt: go for improved straight-line efficiency */
679 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
683 * unmap_vmas - unmap a range of memory covered by a list of vma's
684 * @tlbp: address of the caller's struct mmu_gather
685 * @mm: the controlling mm_struct
686 * @vma: the starting vma
687 * @start_addr: virtual address at which to start unmapping
688 * @end_addr: virtual address at which to end unmapping
689 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
690 * @details: details of nonlinear truncation or shared cache invalidation
692 * Returns the end address of the unmapping (restart addr if interrupted).
694 * Unmap all pages in the vma list. Called under page_table_lock.
696 * We aim to not hold page_table_lock for too long (for scheduling latency
697 * reasons). So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
698 * return the ending mmu_gather to the caller.
700 * Only addresses between `start' and `end' will be unmapped.
702 * The VMA list must be sorted in ascending virtual address order.
704 * unmap_vmas() assumes that the caller will flush the whole unmapped address
705 * range after unmap_vmas() returns. So the only responsibility here is to
706 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
707 * drops the lock and schedules.
709 unsigned long unmap_vmas(struct mmu_gather
**tlbp
, struct mm_struct
*mm
,
710 struct vm_area_struct
*vma
, unsigned long start_addr
,
711 unsigned long end_addr
, unsigned long *nr_accounted
,
712 struct zap_details
*details
)
714 unsigned long zap_bytes
= ZAP_BLOCK_SIZE
;
715 unsigned long tlb_start
= 0; /* For tlb_finish_mmu */
716 int tlb_start_valid
= 0;
717 unsigned long start
= start_addr
;
718 spinlock_t
*i_mmap_lock
= details
? details
->i_mmap_lock
: NULL
;
719 int fullmm
= (*tlbp
)->fullmm
;
721 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
) {
724 start
= max(vma
->vm_start
, start_addr
);
725 if (start
>= vma
->vm_end
)
727 end
= min(vma
->vm_end
, end_addr
);
728 if (end
<= vma
->vm_start
)
731 if (vma
->vm_flags
& VM_ACCOUNT
)
732 *nr_accounted
+= (end
- start
) >> PAGE_SHIFT
;
734 while (start
!= end
) {
737 if (!tlb_start_valid
) {
742 if (is_vm_hugetlb_page(vma
)) {
744 unmap_hugepage_range(vma
, start
, end
);
746 block
= min(zap_bytes
, end
- start
);
747 unmap_page_range(*tlbp
, vma
, start
,
748 start
+ block
, details
);
753 if ((long)zap_bytes
> 0)
756 tlb_finish_mmu(*tlbp
, tlb_start
, start
);
758 if (need_resched() ||
759 need_lockbreak(&mm
->page_table_lock
) ||
760 (i_mmap_lock
&& need_lockbreak(i_mmap_lock
))) {
762 /* must reset count of rss freed */
763 *tlbp
= tlb_gather_mmu(mm
, fullmm
);
766 spin_unlock(&mm
->page_table_lock
);
768 spin_lock(&mm
->page_table_lock
);
771 *tlbp
= tlb_gather_mmu(mm
, fullmm
);
773 zap_bytes
= ZAP_BLOCK_SIZE
;
777 return start
; /* which is now the end (or restart) address */
781 * zap_page_range - remove user pages in a given range
782 * @vma: vm_area_struct holding the applicable pages
783 * @address: starting address of pages to zap
784 * @size: number of bytes to zap
785 * @details: details of nonlinear truncation or shared cache invalidation
787 unsigned long zap_page_range(struct vm_area_struct
*vma
, unsigned long address
,
788 unsigned long size
, struct zap_details
*details
)
790 struct mm_struct
*mm
= vma
->vm_mm
;
791 struct mmu_gather
*tlb
;
792 unsigned long end
= address
+ size
;
793 unsigned long nr_accounted
= 0;
795 if (is_vm_hugetlb_page(vma
)) {
796 zap_hugepage_range(vma
, address
, size
);
801 spin_lock(&mm
->page_table_lock
);
802 tlb
= tlb_gather_mmu(mm
, 0);
803 end
= unmap_vmas(&tlb
, mm
, vma
, address
, end
, &nr_accounted
, details
);
804 tlb_finish_mmu(tlb
, address
, end
);
805 spin_unlock(&mm
->page_table_lock
);
810 * Do a quick page-table lookup for a single page.
811 * mm->page_table_lock must be held.
813 static struct page
*__follow_page(struct mm_struct
*mm
, unsigned long address
,
814 int read
, int write
, int accessed
)
823 page
= follow_huge_addr(mm
, address
, write
);
827 pgd
= pgd_offset(mm
, address
);
828 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
831 pud
= pud_offset(pgd
, address
);
832 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
835 pmd
= pmd_offset(pud
, address
);
836 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
839 return follow_huge_pmd(mm
, address
, pmd
, write
);
841 ptep
= pte_offset_map(pmd
, address
);
847 if (pte_present(pte
)) {
848 if (write
&& !pte_write(pte
))
850 if (read
&& !pte_read(pte
))
853 if (pfn_valid(pfn
)) {
854 page
= pfn_to_page(pfn
);
856 if (write
&& !pte_dirty(pte
) &&!PageDirty(page
))
857 set_page_dirty(page
);
858 mark_page_accessed(page
);
869 follow_page(struct mm_struct
*mm
, unsigned long address
, int write
)
871 return __follow_page(mm
, address
, 0, write
, 1);
875 * check_user_page_readable() can be called frm niterrupt context by oprofile,
876 * so we need to avoid taking any non-irq-safe locks
878 int check_user_page_readable(struct mm_struct
*mm
, unsigned long address
)
880 return __follow_page(mm
, address
, 1, 0, 0) != NULL
;
882 EXPORT_SYMBOL(check_user_page_readable
);
885 untouched_anonymous_page(struct mm_struct
* mm
, struct vm_area_struct
*vma
,
886 unsigned long address
)
892 /* Check if the vma is for an anonymous mapping. */
893 if (vma
->vm_ops
&& vma
->vm_ops
->nopage
)
896 /* Check if page directory entry exists. */
897 pgd
= pgd_offset(mm
, address
);
898 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
901 pud
= pud_offset(pgd
, address
);
902 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
905 /* Check if page middle directory entry exists. */
906 pmd
= pmd_offset(pud
, address
);
907 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
910 /* There is a pte slot for 'address' in 'mm'. */
914 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
915 unsigned long start
, int len
, int write
, int force
,
916 struct page
**pages
, struct vm_area_struct
**vmas
)
922 * Require read or write permissions.
923 * If 'force' is set, we only require the "MAY" flags.
925 flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
926 flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
930 struct vm_area_struct
* vma
;
932 vma
= find_extend_vma(mm
, start
);
933 if (!vma
&& in_gate_area(tsk
, start
)) {
934 unsigned long pg
= start
& PAGE_MASK
;
935 struct vm_area_struct
*gate_vma
= get_gate_vma(tsk
);
940 if (write
) /* user gate pages are read-only */
941 return i
? : -EFAULT
;
943 pgd
= pgd_offset_k(pg
);
945 pgd
= pgd_offset_gate(mm
, pg
);
946 BUG_ON(pgd_none(*pgd
));
947 pud
= pud_offset(pgd
, pg
);
948 BUG_ON(pud_none(*pud
));
949 pmd
= pmd_offset(pud
, pg
);
951 return i
? : -EFAULT
;
952 pte
= pte_offset_map(pmd
, pg
);
953 if (pte_none(*pte
)) {
955 return i
? : -EFAULT
;
958 pages
[i
] = pte_page(*pte
);
970 if (!vma
|| (vma
->vm_flags
& VM_IO
)
971 || !(flags
& vma
->vm_flags
))
972 return i
? : -EFAULT
;
974 if (is_vm_hugetlb_page(vma
)) {
975 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
979 spin_lock(&mm
->page_table_lock
);
981 int write_access
= write
;
984 cond_resched_lock(&mm
->page_table_lock
);
985 while (!(page
= follow_page(mm
, start
, write_access
))) {
989 * Shortcut for anonymous pages. We don't want
990 * to force the creation of pages tables for
991 * insanely big anonymously mapped areas that
992 * nobody touched so far. This is important
993 * for doing a core dump for these mappings.
995 if (!write
&& untouched_anonymous_page(mm
,vma
,start
)) {
996 page
= ZERO_PAGE(start
);
999 spin_unlock(&mm
->page_table_lock
);
1000 ret
= __handle_mm_fault(mm
, vma
, start
, write_access
);
1003 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1004 * broken COW when necessary, even if maybe_mkwrite
1005 * decided not to set pte_write. We can thus safely do
1006 * subsequent page lookups as if they were reads.
1008 if (ret
& VM_FAULT_WRITE
)
1011 switch (ret
& ~VM_FAULT_WRITE
) {
1012 case VM_FAULT_MINOR
:
1015 case VM_FAULT_MAJOR
:
1018 case VM_FAULT_SIGBUS
:
1019 return i
? i
: -EFAULT
;
1021 return i
? i
: -ENOMEM
;
1025 spin_lock(&mm
->page_table_lock
);
1029 flush_dcache_page(page
);
1030 if (!PageReserved(page
))
1031 page_cache_get(page
);
1038 } while (len
&& start
< vma
->vm_end
);
1039 spin_unlock(&mm
->page_table_lock
);
1043 EXPORT_SYMBOL(get_user_pages
);
1045 static int zeromap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1046 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1050 pte
= pte_alloc_map(mm
, pmd
, addr
);
1054 pte_t zero_pte
= pte_wrprotect(mk_pte(ZERO_PAGE(addr
), prot
));
1055 BUG_ON(!pte_none(*pte
));
1056 set_pte_at(mm
, addr
, pte
, zero_pte
);
1057 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1062 static inline int zeromap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1063 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1068 pmd
= pmd_alloc(mm
, pud
, addr
);
1072 next
= pmd_addr_end(addr
, end
);
1073 if (zeromap_pte_range(mm
, pmd
, addr
, next
, prot
))
1075 } while (pmd
++, addr
= next
, addr
!= end
);
1079 static inline int zeromap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1080 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1085 pud
= pud_alloc(mm
, pgd
, addr
);
1089 next
= pud_addr_end(addr
, end
);
1090 if (zeromap_pmd_range(mm
, pud
, addr
, next
, prot
))
1092 } while (pud
++, addr
= next
, addr
!= end
);
1096 int zeromap_page_range(struct vm_area_struct
*vma
,
1097 unsigned long addr
, unsigned long size
, pgprot_t prot
)
1101 unsigned long end
= addr
+ size
;
1102 struct mm_struct
*mm
= vma
->vm_mm
;
1105 BUG_ON(addr
>= end
);
1106 pgd
= pgd_offset(mm
, addr
);
1107 flush_cache_range(vma
, addr
, end
);
1108 spin_lock(&mm
->page_table_lock
);
1110 next
= pgd_addr_end(addr
, end
);
1111 err
= zeromap_pud_range(mm
, pgd
, addr
, next
, prot
);
1114 } while (pgd
++, addr
= next
, addr
!= end
);
1115 spin_unlock(&mm
->page_table_lock
);
1120 * maps a range of physical memory into the requested pages. the old
1121 * mappings are removed. any references to nonexistent pages results
1122 * in null mappings (currently treated as "copy-on-access")
1124 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1125 unsigned long addr
, unsigned long end
,
1126 unsigned long pfn
, pgprot_t prot
)
1130 pte
= pte_alloc_map(mm
, pmd
, addr
);
1134 BUG_ON(!pte_none(*pte
));
1135 if (!pfn_valid(pfn
) || PageReserved(pfn_to_page(pfn
)))
1136 set_pte_at(mm
, addr
, pte
, pfn_pte(pfn
, prot
));
1138 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1143 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1144 unsigned long addr
, unsigned long end
,
1145 unsigned long pfn
, pgprot_t prot
)
1150 pfn
-= addr
>> PAGE_SHIFT
;
1151 pmd
= pmd_alloc(mm
, pud
, addr
);
1155 next
= pmd_addr_end(addr
, end
);
1156 if (remap_pte_range(mm
, pmd
, addr
, next
,
1157 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1159 } while (pmd
++, addr
= next
, addr
!= end
);
1163 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1164 unsigned long addr
, unsigned long end
,
1165 unsigned long pfn
, pgprot_t prot
)
1170 pfn
-= addr
>> PAGE_SHIFT
;
1171 pud
= pud_alloc(mm
, pgd
, addr
);
1175 next
= pud_addr_end(addr
, end
);
1176 if (remap_pmd_range(mm
, pud
, addr
, next
,
1177 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1179 } while (pud
++, addr
= next
, addr
!= end
);
1183 /* Note: this is only safe if the mm semaphore is held when called. */
1184 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1185 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1189 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1190 struct mm_struct
*mm
= vma
->vm_mm
;
1194 * Physically remapped pages are special. Tell the
1195 * rest of the world about it:
1196 * VM_IO tells people not to look at these pages
1197 * (accesses can have side effects).
1198 * VM_RESERVED tells swapout not to try to touch
1201 vma
->vm_flags
|= VM_IO
| VM_RESERVED
;
1203 BUG_ON(addr
>= end
);
1204 pfn
-= addr
>> PAGE_SHIFT
;
1205 pgd
= pgd_offset(mm
, addr
);
1206 flush_cache_range(vma
, addr
, end
);
1207 spin_lock(&mm
->page_table_lock
);
1209 next
= pgd_addr_end(addr
, end
);
1210 err
= remap_pud_range(mm
, pgd
, addr
, next
,
1211 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
1214 } while (pgd
++, addr
= next
, addr
!= end
);
1215 spin_unlock(&mm
->page_table_lock
);
1218 EXPORT_SYMBOL(remap_pfn_range
);
1221 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1222 * servicing faults for write access. In the normal case, do always want
1223 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1224 * that do not have writing enabled, when used by access_process_vm.
1226 static inline pte_t
maybe_mkwrite(pte_t pte
, struct vm_area_struct
*vma
)
1228 if (likely(vma
->vm_flags
& VM_WRITE
))
1229 pte
= pte_mkwrite(pte
);
1234 * This routine handles present pages, when users try to write
1235 * to a shared page. It is done by copying the page to a new address
1236 * and decrementing the shared-page counter for the old page.
1238 * Note that this routine assumes that the protection checks have been
1239 * done by the caller (the low-level page fault routine in most cases).
1240 * Thus we can safely just mark it writable once we've done any necessary
1243 * We also mark the page dirty at this point even though the page will
1244 * change only once the write actually happens. This avoids a few races,
1245 * and potentially makes it more efficient.
1247 * We hold the mm semaphore and the page_table_lock on entry and exit
1248 * with the page_table_lock released.
1250 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1251 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1254 struct page
*old_page
, *new_page
;
1255 unsigned long pfn
= pte_pfn(orig_pte
);
1257 int ret
= VM_FAULT_MINOR
;
1259 if (unlikely(!pfn_valid(pfn
))) {
1261 * Page table corrupted: show pte and kill process.
1263 pte_ERROR(orig_pte
);
1267 old_page
= pfn_to_page(pfn
);
1269 if (PageAnon(old_page
) && !TestSetPageLocked(old_page
)) {
1270 int reuse
= can_share_swap_page(old_page
);
1271 unlock_page(old_page
);
1273 flush_cache_page(vma
, address
, pfn
);
1274 entry
= pte_mkyoung(orig_pte
);
1275 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1276 ptep_set_access_flags(vma
, address
, page_table
, entry
, 1);
1277 update_mmu_cache(vma
, address
, entry
);
1278 lazy_mmu_prot_update(entry
);
1279 ret
|= VM_FAULT_WRITE
;
1285 * Ok, we need to copy. Oh, well..
1287 if (!PageReserved(old_page
))
1288 page_cache_get(old_page
);
1289 pte_unmap(page_table
);
1290 spin_unlock(&mm
->page_table_lock
);
1292 if (unlikely(anon_vma_prepare(vma
)))
1294 if (old_page
== ZERO_PAGE(address
)) {
1295 new_page
= alloc_zeroed_user_highpage(vma
, address
);
1299 new_page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1302 copy_user_highpage(new_page
, old_page
, address
);
1306 * Re-check the pte - we dropped the lock
1308 spin_lock(&mm
->page_table_lock
);
1309 page_table
= pte_offset_map(pmd
, address
);
1310 if (likely(pte_same(*page_table
, orig_pte
))) {
1311 if (PageReserved(old_page
))
1312 inc_mm_counter(mm
, anon_rss
);
1314 page_remove_rmap(old_page
);
1315 if (!PageAnon(old_page
)) {
1316 inc_mm_counter(mm
, anon_rss
);
1317 dec_mm_counter(mm
, file_rss
);
1320 flush_cache_page(vma
, address
, pfn
);
1321 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1322 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1323 ptep_establish(vma
, address
, page_table
, entry
);
1324 update_mmu_cache(vma
, address
, entry
);
1325 lazy_mmu_prot_update(entry
);
1327 lru_cache_add_active(new_page
);
1328 page_add_anon_rmap(new_page
, vma
, address
);
1330 /* Free the old page.. */
1331 new_page
= old_page
;
1332 ret
|= VM_FAULT_WRITE
;
1334 page_cache_release(new_page
);
1335 page_cache_release(old_page
);
1337 pte_unmap(page_table
);
1338 spin_unlock(&mm
->page_table_lock
);
1341 page_cache_release(old_page
);
1342 return VM_FAULT_OOM
;
1346 * Helper functions for unmap_mapping_range().
1348 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1350 * We have to restart searching the prio_tree whenever we drop the lock,
1351 * since the iterator is only valid while the lock is held, and anyway
1352 * a later vma might be split and reinserted earlier while lock dropped.
1354 * The list of nonlinear vmas could be handled more efficiently, using
1355 * a placeholder, but handle it in the same way until a need is shown.
1356 * It is important to search the prio_tree before nonlinear list: a vma
1357 * may become nonlinear and be shifted from prio_tree to nonlinear list
1358 * while the lock is dropped; but never shifted from list to prio_tree.
1360 * In order to make forward progress despite restarting the search,
1361 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1362 * quickly skip it next time around. Since the prio_tree search only
1363 * shows us those vmas affected by unmapping the range in question, we
1364 * can't efficiently keep all vmas in step with mapping->truncate_count:
1365 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1366 * mapping->truncate_count and vma->vm_truncate_count are protected by
1369 * In order to make forward progress despite repeatedly restarting some
1370 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1371 * and restart from that address when we reach that vma again. It might
1372 * have been split or merged, shrunk or extended, but never shifted: so
1373 * restart_addr remains valid so long as it remains in the vma's range.
1374 * unmap_mapping_range forces truncate_count to leap over page-aligned
1375 * values so we can save vma's restart_addr in its truncate_count field.
1377 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1379 static void reset_vma_truncate_counts(struct address_space
*mapping
)
1381 struct vm_area_struct
*vma
;
1382 struct prio_tree_iter iter
;
1384 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, 0, ULONG_MAX
)
1385 vma
->vm_truncate_count
= 0;
1386 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
1387 vma
->vm_truncate_count
= 0;
1390 static int unmap_mapping_range_vma(struct vm_area_struct
*vma
,
1391 unsigned long start_addr
, unsigned long end_addr
,
1392 struct zap_details
*details
)
1394 unsigned long restart_addr
;
1398 restart_addr
= vma
->vm_truncate_count
;
1399 if (is_restart_addr(restart_addr
) && start_addr
< restart_addr
) {
1400 start_addr
= restart_addr
;
1401 if (start_addr
>= end_addr
) {
1402 /* Top of vma has been split off since last time */
1403 vma
->vm_truncate_count
= details
->truncate_count
;
1408 restart_addr
= zap_page_range(vma
, start_addr
,
1409 end_addr
- start_addr
, details
);
1412 * We cannot rely on the break test in unmap_vmas:
1413 * on the one hand, we don't want to restart our loop
1414 * just because that broke out for the page_table_lock;
1415 * on the other hand, it does no test when vma is small.
1417 need_break
= need_resched() ||
1418 need_lockbreak(details
->i_mmap_lock
);
1420 if (restart_addr
>= end_addr
) {
1421 /* We have now completed this vma: mark it so */
1422 vma
->vm_truncate_count
= details
->truncate_count
;
1426 /* Note restart_addr in vma's truncate_count field */
1427 vma
->vm_truncate_count
= restart_addr
;
1432 spin_unlock(details
->i_mmap_lock
);
1434 spin_lock(details
->i_mmap_lock
);
1438 static inline void unmap_mapping_range_tree(struct prio_tree_root
*root
,
1439 struct zap_details
*details
)
1441 struct vm_area_struct
*vma
;
1442 struct prio_tree_iter iter
;
1443 pgoff_t vba
, vea
, zba
, zea
;
1446 vma_prio_tree_foreach(vma
, &iter
, root
,
1447 details
->first_index
, details
->last_index
) {
1448 /* Skip quickly over those we have already dealt with */
1449 if (vma
->vm_truncate_count
== details
->truncate_count
)
1452 vba
= vma
->vm_pgoff
;
1453 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
1454 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1455 zba
= details
->first_index
;
1458 zea
= details
->last_index
;
1462 if (unmap_mapping_range_vma(vma
,
1463 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
1464 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
1470 static inline void unmap_mapping_range_list(struct list_head
*head
,
1471 struct zap_details
*details
)
1473 struct vm_area_struct
*vma
;
1476 * In nonlinear VMAs there is no correspondence between virtual address
1477 * offset and file offset. So we must perform an exhaustive search
1478 * across *all* the pages in each nonlinear VMA, not just the pages
1479 * whose virtual address lies outside the file truncation point.
1482 list_for_each_entry(vma
, head
, shared
.vm_set
.list
) {
1483 /* Skip quickly over those we have already dealt with */
1484 if (vma
->vm_truncate_count
== details
->truncate_count
)
1486 details
->nonlinear_vma
= vma
;
1487 if (unmap_mapping_range_vma(vma
, vma
->vm_start
,
1488 vma
->vm_end
, details
) < 0)
1494 * unmap_mapping_range - unmap the portion of all mmaps
1495 * in the specified address_space corresponding to the specified
1496 * page range in the underlying file.
1497 * @mapping: the address space containing mmaps to be unmapped.
1498 * @holebegin: byte in first page to unmap, relative to the start of
1499 * the underlying file. This will be rounded down to a PAGE_SIZE
1500 * boundary. Note that this is different from vmtruncate(), which
1501 * must keep the partial page. In contrast, we must get rid of
1503 * @holelen: size of prospective hole in bytes. This will be rounded
1504 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1506 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1507 * but 0 when invalidating pagecache, don't throw away private data.
1509 void unmap_mapping_range(struct address_space
*mapping
,
1510 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
1512 struct zap_details details
;
1513 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
1514 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1516 /* Check for overflow. */
1517 if (sizeof(holelen
) > sizeof(hlen
)) {
1519 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1520 if (holeend
& ~(long long)ULONG_MAX
)
1521 hlen
= ULONG_MAX
- hba
+ 1;
1524 details
.check_mapping
= even_cows
? NULL
: mapping
;
1525 details
.nonlinear_vma
= NULL
;
1526 details
.first_index
= hba
;
1527 details
.last_index
= hba
+ hlen
- 1;
1528 if (details
.last_index
< details
.first_index
)
1529 details
.last_index
= ULONG_MAX
;
1530 details
.i_mmap_lock
= &mapping
->i_mmap_lock
;
1532 spin_lock(&mapping
->i_mmap_lock
);
1534 /* serialize i_size write against truncate_count write */
1536 /* Protect against page faults, and endless unmapping loops */
1537 mapping
->truncate_count
++;
1539 * For archs where spin_lock has inclusive semantics like ia64
1540 * this smp_mb() will prevent to read pagetable contents
1541 * before the truncate_count increment is visible to
1545 if (unlikely(is_restart_addr(mapping
->truncate_count
))) {
1546 if (mapping
->truncate_count
== 0)
1547 reset_vma_truncate_counts(mapping
);
1548 mapping
->truncate_count
++;
1550 details
.truncate_count
= mapping
->truncate_count
;
1552 if (unlikely(!prio_tree_empty(&mapping
->i_mmap
)))
1553 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
1554 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
)))
1555 unmap_mapping_range_list(&mapping
->i_mmap_nonlinear
, &details
);
1556 spin_unlock(&mapping
->i_mmap_lock
);
1558 EXPORT_SYMBOL(unmap_mapping_range
);
1561 * Handle all mappings that got truncated by a "truncate()"
1564 * NOTE! We have to be ready to update the memory sharing
1565 * between the file and the memory map for a potential last
1566 * incomplete page. Ugly, but necessary.
1568 int vmtruncate(struct inode
* inode
, loff_t offset
)
1570 struct address_space
*mapping
= inode
->i_mapping
;
1571 unsigned long limit
;
1573 if (inode
->i_size
< offset
)
1576 * truncation of in-use swapfiles is disallowed - it would cause
1577 * subsequent swapout to scribble on the now-freed blocks.
1579 if (IS_SWAPFILE(inode
))
1581 i_size_write(inode
, offset
);
1582 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
1583 truncate_inode_pages(mapping
, offset
);
1587 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1588 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
1590 if (offset
> inode
->i_sb
->s_maxbytes
)
1592 i_size_write(inode
, offset
);
1595 if (inode
->i_op
&& inode
->i_op
->truncate
)
1596 inode
->i_op
->truncate(inode
);
1599 send_sig(SIGXFSZ
, current
, 0);
1606 EXPORT_SYMBOL(vmtruncate
);
1609 * Primitive swap readahead code. We simply read an aligned block of
1610 * (1 << page_cluster) entries in the swap area. This method is chosen
1611 * because it doesn't cost us any seek time. We also make sure to queue
1612 * the 'original' request together with the readahead ones...
1614 * This has been extended to use the NUMA policies from the mm triggering
1617 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1619 void swapin_readahead(swp_entry_t entry
, unsigned long addr
,struct vm_area_struct
*vma
)
1622 struct vm_area_struct
*next_vma
= vma
? vma
->vm_next
: NULL
;
1625 struct page
*new_page
;
1626 unsigned long offset
;
1629 * Get the number of handles we should do readahead io to.
1631 num
= valid_swaphandles(entry
, &offset
);
1632 for (i
= 0; i
< num
; offset
++, i
++) {
1633 /* Ok, do the async read-ahead now */
1634 new_page
= read_swap_cache_async(swp_entry(swp_type(entry
),
1635 offset
), vma
, addr
);
1638 page_cache_release(new_page
);
1641 * Find the next applicable VMA for the NUMA policy.
1647 if (addr
>= vma
->vm_end
) {
1649 next_vma
= vma
? vma
->vm_next
: NULL
;
1651 if (vma
&& addr
< vma
->vm_start
)
1654 if (next_vma
&& addr
>= next_vma
->vm_start
) {
1656 next_vma
= vma
->vm_next
;
1661 lru_add_drain(); /* Push any new pages onto the LRU now */
1665 * We hold the mm semaphore and the page_table_lock on entry and
1666 * should release the pagetable lock on exit..
1668 static int do_swap_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1669 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1670 int write_access
, pte_t orig_pte
)
1675 int ret
= VM_FAULT_MINOR
;
1677 pte_unmap(page_table
);
1678 spin_unlock(&mm
->page_table_lock
);
1680 entry
= pte_to_swp_entry(orig_pte
);
1681 page
= lookup_swap_cache(entry
);
1683 swapin_readahead(entry
, address
, vma
);
1684 page
= read_swap_cache_async(entry
, vma
, address
);
1687 * Back out if somebody else faulted in this pte while
1688 * we released the page table lock.
1690 spin_lock(&mm
->page_table_lock
);
1691 page_table
= pte_offset_map(pmd
, address
);
1692 if (likely(pte_same(*page_table
, orig_pte
)))
1697 /* Had to read the page from swap area: Major fault */
1698 ret
= VM_FAULT_MAJOR
;
1699 inc_page_state(pgmajfault
);
1703 mark_page_accessed(page
);
1707 * Back out if somebody else faulted in this pte while we
1708 * released the page table lock.
1710 spin_lock(&mm
->page_table_lock
);
1711 page_table
= pte_offset_map(pmd
, address
);
1712 if (unlikely(!pte_same(*page_table
, orig_pte
))) {
1713 ret
= VM_FAULT_MINOR
;
1717 if (unlikely(!PageUptodate(page
))) {
1718 ret
= VM_FAULT_SIGBUS
;
1722 /* The page isn't present yet, go ahead with the fault. */
1724 inc_mm_counter(mm
, anon_rss
);
1725 pte
= mk_pte(page
, vma
->vm_page_prot
);
1726 if (write_access
&& can_share_swap_page(page
)) {
1727 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
1731 flush_icache_page(vma
, page
);
1732 set_pte_at(mm
, address
, page_table
, pte
);
1733 page_add_anon_rmap(page
, vma
, address
);
1737 remove_exclusive_swap_page(page
);
1741 if (do_wp_page(mm
, vma
, address
,
1742 page_table
, pmd
, pte
) == VM_FAULT_OOM
)
1747 /* No need to invalidate - it was non-present before */
1748 update_mmu_cache(vma
, address
, pte
);
1749 lazy_mmu_prot_update(pte
);
1751 pte_unmap(page_table
);
1752 spin_unlock(&mm
->page_table_lock
);
1756 pte_unmap(page_table
);
1757 spin_unlock(&mm
->page_table_lock
);
1759 page_cache_release(page
);
1764 * We are called with the MM semaphore and page_table_lock
1765 * spinlock held to protect against concurrent faults in
1766 * multithreaded programs.
1768 static int do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1769 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1774 /* Mapping of ZERO_PAGE - vm_page_prot is readonly */
1775 entry
= mk_pte(ZERO_PAGE(addr
), vma
->vm_page_prot
);
1780 /* Allocate our own private page. */
1781 pte_unmap(page_table
);
1782 spin_unlock(&mm
->page_table_lock
);
1784 if (unlikely(anon_vma_prepare(vma
)))
1786 page
= alloc_zeroed_user_highpage(vma
, address
);
1790 spin_lock(&mm
->page_table_lock
);
1791 page_table
= pte_offset_map(pmd
, address
);
1793 if (!pte_none(*page_table
)) {
1794 page_cache_release(page
);
1797 inc_mm_counter(mm
, anon_rss
);
1798 entry
= mk_pte(page
, vma
->vm_page_prot
);
1799 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1800 lru_cache_add_active(page
);
1801 SetPageReferenced(page
);
1802 page_add_anon_rmap(page
, vma
, address
);
1805 set_pte_at(mm
, address
, page_table
, entry
);
1807 /* No need to invalidate - it was non-present before */
1808 update_mmu_cache(vma
, address
, entry
);
1809 lazy_mmu_prot_update(entry
);
1811 pte_unmap(page_table
);
1812 spin_unlock(&mm
->page_table_lock
);
1813 return VM_FAULT_MINOR
;
1815 return VM_FAULT_OOM
;
1819 * do_no_page() tries to create a new page mapping. It aggressively
1820 * tries to share with existing pages, but makes a separate copy if
1821 * the "write_access" parameter is true in order to avoid the next
1824 * As this is called only for pages that do not currently exist, we
1825 * do not need to flush old virtual caches or the TLB.
1827 * This is called with the MM semaphore held and the page table
1828 * spinlock held. Exit with the spinlock released.
1830 static int do_no_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1831 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1834 struct page
*new_page
;
1835 struct address_space
*mapping
= NULL
;
1837 unsigned int sequence
= 0;
1838 int ret
= VM_FAULT_MINOR
;
1841 pte_unmap(page_table
);
1842 spin_unlock(&mm
->page_table_lock
);
1845 mapping
= vma
->vm_file
->f_mapping
;
1846 sequence
= mapping
->truncate_count
;
1847 smp_rmb(); /* serializes i_size against truncate_count */
1850 new_page
= vma
->vm_ops
->nopage(vma
, address
& PAGE_MASK
, &ret
);
1852 * No smp_rmb is needed here as long as there's a full
1853 * spin_lock/unlock sequence inside the ->nopage callback
1854 * (for the pagecache lookup) that acts as an implicit
1855 * smp_mb() and prevents the i_size read to happen
1856 * after the next truncate_count read.
1859 /* no page was available -- either SIGBUS or OOM */
1860 if (new_page
== NOPAGE_SIGBUS
)
1861 return VM_FAULT_SIGBUS
;
1862 if (new_page
== NOPAGE_OOM
)
1863 return VM_FAULT_OOM
;
1866 * Should we do an early C-O-W break?
1868 if (write_access
&& !(vma
->vm_flags
& VM_SHARED
)) {
1871 if (unlikely(anon_vma_prepare(vma
)))
1873 page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1876 copy_user_highpage(page
, new_page
, address
);
1877 page_cache_release(new_page
);
1882 spin_lock(&mm
->page_table_lock
);
1884 * For a file-backed vma, someone could have truncated or otherwise
1885 * invalidated this page. If unmap_mapping_range got called,
1886 * retry getting the page.
1888 if (mapping
&& unlikely(sequence
!= mapping
->truncate_count
)) {
1889 spin_unlock(&mm
->page_table_lock
);
1890 page_cache_release(new_page
);
1892 sequence
= mapping
->truncate_count
;
1896 page_table
= pte_offset_map(pmd
, address
);
1899 * This silly early PAGE_DIRTY setting removes a race
1900 * due to the bad i386 page protection. But it's valid
1901 * for other architectures too.
1903 * Note that if write_access is true, we either now have
1904 * an exclusive copy of the page, or this is a shared mapping,
1905 * so we can make it writable and dirty to avoid having to
1906 * handle that later.
1908 /* Only go through if we didn't race with anybody else... */
1909 if (pte_none(*page_table
)) {
1910 flush_icache_page(vma
, new_page
);
1911 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1913 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1914 set_pte_at(mm
, address
, page_table
, entry
);
1916 inc_mm_counter(mm
, anon_rss
);
1917 lru_cache_add_active(new_page
);
1918 page_add_anon_rmap(new_page
, vma
, address
);
1919 } else if (!PageReserved(new_page
)) {
1920 inc_mm_counter(mm
, file_rss
);
1921 page_add_file_rmap(new_page
);
1924 /* One of our sibling threads was faster, back out. */
1925 page_cache_release(new_page
);
1929 /* no need to invalidate: a not-present page shouldn't be cached */
1930 update_mmu_cache(vma
, address
, entry
);
1931 lazy_mmu_prot_update(entry
);
1933 pte_unmap(page_table
);
1934 spin_unlock(&mm
->page_table_lock
);
1937 page_cache_release(new_page
);
1938 return VM_FAULT_OOM
;
1942 * Fault of a previously existing named mapping. Repopulate the pte
1943 * from the encoded file_pte if possible. This enables swappable
1946 static int do_file_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1947 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1948 int write_access
, pte_t orig_pte
)
1953 pte_unmap(page_table
);
1954 spin_unlock(&mm
->page_table_lock
);
1956 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
))) {
1958 * Page table corrupted: show pte and kill process.
1960 pte_ERROR(orig_pte
);
1961 return VM_FAULT_OOM
;
1963 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
1965 pgoff
= pte_to_pgoff(orig_pte
);
1966 err
= vma
->vm_ops
->populate(vma
, address
& PAGE_MASK
, PAGE_SIZE
,
1967 vma
->vm_page_prot
, pgoff
, 0);
1969 return VM_FAULT_OOM
;
1971 return VM_FAULT_SIGBUS
;
1972 return VM_FAULT_MAJOR
;
1976 * These routines also need to handle stuff like marking pages dirty
1977 * and/or accessed for architectures that don't do it in hardware (most
1978 * RISC architectures). The early dirtying is also good on the i386.
1980 * There is also a hook called "update_mmu_cache()" that architectures
1981 * with external mmu caches can use to update those (ie the Sparc or
1982 * PowerPC hashed page tables that act as extended TLBs).
1984 * Note the "page_table_lock". It is to protect against kswapd removing
1985 * pages from under us. Note that kswapd only ever _removes_ pages, never
1986 * adds them. As such, once we have noticed that the page is not present,
1987 * we can drop the lock early.
1989 * The adding of pages is protected by the MM semaphore (which we hold),
1990 * so we don't need to worry about a page being suddenly been added into
1993 * We enter with the pagetable spinlock held, we are supposed to
1994 * release it when done.
1996 static inline int handle_pte_fault(struct mm_struct
*mm
,
1997 struct vm_area_struct
*vma
, unsigned long address
,
1998 pte_t
*pte
, pmd_t
*pmd
, int write_access
)
2003 if (!pte_present(entry
)) {
2004 if (pte_none(entry
)) {
2005 if (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
)
2006 return do_anonymous_page(mm
, vma
, address
,
2007 pte
, pmd
, write_access
);
2008 return do_no_page(mm
, vma
, address
,
2009 pte
, pmd
, write_access
);
2011 if (pte_file(entry
))
2012 return do_file_page(mm
, vma
, address
,
2013 pte
, pmd
, write_access
, entry
);
2014 return do_swap_page(mm
, vma
, address
,
2015 pte
, pmd
, write_access
, entry
);
2019 if (!pte_write(entry
))
2020 return do_wp_page(mm
, vma
, address
, pte
, pmd
, entry
);
2021 entry
= pte_mkdirty(entry
);
2023 entry
= pte_mkyoung(entry
);
2024 ptep_set_access_flags(vma
, address
, pte
, entry
, write_access
);
2025 update_mmu_cache(vma
, address
, entry
);
2026 lazy_mmu_prot_update(entry
);
2028 spin_unlock(&mm
->page_table_lock
);
2029 return VM_FAULT_MINOR
;
2033 * By the time we get here, we already hold the mm semaphore
2035 int __handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2036 unsigned long address
, int write_access
)
2043 __set_current_state(TASK_RUNNING
);
2045 inc_page_state(pgfault
);
2047 if (unlikely(is_vm_hugetlb_page(vma
)))
2048 return hugetlb_fault(mm
, vma
, address
, write_access
);
2051 * We need the page table lock to synchronize with kswapd
2052 * and the SMP-safe atomic PTE updates.
2054 pgd
= pgd_offset(mm
, address
);
2055 spin_lock(&mm
->page_table_lock
);
2057 pud
= pud_alloc(mm
, pgd
, address
);
2061 pmd
= pmd_alloc(mm
, pud
, address
);
2065 pte
= pte_alloc_map(mm
, pmd
, address
);
2069 return handle_pte_fault(mm
, vma
, address
, pte
, pmd
, write_access
);
2072 spin_unlock(&mm
->page_table_lock
);
2073 return VM_FAULT_OOM
;
2076 #ifndef __PAGETABLE_PUD_FOLDED
2078 * Allocate page upper directory.
2080 * We've already handled the fast-path in-line, and we own the
2083 pud_t fastcall
*__pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2087 spin_unlock(&mm
->page_table_lock
);
2088 new = pud_alloc_one(mm
, address
);
2089 spin_lock(&mm
->page_table_lock
);
2094 * Because we dropped the lock, we should re-check the
2095 * entry, as somebody else could have populated it..
2097 if (pgd_present(*pgd
)) {
2101 pgd_populate(mm
, pgd
, new);
2103 return pud_offset(pgd
, address
);
2105 #endif /* __PAGETABLE_PUD_FOLDED */
2107 #ifndef __PAGETABLE_PMD_FOLDED
2109 * Allocate page middle directory.
2111 * We've already handled the fast-path in-line, and we own the
2114 pmd_t fastcall
*__pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2118 spin_unlock(&mm
->page_table_lock
);
2119 new = pmd_alloc_one(mm
, address
);
2120 spin_lock(&mm
->page_table_lock
);
2125 * Because we dropped the lock, we should re-check the
2126 * entry, as somebody else could have populated it..
2128 #ifndef __ARCH_HAS_4LEVEL_HACK
2129 if (pud_present(*pud
)) {
2133 pud_populate(mm
, pud
, new);
2135 if (pgd_present(*pud
)) {
2139 pgd_populate(mm
, pud
, new);
2140 #endif /* __ARCH_HAS_4LEVEL_HACK */
2143 return pmd_offset(pud
, address
);
2145 #endif /* __PAGETABLE_PMD_FOLDED */
2147 int make_pages_present(unsigned long addr
, unsigned long end
)
2149 int ret
, len
, write
;
2150 struct vm_area_struct
* vma
;
2152 vma
= find_vma(current
->mm
, addr
);
2155 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
2158 if (end
> vma
->vm_end
)
2160 len
= (end
+PAGE_SIZE
-1)/PAGE_SIZE
-addr
/PAGE_SIZE
;
2161 ret
= get_user_pages(current
, current
->mm
, addr
,
2162 len
, write
, 0, NULL
, NULL
);
2165 return ret
== len
? 0 : -1;
2169 * Map a vmalloc()-space virtual address to the physical page.
2171 struct page
* vmalloc_to_page(void * vmalloc_addr
)
2173 unsigned long addr
= (unsigned long) vmalloc_addr
;
2174 struct page
*page
= NULL
;
2175 pgd_t
*pgd
= pgd_offset_k(addr
);
2180 if (!pgd_none(*pgd
)) {
2181 pud
= pud_offset(pgd
, addr
);
2182 if (!pud_none(*pud
)) {
2183 pmd
= pmd_offset(pud
, addr
);
2184 if (!pmd_none(*pmd
)) {
2185 ptep
= pte_offset_map(pmd
, addr
);
2187 if (pte_present(pte
))
2188 page
= pte_page(pte
);
2196 EXPORT_SYMBOL(vmalloc_to_page
);
2199 * Map a vmalloc()-space virtual address to the physical page frame number.
2201 unsigned long vmalloc_to_pfn(void * vmalloc_addr
)
2203 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
2206 EXPORT_SYMBOL(vmalloc_to_pfn
);
2209 * update_mem_hiwater
2210 * - update per process rss and vm high water data
2212 void update_mem_hiwater(struct task_struct
*tsk
)
2215 unsigned long rss
= get_mm_rss(tsk
->mm
);
2217 if (tsk
->mm
->hiwater_rss
< rss
)
2218 tsk
->mm
->hiwater_rss
= rss
;
2219 if (tsk
->mm
->hiwater_vm
< tsk
->mm
->total_vm
)
2220 tsk
->mm
->hiwater_vm
= tsk
->mm
->total_vm
;
2224 #if !defined(__HAVE_ARCH_GATE_AREA)
2226 #if defined(AT_SYSINFO_EHDR)
2227 static struct vm_area_struct gate_vma
;
2229 static int __init
gate_vma_init(void)
2231 gate_vma
.vm_mm
= NULL
;
2232 gate_vma
.vm_start
= FIXADDR_USER_START
;
2233 gate_vma
.vm_end
= FIXADDR_USER_END
;
2234 gate_vma
.vm_page_prot
= PAGE_READONLY
;
2235 gate_vma
.vm_flags
= 0;
2238 __initcall(gate_vma_init
);
2241 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
2243 #ifdef AT_SYSINFO_EHDR
2250 int in_gate_area_no_task(unsigned long addr
)
2252 #ifdef AT_SYSINFO_EHDR
2253 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
2259 #endif /* __HAVE_ARCH_GATE_AREA */