4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
41 #include <linux/kernel_stat.h>
43 #include <linux/hugetlb.h>
44 #include <linux/mman.h>
45 #include <linux/swap.h>
46 #include <linux/highmem.h>
47 #include <linux/pagemap.h>
48 #include <linux/rmap.h>
49 #include <linux/module.h>
50 #include <linux/init.h>
52 #include <asm/pgalloc.h>
53 #include <asm/uaccess.h>
55 #include <asm/tlbflush.h>
56 #include <asm/pgtable.h>
58 #include <linux/swapops.h>
59 #include <linux/elf.h>
61 #ifndef CONFIG_NEED_MULTIPLE_NODES
62 /* use the per-pgdat data instead for discontigmem - mbligh */
63 unsigned long max_mapnr
;
66 EXPORT_SYMBOL(max_mapnr
);
67 EXPORT_SYMBOL(mem_map
);
70 unsigned long num_physpages
;
72 * A number of key systems in x86 including ioremap() rely on the assumption
73 * that high_memory defines the upper bound on direct map memory, then end
74 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
75 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
79 unsigned long vmalloc_earlyreserve
;
81 EXPORT_SYMBOL(num_physpages
);
82 EXPORT_SYMBOL(high_memory
);
83 EXPORT_SYMBOL(vmalloc_earlyreserve
);
85 int randomize_va_space __read_mostly
= 1;
87 static int __init
disable_randmaps(char *s
)
89 randomize_va_space
= 0;
92 __setup("norandmaps", disable_randmaps
);
96 * If a p?d_bad entry is found while walking page tables, report
97 * the error, before resetting entry to p?d_none. Usually (but
98 * very seldom) called out from the p?d_none_or_clear_bad macros.
101 void pgd_clear_bad(pgd_t
*pgd
)
107 void pud_clear_bad(pud_t
*pud
)
113 void pmd_clear_bad(pmd_t
*pmd
)
120 * Note: this doesn't free the actual pages themselves. That
121 * has been handled earlier when unmapping all the memory regions.
123 static void free_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
)
125 struct page
*page
= pmd_page(*pmd
);
127 pte_lock_deinit(page
);
128 pte_free_tlb(tlb
, page
);
129 dec_page_state(nr_page_table_pages
);
133 static inline void free_pmd_range(struct mmu_gather
*tlb
, pud_t
*pud
,
134 unsigned long addr
, unsigned long end
,
135 unsigned long floor
, unsigned long ceiling
)
142 pmd
= pmd_offset(pud
, addr
);
144 next
= pmd_addr_end(addr
, end
);
145 if (pmd_none_or_clear_bad(pmd
))
147 free_pte_range(tlb
, pmd
);
148 } while (pmd
++, addr
= next
, addr
!= end
);
158 if (end
- 1 > ceiling
- 1)
161 pmd
= pmd_offset(pud
, start
);
163 pmd_free_tlb(tlb
, pmd
);
166 static inline void free_pud_range(struct mmu_gather
*tlb
, pgd_t
*pgd
,
167 unsigned long addr
, unsigned long end
,
168 unsigned long floor
, unsigned long ceiling
)
175 pud
= pud_offset(pgd
, addr
);
177 next
= pud_addr_end(addr
, end
);
178 if (pud_none_or_clear_bad(pud
))
180 free_pmd_range(tlb
, pud
, addr
, next
, floor
, ceiling
);
181 } while (pud
++, addr
= next
, addr
!= end
);
187 ceiling
&= PGDIR_MASK
;
191 if (end
- 1 > ceiling
- 1)
194 pud
= pud_offset(pgd
, start
);
196 pud_free_tlb(tlb
, pud
);
200 * This function frees user-level page tables of a process.
202 * Must be called with pagetable lock held.
204 void free_pgd_range(struct mmu_gather
**tlb
,
205 unsigned long addr
, unsigned long end
,
206 unsigned long floor
, unsigned long ceiling
)
213 * The next few lines have given us lots of grief...
215 * Why are we testing PMD* at this top level? Because often
216 * there will be no work to do at all, and we'd prefer not to
217 * go all the way down to the bottom just to discover that.
219 * Why all these "- 1"s? Because 0 represents both the bottom
220 * of the address space and the top of it (using -1 for the
221 * top wouldn't help much: the masks would do the wrong thing).
222 * The rule is that addr 0 and floor 0 refer to the bottom of
223 * the address space, but end 0 and ceiling 0 refer to the top
224 * Comparisons need to use "end - 1" and "ceiling - 1" (though
225 * that end 0 case should be mythical).
227 * Wherever addr is brought up or ceiling brought down, we must
228 * be careful to reject "the opposite 0" before it confuses the
229 * subsequent tests. But what about where end is brought down
230 * by PMD_SIZE below? no, end can't go down to 0 there.
232 * Whereas we round start (addr) and ceiling down, by different
233 * masks at different levels, in order to test whether a table
234 * now has no other vmas using it, so can be freed, we don't
235 * bother to round floor or end up - the tests don't need that.
249 if (end
- 1 > ceiling
- 1)
255 pgd
= pgd_offset((*tlb
)->mm
, addr
);
257 next
= pgd_addr_end(addr
, end
);
258 if (pgd_none_or_clear_bad(pgd
))
260 free_pud_range(*tlb
, pgd
, addr
, next
, floor
, ceiling
);
261 } while (pgd
++, addr
= next
, addr
!= end
);
264 flush_tlb_pgtables((*tlb
)->mm
, start
, end
);
267 void free_pgtables(struct mmu_gather
**tlb
, struct vm_area_struct
*vma
,
268 unsigned long floor
, unsigned long ceiling
)
271 struct vm_area_struct
*next
= vma
->vm_next
;
272 unsigned long addr
= vma
->vm_start
;
275 * Hide vma from rmap and vmtruncate before freeing pgtables
277 anon_vma_unlink(vma
);
278 unlink_file_vma(vma
);
280 if (is_hugepage_only_range(vma
->vm_mm
, addr
, HPAGE_SIZE
)) {
281 hugetlb_free_pgd_range(tlb
, addr
, vma
->vm_end
,
282 floor
, next
? next
->vm_start
: ceiling
);
285 * Optimization: gather nearby vmas into one call down
287 while (next
&& next
->vm_start
<= vma
->vm_end
+ PMD_SIZE
288 && !is_hugepage_only_range(vma
->vm_mm
, next
->vm_start
,
292 anon_vma_unlink(vma
);
293 unlink_file_vma(vma
);
295 free_pgd_range(tlb
, addr
, vma
->vm_end
,
296 floor
, next
? next
->vm_start
: ceiling
);
302 int __pte_alloc(struct mm_struct
*mm
, pmd_t
*pmd
, unsigned long address
)
304 struct page
*new = pte_alloc_one(mm
, address
);
309 spin_lock(&mm
->page_table_lock
);
310 if (pmd_present(*pmd
)) { /* Another has populated it */
311 pte_lock_deinit(new);
315 inc_page_state(nr_page_table_pages
);
316 pmd_populate(mm
, pmd
, new);
318 spin_unlock(&mm
->page_table_lock
);
322 int __pte_alloc_kernel(pmd_t
*pmd
, unsigned long address
)
324 pte_t
*new = pte_alloc_one_kernel(&init_mm
, address
);
328 spin_lock(&init_mm
.page_table_lock
);
329 if (pmd_present(*pmd
)) /* Another has populated it */
330 pte_free_kernel(new);
332 pmd_populate_kernel(&init_mm
, pmd
, new);
333 spin_unlock(&init_mm
.page_table_lock
);
337 static inline void add_mm_rss(struct mm_struct
*mm
, int file_rss
, int anon_rss
)
340 add_mm_counter(mm
, file_rss
, file_rss
);
342 add_mm_counter(mm
, anon_rss
, anon_rss
);
346 * This function is called to print an error when a bad pte
347 * is found. For example, we might have a PFN-mapped pte in
348 * a region that doesn't allow it.
350 * The calling function must still handle the error.
352 void print_bad_pte(struct vm_area_struct
*vma
, pte_t pte
, unsigned long vaddr
)
354 printk(KERN_ERR
"Bad pte = %08llx, process = %s, "
355 "vm_flags = %lx, vaddr = %lx\n",
356 (long long)pte_val(pte
),
357 (vma
->vm_mm
== current
->mm
? current
->comm
: "???"),
358 vma
->vm_flags
, vaddr
);
362 static inline int is_cow_mapping(unsigned int flags
)
364 return (flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
;
368 * This function gets the "struct page" associated with a pte.
370 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
371 * will have each page table entry just pointing to a raw page frame
372 * number, and as far as the VM layer is concerned, those do not have
373 * pages associated with them - even if the PFN might point to memory
374 * that otherwise is perfectly fine and has a "struct page".
376 * The way we recognize those mappings is through the rules set up
377 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
378 * and the vm_pgoff will point to the first PFN mapped: thus every
379 * page that is a raw mapping will always honor the rule
381 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
383 * and if that isn't true, the page has been COW'ed (in which case it
384 * _does_ have a "struct page" associated with it even if it is in a
387 struct page
*vm_normal_page(struct vm_area_struct
*vma
, unsigned long addr
, pte_t pte
)
389 unsigned long pfn
= pte_pfn(pte
);
391 if (vma
->vm_flags
& VM_PFNMAP
) {
392 unsigned long off
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
393 if (pfn
== vma
->vm_pgoff
+ off
)
395 if (!is_cow_mapping(vma
->vm_flags
))
400 * Add some anal sanity checks for now. Eventually,
401 * we should just do "return pfn_to_page(pfn)", but
402 * in the meantime we check that we get a valid pfn,
403 * and that the resulting page looks ok.
405 * Remove this test eventually!
407 if (unlikely(!pfn_valid(pfn
))) {
408 print_bad_pte(vma
, pte
, addr
);
413 * NOTE! We still have PageReserved() pages in the page
416 * The PAGE_ZERO() pages and various VDSO mappings can
417 * cause them to exist.
419 return pfn_to_page(pfn
);
423 * copy one vm_area from one task to the other. Assumes the page tables
424 * already present in the new task to be cleared in the whole range
425 * covered by this vma.
429 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
430 pte_t
*dst_pte
, pte_t
*src_pte
, struct vm_area_struct
*vma
,
431 unsigned long addr
, int *rss
)
433 unsigned long vm_flags
= vma
->vm_flags
;
434 pte_t pte
= *src_pte
;
437 /* pte contains position in swap or file, so copy. */
438 if (unlikely(!pte_present(pte
))) {
439 if (!pte_file(pte
)) {
440 swap_duplicate(pte_to_swp_entry(pte
));
441 /* make sure dst_mm is on swapoff's mmlist. */
442 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
443 spin_lock(&mmlist_lock
);
444 if (list_empty(&dst_mm
->mmlist
))
445 list_add(&dst_mm
->mmlist
,
447 spin_unlock(&mmlist_lock
);
454 * If it's a COW mapping, write protect it both
455 * in the parent and the child
457 if (is_cow_mapping(vm_flags
)) {
458 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
463 * If it's a shared mapping, mark it clean in
466 if (vm_flags
& VM_SHARED
)
467 pte
= pte_mkclean(pte
);
468 pte
= pte_mkold(pte
);
470 page
= vm_normal_page(vma
, addr
, pte
);
474 rss
[!!PageAnon(page
)]++;
478 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
481 static int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
482 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
483 unsigned long addr
, unsigned long end
)
485 pte_t
*src_pte
, *dst_pte
;
486 spinlock_t
*src_ptl
, *dst_ptl
;
492 dst_pte
= pte_alloc_map_lock(dst_mm
, dst_pmd
, addr
, &dst_ptl
);
495 src_pte
= pte_offset_map_nested(src_pmd
, addr
);
496 src_ptl
= pte_lockptr(src_mm
, src_pmd
);
501 * We are holding two locks at this point - either of them
502 * could generate latencies in another task on another CPU.
504 if (progress
>= 32) {
506 if (need_resched() ||
507 need_lockbreak(src_ptl
) ||
508 need_lockbreak(dst_ptl
))
511 if (pte_none(*src_pte
)) {
515 copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
, vma
, addr
, rss
);
517 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
519 spin_unlock(src_ptl
);
520 pte_unmap_nested(src_pte
- 1);
521 add_mm_rss(dst_mm
, rss
[0], rss
[1]);
522 pte_unmap_unlock(dst_pte
- 1, dst_ptl
);
529 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
530 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
531 unsigned long addr
, unsigned long end
)
533 pmd_t
*src_pmd
, *dst_pmd
;
536 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
539 src_pmd
= pmd_offset(src_pud
, addr
);
541 next
= pmd_addr_end(addr
, end
);
542 if (pmd_none_or_clear_bad(src_pmd
))
544 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
547 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
551 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
552 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
553 unsigned long addr
, unsigned long end
)
555 pud_t
*src_pud
, *dst_pud
;
558 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
561 src_pud
= pud_offset(src_pgd
, addr
);
563 next
= pud_addr_end(addr
, end
);
564 if (pud_none_or_clear_bad(src_pud
))
566 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
569 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
573 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
574 struct vm_area_struct
*vma
)
576 pgd_t
*src_pgd
, *dst_pgd
;
578 unsigned long addr
= vma
->vm_start
;
579 unsigned long end
= vma
->vm_end
;
582 * Don't copy ptes where a page fault will fill them correctly.
583 * Fork becomes much lighter when there are big shared or private
584 * readonly mappings. The tradeoff is that copy_page_range is more
585 * efficient than faulting.
587 if (!(vma
->vm_flags
& (VM_HUGETLB
|VM_NONLINEAR
|VM_PFNMAP
|VM_INSERTPAGE
))) {
592 if (is_vm_hugetlb_page(vma
))
593 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
595 dst_pgd
= pgd_offset(dst_mm
, addr
);
596 src_pgd
= pgd_offset(src_mm
, addr
);
598 next
= pgd_addr_end(addr
, end
);
599 if (pgd_none_or_clear_bad(src_pgd
))
601 if (copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
604 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
608 static unsigned long zap_pte_range(struct mmu_gather
*tlb
,
609 struct vm_area_struct
*vma
, pmd_t
*pmd
,
610 unsigned long addr
, unsigned long end
,
611 long *zap_work
, struct zap_details
*details
)
613 struct mm_struct
*mm
= tlb
->mm
;
619 pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
622 if (pte_none(ptent
)) {
626 if (pte_present(ptent
)) {
629 (*zap_work
) -= PAGE_SIZE
;
631 page
= vm_normal_page(vma
, addr
, ptent
);
632 if (unlikely(details
) && page
) {
634 * unmap_shared_mapping_pages() wants to
635 * invalidate cache without truncating:
636 * unmap shared but keep private pages.
638 if (details
->check_mapping
&&
639 details
->check_mapping
!= page
->mapping
)
642 * Each page->index must be checked when
643 * invalidating or truncating nonlinear.
645 if (details
->nonlinear_vma
&&
646 (page
->index
< details
->first_index
||
647 page
->index
> details
->last_index
))
650 ptent
= ptep_get_and_clear_full(mm
, addr
, pte
,
652 tlb_remove_tlb_entry(tlb
, pte
, addr
);
655 if (unlikely(details
) && details
->nonlinear_vma
656 && linear_page_index(details
->nonlinear_vma
,
657 addr
) != page
->index
)
658 set_pte_at(mm
, addr
, pte
,
659 pgoff_to_pte(page
->index
));
663 if (pte_dirty(ptent
))
664 set_page_dirty(page
);
665 if (pte_young(ptent
))
666 mark_page_accessed(page
);
669 page_remove_rmap(page
);
670 tlb_remove_page(tlb
, page
);
674 * If details->check_mapping, we leave swap entries;
675 * if details->nonlinear_vma, we leave file entries.
677 if (unlikely(details
))
679 if (!pte_file(ptent
))
680 free_swap_and_cache(pte_to_swp_entry(ptent
));
681 pte_clear_full(mm
, addr
, pte
, tlb
->fullmm
);
682 } while (pte
++, addr
+= PAGE_SIZE
, (addr
!= end
&& *zap_work
> 0));
684 add_mm_rss(mm
, file_rss
, anon_rss
);
685 pte_unmap_unlock(pte
- 1, ptl
);
690 static inline unsigned long zap_pmd_range(struct mmu_gather
*tlb
,
691 struct vm_area_struct
*vma
, pud_t
*pud
,
692 unsigned long addr
, unsigned long end
,
693 long *zap_work
, struct zap_details
*details
)
698 pmd
= pmd_offset(pud
, addr
);
700 next
= pmd_addr_end(addr
, end
);
701 if (pmd_none_or_clear_bad(pmd
)) {
705 next
= zap_pte_range(tlb
, vma
, pmd
, addr
, next
,
707 } while (pmd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
712 static inline unsigned long zap_pud_range(struct mmu_gather
*tlb
,
713 struct vm_area_struct
*vma
, pgd_t
*pgd
,
714 unsigned long addr
, unsigned long end
,
715 long *zap_work
, struct zap_details
*details
)
720 pud
= pud_offset(pgd
, addr
);
722 next
= pud_addr_end(addr
, end
);
723 if (pud_none_or_clear_bad(pud
)) {
727 next
= zap_pmd_range(tlb
, vma
, pud
, addr
, next
,
729 } while (pud
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
734 static unsigned long unmap_page_range(struct mmu_gather
*tlb
,
735 struct vm_area_struct
*vma
,
736 unsigned long addr
, unsigned long end
,
737 long *zap_work
, struct zap_details
*details
)
742 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
746 tlb_start_vma(tlb
, vma
);
747 pgd
= pgd_offset(vma
->vm_mm
, addr
);
749 next
= pgd_addr_end(addr
, end
);
750 if (pgd_none_or_clear_bad(pgd
)) {
754 next
= zap_pud_range(tlb
, vma
, pgd
, addr
, next
,
756 } while (pgd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
757 tlb_end_vma(tlb
, vma
);
762 #ifdef CONFIG_PREEMPT
763 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
765 /* No preempt: go for improved straight-line efficiency */
766 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
770 * unmap_vmas - unmap a range of memory covered by a list of vma's
771 * @tlbp: address of the caller's struct mmu_gather
772 * @vma: the starting vma
773 * @start_addr: virtual address at which to start unmapping
774 * @end_addr: virtual address at which to end unmapping
775 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
776 * @details: details of nonlinear truncation or shared cache invalidation
778 * Returns the end address of the unmapping (restart addr if interrupted).
780 * Unmap all pages in the vma list.
782 * We aim to not hold locks for too long (for scheduling latency reasons).
783 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
784 * return the ending mmu_gather to the caller.
786 * Only addresses between `start' and `end' will be unmapped.
788 * The VMA list must be sorted in ascending virtual address order.
790 * unmap_vmas() assumes that the caller will flush the whole unmapped address
791 * range after unmap_vmas() returns. So the only responsibility here is to
792 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
793 * drops the lock and schedules.
795 unsigned long unmap_vmas(struct mmu_gather
**tlbp
,
796 struct vm_area_struct
*vma
, unsigned long start_addr
,
797 unsigned long end_addr
, unsigned long *nr_accounted
,
798 struct zap_details
*details
)
800 long zap_work
= ZAP_BLOCK_SIZE
;
801 unsigned long tlb_start
= 0; /* For tlb_finish_mmu */
802 int tlb_start_valid
= 0;
803 unsigned long start
= start_addr
;
804 spinlock_t
*i_mmap_lock
= details
? details
->i_mmap_lock
: NULL
;
805 int fullmm
= (*tlbp
)->fullmm
;
807 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
) {
810 start
= max(vma
->vm_start
, start_addr
);
811 if (start
>= vma
->vm_end
)
813 end
= min(vma
->vm_end
, end_addr
);
814 if (end
<= vma
->vm_start
)
817 if (vma
->vm_flags
& VM_ACCOUNT
)
818 *nr_accounted
+= (end
- start
) >> PAGE_SHIFT
;
820 while (start
!= end
) {
821 if (!tlb_start_valid
) {
826 if (unlikely(is_vm_hugetlb_page(vma
))) {
827 unmap_hugepage_range(vma
, start
, end
);
828 zap_work
-= (end
- start
) /
829 (HPAGE_SIZE
/ PAGE_SIZE
);
832 start
= unmap_page_range(*tlbp
, vma
,
833 start
, end
, &zap_work
, details
);
836 BUG_ON(start
!= end
);
840 tlb_finish_mmu(*tlbp
, tlb_start
, start
);
842 if (need_resched() ||
843 (i_mmap_lock
&& need_lockbreak(i_mmap_lock
))) {
851 *tlbp
= tlb_gather_mmu(vma
->vm_mm
, fullmm
);
853 zap_work
= ZAP_BLOCK_SIZE
;
857 return start
; /* which is now the end (or restart) address */
861 * zap_page_range - remove user pages in a given range
862 * @vma: vm_area_struct holding the applicable pages
863 * @address: starting address of pages to zap
864 * @size: number of bytes to zap
865 * @details: details of nonlinear truncation or shared cache invalidation
867 unsigned long zap_page_range(struct vm_area_struct
*vma
, unsigned long address
,
868 unsigned long size
, struct zap_details
*details
)
870 struct mm_struct
*mm
= vma
->vm_mm
;
871 struct mmu_gather
*tlb
;
872 unsigned long end
= address
+ size
;
873 unsigned long nr_accounted
= 0;
876 tlb
= tlb_gather_mmu(mm
, 0);
877 update_hiwater_rss(mm
);
878 end
= unmap_vmas(&tlb
, vma
, address
, end
, &nr_accounted
, details
);
880 tlb_finish_mmu(tlb
, address
, end
);
885 * Do a quick page-table lookup for a single page.
887 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
896 struct mm_struct
*mm
= vma
->vm_mm
;
898 page
= follow_huge_addr(mm
, address
, flags
& FOLL_WRITE
);
900 BUG_ON(flags
& FOLL_GET
);
905 pgd
= pgd_offset(mm
, address
);
906 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
909 pud
= pud_offset(pgd
, address
);
910 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
913 pmd
= pmd_offset(pud
, address
);
914 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
917 if (pmd_huge(*pmd
)) {
918 BUG_ON(flags
& FOLL_GET
);
919 page
= follow_huge_pmd(mm
, address
, pmd
, flags
& FOLL_WRITE
);
923 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
928 if (!pte_present(pte
))
930 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
932 page
= vm_normal_page(vma
, address
, pte
);
936 if (flags
& FOLL_GET
)
938 if (flags
& FOLL_TOUCH
) {
939 if ((flags
& FOLL_WRITE
) &&
940 !pte_dirty(pte
) && !PageDirty(page
))
941 set_page_dirty(page
);
942 mark_page_accessed(page
);
945 pte_unmap_unlock(ptep
, ptl
);
951 * When core dumping an enormous anonymous area that nobody
952 * has touched so far, we don't want to allocate page tables.
954 if (flags
& FOLL_ANON
) {
955 page
= ZERO_PAGE(address
);
956 if (flags
& FOLL_GET
)
958 BUG_ON(flags
& FOLL_WRITE
);
963 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
964 unsigned long start
, int len
, int write
, int force
,
965 struct page
**pages
, struct vm_area_struct
**vmas
)
968 unsigned int vm_flags
;
971 * Require read or write permissions.
972 * If 'force' is set, we only require the "MAY" flags.
974 vm_flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
975 vm_flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
979 struct vm_area_struct
*vma
;
980 unsigned int foll_flags
;
982 vma
= find_extend_vma(mm
, start
);
983 if (!vma
&& in_gate_area(tsk
, start
)) {
984 unsigned long pg
= start
& PAGE_MASK
;
985 struct vm_area_struct
*gate_vma
= get_gate_vma(tsk
);
990 if (write
) /* user gate pages are read-only */
991 return i
? : -EFAULT
;
993 pgd
= pgd_offset_k(pg
);
995 pgd
= pgd_offset_gate(mm
, pg
);
996 BUG_ON(pgd_none(*pgd
));
997 pud
= pud_offset(pgd
, pg
);
998 BUG_ON(pud_none(*pud
));
999 pmd
= pmd_offset(pud
, pg
);
1001 return i
? : -EFAULT
;
1002 pte
= pte_offset_map(pmd
, pg
);
1003 if (pte_none(*pte
)) {
1005 return i
? : -EFAULT
;
1008 struct page
*page
= vm_normal_page(gate_vma
, start
, *pte
);
1022 if (!vma
|| (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
))
1023 || !(vm_flags
& vma
->vm_flags
))
1024 return i
? : -EFAULT
;
1026 if (is_vm_hugetlb_page(vma
)) {
1027 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
1032 foll_flags
= FOLL_TOUCH
;
1034 foll_flags
|= FOLL_GET
;
1035 if (!write
&& !(vma
->vm_flags
& VM_LOCKED
) &&
1036 (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
))
1037 foll_flags
|= FOLL_ANON
;
1043 foll_flags
|= FOLL_WRITE
;
1046 while (!(page
= follow_page(vma
, start
, foll_flags
))) {
1048 ret
= __handle_mm_fault(mm
, vma
, start
,
1049 foll_flags
& FOLL_WRITE
);
1051 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1052 * broken COW when necessary, even if maybe_mkwrite
1053 * decided not to set pte_write. We can thus safely do
1054 * subsequent page lookups as if they were reads.
1056 if (ret
& VM_FAULT_WRITE
)
1057 foll_flags
&= ~FOLL_WRITE
;
1059 switch (ret
& ~VM_FAULT_WRITE
) {
1060 case VM_FAULT_MINOR
:
1063 case VM_FAULT_MAJOR
:
1066 case VM_FAULT_SIGBUS
:
1067 return i
? i
: -EFAULT
;
1069 return i
? i
: -ENOMEM
;
1076 flush_dcache_page(page
);
1083 } while (len
&& start
< vma
->vm_end
);
1087 EXPORT_SYMBOL(get_user_pages
);
1089 static int zeromap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1090 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1095 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1099 struct page
*page
= ZERO_PAGE(addr
);
1100 pte_t zero_pte
= pte_wrprotect(mk_pte(page
, prot
));
1101 page_cache_get(page
);
1102 page_add_file_rmap(page
);
1103 inc_mm_counter(mm
, file_rss
);
1104 BUG_ON(!pte_none(*pte
));
1105 set_pte_at(mm
, addr
, pte
, zero_pte
);
1106 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1107 pte_unmap_unlock(pte
- 1, ptl
);
1111 static inline int zeromap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1112 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1117 pmd
= pmd_alloc(mm
, pud
, addr
);
1121 next
= pmd_addr_end(addr
, end
);
1122 if (zeromap_pte_range(mm
, pmd
, addr
, next
, prot
))
1124 } while (pmd
++, addr
= next
, addr
!= end
);
1128 static inline int zeromap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1129 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1134 pud
= pud_alloc(mm
, pgd
, addr
);
1138 next
= pud_addr_end(addr
, end
);
1139 if (zeromap_pmd_range(mm
, pud
, addr
, next
, prot
))
1141 } while (pud
++, addr
= next
, addr
!= end
);
1145 int zeromap_page_range(struct vm_area_struct
*vma
,
1146 unsigned long addr
, unsigned long size
, pgprot_t prot
)
1150 unsigned long end
= addr
+ size
;
1151 struct mm_struct
*mm
= vma
->vm_mm
;
1154 BUG_ON(addr
>= end
);
1155 pgd
= pgd_offset(mm
, addr
);
1156 flush_cache_range(vma
, addr
, end
);
1158 next
= pgd_addr_end(addr
, end
);
1159 err
= zeromap_pud_range(mm
, pgd
, addr
, next
, prot
);
1162 } while (pgd
++, addr
= next
, addr
!= end
);
1166 pte_t
* fastcall
get_locked_pte(struct mm_struct
*mm
, unsigned long addr
, spinlock_t
**ptl
)
1168 pgd_t
* pgd
= pgd_offset(mm
, addr
);
1169 pud_t
* pud
= pud_alloc(mm
, pgd
, addr
);
1171 pmd_t
* pmd
= pmd_alloc(mm
, pud
, addr
);
1173 return pte_alloc_map_lock(mm
, pmd
, addr
, ptl
);
1179 * This is the old fallback for page remapping.
1181 * For historical reasons, it only allows reserved pages. Only
1182 * old drivers should use this, and they needed to mark their
1183 * pages reserved for the old functions anyway.
1185 static int insert_page(struct mm_struct
*mm
, unsigned long addr
, struct page
*page
, pgprot_t prot
)
1195 flush_dcache_page(page
);
1196 pte
= get_locked_pte(mm
, addr
, &ptl
);
1200 if (!pte_none(*pte
))
1203 /* Ok, finally just insert the thing.. */
1205 inc_mm_counter(mm
, file_rss
);
1206 page_add_file_rmap(page
);
1207 set_pte_at(mm
, addr
, pte
, mk_pte(page
, prot
));
1211 pte_unmap_unlock(pte
, ptl
);
1217 * This allows drivers to insert individual pages they've allocated
1220 * The page has to be a nice clean _individual_ kernel allocation.
1221 * If you allocate a compound page, you need to have marked it as
1222 * such (__GFP_COMP), or manually just split the page up yourself
1223 * (which is mainly an issue of doing "set_page_count(page, 1)" for
1224 * each sub-page, and then freeing them one by one when you free
1225 * them rather than freeing it as a compound page).
1227 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1228 * took an arbitrary page protection parameter. This doesn't allow
1229 * that. Your vma protection will have to be set up correctly, which
1230 * means that if you want a shared writable mapping, you'd better
1231 * ask for a shared writable mapping!
1233 * The page does not need to be reserved.
1235 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
, struct page
*page
)
1237 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
1239 if (!page_count(page
))
1241 vma
->vm_flags
|= VM_INSERTPAGE
;
1242 return insert_page(vma
->vm_mm
, addr
, page
, vma
->vm_page_prot
);
1244 EXPORT_SYMBOL(vm_insert_page
);
1247 * maps a range of physical memory into the requested pages. the old
1248 * mappings are removed. any references to nonexistent pages results
1249 * in null mappings (currently treated as "copy-on-access")
1251 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1252 unsigned long addr
, unsigned long end
,
1253 unsigned long pfn
, pgprot_t prot
)
1258 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1262 BUG_ON(!pte_none(*pte
));
1263 set_pte_at(mm
, addr
, pte
, pfn_pte(pfn
, prot
));
1265 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1266 pte_unmap_unlock(pte
- 1, ptl
);
1270 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1271 unsigned long addr
, unsigned long end
,
1272 unsigned long pfn
, pgprot_t prot
)
1277 pfn
-= addr
>> PAGE_SHIFT
;
1278 pmd
= pmd_alloc(mm
, pud
, addr
);
1282 next
= pmd_addr_end(addr
, end
);
1283 if (remap_pte_range(mm
, pmd
, addr
, next
,
1284 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1286 } while (pmd
++, addr
= next
, addr
!= end
);
1290 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1291 unsigned long addr
, unsigned long end
,
1292 unsigned long pfn
, pgprot_t prot
)
1297 pfn
-= addr
>> PAGE_SHIFT
;
1298 pud
= pud_alloc(mm
, pgd
, addr
);
1302 next
= pud_addr_end(addr
, end
);
1303 if (remap_pmd_range(mm
, pud
, addr
, next
,
1304 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1306 } while (pud
++, addr
= next
, addr
!= end
);
1310 /* Note: this is only safe if the mm semaphore is held when called. */
1311 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1312 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1316 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1317 struct mm_struct
*mm
= vma
->vm_mm
;
1321 * Physically remapped pages are special. Tell the
1322 * rest of the world about it:
1323 * VM_IO tells people not to look at these pages
1324 * (accesses can have side effects).
1325 * VM_RESERVED is specified all over the place, because
1326 * in 2.4 it kept swapout's vma scan off this vma; but
1327 * in 2.6 the LRU scan won't even find its pages, so this
1328 * flag means no more than count its pages in reserved_vm,
1329 * and omit it from core dump, even when VM_IO turned off.
1330 * VM_PFNMAP tells the core MM that the base pages are just
1331 * raw PFN mappings, and do not have a "struct page" associated
1334 * There's a horrible special case to handle copy-on-write
1335 * behaviour that some programs depend on. We mark the "original"
1336 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1338 if (is_cow_mapping(vma
->vm_flags
)) {
1339 if (addr
!= vma
->vm_start
|| end
!= vma
->vm_end
)
1341 vma
->vm_pgoff
= pfn
;
1344 vma
->vm_flags
|= VM_IO
| VM_RESERVED
| VM_PFNMAP
;
1346 BUG_ON(addr
>= end
);
1347 pfn
-= addr
>> PAGE_SHIFT
;
1348 pgd
= pgd_offset(mm
, addr
);
1349 flush_cache_range(vma
, addr
, end
);
1351 next
= pgd_addr_end(addr
, end
);
1352 err
= remap_pud_range(mm
, pgd
, addr
, next
,
1353 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
1356 } while (pgd
++, addr
= next
, addr
!= end
);
1359 EXPORT_SYMBOL(remap_pfn_range
);
1362 * handle_pte_fault chooses page fault handler according to an entry
1363 * which was read non-atomically. Before making any commitment, on
1364 * those architectures or configurations (e.g. i386 with PAE) which
1365 * might give a mix of unmatched parts, do_swap_page and do_file_page
1366 * must check under lock before unmapping the pte and proceeding
1367 * (but do_wp_page is only called after already making such a check;
1368 * and do_anonymous_page and do_no_page can safely check later on).
1370 static inline int pte_unmap_same(struct mm_struct
*mm
, pmd_t
*pmd
,
1371 pte_t
*page_table
, pte_t orig_pte
)
1374 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1375 if (sizeof(pte_t
) > sizeof(unsigned long)) {
1376 spinlock_t
*ptl
= pte_lockptr(mm
, pmd
);
1378 same
= pte_same(*page_table
, orig_pte
);
1382 pte_unmap(page_table
);
1387 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1388 * servicing faults for write access. In the normal case, do always want
1389 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1390 * that do not have writing enabled, when used by access_process_vm.
1392 static inline pte_t
maybe_mkwrite(pte_t pte
, struct vm_area_struct
*vma
)
1394 if (likely(vma
->vm_flags
& VM_WRITE
))
1395 pte
= pte_mkwrite(pte
);
1399 static inline void cow_user_page(struct page
*dst
, struct page
*src
, unsigned long va
)
1402 * If the source page was a PFN mapping, we don't have
1403 * a "struct page" for it. We do a best-effort copy by
1404 * just copying from the original user address. If that
1405 * fails, we just zero-fill it. Live with it.
1407 if (unlikely(!src
)) {
1408 void *kaddr
= kmap_atomic(dst
, KM_USER0
);
1409 void __user
*uaddr
= (void __user
*)(va
& PAGE_MASK
);
1412 * This really shouldn't fail, because the page is there
1413 * in the page tables. But it might just be unreadable,
1414 * in which case we just give up and fill the result with
1417 if (__copy_from_user_inatomic(kaddr
, uaddr
, PAGE_SIZE
))
1418 memset(kaddr
, 0, PAGE_SIZE
);
1419 kunmap_atomic(kaddr
, KM_USER0
);
1423 copy_user_highpage(dst
, src
, va
);
1427 * This routine handles present pages, when users try to write
1428 * to a shared page. It is done by copying the page to a new address
1429 * and decrementing the shared-page counter for the old page.
1431 * Note that this routine assumes that the protection checks have been
1432 * done by the caller (the low-level page fault routine in most cases).
1433 * Thus we can safely just mark it writable once we've done any necessary
1436 * We also mark the page dirty at this point even though the page will
1437 * change only once the write actually happens. This avoids a few races,
1438 * and potentially makes it more efficient.
1440 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1441 * but allow concurrent faults), with pte both mapped and locked.
1442 * We return with mmap_sem still held, but pte unmapped and unlocked.
1444 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1445 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1446 spinlock_t
*ptl
, pte_t orig_pte
)
1448 struct page
*old_page
, *new_page
;
1450 int ret
= VM_FAULT_MINOR
;
1452 old_page
= vm_normal_page(vma
, address
, orig_pte
);
1456 if (PageAnon(old_page
) && !TestSetPageLocked(old_page
)) {
1457 int reuse
= can_share_swap_page(old_page
);
1458 unlock_page(old_page
);
1460 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
1461 entry
= pte_mkyoung(orig_pte
);
1462 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1463 ptep_set_access_flags(vma
, address
, page_table
, entry
, 1);
1464 update_mmu_cache(vma
, address
, entry
);
1465 lazy_mmu_prot_update(entry
);
1466 ret
|= VM_FAULT_WRITE
;
1472 * Ok, we need to copy. Oh, well..
1474 page_cache_get(old_page
);
1476 pte_unmap_unlock(page_table
, ptl
);
1478 if (unlikely(anon_vma_prepare(vma
)))
1480 if (old_page
== ZERO_PAGE(address
)) {
1481 new_page
= alloc_zeroed_user_highpage(vma
, address
);
1485 new_page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1488 cow_user_page(new_page
, old_page
, address
);
1492 * Re-check the pte - we dropped the lock
1494 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1495 if (likely(pte_same(*page_table
, orig_pte
))) {
1497 page_remove_rmap(old_page
);
1498 if (!PageAnon(old_page
)) {
1499 dec_mm_counter(mm
, file_rss
);
1500 inc_mm_counter(mm
, anon_rss
);
1503 inc_mm_counter(mm
, anon_rss
);
1504 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
1505 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1506 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1507 ptep_establish(vma
, address
, page_table
, entry
);
1508 update_mmu_cache(vma
, address
, entry
);
1509 lazy_mmu_prot_update(entry
);
1510 lru_cache_add_active(new_page
);
1511 page_add_new_anon_rmap(new_page
, vma
, address
);
1513 /* Free the old page.. */
1514 new_page
= old_page
;
1515 ret
|= VM_FAULT_WRITE
;
1518 page_cache_release(new_page
);
1520 page_cache_release(old_page
);
1522 pte_unmap_unlock(page_table
, ptl
);
1526 page_cache_release(old_page
);
1527 return VM_FAULT_OOM
;
1531 * Helper functions for unmap_mapping_range().
1533 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1535 * We have to restart searching the prio_tree whenever we drop the lock,
1536 * since the iterator is only valid while the lock is held, and anyway
1537 * a later vma might be split and reinserted earlier while lock dropped.
1539 * The list of nonlinear vmas could be handled more efficiently, using
1540 * a placeholder, but handle it in the same way until a need is shown.
1541 * It is important to search the prio_tree before nonlinear list: a vma
1542 * may become nonlinear and be shifted from prio_tree to nonlinear list
1543 * while the lock is dropped; but never shifted from list to prio_tree.
1545 * In order to make forward progress despite restarting the search,
1546 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1547 * quickly skip it next time around. Since the prio_tree search only
1548 * shows us those vmas affected by unmapping the range in question, we
1549 * can't efficiently keep all vmas in step with mapping->truncate_count:
1550 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1551 * mapping->truncate_count and vma->vm_truncate_count are protected by
1554 * In order to make forward progress despite repeatedly restarting some
1555 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1556 * and restart from that address when we reach that vma again. It might
1557 * have been split or merged, shrunk or extended, but never shifted: so
1558 * restart_addr remains valid so long as it remains in the vma's range.
1559 * unmap_mapping_range forces truncate_count to leap over page-aligned
1560 * values so we can save vma's restart_addr in its truncate_count field.
1562 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1564 static void reset_vma_truncate_counts(struct address_space
*mapping
)
1566 struct vm_area_struct
*vma
;
1567 struct prio_tree_iter iter
;
1569 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, 0, ULONG_MAX
)
1570 vma
->vm_truncate_count
= 0;
1571 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
1572 vma
->vm_truncate_count
= 0;
1575 static int unmap_mapping_range_vma(struct vm_area_struct
*vma
,
1576 unsigned long start_addr
, unsigned long end_addr
,
1577 struct zap_details
*details
)
1579 unsigned long restart_addr
;
1583 restart_addr
= vma
->vm_truncate_count
;
1584 if (is_restart_addr(restart_addr
) && start_addr
< restart_addr
) {
1585 start_addr
= restart_addr
;
1586 if (start_addr
>= end_addr
) {
1587 /* Top of vma has been split off since last time */
1588 vma
->vm_truncate_count
= details
->truncate_count
;
1593 restart_addr
= zap_page_range(vma
, start_addr
,
1594 end_addr
- start_addr
, details
);
1595 need_break
= need_resched() ||
1596 need_lockbreak(details
->i_mmap_lock
);
1598 if (restart_addr
>= end_addr
) {
1599 /* We have now completed this vma: mark it so */
1600 vma
->vm_truncate_count
= details
->truncate_count
;
1604 /* Note restart_addr in vma's truncate_count field */
1605 vma
->vm_truncate_count
= restart_addr
;
1610 spin_unlock(details
->i_mmap_lock
);
1612 spin_lock(details
->i_mmap_lock
);
1616 static inline void unmap_mapping_range_tree(struct prio_tree_root
*root
,
1617 struct zap_details
*details
)
1619 struct vm_area_struct
*vma
;
1620 struct prio_tree_iter iter
;
1621 pgoff_t vba
, vea
, zba
, zea
;
1624 vma_prio_tree_foreach(vma
, &iter
, root
,
1625 details
->first_index
, details
->last_index
) {
1626 /* Skip quickly over those we have already dealt with */
1627 if (vma
->vm_truncate_count
== details
->truncate_count
)
1630 vba
= vma
->vm_pgoff
;
1631 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
1632 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1633 zba
= details
->first_index
;
1636 zea
= details
->last_index
;
1640 if (unmap_mapping_range_vma(vma
,
1641 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
1642 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
1648 static inline void unmap_mapping_range_list(struct list_head
*head
,
1649 struct zap_details
*details
)
1651 struct vm_area_struct
*vma
;
1654 * In nonlinear VMAs there is no correspondence between virtual address
1655 * offset and file offset. So we must perform an exhaustive search
1656 * across *all* the pages in each nonlinear VMA, not just the pages
1657 * whose virtual address lies outside the file truncation point.
1660 list_for_each_entry(vma
, head
, shared
.vm_set
.list
) {
1661 /* Skip quickly over those we have already dealt with */
1662 if (vma
->vm_truncate_count
== details
->truncate_count
)
1664 details
->nonlinear_vma
= vma
;
1665 if (unmap_mapping_range_vma(vma
, vma
->vm_start
,
1666 vma
->vm_end
, details
) < 0)
1672 * unmap_mapping_range - unmap the portion of all mmaps
1673 * in the specified address_space corresponding to the specified
1674 * page range in the underlying file.
1675 * @mapping: the address space containing mmaps to be unmapped.
1676 * @holebegin: byte in first page to unmap, relative to the start of
1677 * the underlying file. This will be rounded down to a PAGE_SIZE
1678 * boundary. Note that this is different from vmtruncate(), which
1679 * must keep the partial page. In contrast, we must get rid of
1681 * @holelen: size of prospective hole in bytes. This will be rounded
1682 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1684 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1685 * but 0 when invalidating pagecache, don't throw away private data.
1687 void unmap_mapping_range(struct address_space
*mapping
,
1688 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
1690 struct zap_details details
;
1691 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
1692 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1694 /* Check for overflow. */
1695 if (sizeof(holelen
) > sizeof(hlen
)) {
1697 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1698 if (holeend
& ~(long long)ULONG_MAX
)
1699 hlen
= ULONG_MAX
- hba
+ 1;
1702 details
.check_mapping
= even_cows
? NULL
: mapping
;
1703 details
.nonlinear_vma
= NULL
;
1704 details
.first_index
= hba
;
1705 details
.last_index
= hba
+ hlen
- 1;
1706 if (details
.last_index
< details
.first_index
)
1707 details
.last_index
= ULONG_MAX
;
1708 details
.i_mmap_lock
= &mapping
->i_mmap_lock
;
1710 spin_lock(&mapping
->i_mmap_lock
);
1712 /* serialize i_size write against truncate_count write */
1714 /* Protect against page faults, and endless unmapping loops */
1715 mapping
->truncate_count
++;
1717 * For archs where spin_lock has inclusive semantics like ia64
1718 * this smp_mb() will prevent to read pagetable contents
1719 * before the truncate_count increment is visible to
1723 if (unlikely(is_restart_addr(mapping
->truncate_count
))) {
1724 if (mapping
->truncate_count
== 0)
1725 reset_vma_truncate_counts(mapping
);
1726 mapping
->truncate_count
++;
1728 details
.truncate_count
= mapping
->truncate_count
;
1730 if (unlikely(!prio_tree_empty(&mapping
->i_mmap
)))
1731 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
1732 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
)))
1733 unmap_mapping_range_list(&mapping
->i_mmap_nonlinear
, &details
);
1734 spin_unlock(&mapping
->i_mmap_lock
);
1736 EXPORT_SYMBOL(unmap_mapping_range
);
1739 * Handle all mappings that got truncated by a "truncate()"
1742 * NOTE! We have to be ready to update the memory sharing
1743 * between the file and the memory map for a potential last
1744 * incomplete page. Ugly, but necessary.
1746 int vmtruncate(struct inode
* inode
, loff_t offset
)
1748 struct address_space
*mapping
= inode
->i_mapping
;
1749 unsigned long limit
;
1751 if (inode
->i_size
< offset
)
1754 * truncation of in-use swapfiles is disallowed - it would cause
1755 * subsequent swapout to scribble on the now-freed blocks.
1757 if (IS_SWAPFILE(inode
))
1759 i_size_write(inode
, offset
);
1760 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
1761 truncate_inode_pages(mapping
, offset
);
1765 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1766 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
1768 if (offset
> inode
->i_sb
->s_maxbytes
)
1770 i_size_write(inode
, offset
);
1773 if (inode
->i_op
&& inode
->i_op
->truncate
)
1774 inode
->i_op
->truncate(inode
);
1777 send_sig(SIGXFSZ
, current
, 0);
1783 EXPORT_SYMBOL(vmtruncate
);
1785 int vmtruncate_range(struct inode
*inode
, loff_t offset
, loff_t end
)
1787 struct address_space
*mapping
= inode
->i_mapping
;
1790 * If the underlying filesystem is not going to provide
1791 * a way to truncate a range of blocks (punch a hole) -
1792 * we should return failure right now.
1794 if (!inode
->i_op
|| !inode
->i_op
->truncate_range
)
1797 mutex_lock(&inode
->i_mutex
);
1798 down_write(&inode
->i_alloc_sem
);
1799 unmap_mapping_range(mapping
, offset
, (end
- offset
), 1);
1800 truncate_inode_pages_range(mapping
, offset
, end
);
1801 inode
->i_op
->truncate_range(inode
, offset
, end
);
1802 up_write(&inode
->i_alloc_sem
);
1803 mutex_unlock(&inode
->i_mutex
);
1807 EXPORT_SYMBOL(vmtruncate_range
);
1810 * Primitive swap readahead code. We simply read an aligned block of
1811 * (1 << page_cluster) entries in the swap area. This method is chosen
1812 * because it doesn't cost us any seek time. We also make sure to queue
1813 * the 'original' request together with the readahead ones...
1815 * This has been extended to use the NUMA policies from the mm triggering
1818 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1820 void swapin_readahead(swp_entry_t entry
, unsigned long addr
,struct vm_area_struct
*vma
)
1823 struct vm_area_struct
*next_vma
= vma
? vma
->vm_next
: NULL
;
1826 struct page
*new_page
;
1827 unsigned long offset
;
1830 * Get the number of handles we should do readahead io to.
1832 num
= valid_swaphandles(entry
, &offset
);
1833 for (i
= 0; i
< num
; offset
++, i
++) {
1834 /* Ok, do the async read-ahead now */
1835 new_page
= read_swap_cache_async(swp_entry(swp_type(entry
),
1836 offset
), vma
, addr
);
1839 page_cache_release(new_page
);
1842 * Find the next applicable VMA for the NUMA policy.
1848 if (addr
>= vma
->vm_end
) {
1850 next_vma
= vma
? vma
->vm_next
: NULL
;
1852 if (vma
&& addr
< vma
->vm_start
)
1855 if (next_vma
&& addr
>= next_vma
->vm_start
) {
1857 next_vma
= vma
->vm_next
;
1862 lru_add_drain(); /* Push any new pages onto the LRU now */
1866 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1867 * but allow concurrent faults), and pte mapped but not yet locked.
1868 * We return with mmap_sem still held, but pte unmapped and unlocked.
1870 static int do_swap_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1871 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1872 int write_access
, pte_t orig_pte
)
1878 int ret
= VM_FAULT_MINOR
;
1880 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
1883 entry
= pte_to_swp_entry(orig_pte
);
1885 page
= lookup_swap_cache(entry
);
1887 swapin_readahead(entry
, address
, vma
);
1888 page
= read_swap_cache_async(entry
, vma
, address
);
1891 * Back out if somebody else faulted in this pte
1892 * while we released the pte lock.
1894 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1895 if (likely(pte_same(*page_table
, orig_pte
)))
1900 /* Had to read the page from swap area: Major fault */
1901 ret
= VM_FAULT_MAJOR
;
1902 inc_page_state(pgmajfault
);
1906 mark_page_accessed(page
);
1908 if (!PageSwapCache(page
)) {
1909 /* Page migration has occured */
1911 page_cache_release(page
);
1916 * Back out if somebody else already faulted in this pte.
1918 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1919 if (unlikely(!pte_same(*page_table
, orig_pte
)))
1922 if (unlikely(!PageUptodate(page
))) {
1923 ret
= VM_FAULT_SIGBUS
;
1927 /* The page isn't present yet, go ahead with the fault. */
1929 inc_mm_counter(mm
, anon_rss
);
1930 pte
= mk_pte(page
, vma
->vm_page_prot
);
1931 if (write_access
&& can_share_swap_page(page
)) {
1932 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
1936 flush_icache_page(vma
, page
);
1937 set_pte_at(mm
, address
, page_table
, pte
);
1938 page_add_anon_rmap(page
, vma
, address
);
1942 remove_exclusive_swap_page(page
);
1946 if (do_wp_page(mm
, vma
, address
,
1947 page_table
, pmd
, ptl
, pte
) == VM_FAULT_OOM
)
1952 /* No need to invalidate - it was non-present before */
1953 update_mmu_cache(vma
, address
, pte
);
1954 lazy_mmu_prot_update(pte
);
1956 pte_unmap_unlock(page_table
, ptl
);
1960 pte_unmap_unlock(page_table
, ptl
);
1962 page_cache_release(page
);
1967 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1968 * but allow concurrent faults), and pte mapped but not yet locked.
1969 * We return with mmap_sem still held, but pte unmapped and unlocked.
1971 static int do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1972 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1980 /* Allocate our own private page. */
1981 pte_unmap(page_table
);
1983 if (unlikely(anon_vma_prepare(vma
)))
1985 page
= alloc_zeroed_user_highpage(vma
, address
);
1989 entry
= mk_pte(page
, vma
->vm_page_prot
);
1990 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1992 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1993 if (!pte_none(*page_table
))
1995 inc_mm_counter(mm
, anon_rss
);
1996 lru_cache_add_active(page
);
1997 page_add_new_anon_rmap(page
, vma
, address
);
1999 /* Map the ZERO_PAGE - vm_page_prot is readonly */
2000 page
= ZERO_PAGE(address
);
2001 page_cache_get(page
);
2002 entry
= mk_pte(page
, vma
->vm_page_prot
);
2004 ptl
= pte_lockptr(mm
, pmd
);
2006 if (!pte_none(*page_table
))
2008 inc_mm_counter(mm
, file_rss
);
2009 page_add_file_rmap(page
);
2012 set_pte_at(mm
, address
, page_table
, entry
);
2014 /* No need to invalidate - it was non-present before */
2015 update_mmu_cache(vma
, address
, entry
);
2016 lazy_mmu_prot_update(entry
);
2018 pte_unmap_unlock(page_table
, ptl
);
2019 return VM_FAULT_MINOR
;
2021 page_cache_release(page
);
2024 return VM_FAULT_OOM
;
2028 * do_no_page() tries to create a new page mapping. It aggressively
2029 * tries to share with existing pages, but makes a separate copy if
2030 * the "write_access" parameter is true in order to avoid the next
2033 * As this is called only for pages that do not currently exist, we
2034 * do not need to flush old virtual caches or the TLB.
2036 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2037 * but allow concurrent faults), and pte mapped but not yet locked.
2038 * We return with mmap_sem still held, but pte unmapped and unlocked.
2040 static int do_no_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2041 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2045 struct page
*new_page
;
2046 struct address_space
*mapping
= NULL
;
2048 unsigned int sequence
= 0;
2049 int ret
= VM_FAULT_MINOR
;
2052 pte_unmap(page_table
);
2053 BUG_ON(vma
->vm_flags
& VM_PFNMAP
);
2056 mapping
= vma
->vm_file
->f_mapping
;
2057 sequence
= mapping
->truncate_count
;
2058 smp_rmb(); /* serializes i_size against truncate_count */
2061 new_page
= vma
->vm_ops
->nopage(vma
, address
& PAGE_MASK
, &ret
);
2063 * No smp_rmb is needed here as long as there's a full
2064 * spin_lock/unlock sequence inside the ->nopage callback
2065 * (for the pagecache lookup) that acts as an implicit
2066 * smp_mb() and prevents the i_size read to happen
2067 * after the next truncate_count read.
2070 /* no page was available -- either SIGBUS or OOM */
2071 if (new_page
== NOPAGE_SIGBUS
)
2072 return VM_FAULT_SIGBUS
;
2073 if (new_page
== NOPAGE_OOM
)
2074 return VM_FAULT_OOM
;
2077 * Should we do an early C-O-W break?
2079 if (write_access
&& !(vma
->vm_flags
& VM_SHARED
)) {
2082 if (unlikely(anon_vma_prepare(vma
)))
2084 page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
2087 copy_user_highpage(page
, new_page
, address
);
2088 page_cache_release(new_page
);
2093 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2095 * For a file-backed vma, someone could have truncated or otherwise
2096 * invalidated this page. If unmap_mapping_range got called,
2097 * retry getting the page.
2099 if (mapping
&& unlikely(sequence
!= mapping
->truncate_count
)) {
2100 pte_unmap_unlock(page_table
, ptl
);
2101 page_cache_release(new_page
);
2103 sequence
= mapping
->truncate_count
;
2109 * This silly early PAGE_DIRTY setting removes a race
2110 * due to the bad i386 page protection. But it's valid
2111 * for other architectures too.
2113 * Note that if write_access is true, we either now have
2114 * an exclusive copy of the page, or this is a shared mapping,
2115 * so we can make it writable and dirty to avoid having to
2116 * handle that later.
2118 /* Only go through if we didn't race with anybody else... */
2119 if (pte_none(*page_table
)) {
2120 flush_icache_page(vma
, new_page
);
2121 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
2123 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2124 set_pte_at(mm
, address
, page_table
, entry
);
2126 inc_mm_counter(mm
, anon_rss
);
2127 lru_cache_add_active(new_page
);
2128 page_add_new_anon_rmap(new_page
, vma
, address
);
2130 inc_mm_counter(mm
, file_rss
);
2131 page_add_file_rmap(new_page
);
2134 /* One of our sibling threads was faster, back out. */
2135 page_cache_release(new_page
);
2139 /* no need to invalidate: a not-present page shouldn't be cached */
2140 update_mmu_cache(vma
, address
, entry
);
2141 lazy_mmu_prot_update(entry
);
2143 pte_unmap_unlock(page_table
, ptl
);
2146 page_cache_release(new_page
);
2147 return VM_FAULT_OOM
;
2151 * Fault of a previously existing named mapping. Repopulate the pte
2152 * from the encoded file_pte if possible. This enables swappable
2155 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2156 * but allow concurrent faults), and pte mapped but not yet locked.
2157 * We return with mmap_sem still held, but pte unmapped and unlocked.
2159 static int do_file_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2160 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2161 int write_access
, pte_t orig_pte
)
2166 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
2167 return VM_FAULT_MINOR
;
2169 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
))) {
2171 * Page table corrupted: show pte and kill process.
2173 print_bad_pte(vma
, orig_pte
, address
);
2174 return VM_FAULT_OOM
;
2176 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2178 pgoff
= pte_to_pgoff(orig_pte
);
2179 err
= vma
->vm_ops
->populate(vma
, address
& PAGE_MASK
, PAGE_SIZE
,
2180 vma
->vm_page_prot
, pgoff
, 0);
2182 return VM_FAULT_OOM
;
2184 return VM_FAULT_SIGBUS
;
2185 return VM_FAULT_MAJOR
;
2189 * These routines also need to handle stuff like marking pages dirty
2190 * and/or accessed for architectures that don't do it in hardware (most
2191 * RISC architectures). The early dirtying is also good on the i386.
2193 * There is also a hook called "update_mmu_cache()" that architectures
2194 * with external mmu caches can use to update those (ie the Sparc or
2195 * PowerPC hashed page tables that act as extended TLBs).
2197 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2198 * but allow concurrent faults), and pte mapped but not yet locked.
2199 * We return with mmap_sem still held, but pte unmapped and unlocked.
2201 static inline int handle_pte_fault(struct mm_struct
*mm
,
2202 struct vm_area_struct
*vma
, unsigned long address
,
2203 pte_t
*pte
, pmd_t
*pmd
, int write_access
)
2209 old_entry
= entry
= *pte
;
2210 if (!pte_present(entry
)) {
2211 if (pte_none(entry
)) {
2212 if (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
)
2213 return do_anonymous_page(mm
, vma
, address
,
2214 pte
, pmd
, write_access
);
2215 return do_no_page(mm
, vma
, address
,
2216 pte
, pmd
, write_access
);
2218 if (pte_file(entry
))
2219 return do_file_page(mm
, vma
, address
,
2220 pte
, pmd
, write_access
, entry
);
2221 return do_swap_page(mm
, vma
, address
,
2222 pte
, pmd
, write_access
, entry
);
2225 ptl
= pte_lockptr(mm
, pmd
);
2227 if (unlikely(!pte_same(*pte
, entry
)))
2230 if (!pte_write(entry
))
2231 return do_wp_page(mm
, vma
, address
,
2232 pte
, pmd
, ptl
, entry
);
2233 entry
= pte_mkdirty(entry
);
2235 entry
= pte_mkyoung(entry
);
2236 if (!pte_same(old_entry
, entry
)) {
2237 ptep_set_access_flags(vma
, address
, pte
, entry
, write_access
);
2238 update_mmu_cache(vma
, address
, entry
);
2239 lazy_mmu_prot_update(entry
);
2242 * This is needed only for protection faults but the arch code
2243 * is not yet telling us if this is a protection fault or not.
2244 * This still avoids useless tlb flushes for .text page faults
2248 flush_tlb_page(vma
, address
);
2251 pte_unmap_unlock(pte
, ptl
);
2252 return VM_FAULT_MINOR
;
2256 * By the time we get here, we already hold the mm semaphore
2258 int __handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2259 unsigned long address
, int write_access
)
2266 __set_current_state(TASK_RUNNING
);
2268 inc_page_state(pgfault
);
2270 if (unlikely(is_vm_hugetlb_page(vma
)))
2271 return hugetlb_fault(mm
, vma
, address
, write_access
);
2273 pgd
= pgd_offset(mm
, address
);
2274 pud
= pud_alloc(mm
, pgd
, address
);
2276 return VM_FAULT_OOM
;
2277 pmd
= pmd_alloc(mm
, pud
, address
);
2279 return VM_FAULT_OOM
;
2280 pte
= pte_alloc_map(mm
, pmd
, address
);
2282 return VM_FAULT_OOM
;
2284 return handle_pte_fault(mm
, vma
, address
, pte
, pmd
, write_access
);
2287 EXPORT_SYMBOL_GPL(__handle_mm_fault
);
2289 #ifndef __PAGETABLE_PUD_FOLDED
2291 * Allocate page upper directory.
2292 * We've already handled the fast-path in-line.
2294 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2296 pud_t
*new = pud_alloc_one(mm
, address
);
2300 spin_lock(&mm
->page_table_lock
);
2301 if (pgd_present(*pgd
)) /* Another has populated it */
2304 pgd_populate(mm
, pgd
, new);
2305 spin_unlock(&mm
->page_table_lock
);
2309 /* Workaround for gcc 2.96 */
2310 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2314 #endif /* __PAGETABLE_PUD_FOLDED */
2316 #ifndef __PAGETABLE_PMD_FOLDED
2318 * Allocate page middle directory.
2319 * We've already handled the fast-path in-line.
2321 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2323 pmd_t
*new = pmd_alloc_one(mm
, address
);
2327 spin_lock(&mm
->page_table_lock
);
2328 #ifndef __ARCH_HAS_4LEVEL_HACK
2329 if (pud_present(*pud
)) /* Another has populated it */
2332 pud_populate(mm
, pud
, new);
2334 if (pgd_present(*pud
)) /* Another has populated it */
2337 pgd_populate(mm
, pud
, new);
2338 #endif /* __ARCH_HAS_4LEVEL_HACK */
2339 spin_unlock(&mm
->page_table_lock
);
2343 /* Workaround for gcc 2.96 */
2344 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2348 #endif /* __PAGETABLE_PMD_FOLDED */
2350 int make_pages_present(unsigned long addr
, unsigned long end
)
2352 int ret
, len
, write
;
2353 struct vm_area_struct
* vma
;
2355 vma
= find_vma(current
->mm
, addr
);
2358 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
2361 if (end
> vma
->vm_end
)
2363 len
= (end
+PAGE_SIZE
-1)/PAGE_SIZE
-addr
/PAGE_SIZE
;
2364 ret
= get_user_pages(current
, current
->mm
, addr
,
2365 len
, write
, 0, NULL
, NULL
);
2368 return ret
== len
? 0 : -1;
2372 * Map a vmalloc()-space virtual address to the physical page.
2374 struct page
* vmalloc_to_page(void * vmalloc_addr
)
2376 unsigned long addr
= (unsigned long) vmalloc_addr
;
2377 struct page
*page
= NULL
;
2378 pgd_t
*pgd
= pgd_offset_k(addr
);
2383 if (!pgd_none(*pgd
)) {
2384 pud
= pud_offset(pgd
, addr
);
2385 if (!pud_none(*pud
)) {
2386 pmd
= pmd_offset(pud
, addr
);
2387 if (!pmd_none(*pmd
)) {
2388 ptep
= pte_offset_map(pmd
, addr
);
2390 if (pte_present(pte
))
2391 page
= pte_page(pte
);
2399 EXPORT_SYMBOL(vmalloc_to_page
);
2402 * Map a vmalloc()-space virtual address to the physical page frame number.
2404 unsigned long vmalloc_to_pfn(void * vmalloc_addr
)
2406 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
2409 EXPORT_SYMBOL(vmalloc_to_pfn
);
2411 #if !defined(__HAVE_ARCH_GATE_AREA)
2413 #if defined(AT_SYSINFO_EHDR)
2414 static struct vm_area_struct gate_vma
;
2416 static int __init
gate_vma_init(void)
2418 gate_vma
.vm_mm
= NULL
;
2419 gate_vma
.vm_start
= FIXADDR_USER_START
;
2420 gate_vma
.vm_end
= FIXADDR_USER_END
;
2421 gate_vma
.vm_page_prot
= PAGE_READONLY
;
2422 gate_vma
.vm_flags
= 0;
2425 __initcall(gate_vma_init
);
2428 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
2430 #ifdef AT_SYSINFO_EHDR
2437 int in_gate_area_no_task(unsigned long addr
)
2439 #ifdef AT_SYSINFO_EHDR
2440 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
2446 #endif /* __HAVE_ARCH_GATE_AREA */