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/delayacct.h>
51 #include <linux/init.h>
52 #include <linux/writeback.h>
54 #include <asm/pgalloc.h>
55 #include <asm/uaccess.h>
57 #include <asm/tlbflush.h>
58 #include <asm/pgtable.h>
60 #include <linux/swapops.h>
61 #include <linux/elf.h>
63 #ifndef CONFIG_NEED_MULTIPLE_NODES
64 /* use the per-pgdat data instead for discontigmem - mbligh */
65 unsigned long max_mapnr
;
68 EXPORT_SYMBOL(max_mapnr
);
69 EXPORT_SYMBOL(mem_map
);
72 unsigned long num_physpages
;
74 * A number of key systems in x86 including ioremap() rely on the assumption
75 * that high_memory defines the upper bound on direct map memory, then end
76 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
77 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
81 unsigned long vmalloc_earlyreserve
;
83 EXPORT_SYMBOL(num_physpages
);
84 EXPORT_SYMBOL(high_memory
);
85 EXPORT_SYMBOL(vmalloc_earlyreserve
);
87 int randomize_va_space __read_mostly
= 1;
89 static int __init
disable_randmaps(char *s
)
91 randomize_va_space
= 0;
94 __setup("norandmaps", disable_randmaps
);
98 * If a p?d_bad entry is found while walking page tables, report
99 * the error, before resetting entry to p?d_none. Usually (but
100 * very seldom) called out from the p?d_none_or_clear_bad macros.
103 void pgd_clear_bad(pgd_t
*pgd
)
109 void pud_clear_bad(pud_t
*pud
)
115 void pmd_clear_bad(pmd_t
*pmd
)
122 * Note: this doesn't free the actual pages themselves. That
123 * has been handled earlier when unmapping all the memory regions.
125 static void free_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
)
127 struct page
*page
= pmd_page(*pmd
);
129 pte_lock_deinit(page
);
130 pte_free_tlb(tlb
, page
);
131 dec_zone_page_state(page
, NR_PAGETABLE
);
135 static inline void free_pmd_range(struct mmu_gather
*tlb
, pud_t
*pud
,
136 unsigned long addr
, unsigned long end
,
137 unsigned long floor
, unsigned long ceiling
)
144 pmd
= pmd_offset(pud
, addr
);
146 next
= pmd_addr_end(addr
, end
);
147 if (pmd_none_or_clear_bad(pmd
))
149 free_pte_range(tlb
, pmd
);
150 } while (pmd
++, addr
= next
, addr
!= end
);
160 if (end
- 1 > ceiling
- 1)
163 pmd
= pmd_offset(pud
, start
);
165 pmd_free_tlb(tlb
, pmd
);
168 static inline void free_pud_range(struct mmu_gather
*tlb
, pgd_t
*pgd
,
169 unsigned long addr
, unsigned long end
,
170 unsigned long floor
, unsigned long ceiling
)
177 pud
= pud_offset(pgd
, addr
);
179 next
= pud_addr_end(addr
, end
);
180 if (pud_none_or_clear_bad(pud
))
182 free_pmd_range(tlb
, pud
, addr
, next
, floor
, ceiling
);
183 } while (pud
++, addr
= next
, addr
!= end
);
189 ceiling
&= PGDIR_MASK
;
193 if (end
- 1 > ceiling
- 1)
196 pud
= pud_offset(pgd
, start
);
198 pud_free_tlb(tlb
, pud
);
202 * This function frees user-level page tables of a process.
204 * Must be called with pagetable lock held.
206 void free_pgd_range(struct mmu_gather
**tlb
,
207 unsigned long addr
, unsigned long end
,
208 unsigned long floor
, unsigned long ceiling
)
215 * The next few lines have given us lots of grief...
217 * Why are we testing PMD* at this top level? Because often
218 * there will be no work to do at all, and we'd prefer not to
219 * go all the way down to the bottom just to discover that.
221 * Why all these "- 1"s? Because 0 represents both the bottom
222 * of the address space and the top of it (using -1 for the
223 * top wouldn't help much: the masks would do the wrong thing).
224 * The rule is that addr 0 and floor 0 refer to the bottom of
225 * the address space, but end 0 and ceiling 0 refer to the top
226 * Comparisons need to use "end - 1" and "ceiling - 1" (though
227 * that end 0 case should be mythical).
229 * Wherever addr is brought up or ceiling brought down, we must
230 * be careful to reject "the opposite 0" before it confuses the
231 * subsequent tests. But what about where end is brought down
232 * by PMD_SIZE below? no, end can't go down to 0 there.
234 * Whereas we round start (addr) and ceiling down, by different
235 * masks at different levels, in order to test whether a table
236 * now has no other vmas using it, so can be freed, we don't
237 * bother to round floor or end up - the tests don't need that.
251 if (end
- 1 > ceiling
- 1)
257 pgd
= pgd_offset((*tlb
)->mm
, addr
);
259 next
= pgd_addr_end(addr
, end
);
260 if (pgd_none_or_clear_bad(pgd
))
262 free_pud_range(*tlb
, pgd
, addr
, next
, floor
, ceiling
);
263 } while (pgd
++, addr
= next
, addr
!= end
);
266 flush_tlb_pgtables((*tlb
)->mm
, start
, end
);
269 void free_pgtables(struct mmu_gather
**tlb
, struct vm_area_struct
*vma
,
270 unsigned long floor
, unsigned long ceiling
)
273 struct vm_area_struct
*next
= vma
->vm_next
;
274 unsigned long addr
= vma
->vm_start
;
277 * Hide vma from rmap and vmtruncate before freeing pgtables
279 anon_vma_unlink(vma
);
280 unlink_file_vma(vma
);
282 if (is_vm_hugetlb_page(vma
)) {
283 hugetlb_free_pgd_range(tlb
, addr
, vma
->vm_end
,
284 floor
, next
? next
->vm_start
: ceiling
);
287 * Optimization: gather nearby vmas into one call down
289 while (next
&& next
->vm_start
<= vma
->vm_end
+ PMD_SIZE
290 && !is_vm_hugetlb_page(next
)) {
293 anon_vma_unlink(vma
);
294 unlink_file_vma(vma
);
296 free_pgd_range(tlb
, addr
, vma
->vm_end
,
297 floor
, next
? next
->vm_start
: ceiling
);
303 int __pte_alloc(struct mm_struct
*mm
, pmd_t
*pmd
, unsigned long address
)
305 struct page
*new = pte_alloc_one(mm
, address
);
310 spin_lock(&mm
->page_table_lock
);
311 if (pmd_present(*pmd
)) { /* Another has populated it */
312 pte_lock_deinit(new);
316 inc_zone_page_state(new, NR_PAGETABLE
);
317 pmd_populate(mm
, pmd
, new);
319 spin_unlock(&mm
->page_table_lock
);
323 int __pte_alloc_kernel(pmd_t
*pmd
, unsigned long address
)
325 pte_t
*new = pte_alloc_one_kernel(&init_mm
, address
);
329 spin_lock(&init_mm
.page_table_lock
);
330 if (pmd_present(*pmd
)) /* Another has populated it */
331 pte_free_kernel(new);
333 pmd_populate_kernel(&init_mm
, pmd
, new);
334 spin_unlock(&init_mm
.page_table_lock
);
338 static inline void add_mm_rss(struct mm_struct
*mm
, int file_rss
, int anon_rss
)
341 add_mm_counter(mm
, file_rss
, file_rss
);
343 add_mm_counter(mm
, anon_rss
, anon_rss
);
347 * This function is called to print an error when a bad pte
348 * is found. For example, we might have a PFN-mapped pte in
349 * a region that doesn't allow it.
351 * The calling function must still handle the error.
353 void print_bad_pte(struct vm_area_struct
*vma
, pte_t pte
, unsigned long vaddr
)
355 printk(KERN_ERR
"Bad pte = %08llx, process = %s, "
356 "vm_flags = %lx, vaddr = %lx\n",
357 (long long)pte_val(pte
),
358 (vma
->vm_mm
== current
->mm
? current
->comm
: "???"),
359 vma
->vm_flags
, vaddr
);
363 static inline int is_cow_mapping(unsigned int flags
)
365 return (flags
& (VM_SHARED
| VM_MAYWRITE
)) == VM_MAYWRITE
;
369 * This function gets the "struct page" associated with a pte.
371 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
372 * will have each page table entry just pointing to a raw page frame
373 * number, and as far as the VM layer is concerned, those do not have
374 * pages associated with them - even if the PFN might point to memory
375 * that otherwise is perfectly fine and has a "struct page".
377 * The way we recognize those mappings is through the rules set up
378 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
379 * and the vm_pgoff will point to the first PFN mapped: thus every
380 * page that is a raw mapping will always honor the rule
382 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
384 * and if that isn't true, the page has been COW'ed (in which case it
385 * _does_ have a "struct page" associated with it even if it is in a
388 struct page
*vm_normal_page(struct vm_area_struct
*vma
, unsigned long addr
, pte_t pte
)
390 unsigned long pfn
= pte_pfn(pte
);
392 if (unlikely(vma
->vm_flags
& VM_PFNMAP
)) {
393 unsigned long off
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
394 if (pfn
== vma
->vm_pgoff
+ off
)
396 if (!is_cow_mapping(vma
->vm_flags
))
401 * Add some anal sanity checks for now. Eventually,
402 * we should just do "return pfn_to_page(pfn)", but
403 * in the meantime we check that we get a valid pfn,
404 * and that the resulting page looks ok.
406 if (unlikely(!pfn_valid(pfn
))) {
407 print_bad_pte(vma
, pte
, addr
);
412 * NOTE! We still have PageReserved() pages in the page
415 * The PAGE_ZERO() pages and various VDSO mappings can
416 * cause them to exist.
418 return pfn_to_page(pfn
);
422 * copy one vm_area from one task to the other. Assumes the page tables
423 * already present in the new task to be cleared in the whole range
424 * covered by this vma.
428 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
429 pte_t
*dst_pte
, pte_t
*src_pte
, struct vm_area_struct
*vma
,
430 unsigned long addr
, int *rss
)
432 unsigned long vm_flags
= vma
->vm_flags
;
433 pte_t pte
= *src_pte
;
436 /* pte contains position in swap or file, so copy. */
437 if (unlikely(!pte_present(pte
))) {
438 if (!pte_file(pte
)) {
439 swp_entry_t entry
= pte_to_swp_entry(pte
);
441 swap_duplicate(entry
);
442 /* make sure dst_mm is on swapoff's mmlist. */
443 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
444 spin_lock(&mmlist_lock
);
445 if (list_empty(&dst_mm
->mmlist
))
446 list_add(&dst_mm
->mmlist
,
448 spin_unlock(&mmlist_lock
);
450 if (is_write_migration_entry(entry
) &&
451 is_cow_mapping(vm_flags
)) {
453 * COW mappings require pages in both parent
454 * and child to be set to read.
456 make_migration_entry_read(&entry
);
457 pte
= swp_entry_to_pte(entry
);
458 set_pte_at(src_mm
, addr
, src_pte
, pte
);
465 * If it's a COW mapping, write protect it both
466 * in the parent and the child
468 if (is_cow_mapping(vm_flags
)) {
469 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
474 * If it's a shared mapping, mark it clean in
477 if (vm_flags
& VM_SHARED
)
478 pte
= pte_mkclean(pte
);
479 pte
= pte_mkold(pte
);
481 page
= vm_normal_page(vma
, addr
, pte
);
485 rss
[!!PageAnon(page
)]++;
489 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
492 static int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
493 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
494 unsigned long addr
, unsigned long end
)
496 pte_t
*src_pte
, *dst_pte
;
497 spinlock_t
*src_ptl
, *dst_ptl
;
503 dst_pte
= pte_alloc_map_lock(dst_mm
, dst_pmd
, addr
, &dst_ptl
);
506 src_pte
= pte_offset_map_nested(src_pmd
, addr
);
507 src_ptl
= pte_lockptr(src_mm
, src_pmd
);
508 spin_lock_nested(src_ptl
, SINGLE_DEPTH_NESTING
);
512 * We are holding two locks at this point - either of them
513 * could generate latencies in another task on another CPU.
515 if (progress
>= 32) {
517 if (need_resched() ||
518 need_lockbreak(src_ptl
) ||
519 need_lockbreak(dst_ptl
))
522 if (pte_none(*src_pte
)) {
526 copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
, vma
, addr
, rss
);
528 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
530 spin_unlock(src_ptl
);
531 pte_unmap_nested(src_pte
- 1);
532 add_mm_rss(dst_mm
, rss
[0], rss
[1]);
533 pte_unmap_unlock(dst_pte
- 1, dst_ptl
);
540 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
541 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
542 unsigned long addr
, unsigned long end
)
544 pmd_t
*src_pmd
, *dst_pmd
;
547 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
550 src_pmd
= pmd_offset(src_pud
, addr
);
552 next
= pmd_addr_end(addr
, end
);
553 if (pmd_none_or_clear_bad(src_pmd
))
555 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
558 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
562 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
563 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
564 unsigned long addr
, unsigned long end
)
566 pud_t
*src_pud
, *dst_pud
;
569 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
572 src_pud
= pud_offset(src_pgd
, addr
);
574 next
= pud_addr_end(addr
, end
);
575 if (pud_none_or_clear_bad(src_pud
))
577 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
580 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
584 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
585 struct vm_area_struct
*vma
)
587 pgd_t
*src_pgd
, *dst_pgd
;
589 unsigned long addr
= vma
->vm_start
;
590 unsigned long end
= vma
->vm_end
;
593 * Don't copy ptes where a page fault will fill them correctly.
594 * Fork becomes much lighter when there are big shared or private
595 * readonly mappings. The tradeoff is that copy_page_range is more
596 * efficient than faulting.
598 if (!(vma
->vm_flags
& (VM_HUGETLB
|VM_NONLINEAR
|VM_PFNMAP
|VM_INSERTPAGE
))) {
603 if (is_vm_hugetlb_page(vma
))
604 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
606 dst_pgd
= pgd_offset(dst_mm
, addr
);
607 src_pgd
= pgd_offset(src_mm
, addr
);
609 next
= pgd_addr_end(addr
, end
);
610 if (pgd_none_or_clear_bad(src_pgd
))
612 if (copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
615 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
619 static unsigned long zap_pte_range(struct mmu_gather
*tlb
,
620 struct vm_area_struct
*vma
, pmd_t
*pmd
,
621 unsigned long addr
, unsigned long end
,
622 long *zap_work
, struct zap_details
*details
)
624 struct mm_struct
*mm
= tlb
->mm
;
630 pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
633 if (pte_none(ptent
)) {
638 (*zap_work
) -= PAGE_SIZE
;
640 if (pte_present(ptent
)) {
643 page
= vm_normal_page(vma
, addr
, ptent
);
644 if (unlikely(details
) && page
) {
646 * unmap_shared_mapping_pages() wants to
647 * invalidate cache without truncating:
648 * unmap shared but keep private pages.
650 if (details
->check_mapping
&&
651 details
->check_mapping
!= page
->mapping
)
654 * Each page->index must be checked when
655 * invalidating or truncating nonlinear.
657 if (details
->nonlinear_vma
&&
658 (page
->index
< details
->first_index
||
659 page
->index
> details
->last_index
))
662 ptent
= ptep_get_and_clear_full(mm
, addr
, pte
,
664 tlb_remove_tlb_entry(tlb
, pte
, addr
);
667 if (unlikely(details
) && details
->nonlinear_vma
668 && linear_page_index(details
->nonlinear_vma
,
669 addr
) != page
->index
)
670 set_pte_at(mm
, addr
, pte
,
671 pgoff_to_pte(page
->index
));
675 if (pte_dirty(ptent
))
676 set_page_dirty(page
);
677 if (pte_young(ptent
))
678 mark_page_accessed(page
);
681 page_remove_rmap(page
);
682 tlb_remove_page(tlb
, page
);
686 * If details->check_mapping, we leave swap entries;
687 * if details->nonlinear_vma, we leave file entries.
689 if (unlikely(details
))
691 if (!pte_file(ptent
))
692 free_swap_and_cache(pte_to_swp_entry(ptent
));
693 pte_clear_full(mm
, addr
, pte
, tlb
->fullmm
);
694 } while (pte
++, addr
+= PAGE_SIZE
, (addr
!= end
&& *zap_work
> 0));
696 add_mm_rss(mm
, file_rss
, anon_rss
);
697 pte_unmap_unlock(pte
- 1, ptl
);
702 static inline unsigned long zap_pmd_range(struct mmu_gather
*tlb
,
703 struct vm_area_struct
*vma
, pud_t
*pud
,
704 unsigned long addr
, unsigned long end
,
705 long *zap_work
, struct zap_details
*details
)
710 pmd
= pmd_offset(pud
, addr
);
712 next
= pmd_addr_end(addr
, end
);
713 if (pmd_none_or_clear_bad(pmd
)) {
717 next
= zap_pte_range(tlb
, vma
, pmd
, addr
, next
,
719 } while (pmd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
724 static inline unsigned long zap_pud_range(struct mmu_gather
*tlb
,
725 struct vm_area_struct
*vma
, pgd_t
*pgd
,
726 unsigned long addr
, unsigned long end
,
727 long *zap_work
, struct zap_details
*details
)
732 pud
= pud_offset(pgd
, addr
);
734 next
= pud_addr_end(addr
, end
);
735 if (pud_none_or_clear_bad(pud
)) {
739 next
= zap_pmd_range(tlb
, vma
, pud
, addr
, next
,
741 } while (pud
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
746 static unsigned long unmap_page_range(struct mmu_gather
*tlb
,
747 struct vm_area_struct
*vma
,
748 unsigned long addr
, unsigned long end
,
749 long *zap_work
, struct zap_details
*details
)
754 if (details
&& !details
->check_mapping
&& !details
->nonlinear_vma
)
758 tlb_start_vma(tlb
, vma
);
759 pgd
= pgd_offset(vma
->vm_mm
, addr
);
761 next
= pgd_addr_end(addr
, end
);
762 if (pgd_none_or_clear_bad(pgd
)) {
766 next
= zap_pud_range(tlb
, vma
, pgd
, addr
, next
,
768 } while (pgd
++, addr
= next
, (addr
!= end
&& *zap_work
> 0));
769 tlb_end_vma(tlb
, vma
);
774 #ifdef CONFIG_PREEMPT
775 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
777 /* No preempt: go for improved straight-line efficiency */
778 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
782 * unmap_vmas - unmap a range of memory covered by a list of vma's
783 * @tlbp: address of the caller's struct mmu_gather
784 * @vma: the starting vma
785 * @start_addr: virtual address at which to start unmapping
786 * @end_addr: virtual address at which to end unmapping
787 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
788 * @details: details of nonlinear truncation or shared cache invalidation
790 * Returns the end address of the unmapping (restart addr if interrupted).
792 * Unmap all pages in the vma list.
794 * We aim to not hold locks for too long (for scheduling latency reasons).
795 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
796 * return the ending mmu_gather to the caller.
798 * Only addresses between `start' and `end' will be unmapped.
800 * The VMA list must be sorted in ascending virtual address order.
802 * unmap_vmas() assumes that the caller will flush the whole unmapped address
803 * range after unmap_vmas() returns. So the only responsibility here is to
804 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
805 * drops the lock and schedules.
807 unsigned long unmap_vmas(struct mmu_gather
**tlbp
,
808 struct vm_area_struct
*vma
, unsigned long start_addr
,
809 unsigned long end_addr
, unsigned long *nr_accounted
,
810 struct zap_details
*details
)
812 long zap_work
= ZAP_BLOCK_SIZE
;
813 unsigned long tlb_start
= 0; /* For tlb_finish_mmu */
814 int tlb_start_valid
= 0;
815 unsigned long start
= start_addr
;
816 spinlock_t
*i_mmap_lock
= details
? details
->i_mmap_lock
: NULL
;
817 int fullmm
= (*tlbp
)->fullmm
;
819 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
) {
822 start
= max(vma
->vm_start
, start_addr
);
823 if (start
>= vma
->vm_end
)
825 end
= min(vma
->vm_end
, end_addr
);
826 if (end
<= vma
->vm_start
)
829 if (vma
->vm_flags
& VM_ACCOUNT
)
830 *nr_accounted
+= (end
- start
) >> PAGE_SHIFT
;
832 while (start
!= end
) {
833 if (!tlb_start_valid
) {
838 if (unlikely(is_vm_hugetlb_page(vma
))) {
839 unmap_hugepage_range(vma
, start
, end
);
840 zap_work
-= (end
- start
) /
841 (HPAGE_SIZE
/ PAGE_SIZE
);
844 start
= unmap_page_range(*tlbp
, vma
,
845 start
, end
, &zap_work
, details
);
848 BUG_ON(start
!= end
);
852 tlb_finish_mmu(*tlbp
, tlb_start
, start
);
854 if (need_resched() ||
855 (i_mmap_lock
&& need_lockbreak(i_mmap_lock
))) {
863 *tlbp
= tlb_gather_mmu(vma
->vm_mm
, fullmm
);
865 zap_work
= ZAP_BLOCK_SIZE
;
869 return start
; /* which is now the end (or restart) address */
873 * zap_page_range - remove user pages in a given range
874 * @vma: vm_area_struct holding the applicable pages
875 * @address: starting address of pages to zap
876 * @size: number of bytes to zap
877 * @details: details of nonlinear truncation or shared cache invalidation
879 unsigned long zap_page_range(struct vm_area_struct
*vma
, unsigned long address
,
880 unsigned long size
, struct zap_details
*details
)
882 struct mm_struct
*mm
= vma
->vm_mm
;
883 struct mmu_gather
*tlb
;
884 unsigned long end
= address
+ size
;
885 unsigned long nr_accounted
= 0;
888 tlb
= tlb_gather_mmu(mm
, 0);
889 update_hiwater_rss(mm
);
890 end
= unmap_vmas(&tlb
, vma
, address
, end
, &nr_accounted
, details
);
892 tlb_finish_mmu(tlb
, address
, end
);
897 * Do a quick page-table lookup for a single page.
899 struct page
*follow_page(struct vm_area_struct
*vma
, unsigned long address
,
908 struct mm_struct
*mm
= vma
->vm_mm
;
910 page
= follow_huge_addr(mm
, address
, flags
& FOLL_WRITE
);
912 BUG_ON(flags
& FOLL_GET
);
917 pgd
= pgd_offset(mm
, address
);
918 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
921 pud
= pud_offset(pgd
, address
);
922 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
925 pmd
= pmd_offset(pud
, address
);
926 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
929 if (pmd_huge(*pmd
)) {
930 BUG_ON(flags
& FOLL_GET
);
931 page
= follow_huge_pmd(mm
, address
, pmd
, flags
& FOLL_WRITE
);
935 ptep
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
940 if (!pte_present(pte
))
942 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
944 page
= vm_normal_page(vma
, address
, pte
);
948 if (flags
& FOLL_GET
)
950 if (flags
& FOLL_TOUCH
) {
951 if ((flags
& FOLL_WRITE
) &&
952 !pte_dirty(pte
) && !PageDirty(page
))
953 set_page_dirty(page
);
954 mark_page_accessed(page
);
957 pte_unmap_unlock(ptep
, ptl
);
963 * When core dumping an enormous anonymous area that nobody
964 * has touched so far, we don't want to allocate page tables.
966 if (flags
& FOLL_ANON
) {
967 page
= ZERO_PAGE(address
);
968 if (flags
& FOLL_GET
)
970 BUG_ON(flags
& FOLL_WRITE
);
975 int get_user_pages(struct task_struct
*tsk
, struct mm_struct
*mm
,
976 unsigned long start
, int len
, int write
, int force
,
977 struct page
**pages
, struct vm_area_struct
**vmas
)
980 unsigned int vm_flags
;
983 * Require read or write permissions.
984 * If 'force' is set, we only require the "MAY" flags.
986 vm_flags
= write
? (VM_WRITE
| VM_MAYWRITE
) : (VM_READ
| VM_MAYREAD
);
987 vm_flags
&= force
? (VM_MAYREAD
| VM_MAYWRITE
) : (VM_READ
| VM_WRITE
);
991 struct vm_area_struct
*vma
;
992 unsigned int foll_flags
;
994 vma
= find_extend_vma(mm
, start
);
995 if (!vma
&& in_gate_area(tsk
, start
)) {
996 unsigned long pg
= start
& PAGE_MASK
;
997 struct vm_area_struct
*gate_vma
= get_gate_vma(tsk
);
1002 if (write
) /* user gate pages are read-only */
1003 return i
? : -EFAULT
;
1005 pgd
= pgd_offset_k(pg
);
1007 pgd
= pgd_offset_gate(mm
, pg
);
1008 BUG_ON(pgd_none(*pgd
));
1009 pud
= pud_offset(pgd
, pg
);
1010 BUG_ON(pud_none(*pud
));
1011 pmd
= pmd_offset(pud
, pg
);
1013 return i
? : -EFAULT
;
1014 pte
= pte_offset_map(pmd
, pg
);
1015 if (pte_none(*pte
)) {
1017 return i
? : -EFAULT
;
1020 struct page
*page
= vm_normal_page(gate_vma
, start
, *pte
);
1034 if (!vma
|| (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
))
1035 || !(vm_flags
& vma
->vm_flags
))
1036 return i
? : -EFAULT
;
1038 if (is_vm_hugetlb_page(vma
)) {
1039 i
= follow_hugetlb_page(mm
, vma
, pages
, vmas
,
1044 foll_flags
= FOLL_TOUCH
;
1046 foll_flags
|= FOLL_GET
;
1047 if (!write
&& !(vma
->vm_flags
& VM_LOCKED
) &&
1048 (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
))
1049 foll_flags
|= FOLL_ANON
;
1055 foll_flags
|= FOLL_WRITE
;
1058 while (!(page
= follow_page(vma
, start
, foll_flags
))) {
1060 ret
= __handle_mm_fault(mm
, vma
, start
,
1061 foll_flags
& FOLL_WRITE
);
1063 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1064 * broken COW when necessary, even if maybe_mkwrite
1065 * decided not to set pte_write. We can thus safely do
1066 * subsequent page lookups as if they were reads.
1068 if (ret
& VM_FAULT_WRITE
)
1069 foll_flags
&= ~FOLL_WRITE
;
1071 switch (ret
& ~VM_FAULT_WRITE
) {
1072 case VM_FAULT_MINOR
:
1075 case VM_FAULT_MAJOR
:
1078 case VM_FAULT_SIGBUS
:
1079 return i
? i
: -EFAULT
;
1081 return i
? i
: -ENOMEM
;
1089 flush_anon_page(page
, start
);
1090 flush_dcache_page(page
);
1097 } while (len
&& start
< vma
->vm_end
);
1101 EXPORT_SYMBOL(get_user_pages
);
1103 static int zeromap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1104 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1109 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1113 struct page
*page
= ZERO_PAGE(addr
);
1114 pte_t zero_pte
= pte_wrprotect(mk_pte(page
, prot
));
1115 page_cache_get(page
);
1116 page_add_file_rmap(page
);
1117 inc_mm_counter(mm
, file_rss
);
1118 BUG_ON(!pte_none(*pte
));
1119 set_pte_at(mm
, addr
, pte
, zero_pte
);
1120 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1121 pte_unmap_unlock(pte
- 1, ptl
);
1125 static inline int zeromap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1126 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1131 pmd
= pmd_alloc(mm
, pud
, addr
);
1135 next
= pmd_addr_end(addr
, end
);
1136 if (zeromap_pte_range(mm
, pmd
, addr
, next
, prot
))
1138 } while (pmd
++, addr
= next
, addr
!= end
);
1142 static inline int zeromap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1143 unsigned long addr
, unsigned long end
, pgprot_t prot
)
1148 pud
= pud_alloc(mm
, pgd
, addr
);
1152 next
= pud_addr_end(addr
, end
);
1153 if (zeromap_pmd_range(mm
, pud
, addr
, next
, prot
))
1155 } while (pud
++, addr
= next
, addr
!= end
);
1159 int zeromap_page_range(struct vm_area_struct
*vma
,
1160 unsigned long addr
, unsigned long size
, pgprot_t prot
)
1164 unsigned long end
= addr
+ size
;
1165 struct mm_struct
*mm
= vma
->vm_mm
;
1168 BUG_ON(addr
>= end
);
1169 pgd
= pgd_offset(mm
, addr
);
1170 flush_cache_range(vma
, addr
, end
);
1172 next
= pgd_addr_end(addr
, end
);
1173 err
= zeromap_pud_range(mm
, pgd
, addr
, next
, prot
);
1176 } while (pgd
++, addr
= next
, addr
!= end
);
1180 pte_t
* fastcall
get_locked_pte(struct mm_struct
*mm
, unsigned long addr
, spinlock_t
**ptl
)
1182 pgd_t
* pgd
= pgd_offset(mm
, addr
);
1183 pud_t
* pud
= pud_alloc(mm
, pgd
, addr
);
1185 pmd_t
* pmd
= pmd_alloc(mm
, pud
, addr
);
1187 return pte_alloc_map_lock(mm
, pmd
, addr
, ptl
);
1193 * This is the old fallback for page remapping.
1195 * For historical reasons, it only allows reserved pages. Only
1196 * old drivers should use this, and they needed to mark their
1197 * pages reserved for the old functions anyway.
1199 static int insert_page(struct mm_struct
*mm
, unsigned long addr
, struct page
*page
, pgprot_t prot
)
1209 flush_dcache_page(page
);
1210 pte
= get_locked_pte(mm
, addr
, &ptl
);
1214 if (!pte_none(*pte
))
1217 /* Ok, finally just insert the thing.. */
1219 inc_mm_counter(mm
, file_rss
);
1220 page_add_file_rmap(page
);
1221 set_pte_at(mm
, addr
, pte
, mk_pte(page
, prot
));
1225 pte_unmap_unlock(pte
, ptl
);
1231 * This allows drivers to insert individual pages they've allocated
1234 * The page has to be a nice clean _individual_ kernel allocation.
1235 * If you allocate a compound page, you need to have marked it as
1236 * such (__GFP_COMP), or manually just split the page up yourself
1237 * (see split_page()).
1239 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1240 * took an arbitrary page protection parameter. This doesn't allow
1241 * that. Your vma protection will have to be set up correctly, which
1242 * means that if you want a shared writable mapping, you'd better
1243 * ask for a shared writable mapping!
1245 * The page does not need to be reserved.
1247 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
, struct page
*page
)
1249 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
1251 if (!page_count(page
))
1253 vma
->vm_flags
|= VM_INSERTPAGE
;
1254 return insert_page(vma
->vm_mm
, addr
, page
, vma
->vm_page_prot
);
1256 EXPORT_SYMBOL(vm_insert_page
);
1259 * maps a range of physical memory into the requested pages. the old
1260 * mappings are removed. any references to nonexistent pages results
1261 * in null mappings (currently treated as "copy-on-access")
1263 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1264 unsigned long addr
, unsigned long end
,
1265 unsigned long pfn
, pgprot_t prot
)
1270 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1274 BUG_ON(!pte_none(*pte
));
1275 set_pte_at(mm
, addr
, pte
, pfn_pte(pfn
, prot
));
1277 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1278 pte_unmap_unlock(pte
- 1, ptl
);
1282 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1283 unsigned long addr
, unsigned long end
,
1284 unsigned long pfn
, pgprot_t prot
)
1289 pfn
-= addr
>> PAGE_SHIFT
;
1290 pmd
= pmd_alloc(mm
, pud
, addr
);
1294 next
= pmd_addr_end(addr
, end
);
1295 if (remap_pte_range(mm
, pmd
, addr
, next
,
1296 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1298 } while (pmd
++, addr
= next
, addr
!= end
);
1302 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1303 unsigned long addr
, unsigned long end
,
1304 unsigned long pfn
, pgprot_t prot
)
1309 pfn
-= addr
>> PAGE_SHIFT
;
1310 pud
= pud_alloc(mm
, pgd
, addr
);
1314 next
= pud_addr_end(addr
, end
);
1315 if (remap_pmd_range(mm
, pud
, addr
, next
,
1316 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1318 } while (pud
++, addr
= next
, addr
!= end
);
1322 /* Note: this is only safe if the mm semaphore is held when called. */
1323 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1324 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1328 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1329 struct mm_struct
*mm
= vma
->vm_mm
;
1333 * Physically remapped pages are special. Tell the
1334 * rest of the world about it:
1335 * VM_IO tells people not to look at these pages
1336 * (accesses can have side effects).
1337 * VM_RESERVED is specified all over the place, because
1338 * in 2.4 it kept swapout's vma scan off this vma; but
1339 * in 2.6 the LRU scan won't even find its pages, so this
1340 * flag means no more than count its pages in reserved_vm,
1341 * and omit it from core dump, even when VM_IO turned off.
1342 * VM_PFNMAP tells the core MM that the base pages are just
1343 * raw PFN mappings, and do not have a "struct page" associated
1346 * There's a horrible special case to handle copy-on-write
1347 * behaviour that some programs depend on. We mark the "original"
1348 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1350 if (is_cow_mapping(vma
->vm_flags
)) {
1351 if (addr
!= vma
->vm_start
|| end
!= vma
->vm_end
)
1353 vma
->vm_pgoff
= pfn
;
1356 vma
->vm_flags
|= VM_IO
| VM_RESERVED
| VM_PFNMAP
;
1358 BUG_ON(addr
>= end
);
1359 pfn
-= addr
>> PAGE_SHIFT
;
1360 pgd
= pgd_offset(mm
, addr
);
1361 flush_cache_range(vma
, addr
, end
);
1363 next
= pgd_addr_end(addr
, end
);
1364 err
= remap_pud_range(mm
, pgd
, addr
, next
,
1365 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
1368 } while (pgd
++, addr
= next
, addr
!= end
);
1371 EXPORT_SYMBOL(remap_pfn_range
);
1374 * handle_pte_fault chooses page fault handler according to an entry
1375 * which was read non-atomically. Before making any commitment, on
1376 * those architectures or configurations (e.g. i386 with PAE) which
1377 * might give a mix of unmatched parts, do_swap_page and do_file_page
1378 * must check under lock before unmapping the pte and proceeding
1379 * (but do_wp_page is only called after already making such a check;
1380 * and do_anonymous_page and do_no_page can safely check later on).
1382 static inline int pte_unmap_same(struct mm_struct
*mm
, pmd_t
*pmd
,
1383 pte_t
*page_table
, pte_t orig_pte
)
1386 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1387 if (sizeof(pte_t
) > sizeof(unsigned long)) {
1388 spinlock_t
*ptl
= pte_lockptr(mm
, pmd
);
1390 same
= pte_same(*page_table
, orig_pte
);
1394 pte_unmap(page_table
);
1399 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1400 * servicing faults for write access. In the normal case, do always want
1401 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1402 * that do not have writing enabled, when used by access_process_vm.
1404 static inline pte_t
maybe_mkwrite(pte_t pte
, struct vm_area_struct
*vma
)
1406 if (likely(vma
->vm_flags
& VM_WRITE
))
1407 pte
= pte_mkwrite(pte
);
1411 static inline void cow_user_page(struct page
*dst
, struct page
*src
, unsigned long va
)
1414 * If the source page was a PFN mapping, we don't have
1415 * a "struct page" for it. We do a best-effort copy by
1416 * just copying from the original user address. If that
1417 * fails, we just zero-fill it. Live with it.
1419 if (unlikely(!src
)) {
1420 void *kaddr
= kmap_atomic(dst
, KM_USER0
);
1421 void __user
*uaddr
= (void __user
*)(va
& PAGE_MASK
);
1424 * This really shouldn't fail, because the page is there
1425 * in the page tables. But it might just be unreadable,
1426 * in which case we just give up and fill the result with
1429 if (__copy_from_user_inatomic(kaddr
, uaddr
, PAGE_SIZE
))
1430 memset(kaddr
, 0, PAGE_SIZE
);
1431 kunmap_atomic(kaddr
, KM_USER0
);
1435 copy_user_highpage(dst
, src
, va
);
1439 * This routine handles present pages, when users try to write
1440 * to a shared page. It is done by copying the page to a new address
1441 * and decrementing the shared-page counter for the old page.
1443 * Note that this routine assumes that the protection checks have been
1444 * done by the caller (the low-level page fault routine in most cases).
1445 * Thus we can safely just mark it writable once we've done any necessary
1448 * We also mark the page dirty at this point even though the page will
1449 * change only once the write actually happens. This avoids a few races,
1450 * and potentially makes it more efficient.
1452 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1453 * but allow concurrent faults), with pte both mapped and locked.
1454 * We return with mmap_sem still held, but pte unmapped and unlocked.
1456 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1457 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1458 spinlock_t
*ptl
, pte_t orig_pte
)
1460 struct page
*old_page
, *new_page
;
1462 int reuse
= 0, ret
= VM_FAULT_MINOR
;
1463 struct page
*dirty_page
= NULL
;
1465 old_page
= vm_normal_page(vma
, address
, orig_pte
);
1470 * Only catch write-faults on shared writable pages, read-only
1471 * shared pages can get COWed by get_user_pages(.write=1, .force=1).
1473 if (unlikely((vma
->vm_flags
& (VM_WRITE
|VM_SHARED
)) ==
1474 (VM_WRITE
|VM_SHARED
))) {
1475 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
) {
1477 * Notify the address space that the page is about to
1478 * become writable so that it can prohibit this or wait
1479 * for the page to get into an appropriate state.
1481 * We do this without the lock held, so that it can
1482 * sleep if it needs to.
1484 page_cache_get(old_page
);
1485 pte_unmap_unlock(page_table
, ptl
);
1487 if (vma
->vm_ops
->page_mkwrite(vma
, old_page
) < 0)
1488 goto unwritable_page
;
1490 page_cache_release(old_page
);
1493 * Since we dropped the lock we need to revalidate
1494 * the PTE as someone else may have changed it. If
1495 * they did, we just return, as we can count on the
1496 * MMU to tell us if they didn't also make it writable.
1498 page_table
= pte_offset_map_lock(mm
, pmd
, address
,
1500 if (!pte_same(*page_table
, orig_pte
))
1503 dirty_page
= old_page
;
1504 get_page(dirty_page
);
1506 } else if (PageAnon(old_page
) && !TestSetPageLocked(old_page
)) {
1507 reuse
= can_share_swap_page(old_page
);
1508 unlock_page(old_page
);
1512 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
1513 entry
= pte_mkyoung(orig_pte
);
1514 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1515 ptep_set_access_flags(vma
, address
, page_table
, entry
, 1);
1516 update_mmu_cache(vma
, address
, entry
);
1517 lazy_mmu_prot_update(entry
);
1518 ret
|= VM_FAULT_WRITE
;
1523 * Ok, we need to copy. Oh, well..
1525 page_cache_get(old_page
);
1527 pte_unmap_unlock(page_table
, ptl
);
1529 if (unlikely(anon_vma_prepare(vma
)))
1531 if (old_page
== ZERO_PAGE(address
)) {
1532 new_page
= alloc_zeroed_user_highpage(vma
, address
);
1536 new_page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
1539 cow_user_page(new_page
, old_page
, address
);
1543 * Re-check the pte - we dropped the lock
1545 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1546 if (likely(pte_same(*page_table
, orig_pte
))) {
1548 page_remove_rmap(old_page
);
1549 if (!PageAnon(old_page
)) {
1550 dec_mm_counter(mm
, file_rss
);
1551 inc_mm_counter(mm
, anon_rss
);
1554 inc_mm_counter(mm
, anon_rss
);
1555 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
1556 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
1557 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
1558 lazy_mmu_prot_update(entry
);
1559 ptep_establish(vma
, address
, page_table
, entry
);
1560 update_mmu_cache(vma
, address
, entry
);
1561 lru_cache_add_active(new_page
);
1562 page_add_new_anon_rmap(new_page
, vma
, address
);
1564 /* Free the old page.. */
1565 new_page
= old_page
;
1566 ret
|= VM_FAULT_WRITE
;
1569 page_cache_release(new_page
);
1571 page_cache_release(old_page
);
1573 pte_unmap_unlock(page_table
, ptl
);
1575 set_page_dirty_balance(dirty_page
);
1576 put_page(dirty_page
);
1581 page_cache_release(old_page
);
1582 return VM_FAULT_OOM
;
1585 page_cache_release(old_page
);
1586 return VM_FAULT_SIGBUS
;
1590 * Helper functions for unmap_mapping_range().
1592 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1594 * We have to restart searching the prio_tree whenever we drop the lock,
1595 * since the iterator is only valid while the lock is held, and anyway
1596 * a later vma might be split and reinserted earlier while lock dropped.
1598 * The list of nonlinear vmas could be handled more efficiently, using
1599 * a placeholder, but handle it in the same way until a need is shown.
1600 * It is important to search the prio_tree before nonlinear list: a vma
1601 * may become nonlinear and be shifted from prio_tree to nonlinear list
1602 * while the lock is dropped; but never shifted from list to prio_tree.
1604 * In order to make forward progress despite restarting the search,
1605 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1606 * quickly skip it next time around. Since the prio_tree search only
1607 * shows us those vmas affected by unmapping the range in question, we
1608 * can't efficiently keep all vmas in step with mapping->truncate_count:
1609 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1610 * mapping->truncate_count and vma->vm_truncate_count are protected by
1613 * In order to make forward progress despite repeatedly restarting some
1614 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1615 * and restart from that address when we reach that vma again. It might
1616 * have been split or merged, shrunk or extended, but never shifted: so
1617 * restart_addr remains valid so long as it remains in the vma's range.
1618 * unmap_mapping_range forces truncate_count to leap over page-aligned
1619 * values so we can save vma's restart_addr in its truncate_count field.
1621 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1623 static void reset_vma_truncate_counts(struct address_space
*mapping
)
1625 struct vm_area_struct
*vma
;
1626 struct prio_tree_iter iter
;
1628 vma_prio_tree_foreach(vma
, &iter
, &mapping
->i_mmap
, 0, ULONG_MAX
)
1629 vma
->vm_truncate_count
= 0;
1630 list_for_each_entry(vma
, &mapping
->i_mmap_nonlinear
, shared
.vm_set
.list
)
1631 vma
->vm_truncate_count
= 0;
1634 static int unmap_mapping_range_vma(struct vm_area_struct
*vma
,
1635 unsigned long start_addr
, unsigned long end_addr
,
1636 struct zap_details
*details
)
1638 unsigned long restart_addr
;
1642 restart_addr
= vma
->vm_truncate_count
;
1643 if (is_restart_addr(restart_addr
) && start_addr
< restart_addr
) {
1644 start_addr
= restart_addr
;
1645 if (start_addr
>= end_addr
) {
1646 /* Top of vma has been split off since last time */
1647 vma
->vm_truncate_count
= details
->truncate_count
;
1652 restart_addr
= zap_page_range(vma
, start_addr
,
1653 end_addr
- start_addr
, details
);
1654 need_break
= need_resched() ||
1655 need_lockbreak(details
->i_mmap_lock
);
1657 if (restart_addr
>= end_addr
) {
1658 /* We have now completed this vma: mark it so */
1659 vma
->vm_truncate_count
= details
->truncate_count
;
1663 /* Note restart_addr in vma's truncate_count field */
1664 vma
->vm_truncate_count
= restart_addr
;
1669 spin_unlock(details
->i_mmap_lock
);
1671 spin_lock(details
->i_mmap_lock
);
1675 static inline void unmap_mapping_range_tree(struct prio_tree_root
*root
,
1676 struct zap_details
*details
)
1678 struct vm_area_struct
*vma
;
1679 struct prio_tree_iter iter
;
1680 pgoff_t vba
, vea
, zba
, zea
;
1683 vma_prio_tree_foreach(vma
, &iter
, root
,
1684 details
->first_index
, details
->last_index
) {
1685 /* Skip quickly over those we have already dealt with */
1686 if (vma
->vm_truncate_count
== details
->truncate_count
)
1689 vba
= vma
->vm_pgoff
;
1690 vea
= vba
+ ((vma
->vm_end
- vma
->vm_start
) >> PAGE_SHIFT
) - 1;
1691 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1692 zba
= details
->first_index
;
1695 zea
= details
->last_index
;
1699 if (unmap_mapping_range_vma(vma
,
1700 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
1701 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
1707 static inline void unmap_mapping_range_list(struct list_head
*head
,
1708 struct zap_details
*details
)
1710 struct vm_area_struct
*vma
;
1713 * In nonlinear VMAs there is no correspondence between virtual address
1714 * offset and file offset. So we must perform an exhaustive search
1715 * across *all* the pages in each nonlinear VMA, not just the pages
1716 * whose virtual address lies outside the file truncation point.
1719 list_for_each_entry(vma
, head
, shared
.vm_set
.list
) {
1720 /* Skip quickly over those we have already dealt with */
1721 if (vma
->vm_truncate_count
== details
->truncate_count
)
1723 details
->nonlinear_vma
= vma
;
1724 if (unmap_mapping_range_vma(vma
, vma
->vm_start
,
1725 vma
->vm_end
, details
) < 0)
1731 * unmap_mapping_range - unmap the portion of all mmaps
1732 * in the specified address_space corresponding to the specified
1733 * page range in the underlying file.
1734 * @mapping: the address space containing mmaps to be unmapped.
1735 * @holebegin: byte in first page to unmap, relative to the start of
1736 * the underlying file. This will be rounded down to a PAGE_SIZE
1737 * boundary. Note that this is different from vmtruncate(), which
1738 * must keep the partial page. In contrast, we must get rid of
1740 * @holelen: size of prospective hole in bytes. This will be rounded
1741 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1743 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1744 * but 0 when invalidating pagecache, don't throw away private data.
1746 void unmap_mapping_range(struct address_space
*mapping
,
1747 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
1749 struct zap_details details
;
1750 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
1751 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1753 /* Check for overflow. */
1754 if (sizeof(holelen
) > sizeof(hlen
)) {
1756 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1757 if (holeend
& ~(long long)ULONG_MAX
)
1758 hlen
= ULONG_MAX
- hba
+ 1;
1761 details
.check_mapping
= even_cows
? NULL
: mapping
;
1762 details
.nonlinear_vma
= NULL
;
1763 details
.first_index
= hba
;
1764 details
.last_index
= hba
+ hlen
- 1;
1765 if (details
.last_index
< details
.first_index
)
1766 details
.last_index
= ULONG_MAX
;
1767 details
.i_mmap_lock
= &mapping
->i_mmap_lock
;
1769 spin_lock(&mapping
->i_mmap_lock
);
1771 /* serialize i_size write against truncate_count write */
1773 /* Protect against page faults, and endless unmapping loops */
1774 mapping
->truncate_count
++;
1776 * For archs where spin_lock has inclusive semantics like ia64
1777 * this smp_mb() will prevent to read pagetable contents
1778 * before the truncate_count increment is visible to
1782 if (unlikely(is_restart_addr(mapping
->truncate_count
))) {
1783 if (mapping
->truncate_count
== 0)
1784 reset_vma_truncate_counts(mapping
);
1785 mapping
->truncate_count
++;
1787 details
.truncate_count
= mapping
->truncate_count
;
1789 if (unlikely(!prio_tree_empty(&mapping
->i_mmap
)))
1790 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
1791 if (unlikely(!list_empty(&mapping
->i_mmap_nonlinear
)))
1792 unmap_mapping_range_list(&mapping
->i_mmap_nonlinear
, &details
);
1793 spin_unlock(&mapping
->i_mmap_lock
);
1795 EXPORT_SYMBOL(unmap_mapping_range
);
1798 * Handle all mappings that got truncated by a "truncate()"
1801 * NOTE! We have to be ready to update the memory sharing
1802 * between the file and the memory map for a potential last
1803 * incomplete page. Ugly, but necessary.
1805 int vmtruncate(struct inode
* inode
, loff_t offset
)
1807 struct address_space
*mapping
= inode
->i_mapping
;
1808 unsigned long limit
;
1810 if (inode
->i_size
< offset
)
1813 * truncation of in-use swapfiles is disallowed - it would cause
1814 * subsequent swapout to scribble on the now-freed blocks.
1816 if (IS_SWAPFILE(inode
))
1818 i_size_write(inode
, offset
);
1819 unmap_mapping_range(mapping
, offset
+ PAGE_SIZE
- 1, 0, 1);
1820 truncate_inode_pages(mapping
, offset
);
1824 limit
= current
->signal
->rlim
[RLIMIT_FSIZE
].rlim_cur
;
1825 if (limit
!= RLIM_INFINITY
&& offset
> limit
)
1827 if (offset
> inode
->i_sb
->s_maxbytes
)
1829 i_size_write(inode
, offset
);
1832 if (inode
->i_op
&& inode
->i_op
->truncate
)
1833 inode
->i_op
->truncate(inode
);
1836 send_sig(SIGXFSZ
, current
, 0);
1842 EXPORT_SYMBOL(vmtruncate
);
1844 int vmtruncate_range(struct inode
*inode
, loff_t offset
, loff_t end
)
1846 struct address_space
*mapping
= inode
->i_mapping
;
1849 * If the underlying filesystem is not going to provide
1850 * a way to truncate a range of blocks (punch a hole) -
1851 * we should return failure right now.
1853 if (!inode
->i_op
|| !inode
->i_op
->truncate_range
)
1856 mutex_lock(&inode
->i_mutex
);
1857 down_write(&inode
->i_alloc_sem
);
1858 unmap_mapping_range(mapping
, offset
, (end
- offset
), 1);
1859 truncate_inode_pages_range(mapping
, offset
, end
);
1860 inode
->i_op
->truncate_range(inode
, offset
, end
);
1861 up_write(&inode
->i_alloc_sem
);
1862 mutex_unlock(&inode
->i_mutex
);
1866 EXPORT_UNUSED_SYMBOL(vmtruncate_range
); /* June 2006 */
1869 * Primitive swap readahead code. We simply read an aligned block of
1870 * (1 << page_cluster) entries in the swap area. This method is chosen
1871 * because it doesn't cost us any seek time. We also make sure to queue
1872 * the 'original' request together with the readahead ones...
1874 * This has been extended to use the NUMA policies from the mm triggering
1877 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1879 void swapin_readahead(swp_entry_t entry
, unsigned long addr
,struct vm_area_struct
*vma
)
1882 struct vm_area_struct
*next_vma
= vma
? vma
->vm_next
: NULL
;
1885 struct page
*new_page
;
1886 unsigned long offset
;
1889 * Get the number of handles we should do readahead io to.
1891 num
= valid_swaphandles(entry
, &offset
);
1892 for (i
= 0; i
< num
; offset
++, i
++) {
1893 /* Ok, do the async read-ahead now */
1894 new_page
= read_swap_cache_async(swp_entry(swp_type(entry
),
1895 offset
), vma
, addr
);
1898 page_cache_release(new_page
);
1901 * Find the next applicable VMA for the NUMA policy.
1907 if (addr
>= vma
->vm_end
) {
1909 next_vma
= vma
? vma
->vm_next
: NULL
;
1911 if (vma
&& addr
< vma
->vm_start
)
1914 if (next_vma
&& addr
>= next_vma
->vm_start
) {
1916 next_vma
= vma
->vm_next
;
1921 lru_add_drain(); /* Push any new pages onto the LRU now */
1925 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1926 * but allow concurrent faults), and pte mapped but not yet locked.
1927 * We return with mmap_sem still held, but pte unmapped and unlocked.
1929 static int do_swap_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
1930 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
1931 int write_access
, pte_t orig_pte
)
1937 int ret
= VM_FAULT_MINOR
;
1939 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
1942 entry
= pte_to_swp_entry(orig_pte
);
1943 if (is_migration_entry(entry
)) {
1944 migration_entry_wait(mm
, pmd
, address
);
1947 delayacct_set_flag(DELAYACCT_PF_SWAPIN
);
1948 page
= lookup_swap_cache(entry
);
1950 swapin_readahead(entry
, address
, vma
);
1951 page
= read_swap_cache_async(entry
, vma
, address
);
1954 * Back out if somebody else faulted in this pte
1955 * while we released the pte lock.
1957 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1958 if (likely(pte_same(*page_table
, orig_pte
)))
1960 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
1964 /* Had to read the page from swap area: Major fault */
1965 ret
= VM_FAULT_MAJOR
;
1966 count_vm_event(PGMAJFAULT
);
1970 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
1971 mark_page_accessed(page
);
1975 * Back out if somebody else already faulted in this pte.
1977 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
1978 if (unlikely(!pte_same(*page_table
, orig_pte
)))
1981 if (unlikely(!PageUptodate(page
))) {
1982 ret
= VM_FAULT_SIGBUS
;
1986 /* The page isn't present yet, go ahead with the fault. */
1988 inc_mm_counter(mm
, anon_rss
);
1989 pte
= mk_pte(page
, vma
->vm_page_prot
);
1990 if (write_access
&& can_share_swap_page(page
)) {
1991 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
1995 flush_icache_page(vma
, page
);
1996 set_pte_at(mm
, address
, page_table
, pte
);
1997 page_add_anon_rmap(page
, vma
, address
);
2001 remove_exclusive_swap_page(page
);
2005 if (do_wp_page(mm
, vma
, address
,
2006 page_table
, pmd
, ptl
, pte
) == VM_FAULT_OOM
)
2011 /* No need to invalidate - it was non-present before */
2012 update_mmu_cache(vma
, address
, pte
);
2013 lazy_mmu_prot_update(pte
);
2015 pte_unmap_unlock(page_table
, ptl
);
2019 pte_unmap_unlock(page_table
, ptl
);
2021 page_cache_release(page
);
2026 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2027 * but allow concurrent faults), and pte mapped but not yet locked.
2028 * We return with mmap_sem still held, but pte unmapped and unlocked.
2030 static int do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2031 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2039 /* Allocate our own private page. */
2040 pte_unmap(page_table
);
2042 if (unlikely(anon_vma_prepare(vma
)))
2044 page
= alloc_zeroed_user_highpage(vma
, address
);
2048 entry
= mk_pte(page
, vma
->vm_page_prot
);
2049 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2051 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2052 if (!pte_none(*page_table
))
2054 inc_mm_counter(mm
, anon_rss
);
2055 lru_cache_add_active(page
);
2056 page_add_new_anon_rmap(page
, vma
, address
);
2058 /* Map the ZERO_PAGE - vm_page_prot is readonly */
2059 page
= ZERO_PAGE(address
);
2060 page_cache_get(page
);
2061 entry
= mk_pte(page
, vma
->vm_page_prot
);
2063 ptl
= pte_lockptr(mm
, pmd
);
2065 if (!pte_none(*page_table
))
2067 inc_mm_counter(mm
, file_rss
);
2068 page_add_file_rmap(page
);
2071 set_pte_at(mm
, address
, page_table
, entry
);
2073 /* No need to invalidate - it was non-present before */
2074 update_mmu_cache(vma
, address
, entry
);
2075 lazy_mmu_prot_update(entry
);
2077 pte_unmap_unlock(page_table
, ptl
);
2078 return VM_FAULT_MINOR
;
2080 page_cache_release(page
);
2083 return VM_FAULT_OOM
;
2087 * do_no_page() tries to create a new page mapping. It aggressively
2088 * tries to share with existing pages, but makes a separate copy if
2089 * the "write_access" parameter is true in order to avoid the next
2092 * As this is called only for pages that do not currently exist, we
2093 * do not need to flush old virtual caches or the TLB.
2095 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2096 * but allow concurrent faults), and pte mapped but not yet locked.
2097 * We return with mmap_sem still held, but pte unmapped and unlocked.
2099 static int do_no_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2100 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2104 struct page
*new_page
;
2105 struct address_space
*mapping
= NULL
;
2107 unsigned int sequence
= 0;
2108 int ret
= VM_FAULT_MINOR
;
2110 struct page
*dirty_page
= NULL
;
2112 pte_unmap(page_table
);
2113 BUG_ON(vma
->vm_flags
& VM_PFNMAP
);
2116 mapping
= vma
->vm_file
->f_mapping
;
2117 sequence
= mapping
->truncate_count
;
2118 smp_rmb(); /* serializes i_size against truncate_count */
2121 new_page
= vma
->vm_ops
->nopage(vma
, address
& PAGE_MASK
, &ret
);
2123 * No smp_rmb is needed here as long as there's a full
2124 * spin_lock/unlock sequence inside the ->nopage callback
2125 * (for the pagecache lookup) that acts as an implicit
2126 * smp_mb() and prevents the i_size read to happen
2127 * after the next truncate_count read.
2130 /* no page was available -- either SIGBUS or OOM */
2131 if (new_page
== NOPAGE_SIGBUS
)
2132 return VM_FAULT_SIGBUS
;
2133 if (new_page
== NOPAGE_OOM
)
2134 return VM_FAULT_OOM
;
2137 * Should we do an early C-O-W break?
2140 if (!(vma
->vm_flags
& VM_SHARED
)) {
2143 if (unlikely(anon_vma_prepare(vma
)))
2145 page
= alloc_page_vma(GFP_HIGHUSER
, vma
, address
);
2148 copy_user_highpage(page
, new_page
, address
);
2149 page_cache_release(new_page
);
2154 /* if the page will be shareable, see if the backing
2155 * address space wants to know that the page is about
2156 * to become writable */
2157 if (vma
->vm_ops
->page_mkwrite
&&
2158 vma
->vm_ops
->page_mkwrite(vma
, new_page
) < 0
2160 page_cache_release(new_page
);
2161 return VM_FAULT_SIGBUS
;
2166 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2168 * For a file-backed vma, someone could have truncated or otherwise
2169 * invalidated this page. If unmap_mapping_range got called,
2170 * retry getting the page.
2172 if (mapping
&& unlikely(sequence
!= mapping
->truncate_count
)) {
2173 pte_unmap_unlock(page_table
, ptl
);
2174 page_cache_release(new_page
);
2176 sequence
= mapping
->truncate_count
;
2182 * This silly early PAGE_DIRTY setting removes a race
2183 * due to the bad i386 page protection. But it's valid
2184 * for other architectures too.
2186 * Note that if write_access is true, we either now have
2187 * an exclusive copy of the page, or this is a shared mapping,
2188 * so we can make it writable and dirty to avoid having to
2189 * handle that later.
2191 /* Only go through if we didn't race with anybody else... */
2192 if (pte_none(*page_table
)) {
2193 flush_icache_page(vma
, new_page
);
2194 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
2196 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2197 set_pte_at(mm
, address
, page_table
, entry
);
2199 inc_mm_counter(mm
, anon_rss
);
2200 lru_cache_add_active(new_page
);
2201 page_add_new_anon_rmap(new_page
, vma
, address
);
2203 inc_mm_counter(mm
, file_rss
);
2204 page_add_file_rmap(new_page
);
2206 dirty_page
= new_page
;
2207 get_page(dirty_page
);
2211 /* One of our sibling threads was faster, back out. */
2212 page_cache_release(new_page
);
2216 /* no need to invalidate: a not-present page shouldn't be cached */
2217 update_mmu_cache(vma
, address
, entry
);
2218 lazy_mmu_prot_update(entry
);
2220 pte_unmap_unlock(page_table
, ptl
);
2222 set_page_dirty_balance(dirty_page
);
2223 put_page(dirty_page
);
2227 page_cache_release(new_page
);
2228 return VM_FAULT_OOM
;
2232 * Fault of a previously existing named mapping. Repopulate the pte
2233 * from the encoded file_pte if possible. This enables swappable
2236 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2237 * but allow concurrent faults), and pte mapped but not yet locked.
2238 * We return with mmap_sem still held, but pte unmapped and unlocked.
2240 static int do_file_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2241 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2242 int write_access
, pte_t orig_pte
)
2247 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
2248 return VM_FAULT_MINOR
;
2250 if (unlikely(!(vma
->vm_flags
& VM_NONLINEAR
))) {
2252 * Page table corrupted: show pte and kill process.
2254 print_bad_pte(vma
, orig_pte
, address
);
2255 return VM_FAULT_OOM
;
2257 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2259 pgoff
= pte_to_pgoff(orig_pte
);
2260 err
= vma
->vm_ops
->populate(vma
, address
& PAGE_MASK
, PAGE_SIZE
,
2261 vma
->vm_page_prot
, pgoff
, 0);
2263 return VM_FAULT_OOM
;
2265 return VM_FAULT_SIGBUS
;
2266 return VM_FAULT_MAJOR
;
2270 * These routines also need to handle stuff like marking pages dirty
2271 * and/or accessed for architectures that don't do it in hardware (most
2272 * RISC architectures). The early dirtying is also good on the i386.
2274 * There is also a hook called "update_mmu_cache()" that architectures
2275 * with external mmu caches can use to update those (ie the Sparc or
2276 * PowerPC hashed page tables that act as extended TLBs).
2278 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2279 * but allow concurrent faults), and pte mapped but not yet locked.
2280 * We return with mmap_sem still held, but pte unmapped and unlocked.
2282 static inline int handle_pte_fault(struct mm_struct
*mm
,
2283 struct vm_area_struct
*vma
, unsigned long address
,
2284 pte_t
*pte
, pmd_t
*pmd
, int write_access
)
2290 old_entry
= entry
= *pte
;
2291 if (!pte_present(entry
)) {
2292 if (pte_none(entry
)) {
2293 if (!vma
->vm_ops
|| !vma
->vm_ops
->nopage
)
2294 return do_anonymous_page(mm
, vma
, address
,
2295 pte
, pmd
, write_access
);
2296 return do_no_page(mm
, vma
, address
,
2297 pte
, pmd
, write_access
);
2299 if (pte_file(entry
))
2300 return do_file_page(mm
, vma
, address
,
2301 pte
, pmd
, write_access
, entry
);
2302 return do_swap_page(mm
, vma
, address
,
2303 pte
, pmd
, write_access
, entry
);
2306 ptl
= pte_lockptr(mm
, pmd
);
2308 if (unlikely(!pte_same(*pte
, entry
)))
2311 if (!pte_write(entry
))
2312 return do_wp_page(mm
, vma
, address
,
2313 pte
, pmd
, ptl
, entry
);
2314 entry
= pte_mkdirty(entry
);
2316 entry
= pte_mkyoung(entry
);
2317 if (!pte_same(old_entry
, entry
)) {
2318 ptep_set_access_flags(vma
, address
, pte
, entry
, write_access
);
2319 update_mmu_cache(vma
, address
, entry
);
2320 lazy_mmu_prot_update(entry
);
2323 * This is needed only for protection faults but the arch code
2324 * is not yet telling us if this is a protection fault or not.
2325 * This still avoids useless tlb flushes for .text page faults
2329 flush_tlb_page(vma
, address
);
2332 pte_unmap_unlock(pte
, ptl
);
2333 return VM_FAULT_MINOR
;
2337 * By the time we get here, we already hold the mm semaphore
2339 int __handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2340 unsigned long address
, int write_access
)
2347 __set_current_state(TASK_RUNNING
);
2349 count_vm_event(PGFAULT
);
2351 if (unlikely(is_vm_hugetlb_page(vma
)))
2352 return hugetlb_fault(mm
, vma
, address
, write_access
);
2354 pgd
= pgd_offset(mm
, address
);
2355 pud
= pud_alloc(mm
, pgd
, address
);
2357 return VM_FAULT_OOM
;
2358 pmd
= pmd_alloc(mm
, pud
, address
);
2360 return VM_FAULT_OOM
;
2361 pte
= pte_alloc_map(mm
, pmd
, address
);
2363 return VM_FAULT_OOM
;
2365 return handle_pte_fault(mm
, vma
, address
, pte
, pmd
, write_access
);
2368 EXPORT_SYMBOL_GPL(__handle_mm_fault
);
2370 #ifndef __PAGETABLE_PUD_FOLDED
2372 * Allocate page upper directory.
2373 * We've already handled the fast-path in-line.
2375 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2377 pud_t
*new = pud_alloc_one(mm
, address
);
2381 spin_lock(&mm
->page_table_lock
);
2382 if (pgd_present(*pgd
)) /* Another has populated it */
2385 pgd_populate(mm
, pgd
, new);
2386 spin_unlock(&mm
->page_table_lock
);
2390 /* Workaround for gcc 2.96 */
2391 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
2395 #endif /* __PAGETABLE_PUD_FOLDED */
2397 #ifndef __PAGETABLE_PMD_FOLDED
2399 * Allocate page middle directory.
2400 * We've already handled the fast-path in-line.
2402 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2404 pmd_t
*new = pmd_alloc_one(mm
, address
);
2408 spin_lock(&mm
->page_table_lock
);
2409 #ifndef __ARCH_HAS_4LEVEL_HACK
2410 if (pud_present(*pud
)) /* Another has populated it */
2413 pud_populate(mm
, pud
, new);
2415 if (pgd_present(*pud
)) /* Another has populated it */
2418 pgd_populate(mm
, pud
, new);
2419 #endif /* __ARCH_HAS_4LEVEL_HACK */
2420 spin_unlock(&mm
->page_table_lock
);
2424 /* Workaround for gcc 2.96 */
2425 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
2429 #endif /* __PAGETABLE_PMD_FOLDED */
2431 int make_pages_present(unsigned long addr
, unsigned long end
)
2433 int ret
, len
, write
;
2434 struct vm_area_struct
* vma
;
2436 vma
= find_vma(current
->mm
, addr
);
2439 write
= (vma
->vm_flags
& VM_WRITE
) != 0;
2440 BUG_ON(addr
>= end
);
2441 BUG_ON(end
> vma
->vm_end
);
2442 len
= (end
+PAGE_SIZE
-1)/PAGE_SIZE
-addr
/PAGE_SIZE
;
2443 ret
= get_user_pages(current
, current
->mm
, addr
,
2444 len
, write
, 0, NULL
, NULL
);
2447 return ret
== len
? 0 : -1;
2451 * Map a vmalloc()-space virtual address to the physical page.
2453 struct page
* vmalloc_to_page(void * vmalloc_addr
)
2455 unsigned long addr
= (unsigned long) vmalloc_addr
;
2456 struct page
*page
= NULL
;
2457 pgd_t
*pgd
= pgd_offset_k(addr
);
2462 if (!pgd_none(*pgd
)) {
2463 pud
= pud_offset(pgd
, addr
);
2464 if (!pud_none(*pud
)) {
2465 pmd
= pmd_offset(pud
, addr
);
2466 if (!pmd_none(*pmd
)) {
2467 ptep
= pte_offset_map(pmd
, addr
);
2469 if (pte_present(pte
))
2470 page
= pte_page(pte
);
2478 EXPORT_SYMBOL(vmalloc_to_page
);
2481 * Map a vmalloc()-space virtual address to the physical page frame number.
2483 unsigned long vmalloc_to_pfn(void * vmalloc_addr
)
2485 return page_to_pfn(vmalloc_to_page(vmalloc_addr
));
2488 EXPORT_SYMBOL(vmalloc_to_pfn
);
2490 #if !defined(__HAVE_ARCH_GATE_AREA)
2492 #if defined(AT_SYSINFO_EHDR)
2493 static struct vm_area_struct gate_vma
;
2495 static int __init
gate_vma_init(void)
2497 gate_vma
.vm_mm
= NULL
;
2498 gate_vma
.vm_start
= FIXADDR_USER_START
;
2499 gate_vma
.vm_end
= FIXADDR_USER_END
;
2500 gate_vma
.vm_page_prot
= PAGE_READONLY
;
2501 gate_vma
.vm_flags
= 0;
2504 __initcall(gate_vma_init
);
2507 struct vm_area_struct
*get_gate_vma(struct task_struct
*tsk
)
2509 #ifdef AT_SYSINFO_EHDR
2516 int in_gate_area_no_task(unsigned long addr
)
2518 #ifdef AT_SYSINFO_EHDR
2519 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
2525 #endif /* __HAVE_ARCH_GATE_AREA */