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 if (!(vma->vm_flags & VM_RESERVED))
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 swp_entry_t entry = pte_to_swp_entry(pte);
442 swap_duplicate(entry);
443 /* make sure dst_mm is on swapoff's mmlist. */
444 if (unlikely(list_empty(&dst_mm->mmlist))) {
445 spin_lock(&mmlist_lock);
446 if (list_empty(&dst_mm->mmlist))
447 list_add(&dst_mm->mmlist,
449 spin_unlock(&mmlist_lock);
451 if (is_write_migration_entry(entry) &&
452 is_cow_mapping(vm_flags)) {
454 * COW mappings require pages in both parent
455 * and child to be set to read.
457 make_migration_entry_read(&entry);
458 pte = swp_entry_to_pte(entry);
459 set_pte_at(src_mm, addr, src_pte, pte);
466 * If it's a COW mapping, write protect it both
467 * in the parent and the child
469 if (is_cow_mapping(vm_flags)) {
470 ptep_set_wrprotect(src_mm, addr, src_pte);
471 pte = pte_wrprotect(pte);
475 * If it's a shared mapping, mark it clean in
478 if (vm_flags & VM_SHARED)
479 pte = pte_mkclean(pte);
480 pte = pte_mkold(pte);
482 page = vm_normal_page(vma, addr, pte);
486 rss[!!PageAnon(page)]++;
490 set_pte_at(dst_mm, addr, dst_pte, pte);
493 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
494 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
495 unsigned long addr, unsigned long end)
497 pte_t *src_pte, *dst_pte;
498 spinlock_t *src_ptl, *dst_ptl;
502 if (!vx_rss_avail(dst_mm, ((end - addr)/PAGE_SIZE + 1)))
507 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
510 src_pte = pte_offset_map_nested(src_pmd, addr);
511 src_ptl = pte_lockptr(src_mm, src_pmd);
512 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
513 arch_enter_lazy_mmu_mode();
517 * We are holding two locks at this point - either of them
518 * could generate latencies in another task on another CPU.
520 if (progress >= 32) {
522 if (need_resched() ||
523 need_lockbreak(src_ptl) ||
524 need_lockbreak(dst_ptl))
527 if (pte_none(*src_pte)) {
531 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
533 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
535 arch_leave_lazy_mmu_mode();
536 spin_unlock(src_ptl);
537 pte_unmap_nested(src_pte - 1);
538 add_mm_rss(dst_mm, rss[0], rss[1]);
539 pte_unmap_unlock(dst_pte - 1, dst_ptl);
546 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
547 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
548 unsigned long addr, unsigned long end)
550 pmd_t *src_pmd, *dst_pmd;
553 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
556 src_pmd = pmd_offset(src_pud, addr);
558 next = pmd_addr_end(addr, end);
559 if (pmd_none_or_clear_bad(src_pmd))
561 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
564 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
568 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
569 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
570 unsigned long addr, unsigned long end)
572 pud_t *src_pud, *dst_pud;
575 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
578 src_pud = pud_offset(src_pgd, addr);
580 next = pud_addr_end(addr, end);
581 if (pud_none_or_clear_bad(src_pud))
583 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
586 } while (dst_pud++, src_pud++, addr = next, addr != end);
590 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
591 struct vm_area_struct *vma)
593 pgd_t *src_pgd, *dst_pgd;
595 unsigned long addr = vma->vm_start;
596 unsigned long end = vma->vm_end;
599 * Don't copy ptes where a page fault will fill them correctly.
600 * Fork becomes much lighter when there are big shared or private
601 * readonly mappings. The tradeoff is that copy_page_range is more
602 * efficient than faulting.
604 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
609 if (is_vm_hugetlb_page(vma))
610 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
612 dst_pgd = pgd_offset(dst_mm, addr);
613 src_pgd = pgd_offset(src_mm, addr);
615 next = pgd_addr_end(addr, end);
616 if (pgd_none_or_clear_bad(src_pgd))
618 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
621 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
625 static unsigned long zap_pte_range(struct mmu_gather *tlb,
626 struct vm_area_struct *vma, pmd_t *pmd,
627 unsigned long addr, unsigned long end,
628 long *zap_work, struct zap_details *details)
630 struct mm_struct *mm = tlb->mm;
636 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
637 arch_enter_lazy_mmu_mode();
640 if (pte_none(ptent)) {
645 (*zap_work) -= PAGE_SIZE;
647 if (pte_present(ptent)) {
650 page = vm_normal_page(vma, addr, ptent);
651 if (unlikely(details) && page) {
653 * unmap_shared_mapping_pages() wants to
654 * invalidate cache without truncating:
655 * unmap shared but keep private pages.
657 if (details->check_mapping &&
658 details->check_mapping != page->mapping)
661 * Each page->index must be checked when
662 * invalidating or truncating nonlinear.
664 if (details->nonlinear_vma &&
665 (page->index < details->first_index ||
666 page->index > details->last_index))
669 ptent = ptep_get_and_clear_full(mm, addr, pte,
671 tlb_remove_tlb_entry(tlb, pte, addr);
674 if (unlikely(details) && details->nonlinear_vma
675 && linear_page_index(details->nonlinear_vma,
676 addr) != page->index)
677 set_pte_at(mm, addr, pte,
678 pgoff_to_pte(page->index));
682 if (pte_dirty(ptent))
683 set_page_dirty(page);
684 if (pte_young(ptent))
685 mark_page_accessed(page);
688 page_remove_rmap(page, vma);
689 tlb_remove_page(tlb, page);
693 * If details->check_mapping, we leave swap entries;
694 * if details->nonlinear_vma, we leave file entries.
696 if (unlikely(details))
698 if (!pte_file(ptent))
699 free_swap_and_cache(pte_to_swp_entry(ptent));
700 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
701 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
703 add_mm_rss(mm, file_rss, anon_rss);
704 arch_leave_lazy_mmu_mode();
705 pte_unmap_unlock(pte - 1, ptl);
710 static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
711 struct vm_area_struct *vma, pud_t *pud,
712 unsigned long addr, unsigned long end,
713 long *zap_work, struct zap_details *details)
718 pmd = pmd_offset(pud, addr);
720 next = pmd_addr_end(addr, end);
721 if (pmd_none_or_clear_bad(pmd)) {
725 next = zap_pte_range(tlb, vma, pmd, addr, next,
727 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
732 static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
733 struct vm_area_struct *vma, pgd_t *pgd,
734 unsigned long addr, unsigned long end,
735 long *zap_work, struct zap_details *details)
740 pud = pud_offset(pgd, addr);
742 next = pud_addr_end(addr, end);
743 if (pud_none_or_clear_bad(pud)) {
747 next = zap_pmd_range(tlb, vma, pud, addr, next,
749 } while (pud++, addr = next, (addr != end && *zap_work > 0));
754 static unsigned long unmap_page_range(struct mmu_gather *tlb,
755 struct vm_area_struct *vma,
756 unsigned long addr, unsigned long end,
757 long *zap_work, struct zap_details *details)
762 if (details && !details->check_mapping && !details->nonlinear_vma)
766 tlb_start_vma(tlb, vma);
767 pgd = pgd_offset(vma->vm_mm, addr);
769 next = pgd_addr_end(addr, end);
770 if (pgd_none_or_clear_bad(pgd)) {
774 next = zap_pud_range(tlb, vma, pgd, addr, next,
776 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
777 tlb_end_vma(tlb, vma);
782 #ifdef CONFIG_PREEMPT
783 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
785 /* No preempt: go for improved straight-line efficiency */
786 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
790 * unmap_vmas - unmap a range of memory covered by a list of vma's
791 * @tlbp: address of the caller's struct mmu_gather
792 * @vma: the starting vma
793 * @start_addr: virtual address at which to start unmapping
794 * @end_addr: virtual address at which to end unmapping
795 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
796 * @details: details of nonlinear truncation or shared cache invalidation
798 * Returns the end address of the unmapping (restart addr if interrupted).
800 * Unmap all pages in the vma list.
802 * We aim to not hold locks for too long (for scheduling latency reasons).
803 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
804 * return the ending mmu_gather to the caller.
806 * Only addresses between `start' and `end' will be unmapped.
808 * The VMA list must be sorted in ascending virtual address order.
810 * unmap_vmas() assumes that the caller will flush the whole unmapped address
811 * range after unmap_vmas() returns. So the only responsibility here is to
812 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
813 * drops the lock and schedules.
815 unsigned long unmap_vmas(struct mmu_gather **tlbp,
816 struct vm_area_struct *vma, unsigned long start_addr,
817 unsigned long end_addr, unsigned long *nr_accounted,
818 struct zap_details *details)
820 long zap_work = ZAP_BLOCK_SIZE;
821 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
822 int tlb_start_valid = 0;
823 unsigned long start = start_addr;
824 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
825 int fullmm = (*tlbp)->fullmm;
827 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
830 start = max(vma->vm_start, start_addr);
831 if (start >= vma->vm_end)
833 end = min(vma->vm_end, end_addr);
834 if (end <= vma->vm_start)
837 if (vma->vm_flags & VM_ACCOUNT)
838 *nr_accounted += (end - start) >> PAGE_SHIFT;
840 while (start != end) {
841 if (!tlb_start_valid) {
846 if (unlikely(is_vm_hugetlb_page(vma))) {
847 unmap_hugepage_range(vma, start, end);
848 zap_work -= (end - start) /
849 (HPAGE_SIZE / PAGE_SIZE);
852 start = unmap_page_range(*tlbp, vma,
853 start, end, &zap_work, details);
856 BUG_ON(start != end);
860 tlb_finish_mmu(*tlbp, tlb_start, start);
862 if (need_resched() ||
863 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
871 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
873 zap_work = ZAP_BLOCK_SIZE;
877 return start; /* which is now the end (or restart) address */
881 * zap_page_range - remove user pages in a given range
882 * @vma: vm_area_struct holding the applicable pages
883 * @address: starting address of pages to zap
884 * @size: number of bytes to zap
885 * @details: details of nonlinear truncation or shared cache invalidation
887 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
888 unsigned long size, struct zap_details *details)
890 struct mm_struct *mm = vma->vm_mm;
891 struct mmu_gather *tlb;
892 unsigned long end = address + size;
893 unsigned long nr_accounted = 0;
896 tlb = tlb_gather_mmu(mm, 0);
897 update_hiwater_rss(mm);
898 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
900 tlb_finish_mmu(tlb, address, end);
903 EXPORT_SYMBOL(zap_page_range);
906 * Do a quick page-table lookup for a single page.
908 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
917 struct mm_struct *mm = vma->vm_mm;
919 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
921 BUG_ON(flags & FOLL_GET);
926 pgd = pgd_offset(mm, address);
927 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
930 pud = pud_offset(pgd, address);
931 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
934 pmd = pmd_offset(pud, address);
935 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
938 if (pmd_huge(*pmd)) {
939 BUG_ON(flags & FOLL_GET);
940 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
944 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
949 if (!pte_present(pte))
951 if ((flags & FOLL_WRITE) && !pte_write(pte))
953 page = vm_normal_page(vma, address, pte);
957 if (flags & FOLL_GET)
959 if (flags & FOLL_TOUCH) {
960 if ((flags & FOLL_WRITE) &&
961 !pte_dirty(pte) && !PageDirty(page))
962 set_page_dirty(page);
963 mark_page_accessed(page);
966 pte_unmap_unlock(ptep, ptl);
972 * When core dumping an enormous anonymous area that nobody
973 * has touched so far, we don't want to allocate page tables.
975 if (flags & FOLL_ANON) {
976 page = ZERO_PAGE(address);
977 if (flags & FOLL_GET)
979 BUG_ON(flags & FOLL_WRITE);
984 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
985 unsigned long start, int len, int write, int force,
986 struct page **pages, struct vm_area_struct **vmas)
989 unsigned int vm_flags;
992 * Require read or write permissions.
993 * If 'force' is set, we only require the "MAY" flags.
995 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
996 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1000 struct vm_area_struct *vma;
1001 unsigned int foll_flags;
1003 vma = find_extend_vma(mm, start);
1004 if (!vma && in_gate_area(tsk, start)) {
1005 unsigned long pg = start & PAGE_MASK;
1006 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1011 if (write) /* user gate pages are read-only */
1012 return i ? : -EFAULT;
1014 pgd = pgd_offset_k(pg);
1016 pgd = pgd_offset_gate(mm, pg);
1017 BUG_ON(pgd_none(*pgd));
1018 pud = pud_offset(pgd, pg);
1019 BUG_ON(pud_none(*pud));
1020 pmd = pmd_offset(pud, pg);
1022 return i ? : -EFAULT;
1023 pte = pte_offset_map(pmd, pg);
1024 if (pte_none(*pte)) {
1026 return i ? : -EFAULT;
1029 struct page *page = vm_normal_page(gate_vma, start, *pte);
1044 if (vma && (vma->vm_flags & VM_FOREIGN)) {
1045 struct page **map = vma->vm_private_data;
1046 int offset = (start - vma->vm_start) >> PAGE_SHIFT;
1047 if (map[offset] != NULL) {
1049 struct page *page = map[offset];
1063 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
1064 || !(vm_flags & vma->vm_flags))
1065 return i ? : -EFAULT;
1067 if (is_vm_hugetlb_page(vma)) {
1068 i = follow_hugetlb_page(mm, vma, pages, vmas,
1073 foll_flags = FOLL_TOUCH;
1075 foll_flags |= FOLL_GET;
1076 if (!write && !(vma->vm_flags & VM_LOCKED) &&
1077 (!vma->vm_ops || !vma->vm_ops->nopage))
1078 foll_flags |= FOLL_ANON;
1084 foll_flags |= FOLL_WRITE;
1087 while (!(page = follow_page(vma, start, foll_flags))) {
1089 ret = __handle_mm_fault(mm, vma, start,
1090 foll_flags & FOLL_WRITE);
1092 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1093 * broken COW when necessary, even if maybe_mkwrite
1094 * decided not to set pte_write. We can thus safely do
1095 * subsequent page lookups as if they were reads.
1097 if (ret & VM_FAULT_WRITE)
1098 foll_flags &= ~FOLL_WRITE;
1100 switch (ret & ~VM_FAULT_WRITE) {
1101 case VM_FAULT_MINOR:
1104 case VM_FAULT_MAJOR:
1107 case VM_FAULT_SIGBUS:
1108 return i ? i : -EFAULT;
1110 return i ? i : -ENOMEM;
1119 flush_anon_page(vma, page, start);
1120 flush_dcache_page(page);
1127 } while (len && start < vma->vm_end);
1131 EXPORT_SYMBOL(get_user_pages);
1133 static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1134 unsigned long addr, unsigned long end, pgprot_t prot)
1140 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1143 arch_enter_lazy_mmu_mode();
1145 struct page *page = ZERO_PAGE(addr);
1146 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1148 if (unlikely(!pte_none(*pte))) {
1153 page_cache_get(page);
1154 page_add_file_rmap(page);
1155 inc_mm_counter(mm, file_rss);
1156 set_pte_at(mm, addr, pte, zero_pte);
1157 } while (pte++, addr += PAGE_SIZE, addr != end);
1158 arch_leave_lazy_mmu_mode();
1159 pte_unmap_unlock(pte - 1, ptl);
1163 static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1164 unsigned long addr, unsigned long end, pgprot_t prot)
1170 pmd = pmd_alloc(mm, pud, addr);
1174 next = pmd_addr_end(addr, end);
1175 err = zeromap_pte_range(mm, pmd, addr, next, prot);
1178 } while (pmd++, addr = next, addr != end);
1182 static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1183 unsigned long addr, unsigned long end, pgprot_t prot)
1189 pud = pud_alloc(mm, pgd, addr);
1193 next = pud_addr_end(addr, end);
1194 err = zeromap_pmd_range(mm, pud, addr, next, prot);
1197 } while (pud++, addr = next, addr != end);
1201 int zeromap_page_range(struct vm_area_struct *vma,
1202 unsigned long addr, unsigned long size, pgprot_t prot)
1206 unsigned long end = addr + size;
1207 struct mm_struct *mm = vma->vm_mm;
1210 BUG_ON(addr >= end);
1211 pgd = pgd_offset(mm, addr);
1212 flush_cache_range(vma, addr, end);
1214 next = pgd_addr_end(addr, end);
1215 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1218 } while (pgd++, addr = next, addr != end);
1222 pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
1224 pgd_t * pgd = pgd_offset(mm, addr);
1225 pud_t * pud = pud_alloc(mm, pgd, addr);
1227 pmd_t * pmd = pmd_alloc(mm, pud, addr);
1229 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1235 * This is the old fallback for page remapping.
1237 * For historical reasons, it only allows reserved pages. Only
1238 * old drivers should use this, and they needed to mark their
1239 * pages reserved for the old functions anyway.
1241 static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1251 flush_dcache_page(page);
1252 pte = get_locked_pte(mm, addr, &ptl);
1256 if (!pte_none(*pte))
1259 /* Ok, finally just insert the thing.. */
1261 inc_mm_counter(mm, file_rss);
1262 page_add_file_rmap(page);
1263 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1267 pte_unmap_unlock(pte, ptl);
1273 * vm_insert_page - insert single page into user vma
1274 * @vma: user vma to map to
1275 * @addr: target user address of this page
1276 * @page: source kernel page
1278 * This allows drivers to insert individual pages they've allocated
1281 * The page has to be a nice clean _individual_ kernel allocation.
1282 * If you allocate a compound page, you need to have marked it as
1283 * such (__GFP_COMP), or manually just split the page up yourself
1284 * (see split_page()).
1286 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1287 * took an arbitrary page protection parameter. This doesn't allow
1288 * that. Your vma protection will have to be set up correctly, which
1289 * means that if you want a shared writable mapping, you'd better
1290 * ask for a shared writable mapping!
1292 * The page does not need to be reserved.
1294 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1296 if (addr < vma->vm_start || addr >= vma->vm_end)
1298 if (!page_count(page))
1300 vma->vm_flags |= VM_INSERTPAGE;
1301 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1303 EXPORT_SYMBOL(vm_insert_page);
1306 * maps a range of physical memory into the requested pages. the old
1307 * mappings are removed. any references to nonexistent pages results
1308 * in null mappings (currently treated as "copy-on-access")
1310 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1311 unsigned long addr, unsigned long end,
1312 unsigned long pfn, pgprot_t prot)
1317 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1320 arch_enter_lazy_mmu_mode();
1322 BUG_ON(!pte_none(*pte));
1323 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
1325 } while (pte++, addr += PAGE_SIZE, addr != end);
1326 arch_leave_lazy_mmu_mode();
1327 pte_unmap_unlock(pte - 1, ptl);
1331 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1332 unsigned long addr, unsigned long end,
1333 unsigned long pfn, pgprot_t prot)
1338 pfn -= addr >> PAGE_SHIFT;
1339 pmd = pmd_alloc(mm, pud, addr);
1343 next = pmd_addr_end(addr, end);
1344 if (remap_pte_range(mm, pmd, addr, next,
1345 pfn + (addr >> PAGE_SHIFT), prot))
1347 } while (pmd++, addr = next, addr != end);
1351 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1352 unsigned long addr, unsigned long end,
1353 unsigned long pfn, pgprot_t prot)
1358 pfn -= addr >> PAGE_SHIFT;
1359 pud = pud_alloc(mm, pgd, addr);
1363 next = pud_addr_end(addr, end);
1364 if (remap_pmd_range(mm, pud, addr, next,
1365 pfn + (addr >> PAGE_SHIFT), prot))
1367 } while (pud++, addr = next, addr != end);
1372 * remap_pfn_range - remap kernel memory to userspace
1373 * @vma: user vma to map to
1374 * @addr: target user address to start at
1375 * @pfn: physical address of kernel memory
1376 * @size: size of map area
1377 * @prot: page protection flags for this mapping
1379 * Note: this is only safe if the mm semaphore is held when called.
1381 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1382 unsigned long pfn, unsigned long size, pgprot_t prot)
1386 unsigned long end = addr + PAGE_ALIGN(size);
1387 struct mm_struct *mm = vma->vm_mm;
1391 * Physically remapped pages are special. Tell the
1392 * rest of the world about it:
1393 * VM_IO tells people not to look at these pages
1394 * (accesses can have side effects).
1395 * VM_RESERVED is specified all over the place, because
1396 * in 2.4 it kept swapout's vma scan off this vma; but
1397 * in 2.6 the LRU scan won't even find its pages, so this
1398 * flag means no more than count its pages in reserved_vm,
1399 * and omit it from core dump, even when VM_IO turned off.
1400 * VM_PFNMAP tells the core MM that the base pages are just
1401 * raw PFN mappings, and do not have a "struct page" associated
1404 * There's a horrible special case to handle copy-on-write
1405 * behaviour that some programs depend on. We mark the "original"
1406 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1408 if (is_cow_mapping(vma->vm_flags)) {
1409 if (addr != vma->vm_start || end != vma->vm_end)
1411 vma->vm_pgoff = pfn;
1414 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1416 BUG_ON(addr >= end);
1417 pfn -= addr >> PAGE_SHIFT;
1418 pgd = pgd_offset(mm, addr);
1419 flush_cache_range(vma, addr, end);
1421 next = pgd_addr_end(addr, end);
1422 err = remap_pud_range(mm, pgd, addr, next,
1423 pfn + (addr >> PAGE_SHIFT), prot);
1426 } while (pgd++, addr = next, addr != end);
1429 EXPORT_SYMBOL(remap_pfn_range);
1432 static inline int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1433 unsigned long addr, unsigned long end,
1434 pte_fn_t fn, void *data)
1438 struct page *pmd_page;
1441 pte = (mm == &init_mm) ?
1442 pte_alloc_kernel(pmd, addr) :
1443 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1447 BUG_ON(pmd_huge(*pmd));
1449 pmd_page = pmd_page(*pmd);
1452 err = fn(pte, pmd_page, addr, data);
1455 } while (pte++, addr += PAGE_SIZE, addr != end);
1458 pte_unmap_unlock(pte-1, ptl);
1462 static inline int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1463 unsigned long addr, unsigned long end,
1464 pte_fn_t fn, void *data)
1470 pmd = pmd_alloc(mm, pud, addr);
1474 next = pmd_addr_end(addr, end);
1475 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1478 } while (pmd++, addr = next, addr != end);
1482 static inline int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1483 unsigned long addr, unsigned long end,
1484 pte_fn_t fn, void *data)
1490 pud = pud_alloc(mm, pgd, addr);
1494 next = pud_addr_end(addr, end);
1495 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1498 } while (pud++, addr = next, addr != end);
1503 * Scan a region of virtual memory, filling in page tables as necessary
1504 * and calling a provided function on each leaf page table.
1506 int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1507 unsigned long size, pte_fn_t fn, void *data)
1511 unsigned long end = addr + size;
1514 BUG_ON(addr >= end);
1515 pgd = pgd_offset(mm, addr);
1517 next = pgd_addr_end(addr, end);
1518 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1521 } while (pgd++, addr = next, addr != end);
1524 EXPORT_SYMBOL_GPL(apply_to_page_range);
1528 * handle_pte_fault chooses page fault handler according to an entry
1529 * which was read non-atomically. Before making any commitment, on
1530 * those architectures or configurations (e.g. i386 with PAE) which
1531 * might give a mix of unmatched parts, do_swap_page and do_file_page
1532 * must check under lock before unmapping the pte and proceeding
1533 * (but do_wp_page is only called after already making such a check;
1534 * and do_anonymous_page and do_no_page can safely check later on).
1536 static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
1537 pte_t *page_table, pte_t orig_pte)
1540 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1541 if (sizeof(pte_t) > sizeof(unsigned long)) {
1542 spinlock_t *ptl = pte_lockptr(mm, pmd);
1544 same = pte_same(*page_table, orig_pte);
1548 pte_unmap(page_table);
1553 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1554 * servicing faults for write access. In the normal case, do always want
1555 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1556 * that do not have writing enabled, when used by access_process_vm.
1558 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1560 if (likely(vma->vm_flags & VM_WRITE))
1561 pte = pte_mkwrite(pte);
1565 static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
1568 * If the source page was a PFN mapping, we don't have
1569 * a "struct page" for it. We do a best-effort copy by
1570 * just copying from the original user address. If that
1571 * fails, we just zero-fill it. Live with it.
1573 if (unlikely(!src)) {
1574 void *kaddr = kmap_atomic(dst, KM_USER0);
1575 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1578 * This really shouldn't fail, because the page is there
1579 * in the page tables. But it might just be unreadable,
1580 * in which case we just give up and fill the result with
1583 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
1584 memset(kaddr, 0, PAGE_SIZE);
1585 kunmap_atomic(kaddr, KM_USER0);
1586 flush_dcache_page(dst);
1590 copy_user_highpage(dst, src, va, vma);
1594 * This routine handles present pages, when users try to write
1595 * to a shared page. It is done by copying the page to a new address
1596 * and decrementing the shared-page counter for the old page.
1598 * Note that this routine assumes that the protection checks have been
1599 * done by the caller (the low-level page fault routine in most cases).
1600 * Thus we can safely just mark it writable once we've done any necessary
1603 * We also mark the page dirty at this point even though the page will
1604 * change only once the write actually happens. This avoids a few races,
1605 * and potentially makes it more efficient.
1607 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1608 * but allow concurrent faults), with pte both mapped and locked.
1609 * We return with mmap_sem still held, but pte unmapped and unlocked.
1611 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1612 unsigned long address, pte_t *page_table, pmd_t *pmd,
1613 spinlock_t *ptl, pte_t orig_pte)
1615 struct page *old_page, *new_page;
1617 int reuse = 0, ret = VM_FAULT_MINOR;
1618 struct page *dirty_page = NULL;
1620 old_page = vm_normal_page(vma, address, orig_pte);
1625 * Take out anonymous pages first, anonymous shared vmas are
1626 * not dirty accountable.
1628 if (PageAnon(old_page)) {
1629 if (!TestSetPageLocked(old_page)) {
1630 reuse = can_share_swap_page(old_page);
1631 unlock_page(old_page);
1633 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
1634 (VM_WRITE|VM_SHARED))) {
1636 * Only catch write-faults on shared writable pages,
1637 * read-only shared pages can get COWed by
1638 * get_user_pages(.write=1, .force=1).
1640 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1642 * Notify the address space that the page is about to
1643 * become writable so that it can prohibit this or wait
1644 * for the page to get into an appropriate state.
1646 * We do this without the lock held, so that it can
1647 * sleep if it needs to.
1649 page_cache_get(old_page);
1650 pte_unmap_unlock(page_table, ptl);
1652 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1653 goto unwritable_page;
1655 page_cache_release(old_page);
1658 * Since we dropped the lock we need to revalidate
1659 * the PTE as someone else may have changed it. If
1660 * they did, we just return, as we can count on the
1661 * MMU to tell us if they didn't also make it writable.
1663 page_table = pte_offset_map_lock(mm, pmd, address,
1665 if (!pte_same(*page_table, orig_pte))
1668 dirty_page = old_page;
1669 get_page(dirty_page);
1674 flush_cache_page(vma, address, pte_pfn(orig_pte));
1675 entry = pte_mkyoung(orig_pte);
1676 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1677 ptep_set_access_flags(vma, address, page_table, entry, 1);
1678 update_mmu_cache(vma, address, entry);
1679 lazy_mmu_prot_update(entry);
1680 ret |= VM_FAULT_WRITE;
1685 * Ok, we need to copy. Oh, well..
1687 page_cache_get(old_page);
1689 pte_unmap_unlock(page_table, ptl);
1691 if (unlikely(anon_vma_prepare(vma)))
1693 if (old_page == ZERO_PAGE(address)) {
1694 new_page = alloc_zeroed_user_highpage(vma, address);
1698 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1701 cow_user_page(new_page, old_page, address, vma);
1705 * Re-check the pte - we dropped the lock
1707 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1708 if (likely(pte_same(*page_table, orig_pte))) {
1710 page_remove_rmap(old_page, vma);
1711 if (!PageAnon(old_page)) {
1712 dec_mm_counter(mm, file_rss);
1713 inc_mm_counter(mm, anon_rss);
1716 inc_mm_counter(mm, anon_rss);
1717 flush_cache_page(vma, address, pte_pfn(orig_pte));
1718 entry = mk_pte(new_page, vma->vm_page_prot);
1719 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1720 lazy_mmu_prot_update(entry);
1722 * Clear the pte entry and flush it first, before updating the
1723 * pte with the new entry. This will avoid a race condition
1724 * seen in the presence of one thread doing SMC and another
1727 ptep_clear_flush(vma, address, page_table);
1728 set_pte_at(mm, address, page_table, entry);
1729 update_mmu_cache(vma, address, entry);
1730 lru_cache_add_active(new_page);
1731 page_add_new_anon_rmap(new_page, vma, address);
1733 /* Free the old page.. */
1734 new_page = old_page;
1735 ret |= VM_FAULT_WRITE;
1738 page_cache_release(new_page);
1740 page_cache_release(old_page);
1742 pte_unmap_unlock(page_table, ptl);
1744 set_page_dirty_balance(dirty_page);
1745 put_page(dirty_page);
1750 page_cache_release(old_page);
1751 return VM_FAULT_OOM;
1754 page_cache_release(old_page);
1755 return VM_FAULT_SIGBUS;
1759 * Helper functions for unmap_mapping_range().
1761 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1763 * We have to restart searching the prio_tree whenever we drop the lock,
1764 * since the iterator is only valid while the lock is held, and anyway
1765 * a later vma might be split and reinserted earlier while lock dropped.
1767 * The list of nonlinear vmas could be handled more efficiently, using
1768 * a placeholder, but handle it in the same way until a need is shown.
1769 * It is important to search the prio_tree before nonlinear list: a vma
1770 * may become nonlinear and be shifted from prio_tree to nonlinear list
1771 * while the lock is dropped; but never shifted from list to prio_tree.
1773 * In order to make forward progress despite restarting the search,
1774 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1775 * quickly skip it next time around. Since the prio_tree search only
1776 * shows us those vmas affected by unmapping the range in question, we
1777 * can't efficiently keep all vmas in step with mapping->truncate_count:
1778 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1779 * mapping->truncate_count and vma->vm_truncate_count are protected by
1782 * In order to make forward progress despite repeatedly restarting some
1783 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1784 * and restart from that address when we reach that vma again. It might
1785 * have been split or merged, shrunk or extended, but never shifted: so
1786 * restart_addr remains valid so long as it remains in the vma's range.
1787 * unmap_mapping_range forces truncate_count to leap over page-aligned
1788 * values so we can save vma's restart_addr in its truncate_count field.
1790 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1792 static void reset_vma_truncate_counts(struct address_space *mapping)
1794 struct vm_area_struct *vma;
1795 struct prio_tree_iter iter;
1797 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1798 vma->vm_truncate_count = 0;
1799 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1800 vma->vm_truncate_count = 0;
1803 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1804 unsigned long start_addr, unsigned long end_addr,
1805 struct zap_details *details)
1807 unsigned long restart_addr;
1811 restart_addr = vma->vm_truncate_count;
1812 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1813 start_addr = restart_addr;
1814 if (start_addr >= end_addr) {
1815 /* Top of vma has been split off since last time */
1816 vma->vm_truncate_count = details->truncate_count;
1821 restart_addr = zap_page_range(vma, start_addr,
1822 end_addr - start_addr, details);
1823 need_break = need_resched() ||
1824 need_lockbreak(details->i_mmap_lock);
1826 if (restart_addr >= end_addr) {
1827 /* We have now completed this vma: mark it so */
1828 vma->vm_truncate_count = details->truncate_count;
1832 /* Note restart_addr in vma's truncate_count field */
1833 vma->vm_truncate_count = restart_addr;
1838 spin_unlock(details->i_mmap_lock);
1840 spin_lock(details->i_mmap_lock);
1844 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1845 struct zap_details *details)
1847 struct vm_area_struct *vma;
1848 struct prio_tree_iter iter;
1849 pgoff_t vba, vea, zba, zea;
1852 vma_prio_tree_foreach(vma, &iter, root,
1853 details->first_index, details->last_index) {
1854 /* Skip quickly over those we have already dealt with */
1855 if (vma->vm_truncate_count == details->truncate_count)
1858 vba = vma->vm_pgoff;
1859 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1860 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1861 zba = details->first_index;
1864 zea = details->last_index;
1868 if (unmap_mapping_range_vma(vma,
1869 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1870 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1876 static inline void unmap_mapping_range_list(struct list_head *head,
1877 struct zap_details *details)
1879 struct vm_area_struct *vma;
1882 * In nonlinear VMAs there is no correspondence between virtual address
1883 * offset and file offset. So we must perform an exhaustive search
1884 * across *all* the pages in each nonlinear VMA, not just the pages
1885 * whose virtual address lies outside the file truncation point.
1888 list_for_each_entry(vma, head, shared.vm_set.list) {
1889 /* Skip quickly over those we have already dealt with */
1890 if (vma->vm_truncate_count == details->truncate_count)
1892 details->nonlinear_vma = vma;
1893 if (unmap_mapping_range_vma(vma, vma->vm_start,
1894 vma->vm_end, details) < 0)
1900 * unmap_mapping_range - unmap the portion of all mmaps
1901 * in the specified address_space corresponding to the specified
1902 * page range in the underlying file.
1903 * @mapping: the address space containing mmaps to be unmapped.
1904 * @holebegin: byte in first page to unmap, relative to the start of
1905 * the underlying file. This will be rounded down to a PAGE_SIZE
1906 * boundary. Note that this is different from vmtruncate(), which
1907 * must keep the partial page. In contrast, we must get rid of
1909 * @holelen: size of prospective hole in bytes. This will be rounded
1910 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1912 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1913 * but 0 when invalidating pagecache, don't throw away private data.
1915 void unmap_mapping_range(struct address_space *mapping,
1916 loff_t const holebegin, loff_t const holelen, int even_cows)
1918 struct zap_details details;
1919 pgoff_t hba = holebegin >> PAGE_SHIFT;
1920 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1922 /* Check for overflow. */
1923 if (sizeof(holelen) > sizeof(hlen)) {
1925 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1926 if (holeend & ~(long long)ULONG_MAX)
1927 hlen = ULONG_MAX - hba + 1;
1930 details.check_mapping = even_cows? NULL: mapping;
1931 details.nonlinear_vma = NULL;
1932 details.first_index = hba;
1933 details.last_index = hba + hlen - 1;
1934 if (details.last_index < details.first_index)
1935 details.last_index = ULONG_MAX;
1936 details.i_mmap_lock = &mapping->i_mmap_lock;
1938 spin_lock(&mapping->i_mmap_lock);
1940 /* serialize i_size write against truncate_count write */
1942 /* Protect against page faults, and endless unmapping loops */
1943 mapping->truncate_count++;
1945 * For archs where spin_lock has inclusive semantics like ia64
1946 * this smp_mb() will prevent to read pagetable contents
1947 * before the truncate_count increment is visible to
1951 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1952 if (mapping->truncate_count == 0)
1953 reset_vma_truncate_counts(mapping);
1954 mapping->truncate_count++;
1956 details.truncate_count = mapping->truncate_count;
1958 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1959 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1960 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1961 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1962 spin_unlock(&mapping->i_mmap_lock);
1964 EXPORT_SYMBOL(unmap_mapping_range);
1967 * vmtruncate - unmap mappings "freed" by truncate() syscall
1968 * @inode: inode of the file used
1969 * @offset: file offset to start truncating
1971 * NOTE! We have to be ready to update the memory sharing
1972 * between the file and the memory map for a potential last
1973 * incomplete page. Ugly, but necessary.
1975 int vmtruncate(struct inode * inode, loff_t offset)
1977 struct address_space *mapping = inode->i_mapping;
1978 unsigned long limit;
1980 if (inode->i_size < offset)
1983 * truncation of in-use swapfiles is disallowed - it would cause
1984 * subsequent swapout to scribble on the now-freed blocks.
1986 if (IS_SWAPFILE(inode))
1988 i_size_write(inode, offset);
1989 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1990 truncate_inode_pages(mapping, offset);
1994 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1995 if (limit != RLIM_INFINITY && offset > limit)
1997 if (offset > inode->i_sb->s_maxbytes)
1999 i_size_write(inode, offset);
2002 if (inode->i_op && inode->i_op->truncate)
2003 inode->i_op->truncate(inode);
2006 send_sig(SIGXFSZ, current, 0);
2012 EXPORT_SYMBOL(vmtruncate);
2014 int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2016 struct address_space *mapping = inode->i_mapping;
2019 * If the underlying filesystem is not going to provide
2020 * a way to truncate a range of blocks (punch a hole) -
2021 * we should return failure right now.
2023 if (!inode->i_op || !inode->i_op->truncate_range)
2026 mutex_lock(&inode->i_mutex);
2027 down_write(&inode->i_alloc_sem);
2028 unmap_mapping_range(mapping, offset, (end - offset), 1);
2029 truncate_inode_pages_range(mapping, offset, end);
2030 inode->i_op->truncate_range(inode, offset, end);
2031 up_write(&inode->i_alloc_sem);
2032 mutex_unlock(&inode->i_mutex);
2038 * swapin_readahead - swap in pages in hope we need them soon
2039 * @entry: swap entry of this memory
2040 * @addr: address to start
2041 * @vma: user vma this addresses belong to
2043 * Primitive swap readahead code. We simply read an aligned block of
2044 * (1 << page_cluster) entries in the swap area. This method is chosen
2045 * because it doesn't cost us any seek time. We also make sure to queue
2046 * the 'original' request together with the readahead ones...
2048 * This has been extended to use the NUMA policies from the mm triggering
2051 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
2053 void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
2056 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
2059 struct page *new_page;
2060 unsigned long offset;
2063 * Get the number of handles we should do readahead io to.
2065 num = valid_swaphandles(entry, &offset);
2066 for (i = 0; i < num; offset++, i++) {
2067 /* Ok, do the async read-ahead now */
2068 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
2069 offset), vma, addr);
2072 page_cache_release(new_page);
2075 * Find the next applicable VMA for the NUMA policy.
2081 if (addr >= vma->vm_end) {
2083 next_vma = vma ? vma->vm_next : NULL;
2085 if (vma && addr < vma->vm_start)
2088 if (next_vma && addr >= next_vma->vm_start) {
2090 next_vma = vma->vm_next;
2095 lru_add_drain(); /* Push any new pages onto the LRU now */
2099 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2100 * but allow concurrent faults), and pte mapped but not yet locked.
2101 * We return with mmap_sem still held, but pte unmapped and unlocked.
2103 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2104 unsigned long address, pte_t *page_table, pmd_t *pmd,
2105 int write_access, pte_t orig_pte)
2111 int ret = VM_FAULT_MINOR;
2113 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2116 entry = pte_to_swp_entry(orig_pte);
2117 if (is_migration_entry(entry)) {
2118 migration_entry_wait(mm, pmd, address);
2121 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
2122 page = lookup_swap_cache(entry);
2124 grab_swap_token(); /* Contend for token _before_ read-in */
2125 swapin_readahead(entry, address, vma);
2126 page = read_swap_cache_async(entry, vma, address);
2129 * Back out if somebody else faulted in this pte
2130 * while we released the pte lock.
2132 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2133 if (likely(pte_same(*page_table, orig_pte)))
2135 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2139 /* Had to read the page from swap area: Major fault */
2140 ret = VM_FAULT_MAJOR;
2141 count_vm_event(PGMAJFAULT);
2144 if (!vx_rss_avail(mm, 1)) {
2149 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2150 mark_page_accessed(page);
2154 * Back out if somebody else already faulted in this pte.
2156 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2157 if (unlikely(!pte_same(*page_table, orig_pte)))
2160 if (unlikely(!PageUptodate(page))) {
2161 ret = VM_FAULT_SIGBUS;
2165 /* The page isn't present yet, go ahead with the fault. */
2167 inc_mm_counter(mm, anon_rss);
2168 pte = mk_pte(page, vma->vm_page_prot);
2169 if (write_access && can_share_swap_page(page)) {
2170 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2174 flush_icache_page(vma, page);
2175 set_pte_at(mm, address, page_table, pte);
2176 page_add_anon_rmap(page, vma, address);
2180 remove_exclusive_swap_page(page);
2184 if (do_wp_page(mm, vma, address,
2185 page_table, pmd, ptl, pte) == VM_FAULT_OOM)
2190 /* No need to invalidate - it was non-present before */
2191 update_mmu_cache(vma, address, pte);
2192 lazy_mmu_prot_update(pte);
2194 pte_unmap_unlock(page_table, ptl);
2198 pte_unmap_unlock(page_table, ptl);
2200 page_cache_release(page);
2205 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2206 * but allow concurrent faults), and pte mapped but not yet locked.
2207 * We return with mmap_sem still held, but pte unmapped and unlocked.
2209 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2210 unsigned long address, pte_t *page_table, pmd_t *pmd,
2218 /* Allocate our own private page. */
2219 pte_unmap(page_table);
2221 if (!vx_rss_avail(mm, 1))
2223 if (unlikely(anon_vma_prepare(vma)))
2225 page = alloc_zeroed_user_highpage(vma, address);
2229 entry = mk_pte(page, vma->vm_page_prot);
2230 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2232 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2233 if (!pte_none(*page_table))
2235 inc_mm_counter(mm, anon_rss);
2236 lru_cache_add_active(page);
2237 page_add_new_anon_rmap(page, vma, address);
2239 /* Map the ZERO_PAGE - vm_page_prot is readonly */
2240 page = ZERO_PAGE(address);
2241 page_cache_get(page);
2242 entry = mk_pte(page, vma->vm_page_prot);
2244 ptl = pte_lockptr(mm, pmd);
2246 if (!pte_none(*page_table))
2248 inc_mm_counter(mm, file_rss);
2249 page_add_file_rmap(page);
2252 set_pte_at(mm, address, page_table, entry);
2254 /* No need to invalidate - it was non-present before */
2255 update_mmu_cache(vma, address, entry);
2256 lazy_mmu_prot_update(entry);
2258 pte_unmap_unlock(page_table, ptl);
2259 return VM_FAULT_MINOR;
2261 page_cache_release(page);
2264 return VM_FAULT_OOM;
2268 * do_no_page() tries to create a new page mapping. It aggressively
2269 * tries to share with existing pages, but makes a separate copy if
2270 * the "write_access" parameter is true in order to avoid the next
2273 * As this is called only for pages that do not currently exist, we
2274 * do not need to flush old virtual caches or the TLB.
2276 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2277 * but allow concurrent faults), and pte mapped but not yet locked.
2278 * We return with mmap_sem still held, but pte unmapped and unlocked.
2280 static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
2281 unsigned long address, pte_t *page_table, pmd_t *pmd,
2285 struct page *new_page;
2286 struct address_space *mapping = NULL;
2288 unsigned int sequence = 0;
2289 int ret = VM_FAULT_MINOR;
2291 struct page *dirty_page = NULL;
2293 pte_unmap(page_table);
2294 BUG_ON(vma->vm_flags & VM_PFNMAP);
2296 if (!vx_rss_avail(mm, 1))
2297 return VM_FAULT_OOM;
2300 mapping = vma->vm_file->f_mapping;
2301 sequence = mapping->truncate_count;
2302 smp_rmb(); /* serializes i_size against truncate_count */
2305 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
2307 * No smp_rmb is needed here as long as there's a full
2308 * spin_lock/unlock sequence inside the ->nopage callback
2309 * (for the pagecache lookup) that acts as an implicit
2310 * smp_mb() and prevents the i_size read to happen
2311 * after the next truncate_count read.
2314 /* no page was available -- either SIGBUS, OOM or REFAULT */
2315 if (unlikely(new_page == NOPAGE_SIGBUS))
2316 return VM_FAULT_SIGBUS;
2317 else if (unlikely(new_page == NOPAGE_OOM))
2318 return VM_FAULT_OOM;
2319 else if (unlikely(new_page == NOPAGE_REFAULT))
2320 return VM_FAULT_MINOR;
2323 * Should we do an early C-O-W break?
2326 if (!(vma->vm_flags & VM_SHARED)) {
2329 if (unlikely(anon_vma_prepare(vma)))
2331 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
2334 copy_user_highpage(page, new_page, address, vma);
2335 page_cache_release(new_page);
2340 /* if the page will be shareable, see if the backing
2341 * address space wants to know that the page is about
2342 * to become writable */
2343 if (vma->vm_ops->page_mkwrite &&
2344 vma->vm_ops->page_mkwrite(vma, new_page) < 0
2346 page_cache_release(new_page);
2347 return VM_FAULT_SIGBUS;
2352 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2354 * For a file-backed vma, someone could have truncated or otherwise
2355 * invalidated this page. If unmap_mapping_range got called,
2356 * retry getting the page.
2358 if (mapping && unlikely(sequence != mapping->truncate_count)) {
2359 pte_unmap_unlock(page_table, ptl);
2360 page_cache_release(new_page);
2362 sequence = mapping->truncate_count;
2368 * This silly early PAGE_DIRTY setting removes a race
2369 * due to the bad i386 page protection. But it's valid
2370 * for other architectures too.
2372 * Note that if write_access is true, we either now have
2373 * an exclusive copy of the page, or this is a shared mapping,
2374 * so we can make it writable and dirty to avoid having to
2375 * handle that later.
2377 /* Only go through if we didn't race with anybody else... */
2378 if (pte_none(*page_table)) {
2379 flush_icache_page(vma, new_page);
2380 entry = mk_pte(new_page, vma->vm_page_prot);
2382 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2383 set_pte_at(mm, address, page_table, entry);
2385 inc_mm_counter(mm, anon_rss);
2386 lru_cache_add_active(new_page);
2387 page_add_new_anon_rmap(new_page, vma, address);
2389 inc_mm_counter(mm, file_rss);
2390 page_add_file_rmap(new_page);
2392 dirty_page = new_page;
2393 get_page(dirty_page);
2397 /* One of our sibling threads was faster, back out. */
2398 page_cache_release(new_page);
2402 /* no need to invalidate: a not-present page shouldn't be cached */
2403 update_mmu_cache(vma, address, entry);
2404 lazy_mmu_prot_update(entry);
2406 pte_unmap_unlock(page_table, ptl);
2408 set_page_dirty_balance(dirty_page);
2409 put_page(dirty_page);
2413 page_cache_release(new_page);
2414 return VM_FAULT_OOM;
2418 * do_no_pfn() tries to create a new page mapping for a page without
2419 * a struct_page backing it
2421 * As this is called only for pages that do not currently exist, we
2422 * do not need to flush old virtual caches or the TLB.
2424 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2425 * but allow concurrent faults), and pte mapped but not yet locked.
2426 * We return with mmap_sem still held, but pte unmapped and unlocked.
2428 * It is expected that the ->nopfn handler always returns the same pfn
2429 * for a given virtual mapping.
2431 * Mark this `noinline' to prevent it from bloating the main pagefault code.
2433 static noinline int do_no_pfn(struct mm_struct *mm, struct vm_area_struct *vma,
2434 unsigned long address, pte_t *page_table, pmd_t *pmd,
2440 int ret = VM_FAULT_MINOR;
2442 pte_unmap(page_table);
2443 BUG_ON(!(vma->vm_flags & VM_PFNMAP));
2444 BUG_ON(is_cow_mapping(vma->vm_flags));
2446 pfn = vma->vm_ops->nopfn(vma, address & PAGE_MASK);
2447 if (pfn == NOPFN_OOM)
2448 return VM_FAULT_OOM;
2449 if (pfn == NOPFN_SIGBUS)
2450 return VM_FAULT_SIGBUS;
2452 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2454 /* Only go through if we didn't race with anybody else... */
2455 if (pte_none(*page_table)) {
2456 entry = pfn_pte(pfn, vma->vm_page_prot);
2458 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2459 set_pte_at(mm, address, page_table, entry);
2461 pte_unmap_unlock(page_table, ptl);
2466 * Fault of a previously existing named mapping. Repopulate the pte
2467 * from the encoded file_pte if possible. This enables swappable
2470 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2471 * but allow concurrent faults), and pte mapped but not yet locked.
2472 * We return with mmap_sem still held, but pte unmapped and unlocked.
2474 static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
2475 unsigned long address, pte_t *page_table, pmd_t *pmd,
2476 int write_access, pte_t orig_pte)
2481 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2482 return VM_FAULT_MINOR;
2484 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
2486 * Page table corrupted: show pte and kill process.
2488 print_bad_pte(vma, orig_pte, address);
2489 return VM_FAULT_OOM;
2491 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2493 pgoff = pte_to_pgoff(orig_pte);
2494 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
2495 vma->vm_page_prot, pgoff, 0);
2497 return VM_FAULT_OOM;
2499 return VM_FAULT_SIGBUS;
2500 return VM_FAULT_MAJOR;
2504 * These routines also need to handle stuff like marking pages dirty
2505 * and/or accessed for architectures that don't do it in hardware (most
2506 * RISC architectures). The early dirtying is also good on the i386.
2508 * There is also a hook called "update_mmu_cache()" that architectures
2509 * with external mmu caches can use to update those (ie the Sparc or
2510 * PowerPC hashed page tables that act as extended TLBs).
2512 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2513 * but allow concurrent faults), and pte mapped but not yet locked.
2514 * We return with mmap_sem still held, but pte unmapped and unlocked.
2516 static inline int handle_pte_fault(struct mm_struct *mm,
2517 struct vm_area_struct *vma, unsigned long address,
2518 pte_t *pte, pmd_t *pmd, int write_access)
2523 int ret, type = VXPT_UNKNOWN;
2525 old_entry = entry = *pte;
2526 if (!pte_present(entry)) {
2527 if (pte_none(entry)) {
2529 if (vma->vm_ops->nopage)
2530 return do_no_page(mm, vma, address,
2533 if (unlikely(vma->vm_ops->nopfn))
2534 return do_no_pfn(mm, vma, address, pte,
2537 return do_anonymous_page(mm, vma, address,
2538 pte, pmd, write_access);
2540 if (pte_file(entry))
2541 return do_file_page(mm, vma, address,
2542 pte, pmd, write_access, entry);
2543 return do_swap_page(mm, vma, address,
2544 pte, pmd, write_access, entry);
2547 ptl = pte_lockptr(mm, pmd);
2549 if (unlikely(!pte_same(*pte, entry)))
2552 if (!pte_write(entry)) {
2553 ret = do_wp_page(mm, vma, address,
2554 pte, pmd, ptl, entry);
2558 entry = pte_mkdirty(entry);
2560 entry = pte_mkyoung(entry);
2561 if (!pte_same(old_entry, entry)) {
2562 ptep_set_access_flags(vma, address, pte, entry, write_access);
2563 update_mmu_cache(vma, address, entry);
2564 lazy_mmu_prot_update(entry);
2567 * This is needed only for protection faults but the arch code
2568 * is not yet telling us if this is a protection fault or not.
2569 * This still avoids useless tlb flushes for .text page faults
2573 flush_tlb_page(vma, address);
2576 pte_unmap_unlock(pte, ptl);
2577 ret = VM_FAULT_MINOR;
2579 vx_page_fault(mm, vma, type, ret);
2584 * By the time we get here, we already hold the mm semaphore
2586 int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2587 unsigned long address, int write_access)
2594 __set_current_state(TASK_RUNNING);
2596 count_vm_event(PGFAULT);
2598 if (unlikely(is_vm_hugetlb_page(vma)))
2599 return hugetlb_fault(mm, vma, address, write_access);
2601 pgd = pgd_offset(mm, address);
2602 pud = pud_alloc(mm, pgd, address);
2604 return VM_FAULT_OOM;
2605 pmd = pmd_alloc(mm, pud, address);
2607 return VM_FAULT_OOM;
2608 pte = pte_alloc_map(mm, pmd, address);
2610 return VM_FAULT_OOM;
2612 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
2615 EXPORT_SYMBOL_GPL(__handle_mm_fault);
2617 #ifndef __PAGETABLE_PUD_FOLDED
2619 * Allocate page upper directory.
2620 * We've already handled the fast-path in-line.
2622 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2624 pud_t *new = pud_alloc_one(mm, address);
2628 spin_lock(&mm->page_table_lock);
2629 if (pgd_present(*pgd)) /* Another has populated it */
2632 pgd_populate(mm, pgd, new);
2633 spin_unlock(&mm->page_table_lock);
2637 /* Workaround for gcc 2.96 */
2638 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2642 #endif /* __PAGETABLE_PUD_FOLDED */
2644 #ifndef __PAGETABLE_PMD_FOLDED
2646 * Allocate page middle directory.
2647 * We've already handled the fast-path in-line.
2649 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2651 pmd_t *new = pmd_alloc_one(mm, address);
2655 spin_lock(&mm->page_table_lock);
2656 #ifndef __ARCH_HAS_4LEVEL_HACK
2657 if (pud_present(*pud)) /* Another has populated it */
2660 pud_populate(mm, pud, new);
2662 if (pgd_present(*pud)) /* Another has populated it */
2665 pgd_populate(mm, pud, new);
2666 #endif /* __ARCH_HAS_4LEVEL_HACK */
2667 spin_unlock(&mm->page_table_lock);
2671 /* Workaround for gcc 2.96 */
2672 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2676 #endif /* __PAGETABLE_PMD_FOLDED */
2678 int make_pages_present(unsigned long addr, unsigned long end)
2680 int ret, len, write;
2681 struct vm_area_struct * vma;
2683 vma = find_vma(current->mm, addr);
2686 write = (vma->vm_flags & VM_WRITE) != 0;
2687 BUG_ON(addr >= end);
2688 BUG_ON(end > vma->vm_end);
2689 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2690 ret = get_user_pages(current, current->mm, addr,
2691 len, write, 0, NULL, NULL);
2694 return ret == len ? 0 : -1;
2698 * Map a vmalloc()-space virtual address to the physical page.
2700 struct page * vmalloc_to_page(void * vmalloc_addr)
2702 unsigned long addr = (unsigned long) vmalloc_addr;
2703 struct page *page = NULL;
2704 pgd_t *pgd = pgd_offset_k(addr);
2709 if (!pgd_none(*pgd)) {
2710 pud = pud_offset(pgd, addr);
2711 if (!pud_none(*pud)) {
2712 pmd = pmd_offset(pud, addr);
2713 if (!pmd_none(*pmd)) {
2714 ptep = pte_offset_map(pmd, addr);
2716 if (pte_present(pte))
2717 page = pte_page(pte);
2725 EXPORT_SYMBOL(vmalloc_to_page);
2728 * Map a vmalloc()-space virtual address to the physical page frame number.
2730 unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2732 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2735 EXPORT_SYMBOL(vmalloc_to_pfn);
2737 #if !defined(__HAVE_ARCH_GATE_AREA)
2739 #if defined(AT_SYSINFO_EHDR)
2740 static struct vm_area_struct gate_vma;
2742 static int __init gate_vma_init(void)
2744 gate_vma.vm_mm = NULL;
2745 gate_vma.vm_start = FIXADDR_USER_START;
2746 gate_vma.vm_end = FIXADDR_USER_END;
2747 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
2748 gate_vma.vm_page_prot = __P101;
2750 * Make sure the vDSO gets into every core dump.
2751 * Dumping its contents makes post-mortem fully interpretable later
2752 * without matching up the same kernel and hardware config to see
2753 * what PC values meant.
2755 gate_vma.vm_flags |= VM_ALWAYSDUMP;
2758 __initcall(gate_vma_init);
2761 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2763 #ifdef AT_SYSINFO_EHDR
2770 int in_gate_area_no_task(unsigned long addr)
2772 #ifdef AT_SYSINFO_EHDR
2773 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2779 #endif /* __HAVE_ARCH_GATE_AREA */
2782 * Access another process' address space.
2783 * Source/target buffer must be kernel space,
2784 * Do not walk the page table directly, use get_user_pages
2786 int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
2788 struct mm_struct *mm;
2789 struct vm_area_struct *vma;
2791 void *old_buf = buf;
2793 mm = get_task_mm(tsk);
2797 down_read(&mm->mmap_sem);
2798 /* ignore errors, just check how much was sucessfully transfered */
2800 int bytes, ret, offset;
2803 ret = get_user_pages(tsk, mm, addr, 1,
2804 write, 1, &page, &vma);
2809 offset = addr & (PAGE_SIZE-1);
2810 if (bytes > PAGE_SIZE-offset)
2811 bytes = PAGE_SIZE-offset;
2815 copy_to_user_page(vma, page, addr,
2816 maddr + offset, buf, bytes);
2817 set_page_dirty_lock(page);
2819 copy_from_user_page(vma, page, addr,
2820 buf, maddr + offset, bytes);
2823 page_cache_release(page);
2828 up_read(&mm->mmap_sem);
2831 return buf - old_buf;