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/ksm.h>
49 #include <linux/rmap.h>
50 #include <linux/export.h>
51 #include <linux/delayacct.h>
52 #include <linux/init.h>
53 #include <linux/pfn_t.h>
54 #include <linux/writeback.h>
55 #include <linux/memcontrol.h>
56 #include <linux/mmu_notifier.h>
57 #include <linux/kallsyms.h>
58 #include <linux/swapops.h>
59 #include <linux/elf.h>
60 #include <linux/gfp.h>
61 #include <linux/migrate.h>
62 #include <linux/string.h>
63 #include <linux/dma-debug.h>
64 #include <linux/debugfs.h>
65 #include <linux/userfaultfd_k.h>
68 #include <asm/pgalloc.h>
69 #include <asm/uaccess.h>
71 #include <asm/tlbflush.h>
72 #include <asm/pgtable.h>
76 #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
77 #warning Unfortunate NUMA and NUMA Balancing config, growing page-frame for last_cpupid.
80 #ifndef CONFIG_NEED_MULTIPLE_NODES
81 /* use the per-pgdat data instead for discontigmem - mbligh */
82 unsigned long max_mapnr
;
85 EXPORT_SYMBOL(max_mapnr
);
86 EXPORT_SYMBOL(mem_map
);
90 * A number of key systems in x86 including ioremap() rely on the assumption
91 * that high_memory defines the upper bound on direct map memory, then end
92 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
93 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
98 EXPORT_SYMBOL(high_memory
);
101 * Randomize the address space (stacks, mmaps, brk, etc.).
103 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
104 * as ancient (libc5 based) binaries can segfault. )
106 int randomize_va_space __read_mostly
=
107 #ifdef CONFIG_COMPAT_BRK
113 static int __init
disable_randmaps(char *s
)
115 randomize_va_space
= 0;
118 __setup("norandmaps", disable_randmaps
);
120 unsigned long zero_pfn __read_mostly
;
121 unsigned long highest_memmap_pfn __read_mostly
;
123 EXPORT_SYMBOL(zero_pfn
);
126 * CONFIG_MMU architectures set up ZERO_PAGE in their paging_init()
128 static int __init
init_zero_pfn(void)
130 zero_pfn
= page_to_pfn(ZERO_PAGE(0));
133 core_initcall(init_zero_pfn
);
136 #if defined(SPLIT_RSS_COUNTING)
138 void sync_mm_rss(struct mm_struct
*mm
)
142 for (i
= 0; i
< NR_MM_COUNTERS
; i
++) {
143 if (current
->rss_stat
.count
[i
]) {
144 add_mm_counter(mm
, i
, current
->rss_stat
.count
[i
]);
145 current
->rss_stat
.count
[i
] = 0;
148 current
->rss_stat
.events
= 0;
151 static void add_mm_counter_fast(struct mm_struct
*mm
, int member
, int val
)
153 struct task_struct
*task
= current
;
155 if (likely(task
->mm
== mm
))
156 task
->rss_stat
.count
[member
] += val
;
158 add_mm_counter(mm
, member
, val
);
160 #define inc_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, 1)
161 #define dec_mm_counter_fast(mm, member) add_mm_counter_fast(mm, member, -1)
163 /* sync counter once per 64 page faults */
164 #define TASK_RSS_EVENTS_THRESH (64)
165 static void check_sync_rss_stat(struct task_struct
*task
)
167 if (unlikely(task
!= current
))
169 if (unlikely(task
->rss_stat
.events
++ > TASK_RSS_EVENTS_THRESH
))
170 sync_mm_rss(task
->mm
);
172 #else /* SPLIT_RSS_COUNTING */
174 #define inc_mm_counter_fast(mm, member) inc_mm_counter(mm, member)
175 #define dec_mm_counter_fast(mm, member) dec_mm_counter(mm, member)
177 static void check_sync_rss_stat(struct task_struct
*task
)
181 #endif /* SPLIT_RSS_COUNTING */
183 #ifdef HAVE_GENERIC_MMU_GATHER
185 static bool tlb_next_batch(struct mmu_gather
*tlb
)
187 struct mmu_gather_batch
*batch
;
191 tlb
->active
= batch
->next
;
195 if (tlb
->batch_count
== MAX_GATHER_BATCH_COUNT
)
198 batch
= (void *)__get_free_pages(GFP_NOWAIT
| __GFP_NOWARN
, 0);
205 batch
->max
= MAX_GATHER_BATCH
;
207 tlb
->active
->next
= batch
;
214 * Called to initialize an (on-stack) mmu_gather structure for page-table
215 * tear-down from @mm. The @fullmm argument is used when @mm is without
216 * users and we're going to destroy the full address space (exit/execve).
218 void tlb_gather_mmu(struct mmu_gather
*tlb
, struct mm_struct
*mm
, unsigned long start
, unsigned long end
)
222 /* Is it from 0 to ~0? */
223 tlb
->fullmm
= !(start
| (end
+1));
224 tlb
->need_flush_all
= 0;
225 tlb
->local
.next
= NULL
;
227 tlb
->local
.max
= ARRAY_SIZE(tlb
->__pages
);
228 tlb
->active
= &tlb
->local
;
229 tlb
->batch_count
= 0;
231 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
235 __tlb_reset_range(tlb
);
238 static void tlb_flush_mmu_tlbonly(struct mmu_gather
*tlb
)
244 mmu_notifier_invalidate_range(tlb
->mm
, tlb
->start
, tlb
->end
);
245 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
246 tlb_table_flush(tlb
);
248 __tlb_reset_range(tlb
);
251 static void tlb_flush_mmu_free(struct mmu_gather
*tlb
)
253 struct mmu_gather_batch
*batch
;
255 for (batch
= &tlb
->local
; batch
&& batch
->nr
; batch
= batch
->next
) {
256 free_pages_and_swap_cache(batch
->pages
, batch
->nr
);
259 tlb
->active
= &tlb
->local
;
262 void tlb_flush_mmu(struct mmu_gather
*tlb
)
264 tlb_flush_mmu_tlbonly(tlb
);
265 tlb_flush_mmu_free(tlb
);
269 * Called at the end of the shootdown operation to free up any resources
270 * that were required.
272 void tlb_finish_mmu(struct mmu_gather
*tlb
, unsigned long start
, unsigned long end
)
274 struct mmu_gather_batch
*batch
, *next
;
278 /* keep the page table cache within bounds */
281 for (batch
= tlb
->local
.next
; batch
; batch
= next
) {
283 free_pages((unsigned long)batch
, 0);
285 tlb
->local
.next
= NULL
;
289 * Must perform the equivalent to __free_pte(pte_get_and_clear(ptep)), while
290 * handling the additional races in SMP caused by other CPUs caching valid
291 * mappings in their TLBs. Returns the number of free page slots left.
292 * When out of page slots we must call tlb_flush_mmu().
294 int __tlb_remove_page(struct mmu_gather
*tlb
, struct page
*page
)
296 struct mmu_gather_batch
*batch
;
298 VM_BUG_ON(!tlb
->end
);
301 batch
->pages
[batch
->nr
++] = page
;
302 if (batch
->nr
== batch
->max
) {
303 if (!tlb_next_batch(tlb
))
307 VM_BUG_ON_PAGE(batch
->nr
> batch
->max
, page
);
309 return batch
->max
- batch
->nr
;
312 #endif /* HAVE_GENERIC_MMU_GATHER */
314 #ifdef CONFIG_HAVE_RCU_TABLE_FREE
317 * See the comment near struct mmu_table_batch.
320 static void tlb_remove_table_smp_sync(void *arg
)
322 /* Simply deliver the interrupt */
325 static void tlb_remove_table_one(void *table
)
328 * This isn't an RCU grace period and hence the page-tables cannot be
329 * assumed to be actually RCU-freed.
331 * It is however sufficient for software page-table walkers that rely on
332 * IRQ disabling. See the comment near struct mmu_table_batch.
334 smp_call_function(tlb_remove_table_smp_sync
, NULL
, 1);
335 __tlb_remove_table(table
);
338 static void tlb_remove_table_rcu(struct rcu_head
*head
)
340 struct mmu_table_batch
*batch
;
343 batch
= container_of(head
, struct mmu_table_batch
, rcu
);
345 for (i
= 0; i
< batch
->nr
; i
++)
346 __tlb_remove_table(batch
->tables
[i
]);
348 free_page((unsigned long)batch
);
351 void tlb_table_flush(struct mmu_gather
*tlb
)
353 struct mmu_table_batch
**batch
= &tlb
->batch
;
356 call_rcu_sched(&(*batch
)->rcu
, tlb_remove_table_rcu
);
361 void tlb_remove_table(struct mmu_gather
*tlb
, void *table
)
363 struct mmu_table_batch
**batch
= &tlb
->batch
;
366 * When there's less then two users of this mm there cannot be a
367 * concurrent page-table walk.
369 if (atomic_read(&tlb
->mm
->mm_users
) < 2) {
370 __tlb_remove_table(table
);
374 if (*batch
== NULL
) {
375 *batch
= (struct mmu_table_batch
*)__get_free_page(GFP_NOWAIT
| __GFP_NOWARN
);
376 if (*batch
== NULL
) {
377 tlb_remove_table_one(table
);
382 (*batch
)->tables
[(*batch
)->nr
++] = table
;
383 if ((*batch
)->nr
== MAX_TABLE_BATCH
)
384 tlb_table_flush(tlb
);
387 #endif /* CONFIG_HAVE_RCU_TABLE_FREE */
390 * Note: this doesn't free the actual pages themselves. That
391 * has been handled earlier when unmapping all the memory regions.
393 static void free_pte_range(struct mmu_gather
*tlb
, pmd_t
*pmd
,
396 pgtable_t token
= pmd_pgtable(*pmd
);
398 pte_free_tlb(tlb
, token
, addr
);
399 atomic_long_dec(&tlb
->mm
->nr_ptes
);
402 static inline void free_pmd_range(struct mmu_gather
*tlb
, pud_t
*pud
,
403 unsigned long addr
, unsigned long end
,
404 unsigned long floor
, unsigned long ceiling
)
411 pmd
= pmd_offset(pud
, addr
);
413 next
= pmd_addr_end(addr
, end
);
414 if (pmd_none_or_clear_bad(pmd
))
416 free_pte_range(tlb
, pmd
, addr
);
417 } while (pmd
++, addr
= next
, addr
!= end
);
427 if (end
- 1 > ceiling
- 1)
430 pmd
= pmd_offset(pud
, start
);
432 pmd_free_tlb(tlb
, pmd
, start
);
433 mm_dec_nr_pmds(tlb
->mm
);
436 static inline void free_pud_range(struct mmu_gather
*tlb
, pgd_t
*pgd
,
437 unsigned long addr
, unsigned long end
,
438 unsigned long floor
, unsigned long ceiling
)
445 pud
= pud_offset(pgd
, addr
);
447 next
= pud_addr_end(addr
, end
);
448 if (pud_none_or_clear_bad(pud
))
450 free_pmd_range(tlb
, pud
, addr
, next
, floor
, ceiling
);
451 } while (pud
++, addr
= next
, addr
!= end
);
457 ceiling
&= PGDIR_MASK
;
461 if (end
- 1 > ceiling
- 1)
464 pud
= pud_offset(pgd
, start
);
466 pud_free_tlb(tlb
, pud
, start
);
470 * This function frees user-level page tables of a process.
472 void free_pgd_range(struct mmu_gather
*tlb
,
473 unsigned long addr
, unsigned long end
,
474 unsigned long floor
, unsigned long ceiling
)
480 * The next few lines have given us lots of grief...
482 * Why are we testing PMD* at this top level? Because often
483 * there will be no work to do at all, and we'd prefer not to
484 * go all the way down to the bottom just to discover that.
486 * Why all these "- 1"s? Because 0 represents both the bottom
487 * of the address space and the top of it (using -1 for the
488 * top wouldn't help much: the masks would do the wrong thing).
489 * The rule is that addr 0 and floor 0 refer to the bottom of
490 * the address space, but end 0 and ceiling 0 refer to the top
491 * Comparisons need to use "end - 1" and "ceiling - 1" (though
492 * that end 0 case should be mythical).
494 * Wherever addr is brought up or ceiling brought down, we must
495 * be careful to reject "the opposite 0" before it confuses the
496 * subsequent tests. But what about where end is brought down
497 * by PMD_SIZE below? no, end can't go down to 0 there.
499 * Whereas we round start (addr) and ceiling down, by different
500 * masks at different levels, in order to test whether a table
501 * now has no other vmas using it, so can be freed, we don't
502 * bother to round floor or end up - the tests don't need that.
516 if (end
- 1 > ceiling
- 1)
521 pgd
= pgd_offset(tlb
->mm
, addr
);
523 next
= pgd_addr_end(addr
, end
);
524 if (pgd_none_or_clear_bad(pgd
))
526 free_pud_range(tlb
, pgd
, addr
, next
, floor
, ceiling
);
527 } while (pgd
++, addr
= next
, addr
!= end
);
530 void free_pgtables(struct mmu_gather
*tlb
, struct vm_area_struct
*vma
,
531 unsigned long floor
, unsigned long ceiling
)
534 struct vm_area_struct
*next
= vma
->vm_next
;
535 unsigned long addr
= vma
->vm_start
;
538 * Hide vma from rmap and truncate_pagecache before freeing
541 unlink_anon_vmas(vma
);
542 unlink_file_vma(vma
);
544 if (is_vm_hugetlb_page(vma
)) {
545 hugetlb_free_pgd_range(tlb
, addr
, vma
->vm_end
,
546 floor
, next
? next
->vm_start
: ceiling
);
549 * Optimization: gather nearby vmas into one call down
551 while (next
&& next
->vm_start
<= vma
->vm_end
+ PMD_SIZE
552 && !is_vm_hugetlb_page(next
)) {
555 unlink_anon_vmas(vma
);
556 unlink_file_vma(vma
);
558 free_pgd_range(tlb
, addr
, vma
->vm_end
,
559 floor
, next
? next
->vm_start
: ceiling
);
565 int __pte_alloc(struct mm_struct
*mm
, pmd_t
*pmd
, unsigned long address
)
568 pgtable_t
new = pte_alloc_one(mm
, address
);
573 * Ensure all pte setup (eg. pte page lock and page clearing) are
574 * visible before the pte is made visible to other CPUs by being
575 * put into page tables.
577 * The other side of the story is the pointer chasing in the page
578 * table walking code (when walking the page table without locking;
579 * ie. most of the time). Fortunately, these data accesses consist
580 * of a chain of data-dependent loads, meaning most CPUs (alpha
581 * being the notable exception) will already guarantee loads are
582 * seen in-order. See the alpha page table accessors for the
583 * smp_read_barrier_depends() barriers in page table walking code.
585 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
587 ptl
= pmd_lock(mm
, pmd
);
588 if (likely(pmd_none(*pmd
))) { /* Has another populated it ? */
589 atomic_long_inc(&mm
->nr_ptes
);
590 pmd_populate(mm
, pmd
, new);
599 int __pte_alloc_kernel(pmd_t
*pmd
, unsigned long address
)
601 pte_t
*new = pte_alloc_one_kernel(&init_mm
, address
);
605 smp_wmb(); /* See comment in __pte_alloc */
607 spin_lock(&init_mm
.page_table_lock
);
608 if (likely(pmd_none(*pmd
))) { /* Has another populated it ? */
609 pmd_populate_kernel(&init_mm
, pmd
, new);
612 spin_unlock(&init_mm
.page_table_lock
);
614 pte_free_kernel(&init_mm
, new);
618 static inline void init_rss_vec(int *rss
)
620 memset(rss
, 0, sizeof(int) * NR_MM_COUNTERS
);
623 static inline void add_mm_rss_vec(struct mm_struct
*mm
, int *rss
)
627 if (current
->mm
== mm
)
629 for (i
= 0; i
< NR_MM_COUNTERS
; i
++)
631 add_mm_counter(mm
, i
, rss
[i
]);
635 * This function is called to print an error when a bad pte
636 * is found. For example, we might have a PFN-mapped pte in
637 * a region that doesn't allow it.
639 * The calling function must still handle the error.
641 static void print_bad_pte(struct vm_area_struct
*vma
, unsigned long addr
,
642 pte_t pte
, struct page
*page
)
644 pgd_t
*pgd
= pgd_offset(vma
->vm_mm
, addr
);
645 pud_t
*pud
= pud_offset(pgd
, addr
);
646 pmd_t
*pmd
= pmd_offset(pud
, addr
);
647 struct address_space
*mapping
;
649 static unsigned long resume
;
650 static unsigned long nr_shown
;
651 static unsigned long nr_unshown
;
654 * Allow a burst of 60 reports, then keep quiet for that minute;
655 * or allow a steady drip of one report per second.
657 if (nr_shown
== 60) {
658 if (time_before(jiffies
, resume
)) {
663 pr_alert("BUG: Bad page map: %lu messages suppressed\n",
670 resume
= jiffies
+ 60 * HZ
;
672 mapping
= vma
->vm_file
? vma
->vm_file
->f_mapping
: NULL
;
673 index
= linear_page_index(vma
, addr
);
675 pr_alert("BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
677 (long long)pte_val(pte
), (long long)pmd_val(*pmd
));
679 dump_page(page
, "bad pte");
680 pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
681 (void *)addr
, vma
->vm_flags
, vma
->anon_vma
, mapping
, index
);
683 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
685 pr_alert("file:%pD fault:%pf mmap:%pf readpage:%pf\n",
687 vma
->vm_ops
? vma
->vm_ops
->fault
: NULL
,
688 vma
->vm_file
? vma
->vm_file
->f_op
->mmap
: NULL
,
689 mapping
? mapping
->a_ops
->readpage
: NULL
);
691 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
695 * vm_normal_page -- This function gets the "struct page" associated with a pte.
697 * "Special" mappings do not wish to be associated with a "struct page" (either
698 * it doesn't exist, or it exists but they don't want to touch it). In this
699 * case, NULL is returned here. "Normal" mappings do have a struct page.
701 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
702 * pte bit, in which case this function is trivial. Secondly, an architecture
703 * may not have a spare pte bit, which requires a more complicated scheme,
706 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
707 * special mapping (even if there are underlying and valid "struct pages").
708 * COWed pages of a VM_PFNMAP are always normal.
710 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
711 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
712 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
713 * mapping will always honor the rule
715 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
717 * And for normal mappings this is false.
719 * This restricts such mappings to be a linear translation from virtual address
720 * to pfn. To get around this restriction, we allow arbitrary mappings so long
721 * as the vma is not a COW mapping; in that case, we know that all ptes are
722 * special (because none can have been COWed).
725 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
727 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
728 * page" backing, however the difference is that _all_ pages with a struct
729 * page (that is, those where pfn_valid is true) are refcounted and considered
730 * normal pages by the VM. The disadvantage is that pages are refcounted
731 * (which can be slower and simply not an option for some PFNMAP users). The
732 * advantage is that we don't have to follow the strict linearity rule of
733 * PFNMAP mappings in order to support COWable mappings.
736 #ifdef __HAVE_ARCH_PTE_SPECIAL
737 # define HAVE_PTE_SPECIAL 1
739 # define HAVE_PTE_SPECIAL 0
741 struct page
*vm_normal_page(struct vm_area_struct
*vma
, unsigned long addr
,
744 unsigned long pfn
= pte_pfn(pte
);
746 if (HAVE_PTE_SPECIAL
) {
747 if (likely(!pte_special(pte
)))
749 if (vma
->vm_ops
&& vma
->vm_ops
->find_special_page
)
750 return vma
->vm_ops
->find_special_page(vma
, addr
);
751 if (vma
->vm_flags
& (VM_PFNMAP
| VM_MIXEDMAP
))
753 if (!is_zero_pfn(pfn
))
754 print_bad_pte(vma
, addr
, pte
, NULL
);
758 /* !HAVE_PTE_SPECIAL case follows: */
760 if (unlikely(vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
))) {
761 if (vma
->vm_flags
& VM_MIXEDMAP
) {
767 off
= (addr
- vma
->vm_start
) >> PAGE_SHIFT
;
768 if (pfn
== vma
->vm_pgoff
+ off
)
770 if (!is_cow_mapping(vma
->vm_flags
))
775 if (is_zero_pfn(pfn
))
778 if (unlikely(pfn
> highest_memmap_pfn
)) {
779 print_bad_pte(vma
, addr
, pte
, NULL
);
784 * NOTE! We still have PageReserved() pages in the page tables.
785 * eg. VDSO mappings can cause them to exist.
788 return pfn_to_page(pfn
);
792 * copy one vm_area from one task to the other. Assumes the page tables
793 * already present in the new task to be cleared in the whole range
794 * covered by this vma.
797 static inline unsigned long
798 copy_one_pte(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
799 pte_t
*dst_pte
, pte_t
*src_pte
, struct vm_area_struct
*vma
,
800 unsigned long addr
, int *rss
)
802 unsigned long vm_flags
= vma
->vm_flags
;
803 pte_t pte
= *src_pte
;
806 /* pte contains position in swap or file, so copy. */
807 if (unlikely(!pte_present(pte
))) {
808 swp_entry_t entry
= pte_to_swp_entry(pte
);
810 if (likely(!non_swap_entry(entry
))) {
811 if (swap_duplicate(entry
) < 0)
814 /* make sure dst_mm is on swapoff's mmlist. */
815 if (unlikely(list_empty(&dst_mm
->mmlist
))) {
816 spin_lock(&mmlist_lock
);
817 if (list_empty(&dst_mm
->mmlist
))
818 list_add(&dst_mm
->mmlist
,
820 spin_unlock(&mmlist_lock
);
823 } else if (is_migration_entry(entry
)) {
824 page
= migration_entry_to_page(entry
);
826 rss
[mm_counter(page
)]++;
828 if (is_write_migration_entry(entry
) &&
829 is_cow_mapping(vm_flags
)) {
831 * COW mappings require pages in both
832 * parent and child to be set to read.
834 make_migration_entry_read(&entry
);
835 pte
= swp_entry_to_pte(entry
);
836 if (pte_swp_soft_dirty(*src_pte
))
837 pte
= pte_swp_mksoft_dirty(pte
);
838 set_pte_at(src_mm
, addr
, src_pte
, pte
);
845 * If it's a COW mapping, write protect it both
846 * in the parent and the child
848 if (is_cow_mapping(vm_flags
)) {
849 ptep_set_wrprotect(src_mm
, addr
, src_pte
);
850 pte
= pte_wrprotect(pte
);
854 * If it's a shared mapping, mark it clean in
857 if (vm_flags
& VM_SHARED
)
858 pte
= pte_mkclean(pte
);
859 pte
= pte_mkold(pte
);
861 page
= vm_normal_page(vma
, addr
, pte
);
864 page_dup_rmap(page
, false);
865 rss
[mm_counter(page
)]++;
869 set_pte_at(dst_mm
, addr
, dst_pte
, pte
);
873 static int copy_pte_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
874 pmd_t
*dst_pmd
, pmd_t
*src_pmd
, struct vm_area_struct
*vma
,
875 unsigned long addr
, unsigned long end
)
877 pte_t
*orig_src_pte
, *orig_dst_pte
;
878 pte_t
*src_pte
, *dst_pte
;
879 spinlock_t
*src_ptl
, *dst_ptl
;
881 int rss
[NR_MM_COUNTERS
];
882 swp_entry_t entry
= (swp_entry_t
){0};
887 dst_pte
= pte_alloc_map_lock(dst_mm
, dst_pmd
, addr
, &dst_ptl
);
890 src_pte
= pte_offset_map(src_pmd
, addr
);
891 src_ptl
= pte_lockptr(src_mm
, src_pmd
);
892 spin_lock_nested(src_ptl
, SINGLE_DEPTH_NESTING
);
893 orig_src_pte
= src_pte
;
894 orig_dst_pte
= dst_pte
;
895 arch_enter_lazy_mmu_mode();
899 * We are holding two locks at this point - either of them
900 * could generate latencies in another task on another CPU.
902 if (progress
>= 32) {
904 if (need_resched() ||
905 spin_needbreak(src_ptl
) || spin_needbreak(dst_ptl
))
908 if (pte_none(*src_pte
)) {
912 entry
.val
= copy_one_pte(dst_mm
, src_mm
, dst_pte
, src_pte
,
917 } while (dst_pte
++, src_pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
919 arch_leave_lazy_mmu_mode();
920 spin_unlock(src_ptl
);
921 pte_unmap(orig_src_pte
);
922 add_mm_rss_vec(dst_mm
, rss
);
923 pte_unmap_unlock(orig_dst_pte
, dst_ptl
);
927 if (add_swap_count_continuation(entry
, GFP_KERNEL
) < 0)
936 static inline int copy_pmd_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
937 pud_t
*dst_pud
, pud_t
*src_pud
, struct vm_area_struct
*vma
,
938 unsigned long addr
, unsigned long end
)
940 pmd_t
*src_pmd
, *dst_pmd
;
943 dst_pmd
= pmd_alloc(dst_mm
, dst_pud
, addr
);
946 src_pmd
= pmd_offset(src_pud
, addr
);
948 next
= pmd_addr_end(addr
, end
);
949 if (pmd_trans_huge(*src_pmd
) || pmd_devmap(*src_pmd
)) {
951 VM_BUG_ON(next
-addr
!= HPAGE_PMD_SIZE
);
952 err
= copy_huge_pmd(dst_mm
, src_mm
,
953 dst_pmd
, src_pmd
, addr
, vma
);
960 if (pmd_none_or_clear_bad(src_pmd
))
962 if (copy_pte_range(dst_mm
, src_mm
, dst_pmd
, src_pmd
,
965 } while (dst_pmd
++, src_pmd
++, addr
= next
, addr
!= end
);
969 static inline int copy_pud_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
970 pgd_t
*dst_pgd
, pgd_t
*src_pgd
, struct vm_area_struct
*vma
,
971 unsigned long addr
, unsigned long end
)
973 pud_t
*src_pud
, *dst_pud
;
976 dst_pud
= pud_alloc(dst_mm
, dst_pgd
, addr
);
979 src_pud
= pud_offset(src_pgd
, addr
);
981 next
= pud_addr_end(addr
, end
);
982 if (pud_none_or_clear_bad(src_pud
))
984 if (copy_pmd_range(dst_mm
, src_mm
, dst_pud
, src_pud
,
987 } while (dst_pud
++, src_pud
++, addr
= next
, addr
!= end
);
991 int copy_page_range(struct mm_struct
*dst_mm
, struct mm_struct
*src_mm
,
992 struct vm_area_struct
*vma
)
994 pgd_t
*src_pgd
, *dst_pgd
;
996 unsigned long addr
= vma
->vm_start
;
997 unsigned long end
= vma
->vm_end
;
998 unsigned long mmun_start
; /* For mmu_notifiers */
999 unsigned long mmun_end
; /* For mmu_notifiers */
1004 * Don't copy ptes where a page fault will fill them correctly.
1005 * Fork becomes much lighter when there are big shared or private
1006 * readonly mappings. The tradeoff is that copy_page_range is more
1007 * efficient than faulting.
1009 if (!(vma
->vm_flags
& (VM_HUGETLB
| VM_PFNMAP
| VM_MIXEDMAP
)) &&
1013 if (is_vm_hugetlb_page(vma
))
1014 return copy_hugetlb_page_range(dst_mm
, src_mm
, vma
);
1016 if (unlikely(vma
->vm_flags
& VM_PFNMAP
)) {
1018 * We do not free on error cases below as remove_vma
1019 * gets called on error from higher level routine
1021 ret
= track_pfn_copy(vma
);
1027 * We need to invalidate the secondary MMU mappings only when
1028 * there could be a permission downgrade on the ptes of the
1029 * parent mm. And a permission downgrade will only happen if
1030 * is_cow_mapping() returns true.
1032 is_cow
= is_cow_mapping(vma
->vm_flags
);
1036 mmu_notifier_invalidate_range_start(src_mm
, mmun_start
,
1040 dst_pgd
= pgd_offset(dst_mm
, addr
);
1041 src_pgd
= pgd_offset(src_mm
, addr
);
1043 next
= pgd_addr_end(addr
, end
);
1044 if (pgd_none_or_clear_bad(src_pgd
))
1046 if (unlikely(copy_pud_range(dst_mm
, src_mm
, dst_pgd
, src_pgd
,
1047 vma
, addr
, next
))) {
1051 } while (dst_pgd
++, src_pgd
++, addr
= next
, addr
!= end
);
1054 mmu_notifier_invalidate_range_end(src_mm
, mmun_start
, mmun_end
);
1058 static unsigned long zap_pte_range(struct mmu_gather
*tlb
,
1059 struct vm_area_struct
*vma
, pmd_t
*pmd
,
1060 unsigned long addr
, unsigned long end
,
1061 struct zap_details
*details
)
1063 struct mm_struct
*mm
= tlb
->mm
;
1064 int force_flush
= 0;
1065 int rss
[NR_MM_COUNTERS
];
1073 start_pte
= pte_offset_map_lock(mm
, pmd
, addr
, &ptl
);
1075 arch_enter_lazy_mmu_mode();
1078 if (pte_none(ptent
)) {
1082 if (pte_present(ptent
)) {
1085 page
= vm_normal_page(vma
, addr
, ptent
);
1086 if (unlikely(details
) && page
) {
1088 * unmap_shared_mapping_pages() wants to
1089 * invalidate cache without truncating:
1090 * unmap shared but keep private pages.
1092 if (details
->check_mapping
&&
1093 details
->check_mapping
!= page
->mapping
)
1096 ptent
= ptep_get_and_clear_full(mm
, addr
, pte
,
1098 tlb_remove_tlb_entry(tlb
, pte
, addr
);
1099 if (unlikely(!page
))
1102 if (!PageAnon(page
)) {
1103 if (pte_dirty(ptent
)) {
1105 set_page_dirty(page
);
1107 if (pte_young(ptent
) &&
1108 likely(!(vma
->vm_flags
& VM_SEQ_READ
)))
1109 mark_page_accessed(page
);
1111 rss
[mm_counter(page
)]--;
1112 page_remove_rmap(page
, false);
1113 if (unlikely(page_mapcount(page
) < 0))
1114 print_bad_pte(vma
, addr
, ptent
, page
);
1115 if (unlikely(!__tlb_remove_page(tlb
, page
))) {
1122 /* If details->check_mapping, we leave swap entries. */
1123 if (unlikely(details
))
1126 entry
= pte_to_swp_entry(ptent
);
1127 if (!non_swap_entry(entry
))
1129 else if (is_migration_entry(entry
)) {
1132 page
= migration_entry_to_page(entry
);
1133 rss
[mm_counter(page
)]--;
1135 if (unlikely(!free_swap_and_cache(entry
)))
1136 print_bad_pte(vma
, addr
, ptent
, NULL
);
1137 pte_clear_not_present_full(mm
, addr
, pte
, tlb
->fullmm
);
1138 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1140 add_mm_rss_vec(mm
, rss
);
1141 arch_leave_lazy_mmu_mode();
1143 /* Do the actual TLB flush before dropping ptl */
1145 tlb_flush_mmu_tlbonly(tlb
);
1146 pte_unmap_unlock(start_pte
, ptl
);
1149 * If we forced a TLB flush (either due to running out of
1150 * batch buffers or because we needed to flush dirty TLB
1151 * entries before releasing the ptl), free the batched
1152 * memory too. Restart if we didn't do everything.
1156 tlb_flush_mmu_free(tlb
);
1165 static inline unsigned long zap_pmd_range(struct mmu_gather
*tlb
,
1166 struct vm_area_struct
*vma
, pud_t
*pud
,
1167 unsigned long addr
, unsigned long end
,
1168 struct zap_details
*details
)
1173 pmd
= pmd_offset(pud
, addr
);
1175 next
= pmd_addr_end(addr
, end
);
1176 if (pmd_trans_huge(*pmd
) || pmd_devmap(*pmd
)) {
1177 if (next
- addr
!= HPAGE_PMD_SIZE
) {
1178 #ifdef CONFIG_DEBUG_VM
1179 if (!rwsem_is_locked(&tlb
->mm
->mmap_sem
)) {
1180 pr_err("%s: mmap_sem is unlocked! addr=0x%lx end=0x%lx vma->vm_start=0x%lx vma->vm_end=0x%lx\n",
1181 __func__
, addr
, end
,
1187 split_huge_pmd(vma
, pmd
, addr
);
1188 } else if (zap_huge_pmd(tlb
, vma
, pmd
, addr
))
1193 * Here there can be other concurrent MADV_DONTNEED or
1194 * trans huge page faults running, and if the pmd is
1195 * none or trans huge it can change under us. This is
1196 * because MADV_DONTNEED holds the mmap_sem in read
1199 if (pmd_none_or_trans_huge_or_clear_bad(pmd
))
1201 next
= zap_pte_range(tlb
, vma
, pmd
, addr
, next
, details
);
1204 } while (pmd
++, addr
= next
, addr
!= end
);
1209 static inline unsigned long zap_pud_range(struct mmu_gather
*tlb
,
1210 struct vm_area_struct
*vma
, pgd_t
*pgd
,
1211 unsigned long addr
, unsigned long end
,
1212 struct zap_details
*details
)
1217 pud
= pud_offset(pgd
, addr
);
1219 next
= pud_addr_end(addr
, end
);
1220 if (pud_none_or_clear_bad(pud
))
1222 next
= zap_pmd_range(tlb
, vma
, pud
, addr
, next
, details
);
1223 } while (pud
++, addr
= next
, addr
!= end
);
1228 static void unmap_page_range(struct mmu_gather
*tlb
,
1229 struct vm_area_struct
*vma
,
1230 unsigned long addr
, unsigned long end
,
1231 struct zap_details
*details
)
1236 if (details
&& !details
->check_mapping
)
1239 BUG_ON(addr
>= end
);
1240 tlb_start_vma(tlb
, vma
);
1241 pgd
= pgd_offset(vma
->vm_mm
, addr
);
1243 next
= pgd_addr_end(addr
, end
);
1244 if (pgd_none_or_clear_bad(pgd
))
1246 next
= zap_pud_range(tlb
, vma
, pgd
, addr
, next
, details
);
1247 } while (pgd
++, addr
= next
, addr
!= end
);
1248 tlb_end_vma(tlb
, vma
);
1252 static void unmap_single_vma(struct mmu_gather
*tlb
,
1253 struct vm_area_struct
*vma
, unsigned long start_addr
,
1254 unsigned long end_addr
,
1255 struct zap_details
*details
)
1257 unsigned long start
= max(vma
->vm_start
, start_addr
);
1260 if (start
>= vma
->vm_end
)
1262 end
= min(vma
->vm_end
, end_addr
);
1263 if (end
<= vma
->vm_start
)
1267 uprobe_munmap(vma
, start
, end
);
1269 if (unlikely(vma
->vm_flags
& VM_PFNMAP
))
1270 untrack_pfn(vma
, 0, 0);
1273 if (unlikely(is_vm_hugetlb_page(vma
))) {
1275 * It is undesirable to test vma->vm_file as it
1276 * should be non-null for valid hugetlb area.
1277 * However, vm_file will be NULL in the error
1278 * cleanup path of mmap_region. When
1279 * hugetlbfs ->mmap method fails,
1280 * mmap_region() nullifies vma->vm_file
1281 * before calling this function to clean up.
1282 * Since no pte has actually been setup, it is
1283 * safe to do nothing in this case.
1286 i_mmap_lock_write(vma
->vm_file
->f_mapping
);
1287 __unmap_hugepage_range_final(tlb
, vma
, start
, end
, NULL
);
1288 i_mmap_unlock_write(vma
->vm_file
->f_mapping
);
1291 unmap_page_range(tlb
, vma
, start
, end
, details
);
1296 * unmap_vmas - unmap a range of memory covered by a list of vma's
1297 * @tlb: address of the caller's struct mmu_gather
1298 * @vma: the starting vma
1299 * @start_addr: virtual address at which to start unmapping
1300 * @end_addr: virtual address at which to end unmapping
1302 * Unmap all pages in the vma list.
1304 * Only addresses between `start' and `end' will be unmapped.
1306 * The VMA list must be sorted in ascending virtual address order.
1308 * unmap_vmas() assumes that the caller will flush the whole unmapped address
1309 * range after unmap_vmas() returns. So the only responsibility here is to
1310 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
1311 * drops the lock and schedules.
1313 void unmap_vmas(struct mmu_gather
*tlb
,
1314 struct vm_area_struct
*vma
, unsigned long start_addr
,
1315 unsigned long end_addr
)
1317 struct mm_struct
*mm
= vma
->vm_mm
;
1319 mmu_notifier_invalidate_range_start(mm
, start_addr
, end_addr
);
1320 for ( ; vma
&& vma
->vm_start
< end_addr
; vma
= vma
->vm_next
)
1321 unmap_single_vma(tlb
, vma
, start_addr
, end_addr
, NULL
);
1322 mmu_notifier_invalidate_range_end(mm
, start_addr
, end_addr
);
1326 * zap_page_range - remove user pages in a given range
1327 * @vma: vm_area_struct holding the applicable pages
1328 * @start: starting address of pages to zap
1329 * @size: number of bytes to zap
1330 * @details: details of shared cache invalidation
1332 * Caller must protect the VMA list
1334 void zap_page_range(struct vm_area_struct
*vma
, unsigned long start
,
1335 unsigned long size
, struct zap_details
*details
)
1337 struct mm_struct
*mm
= vma
->vm_mm
;
1338 struct mmu_gather tlb
;
1339 unsigned long end
= start
+ size
;
1342 tlb_gather_mmu(&tlb
, mm
, start
, end
);
1343 update_hiwater_rss(mm
);
1344 mmu_notifier_invalidate_range_start(mm
, start
, end
);
1345 for ( ; vma
&& vma
->vm_start
< end
; vma
= vma
->vm_next
)
1346 unmap_single_vma(&tlb
, vma
, start
, end
, details
);
1347 mmu_notifier_invalidate_range_end(mm
, start
, end
);
1348 tlb_finish_mmu(&tlb
, start
, end
);
1352 * zap_page_range_single - remove user pages in a given range
1353 * @vma: vm_area_struct holding the applicable pages
1354 * @address: starting address of pages to zap
1355 * @size: number of bytes to zap
1356 * @details: details of shared cache invalidation
1358 * The range must fit into one VMA.
1360 static void zap_page_range_single(struct vm_area_struct
*vma
, unsigned long address
,
1361 unsigned long size
, struct zap_details
*details
)
1363 struct mm_struct
*mm
= vma
->vm_mm
;
1364 struct mmu_gather tlb
;
1365 unsigned long end
= address
+ size
;
1368 tlb_gather_mmu(&tlb
, mm
, address
, end
);
1369 update_hiwater_rss(mm
);
1370 mmu_notifier_invalidate_range_start(mm
, address
, end
);
1371 unmap_single_vma(&tlb
, vma
, address
, end
, details
);
1372 mmu_notifier_invalidate_range_end(mm
, address
, end
);
1373 tlb_finish_mmu(&tlb
, address
, end
);
1377 * zap_vma_ptes - remove ptes mapping the vma
1378 * @vma: vm_area_struct holding ptes to be zapped
1379 * @address: starting address of pages to zap
1380 * @size: number of bytes to zap
1382 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1384 * The entire address range must be fully contained within the vma.
1386 * Returns 0 if successful.
1388 int zap_vma_ptes(struct vm_area_struct
*vma
, unsigned long address
,
1391 if (address
< vma
->vm_start
|| address
+ size
> vma
->vm_end
||
1392 !(vma
->vm_flags
& VM_PFNMAP
))
1394 zap_page_range_single(vma
, address
, size
, NULL
);
1397 EXPORT_SYMBOL_GPL(zap_vma_ptes
);
1399 pte_t
*__get_locked_pte(struct mm_struct
*mm
, unsigned long addr
,
1402 pgd_t
* pgd
= pgd_offset(mm
, addr
);
1403 pud_t
* pud
= pud_alloc(mm
, pgd
, addr
);
1405 pmd_t
* pmd
= pmd_alloc(mm
, pud
, addr
);
1407 VM_BUG_ON(pmd_trans_huge(*pmd
));
1408 return pte_alloc_map_lock(mm
, pmd
, addr
, ptl
);
1415 * This is the old fallback for page remapping.
1417 * For historical reasons, it only allows reserved pages. Only
1418 * old drivers should use this, and they needed to mark their
1419 * pages reserved for the old functions anyway.
1421 static int insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
1422 struct page
*page
, pgprot_t prot
)
1424 struct mm_struct
*mm
= vma
->vm_mm
;
1433 flush_dcache_page(page
);
1434 pte
= get_locked_pte(mm
, addr
, &ptl
);
1438 if (!pte_none(*pte
))
1441 /* Ok, finally just insert the thing.. */
1443 inc_mm_counter_fast(mm
, mm_counter_file(page
));
1444 page_add_file_rmap(page
);
1445 set_pte_at(mm
, addr
, pte
, mk_pte(page
, prot
));
1448 pte_unmap_unlock(pte
, ptl
);
1451 pte_unmap_unlock(pte
, ptl
);
1457 * vm_insert_page - insert single page into user vma
1458 * @vma: user vma to map to
1459 * @addr: target user address of this page
1460 * @page: source kernel page
1462 * This allows drivers to insert individual pages they've allocated
1465 * The page has to be a nice clean _individual_ kernel allocation.
1466 * If you allocate a compound page, you need to have marked it as
1467 * such (__GFP_COMP), or manually just split the page up yourself
1468 * (see split_page()).
1470 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1471 * took an arbitrary page protection parameter. This doesn't allow
1472 * that. Your vma protection will have to be set up correctly, which
1473 * means that if you want a shared writable mapping, you'd better
1474 * ask for a shared writable mapping!
1476 * The page does not need to be reserved.
1478 * Usually this function is called from f_op->mmap() handler
1479 * under mm->mmap_sem write-lock, so it can change vma->vm_flags.
1480 * Caller must set VM_MIXEDMAP on vma if it wants to call this
1481 * function from other places, for example from page-fault handler.
1483 int vm_insert_page(struct vm_area_struct
*vma
, unsigned long addr
,
1486 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
1488 if (!page_count(page
))
1490 if (!(vma
->vm_flags
& VM_MIXEDMAP
)) {
1491 BUG_ON(down_read_trylock(&vma
->vm_mm
->mmap_sem
));
1492 BUG_ON(vma
->vm_flags
& VM_PFNMAP
);
1493 vma
->vm_flags
|= VM_MIXEDMAP
;
1495 return insert_page(vma
, addr
, page
, vma
->vm_page_prot
);
1497 EXPORT_SYMBOL(vm_insert_page
);
1499 static int insert_pfn(struct vm_area_struct
*vma
, unsigned long addr
,
1500 pfn_t pfn
, pgprot_t prot
)
1502 struct mm_struct
*mm
= vma
->vm_mm
;
1508 pte
= get_locked_pte(mm
, addr
, &ptl
);
1512 if (!pte_none(*pte
))
1515 /* Ok, finally just insert the thing.. */
1516 if (pfn_t_devmap(pfn
))
1517 entry
= pte_mkdevmap(pfn_t_pte(pfn
, prot
));
1519 entry
= pte_mkspecial(pfn_t_pte(pfn
, prot
));
1520 set_pte_at(mm
, addr
, pte
, entry
);
1521 update_mmu_cache(vma
, addr
, pte
); /* XXX: why not for insert_page? */
1525 pte_unmap_unlock(pte
, ptl
);
1531 * vm_insert_pfn - insert single pfn into user vma
1532 * @vma: user vma to map to
1533 * @addr: target user address of this page
1534 * @pfn: source kernel pfn
1536 * Similar to vm_insert_page, this allows drivers to insert individual pages
1537 * they've allocated into a user vma. Same comments apply.
1539 * This function should only be called from a vm_ops->fault handler, and
1540 * in that case the handler should return NULL.
1542 * vma cannot be a COW mapping.
1544 * As this is called only for pages that do not currently exist, we
1545 * do not need to flush old virtual caches or the TLB.
1547 int vm_insert_pfn(struct vm_area_struct
*vma
, unsigned long addr
,
1550 return vm_insert_pfn_prot(vma
, addr
, pfn
, vma
->vm_page_prot
);
1552 EXPORT_SYMBOL(vm_insert_pfn
);
1555 * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot
1556 * @vma: user vma to map to
1557 * @addr: target user address of this page
1558 * @pfn: source kernel pfn
1559 * @pgprot: pgprot flags for the inserted page
1561 * This is exactly like vm_insert_pfn, except that it allows drivers to
1562 * to override pgprot on a per-page basis.
1564 * This only makes sense for IO mappings, and it makes no sense for
1565 * cow mappings. In general, using multiple vmas is preferable;
1566 * vm_insert_pfn_prot should only be used if using multiple VMAs is
1569 int vm_insert_pfn_prot(struct vm_area_struct
*vma
, unsigned long addr
,
1570 unsigned long pfn
, pgprot_t pgprot
)
1574 * Technically, architectures with pte_special can avoid all these
1575 * restrictions (same for remap_pfn_range). However we would like
1576 * consistency in testing and feature parity among all, so we should
1577 * try to keep these invariants in place for everybody.
1579 BUG_ON(!(vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
)));
1580 BUG_ON((vma
->vm_flags
& (VM_PFNMAP
|VM_MIXEDMAP
)) ==
1581 (VM_PFNMAP
|VM_MIXEDMAP
));
1582 BUG_ON((vma
->vm_flags
& VM_PFNMAP
) && is_cow_mapping(vma
->vm_flags
));
1583 BUG_ON((vma
->vm_flags
& VM_MIXEDMAP
) && pfn_valid(pfn
));
1585 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
1587 if (track_pfn_insert(vma
, &pgprot
, __pfn_to_pfn_t(pfn
, PFN_DEV
)))
1590 ret
= insert_pfn(vma
, addr
, __pfn_to_pfn_t(pfn
, PFN_DEV
), pgprot
);
1594 EXPORT_SYMBOL(vm_insert_pfn_prot
);
1596 int vm_insert_mixed(struct vm_area_struct
*vma
, unsigned long addr
,
1599 BUG_ON(!(vma
->vm_flags
& VM_MIXEDMAP
));
1601 if (addr
< vma
->vm_start
|| addr
>= vma
->vm_end
)
1605 * If we don't have pte special, then we have to use the pfn_valid()
1606 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1607 * refcount the page if pfn_valid is true (hence insert_page rather
1608 * than insert_pfn). If a zero_pfn were inserted into a VM_MIXEDMAP
1609 * without pte special, it would there be refcounted as a normal page.
1611 if (!HAVE_PTE_SPECIAL
&& !pfn_t_devmap(pfn
) && pfn_t_valid(pfn
)) {
1615 * At this point we are committed to insert_page()
1616 * regardless of whether the caller specified flags that
1617 * result in pfn_t_has_page() == false.
1619 page
= pfn_to_page(pfn_t_to_pfn(pfn
));
1620 return insert_page(vma
, addr
, page
, vma
->vm_page_prot
);
1622 return insert_pfn(vma
, addr
, pfn
, vma
->vm_page_prot
);
1624 EXPORT_SYMBOL(vm_insert_mixed
);
1627 * maps a range of physical memory into the requested pages. the old
1628 * mappings are removed. any references to nonexistent pages results
1629 * in null mappings (currently treated as "copy-on-access")
1631 static int remap_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1632 unsigned long addr
, unsigned long end
,
1633 unsigned long pfn
, pgprot_t prot
)
1638 pte
= pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1641 arch_enter_lazy_mmu_mode();
1643 BUG_ON(!pte_none(*pte
));
1644 set_pte_at(mm
, addr
, pte
, pte_mkspecial(pfn_pte(pfn
, prot
)));
1646 } while (pte
++, addr
+= PAGE_SIZE
, addr
!= end
);
1647 arch_leave_lazy_mmu_mode();
1648 pte_unmap_unlock(pte
- 1, ptl
);
1652 static inline int remap_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1653 unsigned long addr
, unsigned long end
,
1654 unsigned long pfn
, pgprot_t prot
)
1659 pfn
-= addr
>> PAGE_SHIFT
;
1660 pmd
= pmd_alloc(mm
, pud
, addr
);
1663 VM_BUG_ON(pmd_trans_huge(*pmd
));
1665 next
= pmd_addr_end(addr
, end
);
1666 if (remap_pte_range(mm
, pmd
, addr
, next
,
1667 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1669 } while (pmd
++, addr
= next
, addr
!= end
);
1673 static inline int remap_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1674 unsigned long addr
, unsigned long end
,
1675 unsigned long pfn
, pgprot_t prot
)
1680 pfn
-= addr
>> PAGE_SHIFT
;
1681 pud
= pud_alloc(mm
, pgd
, addr
);
1685 next
= pud_addr_end(addr
, end
);
1686 if (remap_pmd_range(mm
, pud
, addr
, next
,
1687 pfn
+ (addr
>> PAGE_SHIFT
), prot
))
1689 } while (pud
++, addr
= next
, addr
!= end
);
1694 * remap_pfn_range - remap kernel memory to userspace
1695 * @vma: user vma to map to
1696 * @addr: target user address to start at
1697 * @pfn: physical address of kernel memory
1698 * @size: size of map area
1699 * @prot: page protection flags for this mapping
1701 * Note: this is only safe if the mm semaphore is held when called.
1703 int remap_pfn_range(struct vm_area_struct
*vma
, unsigned long addr
,
1704 unsigned long pfn
, unsigned long size
, pgprot_t prot
)
1708 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1709 struct mm_struct
*mm
= vma
->vm_mm
;
1713 * Physically remapped pages are special. Tell the
1714 * rest of the world about it:
1715 * VM_IO tells people not to look at these pages
1716 * (accesses can have side effects).
1717 * VM_PFNMAP tells the core MM that the base pages are just
1718 * raw PFN mappings, and do not have a "struct page" associated
1721 * Disable vma merging and expanding with mremap().
1723 * Omit vma from core dump, even when VM_IO turned off.
1725 * There's a horrible special case to handle copy-on-write
1726 * behaviour that some programs depend on. We mark the "original"
1727 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1728 * See vm_normal_page() for details.
1730 if (is_cow_mapping(vma
->vm_flags
)) {
1731 if (addr
!= vma
->vm_start
|| end
!= vma
->vm_end
)
1733 vma
->vm_pgoff
= pfn
;
1736 err
= track_pfn_remap(vma
, &prot
, pfn
, addr
, PAGE_ALIGN(size
));
1740 vma
->vm_flags
|= VM_IO
| VM_PFNMAP
| VM_DONTEXPAND
| VM_DONTDUMP
;
1742 BUG_ON(addr
>= end
);
1743 pfn
-= addr
>> PAGE_SHIFT
;
1744 pgd
= pgd_offset(mm
, addr
);
1745 flush_cache_range(vma
, addr
, end
);
1747 next
= pgd_addr_end(addr
, end
);
1748 err
= remap_pud_range(mm
, pgd
, addr
, next
,
1749 pfn
+ (addr
>> PAGE_SHIFT
), prot
);
1752 } while (pgd
++, addr
= next
, addr
!= end
);
1755 untrack_pfn(vma
, pfn
, PAGE_ALIGN(size
));
1759 EXPORT_SYMBOL(remap_pfn_range
);
1762 * vm_iomap_memory - remap memory to userspace
1763 * @vma: user vma to map to
1764 * @start: start of area
1765 * @len: size of area
1767 * This is a simplified io_remap_pfn_range() for common driver use. The
1768 * driver just needs to give us the physical memory range to be mapped,
1769 * we'll figure out the rest from the vma information.
1771 * NOTE! Some drivers might want to tweak vma->vm_page_prot first to get
1772 * whatever write-combining details or similar.
1774 int vm_iomap_memory(struct vm_area_struct
*vma
, phys_addr_t start
, unsigned long len
)
1776 unsigned long vm_len
, pfn
, pages
;
1778 /* Check that the physical memory area passed in looks valid */
1779 if (start
+ len
< start
)
1782 * You *really* shouldn't map things that aren't page-aligned,
1783 * but we've historically allowed it because IO memory might
1784 * just have smaller alignment.
1786 len
+= start
& ~PAGE_MASK
;
1787 pfn
= start
>> PAGE_SHIFT
;
1788 pages
= (len
+ ~PAGE_MASK
) >> PAGE_SHIFT
;
1789 if (pfn
+ pages
< pfn
)
1792 /* We start the mapping 'vm_pgoff' pages into the area */
1793 if (vma
->vm_pgoff
> pages
)
1795 pfn
+= vma
->vm_pgoff
;
1796 pages
-= vma
->vm_pgoff
;
1798 /* Can we fit all of the mapping? */
1799 vm_len
= vma
->vm_end
- vma
->vm_start
;
1800 if (vm_len
>> PAGE_SHIFT
> pages
)
1803 /* Ok, let it rip */
1804 return io_remap_pfn_range(vma
, vma
->vm_start
, pfn
, vm_len
, vma
->vm_page_prot
);
1806 EXPORT_SYMBOL(vm_iomap_memory
);
1808 static int apply_to_pte_range(struct mm_struct
*mm
, pmd_t
*pmd
,
1809 unsigned long addr
, unsigned long end
,
1810 pte_fn_t fn
, void *data
)
1815 spinlock_t
*uninitialized_var(ptl
);
1817 pte
= (mm
== &init_mm
) ?
1818 pte_alloc_kernel(pmd
, addr
) :
1819 pte_alloc_map_lock(mm
, pmd
, addr
, &ptl
);
1823 BUG_ON(pmd_huge(*pmd
));
1825 arch_enter_lazy_mmu_mode();
1827 token
= pmd_pgtable(*pmd
);
1830 err
= fn(pte
++, token
, addr
, data
);
1833 } while (addr
+= PAGE_SIZE
, addr
!= end
);
1835 arch_leave_lazy_mmu_mode();
1838 pte_unmap_unlock(pte
-1, ptl
);
1842 static int apply_to_pmd_range(struct mm_struct
*mm
, pud_t
*pud
,
1843 unsigned long addr
, unsigned long end
,
1844 pte_fn_t fn
, void *data
)
1850 BUG_ON(pud_huge(*pud
));
1852 pmd
= pmd_alloc(mm
, pud
, addr
);
1856 next
= pmd_addr_end(addr
, end
);
1857 err
= apply_to_pte_range(mm
, pmd
, addr
, next
, fn
, data
);
1860 } while (pmd
++, addr
= next
, addr
!= end
);
1864 static int apply_to_pud_range(struct mm_struct
*mm
, pgd_t
*pgd
,
1865 unsigned long addr
, unsigned long end
,
1866 pte_fn_t fn
, void *data
)
1872 pud
= pud_alloc(mm
, pgd
, addr
);
1876 next
= pud_addr_end(addr
, end
);
1877 err
= apply_to_pmd_range(mm
, pud
, addr
, next
, fn
, data
);
1880 } while (pud
++, addr
= next
, addr
!= end
);
1885 * Scan a region of virtual memory, filling in page tables as necessary
1886 * and calling a provided function on each leaf page table.
1888 int apply_to_page_range(struct mm_struct
*mm
, unsigned long addr
,
1889 unsigned long size
, pte_fn_t fn
, void *data
)
1893 unsigned long end
= addr
+ size
;
1896 if (WARN_ON(addr
>= end
))
1899 pgd
= pgd_offset(mm
, addr
);
1901 next
= pgd_addr_end(addr
, end
);
1902 err
= apply_to_pud_range(mm
, pgd
, addr
, next
, fn
, data
);
1905 } while (pgd
++, addr
= next
, addr
!= end
);
1909 EXPORT_SYMBOL_GPL(apply_to_page_range
);
1912 * handle_pte_fault chooses page fault handler according to an entry which was
1913 * read non-atomically. Before making any commitment, on those architectures
1914 * or configurations (e.g. i386 with PAE) which might give a mix of unmatched
1915 * parts, do_swap_page must check under lock before unmapping the pte and
1916 * proceeding (but do_wp_page is only called after already making such a check;
1917 * and do_anonymous_page can safely check later on).
1919 static inline int pte_unmap_same(struct mm_struct
*mm
, pmd_t
*pmd
,
1920 pte_t
*page_table
, pte_t orig_pte
)
1923 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1924 if (sizeof(pte_t
) > sizeof(unsigned long)) {
1925 spinlock_t
*ptl
= pte_lockptr(mm
, pmd
);
1927 same
= pte_same(*page_table
, orig_pte
);
1931 pte_unmap(page_table
);
1935 static inline void cow_user_page(struct page
*dst
, struct page
*src
, unsigned long va
, struct vm_area_struct
*vma
)
1937 debug_dma_assert_idle(src
);
1940 * If the source page was a PFN mapping, we don't have
1941 * a "struct page" for it. We do a best-effort copy by
1942 * just copying from the original user address. If that
1943 * fails, we just zero-fill it. Live with it.
1945 if (unlikely(!src
)) {
1946 void *kaddr
= kmap_atomic(dst
);
1947 void __user
*uaddr
= (void __user
*)(va
& PAGE_MASK
);
1950 * This really shouldn't fail, because the page is there
1951 * in the page tables. But it might just be unreadable,
1952 * in which case we just give up and fill the result with
1955 if (__copy_from_user_inatomic(kaddr
, uaddr
, PAGE_SIZE
))
1957 kunmap_atomic(kaddr
);
1958 flush_dcache_page(dst
);
1960 copy_user_highpage(dst
, src
, va
, vma
);
1963 static gfp_t
__get_fault_gfp_mask(struct vm_area_struct
*vma
)
1965 struct file
*vm_file
= vma
->vm_file
;
1968 return mapping_gfp_mask(vm_file
->f_mapping
) | __GFP_FS
| __GFP_IO
;
1971 * Special mappings (e.g. VDSO) do not have any file so fake
1972 * a default GFP_KERNEL for them.
1978 * Notify the address space that the page is about to become writable so that
1979 * it can prohibit this or wait for the page to get into an appropriate state.
1981 * We do this without the lock held, so that it can sleep if it needs to.
1983 static int do_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
,
1984 unsigned long address
)
1986 struct vm_fault vmf
;
1989 vmf
.virtual_address
= (void __user
*)(address
& PAGE_MASK
);
1990 vmf
.pgoff
= page
->index
;
1991 vmf
.flags
= FAULT_FLAG_WRITE
|FAULT_FLAG_MKWRITE
;
1992 vmf
.gfp_mask
= __get_fault_gfp_mask(vma
);
1994 vmf
.cow_page
= NULL
;
1996 ret
= vma
->vm_ops
->page_mkwrite(vma
, &vmf
);
1997 if (unlikely(ret
& (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
)))
1999 if (unlikely(!(ret
& VM_FAULT_LOCKED
))) {
2001 if (!page
->mapping
) {
2003 return 0; /* retry */
2005 ret
|= VM_FAULT_LOCKED
;
2007 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
2012 * Handle write page faults for pages that can be reused in the current vma
2014 * This can happen either due to the mapping being with the VM_SHARED flag,
2015 * or due to us being the last reference standing to the page. In either
2016 * case, all we need to do here is to mark the page as writable and update
2017 * any related book-keeping.
2019 static inline int wp_page_reuse(struct mm_struct
*mm
,
2020 struct vm_area_struct
*vma
, unsigned long address
,
2021 pte_t
*page_table
, spinlock_t
*ptl
, pte_t orig_pte
,
2022 struct page
*page
, int page_mkwrite
,
2028 * Clear the pages cpupid information as the existing
2029 * information potentially belongs to a now completely
2030 * unrelated process.
2033 page_cpupid_xchg_last(page
, (1 << LAST_CPUPID_SHIFT
) - 1);
2035 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
2036 entry
= pte_mkyoung(orig_pte
);
2037 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2038 if (ptep_set_access_flags(vma
, address
, page_table
, entry
, 1))
2039 update_mmu_cache(vma
, address
, page_table
);
2040 pte_unmap_unlock(page_table
, ptl
);
2043 struct address_space
*mapping
;
2049 dirtied
= set_page_dirty(page
);
2050 VM_BUG_ON_PAGE(PageAnon(page
), page
);
2051 mapping
= page
->mapping
;
2053 page_cache_release(page
);
2055 if ((dirtied
|| page_mkwrite
) && mapping
) {
2057 * Some device drivers do not set page.mapping
2058 * but still dirty their pages
2060 balance_dirty_pages_ratelimited(mapping
);
2064 file_update_time(vma
->vm_file
);
2067 return VM_FAULT_WRITE
;
2071 * Handle the case of a page which we actually need to copy to a new page.
2073 * Called with mmap_sem locked and the old page referenced, but
2074 * without the ptl held.
2076 * High level logic flow:
2078 * - Allocate a page, copy the content of the old page to the new one.
2079 * - Handle book keeping and accounting - cgroups, mmu-notifiers, etc.
2080 * - Take the PTL. If the pte changed, bail out and release the allocated page
2081 * - If the pte is still the way we remember it, update the page table and all
2082 * relevant references. This includes dropping the reference the page-table
2083 * held to the old page, as well as updating the rmap.
2084 * - In any case, unlock the PTL and drop the reference we took to the old page.
2086 static int wp_page_copy(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2087 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2088 pte_t orig_pte
, struct page
*old_page
)
2090 struct page
*new_page
= NULL
;
2091 spinlock_t
*ptl
= NULL
;
2093 int page_copied
= 0;
2094 const unsigned long mmun_start
= address
& PAGE_MASK
; /* For mmu_notifiers */
2095 const unsigned long mmun_end
= mmun_start
+ PAGE_SIZE
; /* For mmu_notifiers */
2096 struct mem_cgroup
*memcg
;
2098 if (unlikely(anon_vma_prepare(vma
)))
2101 if (is_zero_pfn(pte_pfn(orig_pte
))) {
2102 new_page
= alloc_zeroed_user_highpage_movable(vma
, address
);
2106 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, address
);
2109 cow_user_page(new_page
, old_page
, address
, vma
);
2112 if (mem_cgroup_try_charge(new_page
, mm
, GFP_KERNEL
, &memcg
, false))
2115 __SetPageUptodate(new_page
);
2117 mmu_notifier_invalidate_range_start(mm
, mmun_start
, mmun_end
);
2120 * Re-check the pte - we dropped the lock
2122 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2123 if (likely(pte_same(*page_table
, orig_pte
))) {
2125 if (!PageAnon(old_page
)) {
2126 dec_mm_counter_fast(mm
,
2127 mm_counter_file(old_page
));
2128 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
2131 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
2133 flush_cache_page(vma
, address
, pte_pfn(orig_pte
));
2134 entry
= mk_pte(new_page
, vma
->vm_page_prot
);
2135 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2137 * Clear the pte entry and flush it first, before updating the
2138 * pte with the new entry. This will avoid a race condition
2139 * seen in the presence of one thread doing SMC and another
2142 ptep_clear_flush_notify(vma
, address
, page_table
);
2143 page_add_new_anon_rmap(new_page
, vma
, address
, false);
2144 mem_cgroup_commit_charge(new_page
, memcg
, false, false);
2145 lru_cache_add_active_or_unevictable(new_page
, vma
);
2147 * We call the notify macro here because, when using secondary
2148 * mmu page tables (such as kvm shadow page tables), we want the
2149 * new page to be mapped directly into the secondary page table.
2151 set_pte_at_notify(mm
, address
, page_table
, entry
);
2152 update_mmu_cache(vma
, address
, page_table
);
2155 * Only after switching the pte to the new page may
2156 * we remove the mapcount here. Otherwise another
2157 * process may come and find the rmap count decremented
2158 * before the pte is switched to the new page, and
2159 * "reuse" the old page writing into it while our pte
2160 * here still points into it and can be read by other
2163 * The critical issue is to order this
2164 * page_remove_rmap with the ptp_clear_flush above.
2165 * Those stores are ordered by (if nothing else,)
2166 * the barrier present in the atomic_add_negative
2167 * in page_remove_rmap.
2169 * Then the TLB flush in ptep_clear_flush ensures that
2170 * no process can access the old page before the
2171 * decremented mapcount is visible. And the old page
2172 * cannot be reused until after the decremented
2173 * mapcount is visible. So transitively, TLBs to
2174 * old page will be flushed before it can be reused.
2176 page_remove_rmap(old_page
, false);
2179 /* Free the old page.. */
2180 new_page
= old_page
;
2183 mem_cgroup_cancel_charge(new_page
, memcg
, false);
2187 page_cache_release(new_page
);
2189 pte_unmap_unlock(page_table
, ptl
);
2190 mmu_notifier_invalidate_range_end(mm
, mmun_start
, mmun_end
);
2193 * Don't let another task, with possibly unlocked vma,
2194 * keep the mlocked page.
2196 if (page_copied
&& (vma
->vm_flags
& VM_LOCKED
)) {
2197 lock_page(old_page
); /* LRU manipulation */
2198 if (PageMlocked(old_page
))
2199 munlock_vma_page(old_page
);
2200 unlock_page(old_page
);
2202 page_cache_release(old_page
);
2204 return page_copied
? VM_FAULT_WRITE
: 0;
2206 page_cache_release(new_page
);
2209 page_cache_release(old_page
);
2210 return VM_FAULT_OOM
;
2214 * Handle write page faults for VM_MIXEDMAP or VM_PFNMAP for a VM_SHARED
2217 static int wp_pfn_shared(struct mm_struct
*mm
,
2218 struct vm_area_struct
*vma
, unsigned long address
,
2219 pte_t
*page_table
, spinlock_t
*ptl
, pte_t orig_pte
,
2222 if (vma
->vm_ops
&& vma
->vm_ops
->pfn_mkwrite
) {
2223 struct vm_fault vmf
= {
2225 .pgoff
= linear_page_index(vma
, address
),
2226 .virtual_address
= (void __user
*)(address
& PAGE_MASK
),
2227 .flags
= FAULT_FLAG_WRITE
| FAULT_FLAG_MKWRITE
,
2231 pte_unmap_unlock(page_table
, ptl
);
2232 ret
= vma
->vm_ops
->pfn_mkwrite(vma
, &vmf
);
2233 if (ret
& VM_FAULT_ERROR
)
2235 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2237 * We might have raced with another page fault while we
2238 * released the pte_offset_map_lock.
2240 if (!pte_same(*page_table
, orig_pte
)) {
2241 pte_unmap_unlock(page_table
, ptl
);
2245 return wp_page_reuse(mm
, vma
, address
, page_table
, ptl
, orig_pte
,
2249 static int wp_page_shared(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2250 unsigned long address
, pte_t
*page_table
,
2251 pmd_t
*pmd
, spinlock_t
*ptl
, pte_t orig_pte
,
2252 struct page
*old_page
)
2255 int page_mkwrite
= 0;
2257 page_cache_get(old_page
);
2259 if (vma
->vm_ops
&& vma
->vm_ops
->page_mkwrite
) {
2262 pte_unmap_unlock(page_table
, ptl
);
2263 tmp
= do_page_mkwrite(vma
, old_page
, address
);
2264 if (unlikely(!tmp
|| (tmp
&
2265 (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
)))) {
2266 page_cache_release(old_page
);
2270 * Since we dropped the lock we need to revalidate
2271 * the PTE as someone else may have changed it. If
2272 * they did, we just return, as we can count on the
2273 * MMU to tell us if they didn't also make it writable.
2275 page_table
= pte_offset_map_lock(mm
, pmd
, address
,
2277 if (!pte_same(*page_table
, orig_pte
)) {
2278 unlock_page(old_page
);
2279 pte_unmap_unlock(page_table
, ptl
);
2280 page_cache_release(old_page
);
2286 return wp_page_reuse(mm
, vma
, address
, page_table
, ptl
,
2287 orig_pte
, old_page
, page_mkwrite
, 1);
2291 * This routine handles present pages, when users try to write
2292 * to a shared page. It is done by copying the page to a new address
2293 * and decrementing the shared-page counter for the old page.
2295 * Note that this routine assumes that the protection checks have been
2296 * done by the caller (the low-level page fault routine in most cases).
2297 * Thus we can safely just mark it writable once we've done any necessary
2300 * We also mark the page dirty at this point even though the page will
2301 * change only once the write actually happens. This avoids a few races,
2302 * and potentially makes it more efficient.
2304 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2305 * but allow concurrent faults), with pte both mapped and locked.
2306 * We return with mmap_sem still held, but pte unmapped and unlocked.
2308 static int do_wp_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2309 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2310 spinlock_t
*ptl
, pte_t orig_pte
)
2313 struct page
*old_page
;
2315 old_page
= vm_normal_page(vma
, address
, orig_pte
);
2318 * VM_MIXEDMAP !pfn_valid() case, or VM_SOFTDIRTY clear on a
2321 * We should not cow pages in a shared writeable mapping.
2322 * Just mark the pages writable and/or call ops->pfn_mkwrite.
2324 if ((vma
->vm_flags
& (VM_WRITE
|VM_SHARED
)) ==
2325 (VM_WRITE
|VM_SHARED
))
2326 return wp_pfn_shared(mm
, vma
, address
, page_table
, ptl
,
2329 pte_unmap_unlock(page_table
, ptl
);
2330 return wp_page_copy(mm
, vma
, address
, page_table
, pmd
,
2331 orig_pte
, old_page
);
2335 * Take out anonymous pages first, anonymous shared vmas are
2336 * not dirty accountable.
2338 if (PageAnon(old_page
) && !PageKsm(old_page
)) {
2339 if (!trylock_page(old_page
)) {
2340 page_cache_get(old_page
);
2341 pte_unmap_unlock(page_table
, ptl
);
2342 lock_page(old_page
);
2343 page_table
= pte_offset_map_lock(mm
, pmd
, address
,
2345 if (!pte_same(*page_table
, orig_pte
)) {
2346 unlock_page(old_page
);
2347 pte_unmap_unlock(page_table
, ptl
);
2348 page_cache_release(old_page
);
2351 page_cache_release(old_page
);
2353 if (reuse_swap_page(old_page
)) {
2355 * The page is all ours. Move it to our anon_vma so
2356 * the rmap code will not search our parent or siblings.
2357 * Protected against the rmap code by the page lock.
2359 page_move_anon_rmap(old_page
, vma
, address
);
2360 unlock_page(old_page
);
2361 return wp_page_reuse(mm
, vma
, address
, page_table
, ptl
,
2362 orig_pte
, old_page
, 0, 0);
2364 unlock_page(old_page
);
2365 } else if (unlikely((vma
->vm_flags
& (VM_WRITE
|VM_SHARED
)) ==
2366 (VM_WRITE
|VM_SHARED
))) {
2367 return wp_page_shared(mm
, vma
, address
, page_table
, pmd
,
2368 ptl
, orig_pte
, old_page
);
2372 * Ok, we need to copy. Oh, well..
2374 page_cache_get(old_page
);
2376 pte_unmap_unlock(page_table
, ptl
);
2377 return wp_page_copy(mm
, vma
, address
, page_table
, pmd
,
2378 orig_pte
, old_page
);
2381 static void unmap_mapping_range_vma(struct vm_area_struct
*vma
,
2382 unsigned long start_addr
, unsigned long end_addr
,
2383 struct zap_details
*details
)
2385 zap_page_range_single(vma
, start_addr
, end_addr
- start_addr
, details
);
2388 static inline void unmap_mapping_range_tree(struct rb_root
*root
,
2389 struct zap_details
*details
)
2391 struct vm_area_struct
*vma
;
2392 pgoff_t vba
, vea
, zba
, zea
;
2394 vma_interval_tree_foreach(vma
, root
,
2395 details
->first_index
, details
->last_index
) {
2397 vba
= vma
->vm_pgoff
;
2398 vea
= vba
+ vma_pages(vma
) - 1;
2399 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2400 zba
= details
->first_index
;
2403 zea
= details
->last_index
;
2407 unmap_mapping_range_vma(vma
,
2408 ((zba
- vba
) << PAGE_SHIFT
) + vma
->vm_start
,
2409 ((zea
- vba
+ 1) << PAGE_SHIFT
) + vma
->vm_start
,
2415 * unmap_mapping_range - unmap the portion of all mmaps in the specified
2416 * address_space corresponding to the specified page range in the underlying
2419 * @mapping: the address space containing mmaps to be unmapped.
2420 * @holebegin: byte in first page to unmap, relative to the start of
2421 * the underlying file. This will be rounded down to a PAGE_SIZE
2422 * boundary. Note that this is different from truncate_pagecache(), which
2423 * must keep the partial page. In contrast, we must get rid of
2425 * @holelen: size of prospective hole in bytes. This will be rounded
2426 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2428 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2429 * but 0 when invalidating pagecache, don't throw away private data.
2431 void unmap_mapping_range(struct address_space
*mapping
,
2432 loff_t
const holebegin
, loff_t
const holelen
, int even_cows
)
2434 struct zap_details details
;
2435 pgoff_t hba
= holebegin
>> PAGE_SHIFT
;
2436 pgoff_t hlen
= (holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2438 /* Check for overflow. */
2439 if (sizeof(holelen
) > sizeof(hlen
)) {
2441 (holebegin
+ holelen
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
2442 if (holeend
& ~(long long)ULONG_MAX
)
2443 hlen
= ULONG_MAX
- hba
+ 1;
2446 details
.check_mapping
= even_cows
? NULL
: mapping
;
2447 details
.first_index
= hba
;
2448 details
.last_index
= hba
+ hlen
- 1;
2449 if (details
.last_index
< details
.first_index
)
2450 details
.last_index
= ULONG_MAX
;
2453 /* DAX uses i_mmap_lock to serialise file truncate vs page fault */
2454 i_mmap_lock_write(mapping
);
2455 if (unlikely(!RB_EMPTY_ROOT(&mapping
->i_mmap
)))
2456 unmap_mapping_range_tree(&mapping
->i_mmap
, &details
);
2457 i_mmap_unlock_write(mapping
);
2459 EXPORT_SYMBOL(unmap_mapping_range
);
2462 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2463 * but allow concurrent faults), and pte mapped but not yet locked.
2464 * We return with pte unmapped and unlocked.
2466 * We return with the mmap_sem locked or unlocked in the same cases
2467 * as does filemap_fault().
2469 static int do_swap_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2470 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2471 unsigned int flags
, pte_t orig_pte
)
2474 struct page
*page
, *swapcache
;
2475 struct mem_cgroup
*memcg
;
2482 if (!pte_unmap_same(mm
, pmd
, page_table
, orig_pte
))
2485 entry
= pte_to_swp_entry(orig_pte
);
2486 if (unlikely(non_swap_entry(entry
))) {
2487 if (is_migration_entry(entry
)) {
2488 migration_entry_wait(mm
, pmd
, address
);
2489 } else if (is_hwpoison_entry(entry
)) {
2490 ret
= VM_FAULT_HWPOISON
;
2492 print_bad_pte(vma
, address
, orig_pte
, NULL
);
2493 ret
= VM_FAULT_SIGBUS
;
2497 delayacct_set_flag(DELAYACCT_PF_SWAPIN
);
2498 page
= lookup_swap_cache(entry
);
2500 page
= swapin_readahead(entry
,
2501 GFP_HIGHUSER_MOVABLE
, vma
, address
);
2504 * Back out if somebody else faulted in this pte
2505 * while we released the pte lock.
2507 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2508 if (likely(pte_same(*page_table
, orig_pte
)))
2510 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2514 /* Had to read the page from swap area: Major fault */
2515 ret
= VM_FAULT_MAJOR
;
2516 count_vm_event(PGMAJFAULT
);
2517 mem_cgroup_count_vm_event(mm
, PGMAJFAULT
);
2518 } else if (PageHWPoison(page
)) {
2520 * hwpoisoned dirty swapcache pages are kept for killing
2521 * owner processes (which may be unknown at hwpoison time)
2523 ret
= VM_FAULT_HWPOISON
;
2524 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2530 locked
= lock_page_or_retry(page
, mm
, flags
);
2532 delayacct_clear_flag(DELAYACCT_PF_SWAPIN
);
2534 ret
|= VM_FAULT_RETRY
;
2539 * Make sure try_to_free_swap or reuse_swap_page or swapoff did not
2540 * release the swapcache from under us. The page pin, and pte_same
2541 * test below, are not enough to exclude that. Even if it is still
2542 * swapcache, we need to check that the page's swap has not changed.
2544 if (unlikely(!PageSwapCache(page
) || page_private(page
) != entry
.val
))
2547 page
= ksm_might_need_to_copy(page
, vma
, address
);
2548 if (unlikely(!page
)) {
2554 if (mem_cgroup_try_charge(page
, mm
, GFP_KERNEL
, &memcg
, false)) {
2560 * Back out if somebody else already faulted in this pte.
2562 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2563 if (unlikely(!pte_same(*page_table
, orig_pte
)))
2566 if (unlikely(!PageUptodate(page
))) {
2567 ret
= VM_FAULT_SIGBUS
;
2572 * The page isn't present yet, go ahead with the fault.
2574 * Be careful about the sequence of operations here.
2575 * To get its accounting right, reuse_swap_page() must be called
2576 * while the page is counted on swap but not yet in mapcount i.e.
2577 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2578 * must be called after the swap_free(), or it will never succeed.
2581 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
2582 dec_mm_counter_fast(mm
, MM_SWAPENTS
);
2583 pte
= mk_pte(page
, vma
->vm_page_prot
);
2584 if ((flags
& FAULT_FLAG_WRITE
) && reuse_swap_page(page
)) {
2585 pte
= maybe_mkwrite(pte_mkdirty(pte
), vma
);
2586 flags
&= ~FAULT_FLAG_WRITE
;
2587 ret
|= VM_FAULT_WRITE
;
2588 exclusive
= RMAP_EXCLUSIVE
;
2590 flush_icache_page(vma
, page
);
2591 if (pte_swp_soft_dirty(orig_pte
))
2592 pte
= pte_mksoft_dirty(pte
);
2593 set_pte_at(mm
, address
, page_table
, pte
);
2594 if (page
== swapcache
) {
2595 do_page_add_anon_rmap(page
, vma
, address
, exclusive
);
2596 mem_cgroup_commit_charge(page
, memcg
, true, false);
2597 } else { /* ksm created a completely new copy */
2598 page_add_new_anon_rmap(page
, vma
, address
, false);
2599 mem_cgroup_commit_charge(page
, memcg
, false, false);
2600 lru_cache_add_active_or_unevictable(page
, vma
);
2604 if (mem_cgroup_swap_full(page
) ||
2605 (vma
->vm_flags
& VM_LOCKED
) || PageMlocked(page
))
2606 try_to_free_swap(page
);
2608 if (page
!= swapcache
) {
2610 * Hold the lock to avoid the swap entry to be reused
2611 * until we take the PT lock for the pte_same() check
2612 * (to avoid false positives from pte_same). For
2613 * further safety release the lock after the swap_free
2614 * so that the swap count won't change under a
2615 * parallel locked swapcache.
2617 unlock_page(swapcache
);
2618 page_cache_release(swapcache
);
2621 if (flags
& FAULT_FLAG_WRITE
) {
2622 ret
|= do_wp_page(mm
, vma
, address
, page_table
, pmd
, ptl
, pte
);
2623 if (ret
& VM_FAULT_ERROR
)
2624 ret
&= VM_FAULT_ERROR
;
2628 /* No need to invalidate - it was non-present before */
2629 update_mmu_cache(vma
, address
, page_table
);
2631 pte_unmap_unlock(page_table
, ptl
);
2635 mem_cgroup_cancel_charge(page
, memcg
, false);
2636 pte_unmap_unlock(page_table
, ptl
);
2640 page_cache_release(page
);
2641 if (page
!= swapcache
) {
2642 unlock_page(swapcache
);
2643 page_cache_release(swapcache
);
2649 * This is like a special single-page "expand_{down|up}wards()",
2650 * except we must first make sure that 'address{-|+}PAGE_SIZE'
2651 * doesn't hit another vma.
2653 static inline int check_stack_guard_page(struct vm_area_struct
*vma
, unsigned long address
)
2655 address
&= PAGE_MASK
;
2656 if ((vma
->vm_flags
& VM_GROWSDOWN
) && address
== vma
->vm_start
) {
2657 struct vm_area_struct
*prev
= vma
->vm_prev
;
2660 * Is there a mapping abutting this one below?
2662 * That's only ok if it's the same stack mapping
2663 * that has gotten split..
2665 if (prev
&& prev
->vm_end
== address
)
2666 return prev
->vm_flags
& VM_GROWSDOWN
? 0 : -ENOMEM
;
2668 return expand_downwards(vma
, address
- PAGE_SIZE
);
2670 if ((vma
->vm_flags
& VM_GROWSUP
) && address
+ PAGE_SIZE
== vma
->vm_end
) {
2671 struct vm_area_struct
*next
= vma
->vm_next
;
2673 /* As VM_GROWSDOWN but s/below/above/ */
2674 if (next
&& next
->vm_start
== address
+ PAGE_SIZE
)
2675 return next
->vm_flags
& VM_GROWSUP
? 0 : -ENOMEM
;
2677 return expand_upwards(vma
, address
+ PAGE_SIZE
);
2683 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2684 * but allow concurrent faults), and pte mapped but not yet locked.
2685 * We return with mmap_sem still held, but pte unmapped and unlocked.
2687 static int do_anonymous_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2688 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
2691 struct mem_cgroup
*memcg
;
2696 pte_unmap(page_table
);
2698 /* File mapping without ->vm_ops ? */
2699 if (vma
->vm_flags
& VM_SHARED
)
2700 return VM_FAULT_SIGBUS
;
2702 /* Check if we need to add a guard page to the stack */
2703 if (check_stack_guard_page(vma
, address
) < 0)
2704 return VM_FAULT_SIGSEGV
;
2706 /* Use the zero-page for reads */
2707 if (!(flags
& FAULT_FLAG_WRITE
) && !mm_forbids_zeropage(mm
)) {
2708 entry
= pte_mkspecial(pfn_pte(my_zero_pfn(address
),
2709 vma
->vm_page_prot
));
2710 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2711 if (!pte_none(*page_table
))
2713 /* Deliver the page fault to userland, check inside PT lock */
2714 if (userfaultfd_missing(vma
)) {
2715 pte_unmap_unlock(page_table
, ptl
);
2716 return handle_userfault(vma
, address
, flags
,
2722 /* Allocate our own private page. */
2723 if (unlikely(anon_vma_prepare(vma
)))
2725 page
= alloc_zeroed_user_highpage_movable(vma
, address
);
2729 if (mem_cgroup_try_charge(page
, mm
, GFP_KERNEL
, &memcg
, false))
2733 * The memory barrier inside __SetPageUptodate makes sure that
2734 * preceeding stores to the page contents become visible before
2735 * the set_pte_at() write.
2737 __SetPageUptodate(page
);
2739 entry
= mk_pte(page
, vma
->vm_page_prot
);
2740 if (vma
->vm_flags
& VM_WRITE
)
2741 entry
= pte_mkwrite(pte_mkdirty(entry
));
2743 page_table
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2744 if (!pte_none(*page_table
))
2747 /* Deliver the page fault to userland, check inside PT lock */
2748 if (userfaultfd_missing(vma
)) {
2749 pte_unmap_unlock(page_table
, ptl
);
2750 mem_cgroup_cancel_charge(page
, memcg
, false);
2751 page_cache_release(page
);
2752 return handle_userfault(vma
, address
, flags
,
2756 inc_mm_counter_fast(mm
, MM_ANONPAGES
);
2757 page_add_new_anon_rmap(page
, vma
, address
, false);
2758 mem_cgroup_commit_charge(page
, memcg
, false, false);
2759 lru_cache_add_active_or_unevictable(page
, vma
);
2761 set_pte_at(mm
, address
, page_table
, entry
);
2763 /* No need to invalidate - it was non-present before */
2764 update_mmu_cache(vma
, address
, page_table
);
2766 pte_unmap_unlock(page_table
, ptl
);
2769 mem_cgroup_cancel_charge(page
, memcg
, false);
2770 page_cache_release(page
);
2773 page_cache_release(page
);
2775 return VM_FAULT_OOM
;
2779 * The mmap_sem must have been held on entry, and may have been
2780 * released depending on flags and vma->vm_ops->fault() return value.
2781 * See filemap_fault() and __lock_page_retry().
2783 static int __do_fault(struct vm_area_struct
*vma
, unsigned long address
,
2784 pgoff_t pgoff
, unsigned int flags
,
2785 struct page
*cow_page
, struct page
**page
)
2787 struct vm_fault vmf
;
2790 vmf
.virtual_address
= (void __user
*)(address
& PAGE_MASK
);
2794 vmf
.gfp_mask
= __get_fault_gfp_mask(vma
);
2795 vmf
.cow_page
= cow_page
;
2797 ret
= vma
->vm_ops
->fault(vma
, &vmf
);
2798 if (unlikely(ret
& (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
| VM_FAULT_RETRY
)))
2803 if (unlikely(PageHWPoison(vmf
.page
))) {
2804 if (ret
& VM_FAULT_LOCKED
)
2805 unlock_page(vmf
.page
);
2806 page_cache_release(vmf
.page
);
2807 return VM_FAULT_HWPOISON
;
2810 if (unlikely(!(ret
& VM_FAULT_LOCKED
)))
2811 lock_page(vmf
.page
);
2813 VM_BUG_ON_PAGE(!PageLocked(vmf
.page
), vmf
.page
);
2821 * do_set_pte - setup new PTE entry for given page and add reverse page mapping.
2823 * @vma: virtual memory area
2824 * @address: user virtual address
2825 * @page: page to map
2826 * @pte: pointer to target page table entry
2827 * @write: true, if new entry is writable
2828 * @anon: true, if it's anonymous page
2830 * Caller must hold page table lock relevant for @pte.
2832 * Target users are page handler itself and implementations of
2833 * vm_ops->map_pages.
2835 void do_set_pte(struct vm_area_struct
*vma
, unsigned long address
,
2836 struct page
*page
, pte_t
*pte
, bool write
, bool anon
)
2840 flush_icache_page(vma
, page
);
2841 entry
= mk_pte(page
, vma
->vm_page_prot
);
2843 entry
= maybe_mkwrite(pte_mkdirty(entry
), vma
);
2845 inc_mm_counter_fast(vma
->vm_mm
, MM_ANONPAGES
);
2846 page_add_new_anon_rmap(page
, vma
, address
, false);
2848 inc_mm_counter_fast(vma
->vm_mm
, mm_counter_file(page
));
2849 page_add_file_rmap(page
);
2851 set_pte_at(vma
->vm_mm
, address
, pte
, entry
);
2853 /* no need to invalidate: a not-present page won't be cached */
2854 update_mmu_cache(vma
, address
, pte
);
2857 static unsigned long fault_around_bytes __read_mostly
=
2858 rounddown_pow_of_two(65536);
2860 #ifdef CONFIG_DEBUG_FS
2861 static int fault_around_bytes_get(void *data
, u64
*val
)
2863 *val
= fault_around_bytes
;
2868 * fault_around_pages() and fault_around_mask() expects fault_around_bytes
2869 * rounded down to nearest page order. It's what do_fault_around() expects to
2872 static int fault_around_bytes_set(void *data
, u64 val
)
2874 if (val
/ PAGE_SIZE
> PTRS_PER_PTE
)
2876 if (val
> PAGE_SIZE
)
2877 fault_around_bytes
= rounddown_pow_of_two(val
);
2879 fault_around_bytes
= PAGE_SIZE
; /* rounddown_pow_of_two(0) is undefined */
2882 DEFINE_SIMPLE_ATTRIBUTE(fault_around_bytes_fops
,
2883 fault_around_bytes_get
, fault_around_bytes_set
, "%llu\n");
2885 static int __init
fault_around_debugfs(void)
2889 ret
= debugfs_create_file("fault_around_bytes", 0644, NULL
, NULL
,
2890 &fault_around_bytes_fops
);
2892 pr_warn("Failed to create fault_around_bytes in debugfs");
2895 late_initcall(fault_around_debugfs
);
2899 * do_fault_around() tries to map few pages around the fault address. The hope
2900 * is that the pages will be needed soon and this will lower the number of
2903 * It uses vm_ops->map_pages() to map the pages, which skips the page if it's
2904 * not ready to be mapped: not up-to-date, locked, etc.
2906 * This function is called with the page table lock taken. In the split ptlock
2907 * case the page table lock only protects only those entries which belong to
2908 * the page table corresponding to the fault address.
2910 * This function doesn't cross the VMA boundaries, in order to call map_pages()
2913 * fault_around_pages() defines how many pages we'll try to map.
2914 * do_fault_around() expects it to return a power of two less than or equal to
2917 * The virtual address of the area that we map is naturally aligned to the
2918 * fault_around_pages() value (and therefore to page order). This way it's
2919 * easier to guarantee that we don't cross page table boundaries.
2921 static void do_fault_around(struct vm_area_struct
*vma
, unsigned long address
,
2922 pte_t
*pte
, pgoff_t pgoff
, unsigned int flags
)
2924 unsigned long start_addr
, nr_pages
, mask
;
2926 struct vm_fault vmf
;
2929 nr_pages
= READ_ONCE(fault_around_bytes
) >> PAGE_SHIFT
;
2930 mask
= ~(nr_pages
* PAGE_SIZE
- 1) & PAGE_MASK
;
2932 start_addr
= max(address
& mask
, vma
->vm_start
);
2933 off
= ((address
- start_addr
) >> PAGE_SHIFT
) & (PTRS_PER_PTE
- 1);
2938 * max_pgoff is either end of page table or end of vma
2939 * or fault_around_pages() from pgoff, depending what is nearest.
2941 max_pgoff
= pgoff
- ((start_addr
>> PAGE_SHIFT
) & (PTRS_PER_PTE
- 1)) +
2943 max_pgoff
= min3(max_pgoff
, vma_pages(vma
) + vma
->vm_pgoff
- 1,
2944 pgoff
+ nr_pages
- 1);
2946 /* Check if it makes any sense to call ->map_pages */
2947 while (!pte_none(*pte
)) {
2948 if (++pgoff
> max_pgoff
)
2950 start_addr
+= PAGE_SIZE
;
2951 if (start_addr
>= vma
->vm_end
)
2956 vmf
.virtual_address
= (void __user
*) start_addr
;
2959 vmf
.max_pgoff
= max_pgoff
;
2961 vmf
.gfp_mask
= __get_fault_gfp_mask(vma
);
2962 vma
->vm_ops
->map_pages(vma
, &vmf
);
2965 static int do_read_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
2966 unsigned long address
, pmd_t
*pmd
,
2967 pgoff_t pgoff
, unsigned int flags
, pte_t orig_pte
)
2969 struct page
*fault_page
;
2975 * Let's call ->map_pages() first and use ->fault() as fallback
2976 * if page by the offset is not ready to be mapped (cold cache or
2979 if (vma
->vm_ops
->map_pages
&& fault_around_bytes
>> PAGE_SHIFT
> 1) {
2980 pte
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2981 do_fault_around(vma
, address
, pte
, pgoff
, flags
);
2982 if (!pte_same(*pte
, orig_pte
))
2984 pte_unmap_unlock(pte
, ptl
);
2987 ret
= __do_fault(vma
, address
, pgoff
, flags
, NULL
, &fault_page
);
2988 if (unlikely(ret
& (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
| VM_FAULT_RETRY
)))
2991 pte
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
2992 if (unlikely(!pte_same(*pte
, orig_pte
))) {
2993 pte_unmap_unlock(pte
, ptl
);
2994 unlock_page(fault_page
);
2995 page_cache_release(fault_page
);
2998 do_set_pte(vma
, address
, fault_page
, pte
, false, false);
2999 unlock_page(fault_page
);
3001 pte_unmap_unlock(pte
, ptl
);
3005 static int do_cow_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3006 unsigned long address
, pmd_t
*pmd
,
3007 pgoff_t pgoff
, unsigned int flags
, pte_t orig_pte
)
3009 struct page
*fault_page
, *new_page
;
3010 struct mem_cgroup
*memcg
;
3015 if (unlikely(anon_vma_prepare(vma
)))
3016 return VM_FAULT_OOM
;
3018 new_page
= alloc_page_vma(GFP_HIGHUSER_MOVABLE
, vma
, address
);
3020 return VM_FAULT_OOM
;
3022 if (mem_cgroup_try_charge(new_page
, mm
, GFP_KERNEL
, &memcg
, false)) {
3023 page_cache_release(new_page
);
3024 return VM_FAULT_OOM
;
3027 ret
= __do_fault(vma
, address
, pgoff
, flags
, new_page
, &fault_page
);
3028 if (unlikely(ret
& (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
| VM_FAULT_RETRY
)))
3032 copy_user_highpage(new_page
, fault_page
, address
, vma
);
3033 __SetPageUptodate(new_page
);
3035 pte
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
3036 if (unlikely(!pte_same(*pte
, orig_pte
))) {
3037 pte_unmap_unlock(pte
, ptl
);
3039 unlock_page(fault_page
);
3040 page_cache_release(fault_page
);
3043 * The fault handler has no page to lock, so it holds
3044 * i_mmap_lock for read to protect against truncate.
3046 i_mmap_unlock_read(vma
->vm_file
->f_mapping
);
3050 do_set_pte(vma
, address
, new_page
, pte
, true, true);
3051 mem_cgroup_commit_charge(new_page
, memcg
, false, false);
3052 lru_cache_add_active_or_unevictable(new_page
, vma
);
3053 pte_unmap_unlock(pte
, ptl
);
3055 unlock_page(fault_page
);
3056 page_cache_release(fault_page
);
3059 * The fault handler has no page to lock, so it holds
3060 * i_mmap_lock for read to protect against truncate.
3062 i_mmap_unlock_read(vma
->vm_file
->f_mapping
);
3066 mem_cgroup_cancel_charge(new_page
, memcg
, false);
3067 page_cache_release(new_page
);
3071 static int do_shared_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3072 unsigned long address
, pmd_t
*pmd
,
3073 pgoff_t pgoff
, unsigned int flags
, pte_t orig_pte
)
3075 struct page
*fault_page
;
3076 struct address_space
*mapping
;
3082 ret
= __do_fault(vma
, address
, pgoff
, flags
, NULL
, &fault_page
);
3083 if (unlikely(ret
& (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
| VM_FAULT_RETRY
)))
3087 * Check if the backing address space wants to know that the page is
3088 * about to become writable
3090 if (vma
->vm_ops
->page_mkwrite
) {
3091 unlock_page(fault_page
);
3092 tmp
= do_page_mkwrite(vma
, fault_page
, address
);
3093 if (unlikely(!tmp
||
3094 (tmp
& (VM_FAULT_ERROR
| VM_FAULT_NOPAGE
)))) {
3095 page_cache_release(fault_page
);
3100 pte
= pte_offset_map_lock(mm
, pmd
, address
, &ptl
);
3101 if (unlikely(!pte_same(*pte
, orig_pte
))) {
3102 pte_unmap_unlock(pte
, ptl
);
3103 unlock_page(fault_page
);
3104 page_cache_release(fault_page
);
3107 do_set_pte(vma
, address
, fault_page
, pte
, true, false);
3108 pte_unmap_unlock(pte
, ptl
);
3110 if (set_page_dirty(fault_page
))
3113 * Take a local copy of the address_space - page.mapping may be zeroed
3114 * by truncate after unlock_page(). The address_space itself remains
3115 * pinned by vma->vm_file's reference. We rely on unlock_page()'s
3116 * release semantics to prevent the compiler from undoing this copying.
3118 mapping
= page_rmapping(fault_page
);
3119 unlock_page(fault_page
);
3120 if ((dirtied
|| vma
->vm_ops
->page_mkwrite
) && mapping
) {
3122 * Some device drivers do not set page.mapping but still
3125 balance_dirty_pages_ratelimited(mapping
);
3128 if (!vma
->vm_ops
->page_mkwrite
)
3129 file_update_time(vma
->vm_file
);
3135 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3136 * but allow concurrent faults).
3137 * The mmap_sem may have been released depending on flags and our
3138 * return value. See filemap_fault() and __lock_page_or_retry().
3140 static int do_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3141 unsigned long address
, pte_t
*page_table
, pmd_t
*pmd
,
3142 unsigned int flags
, pte_t orig_pte
)
3144 pgoff_t pgoff
= linear_page_index(vma
, address
);
3146 pte_unmap(page_table
);
3147 /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */
3148 if (!vma
->vm_ops
->fault
)
3149 return VM_FAULT_SIGBUS
;
3150 if (!(flags
& FAULT_FLAG_WRITE
))
3151 return do_read_fault(mm
, vma
, address
, pmd
, pgoff
, flags
,
3153 if (!(vma
->vm_flags
& VM_SHARED
))
3154 return do_cow_fault(mm
, vma
, address
, pmd
, pgoff
, flags
,
3156 return do_shared_fault(mm
, vma
, address
, pmd
, pgoff
, flags
, orig_pte
);
3159 static int numa_migrate_prep(struct page
*page
, struct vm_area_struct
*vma
,
3160 unsigned long addr
, int page_nid
,
3165 count_vm_numa_event(NUMA_HINT_FAULTS
);
3166 if (page_nid
== numa_node_id()) {
3167 count_vm_numa_event(NUMA_HINT_FAULTS_LOCAL
);
3168 *flags
|= TNF_FAULT_LOCAL
;
3171 return mpol_misplaced(page
, vma
, addr
);
3174 static int do_numa_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3175 unsigned long addr
, pte_t pte
, pte_t
*ptep
, pmd_t
*pmd
)
3177 struct page
*page
= NULL
;
3182 bool migrated
= false;
3183 bool was_writable
= pte_write(pte
);
3186 /* A PROT_NONE fault should not end up here */
3187 BUG_ON(!(vma
->vm_flags
& (VM_READ
| VM_EXEC
| VM_WRITE
)));
3190 * The "pte" at this point cannot be used safely without
3191 * validation through pte_unmap_same(). It's of NUMA type but
3192 * the pfn may be screwed if the read is non atomic.
3194 * We can safely just do a "set_pte_at()", because the old
3195 * page table entry is not accessible, so there would be no
3196 * concurrent hardware modifications to the PTE.
3198 ptl
= pte_lockptr(mm
, pmd
);
3200 if (unlikely(!pte_same(*ptep
, pte
))) {
3201 pte_unmap_unlock(ptep
, ptl
);
3205 /* Make it present again */
3206 pte
= pte_modify(pte
, vma
->vm_page_prot
);
3207 pte
= pte_mkyoung(pte
);
3209 pte
= pte_mkwrite(pte
);
3210 set_pte_at(mm
, addr
, ptep
, pte
);
3211 update_mmu_cache(vma
, addr
, ptep
);
3213 page
= vm_normal_page(vma
, addr
, pte
);
3215 pte_unmap_unlock(ptep
, ptl
);
3219 /* TODO: handle PTE-mapped THP */
3220 if (PageCompound(page
)) {
3221 pte_unmap_unlock(ptep
, ptl
);
3226 * Avoid grouping on RO pages in general. RO pages shouldn't hurt as
3227 * much anyway since they can be in shared cache state. This misses
3228 * the case where a mapping is writable but the process never writes
3229 * to it but pte_write gets cleared during protection updates and
3230 * pte_dirty has unpredictable behaviour between PTE scan updates,
3231 * background writeback, dirty balancing and application behaviour.
3233 if (!(vma
->vm_flags
& VM_WRITE
))
3234 flags
|= TNF_NO_GROUP
;
3237 * Flag if the page is shared between multiple address spaces. This
3238 * is later used when determining whether to group tasks together
3240 if (page_mapcount(page
) > 1 && (vma
->vm_flags
& VM_SHARED
))
3241 flags
|= TNF_SHARED
;
3243 last_cpupid
= page_cpupid_last(page
);
3244 page_nid
= page_to_nid(page
);
3245 target_nid
= numa_migrate_prep(page
, vma
, addr
, page_nid
, &flags
);
3246 pte_unmap_unlock(ptep
, ptl
);
3247 if (target_nid
== -1) {
3252 /* Migrate to the requested node */
3253 migrated
= migrate_misplaced_page(page
, vma
, target_nid
);
3255 page_nid
= target_nid
;
3256 flags
|= TNF_MIGRATED
;
3258 flags
|= TNF_MIGRATE_FAIL
;
3262 task_numa_fault(last_cpupid
, page_nid
, 1, flags
);
3266 static int create_huge_pmd(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3267 unsigned long address
, pmd_t
*pmd
, unsigned int flags
)
3269 if (vma_is_anonymous(vma
))
3270 return do_huge_pmd_anonymous_page(mm
, vma
, address
, pmd
, flags
);
3271 if (vma
->vm_ops
->pmd_fault
)
3272 return vma
->vm_ops
->pmd_fault(vma
, address
, pmd
, flags
);
3273 return VM_FAULT_FALLBACK
;
3276 static int wp_huge_pmd(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3277 unsigned long address
, pmd_t
*pmd
, pmd_t orig_pmd
,
3280 if (vma_is_anonymous(vma
))
3281 return do_huge_pmd_wp_page(mm
, vma
, address
, pmd
, orig_pmd
);
3282 if (vma
->vm_ops
->pmd_fault
)
3283 return vma
->vm_ops
->pmd_fault(vma
, address
, pmd
, flags
);
3284 return VM_FAULT_FALLBACK
;
3288 * These routines also need to handle stuff like marking pages dirty
3289 * and/or accessed for architectures that don't do it in hardware (most
3290 * RISC architectures). The early dirtying is also good on the i386.
3292 * There is also a hook called "update_mmu_cache()" that architectures
3293 * with external mmu caches can use to update those (ie the Sparc or
3294 * PowerPC hashed page tables that act as extended TLBs).
3296 * We enter with non-exclusive mmap_sem (to exclude vma changes,
3297 * but allow concurrent faults), and pte mapped but not yet locked.
3298 * We return with pte unmapped and unlocked.
3300 * The mmap_sem may have been released depending on flags and our
3301 * return value. See filemap_fault() and __lock_page_or_retry().
3303 static int handle_pte_fault(struct mm_struct
*mm
,
3304 struct vm_area_struct
*vma
, unsigned long address
,
3305 pte_t
*pte
, pmd_t
*pmd
, unsigned int flags
)
3311 * some architectures can have larger ptes than wordsize,
3312 * e.g.ppc44x-defconfig has CONFIG_PTE_64BIT=y and CONFIG_32BIT=y,
3313 * so READ_ONCE or ACCESS_ONCE cannot guarantee atomic accesses.
3314 * The code below just needs a consistent view for the ifs and
3315 * we later double check anyway with the ptl lock held. So here
3316 * a barrier will do.
3320 if (!pte_present(entry
)) {
3321 if (pte_none(entry
)) {
3322 if (vma_is_anonymous(vma
))
3323 return do_anonymous_page(mm
, vma
, address
,
3326 return do_fault(mm
, vma
, address
, pte
, pmd
,
3329 return do_swap_page(mm
, vma
, address
,
3330 pte
, pmd
, flags
, entry
);
3333 if (pte_protnone(entry
))
3334 return do_numa_page(mm
, vma
, address
, entry
, pte
, pmd
);
3336 ptl
= pte_lockptr(mm
, pmd
);
3338 if (unlikely(!pte_same(*pte
, entry
)))
3340 if (flags
& FAULT_FLAG_WRITE
) {
3341 if (!pte_write(entry
))
3342 return do_wp_page(mm
, vma
, address
,
3343 pte
, pmd
, ptl
, entry
);
3344 entry
= pte_mkdirty(entry
);
3346 entry
= pte_mkyoung(entry
);
3347 if (ptep_set_access_flags(vma
, address
, pte
, entry
, flags
& FAULT_FLAG_WRITE
)) {
3348 update_mmu_cache(vma
, address
, pte
);
3351 * This is needed only for protection faults but the arch code
3352 * is not yet telling us if this is a protection fault or not.
3353 * This still avoids useless tlb flushes for .text page faults
3356 if (flags
& FAULT_FLAG_WRITE
)
3357 flush_tlb_fix_spurious_fault(vma
, address
);
3360 pte_unmap_unlock(pte
, ptl
);
3365 * By the time we get here, we already hold the mm semaphore
3367 * The mmap_sem may have been released depending on flags and our
3368 * return value. See filemap_fault() and __lock_page_or_retry().
3370 static int __handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3371 unsigned long address
, unsigned int flags
)
3378 if (unlikely(is_vm_hugetlb_page(vma
)))
3379 return hugetlb_fault(mm
, vma
, address
, flags
);
3381 pgd
= pgd_offset(mm
, address
);
3382 pud
= pud_alloc(mm
, pgd
, address
);
3384 return VM_FAULT_OOM
;
3385 pmd
= pmd_alloc(mm
, pud
, address
);
3387 return VM_FAULT_OOM
;
3388 if (pmd_none(*pmd
) && transparent_hugepage_enabled(vma
)) {
3389 int ret
= create_huge_pmd(mm
, vma
, address
, pmd
, flags
);
3390 if (!(ret
& VM_FAULT_FALLBACK
))
3393 pmd_t orig_pmd
= *pmd
;
3397 if (pmd_trans_huge(orig_pmd
) || pmd_devmap(orig_pmd
)) {
3398 unsigned int dirty
= flags
& FAULT_FLAG_WRITE
;
3400 if (pmd_protnone(orig_pmd
))
3401 return do_huge_pmd_numa_page(mm
, vma
, address
,
3404 if (dirty
&& !pmd_write(orig_pmd
)) {
3405 ret
= wp_huge_pmd(mm
, vma
, address
, pmd
,
3407 if (!(ret
& VM_FAULT_FALLBACK
))
3410 huge_pmd_set_accessed(mm
, vma
, address
, pmd
,
3418 * Use pte_alloc() instead of pte_alloc_map, because we can't
3419 * run pte_offset_map on the pmd, if an huge pmd could
3420 * materialize from under us from a different thread.
3422 if (unlikely(pte_alloc(mm
, pmd
, address
)))
3423 return VM_FAULT_OOM
;
3425 * If a huge pmd materialized under us just retry later. Use
3426 * pmd_trans_unstable() instead of pmd_trans_huge() to ensure the pmd
3427 * didn't become pmd_trans_huge under us and then back to pmd_none, as
3428 * a result of MADV_DONTNEED running immediately after a huge pmd fault
3429 * in a different thread of this mm, in turn leading to a misleading
3430 * pmd_trans_huge() retval. All we have to ensure is that it is a
3431 * regular pmd that we can walk with pte_offset_map() and we can do that
3432 * through an atomic read in C, which is what pmd_trans_unstable()
3435 if (unlikely(pmd_trans_unstable(pmd
) || pmd_devmap(*pmd
)))
3438 * A regular pmd is established and it can't morph into a huge pmd
3439 * from under us anymore at this point because we hold the mmap_sem
3440 * read mode and khugepaged takes it in write mode. So now it's
3441 * safe to run pte_offset_map().
3443 pte
= pte_offset_map(pmd
, address
);
3445 return handle_pte_fault(mm
, vma
, address
, pte
, pmd
, flags
);
3449 * By the time we get here, we already hold the mm semaphore
3451 * The mmap_sem may have been released depending on flags and our
3452 * return value. See filemap_fault() and __lock_page_or_retry().
3454 int handle_mm_fault(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
3455 unsigned long address
, unsigned int flags
)
3459 __set_current_state(TASK_RUNNING
);
3461 count_vm_event(PGFAULT
);
3462 mem_cgroup_count_vm_event(mm
, PGFAULT
);
3464 /* do counter updates before entering really critical section. */
3465 check_sync_rss_stat(current
);
3468 * Enable the memcg OOM handling for faults triggered in user
3469 * space. Kernel faults are handled more gracefully.
3471 if (flags
& FAULT_FLAG_USER
)
3472 mem_cgroup_oom_enable();
3474 ret
= __handle_mm_fault(mm
, vma
, address
, flags
);
3476 if (flags
& FAULT_FLAG_USER
) {
3477 mem_cgroup_oom_disable();
3479 * The task may have entered a memcg OOM situation but
3480 * if the allocation error was handled gracefully (no
3481 * VM_FAULT_OOM), there is no need to kill anything.
3482 * Just clean up the OOM state peacefully.
3484 if (task_in_memcg_oom(current
) && !(ret
& VM_FAULT_OOM
))
3485 mem_cgroup_oom_synchronize(false);
3490 EXPORT_SYMBOL_GPL(handle_mm_fault
);
3492 #ifndef __PAGETABLE_PUD_FOLDED
3494 * Allocate page upper directory.
3495 * We've already handled the fast-path in-line.
3497 int __pud_alloc(struct mm_struct
*mm
, pgd_t
*pgd
, unsigned long address
)
3499 pud_t
*new = pud_alloc_one(mm
, address
);
3503 smp_wmb(); /* See comment in __pte_alloc */
3505 spin_lock(&mm
->page_table_lock
);
3506 if (pgd_present(*pgd
)) /* Another has populated it */
3509 pgd_populate(mm
, pgd
, new);
3510 spin_unlock(&mm
->page_table_lock
);
3513 #endif /* __PAGETABLE_PUD_FOLDED */
3515 #ifndef __PAGETABLE_PMD_FOLDED
3517 * Allocate page middle directory.
3518 * We've already handled the fast-path in-line.
3520 int __pmd_alloc(struct mm_struct
*mm
, pud_t
*pud
, unsigned long address
)
3522 pmd_t
*new = pmd_alloc_one(mm
, address
);
3526 smp_wmb(); /* See comment in __pte_alloc */
3528 spin_lock(&mm
->page_table_lock
);
3529 #ifndef __ARCH_HAS_4LEVEL_HACK
3530 if (!pud_present(*pud
)) {
3532 pud_populate(mm
, pud
, new);
3533 } else /* Another has populated it */
3536 if (!pgd_present(*pud
)) {
3538 pgd_populate(mm
, pud
, new);
3539 } else /* Another has populated it */
3541 #endif /* __ARCH_HAS_4LEVEL_HACK */
3542 spin_unlock(&mm
->page_table_lock
);
3545 #endif /* __PAGETABLE_PMD_FOLDED */
3547 static int __follow_pte(struct mm_struct
*mm
, unsigned long address
,
3548 pte_t
**ptepp
, spinlock_t
**ptlp
)
3555 pgd
= pgd_offset(mm
, address
);
3556 if (pgd_none(*pgd
) || unlikely(pgd_bad(*pgd
)))
3559 pud
= pud_offset(pgd
, address
);
3560 if (pud_none(*pud
) || unlikely(pud_bad(*pud
)))
3563 pmd
= pmd_offset(pud
, address
);
3564 VM_BUG_ON(pmd_trans_huge(*pmd
));
3565 if (pmd_none(*pmd
) || unlikely(pmd_bad(*pmd
)))
3568 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3572 ptep
= pte_offset_map_lock(mm
, pmd
, address
, ptlp
);
3575 if (!pte_present(*ptep
))
3580 pte_unmap_unlock(ptep
, *ptlp
);
3585 static inline int follow_pte(struct mm_struct
*mm
, unsigned long address
,
3586 pte_t
**ptepp
, spinlock_t
**ptlp
)
3590 /* (void) is needed to make gcc happy */
3591 (void) __cond_lock(*ptlp
,
3592 !(res
= __follow_pte(mm
, address
, ptepp
, ptlp
)));
3597 * follow_pfn - look up PFN at a user virtual address
3598 * @vma: memory mapping
3599 * @address: user virtual address
3600 * @pfn: location to store found PFN
3602 * Only IO mappings and raw PFN mappings are allowed.
3604 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3606 int follow_pfn(struct vm_area_struct
*vma
, unsigned long address
,
3613 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
3616 ret
= follow_pte(vma
->vm_mm
, address
, &ptep
, &ptl
);
3619 *pfn
= pte_pfn(*ptep
);
3620 pte_unmap_unlock(ptep
, ptl
);
3623 EXPORT_SYMBOL(follow_pfn
);
3625 #ifdef CONFIG_HAVE_IOREMAP_PROT
3626 int follow_phys(struct vm_area_struct
*vma
,
3627 unsigned long address
, unsigned int flags
,
3628 unsigned long *prot
, resource_size_t
*phys
)
3634 if (!(vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)))
3637 if (follow_pte(vma
->vm_mm
, address
, &ptep
, &ptl
))
3641 if ((flags
& FOLL_WRITE
) && !pte_write(pte
))
3644 *prot
= pgprot_val(pte_pgprot(pte
));
3645 *phys
= (resource_size_t
)pte_pfn(pte
) << PAGE_SHIFT
;
3649 pte_unmap_unlock(ptep
, ptl
);
3654 int generic_access_phys(struct vm_area_struct
*vma
, unsigned long addr
,
3655 void *buf
, int len
, int write
)
3657 resource_size_t phys_addr
;
3658 unsigned long prot
= 0;
3659 void __iomem
*maddr
;
3660 int offset
= addr
& (PAGE_SIZE
-1);
3662 if (follow_phys(vma
, addr
, write
, &prot
, &phys_addr
))
3665 maddr
= ioremap_prot(phys_addr
, PAGE_ALIGN(len
+ offset
), prot
);
3667 memcpy_toio(maddr
+ offset
, buf
, len
);
3669 memcpy_fromio(buf
, maddr
+ offset
, len
);
3674 EXPORT_SYMBOL_GPL(generic_access_phys
);
3678 * Access another process' address space as given in mm. If non-NULL, use the
3679 * given task for page fault accounting.
3681 static int __access_remote_vm(struct task_struct
*tsk
, struct mm_struct
*mm
,
3682 unsigned long addr
, void *buf
, int len
, int write
)
3684 struct vm_area_struct
*vma
;
3685 void *old_buf
= buf
;
3687 down_read(&mm
->mmap_sem
);
3688 /* ignore errors, just check how much was successfully transferred */
3690 int bytes
, ret
, offset
;
3692 struct page
*page
= NULL
;
3694 ret
= get_user_pages(tsk
, mm
, addr
, 1,
3695 write
, 1, &page
, &vma
);
3697 #ifndef CONFIG_HAVE_IOREMAP_PROT
3701 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3702 * we can access using slightly different code.
3704 vma
= find_vma(mm
, addr
);
3705 if (!vma
|| vma
->vm_start
> addr
)
3707 if (vma
->vm_ops
&& vma
->vm_ops
->access
)
3708 ret
= vma
->vm_ops
->access(vma
, addr
, buf
,
3716 offset
= addr
& (PAGE_SIZE
-1);
3717 if (bytes
> PAGE_SIZE
-offset
)
3718 bytes
= PAGE_SIZE
-offset
;
3722 copy_to_user_page(vma
, page
, addr
,
3723 maddr
+ offset
, buf
, bytes
);
3724 set_page_dirty_lock(page
);
3726 copy_from_user_page(vma
, page
, addr
,
3727 buf
, maddr
+ offset
, bytes
);
3730 page_cache_release(page
);
3736 up_read(&mm
->mmap_sem
);
3738 return buf
- old_buf
;
3742 * access_remote_vm - access another process' address space
3743 * @mm: the mm_struct of the target address space
3744 * @addr: start address to access
3745 * @buf: source or destination buffer
3746 * @len: number of bytes to transfer
3747 * @write: whether the access is a write
3749 * The caller must hold a reference on @mm.
3751 int access_remote_vm(struct mm_struct
*mm
, unsigned long addr
,
3752 void *buf
, int len
, int write
)
3754 return __access_remote_vm(NULL
, mm
, addr
, buf
, len
, write
);
3758 * Access another process' address space.
3759 * Source/target buffer must be kernel space,
3760 * Do not walk the page table directly, use get_user_pages
3762 int access_process_vm(struct task_struct
*tsk
, unsigned long addr
,
3763 void *buf
, int len
, int write
)
3765 struct mm_struct
*mm
;
3768 mm
= get_task_mm(tsk
);
3772 ret
= __access_remote_vm(tsk
, mm
, addr
, buf
, len
, write
);
3779 * Print the name of a VMA.
3781 void print_vma_addr(char *prefix
, unsigned long ip
)
3783 struct mm_struct
*mm
= current
->mm
;
3784 struct vm_area_struct
*vma
;
3787 * Do not print if we are in atomic
3788 * contexts (in exception stacks, etc.):
3790 if (preempt_count())
3793 down_read(&mm
->mmap_sem
);
3794 vma
= find_vma(mm
, ip
);
3795 if (vma
&& vma
->vm_file
) {
3796 struct file
*f
= vma
->vm_file
;
3797 char *buf
= (char *)__get_free_page(GFP_KERNEL
);
3801 p
= file_path(f
, buf
, PAGE_SIZE
);
3804 printk("%s%s[%lx+%lx]", prefix
, kbasename(p
),
3806 vma
->vm_end
- vma
->vm_start
);
3807 free_page((unsigned long)buf
);
3810 up_read(&mm
->mmap_sem
);
3813 #if defined(CONFIG_PROVE_LOCKING) || defined(CONFIG_DEBUG_ATOMIC_SLEEP)
3814 void __might_fault(const char *file
, int line
)
3817 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3818 * holding the mmap_sem, this is safe because kernel memory doesn't
3819 * get paged out, therefore we'll never actually fault, and the
3820 * below annotations will generate false positives.
3822 if (segment_eq(get_fs(), KERNEL_DS
))
3824 if (pagefault_disabled())
3826 __might_sleep(file
, line
, 0);
3827 #if defined(CONFIG_DEBUG_ATOMIC_SLEEP)
3829 might_lock_read(¤t
->mm
->mmap_sem
);
3832 EXPORT_SYMBOL(__might_fault
);
3835 #if defined(CONFIG_TRANSPARENT_HUGEPAGE) || defined(CONFIG_HUGETLBFS)
3836 static void clear_gigantic_page(struct page
*page
,
3838 unsigned int pages_per_huge_page
)
3841 struct page
*p
= page
;
3844 for (i
= 0; i
< pages_per_huge_page
;
3845 i
++, p
= mem_map_next(p
, page
, i
)) {
3847 clear_user_highpage(p
, addr
+ i
* PAGE_SIZE
);
3850 void clear_huge_page(struct page
*page
,
3851 unsigned long addr
, unsigned int pages_per_huge_page
)
3855 if (unlikely(pages_per_huge_page
> MAX_ORDER_NR_PAGES
)) {
3856 clear_gigantic_page(page
, addr
, pages_per_huge_page
);
3861 for (i
= 0; i
< pages_per_huge_page
; i
++) {
3863 clear_user_highpage(page
+ i
, addr
+ i
* PAGE_SIZE
);
3867 static void copy_user_gigantic_page(struct page
*dst
, struct page
*src
,
3869 struct vm_area_struct
*vma
,
3870 unsigned int pages_per_huge_page
)
3873 struct page
*dst_base
= dst
;
3874 struct page
*src_base
= src
;
3876 for (i
= 0; i
< pages_per_huge_page
; ) {
3878 copy_user_highpage(dst
, src
, addr
+ i
*PAGE_SIZE
, vma
);
3881 dst
= mem_map_next(dst
, dst_base
, i
);
3882 src
= mem_map_next(src
, src_base
, i
);
3886 void copy_user_huge_page(struct page
*dst
, struct page
*src
,
3887 unsigned long addr
, struct vm_area_struct
*vma
,
3888 unsigned int pages_per_huge_page
)
3892 if (unlikely(pages_per_huge_page
> MAX_ORDER_NR_PAGES
)) {
3893 copy_user_gigantic_page(dst
, src
, addr
, vma
,
3894 pages_per_huge_page
);
3899 for (i
= 0; i
< pages_per_huge_page
; i
++) {
3901 copy_user_highpage(dst
+ i
, src
+ i
, addr
+ i
*PAGE_SIZE
, vma
);
3904 #endif /* CONFIG_TRANSPARENT_HUGEPAGE || CONFIG_HUGETLBFS */
3906 #if USE_SPLIT_PTE_PTLOCKS && ALLOC_SPLIT_PTLOCKS
3908 static struct kmem_cache
*page_ptl_cachep
;
3910 void __init
ptlock_cache_init(void)
3912 page_ptl_cachep
= kmem_cache_create("page->ptl", sizeof(spinlock_t
), 0,
3916 bool ptlock_alloc(struct page
*page
)
3920 ptl
= kmem_cache_alloc(page_ptl_cachep
, GFP_KERNEL
);
3927 void ptlock_free(struct page
*page
)
3929 kmem_cache_free(page_ptl_cachep
, page
->ptl
);