4 * Explicit pagetable population and nonlinear (random) mappings support.
6 * started by Ingo Molnar, Copyright (C) 2002, 2003
10 #include <linux/swap.h>
11 #include <linux/file.h>
12 #include <linux/mman.h>
13 #include <linux/pagemap.h>
14 #include <linux/swapops.h>
15 #include <linux/rmap-locking.h>
16 #include <linux/module.h>
18 #include <asm/mmu_context.h>
19 #include <asm/cacheflush.h>
20 #include <asm/tlbflush.h>
22 static inline int zap_pte(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
23 unsigned long addr
, pte_t
*ptep
)
29 if (pte_present(pte
)) {
30 unsigned long pfn
= pte_pfn(pte
);
32 flush_cache_page(vma
, addr
);
33 pte
= ptep_get_and_clear(ptep
);
35 struct page
*page
= pfn_to_page(pfn
);
36 if (!PageReserved(page
)) {
39 page_remove_rmap(page
, ptep
);
40 page_cache_release(page
);
47 free_swap_and_cache(pte_to_swp_entry(pte
));
54 * Install a page to a given virtual memory address, release any
55 * previously existing mapping.
57 int install_page(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
58 unsigned long addr
, struct page
*page
, pgprot_t prot
)
60 int err
= -ENOMEM
, flush
;
65 struct pte_chain
*pte_chain
;
67 pte_chain
= pte_chain_alloc(GFP_KERNEL
);
70 pgd
= pgd_offset(mm
, addr
);
71 spin_lock(&mm
->page_table_lock
);
73 pmd
= pmd_alloc(mm
, pgd
, addr
);
77 pte
= pte_alloc_map(mm
, pmd
, addr
);
81 flush
= zap_pte(mm
, vma
, addr
, pte
);
84 flush_icache_page(vma
, page
);
85 set_pte(pte
, mk_pte(page
, prot
));
86 pte_chain
= page_add_rmap(page
, pte
, pte_chain
);
90 flush_tlb_page(vma
, addr
);
91 update_mmu_cache(vma
, addr
, pte_val
);
92 spin_unlock(&mm
->page_table_lock
);
93 pte_chain_free(pte_chain
);
97 spin_unlock(&mm
->page_table_lock
);
98 pte_chain_free(pte_chain
);
102 EXPORT_SYMBOL(install_page
);
106 * Install a file pte to a given virtual memory address, release any
107 * previously existing mapping.
109 int install_file_pte(struct mm_struct
*mm
, struct vm_area_struct
*vma
,
110 unsigned long addr
, unsigned long pgoff
, pgprot_t prot
)
112 int err
= -ENOMEM
, flush
;
118 pgd
= pgd_offset(mm
, addr
);
119 spin_lock(&mm
->page_table_lock
);
121 pmd
= pmd_alloc(mm
, pgd
, addr
);
125 pte
= pte_alloc_map(mm
, pmd
, addr
);
129 flush
= zap_pte(mm
, vma
, addr
, pte
);
131 set_pte(pte
, pgoff_to_pte(pgoff
));
135 flush_tlb_page(vma
, addr
);
136 update_mmu_cache(vma
, addr
, pte_val
);
137 spin_unlock(&mm
->page_table_lock
);
141 spin_unlock(&mm
->page_table_lock
);
147 * sys_remap_file_pages - remap arbitrary pages of a shared backing store
148 * file within an existing vma.
149 * @start: start of the remapped virtual memory range
150 * @size: size of the remapped virtual memory range
151 * @prot: new protection bits of the range
152 * @pgoff: to be mapped page of the backing store file
153 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
155 * this syscall works purely via pagetables, so it's the most efficient
156 * way to map the same (large) file into a given virtual window. Unlike
157 * mmap()/mremap() it does not create any new vmas. The new mappings are
158 * also safe across swapout.
160 * NOTE: the 'prot' parameter right now is ignored, and the vma's default
161 * protection is used. Arbitrary protections might be implemented in the
164 long sys_remap_file_pages(unsigned long start
, unsigned long size
,
165 unsigned long __prot
, unsigned long pgoff
, unsigned long flags
)
167 struct mm_struct
*mm
= current
->mm
;
168 unsigned long end
= start
+ size
;
169 struct vm_area_struct
*vma
;
175 * Sanitize the syscall parameters:
177 start
= start
& PAGE_MASK
;
178 size
= size
& PAGE_MASK
;
180 /* Does the address range wrap, or is the span zero-sized? */
181 if (start
+ size
<= start
)
184 /* Can we represent this offset inside this architecture's pte's? */
185 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
186 if (pgoff
+ (size
>> PAGE_SHIFT
) >= (1UL << PTE_FILE_MAX_BITS
))
190 /* We need down_write() to change vma->vm_flags. */
191 down_write(&mm
->mmap_sem
);
192 vma
= find_vma(mm
, start
);
195 * Make sure the vma is shared, that it supports prefaulting,
196 * and that the remapped range is valid and fully within
197 * the single existing vma:
199 if (vma
&& (vma
->vm_flags
& VM_SHARED
) &&
200 vma
->vm_ops
&& vma
->vm_ops
->populate
&&
201 end
> start
&& start
>= vma
->vm_start
&&
202 end
<= vma
->vm_end
) {
204 /* Must set VM_NONLINEAR before any pages are populated. */
205 if (pgoff
!= ((start
- vma
->vm_start
) >> PAGE_SHIFT
) + vma
->vm_pgoff
)
206 vma
->vm_flags
|= VM_NONLINEAR
;
208 /* ->populate can take a long time, so downgrade the lock. */
209 downgrade_write(&mm
->mmap_sem
);
210 err
= vma
->vm_ops
->populate(vma
, start
, size
,
212 pgoff
, flags
& MAP_NONBLOCK
);
215 * We can't clear VM_NONLINEAR because we'd have to do
216 * it after ->populate completes, and that would prevent
217 * downgrading the lock. (Locks can't be upgraded).
219 up_read(&mm
->mmap_sem
);
221 up_write(&mm
->mmap_sem
);