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.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 void 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_clear_flush(vma, addr, ptep);
35 struct page *page = pfn_to_page(pfn);
36 if (!PageReserved(page)) {
39 page_remove_rmap(page);
40 page_cache_release(page);
47 free_swap_and_cache(pte_to_swp_entry(pte));
53 * Install a file page to a given virtual memory address, release any
54 * previously existing mapping.
56 int install_page(struct mm_struct *mm, struct vm_area_struct *vma,
57 unsigned long addr, struct page *page, pgprot_t prot)
65 pgd = pgd_offset(mm, addr);
66 spin_lock(&mm->page_table_lock);
68 if (!vx_rsspages_avail(mm, 1))
71 pmd = pmd_alloc(mm, pgd, addr);
75 pte = pte_alloc_map(mm, pmd, addr);
79 zap_pte(mm, vma, addr, pte);
83 flush_icache_page(vma, page);
84 set_pte(pte, mk_pte(page, prot));
85 page_add_file_rmap(page);
88 update_mmu_cache(vma, addr, pte_val);
92 spin_unlock(&mm->page_table_lock);
95 EXPORT_SYMBOL(install_page);
99 * Install a file pte to a given virtual memory address, release any
100 * previously existing mapping.
102 int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma,
103 unsigned long addr, unsigned long pgoff, pgprot_t prot)
111 pgd = pgd_offset(mm, addr);
112 spin_lock(&mm->page_table_lock);
114 pmd = pmd_alloc(mm, pgd, addr);
118 pte = pte_alloc_map(mm, pmd, addr);
122 zap_pte(mm, vma, addr, pte);
124 set_pte(pte, pgoff_to_pte(pgoff));
127 update_mmu_cache(vma, addr, pte_val);
128 spin_unlock(&mm->page_table_lock);
132 spin_unlock(&mm->page_table_lock);
138 * sys_remap_file_pages - remap arbitrary pages of a shared backing store
139 * file within an existing vma.
140 * @start: start of the remapped virtual memory range
141 * @size: size of the remapped virtual memory range
142 * @prot: new protection bits of the range
143 * @pgoff: to be mapped page of the backing store file
144 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
146 * this syscall works purely via pagetables, so it's the most efficient
147 * way to map the same (large) file into a given virtual window. Unlike
148 * mmap()/mremap() it does not create any new vmas. The new mappings are
149 * also safe across swapout.
151 * NOTE: the 'prot' parameter right now is ignored, and the vma's default
152 * protection is used. Arbitrary protections might be implemented in the
155 asmlinkage long sys_remap_file_pages(unsigned long start, unsigned long size,
156 unsigned long __prot, unsigned long pgoff, unsigned long flags)
158 struct mm_struct *mm = current->mm;
159 struct address_space *mapping;
160 unsigned long end = start + size;
161 struct vm_area_struct *vma;
167 * Sanitize the syscall parameters:
169 start = start & PAGE_MASK;
170 size = size & PAGE_MASK;
172 /* Does the address range wrap, or is the span zero-sized? */
173 if (start + size <= start)
176 /* Can we represent this offset inside this architecture's pte's? */
177 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
178 if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
182 /* We need down_write() to change vma->vm_flags. */
183 down_write(&mm->mmap_sem);
184 vma = find_vma(mm, start);
187 * Make sure the vma is shared, that it supports prefaulting,
188 * and that the remapped range is valid and fully within
189 * the single existing vma. vm_private_data is used as a
190 * swapout cursor in a VM_NONLINEAR vma (unless VM_RESERVED
191 * or VM_LOCKED, but VM_LOCKED could be revoked later on).
193 if (vma && (vma->vm_flags & VM_SHARED) &&
194 (!vma->vm_private_data || (vma->vm_flags & VM_RESERVED)) &&
195 vma->vm_ops && vma->vm_ops->populate &&
196 end > start && start >= vma->vm_start &&
197 end <= vma->vm_end) {
199 /* Must set VM_NONLINEAR before any pages are populated. */
200 if (pgoff != linear_page_index(vma, start) &&
201 !(vma->vm_flags & VM_NONLINEAR)) {
202 mapping = vma->vm_file->f_mapping;
203 spin_lock(&mapping->i_mmap_lock);
204 flush_dcache_mmap_lock(mapping);
205 vma->vm_flags |= VM_NONLINEAR;
206 vma_prio_tree_remove(vma, &mapping->i_mmap);
207 vma_prio_tree_init(vma);
208 list_add_tail(&vma->shared.vm_set.list,
209 &mapping->i_mmap_nonlinear);
210 flush_dcache_mmap_unlock(mapping);
211 spin_unlock(&mapping->i_mmap_lock);
214 /* ->populate can take a long time, so downgrade the lock. */
215 downgrade_write(&mm->mmap_sem);
216 err = vma->vm_ops->populate(vma, start, size,
218 pgoff, flags & MAP_NONBLOCK);
221 * We can't clear VM_NONLINEAR because we'd have to do
222 * it after ->populate completes, and that would prevent
223 * downgrading the lock. (Locks can't be upgraded).
225 up_read(&mm->mmap_sem);
227 up_write(&mm->mmap_sem);