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>
17 #include <linux/syscalls.h>
18 #include <linux/vs_memory.h>
20 #include <asm/mmu_context.h>
21 #include <asm/cacheflush.h>
22 #include <asm/tlbflush.h>
24 static inline void zap_pte(struct mm_struct *mm, struct vm_area_struct *vma,
25 unsigned long addr, pte_t *ptep)
31 if (pte_present(pte)) {
32 unsigned long pfn = pte_pfn(pte);
34 flush_cache_page(vma, addr, pfn);
35 pte = ptep_clear_flush(vma, addr, ptep);
37 struct page *page = pfn_to_page(pfn);
38 if (!PageReserved(page)) {
41 page_remove_rmap(page);
42 page_cache_release(page);
43 dec_mm_counter(mm, rss);
48 free_swap_and_cache(pte_to_swp_entry(pte));
49 pte_clear(mm, addr, ptep);
54 * Install a file 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)
69 pgd = pgd_offset(mm, addr);
70 spin_lock(&mm->page_table_lock);
72 if (!vx_rsspages_avail(mm, 1))
75 pud = pud_alloc(mm, pgd, addr);
79 pmd = pmd_alloc(mm, pud, addr);
83 pte = pte_alloc_map(mm, pmd, addr);
88 * This page may have been truncated. Tell the
93 inode = vma->vm_file->f_mapping->host;
94 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
95 if (!page->mapping || page->index >= size)
99 zap_pte(mm, vma, addr, pte);
101 inc_mm_counter(mm,rss);
102 flush_icache_page(vma, page);
103 set_pte_at(mm, addr, pte, mk_pte(page, prot));
104 page_add_file_rmap(page);
107 update_mmu_cache(vma, addr, pte_val);
111 spin_unlock(&mm->page_table_lock);
114 EXPORT_SYMBOL(install_page);
118 * Install a file pte to a given virtual memory address, release any
119 * previously existing mapping.
121 int install_file_pte(struct mm_struct *mm, struct vm_area_struct *vma,
122 unsigned long addr, unsigned long pgoff, pgprot_t prot)
131 pgd = pgd_offset(mm, addr);
132 spin_lock(&mm->page_table_lock);
134 pud = pud_alloc(mm, pgd, addr);
138 pmd = pmd_alloc(mm, pud, addr);
142 pte = pte_alloc_map(mm, pmd, addr);
146 zap_pte(mm, vma, addr, pte);
148 set_pte_at(mm, addr, pte, pgoff_to_pte(pgoff));
151 update_mmu_cache(vma, addr, pte_val);
152 spin_unlock(&mm->page_table_lock);
156 spin_unlock(&mm->page_table_lock);
162 * sys_remap_file_pages - remap arbitrary pages of a shared backing store
163 * file within an existing vma.
164 * @start: start of the remapped virtual memory range
165 * @size: size of the remapped virtual memory range
166 * @prot: new protection bits of the range
167 * @pgoff: to be mapped page of the backing store file
168 * @flags: 0 or MAP_NONBLOCKED - the later will cause no IO.
170 * this syscall works purely via pagetables, so it's the most efficient
171 * way to map the same (large) file into a given virtual window. Unlike
172 * mmap()/mremap() it does not create any new vmas. The new mappings are
173 * also safe across swapout.
175 * NOTE: the 'prot' parameter right now is ignored, and the vma's default
176 * protection is used. Arbitrary protections might be implemented in the
179 asmlinkage long sys_remap_file_pages(unsigned long start, unsigned long size,
180 unsigned long __prot, unsigned long pgoff, unsigned long flags)
182 struct mm_struct *mm = current->mm;
183 struct address_space *mapping;
184 unsigned long end = start + size;
185 struct vm_area_struct *vma;
187 int has_write_lock = 0;
192 * Sanitize the syscall parameters:
194 start = start & PAGE_MASK;
195 size = size & PAGE_MASK;
197 /* Does the address range wrap, or is the span zero-sized? */
198 if (start + size <= start)
201 /* Can we represent this offset inside this architecture's pte's? */
202 #if PTE_FILE_MAX_BITS < BITS_PER_LONG
203 if (pgoff + (size >> PAGE_SHIFT) >= (1UL << PTE_FILE_MAX_BITS))
207 /* We need down_write() to change vma->vm_flags. */
208 down_read(&mm->mmap_sem);
210 vma = find_vma(mm, start);
213 * Make sure the vma is shared, that it supports prefaulting,
214 * and that the remapped range is valid and fully within
215 * the single existing vma. vm_private_data is used as a
216 * swapout cursor in a VM_NONLINEAR vma (unless VM_RESERVED
217 * or VM_LOCKED, but VM_LOCKED could be revoked later on).
219 if (vma && (vma->vm_flags & VM_SHARED) &&
220 (!vma->vm_private_data ||
221 (vma->vm_flags & (VM_NONLINEAR|VM_RESERVED))) &&
222 vma->vm_ops && vma->vm_ops->populate &&
223 end > start && start >= vma->vm_start &&
224 end <= vma->vm_end) {
226 /* Must set VM_NONLINEAR before any pages are populated. */
227 if (pgoff != linear_page_index(vma, start) &&
228 !(vma->vm_flags & VM_NONLINEAR)) {
229 if (!has_write_lock) {
230 up_read(&mm->mmap_sem);
231 down_write(&mm->mmap_sem);
235 mapping = vma->vm_file->f_mapping;
236 spin_lock(&mapping->i_mmap_lock);
237 flush_dcache_mmap_lock(mapping);
238 vma->vm_flags |= VM_NONLINEAR;
239 vma_prio_tree_remove(vma, &mapping->i_mmap);
240 vma_nonlinear_insert(vma, &mapping->i_mmap_nonlinear);
241 flush_dcache_mmap_unlock(mapping);
242 spin_unlock(&mapping->i_mmap_lock);
245 err = vma->vm_ops->populate(vma, start, size,
247 pgoff, flags & MAP_NONBLOCK);
250 * We can't clear VM_NONLINEAR because we'd have to do
251 * it after ->populate completes, and that would prevent
252 * downgrading the lock. (Locks can't be upgraded).
255 if (likely(!has_write_lock))
256 up_read(&mm->mmap_sem);
258 up_write(&mm->mmap_sem);