2 * IA-32 Huge TLB Page Support for Kernel.
4 * Copyright (C) 2002, Rohit Seth <rohit.seth@intel.com>
7 #include <linux/config.h>
8 #include <linux/init.h>
11 #include <linux/hugetlb.h>
12 #include <linux/pagemap.h>
13 #include <linux/smp_lock.h>
14 #include <linux/slab.h>
15 #include <linux/err.h>
16 #include <linux/sysctl.h>
18 #include <asm/pgalloc.h>
20 #include <asm/tlbflush.h>
22 static pte_t *huge_pte_alloc(struct mm_struct *mm, unsigned long addr)
27 pgd = pgd_offset(mm, addr);
28 pmd = pmd_alloc(mm, pgd, addr);
32 static pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
37 pgd = pgd_offset(mm, addr);
38 pmd = pmd_offset(pgd, addr);
42 static void set_huge_pte(struct mm_struct *mm, struct vm_area_struct *vma, struct page *page, pte_t * page_table, int write_access)
46 mm->rss += (HPAGE_SIZE / PAGE_SIZE);
49 pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot)));
51 entry = pte_wrprotect(mk_pte(page, vma->vm_page_prot));
52 entry = pte_mkyoung(entry);
54 set_pte(page_table, entry);
58 * This function checks for proper alignment of input addr and len parameters.
60 int is_aligned_hugepage_range(unsigned long addr, unsigned long len)
62 if (len & ~HPAGE_MASK)
64 if (addr & ~HPAGE_MASK)
69 int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
70 struct vm_area_struct *vma)
72 pte_t *src_pte, *dst_pte, entry;
74 unsigned long addr = vma->vm_start;
75 unsigned long end = vma->vm_end;
78 dst_pte = huge_pte_alloc(dst, addr);
81 src_pte = huge_pte_offset(src, addr);
83 ptepage = pte_page(entry);
85 set_pte(dst_pte, entry);
86 dst->rss += (HPAGE_SIZE / PAGE_SIZE);
96 follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
97 struct page **pages, struct vm_area_struct **vmas,
98 unsigned long *position, int *length, int i)
100 unsigned long vpfn, vaddr = *position;
101 int remainder = *length;
103 WARN_ON(!is_vm_hugetlb_page(vma));
105 vpfn = vaddr/PAGE_SIZE;
106 while (vaddr < vma->vm_end && remainder) {
112 pte = huge_pte_offset(mm, vaddr);
114 /* hugetlb should be locked, and hence, prefaulted */
115 WARN_ON(!pte || pte_none(*pte));
117 page = &pte_page(*pte)[vpfn % (HPAGE_SIZE/PAGE_SIZE)];
119 WARN_ON(!PageCompound(page));
140 #if 0 /* This is just for testing */
142 follow_huge_addr(struct mm_struct *mm, unsigned long address, int write)
144 unsigned long start = address;
148 struct vm_area_struct *vma;
150 if (! mm->used_hugetlb)
151 return ERR_PTR(-EINVAL);
153 vma = find_vma(mm, addr);
154 if (!vma || !is_vm_hugetlb_page(vma))
155 return ERR_PTR(-EINVAL);
157 pte = huge_pte_offset(mm, address);
159 /* hugetlb should be locked, and hence, prefaulted */
160 WARN_ON(!pte || pte_none(*pte));
162 page = &pte_page(*pte)[vpfn % (HPAGE_SIZE/PAGE_SIZE)];
164 WARN_ON(!PageCompound(page));
169 int pmd_huge(pmd_t pmd)
175 follow_huge_pmd(struct mm_struct *mm, unsigned long address,
176 pmd_t *pmd, int write)
184 follow_huge_addr(struct mm_struct *mm, unsigned long address, int write)
186 return ERR_PTR(-EINVAL);
189 int pmd_huge(pmd_t pmd)
191 return !!(pmd_val(pmd) & _PAGE_PSE);
195 follow_huge_pmd(struct mm_struct *mm, unsigned long address,
196 pmd_t *pmd, int write)
200 page = pte_page(*(pte_t *)pmd);
202 page += ((address & ~HPAGE_MASK) >> PAGE_SHIFT);
207 void unmap_hugepage_range(struct vm_area_struct *vma,
208 unsigned long start, unsigned long end)
210 struct mm_struct *mm = vma->vm_mm;
211 unsigned long address;
215 BUG_ON(start & (HPAGE_SIZE - 1));
216 BUG_ON(end & (HPAGE_SIZE - 1));
218 for (address = start; address < end; address += HPAGE_SIZE) {
219 pte = ptep_get_and_clear(huge_pte_offset(mm, address));
222 page = pte_page(pte);
225 mm->rss -= (end - start) >> PAGE_SHIFT;
226 flush_tlb_range(vma, start, end);
229 int hugetlb_prefault(struct address_space *mapping, struct vm_area_struct *vma)
231 struct mm_struct *mm = current->mm;
235 BUG_ON(vma->vm_start & ~HPAGE_MASK);
236 BUG_ON(vma->vm_end & ~HPAGE_MASK);
238 spin_lock(&mm->page_table_lock);
239 for (addr = vma->vm_start; addr < vma->vm_end; addr += HPAGE_SIZE) {
241 pte_t *pte = huge_pte_alloc(mm, addr);
251 idx = ((addr - vma->vm_start) >> HPAGE_SHIFT)
252 + (vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT));
253 page = find_get_page(mapping, idx);
255 /* charge the fs quota first */
256 if (hugetlb_get_quota(mapping)) {
260 page = alloc_huge_page();
262 hugetlb_put_quota(mapping);
266 ret = add_to_page_cache(page, mapping, idx, GFP_ATOMIC);
269 hugetlb_put_quota(mapping);
270 free_huge_page(page);
274 set_huge_pte(mm, vma, page, pte, vma->vm_flags & VM_WRITE);
277 spin_unlock(&mm->page_table_lock);