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);
47 vx_rsspages_add(mm, HPAGE_SIZE / PAGE_SIZE);
50 pte_mkwrite(pte_mkdirty(mk_pte(page, vma->vm_page_prot)));
52 entry = pte_wrprotect(mk_pte(page, vma->vm_page_prot));
53 entry = pte_mkyoung(entry);
55 set_pte(page_table, entry);
59 * This function checks for proper alignment of input addr and len parameters.
61 int is_aligned_hugepage_range(unsigned long addr, unsigned long len)
63 if (len & ~HPAGE_MASK)
65 if (addr & ~HPAGE_MASK)
70 int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
71 struct vm_area_struct *vma)
73 pte_t *src_pte, *dst_pte, entry;
75 unsigned long addr = vma->vm_start;
76 unsigned long end = vma->vm_end;
79 dst_pte = huge_pte_alloc(dst, addr);
82 src_pte = huge_pte_offset(src, addr);
84 ptepage = pte_page(entry);
86 set_pte(dst_pte, entry);
87 // dst->rss += (HPAGE_SIZE / PAGE_SIZE);
88 vx_rsspages_add(dst, HPAGE_SIZE / PAGE_SIZE);
98 follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
99 struct page **pages, struct vm_area_struct **vmas,
100 unsigned long *position, int *length, int i)
102 unsigned long vpfn, vaddr = *position;
103 int remainder = *length;
105 WARN_ON(!is_vm_hugetlb_page(vma));
107 vpfn = vaddr/PAGE_SIZE;
108 while (vaddr < vma->vm_end && remainder) {
114 pte = huge_pte_offset(mm, vaddr);
116 /* hugetlb should be locked, and hence, prefaulted */
117 WARN_ON(!pte || pte_none(*pte));
119 page = &pte_page(*pte)[vpfn % (HPAGE_SIZE/PAGE_SIZE)];
121 WARN_ON(!PageCompound(page));
142 #if 0 /* This is just for testing */
144 follow_huge_addr(struct mm_struct *mm, unsigned long address, int write)
146 unsigned long start = address;
150 struct vm_area_struct *vma;
152 if (! mm->used_hugetlb)
153 return ERR_PTR(-EINVAL);
155 vma = find_vma(mm, addr);
156 if (!vma || !is_vm_hugetlb_page(vma))
157 return ERR_PTR(-EINVAL);
159 pte = huge_pte_offset(mm, address);
161 /* hugetlb should be locked, and hence, prefaulted */
162 WARN_ON(!pte || pte_none(*pte));
164 page = &pte_page(*pte)[vpfn % (HPAGE_SIZE/PAGE_SIZE)];
166 WARN_ON(!PageCompound(page));
171 int pmd_huge(pmd_t pmd)
177 follow_huge_pmd(struct mm_struct *mm, unsigned long address,
178 pmd_t *pmd, int write)
186 follow_huge_addr(struct mm_struct *mm, unsigned long address, int write)
188 return ERR_PTR(-EINVAL);
191 int pmd_huge(pmd_t pmd)
193 return !!(pmd_val(pmd) & _PAGE_PSE);
197 follow_huge_pmd(struct mm_struct *mm, unsigned long address,
198 pmd_t *pmd, int write)
202 page = pte_page(*(pte_t *)pmd);
204 page += ((address & ~HPAGE_MASK) >> PAGE_SHIFT);
209 void unmap_hugepage_range(struct vm_area_struct *vma,
210 unsigned long start, unsigned long end)
212 struct mm_struct *mm = vma->vm_mm;
213 unsigned long address;
217 BUG_ON(start & (HPAGE_SIZE - 1));
218 BUG_ON(end & (HPAGE_SIZE - 1));
220 for (address = start; address < end; address += HPAGE_SIZE) {
221 pte = ptep_get_and_clear(huge_pte_offset(mm, address));
224 page = pte_page(pte);
227 // mm->rss -= (end - start) >> PAGE_SHIFT;
228 vx_rsspages_sub(mm, (end - start) >> PAGE_SHIFT);
229 flush_tlb_range(vma, start, end);
232 int hugetlb_prefault(struct address_space *mapping, struct vm_area_struct *vma)
234 struct mm_struct *mm = current->mm;
238 BUG_ON(vma->vm_start & ~HPAGE_MASK);
239 BUG_ON(vma->vm_end & ~HPAGE_MASK);
241 spin_lock(&mm->page_table_lock);
242 for (addr = vma->vm_start; addr < vma->vm_end; addr += HPAGE_SIZE) {
244 pte_t *pte = huge_pte_alloc(mm, addr);
254 idx = ((addr - vma->vm_start) >> HPAGE_SHIFT)
255 + (vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT));
256 page = find_get_page(mapping, idx);
258 /* charge the fs quota first */
259 if (hugetlb_get_quota(mapping)) {
263 page = alloc_huge_page();
265 hugetlb_put_quota(mapping);
269 ret = add_to_page_cache(page, mapping, idx, GFP_ATOMIC);
272 hugetlb_put_quota(mapping);
273 free_huge_page(page);
277 set_huge_pte(mm, vma, page, pte, vma->vm_flags & VM_WRITE);
280 spin_unlock(&mm->page_table_lock);