arch/sparc64/mm/hugetlbpage.c

   1 /*
   2  * SPARC64 Huge TLB page support.
   3  *
   4  * Copyright (C) 2002, 2003 David S. Miller (davem@redhat.com)
   5  */
   6
   7 #include <linux/config.h>
   8 #include <linux/init.h>
   9 #include <linux/module.h>
  10 #include <linux/fs.h>
  11 #include <linux/mm.h>
  12 #include <linux/hugetlb.h>
  13 #include <linux/pagemap.h>
  14 #include <linux/smp_lock.h>
  15 #include <linux/slab.h>
  16 #include <linux/sysctl.h>
  17
  18 #include <asm/mman.h>
  19 #include <asm/pgalloc.h>
  20 #include <asm/tlb.h>
  21 #include <asm/tlbflush.h>
  22 #include <asm/cacheflush.h>
  23
  24 static pte_t *huge_pte_alloc(struct mm_struct *mm, unsigned long addr)
  25 {
  26         pgd_t *pgd;
  27         pmd_t *pmd;
  28         pte_t *pte = NULL;
  29
  30         pgd = pgd_offset(mm, addr);
  31         if (pgd) {
  32                 pmd = pmd_alloc(mm, pgd, addr);
  33                 if (pmd)
  34                         pte = pte_alloc_map(mm, pmd, addr);
  35         }
  36         return pte;
  37 }
  38
  39 static pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr)
  40 {
  41         pgd_t *pgd;
  42         pmd_t *pmd;
  43         pte_t *pte = NULL;
  44
  45         pgd = pgd_offset(mm, addr);
  46         if (pgd) {
  47                 pmd = pmd_offset(pgd, addr);
  48                 if (pmd)
  49                         pte = pte_offset_map(pmd, addr);
  50         }
  51         return pte;
  52 }
  53
  54 #define mk_pte_huge(entry) do { pte_val(entry) |= _PAGE_SZHUGE; } while (0)
  55
  56 static void set_huge_pte(struct mm_struct *mm, struct vm_area_struct *vma,
  57                          struct page *page, pte_t * page_table, int write_access)
  58 {
  59         unsigned long i;
  60         pte_t entry;
  61
  62         mm->rss += (HPAGE_SIZE / PAGE_SIZE);
  63
  64         if (write_access)
  65                 entry = pte_mkwrite(pte_mkdirty(mk_pte(page,
  66                                                        vma->vm_page_prot)));
  67         else
  68                 entry = pte_wrprotect(mk_pte(page, vma->vm_page_prot));
  69         entry = pte_mkyoung(entry);
  70         mk_pte_huge(entry);
  71
  72         for (i = 0; i < (1 << HUGETLB_PAGE_ORDER); i++) {
  73                 set_pte(page_table, entry);
  74                 page_table++;
  75
  76                 pte_val(entry) += PAGE_SIZE;
  77         }
  78 }
  79
  80 /*
  81  * This function checks for proper alignment of input addr and len parameters.
  82  */
  83 int is_aligned_hugepage_range(unsigned long addr, unsigned long len)
  84 {
  85         if (len & ~HPAGE_MASK)
  86                 return -EINVAL;
  87         if (addr & ~HPAGE_MASK)
  88                 return -EINVAL;
  89         return 0;
  90 }
  91
  92 int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
  93                             struct vm_area_struct *vma)
  94 {
  95         pte_t *src_pte, *dst_pte, entry;
  96         struct page *ptepage;
  97         unsigned long addr = vma->vm_start;
  98         unsigned long end = vma->vm_end;
  99         int i;
 100
 101         while (addr < end) {
 102                 dst_pte = huge_pte_alloc(dst, addr);
 103                 if (!dst_pte)
 104                         goto nomem;
 105                 src_pte = huge_pte_offset(src, addr);
 106                 BUG_ON(!src_pte || pte_none(*src_pte));
 107                 entry = *src_pte;
 108                 ptepage = pte_page(entry);
 109                 get_page(ptepage);
 110                 for (i = 0; i < (1 << HUGETLB_PAGE_ORDER); i++) {
 111                         set_pte(dst_pte, entry);
 112                         pte_val(entry) += PAGE_SIZE;
 113                         dst_pte++;
 114                 }
 115                 dst->rss += (HPAGE_SIZE / PAGE_SIZE);
 116                 addr += HPAGE_SIZE;
 117         }
 118         return 0;
 119
 120 nomem:
 121         return -ENOMEM;
 122 }
 123
 124 int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
 125                         struct page **pages, struct vm_area_struct **vmas,
 126                         unsigned long *position, int *length, int i)
 127 {
 128         unsigned long vaddr = *position;
 129         int remainder = *length;
 130
 131         WARN_ON(!is_vm_hugetlb_page(vma));
 132
 133         while (vaddr < vma->vm_end && remainder) {
 134                 if (pages) {
 135                         pte_t *pte;
 136                         struct page *page;
 137
 138                         pte = huge_pte_offset(mm, vaddr);
 139
 140                         /* hugetlb should be locked, and hence, prefaulted */
 141                         BUG_ON(!pte || pte_none(*pte));
 142
 143                         page = pte_page(*pte);
 144
 145                         WARN_ON(!PageCompound(page));
 146
 147                         get_page(page);
 148                         pages[i] = page;
 149                 }
 150
 151                 if (vmas)
 152                         vmas[i] = vma;
 153
 154                 vaddr += PAGE_SIZE;
 155                 --remainder;
 156                 ++i;
 157         }
 158
 159         *length = remainder;
 160         *position = vaddr;
 161
 162         return i;
 163 }
 164
 165 struct page *follow_huge_addr(struct mm_struct *mm,
 166                               unsigned long address, int write)
 167 {
 168         return ERR_PTR(-EINVAL);
 169 }
 170
 171 int pmd_huge(pmd_t pmd)
 172 {
 173         return 0;
 174 }
 175
 176 struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address,
 177                              pmd_t *pmd, int write)
 178 {
 179         return NULL;
 180 }
 181
 182 void unmap_hugepage_range(struct vm_area_struct *vma,
 183                           unsigned long start, unsigned long end)
 184 {
 185         struct mm_struct *mm = vma->vm_mm;
 186         unsigned long address;
 187         pte_t *pte;
 188         struct page *page;
 189         int i;
 190
 191         BUG_ON(start & (HPAGE_SIZE - 1));
 192         BUG_ON(end & (HPAGE_SIZE - 1));
 193
 194         for (address = start; address < end; address += HPAGE_SIZE) {
 195                 pte = huge_pte_offset(mm, address);
 196                 BUG_ON(!pte);
 197                 if (pte_none(*pte))
 198                         continue;
 199                 page = pte_page(*pte);
 200                 put_page(page);
 201                 for (i = 0; i < (1 << HUGETLB_PAGE_ORDER); i++) {
 202                         pte_clear(pte);
 203                         pte++;
 204                 }
 205         }
 206         mm->rss -= (end - start) >> PAGE_SHIFT;
 207         flush_tlb_range(vma, start, end);
 208 }
 209
 210 int hugetlb_prefault(struct address_space *mapping, struct vm_area_struct *vma)
 211 {
 212         struct mm_struct *mm = current->mm;
 213         unsigned long addr;
 214         int ret = 0;
 215
 216         BUG_ON(vma->vm_start & ~HPAGE_MASK);
 217         BUG_ON(vma->vm_end & ~HPAGE_MASK);
 218
 219         spin_lock(&mm->page_table_lock);
 220         for (addr = vma->vm_start; addr < vma->vm_end; addr += HPAGE_SIZE) {
 221                 unsigned long idx;
 222                 pte_t *pte = huge_pte_alloc(mm, addr);
 223                 struct page *page;
 224
 225                 if (!pte) {
 226                         ret = -ENOMEM;
 227                         goto out;
 228                 }
 229                 if (!pte_none(*pte))
 230                         continue;
 231
 232                 idx = ((addr - vma->vm_start) >> HPAGE_SHIFT)
 233                         + (vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT));
 234                 page = find_get_page(mapping, idx);
 235                 if (!page) {
 236                         /* charge the fs quota first */
 237                         if (hugetlb_get_quota(mapping)) {
 238                                 ret = -ENOMEM;
 239                                 goto out;
 240                         }
 241                         page = alloc_huge_page();
 242                         if (!page) {
 243                                 hugetlb_put_quota(mapping);
 244                                 ret = -ENOMEM;
 245                                 goto out;
 246                         }
 247                         ret = add_to_page_cache(page, mapping, idx, GFP_ATOMIC);
 248                         if (! ret) {
 249                                 unlock_page(page);
 250                         } else {
 251                                 hugetlb_put_quota(mapping);
 252                                 free_huge_page(page);
 253                                 goto out;
 254                         }
 255                 }
 256                 set_huge_pte(mm, vma, page, pte, vma->vm_flags & VM_WRITE);
 257         }
 258 out:
 259         spin_unlock(&mm->page_table_lock);
 260         return ret;
 261 }