arch/x86/mm/pgtable.c

   1 #include <linux/mm.h>
   2 #include <asm/pgalloc.h>
   3 #include <asm/pgtable.h>
   4 #include <asm/tlb.h>
   5
   6 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
   7 {
   8         return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
   9 }
  10
  11 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
  12 {
  13         struct page *pte;
  14
  15 #ifdef CONFIG_HIGHPTE
  16         pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
  17 #else
  18         pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
  19 #endif
  20         if (pte)
  21                 pgtable_page_ctor(pte);
  22         return pte;
  23 }
  24
  25 void __pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
  26 {
  27         pgtable_page_dtor(pte);
  28         paravirt_release_pte(page_to_pfn(pte));
  29         tlb_remove_page(tlb, pte);
  30 }
  31
  32 #if PAGETABLE_LEVELS > 2
  33 void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
  34 {
  35         paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT);
  36         tlb_remove_page(tlb, virt_to_page(pmd));
  37 }
  38
  39 #if PAGETABLE_LEVELS > 3
  40 void __pud_free_tlb(struct mmu_gather *tlb, pud_t *pud)
  41 {
  42         paravirt_release_pud(__pa(pud) >> PAGE_SHIFT);
  43         tlb_remove_page(tlb, virt_to_page(pud));
  44 }
  45 #endif  /* PAGETABLE_LEVELS > 3 */
  46 #endif  /* PAGETABLE_LEVELS > 2 */
  47
  48 static inline void pgd_list_add(pgd_t *pgd)
  49 {
  50         struct page *page = virt_to_page(pgd);
  51
  52         list_add(&page->lru, &pgd_list);
  53 }
  54
  55 static inline void pgd_list_del(pgd_t *pgd)
  56 {
  57         struct page *page = virt_to_page(pgd);
  58
  59         list_del(&page->lru);
  60 }
  61
  62 #ifdef CONFIG_X86_64
  63 pgd_t *pgd_alloc(struct mm_struct *mm)
  64 {
  65         unsigned boundary;
  66         pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT);
  67         unsigned long flags;
  68         if (!pgd)
  69                 return NULL;
  70         spin_lock_irqsave(&pgd_lock, flags);
  71         pgd_list_add(pgd);
  72         spin_unlock_irqrestore(&pgd_lock, flags);
  73         /*
  74          * Copy kernel pointers in from init.
  75          * Could keep a freelist or slab cache of those because the kernel
  76          * part never changes.
  77          */
  78         boundary = pgd_index(__PAGE_OFFSET);
  79         memset(pgd, 0, boundary * sizeof(pgd_t));
  80         memcpy(pgd + boundary,
  81                init_level4_pgt + boundary,
  82                (PTRS_PER_PGD - boundary) * sizeof(pgd_t));
  83         return pgd;
  84 }
  85
  86 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
  87 {
  88         unsigned long flags;
  89         BUG_ON((unsigned long)pgd & (PAGE_SIZE-1));
  90         spin_lock_irqsave(&pgd_lock, flags);
  91         pgd_list_del(pgd);
  92         spin_unlock_irqrestore(&pgd_lock, flags);
  93         free_page((unsigned long)pgd);
  94 }
  95 #else
  96 /*
  97  * List of all pgd's needed for non-PAE so it can invalidate entries
  98  * in both cached and uncached pgd's; not needed for PAE since the
  99  * kernel pmd is shared. If PAE were not to share the pmd a similar
 100  * tactic would be needed. This is essentially codepath-based locking
 101  * against pageattr.c; it is the unique case in which a valid change
 102  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 103  * vmalloc faults work because attached pagetables are never freed.
 104  * -- wli
 105  */
 106 #define UNSHARED_PTRS_PER_PGD                           \
 107         (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
 108
 109 static void pgd_ctor(void *p)
 110 {
 111         pgd_t *pgd = p;
 112         unsigned long flags;
 113
 114         /* Clear usermode parts of PGD */
 115         memset(pgd, 0, KERNEL_PGD_BOUNDARY*sizeof(pgd_t));
 116
 117         spin_lock_irqsave(&pgd_lock, flags);
 118
 119         /* If the pgd points to a shared pagetable level (either the
 120            ptes in non-PAE, or shared PMD in PAE), then just copy the
 121            references from swapper_pg_dir. */
 122         if (PAGETABLE_LEVELS == 2 ||
 123             (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD)) {
 124                 clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY,
 125                                 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
 126                                 KERNEL_PGD_PTRS);
 127                 paravirt_alloc_pmd_clone(__pa(pgd) >> PAGE_SHIFT,
 128                                          __pa(swapper_pg_dir) >> PAGE_SHIFT,
 129                                          KERNEL_PGD_BOUNDARY,
 130                                          KERNEL_PGD_PTRS);
 131         }
 132
 133         /* list required to sync kernel mapping updates */
 134         if (!SHARED_KERNEL_PMD)
 135                 pgd_list_add(pgd);
 136
 137         spin_unlock_irqrestore(&pgd_lock, flags);
 138 }
 139
 140 static void pgd_dtor(void *pgd)
 141 {
 142         unsigned long flags; /* can be called from interrupt context */
 143
 144         if (SHARED_KERNEL_PMD)
 145                 return;
 146
 147         spin_lock_irqsave(&pgd_lock, flags);
 148         pgd_list_del(pgd);
 149         spin_unlock_irqrestore(&pgd_lock, flags);
 150 }
 151
 152 #ifdef CONFIG_X86_PAE
 153 /*
 154  * Mop up any pmd pages which may still be attached to the pgd.
 155  * Normally they will be freed by munmap/exit_mmap, but any pmd we
 156  * preallocate which never got a corresponding vma will need to be
 157  * freed manually.
 158  */
 159 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
 160 {
 161         int i;
 162
 163         for(i = 0; i < UNSHARED_PTRS_PER_PGD; i++) {
 164                 pgd_t pgd = pgdp[i];
 165
 166                 if (pgd_val(pgd) != 0) {
 167                         pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
 168
 169                         pgdp[i] = native_make_pgd(0);
 170
 171                         paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
 172                         pmd_free(mm, pmd);
 173                 }
 174         }
 175 }
 176
 177 /*
 178  * In PAE mode, we need to do a cr3 reload (=tlb flush) when
 179  * updating the top-level pagetable entries to guarantee the
 180  * processor notices the update.  Since this is expensive, and
 181  * all 4 top-level entries are used almost immediately in a
 182  * new process's life, we just pre-populate them here.
 183  *
 184  * Also, if we're in a paravirt environment where the kernel pmd is
 185  * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
 186  * and initialize the kernel pmds here.
 187  */
 188 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
 189 {
 190         pud_t *pud;
 191         unsigned long addr;
 192         int i;
 193
 194         pud = pud_offset(pgd, 0);
 195         for (addr = i = 0; i < UNSHARED_PTRS_PER_PGD;
 196              i++, pud++, addr += PUD_SIZE) {
 197                 pmd_t *pmd = pmd_alloc_one(mm, addr);
 198
 199                 if (!pmd) {
 200                         pgd_mop_up_pmds(mm, pgd);
 201                         return 0;
 202                 }
 203
 204                 if (i >= KERNEL_PGD_BOUNDARY)
 205                         memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
 206                                sizeof(pmd_t) * PTRS_PER_PMD);
 207
 208                 pud_populate(mm, pud, pmd);
 209         }
 210
 211         return 1;
 212 }
 213
 214 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
 215 {
 216         paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
 217
 218         /* Note: almost everything apart from _PAGE_PRESENT is
 219            reserved at the pmd (PDPT) level. */
 220         set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
 221
 222         /*
 223          * According to Intel App note "TLBs, Paging-Structure Caches,
 224          * and Their Invalidation", April 2007, document 317080-001,
 225          * section 8.1: in PAE mode we explicitly have to flush the
 226          * TLB via cr3 if the top-level pgd is changed...
 227          */
 228         if (mm == current->active_mm)
 229                 write_cr3(read_cr3());
 230 }
 231 #else  /* !CONFIG_X86_PAE */
 232 /* No need to prepopulate any pagetable entries in non-PAE modes. */
 233 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
 234 {
 235         return 1;
 236 }
 237
 238 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgd)
 239 {
 240 }
 241 #endif  /* CONFIG_X86_PAE */
 242
 243 pgd_t *pgd_alloc(struct mm_struct *mm)
 244 {
 245         pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
 246
 247         /* so that alloc_pmd can use it */
 248         mm->pgd = pgd;
 249         if (pgd)
 250                 pgd_ctor(pgd);
 251
 252         if (pgd && !pgd_prepopulate_pmd(mm, pgd)) {
 253                 pgd_dtor(pgd);
 254                 free_page((unsigned long)pgd);
 255                 pgd = NULL;
 256         }
 257
 258         return pgd;
 259 }
 260
 261 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
 262 {
 263         pgd_mop_up_pmds(mm, pgd);
 264         pgd_dtor(pgd);
 265         free_page((unsigned long)pgd);
 266 }
 267 #endif
 268
 269 int ptep_set_access_flags(struct vm_area_struct *vma,
 270                           unsigned long address, pte_t *ptep,
 271                           pte_t entry, int dirty)
 272 {
 273         int changed = !pte_same(*ptep, entry);
 274
 275         if (changed && dirty) {
 276                 *ptep = entry;
 277                 pte_update_defer(vma->vm_mm, address, ptep);
 278                 flush_tlb_page(vma, address);
 279         }
 280
 281         return changed;
 282 }
 283
 284 int ptep_test_and_clear_young(struct vm_area_struct *vma,
 285                               unsigned long addr, pte_t *ptep)
 286 {
 287         int ret = 0;
 288
 289         if (pte_young(*ptep))
 290                 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
 291                                          &ptep->pte);
 292
 293         if (ret)
 294                 pte_update(vma->vm_mm, addr, ptep);
 295
 296         return ret;
 297 }
 298
 299 int ptep_clear_flush_young(struct vm_area_struct *vma,
 300                            unsigned long address, pte_t *ptep)
 301 {
 302         int young;
 303
 304         young = ptep_test_and_clear_young(vma, address, ptep);
 305         if (young)
 306                 flush_tlb_page(vma, address);
 307
 308         return young;
 309 }