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 #define UNSHARED_PTRS_PER_PGD                           \
  63         (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
  64
  65 static void pgd_ctor(void *p)
  66 {
  67         pgd_t *pgd = p;
  68         unsigned long flags;
  69
  70         /* Clear usermode parts of PGD */
  71         memset(pgd, 0, KERNEL_PGD_BOUNDARY*sizeof(pgd_t));
  72
  73         spin_lock_irqsave(&pgd_lock, flags);
  74
  75         /* If the pgd points to a shared pagetable level (either the
  76            ptes in non-PAE, or shared PMD in PAE), then just copy the
  77            references from swapper_pg_dir. */
  78         if (PAGETABLE_LEVELS == 2 ||
  79             (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD) ||
  80             PAGETABLE_LEVELS == 4) {
  81                 clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY,
  82                                 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
  83                                 KERNEL_PGD_PTRS);
  84                 paravirt_alloc_pmd_clone(__pa(pgd) >> PAGE_SHIFT,
  85                                          __pa(swapper_pg_dir) >> PAGE_SHIFT,
  86                                          KERNEL_PGD_BOUNDARY,
  87                                          KERNEL_PGD_PTRS);
  88         }
  89
  90         /* list required to sync kernel mapping updates */
  91         if (!SHARED_KERNEL_PMD)
  92                 pgd_list_add(pgd);
  93
  94         spin_unlock_irqrestore(&pgd_lock, flags);
  95 }
  96
  97 static void pgd_dtor(void *pgd)
  98 {
  99         unsigned long flags; /* can be called from interrupt context */
 100
 101         if (SHARED_KERNEL_PMD)
 102                 return;
 103
 104         spin_lock_irqsave(&pgd_lock, flags);
 105         pgd_list_del(pgd);
 106         spin_unlock_irqrestore(&pgd_lock, flags);
 107 }
 108
 109 /*
 110  * List of all pgd's needed for non-PAE so it can invalidate entries
 111  * in both cached and uncached pgd's; not needed for PAE since the
 112  * kernel pmd is shared. If PAE were not to share the pmd a similar
 113  * tactic would be needed. This is essentially codepath-based locking
 114  * against pageattr.c; it is the unique case in which a valid change
 115  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 116  * vmalloc faults work because attached pagetables are never freed.
 117  * -- wli
 118  */
 119
 120 #ifdef CONFIG_X86_PAE
 121 /*
 122  * Mop up any pmd pages which may still be attached to the pgd.
 123  * Normally they will be freed by munmap/exit_mmap, but any pmd we
 124  * preallocate which never got a corresponding vma will need to be
 125  * freed manually.
 126  */
 127 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
 128 {
 129         int i;
 130
 131         for(i = 0; i < UNSHARED_PTRS_PER_PGD; i++) {
 132                 pgd_t pgd = pgdp[i];
 133
 134                 if (pgd_val(pgd) != 0) {
 135                         pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
 136
 137                         pgdp[i] = native_make_pgd(0);
 138
 139                         paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
 140                         pmd_free(mm, pmd);
 141                 }
 142         }
 143 }
 144
 145 /*
 146  * In PAE mode, we need to do a cr3 reload (=tlb flush) when
 147  * updating the top-level pagetable entries to guarantee the
 148  * processor notices the update.  Since this is expensive, and
 149  * all 4 top-level entries are used almost immediately in a
 150  * new process's life, we just pre-populate them here.
 151  *
 152  * Also, if we're in a paravirt environment where the kernel pmd is
 153  * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
 154  * and initialize the kernel pmds here.
 155  */
 156 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
 157 {
 158         pud_t *pud;
 159         unsigned long addr;
 160         int i;
 161
 162         pud = pud_offset(pgd, 0);
 163         for (addr = i = 0; i < UNSHARED_PTRS_PER_PGD;
 164              i++, pud++, addr += PUD_SIZE) {
 165                 pmd_t *pmd = pmd_alloc_one(mm, addr);
 166
 167                 if (!pmd) {
 168                         pgd_mop_up_pmds(mm, pgd);
 169                         return 0;
 170                 }
 171
 172                 if (i >= KERNEL_PGD_BOUNDARY)
 173                         memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
 174                                sizeof(pmd_t) * PTRS_PER_PMD);
 175
 176                 pud_populate(mm, pud, pmd);
 177         }
 178
 179         return 1;
 180 }
 181
 182 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
 183 {
 184         paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
 185
 186         /* Note: almost everything apart from _PAGE_PRESENT is
 187            reserved at the pmd (PDPT) level. */
 188         set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
 189
 190         /*
 191          * According to Intel App note "TLBs, Paging-Structure Caches,
 192          * and Their Invalidation", April 2007, document 317080-001,
 193          * section 8.1: in PAE mode we explicitly have to flush the
 194          * TLB via cr3 if the top-level pgd is changed...
 195          */
 196         if (mm == current->active_mm)
 197                 write_cr3(read_cr3());
 198 }
 199 #else  /* !CONFIG_X86_PAE */
 200 /* No need to prepopulate any pagetable entries in non-PAE modes. */
 201 static int pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd)
 202 {
 203         return 1;
 204 }
 205
 206 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgd)
 207 {
 208 }
 209 #endif  /* CONFIG_X86_PAE */
 210
 211 pgd_t *pgd_alloc(struct mm_struct *mm)
 212 {
 213         pgd_t *pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
 214
 215         /* so that alloc_pmd can use it */
 216         mm->pgd = pgd;
 217         if (pgd)
 218                 pgd_ctor(pgd);
 219
 220         if (pgd && !pgd_prepopulate_pmd(mm, pgd)) {
 221                 pgd_dtor(pgd);
 222                 free_page((unsigned long)pgd);
 223                 pgd = NULL;
 224         }
 225
 226         return pgd;
 227 }
 228
 229 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
 230 {
 231         pgd_mop_up_pmds(mm, pgd);
 232         pgd_dtor(pgd);
 233         free_page((unsigned long)pgd);
 234 }
 235
 236 int ptep_set_access_flags(struct vm_area_struct *vma,
 237                           unsigned long address, pte_t *ptep,
 238                           pte_t entry, int dirty)
 239 {
 240         int changed = !pte_same(*ptep, entry);
 241
 242         if (changed && dirty) {
 243                 *ptep = entry;
 244                 pte_update_defer(vma->vm_mm, address, ptep);
 245                 flush_tlb_page(vma, address);
 246         }
 247
 248         return changed;
 249 }
 250
 251 int ptep_test_and_clear_young(struct vm_area_struct *vma,
 252                               unsigned long addr, pte_t *ptep)
 253 {
 254         int ret = 0;
 255
 256         if (pte_young(*ptep))
 257                 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
 258                                          &ptep->pte);
 259
 260         if (ret)
 261                 pte_update(vma->vm_mm, addr, ptep);
 262
 263         return ret;
 264 }
 265
 266 int ptep_clear_flush_young(struct vm_area_struct *vma,
 267                            unsigned long address, pte_t *ptep)
 268 {
 269         int young;
 270
 271         young = ptep_test_and_clear_young(vma, address, ptep);
 272         if (young)
 273                 flush_tlb_page(vma, address);
 274
 275         return young;
 276 }