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 #include <asm/fixmap.h>
   6
   7 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
   8 {
   9         return (pte_t *)__get_free_page(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO);
  10 }
  11
  12 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
  13 {
  14         struct page *pte;
  15
  16 #ifdef CONFIG_HIGHPTE
  17         pte = alloc_pages(GFP_KERNEL|__GFP_HIGHMEM|__GFP_REPEAT|__GFP_ZERO, 0);
  18 #else
  19         pte = alloc_pages(GFP_KERNEL|__GFP_REPEAT|__GFP_ZERO, 0);
  20 #endif
  21         if (pte)
  22                 pgtable_page_ctor(pte);
  23         return pte;
  24 }
  25
  26 void __pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
  27 {
  28         pgtable_page_dtor(pte);
  29         paravirt_release_pte(page_to_pfn(pte));
  30         tlb_remove_page(tlb, pte);
  31 }
  32
  33 #if PAGETABLE_LEVELS > 2
  34 void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
  35 {
  36         paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT);
  37         tlb_remove_page(tlb, virt_to_page(pmd));
  38 }
  39
  40 #if PAGETABLE_LEVELS > 3
  41 void __pud_free_tlb(struct mmu_gather *tlb, pud_t *pud)
  42 {
  43         paravirt_release_pud(__pa(pud) >> PAGE_SHIFT);
  44         tlb_remove_page(tlb, virt_to_page(pud));
  45 }
  46 #endif  /* PAGETABLE_LEVELS > 3 */
  47 #endif  /* PAGETABLE_LEVELS > 2 */
  48
  49 static inline void pgd_list_add(pgd_t *pgd)
  50 {
  51         struct page *page = virt_to_page(pgd);
  52
  53         list_add(&page->lru, &pgd_list);
  54 }
  55
  56 static inline void pgd_list_del(pgd_t *pgd)
  57 {
  58         struct page *page = virt_to_page(pgd);
  59
  60         list_del(&page->lru);
  61 }
  62
  63 #define UNSHARED_PTRS_PER_PGD                           \
  64         (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
  65
  66 static void pgd_ctor(pgd_t *pgd)
  67 {
  68         /* If the pgd points to a shared pagetable level (either the
  69            ptes in non-PAE, or shared PMD in PAE), then just copy the
  70            references from swapper_pg_dir. */
  71         if (PAGETABLE_LEVELS == 2 ||
  72             (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD) ||
  73             PAGETABLE_LEVELS == 4) {
  74                 clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY,
  75                                 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
  76                                 KERNEL_PGD_PTRS);
  77                 paravirt_alloc_pmd_clone(__pa(pgd) >> PAGE_SHIFT,
  78                                          __pa(swapper_pg_dir) >> PAGE_SHIFT,
  79                                          KERNEL_PGD_BOUNDARY,
  80                                          KERNEL_PGD_PTRS);
  81         }
  82
  83         /* list required to sync kernel mapping updates */
  84         if (!SHARED_KERNEL_PMD)
  85                 pgd_list_add(pgd);
  86 }
  87
  88 static void pgd_dtor(pgd_t *pgd)
  89 {
  90         unsigned long flags; /* can be called from interrupt context */
  91
  92         if (SHARED_KERNEL_PMD)
  93                 return;
  94
  95         spin_lock_irqsave(&pgd_lock, flags);
  96         pgd_list_del(pgd);
  97         spin_unlock_irqrestore(&pgd_lock, flags);
  98 }
  99
 100 /*
 101  * List of all pgd's needed for non-PAE so it can invalidate entries
 102  * in both cached and uncached pgd's; not needed for PAE since the
 103  * kernel pmd is shared. If PAE were not to share the pmd a similar
 104  * tactic would be needed. This is essentially codepath-based locking
 105  * against pageattr.c; it is the unique case in which a valid change
 106  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 107  * vmalloc faults work because attached pagetables are never freed.
 108  * -- wli
 109  */
 110
 111 #ifdef CONFIG_X86_PAE
 112 /*
 113  * In PAE mode, we need to do a cr3 reload (=tlb flush) when
 114  * updating the top-level pagetable entries to guarantee the
 115  * processor notices the update.  Since this is expensive, and
 116  * all 4 top-level entries are used almost immediately in a
 117  * new process's life, we just pre-populate them here.
 118  *
 119  * Also, if we're in a paravirt environment where the kernel pmd is
 120  * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
 121  * and initialize the kernel pmds here.
 122  */
 123 #define PREALLOCATED_PMDS       UNSHARED_PTRS_PER_PGD
 124
 125 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
 126 {
 127         paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
 128
 129         /* Note: almost everything apart from _PAGE_PRESENT is
 130            reserved at the pmd (PDPT) level. */
 131         set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
 132
 133         /*
 134          * According to Intel App note "TLBs, Paging-Structure Caches,
 135          * and Their Invalidation", April 2007, document 317080-001,
 136          * section 8.1: in PAE mode we explicitly have to flush the
 137          * TLB via cr3 if the top-level pgd is changed...
 138          */
 139         if (mm == current->active_mm)
 140                 write_cr3(read_cr3());
 141 }
 142 #else  /* !CONFIG_X86_PAE */
 143
 144 /* No need to prepopulate any pagetable entries in non-PAE modes. */
 145 #define PREALLOCATED_PMDS       0
 146
 147 #endif  /* CONFIG_X86_PAE */
 148
 149 static void free_pmds(pmd_t *pmds[])
 150 {
 151         int i;
 152
 153         for(i = 0; i < PREALLOCATED_PMDS; i++)
 154                 if (pmds[i])
 155                         free_page((unsigned long)pmds[i]);
 156 }
 157
 158 static int preallocate_pmds(pmd_t *pmds[])
 159 {
 160         int i;
 161         bool failed = false;
 162
 163         for(i = 0; i < PREALLOCATED_PMDS; i++) {
 164                 pmd_t *pmd = (pmd_t *)get_zeroed_page(GFP_KERNEL|__GFP_REPEAT);
 165                 if (pmd == NULL)
 166                         failed = true;
 167                 pmds[i] = pmd;
 168         }
 169
 170         if (failed) {
 171                 free_pmds(pmds);
 172                 return -ENOMEM;
 173         }
 174
 175         return 0;
 176 }
 177
 178 /*
 179  * Mop up any pmd pages which may still be attached to the pgd.
 180  * Normally they will be freed by munmap/exit_mmap, but any pmd we
 181  * preallocate which never got a corresponding vma will need to be
 182  * freed manually.
 183  */
 184 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
 185 {
 186         int i;
 187
 188         for(i = 0; i < PREALLOCATED_PMDS; i++) {
 189                 pgd_t pgd = pgdp[i];
 190
 191                 if (pgd_val(pgd) != 0) {
 192                         pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
 193
 194                         pgdp[i] = native_make_pgd(0);
 195
 196                         paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
 197                         pmd_free(mm, pmd);
 198                 }
 199         }
 200 }
 201
 202 static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[])
 203 {
 204         pud_t *pud;
 205         unsigned long addr;
 206         int i;
 207
 208         if (PREALLOCATED_PMDS == 0) /* Work around gcc-3.4.x bug */
 209                 return;
 210
 211         pud = pud_offset(pgd, 0);
 212
 213         for (addr = i = 0; i < PREALLOCATED_PMDS;
 214              i++, pud++, addr += PUD_SIZE) {
 215                 pmd_t *pmd = pmds[i];
 216
 217                 if (i >= KERNEL_PGD_BOUNDARY)
 218                         memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
 219                                sizeof(pmd_t) * PTRS_PER_PMD);
 220
 221                 pud_populate(mm, pud, pmd);
 222         }
 223 }
 224
 225 pgd_t *pgd_alloc(struct mm_struct *mm)
 226 {
 227         pgd_t *pgd;
 228         pmd_t *pmds[PREALLOCATED_PMDS];
 229         unsigned long flags;
 230
 231         pgd = (pgd_t *)__get_free_page(GFP_KERNEL | __GFP_ZERO);
 232
 233         if (pgd == NULL)
 234                 goto out;
 235
 236         mm->pgd = pgd;
 237
 238         if (preallocate_pmds(pmds) != 0)
 239                 goto out_free_pgd;
 240
 241         if (paravirt_pgd_alloc(mm) != 0)
 242                 goto out_free_pmds;
 243
 244         /*
 245          * Make sure that pre-populating the pmds is atomic with
 246          * respect to anything walking the pgd_list, so that they
 247          * never see a partially populated pgd.
 248          */
 249         spin_lock_irqsave(&pgd_lock, flags);
 250
 251         pgd_ctor(pgd);
 252         pgd_prepopulate_pmd(mm, pgd, pmds);
 253
 254         spin_unlock_irqrestore(&pgd_lock, flags);
 255
 256         return pgd;
 257
 258 out_free_pmds:
 259         free_pmds(pmds);
 260 out_free_pgd:
 261         free_page((unsigned long)pgd);
 262 out:
 263         return NULL;
 264 }
 265
 266 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
 267 {
 268         pgd_mop_up_pmds(mm, pgd);
 269         pgd_dtor(pgd);
 270         paravirt_pgd_free(mm, pgd);
 271         free_page((unsigned long)pgd);
 272 }
 273
 274 int ptep_set_access_flags(struct vm_area_struct *vma,
 275                           unsigned long address, pte_t *ptep,
 276                           pte_t entry, int dirty)
 277 {
 278         int changed = !pte_same(*ptep, entry);
 279
 280         if (changed && dirty) {
 281                 *ptep = entry;
 282                 pte_update_defer(vma->vm_mm, address, ptep);
 283                 flush_tlb_page(vma, address);
 284         }
 285
 286         return changed;
 287 }
 288
 289 int ptep_test_and_clear_young(struct vm_area_struct *vma,
 290                               unsigned long addr, pte_t *ptep)
 291 {
 292         int ret = 0;
 293
 294         if (pte_young(*ptep))
 295                 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
 296                                          (unsigned long *) &ptep->pte);
 297
 298         if (ret)
 299                 pte_update(vma->vm_mm, addr, ptep);
 300
 301         return ret;
 302 }
 303
 304 int ptep_clear_flush_young(struct vm_area_struct *vma,
 305                            unsigned long address, pte_t *ptep)
 306 {
 307         int young;
 308
 309         young = ptep_test_and_clear_young(vma, address, ptep);
 310         if (young)
 311                 flush_tlb_page(vma, address);
 312
 313         return young;
 314 }
 315
 316 /**
 317  * reserve_top_address - reserves a hole in the top of kernel address space
 318  * @reserve - size of hole to reserve
 319  *
 320  * Can be used to relocate the fixmap area and poke a hole in the top
 321  * of kernel address space to make room for a hypervisor.
 322  */
 323 void __init reserve_top_address(unsigned long reserve)
 324 {
 325 #ifdef CONFIG_X86_32
 326         BUG_ON(fixmaps_set > 0);
 327         printk(KERN_INFO "Reserving virtual address space above 0x%08x\n",
 328                (int)-reserve);
 329         __FIXADDR_TOP = -reserve - PAGE_SIZE;
 330         __VMALLOC_RESERVE += reserve;
 331 #endif
 332 }
 333
 334 int fixmaps_set;
 335
 336 void __native_set_fixmap(enum fixed_addresses idx, pte_t pte)
 337 {
 338         unsigned long address = __fix_to_virt(idx);
 339
 340         if (idx >= __end_of_fixed_addresses) {
 341                 BUG();
 342                 return;
 343         }
 344         set_pte_vaddr(address, pte);
 345         fixmaps_set++;
 346 }
 347
 348 void native_set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t flags)
 349 {
 350         __native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, flags));
 351 }