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 #define PGALLOC_GFP GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO
   8
   9 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
  10 {
  11         return (pte_t *)__get_free_page(PGALLOC_GFP);
  12 }
  13
  14 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
  15 {
  16         struct page *pte;
  17
  18 #ifdef CONFIG_HIGHPTE
  19         pte = alloc_pages(PGALLOC_GFP | __GFP_HIGHMEM, 0);
  20 #else
  21         pte = alloc_pages(PGALLOC_GFP, 0);
  22 #endif
  23         if (pte)
  24                 pgtable_page_ctor(pte);
  25         return pte;
  26 }
  27
  28 void __pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
  29 {
  30         pgtable_page_dtor(pte);
  31         paravirt_release_pte(page_to_pfn(pte));
  32         tlb_remove_page(tlb, pte);
  33 }
  34
  35 #if PAGETABLE_LEVELS > 2
  36 void __pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
  37 {
  38         paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT);
  39         tlb_remove_page(tlb, virt_to_page(pmd));
  40 }
  41
  42 #if PAGETABLE_LEVELS > 3
  43 void __pud_free_tlb(struct mmu_gather *tlb, pud_t *pud)
  44 {
  45         paravirt_release_pud(__pa(pud) >> PAGE_SHIFT);
  46         tlb_remove_page(tlb, virt_to_page(pud));
  47 }
  48 #endif  /* PAGETABLE_LEVELS > 3 */
  49 #endif  /* PAGETABLE_LEVELS > 2 */
  50
  51 static inline void pgd_list_add(pgd_t *pgd)
  52 {
  53         struct page *page = virt_to_page(pgd);
  54
  55         list_add(&page->lru, &pgd_list);
  56 }
  57
  58 static inline void pgd_list_del(pgd_t *pgd)
  59 {
  60         struct page *page = virt_to_page(pgd);
  61
  62         list_del(&page->lru);
  63 }
  64
  65 #define UNSHARED_PTRS_PER_PGD                           \
  66         (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
  67
  68 static void pgd_ctor(pgd_t *pgd)
  69 {
  70         /* If the pgd points to a shared pagetable level (either the
  71            ptes in non-PAE, or shared PMD in PAE), then just copy the
  72            references from swapper_pg_dir. */
  73         if (PAGETABLE_LEVELS == 2 ||
  74             (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD) ||
  75             PAGETABLE_LEVELS == 4) {
  76                 clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY,
  77                                 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
  78                                 KERNEL_PGD_PTRS);
  79                 paravirt_alloc_pmd_clone(__pa(pgd) >> PAGE_SHIFT,
  80                                          __pa(swapper_pg_dir) >> PAGE_SHIFT,
  81                                          KERNEL_PGD_BOUNDARY,
  82                                          KERNEL_PGD_PTRS);
  83         }
  84
  85         /* list required to sync kernel mapping updates */
  86         if (!SHARED_KERNEL_PMD)
  87                 pgd_list_add(pgd);
  88 }
  89
  90 static void pgd_dtor(pgd_t *pgd)
  91 {
  92         unsigned long flags; /* can be called from interrupt context */
  93
  94         if (SHARED_KERNEL_PMD)
  95                 return;
  96
  97         spin_lock_irqsave(&pgd_lock, flags);
  98         pgd_list_del(pgd);
  99         spin_unlock_irqrestore(&pgd_lock, flags);
 100 }
 101
 102 /*
 103  * List of all pgd's needed for non-PAE so it can invalidate entries
 104  * in both cached and uncached pgd's; not needed for PAE since the
 105  * kernel pmd is shared. If PAE were not to share the pmd a similar
 106  * tactic would be needed. This is essentially codepath-based locking
 107  * against pageattr.c; it is the unique case in which a valid change
 108  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 109  * vmalloc faults work because attached pagetables are never freed.
 110  * -- wli
 111  */
 112
 113 #ifdef CONFIG_X86_PAE
 114 /*
 115  * In PAE mode, we need to do a cr3 reload (=tlb flush) when
 116  * updating the top-level pagetable entries to guarantee the
 117  * processor notices the update.  Since this is expensive, and
 118  * all 4 top-level entries are used almost immediately in a
 119  * new process's life, we just pre-populate them here.
 120  *
 121  * Also, if we're in a paravirt environment where the kernel pmd is
 122  * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
 123  * and initialize the kernel pmds here.
 124  */
 125 #define PREALLOCATED_PMDS       UNSHARED_PTRS_PER_PGD
 126
 127 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
 128 {
 129         paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
 130
 131         /* Note: almost everything apart from _PAGE_PRESENT is
 132            reserved at the pmd (PDPT) level. */
 133         set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
 134
 135         preempt_disable();
 136         /*
 137          * According to Intel App note "TLBs, Paging-Structure Caches,
 138          * and Their Invalidation", April 2007, document 317080-001,
 139          * section 8.1: in PAE mode we explicitly have to flush the
 140          * TLB via cr3 if the top-level pgd is changed...
 141          */
 142         if (mm == current->active_mm)
 143                 write_cr3(read_cr3());
 144         preempt_enable();
 145 }
 146 #else  /* !CONFIG_X86_PAE */
 147
 148 /* No need to prepopulate any pagetable entries in non-PAE modes. */
 149 #define PREALLOCATED_PMDS       0
 150
 151 #endif  /* CONFIG_X86_PAE */
 152
 153 static void free_pmds(pmd_t *pmds[])
 154 {
 155         int i;
 156
 157         for(i = 0; i < PREALLOCATED_PMDS; i++)
 158                 if (pmds[i])
 159                         free_page((unsigned long)pmds[i]);
 160 }
 161
 162 static int preallocate_pmds(pmd_t *pmds[])
 163 {
 164         int i;
 165         bool failed = false;
 166
 167         for(i = 0; i < PREALLOCATED_PMDS; i++) {
 168                 pmd_t *pmd = (pmd_t *)__get_free_page(PGALLOC_GFP);
 169                 if (pmd == NULL)
 170                         failed = true;
 171                 pmds[i] = pmd;
 172         }
 173
 174         if (failed) {
 175                 free_pmds(pmds);
 176                 return -ENOMEM;
 177         }
 178
 179         return 0;
 180 }
 181
 182 /*
 183  * Mop up any pmd pages which may still be attached to the pgd.
 184  * Normally they will be freed by munmap/exit_mmap, but any pmd we
 185  * preallocate which never got a corresponding vma will need to be
 186  * freed manually.
 187  */
 188 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
 189 {
 190         int i;
 191
 192         for(i = 0; i < PREALLOCATED_PMDS; i++) {
 193                 pgd_t pgd = pgdp[i];
 194
 195                 if (pgd_val(pgd) != 0) {
 196                         pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
 197
 198                         pgdp[i] = native_make_pgd(0);
 199
 200                         paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
 201                         pmd_free(mm, pmd);
 202                 }
 203         }
 204 }
 205
 206 static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[])
 207 {
 208         pud_t *pud;
 209         unsigned long addr;
 210         int i;
 211
 212         if (PREALLOCATED_PMDS == 0) /* Work around gcc-3.4.x bug */
 213                 return;
 214
 215         pud = pud_offset(pgd, 0);
 216
 217         for (addr = i = 0; i < PREALLOCATED_PMDS;
 218              i++, pud++, addr += PUD_SIZE) {
 219                 pmd_t *pmd = pmds[i];
 220
 221                 if (i >= KERNEL_PGD_BOUNDARY)
 222                         memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
 223                                sizeof(pmd_t) * PTRS_PER_PMD);
 224
 225                 pud_populate(mm, pud, pmd);
 226         }
 227 }
 228
 229 pgd_t *pgd_alloc(struct mm_struct *mm)
 230 {
 231         pgd_t *pgd;
 232         pmd_t *pmds[PREALLOCATED_PMDS];
 233         unsigned long flags;
 234
 235         pgd = (pgd_t *)__get_free_page(PGALLOC_GFP);
 236
 237         if (pgd == NULL)
 238                 goto out;
 239
 240         mm->pgd = pgd;
 241
 242         if (preallocate_pmds(pmds) != 0)
 243                 goto out_free_pgd;
 244
 245         if (paravirt_pgd_alloc(mm) != 0)
 246                 goto out_free_pmds;
 247
 248         /*
 249          * Make sure that pre-populating the pmds is atomic with
 250          * respect to anything walking the pgd_list, so that they
 251          * never see a partially populated pgd.
 252          */
 253         spin_lock_irqsave(&pgd_lock, flags);
 254
 255         pgd_ctor(pgd);
 256         pgd_prepopulate_pmd(mm, pgd, pmds);
 257
 258         spin_unlock_irqrestore(&pgd_lock, flags);
 259
 260         return pgd;
 261
 262 out_free_pmds:
 263         free_pmds(pmds);
 264 out_free_pgd:
 265         free_page((unsigned long)pgd);
 266 out:
 267         return NULL;
 268 }
 269
 270 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
 271 {
 272         pgd_mop_up_pmds(mm, pgd);
 273         pgd_dtor(pgd);
 274         paravirt_pgd_free(mm, pgd);
 275         free_page((unsigned long)pgd);
 276 }
 277
 278 int ptep_set_access_flags(struct vm_area_struct *vma,
 279                           unsigned long address, pte_t *ptep,
 280                           pte_t entry, int dirty)
 281 {
 282         int changed = !pte_same(*ptep, entry);
 283
 284         if (changed && dirty) {
 285                 *ptep = entry;
 286                 pte_update_defer(vma->vm_mm, address, ptep);
 287                 flush_tlb_page(vma, address);
 288         }
 289
 290         return changed;
 291 }
 292
 293 int ptep_test_and_clear_young(struct vm_area_struct *vma,
 294                               unsigned long addr, pte_t *ptep)
 295 {
 296         int ret = 0;
 297
 298         if (pte_young(*ptep))
 299                 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
 300                                          (unsigned long *) &ptep->pte);
 301
 302         if (ret)
 303                 pte_update(vma->vm_mm, addr, ptep);
 304
 305         return ret;
 306 }
 307
 308 int ptep_clear_flush_young(struct vm_area_struct *vma,
 309                            unsigned long address, pte_t *ptep)
 310 {
 311         int young;
 312
 313         young = ptep_test_and_clear_young(vma, address, ptep);
 314         if (young)
 315                 flush_tlb_page(vma, address);
 316
 317         return young;
 318 }
 319
 320 /**
 321  * reserve_top_address - reserves a hole in the top of kernel address space
 322  * @reserve - size of hole to reserve
 323  *
 324  * Can be used to relocate the fixmap area and poke a hole in the top
 325  * of kernel address space to make room for a hypervisor.
 326  */
 327 void __init reserve_top_address(unsigned long reserve)
 328 {
 329 #ifdef CONFIG_X86_32
 330         BUG_ON(fixmaps_set > 0);
 331         printk(KERN_INFO "Reserving virtual address space above 0x%08x\n",
 332                (int)-reserve);
 333         __FIXADDR_TOP = -reserve - PAGE_SIZE;
 334         __VMALLOC_RESERVE += reserve;
 335 #endif
 336 }
 337
 338 int fixmaps_set;
 339
 340 void __native_set_fixmap(enum fixed_addresses idx, pte_t pte)
 341 {
 342         unsigned long address = __fix_to_virt(idx);
 343
 344         if (idx >= __end_of_fixed_addresses) {
 345                 BUG();
 346                 return;
 347         }
 348         set_pte_vaddr(address, pte);
 349         fixmaps_set++;
 350 }
 351
 352 void native_set_fixmap(enum fixed_addresses idx, phys_addr_t phys,
 353                        pgprot_t flags)
 354 {
 355         __native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, flags));
 356 }