arch/x86/mm/pgtable.c

   1 #include <linux/mm.h>
   2 #include <linux/gfp.h>
   3 #include <asm/pgalloc.h>
   4 #include <asm/pgtable.h>
   5 #include <asm/tlb.h>
   6 #include <asm/fixmap.h>
   7
   8 #define PGALLOC_GFP GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO
   9
  10 #ifdef CONFIG_HIGHPTE
  11 #define PGALLOC_USER_GFP __GFP_HIGHMEM
  12 #else
  13 #define PGALLOC_USER_GFP 0
  14 #endif
  15
  16 gfp_t __userpte_alloc_gfp = PGALLOC_GFP | PGALLOC_USER_GFP;
  17
  18 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
  19 {
  20         return (pte_t *)__get_free_page(PGALLOC_GFP);
  21 }
  22
  23 pgtable_t pte_alloc_one(struct mm_struct *mm, unsigned long address)
  24 {
  25         struct page *pte;
  26
  27         pte = alloc_pages(__userpte_alloc_gfp, 0);
  28         if (pte)
  29                 pgtable_page_ctor(pte);
  30         return pte;
  31 }
  32
  33 static int __init setup_userpte(char *arg)
  34 {
  35         if (!arg)
  36                 return -EINVAL;
  37
  38         /*
  39          * "userpte=nohigh" disables allocation of user pagetables in
  40          * high memory.
  41          */
  42         if (strcmp(arg, "nohigh") == 0)
  43                 __userpte_alloc_gfp &= ~__GFP_HIGHMEM;
  44         else
  45                 return -EINVAL;
  46         return 0;
  47 }
  48 early_param("userpte", setup_userpte);
  49
  50 void ___pte_free_tlb(struct mmu_gather *tlb, struct page *pte)
  51 {
  52         pgtable_page_dtor(pte);
  53         paravirt_release_pte(page_to_pfn(pte));
  54         tlb_remove_page(tlb, pte);
  55 }
  56
  57 #if PAGETABLE_LEVELS > 2
  58 void ___pmd_free_tlb(struct mmu_gather *tlb, pmd_t *pmd)
  59 {
  60         paravirt_release_pmd(__pa(pmd) >> PAGE_SHIFT);
  61         tlb_remove_page(tlb, virt_to_page(pmd));
  62 }
  63
  64 #if PAGETABLE_LEVELS > 3
  65 void ___pud_free_tlb(struct mmu_gather *tlb, pud_t *pud)
  66 {
  67         paravirt_release_pud(__pa(pud) >> PAGE_SHIFT);
  68         tlb_remove_page(tlb, virt_to_page(pud));
  69 }
  70 #endif  /* PAGETABLE_LEVELS > 3 */
  71 #endif  /* PAGETABLE_LEVELS > 2 */
  72
  73 static inline void pgd_list_add(pgd_t *pgd)
  74 {
  75         struct page *page = virt_to_page(pgd);
  76
  77         list_add(&page->lru, &pgd_list);
  78 }
  79
  80 static inline void pgd_list_del(pgd_t *pgd)
  81 {
  82         struct page *page = virt_to_page(pgd);
  83
  84         list_del(&page->lru);
  85 }
  86
  87 #define UNSHARED_PTRS_PER_PGD                           \
  88         (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
  89
  90 static void pgd_ctor(pgd_t *pgd)
  91 {
  92         /* If the pgd points to a shared pagetable level (either the
  93            ptes in non-PAE, or shared PMD in PAE), then just copy the
  94            references from swapper_pg_dir. */
  95         if (PAGETABLE_LEVELS == 2 ||
  96             (PAGETABLE_LEVELS == 3 && SHARED_KERNEL_PMD) ||
  97             PAGETABLE_LEVELS == 4) {
  98                 clone_pgd_range(pgd + KERNEL_PGD_BOUNDARY,
  99                                 swapper_pg_dir + KERNEL_PGD_BOUNDARY,
 100                                 KERNEL_PGD_PTRS);
 101                 paravirt_alloc_pmd_clone(__pa(pgd) >> PAGE_SHIFT,
 102                                          __pa(swapper_pg_dir) >> PAGE_SHIFT,
 103                                          KERNEL_PGD_BOUNDARY,
 104                                          KERNEL_PGD_PTRS);
 105         }
 106
 107         /* list required to sync kernel mapping updates */
 108         if (!SHARED_KERNEL_PMD)
 109                 pgd_list_add(pgd);
 110 }
 111
 112 static void pgd_dtor(pgd_t *pgd)
 113 {
 114         unsigned long flags; /* can be called from interrupt context */
 115
 116         if (SHARED_KERNEL_PMD)
 117                 return;
 118
 119         spin_lock_irqsave(&pgd_lock, flags);
 120         pgd_list_del(pgd);
 121         spin_unlock_irqrestore(&pgd_lock, flags);
 122 }
 123
 124 /*
 125  * List of all pgd's needed for non-PAE so it can invalidate entries
 126  * in both cached and uncached pgd's; not needed for PAE since the
 127  * kernel pmd is shared. If PAE were not to share the pmd a similar
 128  * tactic would be needed. This is essentially codepath-based locking
 129  * against pageattr.c; it is the unique case in which a valid change
 130  * of kernel pagetables can't be lazily synchronized by vmalloc faults.
 131  * vmalloc faults work because attached pagetables are never freed.
 132  * -- wli
 133  */
 134
 135 #ifdef CONFIG_X86_PAE
 136 /*
 137  * In PAE mode, we need to do a cr3 reload (=tlb flush) when
 138  * updating the top-level pagetable entries to guarantee the
 139  * processor notices the update.  Since this is expensive, and
 140  * all 4 top-level entries are used almost immediately in a
 141  * new process's life, we just pre-populate them here.
 142  *
 143  * Also, if we're in a paravirt environment where the kernel pmd is
 144  * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
 145  * and initialize the kernel pmds here.
 146  */
 147 #define PREALLOCATED_PMDS       UNSHARED_PTRS_PER_PGD
 148
 149 void pud_populate(struct mm_struct *mm, pud_t *pudp, pmd_t *pmd)
 150 {
 151         paravirt_alloc_pmd(mm, __pa(pmd) >> PAGE_SHIFT);
 152
 153         /* Note: almost everything apart from _PAGE_PRESENT is
 154            reserved at the pmd (PDPT) level. */
 155         set_pud(pudp, __pud(__pa(pmd) | _PAGE_PRESENT));
 156
 157         /*
 158          * According to Intel App note "TLBs, Paging-Structure Caches,
 159          * and Their Invalidation", April 2007, document 317080-001,
 160          * section 8.1: in PAE mode we explicitly have to flush the
 161          * TLB via cr3 if the top-level pgd is changed...
 162          */
 163         if (mm == current->active_mm)
 164                 write_cr3(read_cr3());
 165 }
 166 #else  /* !CONFIG_X86_PAE */
 167
 168 /* No need to prepopulate any pagetable entries in non-PAE modes. */
 169 #define PREALLOCATED_PMDS       0
 170
 171 #endif  /* CONFIG_X86_PAE */
 172
 173 static void free_pmds(pmd_t *pmds[])
 174 {
 175         int i;
 176
 177         for(i = 0; i < PREALLOCATED_PMDS; i++)
 178                 if (pmds[i])
 179                         free_page((unsigned long)pmds[i]);
 180 }
 181
 182 static int preallocate_pmds(pmd_t *pmds[])
 183 {
 184         int i;
 185         bool failed = false;
 186
 187         for(i = 0; i < PREALLOCATED_PMDS; i++) {
 188                 pmd_t *pmd = (pmd_t *)__get_free_page(PGALLOC_GFP);
 189                 if (pmd == NULL)
 190                         failed = true;
 191                 pmds[i] = pmd;
 192         }
 193
 194         if (failed) {
 195                 free_pmds(pmds);
 196                 return -ENOMEM;
 197         }
 198
 199         return 0;
 200 }
 201
 202 /*
 203  * Mop up any pmd pages which may still be attached to the pgd.
 204  * Normally they will be freed by munmap/exit_mmap, but any pmd we
 205  * preallocate which never got a corresponding vma will need to be
 206  * freed manually.
 207  */
 208 static void pgd_mop_up_pmds(struct mm_struct *mm, pgd_t *pgdp)
 209 {
 210         int i;
 211
 212         for(i = 0; i < PREALLOCATED_PMDS; i++) {
 213                 pgd_t pgd = pgdp[i];
 214
 215                 if (pgd_val(pgd) != 0) {
 216                         pmd_t *pmd = (pmd_t *)pgd_page_vaddr(pgd);
 217
 218                         pgdp[i] = native_make_pgd(0);
 219
 220                         paravirt_release_pmd(pgd_val(pgd) >> PAGE_SHIFT);
 221                         pmd_free(mm, pmd);
 222                 }
 223         }
 224 }
 225
 226 static void pgd_prepopulate_pmd(struct mm_struct *mm, pgd_t *pgd, pmd_t *pmds[])
 227 {
 228         pud_t *pud;
 229         unsigned long addr;
 230         int i;
 231
 232         if (PREALLOCATED_PMDS == 0) /* Work around gcc-3.4.x bug */
 233                 return;
 234
 235         pud = pud_offset(pgd, 0);
 236
 237         for (addr = i = 0; i < PREALLOCATED_PMDS;
 238              i++, pud++, addr += PUD_SIZE) {
 239                 pmd_t *pmd = pmds[i];
 240
 241                 if (i >= KERNEL_PGD_BOUNDARY)
 242                         memcpy(pmd, (pmd_t *)pgd_page_vaddr(swapper_pg_dir[i]),
 243                                sizeof(pmd_t) * PTRS_PER_PMD);
 244
 245                 pud_populate(mm, pud, pmd);
 246         }
 247 }
 248
 249 pgd_t *pgd_alloc(struct mm_struct *mm)
 250 {
 251         pgd_t *pgd;
 252         pmd_t *pmds[PREALLOCATED_PMDS];
 253         unsigned long flags;
 254
 255         pgd = (pgd_t *)__get_free_page(PGALLOC_GFP);
 256
 257         if (pgd == NULL)
 258                 goto out;
 259
 260         mm->pgd = pgd;
 261
 262         if (preallocate_pmds(pmds) != 0)
 263                 goto out_free_pgd;
 264
 265         if (paravirt_pgd_alloc(mm) != 0)
 266                 goto out_free_pmds;
 267
 268         /*
 269          * Make sure that pre-populating the pmds is atomic with
 270          * respect to anything walking the pgd_list, so that they
 271          * never see a partially populated pgd.
 272          */
 273         spin_lock_irqsave(&pgd_lock, flags);
 274
 275         pgd_ctor(pgd);
 276         pgd_prepopulate_pmd(mm, pgd, pmds);
 277
 278         spin_unlock_irqrestore(&pgd_lock, flags);
 279
 280         return pgd;
 281
 282 out_free_pmds:
 283         free_pmds(pmds);
 284 out_free_pgd:
 285         free_page((unsigned long)pgd);
 286 out:
 287         return NULL;
 288 }
 289
 290 void pgd_free(struct mm_struct *mm, pgd_t *pgd)
 291 {
 292         pgd_mop_up_pmds(mm, pgd);
 293         pgd_dtor(pgd);
 294         paravirt_pgd_free(mm, pgd);
 295         free_page((unsigned long)pgd);
 296 }
 297
 298 int ptep_set_access_flags(struct vm_area_struct *vma,
 299                           unsigned long address, pte_t *ptep,
 300                           pte_t entry, int dirty)
 301 {
 302         int changed = !pte_same(*ptep, entry);
 303
 304         if (changed && dirty) {
 305                 *ptep = entry;
 306                 pte_update_defer(vma->vm_mm, address, ptep);
 307                 flush_tlb_page(vma, address);
 308         }
 309
 310         return changed;
 311 }
 312
 313 int ptep_test_and_clear_young(struct vm_area_struct *vma,
 314                               unsigned long addr, pte_t *ptep)
 315 {
 316         int ret = 0;
 317
 318         if (pte_young(*ptep))
 319                 ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,
 320                                          (unsigned long *) &ptep->pte);
 321
 322         if (ret)
 323                 pte_update(vma->vm_mm, addr, ptep);
 324
 325         return ret;
 326 }
 327
 328 int ptep_clear_flush_young(struct vm_area_struct *vma,
 329                            unsigned long address, pte_t *ptep)
 330 {
 331         int young;
 332
 333         young = ptep_test_and_clear_young(vma, address, ptep);
 334         if (young)
 335                 flush_tlb_page(vma, address);
 336
 337         return young;
 338 }
 339
 340 /**
 341  * reserve_top_address - reserves a hole in the top of kernel address space
 342  * @reserve - size of hole to reserve
 343  *
 344  * Can be used to relocate the fixmap area and poke a hole in the top
 345  * of kernel address space to make room for a hypervisor.
 346  */
 347 void __init reserve_top_address(unsigned long reserve)
 348 {
 349 #ifdef CONFIG_X86_32
 350         BUG_ON(fixmaps_set > 0);
 351         printk(KERN_INFO "Reserving virtual address space above 0x%08x\n",
 352                (int)-reserve);
 353         __FIXADDR_TOP = -reserve - PAGE_SIZE;
 354 #endif
 355 }
 356
 357 int fixmaps_set;
 358
 359 void __native_set_fixmap(enum fixed_addresses idx, pte_t pte)
 360 {
 361         unsigned long address = __fix_to_virt(idx);
 362
 363         if (idx >= __end_of_fixed_addresses) {
 364                 BUG();
 365                 return;
 366         }
 367         set_pte_vaddr(address, pte);
 368         fixmaps_set++;
 369 }
 370
 371 void native_set_fixmap(enum fixed_addresses idx, phys_addr_t phys,
 372                        pgprot_t flags)
 373 {
 374         __native_set_fixmap(idx, pfn_pte(phys >> PAGE_SHIFT, flags));
 375 }