arch/x86/mm/init_32.c

   1 /*
   2  *
   3  *  Copyright (C) 1995  Linus Torvalds
   4  *
   5  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/signal.h>
  10 #include <linux/sched.h>
  11 #include <linux/kernel.h>
  12 #include <linux/errno.h>
  13 #include <linux/string.h>
  14 #include <linux/types.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/mman.h>
  17 #include <linux/mm.h>
  18 #include <linux/hugetlb.h>
  19 #include <linux/swap.h>
  20 #include <linux/smp.h>
  21 #include <linux/init.h>
  22 #include <linux/highmem.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/pci.h>
  25 #include <linux/pfn.h>
  26 #include <linux/poison.h>
  27 #include <linux/bootmem.h>
  28 #include <linux/slab.h>
  29 #include <linux/proc_fs.h>
  30 #include <linux/memory_hotplug.h>
  31 #include <linux/initrd.h>
  32 #include <linux/cpumask.h>
  33
  34 #include <asm/asm.h>
  35 #include <asm/bios_ebda.h>
  36 #include <asm/processor.h>
  37 #include <asm/system.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/dma.h>
  41 #include <asm/fixmap.h>
  42 #include <asm/e820.h>
  43 #include <asm/apic.h>
  44 #include <asm/bugs.h>
  45 #include <asm/tlb.h>
  46 #include <asm/tlbflush.h>
  47 #include <asm/pgalloc.h>
  48 #include <asm/sections.h>
  49 #include <asm/paravirt.h>
  50 #include <asm/setup.h>
  51 #include <asm/cacheflush.h>
  52
  53 unsigned long max_low_pfn_mapped;
  54 unsigned long max_pfn_mapped;
  55
  56 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  57 unsigned long highstart_pfn, highend_pfn;
  58
  59 static noinline int do_test_wp_bit(void);
  60
  61
  62 static unsigned long __initdata table_start;
  63 static unsigned long __meminitdata table_end;
  64 static unsigned long __meminitdata table_top;
  65
  66 static int __initdata after_init_bootmem;
  67
  68 int direct_gbpages;
  69
  70 static __init void *alloc_low_page(void)
  71 {
  72         unsigned long pfn = table_end++;
  73         void *adr;
  74
  75         if (pfn >= table_top)
  76                 panic("alloc_low_page: ran out of memory");
  77
  78         adr = __va(pfn * PAGE_SIZE);
  79         memset(adr, 0, PAGE_SIZE);
  80         return adr;
  81 }
  82
  83 /*
  84  * Creates a middle page table and puts a pointer to it in the
  85  * given global directory entry. This only returns the gd entry
  86  * in non-PAE compilation mode, since the middle layer is folded.
  87  */
  88 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  89 {
  90         pud_t *pud;
  91         pmd_t *pmd_table;
  92
  93 #ifdef CONFIG_X86_PAE
  94         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
  95                 if (after_init_bootmem)
  96                         pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
  97                 else
  98                         pmd_table = (pmd_t *)alloc_low_page();
  99                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
 100                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
 101                 pud = pud_offset(pgd, 0);
 102                 BUG_ON(pmd_table != pmd_offset(pud, 0));
 103
 104                 return pmd_table;
 105         }
 106 #endif
 107         pud = pud_offset(pgd, 0);
 108         pmd_table = pmd_offset(pud, 0);
 109
 110         return pmd_table;
 111 }
 112
 113 /*
 114  * Create a page table and place a pointer to it in a middle page
 115  * directory entry:
 116  */
 117 static pte_t * __init one_page_table_init(pmd_t *pmd)
 118 {
 119         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
 120                 pte_t *page_table = NULL;
 121
 122                 if (after_init_bootmem) {
 123 #ifdef CONFIG_DEBUG_PAGEALLOC
 124                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
 125 #endif
 126                         if (!page_table)
 127                                 page_table =
 128                                 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 129                 } else
 130                         page_table = (pte_t *)alloc_low_page();
 131
 132                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
 133                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
 134                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
 135         }
 136
 137         return pte_offset_kernel(pmd, 0);
 138 }
 139
 140 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
 141                                            unsigned long vaddr, pte_t *lastpte)
 142 {
 143 #ifdef CONFIG_HIGHMEM
 144         /*
 145          * Something (early fixmap) may already have put a pte
 146          * page here, which causes the page table allocation
 147          * to become nonlinear. Attempt to fix it, and if it
 148          * is still nonlinear then we have to bug.
 149          */
 150         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
 151         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
 152
 153         if (pmd_idx_kmap_begin != pmd_idx_kmap_end
 154             && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
 155             && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
 156             && ((__pa(pte) >> PAGE_SHIFT) < table_start
 157                 || (__pa(pte) >> PAGE_SHIFT) >= table_end)) {
 158                 pte_t *newpte;
 159                 int i;
 160
 161                 BUG_ON(after_init_bootmem);
 162                 newpte = alloc_low_page();
 163                 for (i = 0; i < PTRS_PER_PTE; i++)
 164                         set_pte(newpte + i, pte[i]);
 165
 166                 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
 167                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
 168                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
 169                 __flush_tlb_all();
 170
 171                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
 172                 pte = newpte;
 173         }
 174         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
 175                && vaddr > fix_to_virt(FIX_KMAP_END)
 176                && lastpte && lastpte + PTRS_PER_PTE != pte);
 177 #endif
 178         return pte;
 179 }
 180
 181 /*
 182  * This function initializes a certain range of kernel virtual memory
 183  * with new bootmem page tables, everywhere page tables are missing in
 184  * the given range.
 185  *
 186  * NOTE: The pagetables are allocated contiguous on the physical space
 187  * so we can cache the place of the first one and move around without
 188  * checking the pgd every time.
 189  */
 190 static void __init
 191 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 192 {
 193         int pgd_idx, pmd_idx;
 194         unsigned long vaddr;
 195         pgd_t *pgd;
 196         pmd_t *pmd;
 197         pte_t *pte = NULL;
 198
 199         vaddr = start;
 200         pgd_idx = pgd_index(vaddr);
 201         pmd_idx = pmd_index(vaddr);
 202         pgd = pgd_base + pgd_idx;
 203
 204         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 205                 pmd = one_md_table_init(pgd);
 206                 pmd = pmd + pmd_index(vaddr);
 207                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 208                                                         pmd++, pmd_idx++) {
 209                         pte = page_table_kmap_check(one_page_table_init(pmd),
 210                                                     pmd, vaddr, pte);
 211
 212                         vaddr += PMD_SIZE;
 213                 }
 214                 pmd_idx = 0;
 215         }
 216 }
 217
 218 static inline int is_kernel_text(unsigned long addr)
 219 {
 220         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 221                 return 1;
 222         return 0;
 223 }
 224
 225 /*
 226  * This maps the physical memory to kernel virtual address space, a total
 227  * of max_low_pfn pages, by creating page tables starting from address
 228  * PAGE_OFFSET:
 229  */
 230 static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
 231                                                 unsigned long start_pfn,
 232                                                 unsigned long end_pfn,
 233                                                 int use_pse)
 234 {
 235         int pgd_idx, pmd_idx, pte_ofs;
 236         unsigned long pfn;
 237         pgd_t *pgd;
 238         pmd_t *pmd;
 239         pte_t *pte;
 240         unsigned pages_2m, pages_4k;
 241         int mapping_iter;
 242
 243         /*
 244          * First iteration will setup identity mapping using large/small pages
 245          * based on use_pse, with other attributes same as set by
 246          * the early code in head_32.S
 247          *
 248          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
 249          * as desired for the kernel identity mapping.
 250          *
 251          * This two pass mechanism conforms to the TLB app note which says:
 252          *
 253          *     "Software should not write to a paging-structure entry in a way
 254          *      that would change, for any linear address, both the page size
 255          *      and either the page frame or attributes."
 256          */
 257         mapping_iter = 1;
 258
 259         if (!cpu_has_pse)
 260                 use_pse = 0;
 261
 262 repeat:
 263         pages_2m = pages_4k = 0;
 264         pfn = start_pfn;
 265         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 266         pgd = pgd_base + pgd_idx;
 267         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 268                 pmd = one_md_table_init(pgd);
 269
 270                 if (pfn >= end_pfn)
 271                         continue;
 272 #ifdef CONFIG_X86_PAE
 273                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 274                 pmd += pmd_idx;
 275 #else
 276                 pmd_idx = 0;
 277 #endif
 278                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
 279                      pmd++, pmd_idx++) {
 280                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 281
 282                         /*
 283                          * Map with big pages if possible, otherwise
 284                          * create normal page tables:
 285                          */
 286                         if (use_pse) {
 287                                 unsigned int addr2;
 288                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 289                                 /*
 290                                  * first pass will use the same initial
 291                                  * identity mapping attribute + _PAGE_PSE.
 292                                  */
 293                                 pgprot_t init_prot =
 294                                         __pgprot(PTE_IDENT_ATTR |
 295                                                  _PAGE_PSE);
 296
 297                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 298                                         PAGE_OFFSET + PAGE_SIZE-1;
 299
 300                                 if (is_kernel_text(addr) ||
 301                                     is_kernel_text(addr2))
 302                                         prot = PAGE_KERNEL_LARGE_EXEC;
 303
 304                                 pages_2m++;
 305                                 if (mapping_iter == 1)
 306                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
 307                                 else
 308                                         set_pmd(pmd, pfn_pmd(pfn, prot));
 309
 310                                 pfn += PTRS_PER_PTE;
 311                                 continue;
 312                         }
 313                         pte = one_page_table_init(pmd);
 314
 315                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 316                         pte += pte_ofs;
 317                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 318                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 319                                 pgprot_t prot = PAGE_KERNEL;
 320                                 /*
 321                                  * first pass will use the same initial
 322                                  * identity mapping attribute.
 323                                  */
 324                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
 325
 326                                 if (is_kernel_text(addr))
 327                                         prot = PAGE_KERNEL_EXEC;
 328
 329                                 pages_4k++;
 330                                 if (mapping_iter == 1)
 331                                         set_pte(pte, pfn_pte(pfn, init_prot));
 332                                 else
 333                                         set_pte(pte, pfn_pte(pfn, prot));
 334                         }
 335                 }
 336         }
 337         if (mapping_iter == 1) {
 338                 /*
 339                  * update direct mapping page count only in the first
 340                  * iteration.
 341                  */
 342                 update_page_count(PG_LEVEL_2M, pages_2m);
 343                 update_page_count(PG_LEVEL_4K, pages_4k);
 344
 345                 /*
 346                  * local global flush tlb, which will flush the previous
 347                  * mappings present in both small and large page TLB's.
 348                  */
 349                 __flush_tlb_all();
 350
 351                 /*
 352                  * Second iteration will set the actual desired PTE attributes.
 353                  */
 354                 mapping_iter = 2;
 355                 goto repeat;
 356         }
 357 }
 358
 359 pte_t *kmap_pte;
 360 pgprot_t kmap_prot;
 361
 362 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 363 {
 364         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 365                         vaddr), vaddr), vaddr);
 366 }
 367
 368 static void __init kmap_init(void)
 369 {
 370         unsigned long kmap_vstart;
 371
 372         /*
 373          * Cache the first kmap pte:
 374          */
 375         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 376         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 377
 378         kmap_prot = PAGE_KERNEL;
 379 }
 380
 381 #ifdef CONFIG_HIGHMEM
 382 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 383 {
 384         unsigned long vaddr;
 385         pgd_t *pgd;
 386         pud_t *pud;
 387         pmd_t *pmd;
 388         pte_t *pte;
 389
 390         vaddr = PKMAP_BASE;
 391         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 392
 393         pgd = swapper_pg_dir + pgd_index(vaddr);
 394         pud = pud_offset(pgd, vaddr);
 395         pmd = pmd_offset(pud, vaddr);
 396         pte = pte_offset_kernel(pmd, vaddr);
 397         pkmap_page_table = pte;
 398 }
 399
 400 static void __init add_one_highpage_init(struct page *page, int pfn)
 401 {
 402         ClearPageReserved(page);
 403         init_page_count(page);
 404         __free_page(page);
 405         totalhigh_pages++;
 406 }
 407
 408 struct add_highpages_data {
 409         unsigned long start_pfn;
 410         unsigned long end_pfn;
 411 };
 412
 413 static int __init add_highpages_work_fn(unsigned long start_pfn,
 414                                          unsigned long end_pfn, void *datax)
 415 {
 416         int node_pfn;
 417         struct page *page;
 418         unsigned long final_start_pfn, final_end_pfn;
 419         struct add_highpages_data *data;
 420
 421         data = (struct add_highpages_data *)datax;
 422
 423         final_start_pfn = max(start_pfn, data->start_pfn);
 424         final_end_pfn = min(end_pfn, data->end_pfn);
 425         if (final_start_pfn >= final_end_pfn)
 426                 return 0;
 427
 428         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
 429              node_pfn++) {
 430                 if (!pfn_valid(node_pfn))
 431                         continue;
 432                 page = pfn_to_page(node_pfn);
 433                 add_one_highpage_init(page, node_pfn);
 434         }
 435
 436         return 0;
 437
 438 }
 439
 440 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
 441                                               unsigned long end_pfn)
 442 {
 443         struct add_highpages_data data;
 444
 445         data.start_pfn = start_pfn;
 446         data.end_pfn = end_pfn;
 447
 448         work_with_active_regions(nid, add_highpages_work_fn, &data);
 449 }
 450
 451 #else
 452 static inline void permanent_kmaps_init(pgd_t *pgd_base)
 453 {
 454 }
 455 #endif /* CONFIG_HIGHMEM */
 456
 457 void __init native_pagetable_setup_start(pgd_t *base)
 458 {
 459         unsigned long pfn, va;
 460         pgd_t *pgd;
 461         pud_t *pud;
 462         pmd_t *pmd;
 463         pte_t *pte;
 464
 465         /*
 466          * Remove any mappings which extend past the end of physical
 467          * memory from the boot time page table:
 468          */
 469         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 470                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 471                 pgd = base + pgd_index(va);
 472                 if (!pgd_present(*pgd))
 473                         break;
 474
 475                 pud = pud_offset(pgd, va);
 476                 pmd = pmd_offset(pud, va);
 477                 if (!pmd_present(*pmd))
 478                         break;
 479
 480                 pte = pte_offset_kernel(pmd, va);
 481                 if (!pte_present(*pte))
 482                         break;
 483
 484                 pte_clear(NULL, va, pte);
 485         }
 486         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 487 }
 488
 489 void __init native_pagetable_setup_done(pgd_t *base)
 490 {
 491 }
 492
 493 /*
 494  * Build a proper pagetable for the kernel mappings.  Up until this
 495  * point, we've been running on some set of pagetables constructed by
 496  * the boot process.
 497  *
 498  * If we're booting on native hardware, this will be a pagetable
 499  * constructed in arch/x86/kernel/head_32.S.  The root of the
 500  * pagetable will be swapper_pg_dir.
 501  *
 502  * If we're booting paravirtualized under a hypervisor, then there are
 503  * more options: we may already be running PAE, and the pagetable may
 504  * or may not be based in swapper_pg_dir.  In any case,
 505  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 506  * appropriately for the rest of the initialization to work.
 507  *
 508  * In general, pagetable_init() assumes that the pagetable may already
 509  * be partially populated, and so it avoids stomping on any existing
 510  * mappings.
 511  */
 512 static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base)
 513 {
 514         unsigned long vaddr, end;
 515
 516         /*
 517          * Fixed mappings, only the page table structure has to be
 518          * created - mappings will be set by set_fixmap():
 519          */
 520         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 521         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 522         page_table_range_init(vaddr, end, pgd_base);
 523         early_ioremap_reset();
 524 }
 525
 526 static void __init pagetable_init(void)
 527 {
 528         pgd_t *pgd_base = swapper_pg_dir;
 529
 530         permanent_kmaps_init(pgd_base);
 531 }
 532
 533 #ifdef CONFIG_ACPI_SLEEP
 534 /*
 535  * ACPI suspend needs this for resume, because things like the intel-agp
 536  * driver might have split up a kernel 4MB mapping.
 537  */
 538 char swsusp_pg_dir[PAGE_SIZE]
 539         __attribute__ ((aligned(PAGE_SIZE)));
 540
 541 static inline void save_pg_dir(void)
 542 {
 543         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 544 }
 545 #else /* !CONFIG_ACPI_SLEEP */
 546 static inline void save_pg_dir(void)
 547 {
 548 }
 549 #endif /* !CONFIG_ACPI_SLEEP */
 550
 551 void zap_low_mappings(void)
 552 {
 553         int i;
 554
 555         /*
 556          * Zap initial low-memory mappings.
 557          *
 558          * Note that "pgd_clear()" doesn't do it for
 559          * us, because pgd_clear() is a no-op on i386.
 560          */
 561         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 562 #ifdef CONFIG_X86_PAE
 563                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 564 #else
 565                 set_pgd(swapper_pg_dir+i, __pgd(0));
 566 #endif
 567         }
 568         flush_tlb_all();
 569 }
 570
 571 int nx_enabled;
 572
 573 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
 574 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 575
 576 #ifdef CONFIG_X86_PAE
 577
 578 static int disable_nx __initdata;
 579
 580 /*
 581  * noexec = on|off
 582  *
 583  * Control non executable mappings.
 584  *
 585  * on      Enable
 586  * off     Disable
 587  */
 588 static int __init noexec_setup(char *str)
 589 {
 590         if (!str || !strcmp(str, "on")) {
 591                 if (cpu_has_nx) {
 592                         __supported_pte_mask |= _PAGE_NX;
 593                         disable_nx = 0;
 594                 }
 595         } else {
 596                 if (!strcmp(str, "off")) {
 597                         disable_nx = 1;
 598                         __supported_pte_mask &= ~_PAGE_NX;
 599                 } else {
 600                         return -EINVAL;
 601                 }
 602         }
 603
 604         return 0;
 605 }
 606 early_param("noexec", noexec_setup);
 607
 608 static void __init set_nx(void)
 609 {
 610         unsigned int v[4], l, h;
 611
 612         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 613                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 614
 615                 if ((v[3] & (1 << 20)) && !disable_nx) {
 616                         rdmsr(MSR_EFER, l, h);
 617                         l |= EFER_NX;
 618                         wrmsr(MSR_EFER, l, h);
 619                         nx_enabled = 1;
 620                         __supported_pte_mask |= _PAGE_NX;
 621                 }
 622         }
 623 }
 624 #endif
 625
 626 /* user-defined highmem size */
 627 static unsigned int highmem_pages = -1;
 628
 629 /*
 630  * highmem=size forces highmem to be exactly 'size' bytes.
 631  * This works even on boxes that have no highmem otherwise.
 632  * This also works to reduce highmem size on bigger boxes.
 633  */
 634 static int __init parse_highmem(char *arg)
 635 {
 636         if (!arg)
 637                 return -EINVAL;
 638
 639         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 640         return 0;
 641 }
 642 early_param("highmem", parse_highmem);
 643
 644 #define MSG_HIGHMEM_TOO_BIG \
 645         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
 646
 647 #define MSG_LOWMEM_TOO_SMALL \
 648         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
 649 /*
 650  * All of RAM fits into lowmem - but if user wants highmem
 651  * artificially via the highmem=x boot parameter then create
 652  * it:
 653  */
 654 void __init lowmem_pfn_init(void)
 655 {
 656         /* max_low_pfn is 0, we already have early_res support */
 657         max_low_pfn = max_pfn;
 658
 659         if (highmem_pages == -1)
 660                 highmem_pages = 0;
 661 #ifdef CONFIG_HIGHMEM
 662         if (highmem_pages >= max_pfn) {
 663                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
 664                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 665                 highmem_pages = 0;
 666         }
 667         if (highmem_pages) {
 668                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
 669                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
 670                                 pages_to_mb(highmem_pages));
 671                         highmem_pages = 0;
 672                 }
 673                 max_low_pfn -= highmem_pages;
 674         }
 675 #else
 676         if (highmem_pages)
 677                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 678 #endif
 679 }
 680
 681 #define MSG_HIGHMEM_TOO_SMALL \
 682         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
 683
 684 #define MSG_HIGHMEM_TRIMMED \
 685         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
 686 /*
 687  * We have more RAM than fits into lowmem - we try to put it into
 688  * highmem, also taking the highmem=x boot parameter into account:
 689  */
 690 void __init highmem_pfn_init(void)
 691 {
 692         max_low_pfn = MAXMEM_PFN;
 693
 694         if (highmem_pages == -1)
 695                 highmem_pages = max_pfn - MAXMEM_PFN;
 696
 697         if (highmem_pages + MAXMEM_PFN < max_pfn)
 698                 max_pfn = MAXMEM_PFN + highmem_pages;
 699
 700         if (highmem_pages + MAXMEM_PFN > max_pfn) {
 701                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
 702                         pages_to_mb(max_pfn - MAXMEM_PFN),
 703                         pages_to_mb(highmem_pages));
 704                 highmem_pages = 0;
 705         }
 706 #ifndef CONFIG_HIGHMEM
 707         /* Maximum memory usable is what is directly addressable */
 708         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
 709         if (max_pfn > MAX_NONPAE_PFN)
 710                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 711         else
 712                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 713         max_pfn = MAXMEM_PFN;
 714 #else /* !CONFIG_HIGHMEM */
 715 #ifndef CONFIG_HIGHMEM64G
 716         if (max_pfn > MAX_NONPAE_PFN) {
 717                 max_pfn = MAX_NONPAE_PFN;
 718                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
 719         }
 720 #endif /* !CONFIG_HIGHMEM64G */
 721 #endif /* !CONFIG_HIGHMEM */
 722 }
 723
 724 /*
 725  * Determine low and high memory ranges:
 726  */
 727 void __init find_low_pfn_range(void)
 728 {
 729         /* it could update max_pfn */
 730
 731         if (max_pfn <= MAXMEM_PFN)
 732                 lowmem_pfn_init();
 733         else
 734                 highmem_pfn_init();
 735 }
 736
 737 #ifndef CONFIG_NEED_MULTIPLE_NODES
 738 void __init initmem_init(unsigned long start_pfn,
 739                                   unsigned long end_pfn)
 740 {
 741 #ifdef CONFIG_HIGHMEM
 742         highstart_pfn = highend_pfn = max_pfn;
 743         if (max_pfn > max_low_pfn)
 744                 highstart_pfn = max_low_pfn;
 745         memory_present(0, 0, highend_pfn);
 746         e820_register_active_regions(0, 0, highend_pfn);
 747         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 748                 pages_to_mb(highend_pfn - highstart_pfn));
 749         num_physpages = highend_pfn;
 750         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 751 #else
 752         memory_present(0, 0, max_low_pfn);
 753         e820_register_active_regions(0, 0, max_low_pfn);
 754         num_physpages = max_low_pfn;
 755         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 756 #endif
 757 #ifdef CONFIG_FLATMEM
 758         max_mapnr = num_physpages;
 759 #endif
 760         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 761                         pages_to_mb(max_low_pfn));
 762
 763         setup_bootmem_allocator();
 764 }
 765 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 766
 767 static void __init zone_sizes_init(void)
 768 {
 769         unsigned long max_zone_pfns[MAX_NR_ZONES];
 770         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 771         max_zone_pfns[ZONE_DMA] =
 772                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 773         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 774 #ifdef CONFIG_HIGHMEM
 775         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 776 #endif
 777
 778         free_area_init_nodes(max_zone_pfns);
 779 }
 780
 781 static unsigned long __init setup_node_bootmem(int nodeid,
 782                                  unsigned long start_pfn,
 783                                  unsigned long end_pfn,
 784                                  unsigned long bootmap)
 785 {
 786         unsigned long bootmap_size;
 787
 788         if (start_pfn > max_low_pfn)
 789                 return bootmap;
 790         if (end_pfn > max_low_pfn)
 791                 end_pfn = max_low_pfn;
 792
 793         /* don't touch min_low_pfn */
 794         bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
 795                                          bootmap >> PAGE_SHIFT,
 796                                          start_pfn, end_pfn);
 797         printk(KERN_INFO "  node %d low ram: %08lx - %08lx\n",
 798                 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 799         printk(KERN_INFO "  node %d bootmap %08lx - %08lx\n",
 800                  nodeid, bootmap, bootmap + bootmap_size);
 801         free_bootmem_with_active_regions(nodeid, end_pfn);
 802         early_res_to_bootmem(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 803
 804         return bootmap + bootmap_size;
 805 }
 806
 807 void __init setup_bootmem_allocator(void)
 808 {
 809         int nodeid;
 810         unsigned long bootmap_size, bootmap;
 811         /*
 812          * Initialize the boot-time allocator (with low memory only):
 813          */
 814         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 815         bootmap = find_e820_area(0, max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 816                                  PAGE_SIZE);
 817         if (bootmap == -1L)
 818                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 819         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 820
 821         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 822                  max_pfn_mapped<<PAGE_SHIFT);
 823         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
 824
 825 #ifdef CONFIG_NEED_MULTIPLE_NODES
 826         for_each_online_node(nodeid)
 827                 bootmap = setup_node_bootmem(nodeid, node_start_pfn[nodeid],
 828                                         node_end_pfn[nodeid], bootmap);
 829 #else
 830         bootmap = setup_node_bootmem(0, 0, max_low_pfn, bootmap);
 831 #endif
 832
 833         after_init_bootmem = 1;
 834 }
 835
 836 static void __init find_early_table_space(unsigned long end, int use_pse,
 837                                           int use_gbpages)
 838 {
 839         unsigned long puds, pmds, ptes, tables, start;
 840
 841         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 842         tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
 843
 844         if (use_gbpages) {
 845                 unsigned long extra;
 846
 847                 extra = end - ((end>>PUD_SHIFT) << PUD_SHIFT);
 848                 pmds = (extra + PMD_SIZE - 1) >> PMD_SHIFT;
 849         } else
 850                 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 851
 852         tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
 853
 854         if (use_pse) {
 855                 unsigned long extra;
 856
 857                 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
 858                 extra += PMD_SIZE;
 859                 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
 860         } else
 861                 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
 862
 863         tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
 864
 865         /* for fixmap */
 866         tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
 867
 868         /*
 869          * RED-PEN putting page tables only on node 0 could
 870          * cause a hotspot and fill up ZONE_DMA. The page tables
 871          * need roughly 0.5KB per GB.
 872          */
 873         start = 0x7000;
 874         table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
 875                                         tables, PAGE_SIZE);
 876         if (table_start == -1UL)
 877                 panic("Cannot find space for the kernel page tables");
 878
 879         table_start >>= PAGE_SHIFT;
 880         table_end = table_start;
 881         table_top = table_start + (tables >> PAGE_SHIFT);
 882
 883         printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
 884                 end, table_start << PAGE_SHIFT, table_top << PAGE_SHIFT);
 885 }
 886
 887 struct map_range {
 888         unsigned long start;
 889         unsigned long end;
 890         unsigned page_size_mask;
 891 };
 892
 893 #define NR_RANGE_MR 3
 894
 895 static int save_mr(struct map_range *mr, int nr_range,
 896                    unsigned long start_pfn, unsigned long end_pfn,
 897                    unsigned long page_size_mask)
 898 {
 899         if (start_pfn < end_pfn) {
 900                 if (nr_range >= NR_RANGE_MR)
 901                         panic("run out of range for init_memory_mapping\n");
 902                 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
 903                 mr[nr_range].end   = end_pfn<<PAGE_SHIFT;
 904                 mr[nr_range].page_size_mask = page_size_mask;
 905                 nr_range++;
 906         }
 907
 908         return nr_range;
 909 }
 910
 911 /*
 912  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 913  * This runs before bootmem is initialized and gets pages directly from
 914  * the physical memory. To access them they are temporarily mapped.
 915  */
 916 unsigned long __init_refok init_memory_mapping(unsigned long start,
 917                                                unsigned long end)
 918 {
 919         pgd_t *pgd_base = swapper_pg_dir;
 920         unsigned long page_size_mask = 0;
 921         unsigned long start_pfn, end_pfn;
 922         unsigned long pos;
 923
 924         struct map_range mr[NR_RANGE_MR];
 925         int nr_range, i;
 926         int use_pse, use_gbpages;
 927
 928         printk(KERN_INFO "init_memory_mapping: %016lx-%016lx\n", start, end);
 929
 930 #ifdef CONFIG_DEBUG_PAGEALLOC
 931         /*
 932          * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
 933          * This will simplify cpa(), which otherwise needs to support splitting
 934          * large pages into small in interrupt context, etc.
 935          */
 936         use_pse = use_gbpages = 0;
 937 #else
 938         use_pse = cpu_has_pse;
 939         use_gbpages = direct_gbpages;
 940 #endif
 941
 942 #ifdef CONFIG_X86_PAE
 943         set_nx();
 944         if (nx_enabled)
 945                 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
 946 #endif
 947
 948         /* Enable PSE if available */
 949         if (cpu_has_pse)
 950                 set_in_cr4(X86_CR4_PSE);
 951
 952         /* Enable PGE if available */
 953         if (cpu_has_pge) {
 954                 set_in_cr4(X86_CR4_PGE);
 955                 __supported_pte_mask |= _PAGE_GLOBAL;
 956         }
 957
 958         if (use_gbpages)
 959                 page_size_mask |= 1 << PG_LEVEL_1G;
 960         if (use_pse)
 961                 page_size_mask |= 1 << PG_LEVEL_2M;
 962
 963         memset(mr, 0, sizeof(mr));
 964         nr_range = 0;
 965
 966         /*
 967          * Don't use a large page for the first 2/4MB of memory
 968          * because there are often fixed size MTRRs in there
 969          * and overlapping MTRRs into large pages can cause
 970          * slowdowns.
 971          */
 972         /* head if not big page alignment ? */
 973         start_pfn = start >> PAGE_SHIFT;
 974         pos = start_pfn << PAGE_SHIFT;
 975         if (pos == 0)
 976                 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
 977         else
 978                 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 979                                  << (PMD_SHIFT - PAGE_SHIFT);
 980         if (end_pfn > (end >> PAGE_SHIFT))
 981                 end_pfn = end >> PAGE_SHIFT;
 982         if (start_pfn < end_pfn) {
 983                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
 984                 pos = end_pfn << PAGE_SHIFT;
 985         }
 986
 987         /* big page (2M) range */
 988         start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 989                          << (PMD_SHIFT - PAGE_SHIFT);
 990         end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 991         if (start_pfn < end_pfn) {
 992                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
 993                                 page_size_mask & (1<<PG_LEVEL_2M));
 994                 pos = end_pfn << PAGE_SHIFT;
 995         }
 996
 997         /* tail is not big page (2M) alignment */
 998         start_pfn = pos>>PAGE_SHIFT;
 999         end_pfn = end>>PAGE_SHIFT;
1000         if (start_pfn < end_pfn)
1001                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
1002
1003         /* try to merge same page size and continuous */
1004         for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
1005                 unsigned long old_start;
1006                 if (mr[i].end != mr[i+1].start ||
1007                     mr[i].page_size_mask != mr[i+1].page_size_mask)
1008                         continue;
1009                 /* move it */
1010                 old_start = mr[i].start;
1011                 memmove(&mr[i], &mr[i+1],
1012                         (nr_range - 1 - i) * sizeof(struct map_range));
1013                 mr[i--].start = old_start;
1014                 nr_range--;
1015         }
1016
1017         for (i = 0; i < nr_range; i++)
1018                 printk(KERN_DEBUG " %010lx - %010lx page %s\n",
1019                                 mr[i].start, mr[i].end,
1020                         (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
1021                          (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
1022
1023         /*
1024          * Find space for the kernel direct mapping tables.
1025          *
1026          * Later we should allocate these tables in the local node of the
1027          * memory mapped. Unfortunately this is done currently before the
1028          * nodes are discovered.
1029          */
1030         if (!after_init_bootmem)
1031                 find_early_table_space(end, use_pse, use_gbpages);
1032
1033         for (i = 0; i < nr_range; i++)
1034                 kernel_physical_mapping_init(pgd_base,
1035                                 mr[i].start >> PAGE_SHIFT,
1036                                 mr[i].end >> PAGE_SHIFT,
1037                                 mr[i].page_size_mask == (1<<PG_LEVEL_2M));
1038
1039         early_ioremap_page_table_range_init(pgd_base);
1040
1041         load_cr3(swapper_pg_dir);
1042
1043         __flush_tlb_all();
1044
1045         if (!after_init_bootmem)
1046                 reserve_early(table_start << PAGE_SHIFT,
1047                                  table_end << PAGE_SHIFT, "PGTABLE");
1048
1049         if (!after_init_bootmem)
1050                 early_memtest(start, end);
1051
1052         return end >> PAGE_SHIFT;
1053 }
1054
1055
1056 /*
1057  * paging_init() sets up the page tables - note that the first 8MB are
1058  * already mapped by head.S.
1059  *
1060  * This routines also unmaps the page at virtual kernel address 0, so
1061  * that we can trap those pesky NULL-reference errors in the kernel.
1062  */
1063 void __init paging_init(void)
1064 {
1065         pagetable_init();
1066
1067         __flush_tlb_all();
1068
1069         kmap_init();
1070
1071         /*
1072          * NOTE: at this point the bootmem allocator is fully available.
1073          */
1074         sparse_init();
1075         zone_sizes_init();
1076 }
1077
1078 /*
1079  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
1080  * and also on some strange 486's. All 586+'s are OK. This used to involve
1081  * black magic jumps to work around some nasty CPU bugs, but fortunately the
1082  * switch to using exceptions got rid of all that.
1083  */
1084 static void __init test_wp_bit(void)
1085 {
1086         printk(KERN_INFO
1087   "Checking if this processor honours the WP bit even in supervisor mode...");
1088
1089         /* Any page-aligned address will do, the test is non-destructive */
1090         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
1091         boot_cpu_data.wp_works_ok = do_test_wp_bit();
1092         clear_fixmap(FIX_WP_TEST);
1093
1094         if (!boot_cpu_data.wp_works_ok) {
1095                 printk(KERN_CONT "No.\n");
1096 #ifdef CONFIG_X86_WP_WORKS_OK
1097                 panic(
1098   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
1099 #endif
1100         } else {
1101                 printk(KERN_CONT "Ok.\n");
1102         }
1103 }
1104
1105 static struct kcore_list kcore_mem, kcore_vmalloc;
1106
1107 void __init mem_init(void)
1108 {
1109         int codesize, reservedpages, datasize, initsize;
1110         int tmp;
1111
1112         pci_iommu_alloc();
1113
1114 #ifdef CONFIG_FLATMEM
1115         BUG_ON(!mem_map);
1116 #endif
1117         /* this will put all low memory onto the freelists */
1118         totalram_pages += free_all_bootmem();
1119
1120         reservedpages = 0;
1121         for (tmp = 0; tmp < max_low_pfn; tmp++)
1122                 /*
1123                  * Only count reserved RAM pages:
1124                  */
1125                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
1126                         reservedpages++;
1127
1128         set_highmem_pages_init();
1129
1130         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
1131         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
1132         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
1133
1134         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
1135         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
1136                    VMALLOC_END-VMALLOC_START);
1137
1138         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
1139                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
1140                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
1141                 num_physpages << (PAGE_SHIFT-10),
1142                 codesize >> 10,
1143                 reservedpages << (PAGE_SHIFT-10),
1144                 datasize >> 10,
1145                 initsize >> 10,
1146                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
1147                );
1148
1149         printk(KERN_INFO "virtual kernel memory layout:\n"
1150                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1151 #ifdef CONFIG_HIGHMEM
1152                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1153 #endif
1154                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
1155                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
1156                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1157                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1158                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
1159                 FIXADDR_START, FIXADDR_TOP,
1160                 (FIXADDR_TOP - FIXADDR_START) >> 10,
1161
1162 #ifdef CONFIG_HIGHMEM
1163                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
1164                 (LAST_PKMAP*PAGE_SIZE) >> 10,
1165 #endif
1166
1167                 VMALLOC_START, VMALLOC_END,
1168                 (VMALLOC_END - VMALLOC_START) >> 20,
1169
1170                 (unsigned long)__va(0), (unsigned long)high_memory,
1171                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
1172
1173                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
1174                 ((unsigned long)&__init_end -
1175                  (unsigned long)&__init_begin) >> 10,
1176
1177                 (unsigned long)&_etext, (unsigned long)&_edata,
1178                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
1179
1180                 (unsigned long)&_text, (unsigned long)&_etext,
1181                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
1182
1183         /*
1184          * Check boundaries twice: Some fundamental inconsistencies can
1185          * be detected at build time already.
1186          */
1187 #define __FIXADDR_TOP (-PAGE_SIZE)
1188 #ifdef CONFIG_HIGHMEM
1189         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
1190         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
1191 #endif
1192 #define high_memory (-128UL << 20)
1193         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
1194 #undef high_memory
1195 #undef __FIXADDR_TOP
1196
1197 #ifdef CONFIG_HIGHMEM
1198         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
1199         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
1200 #endif
1201         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
1202         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
1203
1204         if (boot_cpu_data.wp_works_ok < 0)
1205                 test_wp_bit();
1206
1207         save_pg_dir();
1208         zap_low_mappings();
1209 }
1210
1211 #ifdef CONFIG_MEMORY_HOTPLUG
1212 int arch_add_memory(int nid, u64 start, u64 size)
1213 {
1214         struct pglist_data *pgdata = NODE_DATA(nid);
1215         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1216         unsigned long start_pfn = start >> PAGE_SHIFT;
1217         unsigned long nr_pages = size >> PAGE_SHIFT;
1218
1219         return __add_pages(nid, zone, start_pfn, nr_pages);
1220 }
1221 #endif
1222
1223 /*
1224  * This function cannot be __init, since exceptions don't work in that
1225  * section.  Put this after the callers, so that it cannot be inlined.
1226  */
1227 static noinline int do_test_wp_bit(void)
1228 {
1229         char tmp_reg;
1230         int flag;
1231
1232         __asm__ __volatile__(
1233                 "       movb %0, %1     \n"
1234                 "1:     movb %1, %0     \n"
1235                 "       xorl %2, %2     \n"
1236                 "2:                     \n"
1237                 _ASM_EXTABLE(1b,2b)
1238                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1239                  "=q" (tmp_reg),
1240                  "=r" (flag)
1241                 :"2" (1)
1242                 :"memory");
1243
1244         return flag;
1245 }
1246
1247 #ifdef CONFIG_DEBUG_RODATA
1248 const int rodata_test_data = 0xC3;
1249 EXPORT_SYMBOL_GPL(rodata_test_data);
1250
1251 void mark_rodata_ro(void)
1252 {
1253         unsigned long start = PFN_ALIGN(_text);
1254         unsigned long size = PFN_ALIGN(_etext) - start;
1255
1256 #ifndef CONFIG_DYNAMIC_FTRACE
1257         /* Dynamic tracing modifies the kernel text section */
1258         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1259         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1260                 size >> 10);
1261
1262 #ifdef CONFIG_CPA_DEBUG
1263         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1264                 start, start+size);
1265         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1266
1267         printk(KERN_INFO "Testing CPA: write protecting again\n");
1268         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1269 #endif
1270 #endif /* CONFIG_DYNAMIC_FTRACE */
1271
1272         start += size;
1273         size = (unsigned long)__end_rodata - start;
1274         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1275         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1276                 size >> 10);
1277         rodata_test();
1278
1279 #ifdef CONFIG_CPA_DEBUG
1280         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1281         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1282
1283         printk(KERN_INFO "Testing CPA: write protecting again\n");
1284         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1285 #endif
1286 }
1287 #endif
1288
1289 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1290                                    int flags)
1291 {
1292         return reserve_bootmem(phys, len, flags);
1293 }