SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/pfn.h>
  25 #include <linux/poison.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/memory_hotplug.h>
  30 #include <linux/initrd.h>
  31 #include <linux/cpumask.h>
  32
  33 #include <asm/asm.h>
  34 #include <asm/processor.h>
  35 #include <asm/system.h>
  36 #include <asm/uaccess.h>
  37 #include <asm/pgtable.h>
  38 #include <asm/dma.h>
  39 #include <asm/fixmap.h>
  40 #include <asm/e820.h>
  41 #include <asm/apic.h>
  42 #include <asm/bugs.h>
  43 #include <asm/tlb.h>
  44 #include <asm/tlbflush.h>
  45 #include <asm/pgalloc.h>
  46 #include <asm/sections.h>
  47 #include <asm/paravirt.h>
  48 #include <asm/setup.h>
  49 #include <asm/cacheflush.h>
  50
  51 unsigned int __VMALLOC_RESERVE = 128 << 20;
  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 static __init void *alloc_low_page(unsigned long *phys)
  69 {
  70         unsigned long pfn = table_end++;
  71         void *adr;
  72
  73         if (pfn >= table_top)
  74                 panic("alloc_low_page: ran out of memory");
  75
  76         adr = __va(pfn * PAGE_SIZE);
  77         memset(adr, 0, PAGE_SIZE);
  78         *phys  = pfn * PAGE_SIZE;
  79         return adr;
  80 }
  81
  82 /*
  83  * Creates a middle page table and puts a pointer to it in the
  84  * given global directory entry. This only returns the gd entry
  85  * in non-PAE compilation mode, since the middle layer is folded.
  86  */
  87 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  88 {
  89         pud_t *pud;
  90         pmd_t *pmd_table;
  91
  92 #ifdef CONFIG_X86_PAE
  93         unsigned long phys;
  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(&phys);
  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 #endif
 105         pud = pud_offset(pgd, 0);
 106         pmd_table = pmd_offset(pud, 0);
 107
 108         return pmd_table;
 109 }
 110
 111 /*
 112  * Create a page table and place a pointer to it in a middle page
 113  * directory entry:
 114  */
 115 static pte_t * __init one_page_table_init(pmd_t *pmd)
 116 {
 117         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
 118                 pte_t *page_table = NULL;
 119
 120                 if (after_init_bootmem) {
 121 #ifdef CONFIG_DEBUG_PAGEALLOC
 122                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
 123 #endif
 124                         if (!page_table)
 125                                 page_table =
 126                                 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 127                 } else {
 128                         unsigned long phys;
 129                         page_table = (pte_t *)alloc_low_page(&phys);
 130                 }
 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 /*
 141  * This function initializes a certain range of kernel virtual memory
 142  * with new bootmem page tables, everywhere page tables are missing in
 143  * the given range.
 144  *
 145  * NOTE: The pagetables are allocated contiguous on the physical space
 146  * so we can cache the place of the first one and move around without
 147  * checking the pgd every time.
 148  */
 149 static void __init
 150 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 151 {
 152         int pgd_idx, pmd_idx;
 153         unsigned long vaddr;
 154         pgd_t *pgd;
 155         pmd_t *pmd;
 156
 157         vaddr = start;
 158         pgd_idx = pgd_index(vaddr);
 159         pmd_idx = pmd_index(vaddr);
 160         pgd = pgd_base + pgd_idx;
 161
 162         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 163                 pmd = one_md_table_init(pgd);
 164                 pmd = pmd + pmd_index(vaddr);
 165                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 166                                                         pmd++, pmd_idx++) {
 167                         one_page_table_init(pmd);
 168
 169                         vaddr += PMD_SIZE;
 170                 }
 171                 pmd_idx = 0;
 172         }
 173 }
 174
 175 static inline int is_kernel_text(unsigned long addr)
 176 {
 177         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 178                 return 1;
 179         return 0;
 180 }
 181
 182 /*
 183  * This maps the physical memory to kernel virtual address space, a total
 184  * of max_low_pfn pages, by creating page tables starting from address
 185  * PAGE_OFFSET:
 186  */
 187 static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
 188                                                 unsigned long start_pfn,
 189                                                 unsigned long end_pfn,
 190                                                 int use_pse)
 191 {
 192         int pgd_idx, pmd_idx, pte_ofs;
 193         unsigned long pfn;
 194         pgd_t *pgd;
 195         pmd_t *pmd;
 196         pte_t *pte;
 197         unsigned pages_2m, pages_4k;
 198         int mapping_iter;
 199
 200         /*
 201          * First iteration will setup identity mapping using large/small pages
 202          * based on use_pse, with other attributes same as set by
 203          * the early code in head_32.S
 204          *
 205          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
 206          * as desired for the kernel identity mapping.
 207          *
 208          * This two pass mechanism conforms to the TLB app note which says:
 209          *
 210          *     "Software should not write to a paging-structure entry in a way
 211          *      that would change, for any linear address, both the page size
 212          *      and either the page frame or attributes."
 213          */
 214         mapping_iter = 1;
 215
 216         if (!cpu_has_pse)
 217                 use_pse = 0;
 218
 219 repeat:
 220         pages_2m = pages_4k = 0;
 221         pfn = start_pfn;
 222         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 223         pgd = pgd_base + pgd_idx;
 224         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 225                 pmd = one_md_table_init(pgd);
 226
 227                 if (pfn >= end_pfn)
 228                         continue;
 229 #ifdef CONFIG_X86_PAE
 230                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 231                 pmd += pmd_idx;
 232 #else
 233                 pmd_idx = 0;
 234 #endif
 235                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
 236                      pmd++, pmd_idx++) {
 237                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 238
 239                         /*
 240                          * Map with big pages if possible, otherwise
 241                          * create normal page tables:
 242                          */
 243                         if (use_pse) {
 244                                 unsigned int addr2;
 245                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 246                                 /*
 247                                  * first pass will use the same initial
 248                                  * identity mapping attribute + _PAGE_PSE.
 249                                  */
 250                                 pgprot_t init_prot =
 251                                         __pgprot(PTE_IDENT_ATTR |
 252                                                  _PAGE_PSE);
 253
 254                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 255                                         PAGE_OFFSET + PAGE_SIZE-1;
 256
 257                                 if (is_kernel_text(addr) ||
 258                                     is_kernel_text(addr2))
 259                                         prot = PAGE_KERNEL_LARGE_EXEC;
 260
 261                                 pages_2m++;
 262                                 if (mapping_iter == 1)
 263                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
 264                                 else
 265                                         set_pmd(pmd, pfn_pmd(pfn, prot));
 266
 267                                 pfn += PTRS_PER_PTE;
 268                                 continue;
 269                         }
 270                         pte = one_page_table_init(pmd);
 271
 272                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 273                         pte += pte_ofs;
 274                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 275                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 276                                 pgprot_t prot = PAGE_KERNEL;
 277                                 /*
 278                                  * first pass will use the same initial
 279                                  * identity mapping attribute.
 280                                  */
 281                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
 282
 283                                 if (is_kernel_text(addr))
 284                                         prot = PAGE_KERNEL_EXEC;
 285
 286                                 pages_4k++;
 287                                 if (mapping_iter == 1)
 288                                         set_pte(pte, pfn_pte(pfn, init_prot));
 289                                 else
 290                                         set_pte(pte, pfn_pte(pfn, prot));
 291                         }
 292                 }
 293         }
 294         if (mapping_iter == 1) {
 295                 /*
 296                  * update direct mapping page count only in the first
 297                  * iteration.
 298                  */
 299                 update_page_count(PG_LEVEL_2M, pages_2m);
 300                 update_page_count(PG_LEVEL_4K, pages_4k);
 301
 302                 /*
 303                  * local global flush tlb, which will flush the previous
 304                  * mappings present in both small and large page TLB's.
 305                  */
 306                 __flush_tlb_all();
 307
 308                 /*
 309                  * Second iteration will set the actual desired PTE attributes.
 310                  */
 311                 mapping_iter = 2;
 312                 goto repeat;
 313         }
 314 }
 315
 316 /*
 317  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
 318  * is valid. The argument is a physical page number.
 319  *
 320  *
 321  * On x86, access has to be given to the first megabyte of ram because that area
 322  * contains bios code and data regions used by X and dosemu and similar apps.
 323  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 324  * mmio resources as well as potential bios/acpi data regions.
 325  */
 326 int devmem_is_allowed(unsigned long pagenr)
 327 {
 328         if (pagenr <= 256)
 329                 return 1;
 330         if (!page_is_ram(pagenr))
 331                 return 1;
 332         return 0;
 333 }
 334
 335 #ifdef CONFIG_HIGHMEM
 336 pte_t *kmap_pte;
 337 pgprot_t kmap_prot;
 338
 339 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 340 {
 341         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 342                         vaddr), vaddr), vaddr);
 343 }
 344
 345 static void __init kmap_init(void)
 346 {
 347         unsigned long kmap_vstart;
 348
 349         /*
 350          * Cache the first kmap pte:
 351          */
 352         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 353         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 354
 355         kmap_prot = PAGE_KERNEL;
 356 }
 357
 358 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 359 {
 360         unsigned long vaddr;
 361         pgd_t *pgd;
 362         pud_t *pud;
 363         pmd_t *pmd;
 364         pte_t *pte;
 365
 366         vaddr = PKMAP_BASE;
 367         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 368
 369         pgd = swapper_pg_dir + pgd_index(vaddr);
 370         pud = pud_offset(pgd, vaddr);
 371         pmd = pmd_offset(pud, vaddr);
 372         pte = pte_offset_kernel(pmd, vaddr);
 373         pkmap_page_table = pte;
 374 }
 375
 376 static void __init add_one_highpage_init(struct page *page, int pfn)
 377 {
 378         ClearPageReserved(page);
 379         init_page_count(page);
 380         __free_page(page);
 381         totalhigh_pages++;
 382 }
 383
 384 struct add_highpages_data {
 385         unsigned long start_pfn;
 386         unsigned long end_pfn;
 387 };
 388
 389 static int __init add_highpages_work_fn(unsigned long start_pfn,
 390                                          unsigned long end_pfn, void *datax)
 391 {
 392         int node_pfn;
 393         struct page *page;
 394         unsigned long final_start_pfn, final_end_pfn;
 395         struct add_highpages_data *data;
 396
 397         data = (struct add_highpages_data *)datax;
 398
 399         final_start_pfn = max(start_pfn, data->start_pfn);
 400         final_end_pfn = min(end_pfn, data->end_pfn);
 401         if (final_start_pfn >= final_end_pfn)
 402                 return 0;
 403
 404         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
 405              node_pfn++) {
 406                 if (!pfn_valid(node_pfn))
 407                         continue;
 408                 page = pfn_to_page(node_pfn);
 409                 add_one_highpage_init(page, node_pfn);
 410         }
 411
 412         return 0;
 413
 414 }
 415
 416 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
 417                                               unsigned long end_pfn)
 418 {
 419         struct add_highpages_data data;
 420
 421         data.start_pfn = start_pfn;
 422         data.end_pfn = end_pfn;
 423
 424         work_with_active_regions(nid, add_highpages_work_fn, &data);
 425 }
 426
 427 #ifndef CONFIG_NUMA
 428 static void __init set_highmem_pages_init(void)
 429 {
 430         add_highpages_with_active_regions(0, highstart_pfn, highend_pfn);
 431
 432         totalram_pages += totalhigh_pages;
 433 }
 434 #endif /* !CONFIG_NUMA */
 435
 436 #else
 437 # define kmap_init()                            do { } while (0)
 438 # define permanent_kmaps_init(pgd_base)         do { } while (0)
 439 # define set_highmem_pages_init()       do { } while (0)
 440 #endif /* CONFIG_HIGHMEM */
 441
 442 void __init native_pagetable_setup_start(pgd_t *base)
 443 {
 444         unsigned long pfn, va;
 445         pgd_t *pgd;
 446         pud_t *pud;
 447         pmd_t *pmd;
 448         pte_t *pte;
 449
 450         /*
 451          * Remove any mappings which extend past the end of physical
 452          * memory from the boot time page table:
 453          */
 454         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 455                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 456                 pgd = base + pgd_index(va);
 457                 if (!pgd_present(*pgd))
 458                         break;
 459
 460                 pud = pud_offset(pgd, va);
 461                 pmd = pmd_offset(pud, va);
 462                 if (!pmd_present(*pmd))
 463                         break;
 464
 465                 pte = pte_offset_kernel(pmd, va);
 466                 if (!pte_present(*pte))
 467                         break;
 468
 469                 pte_clear(NULL, va, pte);
 470         }
 471         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 472 }
 473
 474 void __init native_pagetable_setup_done(pgd_t *base)
 475 {
 476 }
 477
 478 /*
 479  * Build a proper pagetable for the kernel mappings.  Up until this
 480  * point, we've been running on some set of pagetables constructed by
 481  * the boot process.
 482  *
 483  * If we're booting on native hardware, this will be a pagetable
 484  * constructed in arch/x86/kernel/head_32.S.  The root of the
 485  * pagetable will be swapper_pg_dir.
 486  *
 487  * If we're booting paravirtualized under a hypervisor, then there are
 488  * more options: we may already be running PAE, and the pagetable may
 489  * or may not be based in swapper_pg_dir.  In any case,
 490  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 491  * appropriately for the rest of the initialization to work.
 492  *
 493  * In general, pagetable_init() assumes that the pagetable may already
 494  * be partially populated, and so it avoids stomping on any existing
 495  * mappings.
 496  */
 497 static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base)
 498 {
 499         unsigned long vaddr, end;
 500
 501         /*
 502          * Fixed mappings, only the page table structure has to be
 503          * created - mappings will be set by set_fixmap():
 504          */
 505         early_ioremap_clear();
 506         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 507         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 508         page_table_range_init(vaddr, end, pgd_base);
 509         early_ioremap_reset();
 510 }
 511
 512 static void __init pagetable_init(void)
 513 {
 514         pgd_t *pgd_base = swapper_pg_dir;
 515
 516         paravirt_pagetable_setup_start(pgd_base);
 517
 518         permanent_kmaps_init(pgd_base);
 519
 520         paravirt_pagetable_setup_done(pgd_base);
 521 }
 522
 523 #ifdef CONFIG_ACPI_SLEEP
 524 /*
 525  * ACPI suspend needs this for resume, because things like the intel-agp
 526  * driver might have split up a kernel 4MB mapping.
 527  */
 528 char swsusp_pg_dir[PAGE_SIZE]
 529         __attribute__ ((aligned(PAGE_SIZE)));
 530
 531 static inline void save_pg_dir(void)
 532 {
 533         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 534 }
 535 #else /* !CONFIG_ACPI_SLEEP */
 536 static inline void save_pg_dir(void)
 537 {
 538 }
 539 #endif /* !CONFIG_ACPI_SLEEP */
 540
 541 void zap_low_mappings(void)
 542 {
 543         int i;
 544
 545         /*
 546          * Zap initial low-memory mappings.
 547          *
 548          * Note that "pgd_clear()" doesn't do it for
 549          * us, because pgd_clear() is a no-op on i386.
 550          */
 551         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 552 #ifdef CONFIG_X86_PAE
 553                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 554 #else
 555                 set_pgd(swapper_pg_dir+i, __pgd(0));
 556 #endif
 557         }
 558         flush_tlb_all();
 559 }
 560
 561 int nx_enabled;
 562
 563 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL);
 564 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 565
 566 #ifdef CONFIG_X86_PAE
 567
 568 static int disable_nx __initdata;
 569
 570 /*
 571  * noexec = on|off
 572  *
 573  * Control non executable mappings.
 574  *
 575  * on      Enable
 576  * off     Disable
 577  */
 578 static int __init noexec_setup(char *str)
 579 {
 580         if (!str || !strcmp(str, "on")) {
 581                 if (cpu_has_nx) {
 582                         __supported_pte_mask |= _PAGE_NX;
 583                         disable_nx = 0;
 584                 }
 585         } else {
 586                 if (!strcmp(str, "off")) {
 587                         disable_nx = 1;
 588                         __supported_pte_mask &= ~_PAGE_NX;
 589                 } else {
 590                         return -EINVAL;
 591                 }
 592         }
 593
 594         return 0;
 595 }
 596 early_param("noexec", noexec_setup);
 597
 598 static void __init set_nx(void)
 599 {
 600         unsigned int v[4], l, h;
 601
 602         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 603                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 604
 605                 if ((v[3] & (1 << 20)) && !disable_nx) {
 606                         rdmsr(MSR_EFER, l, h);
 607                         l |= EFER_NX;
 608                         wrmsr(MSR_EFER, l, h);
 609                         nx_enabled = 1;
 610                         __supported_pte_mask |= _PAGE_NX;
 611                 }
 612         }
 613 }
 614 #endif
 615
 616 /* user-defined highmem size */
 617 static unsigned int highmem_pages = -1;
 618
 619 /*
 620  * highmem=size forces highmem to be exactly 'size' bytes.
 621  * This works even on boxes that have no highmem otherwise.
 622  * This also works to reduce highmem size on bigger boxes.
 623  */
 624 static int __init parse_highmem(char *arg)
 625 {
 626         if (!arg)
 627                 return -EINVAL;
 628
 629         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 630         return 0;
 631 }
 632 early_param("highmem", parse_highmem);
 633
 634 /*
 635  * Determine low and high memory ranges:
 636  */
 637 void __init find_low_pfn_range(void)
 638 {
 639         /* it could update max_pfn */
 640
 641         /* max_low_pfn is 0, we already have early_res support */
 642
 643         max_low_pfn = max_pfn;
 644         if (max_low_pfn > MAXMEM_PFN) {
 645                 if (highmem_pages == -1)
 646                         highmem_pages = max_pfn - MAXMEM_PFN;
 647                 if (highmem_pages + MAXMEM_PFN < max_pfn)
 648                         max_pfn = MAXMEM_PFN + highmem_pages;
 649                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
 650                         printk(KERN_WARNING "only %luMB highmem pages "
 651                                 "available, ignoring highmem size of %uMB.\n",
 652                                 pages_to_mb(max_pfn - MAXMEM_PFN),
 653                                 pages_to_mb(highmem_pages));
 654                         highmem_pages = 0;
 655                 }
 656                 max_low_pfn = MAXMEM_PFN;
 657 #ifndef CONFIG_HIGHMEM
 658                 /* Maximum memory usable is what is directly addressable */
 659                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
 660                                         MAXMEM>>20);
 661                 if (max_pfn > MAX_NONPAE_PFN)
 662                         printk(KERN_WARNING
 663                                  "Use a HIGHMEM64G enabled kernel.\n");
 664                 else
 665                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 666                 max_pfn = MAXMEM_PFN;
 667 #else /* !CONFIG_HIGHMEM */
 668 #ifndef CONFIG_HIGHMEM64G
 669                 if (max_pfn > MAX_NONPAE_PFN) {
 670                         max_pfn = MAX_NONPAE_PFN;
 671                         printk(KERN_WARNING "Warning only 4GB will be used."
 672                                 "Use a HIGHMEM64G enabled kernel.\n");
 673                 }
 674 #endif /* !CONFIG_HIGHMEM64G */
 675 #endif /* !CONFIG_HIGHMEM */
 676         } else {
 677                 if (highmem_pages == -1)
 678                         highmem_pages = 0;
 679 #ifdef CONFIG_HIGHMEM
 680                 if (highmem_pages >= max_pfn) {
 681                         printk(KERN_ERR "highmem size specified (%uMB) is "
 682                                 "bigger than pages available (%luMB)!.\n",
 683                                 pages_to_mb(highmem_pages),
 684                                 pages_to_mb(max_pfn));
 685                         highmem_pages = 0;
 686                 }
 687                 if (highmem_pages) {
 688                         if (max_low_pfn - highmem_pages <
 689                             64*1024*1024/PAGE_SIZE){
 690                                 printk(KERN_ERR "highmem size %uMB results in "
 691                                 "smaller than 64MB lowmem, ignoring it.\n"
 692                                         , pages_to_mb(highmem_pages));
 693                                 highmem_pages = 0;
 694                         }
 695                         max_low_pfn -= highmem_pages;
 696                 }
 697 #else
 698                 if (highmem_pages)
 699                         printk(KERN_ERR "ignoring highmem size on non-highmem"
 700                                         " kernel!\n");
 701 #endif
 702         }
 703 }
 704
 705 #ifndef CONFIG_NEED_MULTIPLE_NODES
 706 void __init initmem_init(unsigned long start_pfn,
 707                                   unsigned long end_pfn)
 708 {
 709 #ifdef CONFIG_HIGHMEM
 710         highstart_pfn = highend_pfn = max_pfn;
 711         if (max_pfn > max_low_pfn)
 712                 highstart_pfn = max_low_pfn;
 713         memory_present(0, 0, highend_pfn);
 714         e820_register_active_regions(0, 0, highend_pfn);
 715         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 716                 pages_to_mb(highend_pfn - highstart_pfn));
 717         num_physpages = highend_pfn;
 718         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 719 #else
 720         memory_present(0, 0, max_low_pfn);
 721         e820_register_active_regions(0, 0, max_low_pfn);
 722         num_physpages = max_low_pfn;
 723         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 724 #endif
 725 #ifdef CONFIG_FLATMEM
 726         max_mapnr = num_physpages;
 727 #endif
 728         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 729                         pages_to_mb(max_low_pfn));
 730
 731         setup_bootmem_allocator();
 732 }
 733 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 734
 735 static void __init zone_sizes_init(void)
 736 {
 737         unsigned long max_zone_pfns[MAX_NR_ZONES];
 738         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 739         max_zone_pfns[ZONE_DMA] =
 740                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 741         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 742 #ifdef CONFIG_HIGHMEM
 743         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 744 #endif
 745
 746         free_area_init_nodes(max_zone_pfns);
 747 }
 748
 749 void __init setup_bootmem_allocator(void)
 750 {
 751         int i;
 752         unsigned long bootmap_size, bootmap;
 753         /*
 754          * Initialize the boot-time allocator (with low memory only):
 755          */
 756         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 757         bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT,
 758                                  max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 759                                  PAGE_SIZE);
 760         if (bootmap == -1L)
 761                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 762         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 763
 764         /* don't touch min_low_pfn */
 765         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
 766                                          min_low_pfn, max_low_pfn);
 767         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 768                  max_pfn_mapped<<PAGE_SHIFT);
 769         printk(KERN_INFO "  low ram: %08lx - %08lx\n",
 770                  min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT);
 771         printk(KERN_INFO "  bootmap %08lx - %08lx\n",
 772                  bootmap, bootmap + bootmap_size);
 773         for_each_online_node(i)
 774                 free_bootmem_with_active_regions(i, max_low_pfn);
 775         early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
 776
 777         after_init_bootmem = 1;
 778 }
 779
 780 static void __init find_early_table_space(unsigned long end)
 781 {
 782         unsigned long puds, pmds, ptes, tables, start;
 783
 784         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 785         tables = PAGE_ALIGN(puds * sizeof(pud_t));
 786
 787         pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 788         tables += PAGE_ALIGN(pmds * sizeof(pmd_t));
 789
 790         if (cpu_has_pse) {
 791                 unsigned long extra;
 792
 793                 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
 794                 extra += PMD_SIZE;
 795                 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
 796         } else
 797                 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
 798
 799         tables += PAGE_ALIGN(ptes * sizeof(pte_t));
 800
 801         /* for fixmap */
 802         tables += PAGE_SIZE * 2;
 803
 804         /*
 805          * RED-PEN putting page tables only on node 0 could
 806          * cause a hotspot and fill up ZONE_DMA. The page tables
 807          * need roughly 0.5KB per GB.
 808          */
 809         start = 0x7000;
 810         table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
 811                                         tables, PAGE_SIZE);
 812         if (table_start == -1UL)
 813                 panic("Cannot find space for the kernel page tables");
 814
 815         table_start >>= PAGE_SHIFT;
 816         table_end = table_start;
 817         table_top = table_start + (tables>>PAGE_SHIFT);
 818
 819         printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
 820                 end, table_start << PAGE_SHIFT,
 821                 (table_start << PAGE_SHIFT) + tables);
 822 }
 823
 824 unsigned long __init_refok init_memory_mapping(unsigned long start,
 825                                                 unsigned long end)
 826 {
 827         pgd_t *pgd_base = swapper_pg_dir;
 828         unsigned long start_pfn, end_pfn;
 829         unsigned long big_page_start;
 830
 831         /*
 832          * Find space for the kernel direct mapping tables.
 833          */
 834         if (!after_init_bootmem)
 835                 find_early_table_space(end);
 836
 837 #ifdef CONFIG_X86_PAE
 838         set_nx();
 839         if (nx_enabled)
 840                 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
 841 #endif
 842
 843         /* Enable PSE if available */
 844         if (cpu_has_pse)
 845                 set_in_cr4(X86_CR4_PSE);
 846
 847         /* Enable PGE if available */
 848         if (cpu_has_pge) {
 849                 set_in_cr4(X86_CR4_PGE);
 850                 __supported_pte_mask |= _PAGE_GLOBAL;
 851         }
 852
 853         /*
 854          * Don't use a large page for the first 2/4MB of memory
 855          * because there are often fixed size MTRRs in there
 856          * and overlapping MTRRs into large pages can cause
 857          * slowdowns.
 858          */
 859         big_page_start = PMD_SIZE;
 860
 861         if (start < big_page_start) {
 862                 start_pfn = start >> PAGE_SHIFT;
 863                 end_pfn = min(big_page_start>>PAGE_SHIFT, end>>PAGE_SHIFT);
 864         } else {
 865                 /* head is not big page alignment ? */
 866                 start_pfn = start >> PAGE_SHIFT;
 867                 end_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
 868                                  << (PMD_SHIFT - PAGE_SHIFT);
 869         }
 870         if (start_pfn < end_pfn)
 871                 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, 0);
 872
 873         /* big page range */
 874         start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
 875                          << (PMD_SHIFT - PAGE_SHIFT);
 876         if (start_pfn < (big_page_start >> PAGE_SHIFT))
 877                 start_pfn =  big_page_start >> PAGE_SHIFT;
 878         end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 879         if (start_pfn < end_pfn)
 880                 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn,
 881                                                 cpu_has_pse);
 882
 883         /* tail is not big page alignment ? */
 884         start_pfn = end_pfn;
 885         if (start_pfn > (big_page_start>>PAGE_SHIFT)) {
 886                 end_pfn = end >> PAGE_SHIFT;
 887                 if (start_pfn < end_pfn)
 888                         kernel_physical_mapping_init(pgd_base, start_pfn,
 889                                                          end_pfn, 0);
 890         }
 891
 892         early_ioremap_page_table_range_init(pgd_base);
 893
 894         load_cr3(swapper_pg_dir);
 895
 896         __flush_tlb_all();
 897
 898         if (!after_init_bootmem)
 899                 reserve_early(table_start << PAGE_SHIFT,
 900                                  table_end << PAGE_SHIFT, "PGTABLE");
 901
 902         if (!after_init_bootmem)
 903                 early_memtest(start, end);
 904
 905         return end >> PAGE_SHIFT;
 906 }
 907
 908
 909 /*
 910  * paging_init() sets up the page tables - note that the first 8MB are
 911  * already mapped by head.S.
 912  *
 913  * This routines also unmaps the page at virtual kernel address 0, so
 914  * that we can trap those pesky NULL-reference errors in the kernel.
 915  */
 916 void __init paging_init(void)
 917 {
 918         pagetable_init();
 919
 920         __flush_tlb_all();
 921
 922         kmap_init();
 923
 924         /*
 925          * NOTE: at this point the bootmem allocator is fully available.
 926          */
 927         sparse_init();
 928         zone_sizes_init();
 929 }
 930
 931 /*
 932  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 933  * and also on some strange 486's. All 586+'s are OK. This used to involve
 934  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 935  * switch to using exceptions got rid of all that.
 936  */
 937 static void __init test_wp_bit(void)
 938 {
 939         printk(KERN_INFO
 940   "Checking if this processor honours the WP bit even in supervisor mode...");
 941
 942         /* Any page-aligned address will do, the test is non-destructive */
 943         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 944         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 945         clear_fixmap(FIX_WP_TEST);
 946
 947         if (!boot_cpu_data.wp_works_ok) {
 948                 printk(KERN_CONT "No.\n");
 949 #ifdef CONFIG_X86_WP_WORKS_OK
 950                 panic(
 951   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 952 #endif
 953         } else {
 954                 printk(KERN_CONT "Ok.\n");
 955         }
 956 }
 957
 958 static struct kcore_list kcore_mem, kcore_vmalloc;
 959
 960 void __init mem_init(void)
 961 {
 962         int codesize, reservedpages, datasize, initsize;
 963         int tmp;
 964
 965 #ifdef CONFIG_FLATMEM
 966         BUG_ON(!mem_map);
 967 #endif
 968         /* this will put all low memory onto the freelists */
 969         totalram_pages += free_all_bootmem();
 970
 971         reservedpages = 0;
 972         for (tmp = 0; tmp < max_low_pfn; tmp++)
 973                 /*
 974                  * Only count reserved RAM pages:
 975                  */
 976                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 977                         reservedpages++;
 978
 979         set_highmem_pages_init();
 980
 981         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 982         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 983         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 984
 985         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 986         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 987                    VMALLOC_END-VMALLOC_START);
 988
 989         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 990                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 991                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 992                 num_physpages << (PAGE_SHIFT-10),
 993                 codesize >> 10,
 994                 reservedpages << (PAGE_SHIFT-10),
 995                 datasize >> 10,
 996                 initsize >> 10,
 997                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 998                );
 999
1000         printk(KERN_INFO "virtual kernel memory layout:\n"
1001                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1002 #ifdef CONFIG_HIGHMEM
1003                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1004 #endif
1005                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
1006                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
1007                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1008                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1009                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
1010                 FIXADDR_START, FIXADDR_TOP,
1011                 (FIXADDR_TOP - FIXADDR_START) >> 10,
1012
1013 #ifdef CONFIG_HIGHMEM
1014                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
1015                 (LAST_PKMAP*PAGE_SIZE) >> 10,
1016 #endif
1017
1018                 VMALLOC_START, VMALLOC_END,
1019                 (VMALLOC_END - VMALLOC_START) >> 20,
1020
1021                 (unsigned long)__va(0), (unsigned long)high_memory,
1022                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
1023
1024                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
1025                 ((unsigned long)&__init_end -
1026                  (unsigned long)&__init_begin) >> 10,
1027
1028                 (unsigned long)&_etext, (unsigned long)&_edata,
1029                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
1030
1031                 (unsigned long)&_text, (unsigned long)&_etext,
1032                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
1033
1034 #ifdef CONFIG_HIGHMEM
1035         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
1036         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
1037 #endif
1038         BUG_ON(VMALLOC_START                            > VMALLOC_END);
1039         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
1040
1041         if (boot_cpu_data.wp_works_ok < 0)
1042                 test_wp_bit();
1043
1044         cpa_init();
1045         save_pg_dir();
1046         zap_low_mappings();
1047 }
1048
1049 #ifdef CONFIG_MEMORY_HOTPLUG
1050 int arch_add_memory(int nid, u64 start, u64 size)
1051 {
1052         struct pglist_data *pgdata = NODE_DATA(nid);
1053         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1054         unsigned long start_pfn = start >> PAGE_SHIFT;
1055         unsigned long nr_pages = size >> PAGE_SHIFT;
1056
1057         return __add_pages(zone, start_pfn, nr_pages);
1058 }
1059 #endif
1060
1061 /*
1062  * This function cannot be __init, since exceptions don't work in that
1063  * section.  Put this after the callers, so that it cannot be inlined.
1064  */
1065 static noinline int do_test_wp_bit(void)
1066 {
1067         char tmp_reg;
1068         int flag;
1069
1070         __asm__ __volatile__(
1071                 "       movb %0, %1     \n"
1072                 "1:     movb %1, %0     \n"
1073                 "       xorl %2, %2     \n"
1074                 "2:                     \n"
1075                 _ASM_EXTABLE(1b,2b)
1076                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1077                  "=q" (tmp_reg),
1078                  "=r" (flag)
1079                 :"2" (1)
1080                 :"memory");
1081
1082         return flag;
1083 }
1084
1085 #ifdef CONFIG_DEBUG_RODATA
1086 const int rodata_test_data = 0xC3;
1087 EXPORT_SYMBOL_GPL(rodata_test_data);
1088
1089 void mark_rodata_ro(void)
1090 {
1091         unsigned long start = PFN_ALIGN(_text);
1092         unsigned long size = PFN_ALIGN(_etext) - start;
1093
1094 #ifndef CONFIG_DYNAMIC_FTRACE
1095         /* Dynamic tracing modifies the kernel text section */
1096         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1097         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1098                 size >> 10);
1099
1100 #ifdef CONFIG_CPA_DEBUG
1101         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1102                 start, start+size);
1103         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1104
1105         printk(KERN_INFO "Testing CPA: write protecting again\n");
1106         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1107 #endif
1108 #endif /* CONFIG_DYNAMIC_FTRACE */
1109
1110         start += size;
1111         size = (unsigned long)__end_rodata - start;
1112         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1113         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1114                 size >> 10);
1115         rodata_test();
1116
1117 #ifdef CONFIG_CPA_DEBUG
1118         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1119         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1120
1121         printk(KERN_INFO "Testing CPA: write protecting again\n");
1122         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1123 #endif
1124 }
1125 #endif
1126
1127 void free_init_pages(char *what, unsigned long begin, unsigned long end)
1128 {
1129 #ifdef CONFIG_DEBUG_PAGEALLOC
1130         /*
1131          * If debugging page accesses then do not free this memory but
1132          * mark them not present - any buggy init-section access will
1133          * create a kernel page fault:
1134          */
1135         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
1136                 begin, PAGE_ALIGN(end));
1137         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
1138 #else
1139         unsigned long addr;
1140
1141         /*
1142          * We just marked the kernel text read only above, now that
1143          * we are going to free part of that, we need to make that
1144          * writeable first.
1145          */
1146         set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
1147
1148         for (addr = begin; addr < end; addr += PAGE_SIZE) {
1149                 ClearPageReserved(virt_to_page(addr));
1150                 init_page_count(virt_to_page(addr));
1151                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
1152                 free_page(addr);
1153                 totalram_pages++;
1154         }
1155         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
1156 #endif
1157 }
1158
1159 void free_initmem(void)
1160 {
1161         free_init_pages("unused kernel memory",
1162                         (unsigned long)(&__init_begin),
1163                         (unsigned long)(&__init_end));
1164 }
1165
1166 #ifdef CONFIG_BLK_DEV_INITRD
1167 void free_initrd_mem(unsigned long start, unsigned long end)
1168 {
1169         free_init_pages("initrd memory", start, end);
1170 }
1171 #endif
1172
1173 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1174                                    int flags)
1175 {
1176         return reserve_bootmem(phys, len, flags);
1177 }