git.oblomov.eu Git - 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/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 int after_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_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_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_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(unsigned long start_pfn,
 231                                                 unsigned long end_pfn,
 232                                                 int use_pse)
 233 {
 234         pgd_t *pgd_base = swapper_pg_dir;
 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(void)
 513 {
 514         pgd_t *pgd_base = swapper_pg_dir;
 515         unsigned long vaddr, end;
 516
 517         /*
 518          * Fixed mappings, only the page table structure has to be
 519          * created - mappings will be set by set_fixmap():
 520          */
 521         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 522         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 523         page_table_range_init(vaddr, end, pgd_base);
 524         early_ioremap_reset();
 525 }
 526
 527 static void __init pagetable_init(void)
 528 {
 529         pgd_t *pgd_base = swapper_pg_dir;
 530
 531         permanent_kmaps_init(pgd_base);
 532 }
 533
 534 #ifdef CONFIG_ACPI_SLEEP
 535 /*
 536  * ACPI suspend needs this for resume, because things like the intel-agp
 537  * driver might have split up a kernel 4MB mapping.
 538  */
 539 char swsusp_pg_dir[PAGE_SIZE]
 540         __attribute__ ((aligned(PAGE_SIZE)));
 541
 542 static inline void save_pg_dir(void)
 543 {
 544         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 545 }
 546 #else /* !CONFIG_ACPI_SLEEP */
 547 static inline void save_pg_dir(void)
 548 {
 549 }
 550 #endif /* !CONFIG_ACPI_SLEEP */
 551
 552 void zap_low_mappings(void)
 553 {
 554         int i;
 555
 556         /*
 557          * Zap initial low-memory mappings.
 558          *
 559          * Note that "pgd_clear()" doesn't do it for
 560          * us, because pgd_clear() is a no-op on i386.
 561          */
 562         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 563 #ifdef CONFIG_X86_PAE
 564                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 565 #else
 566                 set_pgd(swapper_pg_dir+i, __pgd(0));
 567 #endif
 568         }
 569         flush_tlb_all();
 570 }
 571
 572 int nx_enabled;
 573
 574 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
 575 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 576
 577 #ifdef CONFIG_X86_PAE
 578
 579 static int disable_nx __initdata;
 580
 581 /*
 582  * noexec = on|off
 583  *
 584  * Control non executable mappings.
 585  *
 586  * on      Enable
 587  * off     Disable
 588  */
 589 static int __init noexec_setup(char *str)
 590 {
 591         if (!str || !strcmp(str, "on")) {
 592                 if (cpu_has_nx) {
 593                         __supported_pte_mask |= _PAGE_NX;
 594                         disable_nx = 0;
 595                 }
 596         } else {
 597                 if (!strcmp(str, "off")) {
 598                         disable_nx = 1;
 599                         __supported_pte_mask &= ~_PAGE_NX;
 600                 } else {
 601                         return -EINVAL;
 602                 }
 603         }
 604
 605         return 0;
 606 }
 607 early_param("noexec", noexec_setup);
 608
 609 static void __init set_nx(void)
 610 {
 611         unsigned int v[4], l, h;
 612
 613         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 614                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 615
 616                 if ((v[3] & (1 << 20)) && !disable_nx) {
 617                         rdmsr(MSR_EFER, l, h);
 618                         l |= EFER_NX;
 619                         wrmsr(MSR_EFER, l, h);
 620                         nx_enabled = 1;
 621                         __supported_pte_mask |= _PAGE_NX;
 622                 }
 623         }
 624 }
 625 #endif
 626
 627 /* user-defined highmem size */
 628 static unsigned int highmem_pages = -1;
 629
 630 /*
 631  * highmem=size forces highmem to be exactly 'size' bytes.
 632  * This works even on boxes that have no highmem otherwise.
 633  * This also works to reduce highmem size on bigger boxes.
 634  */
 635 static int __init parse_highmem(char *arg)
 636 {
 637         if (!arg)
 638                 return -EINVAL;
 639
 640         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 641         return 0;
 642 }
 643 early_param("highmem", parse_highmem);
 644
 645 #define MSG_HIGHMEM_TOO_BIG \
 646         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
 647
 648 #define MSG_LOWMEM_TOO_SMALL \
 649         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
 650 /*
 651  * All of RAM fits into lowmem - but if user wants highmem
 652  * artificially via the highmem=x boot parameter then create
 653  * it:
 654  */
 655 void __init lowmem_pfn_init(void)
 656 {
 657         /* max_low_pfn is 0, we already have early_res support */
 658         max_low_pfn = max_pfn;
 659
 660         if (highmem_pages == -1)
 661                 highmem_pages = 0;
 662 #ifdef CONFIG_HIGHMEM
 663         if (highmem_pages >= max_pfn) {
 664                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
 665                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 666                 highmem_pages = 0;
 667         }
 668         if (highmem_pages) {
 669                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
 670                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
 671                                 pages_to_mb(highmem_pages));
 672                         highmem_pages = 0;
 673                 }
 674                 max_low_pfn -= highmem_pages;
 675         }
 676 #else
 677         if (highmem_pages)
 678                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 679 #endif
 680 }
 681
 682 #define MSG_HIGHMEM_TOO_SMALL \
 683         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
 684
 685 #define MSG_HIGHMEM_TRIMMED \
 686         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
 687 /*
 688  * We have more RAM than fits into lowmem - we try to put it into
 689  * highmem, also taking the highmem=x boot parameter into account:
 690  */
 691 void __init highmem_pfn_init(void)
 692 {
 693         max_low_pfn = MAXMEM_PFN;
 694
 695         if (highmem_pages == -1)
 696                 highmem_pages = max_pfn - MAXMEM_PFN;
 697
 698         if (highmem_pages + MAXMEM_PFN < max_pfn)
 699                 max_pfn = MAXMEM_PFN + highmem_pages;
 700
 701         if (highmem_pages + MAXMEM_PFN > max_pfn) {
 702                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
 703                         pages_to_mb(max_pfn - MAXMEM_PFN),
 704                         pages_to_mb(highmem_pages));
 705                 highmem_pages = 0;
 706         }
 707 #ifndef CONFIG_HIGHMEM
 708         /* Maximum memory usable is what is directly addressable */
 709         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
 710         if (max_pfn > MAX_NONPAE_PFN)
 711                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 712         else
 713                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 714         max_pfn = MAXMEM_PFN;
 715 #else /* !CONFIG_HIGHMEM */
 716 #ifndef CONFIG_HIGHMEM64G
 717         if (max_pfn > MAX_NONPAE_PFN) {
 718                 max_pfn = MAX_NONPAE_PFN;
 719                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
 720         }
 721 #endif /* !CONFIG_HIGHMEM64G */
 722 #endif /* !CONFIG_HIGHMEM */
 723 }
 724
 725 /*
 726  * Determine low and high memory ranges:
 727  */
 728 void __init find_low_pfn_range(void)
 729 {
 730         /* it could update max_pfn */
 731
 732         if (max_pfn <= MAXMEM_PFN)
 733                 lowmem_pfn_init();
 734         else
 735                 highmem_pfn_init();
 736 }
 737
 738 #ifndef CONFIG_NEED_MULTIPLE_NODES
 739 void __init initmem_init(unsigned long start_pfn,
 740                                   unsigned long end_pfn)
 741 {
 742 #ifdef CONFIG_HIGHMEM
 743         highstart_pfn = highend_pfn = max_pfn;
 744         if (max_pfn > max_low_pfn)
 745                 highstart_pfn = max_low_pfn;
 746         memory_present(0, 0, highend_pfn);
 747         e820_register_active_regions(0, 0, highend_pfn);
 748         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 749                 pages_to_mb(highend_pfn - highstart_pfn));
 750         num_physpages = highend_pfn;
 751         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 752 #else
 753         memory_present(0, 0, max_low_pfn);
 754         e820_register_active_regions(0, 0, max_low_pfn);
 755         num_physpages = max_low_pfn;
 756         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 757 #endif
 758 #ifdef CONFIG_FLATMEM
 759         max_mapnr = num_physpages;
 760 #endif
 761         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 762                         pages_to_mb(max_low_pfn));
 763
 764         setup_bootmem_allocator();
 765 }
 766 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 767
 768 static void __init zone_sizes_init(void)
 769 {
 770         unsigned long max_zone_pfns[MAX_NR_ZONES];
 771         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 772         max_zone_pfns[ZONE_DMA] =
 773                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 774         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 775 #ifdef CONFIG_HIGHMEM
 776         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 777 #endif
 778
 779         free_area_init_nodes(max_zone_pfns);
 780 }
 781
 782 static unsigned long __init setup_node_bootmem(int nodeid,
 783                                  unsigned long start_pfn,
 784                                  unsigned long end_pfn,
 785                                  unsigned long bootmap)
 786 {
 787         unsigned long bootmap_size;
 788
 789         if (start_pfn > max_low_pfn)
 790                 return bootmap;
 791         if (end_pfn > max_low_pfn)
 792                 end_pfn = max_low_pfn;
 793
 794         /* don't touch min_low_pfn */
 795         bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
 796                                          bootmap >> PAGE_SHIFT,
 797                                          start_pfn, end_pfn);
 798         printk(KERN_INFO "  node %d low ram: %08lx - %08lx\n",
 799                 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 800         printk(KERN_INFO "  node %d bootmap %08lx - %08lx\n",
 801                  nodeid, bootmap, bootmap + bootmap_size);
 802         free_bootmem_with_active_regions(nodeid, end_pfn);
 803         early_res_to_bootmem(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 804
 805         return bootmap + bootmap_size;
 806 }
 807
 808 void __init setup_bootmem_allocator(void)
 809 {
 810         int nodeid;
 811         unsigned long bootmap_size, bootmap;
 812         /*
 813          * Initialize the boot-time allocator (with low memory only):
 814          */
 815         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 816         bootmap = find_e820_area(0, max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 817                                  PAGE_SIZE);
 818         if (bootmap == -1L)
 819                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 820         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 821
 822         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 823                  max_pfn_mapped<<PAGE_SHIFT);
 824         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
 825
 826 #ifdef CONFIG_NEED_MULTIPLE_NODES
 827         for_each_online_node(nodeid)
 828                 bootmap = setup_node_bootmem(nodeid, node_start_pfn[nodeid],
 829                                         node_end_pfn[nodeid], bootmap);
 830 #else
 831         bootmap = setup_node_bootmem(0, 0, max_low_pfn, bootmap);
 832 #endif
 833
 834         after_bootmem = 1;
 835 }
 836
 837 static void __init find_early_table_space(unsigned long end, int use_pse,
 838                                           int use_gbpages)
 839 {
 840         unsigned long puds, pmds, ptes, tables, start;
 841
 842         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 843         tables = roundup(puds * sizeof(pud_t), PAGE_SIZE);
 844
 845         if (use_gbpages) {
 846                 unsigned long extra;
 847
 848                 extra = end - ((end>>PUD_SHIFT) << PUD_SHIFT);
 849                 pmds = (extra + PMD_SIZE - 1) >> PMD_SHIFT;
 850         } else
 851                 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 852
 853         tables += roundup(pmds * sizeof(pmd_t), PAGE_SIZE);
 854
 855         if (use_pse) {
 856                 unsigned long extra;
 857
 858                 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
 859 #ifdef CONFIG_X86_32
 860                 extra += PMD_SIZE;
 861 #endif
 862                 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
 863         } else
 864                 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
 865
 866         tables += roundup(ptes * sizeof(pte_t), PAGE_SIZE);
 867
 868 #ifdef CONFIG_X86_32
 869         /* for fixmap */
 870         tables += roundup(__end_of_fixed_addresses * sizeof(pte_t), PAGE_SIZE);
 871 #endif
 872
 873         /*
 874          * RED-PEN putting page tables only on node 0 could
 875          * cause a hotspot and fill up ZONE_DMA. The page tables
 876          * need roughly 0.5KB per GB.
 877          */
 878 #ifdef CONFIG_X86_32
 879         start = 0x7000;
 880         table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
 881                                         tables, PAGE_SIZE);
 882 #else /* CONFIG_X86_64 */
 883         start = 0x8000;
 884         table_start = find_e820_area(start, end, tables, PAGE_SIZE);
 885 #endif
 886         if (table_start == -1UL)
 887                 panic("Cannot find space for the kernel page tables");
 888
 889         table_start >>= PAGE_SHIFT;
 890         table_end = table_start;
 891         table_top = table_start + (tables >> PAGE_SHIFT);
 892
 893         printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
 894                 end, table_start << PAGE_SHIFT, table_top << PAGE_SHIFT);
 895 }
 896
 897 struct map_range {
 898         unsigned long start;
 899         unsigned long end;
 900         unsigned page_size_mask;
 901 };
 902
 903 #define NR_RANGE_MR 3
 904
 905 static int save_mr(struct map_range *mr, int nr_range,
 906                    unsigned long start_pfn, unsigned long end_pfn,
 907                    unsigned long page_size_mask)
 908 {
 909         if (start_pfn < end_pfn) {
 910                 if (nr_range >= NR_RANGE_MR)
 911                         panic("run out of range for init_memory_mapping\n");
 912                 mr[nr_range].start = start_pfn<<PAGE_SHIFT;
 913                 mr[nr_range].end   = end_pfn<<PAGE_SHIFT;
 914                 mr[nr_range].page_size_mask = page_size_mask;
 915                 nr_range++;
 916         }
 917
 918         return nr_range;
 919 }
 920
 921 /*
 922  * Setup the direct mapping of the physical memory at PAGE_OFFSET.
 923  * This runs before bootmem is initialized and gets pages directly from
 924  * the physical memory. To access them they are temporarily mapped.
 925  */
 926 unsigned long __init_refok init_memory_mapping(unsigned long start,
 927                                                unsigned long end)
 928 {
 929         unsigned long page_size_mask = 0;
 930         unsigned long start_pfn, end_pfn;
 931         unsigned long pos;
 932         unsigned long ret;
 933
 934         struct map_range mr[NR_RANGE_MR];
 935         int nr_range, i;
 936         int use_pse, use_gbpages;
 937
 938         printk(KERN_INFO "init_memory_mapping: %016lx-%016lx\n", start, end);
 939
 940 #ifdef CONFIG_DEBUG_PAGEALLOC
 941         /*
 942          * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
 943          * This will simplify cpa(), which otherwise needs to support splitting
 944          * large pages into small in interrupt context, etc.
 945          */
 946         use_pse = use_gbpages = 0;
 947 #else
 948         use_pse = cpu_has_pse;
 949         use_gbpages = direct_gbpages;
 950 #endif
 951
 952 #ifdef CONFIG_X86_32
 953 #ifdef CONFIG_X86_PAE
 954         set_nx();
 955         if (nx_enabled)
 956                 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
 957 #endif
 958
 959         /* Enable PSE if available */
 960         if (cpu_has_pse)
 961                 set_in_cr4(X86_CR4_PSE);
 962
 963         /* Enable PGE if available */
 964         if (cpu_has_pge) {
 965                 set_in_cr4(X86_CR4_PGE);
 966                 __supported_pte_mask |= _PAGE_GLOBAL;
 967         }
 968 #endif
 969
 970         if (use_gbpages)
 971                 page_size_mask |= 1 << PG_LEVEL_1G;
 972         if (use_pse)
 973                 page_size_mask |= 1 << PG_LEVEL_2M;
 974
 975         memset(mr, 0, sizeof(mr));
 976         nr_range = 0;
 977
 978         /* head if not big page alignment ? */
 979         start_pfn = start >> PAGE_SHIFT;
 980         pos = start_pfn << PAGE_SHIFT;
 981 #ifdef CONFIG_X86_32
 982         /*
 983          * Don't use a large page for the first 2/4MB of memory
 984          * because there are often fixed size MTRRs in there
 985          * and overlapping MTRRs into large pages can cause
 986          * slowdowns.
 987          */
 988         if (pos == 0)
 989                 end_pfn = 1<<(PMD_SHIFT - PAGE_SHIFT);
 990         else
 991                 end_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
 992                                  << (PMD_SHIFT - PAGE_SHIFT);
 993 #else /* CONFIG_X86_64 */
 994         end_pfn = ((pos + (PMD_SIZE - 1)) >> PMD_SHIFT)
 995                         << (PMD_SHIFT - PAGE_SHIFT);
 996 #endif
 997         if (end_pfn > (end >> PAGE_SHIFT))
 998                 end_pfn = end >> PAGE_SHIFT;
 999         if (start_pfn < end_pfn) {
1000                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
1001                 pos = end_pfn << PAGE_SHIFT;
1002         }
1003
1004         /* big page (2M) range */
1005         start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
1006                          << (PMD_SHIFT - PAGE_SHIFT);
1007 #ifdef CONFIG_X86_32
1008         end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
1009 #else /* CONFIG_X86_64 */
1010         end_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
1011                          << (PUD_SHIFT - PAGE_SHIFT);
1012         if (end_pfn > ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT)))
1013                 end_pfn = ((end>>PMD_SHIFT)<<(PMD_SHIFT - PAGE_SHIFT));
1014 #endif
1015
1016         if (start_pfn < end_pfn) {
1017                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
1018                                 page_size_mask & (1<<PG_LEVEL_2M));
1019                 pos = end_pfn << PAGE_SHIFT;
1020         }
1021
1022 #ifdef CONFIG_X86_64
1023         /* big page (1G) range */
1024         start_pfn = ((pos + (PUD_SIZE - 1))>>PUD_SHIFT)
1025                          << (PUD_SHIFT - PAGE_SHIFT);
1026         end_pfn = (end >> PUD_SHIFT) << (PUD_SHIFT - PAGE_SHIFT);
1027         if (start_pfn < end_pfn) {
1028                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
1029                                 page_size_mask &
1030                                  ((1<<PG_LEVEL_2M)|(1<<PG_LEVEL_1G)));
1031                 pos = end_pfn << PAGE_SHIFT;
1032         }
1033
1034         /* tail is not big page (1G) alignment */
1035         start_pfn = ((pos + (PMD_SIZE - 1))>>PMD_SHIFT)
1036                          << (PMD_SHIFT - PAGE_SHIFT);
1037         end_pfn = (end >> PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
1038         if (start_pfn < end_pfn) {
1039                 nr_range = save_mr(mr, nr_range, start_pfn, end_pfn,
1040                                 page_size_mask & (1<<PG_LEVEL_2M));
1041                 pos = end_pfn << PAGE_SHIFT;
1042         }
1043 #endif
1044
1045         /* tail is not big page (2M) alignment */
1046         start_pfn = pos>>PAGE_SHIFT;
1047         end_pfn = end>>PAGE_SHIFT;
1048         nr_range = save_mr(mr, nr_range, start_pfn, end_pfn, 0);
1049
1050         /* try to merge same page size and continuous */
1051         for (i = 0; nr_range > 1 && i < nr_range - 1; i++) {
1052                 unsigned long old_start;
1053                 if (mr[i].end != mr[i+1].start ||
1054                     mr[i].page_size_mask != mr[i+1].page_size_mask)
1055                         continue;
1056                 /* move it */
1057                 old_start = mr[i].start;
1058                 memmove(&mr[i], &mr[i+1],
1059                         (nr_range - 1 - i) * sizeof(struct map_range));
1060                 mr[i--].start = old_start;
1061                 nr_range--;
1062         }
1063
1064         for (i = 0; i < nr_range; i++)
1065                 printk(KERN_DEBUG " %010lx - %010lx page %s\n",
1066                                 mr[i].start, mr[i].end,
1067                         (mr[i].page_size_mask & (1<<PG_LEVEL_1G))?"1G":(
1068                          (mr[i].page_size_mask & (1<<PG_LEVEL_2M))?"2M":"4k"));
1069
1070         /*
1071          * Find space for the kernel direct mapping tables.
1072          *
1073          * Later we should allocate these tables in the local node of the
1074          * memory mapped. Unfortunately this is done currently before the
1075          * nodes are discovered.
1076          */
1077         if (!after_bootmem)
1078                 find_early_table_space(end, use_pse, use_gbpages);
1079
1080 #ifdef CONFIG_X86_32
1081         for (i = 0; i < nr_range; i++)
1082                 kernel_physical_mapping_init(
1083                                 mr[i].start >> PAGE_SHIFT,
1084                                 mr[i].end >> PAGE_SHIFT,
1085                                 mr[i].page_size_mask == (1<<PG_LEVEL_2M));
1086         ret = end;
1087 #else /* CONFIG_X86_64 */
1088         for (i = 0; i < nr_range; i++)
1089                 ret = kernel_physical_mapping_init(mr[i].start, mr[i].end,
1090                                                    mr[i].page_size_mask);
1091 #endif
1092
1093 #ifdef CONFIG_X86_32
1094         early_ioremap_page_table_range_init();
1095
1096         load_cr3(swapper_pg_dir);
1097 #endif
1098
1099 #ifdef CONFIG_X86_64
1100         if (!after_bootmem)
1101                 mmu_cr4_features = read_cr4();
1102 #endif
1103         __flush_tlb_all();
1104
1105         if (!after_bootmem && table_end > table_start)
1106                 reserve_early(table_start << PAGE_SHIFT,
1107                                  table_end << PAGE_SHIFT, "PGTABLE");
1108
1109         if (!after_bootmem)
1110                 early_memtest(start, end);
1111
1112         return ret >> PAGE_SHIFT;
1113 }
1114
1115
1116 /*
1117  * paging_init() sets up the page tables - note that the first 8MB are
1118  * already mapped by head.S.
1119  *
1120  * This routines also unmaps the page at virtual kernel address 0, so
1121  * that we can trap those pesky NULL-reference errors in the kernel.
1122  */
1123 void __init paging_init(void)
1124 {
1125         pagetable_init();
1126
1127         __flush_tlb_all();
1128
1129         kmap_init();
1130
1131         /*
1132          * NOTE: at this point the bootmem allocator is fully available.
1133          */
1134         sparse_init();
1135         zone_sizes_init();
1136 }
1137
1138 /*
1139  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
1140  * and also on some strange 486's. All 586+'s are OK. This used to involve
1141  * black magic jumps to work around some nasty CPU bugs, but fortunately the
1142  * switch to using exceptions got rid of all that.
1143  */
1144 static void __init test_wp_bit(void)
1145 {
1146         printk(KERN_INFO
1147   "Checking if this processor honours the WP bit even in supervisor mode...");
1148
1149         /* Any page-aligned address will do, the test is non-destructive */
1150         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
1151         boot_cpu_data.wp_works_ok = do_test_wp_bit();
1152         clear_fixmap(FIX_WP_TEST);
1153
1154         if (!boot_cpu_data.wp_works_ok) {
1155                 printk(KERN_CONT "No.\n");
1156 #ifdef CONFIG_X86_WP_WORKS_OK
1157                 panic(
1158   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
1159 #endif
1160         } else {
1161                 printk(KERN_CONT "Ok.\n");
1162         }
1163 }
1164
1165 static struct kcore_list kcore_mem, kcore_vmalloc;
1166
1167 void __init mem_init(void)
1168 {
1169         int codesize, reservedpages, datasize, initsize;
1170         int tmp;
1171
1172         pci_iommu_alloc();
1173
1174 #ifdef CONFIG_FLATMEM
1175         BUG_ON(!mem_map);
1176 #endif
1177         /* this will put all low memory onto the freelists */
1178         totalram_pages += free_all_bootmem();
1179
1180         reservedpages = 0;
1181         for (tmp = 0; tmp < max_low_pfn; tmp++)
1182                 /*
1183                  * Only count reserved RAM pages:
1184                  */
1185                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
1186                         reservedpages++;
1187
1188         set_highmem_pages_init();
1189
1190         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
1191         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
1192         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
1193
1194         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
1195         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
1196                    VMALLOC_END-VMALLOC_START);
1197
1198         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
1199                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
1200                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
1201                 num_physpages << (PAGE_SHIFT-10),
1202                 codesize >> 10,
1203                 reservedpages << (PAGE_SHIFT-10),
1204                 datasize >> 10,
1205                 initsize >> 10,
1206                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
1207                );
1208
1209         printk(KERN_INFO "virtual kernel memory layout:\n"
1210                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1211 #ifdef CONFIG_HIGHMEM
1212                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1213 #endif
1214                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
1215                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
1216                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1217                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1218                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
1219                 FIXADDR_START, FIXADDR_TOP,
1220                 (FIXADDR_TOP - FIXADDR_START) >> 10,
1221
1222 #ifdef CONFIG_HIGHMEM
1223                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
1224                 (LAST_PKMAP*PAGE_SIZE) >> 10,
1225 #endif
1226
1227                 VMALLOC_START, VMALLOC_END,
1228                 (VMALLOC_END - VMALLOC_START) >> 20,
1229
1230                 (unsigned long)__va(0), (unsigned long)high_memory,
1231                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
1232
1233                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
1234                 ((unsigned long)&__init_end -
1235                  (unsigned long)&__init_begin) >> 10,
1236
1237                 (unsigned long)&_etext, (unsigned long)&_edata,
1238                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
1239
1240                 (unsigned long)&_text, (unsigned long)&_etext,
1241                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
1242
1243         /*
1244          * Check boundaries twice: Some fundamental inconsistencies can
1245          * be detected at build time already.
1246          */
1247 #define __FIXADDR_TOP (-PAGE_SIZE)
1248 #ifdef CONFIG_HIGHMEM
1249         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
1250         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
1251 #endif
1252 #define high_memory (-128UL << 20)
1253         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
1254 #undef high_memory
1255 #undef __FIXADDR_TOP
1256
1257 #ifdef CONFIG_HIGHMEM
1258         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
1259         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
1260 #endif
1261         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
1262         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
1263
1264         if (boot_cpu_data.wp_works_ok < 0)
1265                 test_wp_bit();
1266
1267         save_pg_dir();
1268         zap_low_mappings();
1269 }
1270
1271 #ifdef CONFIG_MEMORY_HOTPLUG
1272 int arch_add_memory(int nid, u64 start, u64 size)
1273 {
1274         struct pglist_data *pgdata = NODE_DATA(nid);
1275         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1276         unsigned long start_pfn = start >> PAGE_SHIFT;
1277         unsigned long nr_pages = size >> PAGE_SHIFT;
1278
1279         return __add_pages(nid, zone, start_pfn, nr_pages);
1280 }
1281 #endif
1282
1283 /*
1284  * This function cannot be __init, since exceptions don't work in that
1285  * section.  Put this after the callers, so that it cannot be inlined.
1286  */
1287 static noinline int do_test_wp_bit(void)
1288 {
1289         char tmp_reg;
1290         int flag;
1291
1292         __asm__ __volatile__(
1293                 "       movb %0, %1     \n"
1294                 "1:     movb %1, %0     \n"
1295                 "       xorl %2, %2     \n"
1296                 "2:                     \n"
1297                 _ASM_EXTABLE(1b,2b)
1298                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1299                  "=q" (tmp_reg),
1300                  "=r" (flag)
1301                 :"2" (1)
1302                 :"memory");
1303
1304         return flag;
1305 }
1306
1307 #ifdef CONFIG_DEBUG_RODATA
1308 const int rodata_test_data = 0xC3;
1309 EXPORT_SYMBOL_GPL(rodata_test_data);
1310
1311 void mark_rodata_ro(void)
1312 {
1313         unsigned long start = PFN_ALIGN(_text);
1314         unsigned long size = PFN_ALIGN(_etext) - start;
1315
1316 #ifndef CONFIG_DYNAMIC_FTRACE
1317         /* Dynamic tracing modifies the kernel text section */
1318         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1319         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1320                 size >> 10);
1321
1322 #ifdef CONFIG_CPA_DEBUG
1323         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1324                 start, start+size);
1325         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1326
1327         printk(KERN_INFO "Testing CPA: write protecting again\n");
1328         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1329 #endif
1330 #endif /* CONFIG_DYNAMIC_FTRACE */
1331
1332         start += size;
1333         size = (unsigned long)__end_rodata - start;
1334         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1335         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1336                 size >> 10);
1337         rodata_test();
1338
1339 #ifdef CONFIG_CPA_DEBUG
1340         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1341         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1342
1343         printk(KERN_INFO "Testing CPA: write protecting again\n");
1344         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1345 #endif
1346 }
1347 #endif
1348
1349 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1350                                    int flags)
1351 {
1352         return reserve_bootmem(phys, len, flags);
1353 }