arch/i386/mm/init.c

   1 /*
   2  *  linux/arch/i386/mm/init.c
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *
   6  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   7  */
   8
   9 #include <linux/module.h>
  10 #include <linux/signal.h>
  11 #include <linux/sched.h>
  12 #include <linux/kernel.h>
  13 #include <linux/errno.h>
  14 #include <linux/string.h>
  15 #include <linux/types.h>
  16 #include <linux/ptrace.h>
  17 #include <linux/mman.h>
  18 #include <linux/mm.h>
  19 #include <linux/hugetlb.h>
  20 #include <linux/swap.h>
  21 #include <linux/smp.h>
  22 #include <linux/init.h>
  23 #include <linux/highmem.h>
  24 #include <linux/pagemap.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/efi.h>
  30 #include <linux/memory_hotplug.h>
  31 #include <linux/initrd.h>
  32 #include <linux/cpumask.h>
  33
  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/tlb.h>
  43 #include <asm/tlbflush.h>
  44 #include <asm/sections.h>
  45
  46 unsigned int __VMALLOC_RESERVE = 128 << 20;
  47
  48 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  49 unsigned long highstart_pfn, highend_pfn;
  50
  51 static int noinline do_test_wp_bit(void);
  52
  53 /*
  54  * Creates a middle page table and puts a pointer to it in the
  55  * given global directory entry. This only returns the gd entry
  56  * in non-PAE compilation mode, since the middle layer is folded.
  57  */
  58 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  59 {
  60         pud_t *pud;
  61         pmd_t *pmd_table;
  62
  63 #ifdef CONFIG_X86_PAE
  64         pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  65         set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  66         pud = pud_offset(pgd, 0);
  67         if (pmd_table != pmd_offset(pud, 0))
  68                 BUG();
  69 #else
  70         pud = pud_offset(pgd, 0);
  71         pmd_table = pmd_offset(pud, 0);
  72 #endif
  73
  74         return pmd_table;
  75 }
  76
  77 /*
  78  * Create a page table and place a pointer to it in a middle page
  79  * directory entry.
  80  */
  81 static pte_t * __init one_page_table_init(pmd_t *pmd)
  82 {
  83         if (pmd_none(*pmd)) {
  84                 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  85                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
  86                 if (page_table != pte_offset_kernel(pmd, 0))
  87                         BUG();
  88
  89                 return page_table;
  90         }
  91
  92         return pte_offset_kernel(pmd, 0);
  93 }
  94
  95 /*
  96  * This function initializes a certain range of kernel virtual memory
  97  * with new bootmem page tables, everywhere page tables are missing in
  98  * the given range.
  99  */
 100
 101 /*
 102  * NOTE: The pagetables are allocated contiguous on the physical space
 103  * so we can cache the place of the first one and move around without
 104  * checking the pgd every time.
 105  */
 106 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
 107 {
 108         pgd_t *pgd;
 109         pud_t *pud;
 110         pmd_t *pmd;
 111         int pgd_idx, pmd_idx;
 112         unsigned long vaddr;
 113
 114         vaddr = start;
 115         pgd_idx = pgd_index(vaddr);
 116         pmd_idx = pmd_index(vaddr);
 117         pgd = pgd_base + pgd_idx;
 118
 119         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 120                 if (pgd_none(*pgd))
 121                         one_md_table_init(pgd);
 122                 pud = pud_offset(pgd, vaddr);
 123                 pmd = pmd_offset(pud, vaddr);
 124                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
 125                         if (pmd_none(*pmd))
 126                                 one_page_table_init(pmd);
 127
 128                         vaddr += PMD_SIZE;
 129                 }
 130                 pmd_idx = 0;
 131         }
 132 }
 133
 134 static inline int is_kernel_text(unsigned long addr)
 135 {
 136         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 137                 return 1;
 138         return 0;
 139 }
 140
 141 /*
 142  * This maps the physical memory to kernel virtual address space, a total
 143  * of max_low_pfn pages, by creating page tables starting from address
 144  * PAGE_OFFSET.
 145  */
 146 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
 147 {
 148         unsigned long pfn;
 149         pgd_t *pgd;
 150         pmd_t *pmd;
 151         pte_t *pte;
 152         int pgd_idx, pmd_idx, pte_ofs;
 153
 154         pgd_idx = pgd_index(PAGE_OFFSET);
 155         pgd = pgd_base + pgd_idx;
 156         pfn = 0;
 157
 158         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 159                 pmd = one_md_table_init(pgd);
 160                 if (pfn >= max_low_pfn)
 161                         continue;
 162                 for (pmd_idx = 0; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
 163                         unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
 164
 165                         /* Map with big pages if possible, otherwise create normal page tables. */
 166                         if (cpu_has_pse) {
 167                                 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
 168
 169                                 if (is_kernel_text(address) || is_kernel_text(address2))
 170                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
 171                                 else
 172                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
 173                                 pfn += PTRS_PER_PTE;
 174                         } else {
 175                                 pte = one_page_table_init(pmd);
 176
 177                                 for (pte_ofs = 0; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
 178                                                 if (is_kernel_text(address))
 179                                                         set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
 180                                                 else
 181                                                         set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
 182                                 }
 183                         }
 184                 }
 185         }
 186 }
 187
 188 static inline int page_kills_ppro(unsigned long pagenr)
 189 {
 190         if (pagenr >= 0x70000 && pagenr <= 0x7003F)
 191                 return 1;
 192         return 0;
 193 }
 194
 195 extern int is_available_memory(efi_memory_desc_t *);
 196
 197 int page_is_ram(unsigned long pagenr)
 198 {
 199         int i;
 200         unsigned long addr, end;
 201
 202         if (efi_enabled) {
 203                 efi_memory_desc_t *md;
 204                 void *p;
 205
 206                 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 207                         md = p;
 208                         if (!is_available_memory(md))
 209                                 continue;
 210                         addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 211                         end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
 212
 213                         if ((pagenr >= addr) && (pagenr < end))
 214                                 return 1;
 215                 }
 216                 return 0;
 217         }
 218
 219         for (i = 0; i < e820.nr_map; i++) {
 220
 221                 if (e820.map[i].type != E820_RAM)       /* not usable memory */
 222                         continue;
 223                 /*
 224                  *      !!!FIXME!!! Some BIOSen report areas as RAM that
 225                  *      are not. Notably the 640->1Mb area. We need a sanity
 226                  *      check here.
 227                  */
 228                 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 229                 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
 230                 if  ((pagenr >= addr) && (pagenr < end))
 231                         return 1;
 232         }
 233         return 0;
 234 }
 235
 236 #ifdef CONFIG_HIGHMEM
 237 pte_t *kmap_pte;
 238 pgprot_t kmap_prot;
 239
 240 #define kmap_get_fixmap_pte(vaddr)                                      \
 241         pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
 242
 243 static void __init kmap_init(void)
 244 {
 245         unsigned long kmap_vstart;
 246
 247         /* cache the first kmap pte */
 248         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 249         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 250
 251         kmap_prot = PAGE_KERNEL;
 252 }
 253
 254 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 255 {
 256         pgd_t *pgd;
 257         pud_t *pud;
 258         pmd_t *pmd;
 259         pte_t *pte;
 260         unsigned long vaddr;
 261
 262         vaddr = PKMAP_BASE;
 263         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 264
 265         pgd = swapper_pg_dir + pgd_index(vaddr);
 266         pud = pud_offset(pgd, vaddr);
 267         pmd = pmd_offset(pud, vaddr);
 268         pte = pte_offset_kernel(pmd, vaddr);
 269         pkmap_page_table = pte;
 270 }
 271
 272 static void __meminit free_new_highpage(struct page *page)
 273 {
 274         init_page_count(page);
 275         __free_page(page);
 276         totalhigh_pages++;
 277 }
 278
 279 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
 280 {
 281         if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
 282                 ClearPageReserved(page);
 283                 free_new_highpage(page);
 284         } else
 285                 SetPageReserved(page);
 286 }
 287
 288 static int add_one_highpage_hotplug(struct page *page, unsigned long pfn)
 289 {
 290         free_new_highpage(page);
 291         totalram_pages++;
 292 #ifdef CONFIG_FLATMEM
 293         max_mapnr = max(pfn, max_mapnr);
 294 #endif
 295         num_physpages++;
 296         return 0;
 297 }
 298
 299 /*
 300  * Not currently handling the NUMA case.
 301  * Assuming single node and all memory that
 302  * has been added dynamically that would be
 303  * onlined here is in HIGHMEM
 304  */
 305 void online_page(struct page *page)
 306 {
 307         ClearPageReserved(page);
 308         add_one_highpage_hotplug(page, page_to_pfn(page));
 309 }
 310
 311
 312 #ifdef CONFIG_NUMA
 313 extern void set_highmem_pages_init(int);
 314 #else
 315 static void __init set_highmem_pages_init(int bad_ppro)
 316 {
 317         int pfn;
 318         for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
 319                 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
 320         totalram_pages += totalhigh_pages;
 321 }
 322 #endif /* CONFIG_FLATMEM */
 323
 324 #else
 325 #define kmap_init() do { } while (0)
 326 #define permanent_kmaps_init(pgd_base) do { } while (0)
 327 #define set_highmem_pages_init(bad_ppro) do { } while (0)
 328 #endif /* CONFIG_HIGHMEM */
 329
 330 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
 331 EXPORT_SYMBOL(__PAGE_KERNEL);
 332 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
 333
 334 #ifdef CONFIG_NUMA
 335 extern void __init remap_numa_kva(void);
 336 #else
 337 #define remap_numa_kva() do {} while (0)
 338 #endif
 339
 340 static void __init pagetable_init (void)
 341 {
 342         unsigned long vaddr;
 343         pgd_t *pgd_base = swapper_pg_dir;
 344
 345 #ifdef CONFIG_X86_PAE
 346         int i;
 347         /* Init entries of the first-level page table to the zero page */
 348         for (i = 0; i < PTRS_PER_PGD; i++)
 349                 set_pgd(pgd_base + i, __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
 350 #endif
 351
 352         /* Enable PSE if available */
 353         if (cpu_has_pse) {
 354                 set_in_cr4(X86_CR4_PSE);
 355         }
 356
 357         /* Enable PGE if available */
 358         if (cpu_has_pge) {
 359                 set_in_cr4(X86_CR4_PGE);
 360                 __PAGE_KERNEL |= _PAGE_GLOBAL;
 361                 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
 362         }
 363
 364         kernel_physical_mapping_init(pgd_base);
 365         remap_numa_kva();
 366
 367         /*
 368          * Fixed mappings, only the page table structure has to be
 369          * created - mappings will be set by set_fixmap():
 370          */
 371         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 372         page_table_range_init(vaddr, 0, pgd_base);
 373
 374         permanent_kmaps_init(pgd_base);
 375
 376 #ifdef CONFIG_X86_PAE
 377         /*
 378          * Add low memory identity-mappings - SMP needs it when
 379          * starting up on an AP from real-mode. In the non-PAE
 380          * case we already have these mappings through head.S.
 381          * All user-space mappings are explicitly cleared after
 382          * SMP startup.
 383          */
 384         set_pgd(&pgd_base[0], pgd_base[USER_PTRS_PER_PGD]);
 385 #endif
 386 }
 387
 388 #if defined(CONFIG_SOFTWARE_SUSPEND) || defined(CONFIG_ACPI_SLEEP)
 389 /*
 390  * Swap suspend & friends need this for resume because things like the intel-agp
 391  * driver might have split up a kernel 4MB mapping.
 392  */
 393 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
 394         __attribute__ ((aligned (PAGE_SIZE)));
 395
 396 static inline void save_pg_dir(void)
 397 {
 398         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 399 }
 400 #else
 401 static inline void save_pg_dir(void)
 402 {
 403 }
 404 #endif
 405
 406 void zap_low_mappings (void)
 407 {
 408         int i;
 409
 410         save_pg_dir();
 411
 412         /*
 413          * Zap initial low-memory mappings.
 414          *
 415          * Note that "pgd_clear()" doesn't do it for
 416          * us, because pgd_clear() is a no-op on i386.
 417          */
 418         for (i = 0; i < USER_PTRS_PER_PGD; i++)
 419 #ifdef CONFIG_X86_PAE
 420                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 421 #else
 422                 set_pgd(swapper_pg_dir+i, __pgd(0));
 423 #endif
 424         flush_tlb_all();
 425 }
 426
 427 static int disable_nx __initdata = 0;
 428 u64 __supported_pte_mask __read_mostly = ~_PAGE_NX;
 429
 430 /*
 431  * noexec = on|off
 432  *
 433  * Control non executable mappings.
 434  *
 435  * on      Enable
 436  * off     Disable
 437  */
 438 static int __init noexec_setup(char *str)
 439 {
 440         if (!str || !strcmp(str, "on")) {
 441                 if (cpu_has_nx) {
 442                         __supported_pte_mask |= _PAGE_NX;
 443                         disable_nx = 0;
 444                 }
 445         } else if (!strcmp(str,"off")) {
 446                 disable_nx = 1;
 447                 __supported_pte_mask &= ~_PAGE_NX;
 448         } else
 449                 return -EINVAL;
 450
 451         return 0;
 452 }
 453 early_param("noexec", noexec_setup);
 454
 455 int nx_enabled = 0;
 456 #ifdef CONFIG_X86_PAE
 457
 458 static void __init set_nx(void)
 459 {
 460         unsigned int v[4], l, h;
 461
 462         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 463                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 464                 if ((v[3] & (1 << 20)) && !disable_nx) {
 465                         rdmsr(MSR_EFER, l, h);
 466                         l |= EFER_NX;
 467                         wrmsr(MSR_EFER, l, h);
 468                         nx_enabled = 1;
 469                         __supported_pte_mask |= _PAGE_NX;
 470                 }
 471         }
 472 }
 473
 474 /*
 475  * Enables/disables executability of a given kernel page and
 476  * returns the previous setting.
 477  */
 478 int __init set_kernel_exec(unsigned long vaddr, int enable)
 479 {
 480         pte_t *pte;
 481         int ret = 1;
 482
 483         if (!nx_enabled)
 484                 goto out;
 485
 486         pte = lookup_address(vaddr);
 487         BUG_ON(!pte);
 488
 489         if (!pte_exec_kernel(*pte))
 490                 ret = 0;
 491
 492         if (enable)
 493                 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
 494         else
 495                 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
 496         pte_update_defer(&init_mm, vaddr, pte);
 497         __flush_tlb_all();
 498 out:
 499         return ret;
 500 }
 501
 502 #endif
 503
 504 /*
 505  * paging_init() sets up the page tables - note that the first 8MB are
 506  * already mapped by head.S.
 507  *
 508  * This routines also unmaps the page at virtual kernel address 0, so
 509  * that we can trap those pesky NULL-reference errors in the kernel.
 510  */
 511 void __init paging_init(void)
 512 {
 513 #ifdef CONFIG_X86_PAE
 514         set_nx();
 515         if (nx_enabled)
 516                 printk("NX (Execute Disable) protection: active\n");
 517 #endif
 518
 519         pagetable_init();
 520
 521         load_cr3(swapper_pg_dir);
 522
 523 #ifdef CONFIG_X86_PAE
 524         /*
 525          * We will bail out later - printk doesn't work right now so
 526          * the user would just see a hanging kernel.
 527          */
 528         if (cpu_has_pae)
 529                 set_in_cr4(X86_CR4_PAE);
 530 #endif
 531         __flush_tlb_all();
 532
 533         kmap_init();
 534 }
 535
 536 /*
 537  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 538  * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
 539  * used to involve black magic jumps to work around some nasty CPU bugs,
 540  * but fortunately the switch to using exceptions got rid of all that.
 541  */
 542
 543 static void __init test_wp_bit(void)
 544 {
 545         printk("Checking if this processor honours the WP bit even in supervisor mode... ");
 546
 547         /* Any page-aligned address will do, the test is non-destructive */
 548         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 549         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 550         clear_fixmap(FIX_WP_TEST);
 551
 552         if (!boot_cpu_data.wp_works_ok) {
 553                 printk("No.\n");
 554 #ifdef CONFIG_X86_WP_WORKS_OK
 555                 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 556 #endif
 557         } else {
 558                 printk("Ok.\n");
 559         }
 560 }
 561
 562 static struct kcore_list kcore_mem, kcore_vmalloc;
 563
 564 void __init mem_init(void)
 565 {
 566         extern int ppro_with_ram_bug(void);
 567         int codesize, reservedpages, datasize, initsize;
 568         int tmp;
 569         int bad_ppro;
 570
 571 #ifdef CONFIG_FLATMEM
 572         if (!mem_map)
 573                 BUG();
 574 #endif
 575
 576         bad_ppro = ppro_with_ram_bug();
 577
 578 #ifdef CONFIG_HIGHMEM
 579         /* check that fixmap and pkmap do not overlap */
 580         if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
 581                 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
 582                 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
 583                                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
 584                 BUG();
 585         }
 586 #endif
 587
 588         /* this will put all low memory onto the freelists */
 589         totalram_pages += free_all_bootmem();
 590
 591         reservedpages = 0;
 592         for (tmp = 0; tmp < max_low_pfn; tmp++)
 593                 /*
 594                  * Only count reserved RAM pages
 595                  */
 596                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 597                         reservedpages++;
 598
 599         set_highmem_pages_init(bad_ppro);
 600
 601         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 602         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 603         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 604
 605         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 606         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 607                    VMALLOC_END-VMALLOC_START);
 608
 609         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
 610                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 611                 num_physpages << (PAGE_SHIFT-10),
 612                 codesize >> 10,
 613                 reservedpages << (PAGE_SHIFT-10),
 614                 datasize >> 10,
 615                 initsize >> 10,
 616                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 617                );
 618
 619 #if 1 /* double-sanity-check paranoia */
 620         printk("virtual kernel memory layout:\n"
 621                "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 622 #ifdef CONFIG_HIGHMEM
 623                "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 624 #endif
 625                "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 626                "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 627                "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 628                "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 629                "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 630                FIXADDR_START, FIXADDR_TOP,
 631                (FIXADDR_TOP - FIXADDR_START) >> 10,
 632
 633 #ifdef CONFIG_HIGHMEM
 634                PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 635                (LAST_PKMAP*PAGE_SIZE) >> 10,
 636 #endif
 637
 638                VMALLOC_START, VMALLOC_END,
 639                (VMALLOC_END - VMALLOC_START) >> 20,
 640
 641                (unsigned long)__va(0), (unsigned long)high_memory,
 642                ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 643
 644                (unsigned long)&__init_begin, (unsigned long)&__init_end,
 645                ((unsigned long)&__init_end - (unsigned long)&__init_begin) >> 10,
 646
 647                (unsigned long)&_etext, (unsigned long)&_edata,
 648                ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 649
 650                (unsigned long)&_text, (unsigned long)&_etext,
 651                ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 652
 653 #ifdef CONFIG_HIGHMEM
 654         BUG_ON(PKMAP_BASE+LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
 655         BUG_ON(VMALLOC_END                     > PKMAP_BASE);
 656 #endif
 657         BUG_ON(VMALLOC_START                   > VMALLOC_END);
 658         BUG_ON((unsigned long)high_memory      > VMALLOC_START);
 659 #endif /* double-sanity-check paranoia */
 660
 661 #ifdef CONFIG_X86_PAE
 662         if (!cpu_has_pae)
 663                 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
 664 #endif
 665         if (boot_cpu_data.wp_works_ok < 0)
 666                 test_wp_bit();
 667
 668         /*
 669          * Subtle. SMP is doing it's boot stuff late (because it has to
 670          * fork idle threads) - but it also needs low mappings for the
 671          * protected-mode entry to work. We zap these entries only after
 672          * the WP-bit has been tested.
 673          */
 674 #ifndef CONFIG_SMP
 675         zap_low_mappings();
 676 #endif
 677 }
 678
 679 /*
 680  * this is for the non-NUMA, single node SMP system case.
 681  * Specifically, in the case of x86, we will always add
 682  * memory to the highmem for now.
 683  */
 684 #ifdef CONFIG_MEMORY_HOTPLUG
 685 #ifndef CONFIG_NEED_MULTIPLE_NODES
 686 int arch_add_memory(int nid, u64 start, u64 size)
 687 {
 688         struct pglist_data *pgdata = &contig_page_data;
 689         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 690         unsigned long start_pfn = start >> PAGE_SHIFT;
 691         unsigned long nr_pages = size >> PAGE_SHIFT;
 692
 693         return __add_pages(zone, start_pfn, nr_pages);
 694 }
 695
 696 int remove_memory(u64 start, u64 size)
 697 {
 698         return -EINVAL;
 699 }
 700 #endif
 701 #endif
 702
 703 kmem_cache_t *pgd_cache;
 704 kmem_cache_t *pmd_cache;
 705
 706 void __init pgtable_cache_init(void)
 707 {
 708         if (PTRS_PER_PMD > 1) {
 709                 pmd_cache = kmem_cache_create("pmd",
 710                                         PTRS_PER_PMD*sizeof(pmd_t),
 711                                         PTRS_PER_PMD*sizeof(pmd_t),
 712                                         0,
 713                                         pmd_ctor,
 714                                         NULL);
 715                 if (!pmd_cache)
 716                         panic("pgtable_cache_init(): cannot create pmd cache");
 717         }
 718         pgd_cache = kmem_cache_create("pgd",
 719                                 PTRS_PER_PGD*sizeof(pgd_t),
 720                                 PTRS_PER_PGD*sizeof(pgd_t),
 721                                 0,
 722                                 pgd_ctor,
 723                                 PTRS_PER_PMD == 1 ? pgd_dtor : NULL);
 724         if (!pgd_cache)
 725                 panic("pgtable_cache_init(): Cannot create pgd cache");
 726 }
 727
 728 /*
 729  * This function cannot be __init, since exceptions don't work in that
 730  * section.  Put this after the callers, so that it cannot be inlined.
 731  */
 732 static int noinline do_test_wp_bit(void)
 733 {
 734         char tmp_reg;
 735         int flag;
 736
 737         __asm__ __volatile__(
 738                 "       movb %0,%1      \n"
 739                 "1:     movb %1,%0      \n"
 740                 "       xorl %2,%2      \n"
 741                 "2:                     \n"
 742                 ".section __ex_table,\"a\"\n"
 743                 "       .align 4        \n"
 744                 "       .long 1b,2b     \n"
 745                 ".previous              \n"
 746                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 747                  "=q" (tmp_reg),
 748                  "=r" (flag)
 749                 :"2" (1)
 750                 :"memory");
 751
 752         return flag;
 753 }
 754
 755 #ifdef CONFIG_DEBUG_RODATA
 756
 757 void mark_rodata_ro(void)
 758 {
 759         unsigned long addr = (unsigned long)__start_rodata;
 760
 761         for (; addr < (unsigned long)__end_rodata; addr += PAGE_SIZE)
 762                 change_page_attr(virt_to_page(addr), 1, PAGE_KERNEL_RO);
 763
 764         printk("Write protecting the kernel read-only data: %uk\n",
 765                         (__end_rodata - __start_rodata) >> 10);
 766
 767         /*
 768          * change_page_attr() requires a global_flush_tlb() call after it.
 769          * We do this after the printk so that if something went wrong in the
 770          * change, the printk gets out at least to give a better debug hint
 771          * of who is the culprit.
 772          */
 773         global_flush_tlb();
 774 }
 775 #endif
 776
 777 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 778 {
 779         unsigned long addr;
 780
 781         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 782                 ClearPageReserved(virt_to_page(addr));
 783                 init_page_count(virt_to_page(addr));
 784                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
 785                 free_page(addr);
 786                 totalram_pages++;
 787         }
 788         printk(KERN_INFO "Freeing %s: %ldk freed\n", what, (end - begin) >> 10);
 789 }
 790
 791 void free_initmem(void)
 792 {
 793         free_init_pages("unused kernel memory",
 794                         (unsigned long)(&__init_begin),
 795                         (unsigned long)(&__init_end));
 796 }
 797
 798 #ifdef CONFIG_BLK_DEV_INITRD
 799 void free_initrd_mem(unsigned long start, unsigned long end)
 800 {
 801         free_init_pages("initrd memory", start, end);
 802 }
 803 #endif
 804