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