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/config.h>
  10 #include <linux/module.h>
  11 #include <linux/signal.h>
  12 #include <linux/sched.h>
  13 #include <linux/kernel.h>
  14 #include <linux/errno.h>
  15 #include <linux/string.h>
  16 #include <linux/types.h>
  17 #include <linux/ptrace.h>
  18 #include <linux/mman.h>
  19 #include <linux/mm.h>
  20 #include <linux/hugetlb.h>
  21 #include <linux/swap.h>
  22 #include <linux/smp.h>
  23 #include <linux/init.h>
  24 #include <linux/highmem.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/efi.h>
  30
  31 #include <asm/processor.h>
  32 #include <asm/system.h>
  33 #include <asm/uaccess.h>
  34 #include <asm/pgtable.h>
  35 #include <asm/dma.h>
  36 #include <asm/fixmap.h>
  37 #include <asm/e820.h>
  38 #include <asm/apic.h>
  39 #include <asm/tlb.h>
  40 #include <asm/tlbflush.h>
  41 #include <asm/sections.h>
  42 #include <asm/desc.h>
  43
  44 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  45 unsigned long highstart_pfn, highend_pfn;
  46
  47 static int do_test_wp_bit(void);
  48
  49 static inline int page_kills_ppro(unsigned long pagenr)
  50 {
  51         if (pagenr >= 0x70000 && pagenr <= 0x7003F)
  52                 return 1;
  53         return 0;
  54 }
  55
  56 extern int is_available_memory(efi_memory_desc_t *);
  57
  58 static inline int page_is_ram(unsigned long pagenr)
  59 {
  60         int i;
  61         unsigned long addr, end;
  62
  63         if (efi_enabled) {
  64                 efi_memory_desc_t *md;
  65
  66                 for (i = 0; i < memmap.nr_map; i++) {
  67                         md = &memmap.map[i];
  68                         if (!is_available_memory(md))
  69                                 continue;
  70                         addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
  71                         end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
  72
  73                         if ((pagenr >= addr) && (pagenr < end))
  74                                 return 1;
  75                 }
  76                 return 0;
  77         }
  78
  79         for (i = 0; i < e820.nr_map; i++) {
  80
  81                 if (e820.map[i].type != E820_RAM)       /* not usable memory */
  82                         continue;
  83                 /*
  84                  *      !!!FIXME!!! Some BIOSen report areas as RAM that
  85                  *      are not. Notably the 640->1Mb area. We need a sanity
  86                  *      check here.
  87                  */
  88                 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
  89                 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
  90                 if  ((pagenr >= addr) && (pagenr < end))
  91                         return 1;
  92         }
  93         return 0;
  94 }
  95
  96 /* To enable modules to check if a page is in RAM */
  97 int pfn_is_ram(unsigned long pfn)
  98 {
  99         return (page_is_ram(pfn));
 100 }
 101
 102
 103 /*
 104  * devmem_is_allowed() checks to see if /dev/mem access to a certain address is
 105  * valid. The argument is a physical page number.
 106  *
 107  *
 108  * On x86, access has to be given to the first megabyte of ram because that area
 109  * contains bios code and data regions used by X and dosemu and similar apps.
 110  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 111  * mmio resources as well as potential bios/acpi data regions.
 112  */
 113 int devmem_is_allowed(unsigned long pagenr)
 114 {
 115         if (pagenr <= 256)
 116                 return 1;
 117         if (!page_is_ram(pagenr))
 118                 return 1;
 119         return 0;
 120 }
 121
 122
 123 pte_t *kmap_pte;
 124
 125 EXPORT_SYMBOL(kmap_pte);
 126
 127 #define kmap_get_fixmap_pte(vaddr)                                      \
 128         pte_offset_kernel(pmd_offset(pgd_offset_k(vaddr), (vaddr)), (vaddr))
 129
 130 void __init kmap_init(void)
 131 {
 132         kmap_pte = kmap_get_fixmap_pte(__fix_to_virt(FIX_KMAP_BEGIN));
 133 }
 134
 135 void __init one_highpage_init(struct page *page, int pfn, int bad_ppro)
 136 {
 137         if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
 138                 ClearPageReserved(page);
 139                 set_bit(PG_highmem, &page->flags);
 140                 set_page_count(page, 1);
 141                 __free_page(page);
 142                 totalhigh_pages++;
 143         } else
 144                 SetPageReserved(page);
 145 }
 146
 147 EXPORT_SYMBOL_GPL(page_is_ram);
 148
 149 #ifdef CONFIG_HIGHMEM
 150
 151 #ifndef CONFIG_DISCONTIGMEM
 152 void __init set_highmem_pages_init(int bad_ppro)
 153 {
 154         int pfn;
 155         for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
 156                 one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
 157         totalram_pages += totalhigh_pages;
 158 }
 159 #else
 160 extern void set_highmem_pages_init(int);
 161 #endif /* !CONFIG_DISCONTIGMEM */
 162 #else
 163 # define set_highmem_pages_init(bad_ppro) do { } while (0)
 164 #endif
 165
 166 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
 167 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
 168
 169 #ifndef CONFIG_DISCONTIGMEM
 170 #define remap_numa_kva() do {} while (0)
 171 #else
 172 extern void __init remap_numa_kva(void);
 173 #endif
 174
 175 static __init void prepare_pagetables(pgd_t *pgd_base, unsigned long address)
 176 {
 177         pgd_t *pgd;
 178         pmd_t *pmd;
 179         pte_t *pte;
 180
 181         pgd = pgd_base + pgd_index(address);
 182         pmd = pmd_offset(pgd, address);
 183         if (!pmd_present(*pmd)) {
 184                 pte = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
 185                 set_pmd(pmd, __pmd(_PAGE_TABLE + __pa(pte)));
 186         }
 187 }
 188
 189 static void __init fixrange_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
 190 {
 191         unsigned long vaddr;
 192
 193         for (vaddr = start; vaddr != end; vaddr += PAGE_SIZE)
 194                 prepare_pagetables(pgd_base, vaddr);
 195 }
 196
 197 void setup_identity_mappings(pgd_t *pgd_base, unsigned long start, unsigned long end)
 198 {
 199         unsigned long vaddr;
 200         pgd_t *pgd;
 201         int i, j, k;
 202         pmd_t *pmd;
 203         pte_t *pte, *pte_base;
 204
 205         pgd = pgd_base;
 206
 207         for (i = 0; i < PTRS_PER_PGD; pgd++, i++) {
 208                 vaddr = i*PGDIR_SIZE;
 209                 if (end && (vaddr >= end))
 210                         break;
 211                 pmd = pmd_offset(pgd, 0);
 212                 for (j = 0; j < PTRS_PER_PMD; pmd++, j++) {
 213                         vaddr = i*PGDIR_SIZE + j*PMD_SIZE;
 214                         if (end && (vaddr >= end))
 215                                 break;
 216                         if (vaddr < start)
 217                                 continue;
 218                         if (cpu_has_pse) {
 219                                 unsigned long __pe;
 220
 221                                 set_in_cr4(X86_CR4_PSE);
 222                                 boot_cpu_data.wp_works_ok = 1;
 223                                 __pe = _KERNPG_TABLE + _PAGE_PSE + vaddr - start;
 224                                 /* Make it "global" too if supported */
 225                                 if (cpu_has_pge) {
 226                                         set_in_cr4(X86_CR4_PGE);
 227 #if !defined(CONFIG_X86_SWITCH_PAGETABLES)
 228                                         __pe += _PAGE_GLOBAL;
 229                                         __PAGE_KERNEL |= _PAGE_GLOBAL;
 230 #endif
 231                                 }
 232                                 set_pmd(pmd, __pmd(__pe));
 233                                 continue;
 234                         }
 235                         if (!pmd_present(*pmd))
 236                                 pte_base = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
 237                         else
 238                                 pte_base = pte_offset_kernel(pmd, 0);
 239                         pte = pte_base;
 240                         for (k = 0; k < PTRS_PER_PTE; pte++, k++) {
 241                                 vaddr = i*PGDIR_SIZE + j*PMD_SIZE + k*PAGE_SIZE;
 242                                 if (end && (vaddr >= end))
 243                                         break;
 244                                 if (vaddr < start)
 245                                         continue;
 246                                 *pte = mk_pte_phys(vaddr-start, PAGE_KERNEL);
 247                         }
 248                         set_pmd(pmd, __pmd(_KERNPG_TABLE + __pa(pte_base)));
 249                 }
 250         }
 251 }
 252
 253 static void __init pagetable_init (void)
 254 {
 255         unsigned long vaddr, end;
 256         pgd_t *pgd_base;
 257 #ifdef CONFIG_X86_PAE
 258         int i;
 259 #endif
 260
 261         /*
 262          * This can be zero as well - no problem, in that case we exit
 263          * the loops anyway due to the PTRS_PER_* conditions.
 264          */
 265         end = (unsigned long)__va(max_low_pfn*PAGE_SIZE);
 266
 267         pgd_base = swapper_pg_dir;
 268 #ifdef CONFIG_X86_PAE
 269         /*
 270          * It causes too many problems if there's no proper pmd set up
 271          * for all 4 entries of the PGD - so we allocate all of them.
 272          * PAE systems will not miss this extra 4-8K anyway ...
 273          */
 274         for (i = 0; i < PTRS_PER_PGD; i++) {
 275                 pmd_t *pmd = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
 276                 set_pgd(pgd_base + i, __pgd(__pa(pmd) + 0x1));
 277         }
 278 #endif
 279         /*
 280          * Set up lowmem-sized identity mappings at PAGE_OFFSET:
 281          */
 282         setup_identity_mappings(pgd_base, PAGE_OFFSET, end);
 283
 284         /*
 285          * Add flat-mode identity-mappings - SMP needs it when
 286          * starting up on an AP from real-mode. (In the non-PAE
 287          * case we already have these mappings through head.S.)
 288          * All user-space mappings are explicitly cleared after
 289          * SMP startup.
 290          */
 291 #if defined(CONFIG_SMP) && defined(CONFIG_X86_PAE)
 292         setup_identity_mappings(pgd_base, 0, 16*1024*1024);
 293 #endif
 294         remap_numa_kva();
 295
 296         /*
 297          * Fixed mappings, only the page table structure has to be
 298          * created - mappings will be set by set_fixmap():
 299          */
 300         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 301         fixrange_init(vaddr, 0, pgd_base);
 302
 303 #ifdef CONFIG_HIGHMEM
 304         {
 305                 pgd_t *pgd;
 306                 pmd_t *pmd;
 307                 pte_t *pte;
 308
 309                 /*
 310                  * Permanent kmaps:
 311                  */
 312                 vaddr = PKMAP_BASE;
 313                 fixrange_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 314
 315                 pgd = swapper_pg_dir + pgd_index(vaddr);
 316                 pmd = pmd_offset(pgd, vaddr);
 317                 pte = pte_offset_kernel(pmd, vaddr);
 318                 pkmap_page_table = pte;
 319         }
 320 #endif
 321 }
 322
 323 /*
 324  * Clear kernel pagetables in a PMD_SIZE-aligned range.
 325  */
 326 static void clear_mappings(pgd_t *pgd_base, unsigned long start, unsigned long end)
 327 {
 328         unsigned long vaddr;
 329         pgd_t *pgd;
 330         pmd_t *pmd;
 331         int i, j;
 332
 333         pgd = pgd_base;
 334
 335         for (i = 0; i < PTRS_PER_PGD; pgd++, i++) {
 336                 vaddr = i*PGDIR_SIZE;
 337                 if (end && (vaddr >= end))
 338                         break;
 339                 pmd = pmd_offset(pgd, 0);
 340                 for (j = 0; j < PTRS_PER_PMD; pmd++, j++) {
 341                         vaddr = i*PGDIR_SIZE + j*PMD_SIZE;
 342                         if (end && (vaddr >= end))
 343                                 break;
 344                         if (vaddr < start)
 345                                 continue;
 346                         pmd_clear(pmd);
 347                 }
 348         }
 349         flush_tlb_all();
 350 }
 351
 352 void zap_low_mappings(void)
 353 {
 354         printk("zapping low mappings.\n");
 355         /*
 356          * Zap initial low-memory mappings.
 357          */
 358         clear_mappings(swapper_pg_dir, 0, 16*1024*1024);
 359 }
 360
 361 #ifndef CONFIG_DISCONTIGMEM
 362 void __init zone_sizes_init(void)
 363 {
 364         unsigned long zones_size[MAX_NR_ZONES] = {0, 0, 0};
 365         unsigned int max_dma, high, low;
 366
 367         max_dma = virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 368         low = max_low_pfn;
 369         high = highend_pfn;
 370
 371         if (low < max_dma)
 372                 zones_size[ZONE_DMA] = low;
 373         else {
 374                 zones_size[ZONE_DMA] = max_dma;
 375                 zones_size[ZONE_NORMAL] = low - max_dma;
 376 #ifdef CONFIG_HIGHMEM
 377                 zones_size[ZONE_HIGHMEM] = high - low;
 378 #endif
 379         }
 380         free_area_init(zones_size);
 381 }
 382 #else
 383 extern void zone_sizes_init(void);
 384 #endif /* !CONFIG_DISCONTIGMEM */
 385
 386 static int disable_nx __initdata = 0;
 387 u64 __supported_pte_mask = ~_PAGE_NX;
 388
 389 /*
 390  * noexec = on|off
 391  *
 392  * Control non executable mappings.
 393  *
 394  * on      Enable
 395  * off     Disable (disables exec-shield too)
 396  */
 397 static int __init noexec_setup(char *str)
 398 {
 399         if (!strncmp(str, "on",2) && cpu_has_nx) {
 400                 __supported_pte_mask |= _PAGE_NX;
 401                 disable_nx = 0;
 402         } else if (!strncmp(str,"off",3)) {
 403                 disable_nx = 1;
 404                 __supported_pte_mask &= ~_PAGE_NX;
 405                 exec_shield = 0;
 406         }
 407         return 1;
 408 }
 409
 410 __setup("noexec=", noexec_setup);
 411
 412 #ifdef CONFIG_X86_PAE
 413 int nx_enabled = 0;
 414
 415 static void __init set_nx(void)
 416 {
 417         unsigned int v[4], l, h;
 418
 419         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 420                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 421                 if ((v[3] & (1 << 20)) && !disable_nx) {
 422                         rdmsr(MSR_EFER, l, h);
 423                         l |= EFER_NX;
 424                         wrmsr(MSR_EFER, l, h);
 425                         nx_enabled = 1;
 426                         __supported_pte_mask |= _PAGE_NX;
 427                 }
 428         }
 429 }
 430 /*
 431  * Enables/disables executability of a given kernel page and
 432  * returns the previous setting.
 433  */
 434 int __init set_kernel_exec(unsigned long vaddr, int enable)
 435 {
 436         pte_t *pte;
 437         int ret = 1;
 438
 439         if (!nx_enabled)
 440                 goto out;
 441
 442         pte = lookup_address(vaddr);
 443         BUG_ON(!pte);
 444
 445         if (!pte_exec_kernel(*pte))
 446                 ret = 0;
 447
 448         if (enable)
 449                 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
 450         else
 451                 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
 452         __flush_tlb_all();
 453 out:
 454         return ret;
 455 }
 456
 457 #endif
 458
 459 /*
 460  * paging_init() sets up the page tables - note that the first 8MB are
 461  * already mapped by head.S.
 462  *
 463  * This routines also unmaps the page at virtual kernel address 0, so
 464  * that we can trap those pesky NULL-reference errors in the kernel.
 465  */
 466 void __init paging_init(void)
 467 {
 468 #ifdef CONFIG_X86_PAE
 469         set_nx();
 470         if (nx_enabled)
 471                 printk("NX (Execute Disable) protection: active\n");
 472 #endif
 473
 474         pagetable_init();
 475
 476         load_cr3(swapper_pg_dir);
 477
 478 #ifdef CONFIG_X86_PAE
 479         /*
 480          * We will bail out later - printk doesn't work right now so
 481          * the user would just see a hanging kernel.
 482          */
 483         if (cpu_has_pae)
 484                 set_in_cr4(X86_CR4_PAE);
 485 #endif
 486         __flush_tlb_all();
 487         /*
 488          * Subtle. SMP is doing it's boot stuff late (because it has to
 489          * fork idle threads) - but it also needs low mappings for the
 490          * protected-mode entry to work. We zap these entries only after
 491          * the WP-bit has been tested.
 492          */
 493 #ifndef CONFIG_SMP
 494         zap_low_mappings();
 495 #endif
 496         kmap_init();
 497         zone_sizes_init();
 498 }
 499
 500 /*
 501  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 502  * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
 503  * used to involve black magic jumps to work around some nasty CPU bugs,
 504  * but fortunately the switch to using exceptions got rid of all that.
 505  */
 506
 507 void __init test_wp_bit(void)
 508 {
 509         printk("Checking if this processor honours the WP bit even in supervisor mode... ");
 510
 511         /* Any page-aligned address will do, the test is non-destructive */
 512         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 513         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 514         clear_fixmap(FIX_WP_TEST);
 515
 516         if (!boot_cpu_data.wp_works_ok) {
 517                 printk("No.\n");
 518 #ifdef CONFIG_X86_WP_WORKS_OK
 519                 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 520 #endif
 521         } else {
 522                 printk("Ok.\n");
 523         }
 524 }
 525
 526 #ifndef CONFIG_DISCONTIGMEM
 527 static void __init set_max_mapnr_init(void)
 528 {
 529 #ifdef CONFIG_HIGHMEM
 530         highmem_start_page = pfn_to_page(highstart_pfn);
 531         max_mapnr = num_physpages = highend_pfn;
 532 #else
 533         max_mapnr = num_physpages = max_low_pfn;
 534 #endif
 535 }
 536 #define __free_all_bootmem() free_all_bootmem()
 537 #else
 538 #define __free_all_bootmem() free_all_bootmem_node(NODE_DATA(0))
 539 extern void set_max_mapnr_init(void);
 540 #endif /* !CONFIG_DISCONTIGMEM */
 541
 542 static struct kcore_list kcore_mem, kcore_vmalloc;
 543
 544 void __init mem_init(void)
 545 {
 546         extern int ppro_with_ram_bug(void);
 547         int codesize, reservedpages, datasize, initsize;
 548         int tmp;
 549         int bad_ppro;
 550
 551 #ifndef CONFIG_DISCONTIGMEM
 552         if (!mem_map)
 553                 BUG();
 554 #endif
 555
 556         bad_ppro = ppro_with_ram_bug();
 557
 558 #ifdef CONFIG_HIGHMEM
 559         /* check that fixmap and pkmap do not overlap */
 560         if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
 561                 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
 562                 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
 563                                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
 564                 BUG();
 565         }
 566 #endif
 567
 568         set_max_mapnr_init();
 569
 570 #ifdef CONFIG_HIGHMEM
 571         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE);
 572 #else
 573         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE);
 574 #endif
 575
 576         /* this will put all low memory onto the freelists */
 577         totalram_pages += __free_all_bootmem();
 578
 579         reservedpages = 0;
 580         for (tmp = 0; tmp < max_low_pfn; tmp++)
 581                 /*
 582                  * Only count reserved RAM pages
 583                  */
 584                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 585                         reservedpages++;
 586
 587         set_highmem_pages_init(bad_ppro);
 588
 589         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 590         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 591         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 592
 593         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 594         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 595                    VMALLOC_END-VMALLOC_START);
 596
 597         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
 598                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 599                 num_physpages << (PAGE_SHIFT-10),
 600                 codesize >> 10,
 601                 reservedpages << (PAGE_SHIFT-10),
 602                 datasize >> 10,
 603                 initsize >> 10,
 604                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 605                );
 606
 607 #ifdef CONFIG_X86_PAE
 608         if (!cpu_has_pae)
 609                 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
 610 #endif
 611         if (boot_cpu_data.wp_works_ok < 0)
 612                 test_wp_bit();
 613
 614         entry_trampoline_setup();
 615         default_ldt_page = virt_to_page(default_ldt);
 616         load_LDT(&init_mm.context);
 617 }
 618
 619 kmem_cache_t *pgd_cache, *pmd_cache, *kpmd_cache;
 620
 621 void __init pgtable_cache_init(void)
 622 {
 623         void (*ctor)(void *, kmem_cache_t *, unsigned long);
 624         void (*dtor)(void *, kmem_cache_t *, unsigned long);
 625
 626         if (PTRS_PER_PMD > 1) {
 627                 pmd_cache = kmem_cache_create("pmd",
 628                                         PTRS_PER_PMD*sizeof(pmd_t),
 629                                         PTRS_PER_PMD*sizeof(pmd_t),
 630                                         0,
 631                                         pmd_ctor,
 632                                         NULL);
 633                 if (!pmd_cache)
 634                         panic("pgtable_cache_init(): cannot create pmd cache");
 635
 636                 if (TASK_SIZE > PAGE_OFFSET) {
 637                         kpmd_cache = kmem_cache_create("kpmd",
 638                                         PTRS_PER_PMD*sizeof(pmd_t),
 639                                         PTRS_PER_PMD*sizeof(pmd_t),
 640                                         0,
 641                                         kpmd_ctor,
 642                                         NULL);
 643                         if (!kpmd_cache)
 644                                 panic("pgtable_cache_init(): "
 645                                                 "cannot create kpmd cache");
 646                 }
 647         }
 648
 649         if (PTRS_PER_PMD == 1 || TASK_SIZE <= PAGE_OFFSET)
 650                 ctor = pgd_ctor;
 651         else
 652                 ctor = NULL;
 653
 654         if (PTRS_PER_PMD == 1 && TASK_SIZE <= PAGE_OFFSET)
 655                 dtor = pgd_dtor;
 656         else
 657                 dtor = NULL;
 658
 659         pgd_cache = kmem_cache_create("pgd",
 660                                 PTRS_PER_PGD*sizeof(pgd_t),
 661                                 PTRS_PER_PGD*sizeof(pgd_t),
 662                                 0,
 663                                 ctor,
 664                                 dtor);
 665         if (!pgd_cache)
 666                 panic("pgtable_cache_init(): Cannot create pgd cache");
 667 }
 668
 669 /*
 670  * This function cannot be __init, since exceptions don't work in that
 671  * section.  Put this after the callers, so that it cannot be inlined.
 672  */
 673 static int noinline do_test_wp_bit(void)
 674 {
 675         char tmp_reg;
 676         int flag;
 677
 678         __asm__ __volatile__(
 679                 "       movb %0,%1      \n"
 680                 "1:     movb %1,%0      \n"
 681                 "       xorl %2,%2      \n"
 682                 "2:                     \n"
 683                 ".section __ex_table,\"a\"\n"
 684                 "       .align 4        \n"
 685                 "       .long 1b,2b     \n"
 686                 ".previous              \n"
 687                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 688                  "=q" (tmp_reg),
 689                  "=r" (flag)
 690                 :"2" (1)
 691                 :"memory");
 692
 693         return flag;
 694 }
 695
 696 void free_initmem(void)
 697 {
 698         unsigned long addr;
 699
 700         addr = (unsigned long)(&__init_begin);
 701         for (; addr < (unsigned long)(&__init_end); addr += PAGE_SIZE) {
 702                 ClearPageReserved(virt_to_page(addr));
 703                 set_page_count(virt_to_page(addr), 1);
 704                 free_page(addr);
 705                 totalram_pages++;
 706         }
 707         printk (KERN_INFO "Freeing unused kernel memory: %dk freed\n", (__init_end - __init_begin) >> 10);
 708 }
 709
 710 #ifdef CONFIG_BLK_DEV_INITRD
 711 void free_initrd_mem(unsigned long start, unsigned long end)
 712 {
 713         if (start < end)
 714                 printk (KERN_INFO "Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
 715         for (; start < end; start += PAGE_SIZE) {
 716                 ClearPageReserved(virt_to_page(start));
 717                 set_page_count(virt_to_page(start), 1);
 718                 free_page(start);
 719                 totalram_pages++;
 720         }
 721 }
 722 #endif