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/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/efi.h>
  31 #include <linux/memory_hotplug.h>
  32 #include <linux/initrd.h>
  33 #include <linux/cpumask.h>
  34
  35 #include <asm/processor.h>
  36 #include <asm/system.h>
  37 #include <asm/uaccess.h>
  38 #include <asm/pgtable.h>
  39 #include <asm/dma.h>
  40 #include <asm/fixmap.h>
  41 #include <asm/e820.h>
  42 #include <asm/apic.h>
  43 #include <asm/tlb.h>
  44 #include <asm/tlbflush.h>
  45 #include <asm/sections.h>
  46 #include <asm/paravirt.h>
  47
  48 unsigned int __VMALLOC_RESERVE = 128 << 20;
  49
  50 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  51 unsigned long highstart_pfn, highend_pfn;
  52
  53 static int noinline do_test_wp_bit(void);
  54
  55 /*
  56  * Creates a middle page table and puts a pointer to it in the
  57  * given global directory entry. This only returns the gd entry
  58  * in non-PAE compilation mode, since the middle layer is folded.
  59  */
  60 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  61 {
  62         pud_t *pud;
  63         pmd_t *pmd_table;
  64
  65 #ifdef CONFIG_X86_PAE
  66         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
  67                 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  68
  69                 paravirt_alloc_pd(__pa(pmd_table) >> PAGE_SHIFT);
  70                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  71                 pud = pud_offset(pgd, 0);
  72                 if (pmd_table != pmd_offset(pud, 0))
  73                         BUG();
  74         }
  75 #endif
  76         pud = pud_offset(pgd, 0);
  77         pmd_table = pmd_offset(pud, 0);
  78         return pmd_table;
  79 }
  80
  81 /*
  82  * Create a page table and place a pointer to it in a middle page
  83  * directory entry.
  84  */
  85 static pte_t * __init one_page_table_init(pmd_t *pmd)
  86 {
  87         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
  88                 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  89
  90                 paravirt_alloc_pt(__pa(page_table) >> PAGE_SHIFT);
  91                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
  92                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
  93         }
  94
  95         return pte_offset_kernel(pmd, 0);
  96 }
  97
  98 /*
  99  * This function initializes a certain range of kernel virtual memory
 100  * with new bootmem page tables, everywhere page tables are missing in
 101  * the given range.
 102  */
 103
 104 /*
 105  * NOTE: The pagetables are allocated contiguous on the physical space
 106  * so we can cache the place of the first one and move around without
 107  * checking the pgd every time.
 108  */
 109 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
 110 {
 111         pgd_t *pgd;
 112         pmd_t *pmd;
 113         int pgd_idx, pmd_idx;
 114         unsigned long vaddr;
 115
 116         vaddr = start;
 117         pgd_idx = pgd_index(vaddr);
 118         pmd_idx = pmd_index(vaddr);
 119         pgd = pgd_base + pgd_idx;
 120
 121         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 122                 pmd = one_md_table_init(pgd);
 123                 pmd = pmd + pmd_index(vaddr);
 124                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
 125                         one_page_table_init(pmd);
 126
 127                         vaddr += PMD_SIZE;
 128                 }
 129                 pmd_idx = 0;
 130         }
 131 }
 132
 133 static inline int is_kernel_text(unsigned long addr)
 134 {
 135         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 136                 return 1;
 137         return 0;
 138 }
 139
 140 /*
 141  * This maps the physical memory to kernel virtual address space, a total
 142  * of max_low_pfn pages, by creating page tables starting from address
 143  * PAGE_OFFSET.
 144  */
 145 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
 146 {
 147         unsigned long pfn;
 148         pgd_t *pgd;
 149         pmd_t *pmd;
 150         pte_t *pte;
 151         int pgd_idx, pmd_idx, pte_ofs;
 152
 153         pgd_idx = pgd_index(PAGE_OFFSET);
 154         pgd = pgd_base + pgd_idx;
 155         pfn = 0;
 156
 157         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 158                 pmd = one_md_table_init(pgd);
 159                 if (pfn >= max_low_pfn)
 160                         continue;
 161                 for (pmd_idx = 0; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
 162                         unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
 163
 164                         /* Map with big pages if possible, otherwise create normal page tables. */
 165                         if (cpu_has_pse) {
 166                                 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
 167                                 if (is_kernel_text(address) || is_kernel_text(address2))
 168                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
 169                                 else
 170                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
 171
 172                                 pfn += PTRS_PER_PTE;
 173                         } else {
 174                                 pte = one_page_table_init(pmd);
 175
 176                                 for (pte_ofs = 0;
 177                                      pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn;
 178                                      pte++, pfn++, pte_ofs++, address += PAGE_SIZE) {
 179                                         if (is_kernel_text(address))
 180                                                 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
 181                                         else
 182                                                 set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
 183                                 }
 184                         }
 185                 }
 186         }
 187 }
 188
 189 static inline int page_kills_ppro(unsigned long pagenr)
 190 {
 191         if (pagenr >= 0x70000 && pagenr <= 0x7003F)
 192                 return 1;
 193         return 0;
 194 }
 195
 196 int page_is_ram(unsigned long pagenr)
 197 {
 198         int i;
 199         unsigned long addr, end;
 200
 201         if (efi_enabled) {
 202                 efi_memory_desc_t *md;
 203                 void *p;
 204
 205                 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 206                         md = p;
 207                         if (!is_available_memory(md))
 208                                 continue;
 209                         addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 210                         end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
 211
 212                         if ((pagenr >= addr) && (pagenr < end))
 213                                 return 1;
 214                 }
 215                 return 0;
 216         }
 217
 218         for (i = 0; i < e820.nr_map; i++) {
 219
 220                 if (e820.map[i].type != E820_RAM)       /* not usable memory */
 221                         continue;
 222                 /*
 223                  *      !!!FIXME!!! Some BIOSen report areas as RAM that
 224                  *      are not. Notably the 640->1Mb area. We need a sanity
 225                  *      check here.
 226                  */
 227                 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 228                 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
 229                 if  ((pagenr >= addr) && (pagenr < end))
 230                         return 1;
 231         }
 232         return 0;
 233 }
 234
 235 #ifdef CONFIG_HIGHMEM
 236 pte_t *kmap_pte;
 237 pgprot_t kmap_prot;
 238
 239 #define kmap_get_fixmap_pte(vaddr)                                      \
 240         pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
 241
 242 static void __init kmap_init(void)
 243 {
 244         unsigned long kmap_vstart;
 245
 246         /* cache the first kmap pte */
 247         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 248         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 249
 250         kmap_prot = PAGE_KERNEL;
 251 }
 252
 253 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 254 {
 255         pgd_t *pgd;
 256         pud_t *pud;
 257         pmd_t *pmd;
 258         pte_t *pte;
 259         unsigned long vaddr;
 260
 261         vaddr = PKMAP_BASE;
 262         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 263
 264         pgd = swapper_pg_dir + pgd_index(vaddr);
 265         pud = pud_offset(pgd, vaddr);
 266         pmd = pmd_offset(pud, vaddr);
 267         pte = pte_offset_kernel(pmd, vaddr);
 268         pkmap_page_table = pte;
 269 }
 270
 271 static void __meminit free_new_highpage(struct page *page)
 272 {
 273         init_page_count(page);
 274         __free_page(page);
 275         totalhigh_pages++;
 276 }
 277
 278 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
 279 {
 280         if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
 281                 ClearPageReserved(page);
 282                 free_new_highpage(page);
 283         } else
 284                 SetPageReserved(page);
 285 }
 286
 287 static int __meminit add_one_highpage_hotplug(struct page *page, unsigned long pfn)
 288 {
 289         free_new_highpage(page);
 290         totalram_pages++;
 291 #ifdef CONFIG_FLATMEM
 292         max_mapnr = max(pfn, max_mapnr);
 293 #endif
 294         num_physpages++;
 295         return 0;
 296 }
 297
 298 /*
 299  * Not currently handling the NUMA case.
 300  * Assuming single node and all memory that
 301  * has been added dynamically that would be
 302  * onlined here is in HIGHMEM
 303  */
 304 void __meminit online_page(struct page *page)
 305 {
 306         ClearPageReserved(page);
 307         add_one_highpage_hotplug(page, page_to_pfn(page));
 308 }
 309
 310
 311 #ifdef CONFIG_NUMA
 312 extern void set_highmem_pages_init(int);
 313 #else
 314 static void __init set_highmem_pages_init(int bad_ppro)
 315 {
 316         int pfn;
 317         for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
 318                 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
 319         totalram_pages += totalhigh_pages;
 320 }
 321 #endif /* CONFIG_FLATMEM */
 322
 323 #else
 324 #define kmap_init() do { } while (0)
 325 #define permanent_kmaps_init(pgd_base) do { } while (0)
 326 #define set_highmem_pages_init(bad_ppro) do { } while (0)
 327 #endif /* CONFIG_HIGHMEM */
 328
 329 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
 330 EXPORT_SYMBOL(__PAGE_KERNEL);
 331 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
 332
 333 #ifdef CONFIG_NUMA
 334 extern void __init remap_numa_kva(void);
 335 #else
 336 #define remap_numa_kva() do {} while (0)
 337 #endif
 338
 339 void __init native_pagetable_setup_start(pgd_t *base)
 340 {
 341 #ifdef CONFIG_X86_PAE
 342         int i;
 343
 344         /*
 345          * Init entries of the first-level page table to the
 346          * zero page, if they haven't already been set up.
 347          *
 348          * In a normal native boot, we'll be running on a
 349          * pagetable rooted in swapper_pg_dir, but not in PAE
 350          * mode, so this will end up clobbering the mappings
 351          * for the lower 24Mbytes of the address space,
 352          * without affecting the kernel address space.
 353          */
 354         for (i = 0; i < USER_PTRS_PER_PGD; i++)
 355                 set_pgd(&base[i],
 356                         __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
 357
 358         /* Make sure kernel address space is empty so that a pagetable
 359            will be allocated for it. */
 360         memset(&base[USER_PTRS_PER_PGD], 0,
 361                KERNEL_PGD_PTRS * sizeof(pgd_t));
 362 #else
 363         paravirt_alloc_pd(__pa(swapper_pg_dir) >> PAGE_SHIFT);
 364 #endif
 365 }
 366
 367 void __init native_pagetable_setup_done(pgd_t *base)
 368 {
 369 #ifdef CONFIG_X86_PAE
 370         /*
 371          * Add low memory identity-mappings - SMP needs it when
 372          * starting up on an AP from real-mode. In the non-PAE
 373          * case we already have these mappings through head.S.
 374          * All user-space mappings are explicitly cleared after
 375          * SMP startup.
 376          */
 377         set_pgd(&base[0], base[USER_PTRS_PER_PGD]);
 378 #endif
 379 }
 380
 381 /*
 382  * Build a proper pagetable for the kernel mappings.  Up until this
 383  * point, we've been running on some set of pagetables constructed by
 384  * the boot process.
 385  *
 386  * If we're booting on native hardware, this will be a pagetable
 387  * constructed in arch/i386/kernel/head.S, and not running in PAE mode
 388  * (even if we'll end up running in PAE).  The root of the pagetable
 389  * will be swapper_pg_dir.
 390  *
 391  * If we're booting paravirtualized under a hypervisor, then there are
 392  * more options: we may already be running PAE, and the pagetable may
 393  * or may not be based in swapper_pg_dir.  In any case,
 394  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 395  * appropriately for the rest of the initialization to work.
 396  *
 397  * In general, pagetable_init() assumes that the pagetable may already
 398  * be partially populated, and so it avoids stomping on any existing
 399  * mappings.
 400  */
 401 static void __init pagetable_init (void)
 402 {
 403         unsigned long vaddr, end;
 404         pgd_t *pgd_base = swapper_pg_dir;
 405
 406         paravirt_pagetable_setup_start(pgd_base);
 407
 408         /* Enable PSE if available */
 409         if (cpu_has_pse)
 410                 set_in_cr4(X86_CR4_PSE);
 411
 412         /* Enable PGE if available */
 413         if (cpu_has_pge) {
 414                 set_in_cr4(X86_CR4_PGE);
 415                 __PAGE_KERNEL |= _PAGE_GLOBAL;
 416                 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
 417         }
 418
 419         kernel_physical_mapping_init(pgd_base);
 420         remap_numa_kva();
 421
 422         /*
 423          * Fixed mappings, only the page table structure has to be
 424          * created - mappings will be set by set_fixmap():
 425          */
 426         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 427         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 428         page_table_range_init(vaddr, end, pgd_base);
 429
 430         permanent_kmaps_init(pgd_base);
 431
 432         paravirt_pagetable_setup_done(pgd_base);
 433 }
 434
 435 #if defined(CONFIG_SOFTWARE_SUSPEND) || defined(CONFIG_ACPI_SLEEP)
 436 /*
 437  * Swap suspend & friends need this for resume because things like the intel-agp
 438  * driver might have split up a kernel 4MB mapping.
 439  */
 440 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
 441         __attribute__ ((aligned (PAGE_SIZE)));
 442
 443 static inline void save_pg_dir(void)
 444 {
 445         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 446 }
 447 #else
 448 static inline void save_pg_dir(void)
 449 {
 450 }
 451 #endif
 452
 453 void zap_low_mappings (void)
 454 {
 455         int i;
 456
 457         save_pg_dir();
 458
 459         /*
 460          * Zap initial low-memory mappings.
 461          *
 462          * Note that "pgd_clear()" doesn't do it for
 463          * us, because pgd_clear() is a no-op on i386.
 464          */
 465         for (i = 0; i < USER_PTRS_PER_PGD; i++)
 466 #ifdef CONFIG_X86_PAE
 467                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 468 #else
 469                 set_pgd(swapper_pg_dir+i, __pgd(0));
 470 #endif
 471         flush_tlb_all();
 472 }
 473
 474 static int disable_nx __initdata = 0;
 475 u64 __supported_pte_mask __read_mostly = ~_PAGE_NX;
 476
 477 /*
 478  * noexec = on|off
 479  *
 480  * Control non executable mappings.
 481  *
 482  * on      Enable
 483  * off     Disable
 484  */
 485 static int __init noexec_setup(char *str)
 486 {
 487         if (!str || !strcmp(str, "on")) {
 488                 if (cpu_has_nx) {
 489                         __supported_pte_mask |= _PAGE_NX;
 490                         disable_nx = 0;
 491                 }
 492         } else if (!strcmp(str,"off")) {
 493                 disable_nx = 1;
 494                 __supported_pte_mask &= ~_PAGE_NX;
 495         } else
 496                 return -EINVAL;
 497
 498         return 0;
 499 }
 500 early_param("noexec", noexec_setup);
 501
 502 int nx_enabled = 0;
 503 #ifdef CONFIG_X86_PAE
 504
 505 static void __init set_nx(void)
 506 {
 507         unsigned int v[4], l, h;
 508
 509         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 510                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 511                 if ((v[3] & (1 << 20)) && !disable_nx) {
 512                         rdmsr(MSR_EFER, l, h);
 513                         l |= EFER_NX;
 514                         wrmsr(MSR_EFER, l, h);
 515                         nx_enabled = 1;
 516                         __supported_pte_mask |= _PAGE_NX;
 517                 }
 518         }
 519 }
 520
 521 /*
 522  * Enables/disables executability of a given kernel page and
 523  * returns the previous setting.
 524  */
 525 int __init set_kernel_exec(unsigned long vaddr, int enable)
 526 {
 527         pte_t *pte;
 528         int ret = 1;
 529
 530         if (!nx_enabled)
 531                 goto out;
 532
 533         pte = lookup_address(vaddr);
 534         BUG_ON(!pte);
 535
 536         if (!pte_exec_kernel(*pte))
 537                 ret = 0;
 538
 539         if (enable)
 540                 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
 541         else
 542                 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
 543         pte_update_defer(&init_mm, vaddr, pte);
 544         __flush_tlb_all();
 545 out:
 546         return ret;
 547 }
 548
 549 #endif
 550
 551 /*
 552  * paging_init() sets up the page tables - note that the first 8MB are
 553  * already mapped by head.S.
 554  *
 555  * This routines also unmaps the page at virtual kernel address 0, so
 556  * that we can trap those pesky NULL-reference errors in the kernel.
 557  */
 558 void __init paging_init(void)
 559 {
 560 #ifdef CONFIG_X86_PAE
 561         set_nx();
 562         if (nx_enabled)
 563                 printk("NX (Execute Disable) protection: active\n");
 564 #endif
 565
 566         pagetable_init();
 567
 568         load_cr3(swapper_pg_dir);
 569
 570 #ifdef CONFIG_X86_PAE
 571         /*
 572          * We will bail out later - printk doesn't work right now so
 573          * the user would just see a hanging kernel.
 574          */
 575         if (cpu_has_pae)
 576                 set_in_cr4(X86_CR4_PAE);
 577 #endif
 578         __flush_tlb_all();
 579
 580         kmap_init();
 581 }
 582
 583 /*
 584  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 585  * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
 586  * used to involve black magic jumps to work around some nasty CPU bugs,
 587  * but fortunately the switch to using exceptions got rid of all that.
 588  */
 589
 590 static void __init test_wp_bit(void)
 591 {
 592         printk("Checking if this processor honours the WP bit even in supervisor mode... ");
 593
 594         /* Any page-aligned address will do, the test is non-destructive */
 595         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 596         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 597         clear_fixmap(FIX_WP_TEST);
 598
 599         if (!boot_cpu_data.wp_works_ok) {
 600                 printk("No.\n");
 601 #ifdef CONFIG_X86_WP_WORKS_OK
 602                 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 603 #endif
 604         } else {
 605                 printk("Ok.\n");
 606         }
 607 }
 608
 609 static struct kcore_list kcore_mem, kcore_vmalloc;
 610
 611 void __init mem_init(void)
 612 {
 613         extern int ppro_with_ram_bug(void);
 614         int codesize, reservedpages, datasize, initsize;
 615         int tmp;
 616         int bad_ppro;
 617
 618 #ifdef CONFIG_FLATMEM
 619         BUG_ON(!mem_map);
 620 #endif
 621
 622         bad_ppro = ppro_with_ram_bug();
 623
 624 #ifdef CONFIG_HIGHMEM
 625         /* check that fixmap and pkmap do not overlap */
 626         if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
 627                 printk(KERN_ERR "fixmap and kmap areas overlap - this will crash\n");
 628                 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
 629                                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE, FIXADDR_START);
 630                 BUG();
 631         }
 632 #endif
 633
 634         /* this will put all low memory onto the freelists */
 635         totalram_pages += free_all_bootmem();
 636
 637         reservedpages = 0;
 638         for (tmp = 0; tmp < max_low_pfn; tmp++)
 639                 /*
 640                  * Only count reserved RAM pages
 641                  */
 642                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 643                         reservedpages++;
 644
 645         set_highmem_pages_init(bad_ppro);
 646
 647         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 648         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 649         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 650
 651         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 652         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 653                    VMALLOC_END-VMALLOC_START);
 654
 655         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
 656                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 657                 num_physpages << (PAGE_SHIFT-10),
 658                 codesize >> 10,
 659                 reservedpages << (PAGE_SHIFT-10),
 660                 datasize >> 10,
 661                 initsize >> 10,
 662                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 663                );
 664
 665 #if 1 /* double-sanity-check paranoia */
 666         printk("virtual kernel memory layout:\n"
 667                "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 668 #ifdef CONFIG_HIGHMEM
 669                "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 670 #endif
 671                "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 672                "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 673                "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 674                "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 675                "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 676                FIXADDR_START, FIXADDR_TOP,
 677                (FIXADDR_TOP - FIXADDR_START) >> 10,
 678
 679 #ifdef CONFIG_HIGHMEM
 680                PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 681                (LAST_PKMAP*PAGE_SIZE) >> 10,
 682 #endif
 683
 684                VMALLOC_START, VMALLOC_END,
 685                (VMALLOC_END - VMALLOC_START) >> 20,
 686
 687                (unsigned long)__va(0), (unsigned long)high_memory,
 688                ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 689
 690                (unsigned long)&__init_begin, (unsigned long)&__init_end,
 691                ((unsigned long)&__init_end - (unsigned long)&__init_begin) >> 10,
 692
 693                (unsigned long)&_etext, (unsigned long)&_edata,
 694                ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 695
 696                (unsigned long)&_text, (unsigned long)&_etext,
 697                ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 698
 699 #ifdef CONFIG_HIGHMEM
 700         BUG_ON(PKMAP_BASE+LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
 701         BUG_ON(VMALLOC_END                     > PKMAP_BASE);
 702 #endif
 703         BUG_ON(VMALLOC_START                   > VMALLOC_END);
 704         BUG_ON((unsigned long)high_memory      > VMALLOC_START);
 705 #endif /* double-sanity-check paranoia */
 706
 707 #ifdef CONFIG_X86_PAE
 708         if (!cpu_has_pae)
 709                 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
 710 #endif
 711         if (boot_cpu_data.wp_works_ok < 0)
 712                 test_wp_bit();
 713
 714         /*
 715          * Subtle. SMP is doing it's boot stuff late (because it has to
 716          * fork idle threads) - but it also needs low mappings for the
 717          * protected-mode entry to work. We zap these entries only after
 718          * the WP-bit has been tested.
 719          */
 720 #ifndef CONFIG_SMP
 721         zap_low_mappings();
 722 #endif
 723 }
 724
 725 #ifdef CONFIG_MEMORY_HOTPLUG
 726 int arch_add_memory(int nid, u64 start, u64 size)
 727 {
 728         struct pglist_data *pgdata = NODE_DATA(nid);
 729         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 730         unsigned long start_pfn = start >> PAGE_SHIFT;
 731         unsigned long nr_pages = size >> PAGE_SHIFT;
 732
 733         return __add_pages(zone, start_pfn, nr_pages);
 734 }
 735
 736 int remove_memory(u64 start, u64 size)
 737 {
 738         return -EINVAL;
 739 }
 740 EXPORT_SYMBOL_GPL(remove_memory);
 741 #endif
 742
 743 struct kmem_cache *pmd_cache;
 744
 745 void __init pgtable_cache_init(void)
 746 {
 747         size_t pgd_size = PTRS_PER_PGD*sizeof(pgd_t);
 748
 749         if (PTRS_PER_PMD > 1) {
 750                 pmd_cache = kmem_cache_create("pmd",
 751                                         PTRS_PER_PMD*sizeof(pmd_t),
 752                                         PTRS_PER_PMD*sizeof(pmd_t),
 753                                         SLAB_PANIC,
 754                                         pmd_ctor,
 755                                         NULL);
 756                 if (!SHARED_KERNEL_PMD) {
 757                         /* If we're in PAE mode and have a non-shared
 758                            kernel pmd, then the pgd size must be a
 759                            page size.  This is because the pgd_list
 760                            links through the page structure, so there
 761                            can only be one pgd per page for this to
 762                            work. */
 763                         pgd_size = PAGE_SIZE;
 764                 }
 765         }
 766 }
 767
 768 /*
 769  * This function cannot be __init, since exceptions don't work in that
 770  * section.  Put this after the callers, so that it cannot be inlined.
 771  */
 772 static int noinline do_test_wp_bit(void)
 773 {
 774         char tmp_reg;
 775         int flag;
 776
 777         __asm__ __volatile__(
 778                 "       movb %0,%1      \n"
 779                 "1:     movb %1,%0      \n"
 780                 "       xorl %2,%2      \n"
 781                 "2:                     \n"
 782                 ".section __ex_table,\"a\"\n"
 783                 "       .align 4        \n"
 784                 "       .long 1b,2b     \n"
 785                 ".previous              \n"
 786                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 787                  "=q" (tmp_reg),
 788                  "=r" (flag)
 789                 :"2" (1)
 790                 :"memory");
 791
 792         return flag;
 793 }
 794
 795 #ifdef CONFIG_DEBUG_RODATA
 796
 797 void mark_rodata_ro(void)
 798 {
 799         unsigned long start = PFN_ALIGN(_text);
 800         unsigned long size = PFN_ALIGN(_etext) - start;
 801
 802 #ifndef CONFIG_KPROBES
 803 #ifdef CONFIG_HOTPLUG_CPU
 804         /* It must still be possible to apply SMP alternatives. */
 805         if (num_possible_cpus() <= 1)
 806 #endif
 807         {
 808                 change_page_attr(virt_to_page(start),
 809                                  size >> PAGE_SHIFT, PAGE_KERNEL_RX);
 810                 printk("Write protecting the kernel text: %luk\n", size >> 10);
 811         }
 812 #endif
 813         start += size;
 814         size = (unsigned long)__end_rodata - start;
 815         change_page_attr(virt_to_page(start),
 816                          size >> PAGE_SHIFT, PAGE_KERNEL_RO);
 817         printk("Write protecting the kernel read-only data: %luk\n",
 818                size >> 10);
 819
 820         /*
 821          * change_page_attr() requires a global_flush_tlb() call after it.
 822          * We do this after the printk so that if something went wrong in the
 823          * change, the printk gets out at least to give a better debug hint
 824          * of who is the culprit.
 825          */
 826         global_flush_tlb();
 827 }
 828 #endif
 829
 830 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 831 {
 832         unsigned long addr;
 833
 834         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 835                 ClearPageReserved(virt_to_page(addr));
 836                 init_page_count(virt_to_page(addr));
 837                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
 838                 free_page(addr);
 839                 totalram_pages++;
 840         }
 841         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 842 }
 843
 844 void free_initmem(void)
 845 {
 846         free_init_pages("unused kernel memory",
 847                         (unsigned long)(&__init_begin),
 848                         (unsigned long)(&__init_end));
 849 }
 850
 851 #ifdef CONFIG_BLK_DEV_INITRD
 852 void free_initrd_mem(unsigned long start, unsigned long end)
 853 {
 854         free_init_pages("initrd memory", start, end);
 855 }
 856 #endif
 857