err.no Git - linux-2.6/blob - arch/x86/mm/init_32.c

   1 /*
   2  *
   3  *  Copyright (C) 1995  Linus Torvalds
   4  *
   5  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/signal.h>
  10 #include <linux/sched.h>
  11 #include <linux/kernel.h>
  12 #include <linux/errno.h>
  13 #include <linux/string.h>
  14 #include <linux/types.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/mman.h>
  17 #include <linux/mm.h>
  18 #include <linux/hugetlb.h>
  19 #include <linux/swap.h>
  20 #include <linux/smp.h>
  21 #include <linux/init.h>
  22 #include <linux/highmem.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/pfn.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/memory_hotplug.h>
  30 #include <linux/initrd.h>
  31 #include <linux/cpumask.h>
  32
  33 #include <asm/asm.h>
  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/bugs.h>
  43 #include <asm/tlb.h>
  44 #include <asm/tlbflush.h>
  45 #include <asm/pgalloc.h>
  46 #include <asm/sections.h>
  47 #include <asm/paravirt.h>
  48 #include <asm/setup.h>
  49 #include <asm/cacheflush.h>
  50
  51 unsigned int __VMALLOC_RESERVE = 128 << 20;
  52
  53 unsigned long max_pfn_mapped;
  54
  55 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  56 unsigned long highstart_pfn, highend_pfn;
  57
  58 static noinline int do_test_wp_bit(void);
  59
  60
  61 static unsigned long __initdata table_start;
  62 static unsigned long __meminitdata table_end;
  63 static unsigned long __meminitdata table_top;
  64
  65 static int __initdata after_init_bootmem;
  66
  67 static __init void *alloc_low_page(unsigned long *phys)
  68 {
  69         unsigned long pfn = table_end++;
  70         void *adr;
  71
  72         if (pfn >= table_top)
  73                 panic("alloc_low_page: ran out of memory");
  74
  75         adr = __va(pfn * PAGE_SIZE);
  76         memset(adr, 0, PAGE_SIZE);
  77         *phys  = pfn * PAGE_SIZE;
  78         return adr;
  79 }
  80
  81 /*
  82  * Creates a middle page table and puts a pointer to it in the
  83  * given global directory entry. This only returns the gd entry
  84  * in non-PAE compilation mode, since the middle layer is folded.
  85  */
  86 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  87 {
  88         pud_t *pud;
  89         pmd_t *pmd_table;
  90
  91 #ifdef CONFIG_X86_PAE
  92         unsigned long phys;
  93         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
  94                 if (after_init_bootmem)
  95                         pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
  96                 else
  97                         pmd_table = (pmd_t *)alloc_low_page(&phys);
  98                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
  99                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
 100                 pud = pud_offset(pgd, 0);
 101                 BUG_ON(pmd_table != pmd_offset(pud, 0));
 102         }
 103 #endif
 104         pud = pud_offset(pgd, 0);
 105         pmd_table = pmd_offset(pud, 0);
 106
 107         return pmd_table;
 108 }
 109
 110 /*
 111  * Create a page table and place a pointer to it in a middle page
 112  * directory entry:
 113  */
 114 static pte_t * __init one_page_table_init(pmd_t *pmd)
 115 {
 116         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
 117                 pte_t *page_table = NULL;
 118
 119                 if (after_init_bootmem) {
 120 #ifdef CONFIG_DEBUG_PAGEALLOC
 121                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
 122 #endif
 123                         if (!page_table)
 124                                 page_table =
 125                                 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 126                 } else {
 127                         unsigned long phys;
 128                         page_table = (pte_t *)alloc_low_page(&phys);
 129                 }
 130
 131                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
 132                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
 133                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
 134         }
 135
 136         return pte_offset_kernel(pmd, 0);
 137 }
 138
 139 /*
 140  * This function initializes a certain range of kernel virtual memory
 141  * with new bootmem page tables, everywhere page tables are missing in
 142  * the given range.
 143  *
 144  * NOTE: The pagetables are allocated contiguous on the physical space
 145  * so we can cache the place of the first one and move around without
 146  * checking the pgd every time.
 147  */
 148 static void __init
 149 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 150 {
 151         int pgd_idx, pmd_idx;
 152         unsigned long vaddr;
 153         pgd_t *pgd;
 154         pmd_t *pmd;
 155
 156         vaddr = start;
 157         pgd_idx = pgd_index(vaddr);
 158         pmd_idx = pmd_index(vaddr);
 159         pgd = pgd_base + pgd_idx;
 160
 161         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 162                 pmd = one_md_table_init(pgd);
 163                 pmd = pmd + pmd_index(vaddr);
 164                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 165                                                         pmd++, pmd_idx++) {
 166                         one_page_table_init(pmd);
 167
 168                         vaddr += PMD_SIZE;
 169                 }
 170                 pmd_idx = 0;
 171         }
 172 }
 173
 174 static inline int is_kernel_text(unsigned long addr)
 175 {
 176         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 177                 return 1;
 178         return 0;
 179 }
 180
 181 /*
 182  * This maps the physical memory to kernel virtual address space, a total
 183  * of max_low_pfn pages, by creating page tables starting from address
 184  * PAGE_OFFSET:
 185  */
 186 static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
 187                                                 unsigned long start_pfn,
 188                                                 unsigned long end_pfn,
 189                                                 int use_pse)
 190 {
 191         int pgd_idx, pmd_idx, pte_ofs;
 192         unsigned long pfn;
 193         pgd_t *pgd;
 194         pmd_t *pmd;
 195         pte_t *pte;
 196         unsigned pages_2m = 0, pages_4k = 0;
 197
 198         if (!cpu_has_pse)
 199                 use_pse = 0;
 200
 201         pfn = start_pfn;
 202         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 203         pgd = pgd_base + pgd_idx;
 204         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 205                 pmd = one_md_table_init(pgd);
 206
 207                 if (pfn >= end_pfn)
 208                         continue;
 209 #ifdef CONFIG_X86_PAE
 210                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 211                 pmd += pmd_idx;
 212 #else
 213                 pmd_idx = 0;
 214 #endif
 215                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
 216                      pmd++, pmd_idx++) {
 217                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 218
 219                         /*
 220                          * Map with big pages if possible, otherwise
 221                          * create normal page tables:
 222                          */
 223                         if (use_pse) {
 224                                 unsigned int addr2;
 225                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 226
 227                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 228                                         PAGE_OFFSET + PAGE_SIZE-1;
 229
 230                                 if (is_kernel_text(addr) ||
 231                                     is_kernel_text(addr2))
 232                                         prot = PAGE_KERNEL_LARGE_EXEC;
 233
 234                                 pages_2m++;
 235                                 set_pmd(pmd, pfn_pmd(pfn, prot));
 236
 237                                 pfn += PTRS_PER_PTE;
 238                                 continue;
 239                         }
 240                         pte = one_page_table_init(pmd);
 241
 242                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 243                         pte += pte_ofs;
 244                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 245                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 246                                 pgprot_t prot = PAGE_KERNEL;
 247
 248                                 if (is_kernel_text(addr))
 249                                         prot = PAGE_KERNEL_EXEC;
 250
 251                                 pages_4k++;
 252                                 set_pte(pte, pfn_pte(pfn, prot));
 253                         }
 254                 }
 255         }
 256         update_page_count(PG_LEVEL_2M, pages_2m);
 257         update_page_count(PG_LEVEL_4K, pages_4k);
 258 }
 259
 260 /*
 261  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
 262  * is valid. The argument is a physical page number.
 263  *
 264  *
 265  * On x86, access has to be given to the first megabyte of ram because that area
 266  * contains bios code and data regions used by X and dosemu and similar apps.
 267  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 268  * mmio resources as well as potential bios/acpi data regions.
 269  */
 270 int devmem_is_allowed(unsigned long pagenr)
 271 {
 272         if (pagenr <= 256)
 273                 return 1;
 274         if (!page_is_ram(pagenr))
 275                 return 1;
 276         return 0;
 277 }
 278
 279 #ifdef CONFIG_HIGHMEM
 280 pte_t *kmap_pte;
 281 pgprot_t kmap_prot;
 282
 283 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 284 {
 285         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 286                         vaddr), vaddr), vaddr);
 287 }
 288
 289 static void __init kmap_init(void)
 290 {
 291         unsigned long kmap_vstart;
 292
 293         /*
 294          * Cache the first kmap pte:
 295          */
 296         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 297         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 298
 299         kmap_prot = PAGE_KERNEL;
 300 }
 301
 302 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 303 {
 304         unsigned long vaddr;
 305         pgd_t *pgd;
 306         pud_t *pud;
 307         pmd_t *pmd;
 308         pte_t *pte;
 309
 310         vaddr = PKMAP_BASE;
 311         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 312
 313         pgd = swapper_pg_dir + pgd_index(vaddr);
 314         pud = pud_offset(pgd, vaddr);
 315         pmd = pmd_offset(pud, vaddr);
 316         pte = pte_offset_kernel(pmd, vaddr);
 317         pkmap_page_table = pte;
 318 }
 319
 320 static void __init add_one_highpage_init(struct page *page, int pfn)
 321 {
 322         ClearPageReserved(page);
 323         init_page_count(page);
 324         __free_page(page);
 325         totalhigh_pages++;
 326 }
 327
 328 struct add_highpages_data {
 329         unsigned long start_pfn;
 330         unsigned long end_pfn;
 331 };
 332
 333 static int __init add_highpages_work_fn(unsigned long start_pfn,
 334                                          unsigned long end_pfn, void *datax)
 335 {
 336         int node_pfn;
 337         struct page *page;
 338         unsigned long final_start_pfn, final_end_pfn;
 339         struct add_highpages_data *data;
 340
 341         data = (struct add_highpages_data *)datax;
 342
 343         final_start_pfn = max(start_pfn, data->start_pfn);
 344         final_end_pfn = min(end_pfn, data->end_pfn);
 345         if (final_start_pfn >= final_end_pfn)
 346                 return 0;
 347
 348         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
 349              node_pfn++) {
 350                 if (!pfn_valid(node_pfn))
 351                         continue;
 352                 page = pfn_to_page(node_pfn);
 353                 add_one_highpage_init(page, node_pfn);
 354         }
 355
 356         return 0;
 357
 358 }
 359
 360 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
 361                                               unsigned long end_pfn)
 362 {
 363         struct add_highpages_data data;
 364
 365         data.start_pfn = start_pfn;
 366         data.end_pfn = end_pfn;
 367
 368         work_with_active_regions(nid, add_highpages_work_fn, &data);
 369 }
 370
 371 #ifndef CONFIG_NUMA
 372 static void __init set_highmem_pages_init(void)
 373 {
 374         add_highpages_with_active_regions(0, highstart_pfn, highend_pfn);
 375
 376         totalram_pages += totalhigh_pages;
 377 }
 378 #endif /* !CONFIG_NUMA */
 379
 380 #else
 381 # define kmap_init()                            do { } while (0)
 382 # define permanent_kmaps_init(pgd_base)         do { } while (0)
 383 # define set_highmem_pages_init()       do { } while (0)
 384 #endif /* CONFIG_HIGHMEM */
 385
 386 pteval_t __PAGE_KERNEL = _PAGE_KERNEL;
 387 EXPORT_SYMBOL(__PAGE_KERNEL);
 388
 389 pteval_t __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
 390
 391 void __init native_pagetable_setup_start(pgd_t *base)
 392 {
 393         unsigned long pfn, va;
 394         pgd_t *pgd;
 395         pud_t *pud;
 396         pmd_t *pmd;
 397         pte_t *pte;
 398
 399         /*
 400          * Remove any mappings which extend past the end of physical
 401          * memory from the boot time page table:
 402          */
 403         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 404                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 405                 pgd = base + pgd_index(va);
 406                 if (!pgd_present(*pgd))
 407                         break;
 408
 409                 pud = pud_offset(pgd, va);
 410                 pmd = pmd_offset(pud, va);
 411                 if (!pmd_present(*pmd))
 412                         break;
 413
 414                 pte = pte_offset_kernel(pmd, va);
 415                 if (!pte_present(*pte))
 416                         break;
 417
 418                 pte_clear(NULL, va, pte);
 419         }
 420         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 421 }
 422
 423 void __init native_pagetable_setup_done(pgd_t *base)
 424 {
 425 }
 426
 427 /*
 428  * Build a proper pagetable for the kernel mappings.  Up until this
 429  * point, we've been running on some set of pagetables constructed by
 430  * the boot process.
 431  *
 432  * If we're booting on native hardware, this will be a pagetable
 433  * constructed in arch/x86/kernel/head_32.S.  The root of the
 434  * pagetable will be swapper_pg_dir.
 435  *
 436  * If we're booting paravirtualized under a hypervisor, then there are
 437  * more options: we may already be running PAE, and the pagetable may
 438  * or may not be based in swapper_pg_dir.  In any case,
 439  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 440  * appropriately for the rest of the initialization to work.
 441  *
 442  * In general, pagetable_init() assumes that the pagetable may already
 443  * be partially populated, and so it avoids stomping on any existing
 444  * mappings.
 445  */
 446 static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base)
 447 {
 448         unsigned long vaddr, end;
 449
 450         /*
 451          * Fixed mappings, only the page table structure has to be
 452          * created - mappings will be set by set_fixmap():
 453          */
 454         early_ioremap_clear();
 455         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 456         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 457         page_table_range_init(vaddr, end, pgd_base);
 458         early_ioremap_reset();
 459 }
 460
 461 static void __init pagetable_init(void)
 462 {
 463         pgd_t *pgd_base = swapper_pg_dir;
 464
 465         paravirt_pagetable_setup_start(pgd_base);
 466
 467         permanent_kmaps_init(pgd_base);
 468
 469         paravirt_pagetable_setup_done(pgd_base);
 470 }
 471
 472 #ifdef CONFIG_ACPI_SLEEP
 473 /*
 474  * ACPI suspend needs this for resume, because things like the intel-agp
 475  * driver might have split up a kernel 4MB mapping.
 476  */
 477 char swsusp_pg_dir[PAGE_SIZE]
 478         __attribute__ ((aligned(PAGE_SIZE)));
 479
 480 static inline void save_pg_dir(void)
 481 {
 482         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 483 }
 484 #else /* !CONFIG_ACPI_SLEEP */
 485 static inline void save_pg_dir(void)
 486 {
 487 }
 488 #endif /* !CONFIG_ACPI_SLEEP */
 489
 490 void zap_low_mappings(void)
 491 {
 492         int i;
 493
 494         /*
 495          * Zap initial low-memory mappings.
 496          *
 497          * Note that "pgd_clear()" doesn't do it for
 498          * us, because pgd_clear() is a no-op on i386.
 499          */
 500         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 501 #ifdef CONFIG_X86_PAE
 502                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 503 #else
 504                 set_pgd(swapper_pg_dir+i, __pgd(0));
 505 #endif
 506         }
 507         flush_tlb_all();
 508 }
 509
 510 int nx_enabled;
 511
 512 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_NX;
 513 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 514
 515 #ifdef CONFIG_X86_PAE
 516
 517 static int disable_nx __initdata;
 518
 519 /*
 520  * noexec = on|off
 521  *
 522  * Control non executable mappings.
 523  *
 524  * on      Enable
 525  * off     Disable
 526  */
 527 static int __init noexec_setup(char *str)
 528 {
 529         if (!str || !strcmp(str, "on")) {
 530                 if (cpu_has_nx) {
 531                         __supported_pte_mask |= _PAGE_NX;
 532                         disable_nx = 0;
 533                 }
 534         } else {
 535                 if (!strcmp(str, "off")) {
 536                         disable_nx = 1;
 537                         __supported_pte_mask &= ~_PAGE_NX;
 538                 } else {
 539                         return -EINVAL;
 540                 }
 541         }
 542
 543         return 0;
 544 }
 545 early_param("noexec", noexec_setup);
 546
 547 static void __init set_nx(void)
 548 {
 549         unsigned int v[4], l, h;
 550
 551         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 552                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 553
 554                 if ((v[3] & (1 << 20)) && !disable_nx) {
 555                         rdmsr(MSR_EFER, l, h);
 556                         l |= EFER_NX;
 557                         wrmsr(MSR_EFER, l, h);
 558                         nx_enabled = 1;
 559                         __supported_pte_mask |= _PAGE_NX;
 560                 }
 561         }
 562 }
 563 #endif
 564
 565 /* user-defined highmem size */
 566 static unsigned int highmem_pages = -1;
 567
 568 /*
 569  * highmem=size forces highmem to be exactly 'size' bytes.
 570  * This works even on boxes that have no highmem otherwise.
 571  * This also works to reduce highmem size on bigger boxes.
 572  */
 573 static int __init parse_highmem(char *arg)
 574 {
 575         if (!arg)
 576                 return -EINVAL;
 577
 578         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 579         return 0;
 580 }
 581 early_param("highmem", parse_highmem);
 582
 583 /*
 584  * Determine low and high memory ranges:
 585  */
 586 void __init find_low_pfn_range(void)
 587 {
 588         /* it could update max_pfn */
 589
 590         /* max_low_pfn is 0, we already have early_res support */
 591
 592         max_low_pfn = max_pfn;
 593         if (max_low_pfn > MAXMEM_PFN) {
 594                 if (highmem_pages == -1)
 595                         highmem_pages = max_pfn - MAXMEM_PFN;
 596                 if (highmem_pages + MAXMEM_PFN < max_pfn)
 597                         max_pfn = MAXMEM_PFN + highmem_pages;
 598                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
 599                         printk(KERN_WARNING "only %luMB highmem pages "
 600                                 "available, ignoring highmem size of %uMB.\n",
 601                                 pages_to_mb(max_pfn - MAXMEM_PFN),
 602                                 pages_to_mb(highmem_pages));
 603                         highmem_pages = 0;
 604                 }
 605                 max_low_pfn = MAXMEM_PFN;
 606 #ifndef CONFIG_HIGHMEM
 607                 /* Maximum memory usable is what is directly addressable */
 608                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
 609                                         MAXMEM>>20);
 610                 if (max_pfn > MAX_NONPAE_PFN)
 611                         printk(KERN_WARNING
 612                                  "Use a HIGHMEM64G enabled kernel.\n");
 613                 else
 614                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 615                 max_pfn = MAXMEM_PFN;
 616 #else /* !CONFIG_HIGHMEM */
 617 #ifndef CONFIG_HIGHMEM64G
 618                 if (max_pfn > MAX_NONPAE_PFN) {
 619                         max_pfn = MAX_NONPAE_PFN;
 620                         printk(KERN_WARNING "Warning only 4GB will be used."
 621                                 "Use a HIGHMEM64G enabled kernel.\n");
 622                 }
 623 #endif /* !CONFIG_HIGHMEM64G */
 624 #endif /* !CONFIG_HIGHMEM */
 625         } else {
 626                 if (highmem_pages == -1)
 627                         highmem_pages = 0;
 628 #ifdef CONFIG_HIGHMEM
 629                 if (highmem_pages >= max_pfn) {
 630                         printk(KERN_ERR "highmem size specified (%uMB) is "
 631                                 "bigger than pages available (%luMB)!.\n",
 632                                 pages_to_mb(highmem_pages),
 633                                 pages_to_mb(max_pfn));
 634                         highmem_pages = 0;
 635                 }
 636                 if (highmem_pages) {
 637                         if (max_low_pfn - highmem_pages <
 638                             64*1024*1024/PAGE_SIZE){
 639                                 printk(KERN_ERR "highmem size %uMB results in "
 640                                 "smaller than 64MB lowmem, ignoring it.\n"
 641                                         , pages_to_mb(highmem_pages));
 642                                 highmem_pages = 0;
 643                         }
 644                         max_low_pfn -= highmem_pages;
 645                 }
 646 #else
 647                 if (highmem_pages)
 648                         printk(KERN_ERR "ignoring highmem size on non-highmem"
 649                                         " kernel!\n");
 650 #endif
 651         }
 652 }
 653
 654 #ifndef CONFIG_NEED_MULTIPLE_NODES
 655 void __init initmem_init(unsigned long start_pfn,
 656                                   unsigned long end_pfn)
 657 {
 658 #ifdef CONFIG_HIGHMEM
 659         highstart_pfn = highend_pfn = max_pfn;
 660         if (max_pfn > max_low_pfn)
 661                 highstart_pfn = max_low_pfn;
 662         memory_present(0, 0, highend_pfn);
 663         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 664                 pages_to_mb(highend_pfn - highstart_pfn));
 665         num_physpages = highend_pfn;
 666         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 667 #else
 668         memory_present(0, 0, max_low_pfn);
 669         num_physpages = max_low_pfn;
 670         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 671 #endif
 672 #ifdef CONFIG_FLATMEM
 673         max_mapnr = num_physpages;
 674 #endif
 675         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 676                         pages_to_mb(max_low_pfn));
 677
 678         setup_bootmem_allocator();
 679 }
 680
 681 void __init zone_sizes_init(void)
 682 {
 683         unsigned long max_zone_pfns[MAX_NR_ZONES];
 684         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 685         max_zone_pfns[ZONE_DMA] =
 686                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 687         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 688         remove_all_active_ranges();
 689 #ifdef CONFIG_HIGHMEM
 690         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 691         e820_register_active_regions(0, 0, highend_pfn);
 692 #else
 693         e820_register_active_regions(0, 0, max_low_pfn);
 694 #endif
 695
 696         free_area_init_nodes(max_zone_pfns);
 697 }
 698 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 699
 700 void __init setup_bootmem_allocator(void)
 701 {
 702         int i;
 703         unsigned long bootmap_size, bootmap;
 704         /*
 705          * Initialize the boot-time allocator (with low memory only):
 706          */
 707         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 708         bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT,
 709                                  max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 710                                  PAGE_SIZE);
 711         if (bootmap == -1L)
 712                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 713         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 714
 715         /* don't touch min_low_pfn */
 716         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
 717                                          min_low_pfn, max_low_pfn);
 718         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 719                  max_pfn_mapped<<PAGE_SHIFT);
 720         printk(KERN_INFO "  low ram: %08lx - %08lx\n",
 721                  min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT);
 722         printk(KERN_INFO "  bootmap %08lx - %08lx\n",
 723                  bootmap, bootmap + bootmap_size);
 724         for_each_online_node(i)
 725                 free_bootmem_with_active_regions(i, max_low_pfn);
 726         early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
 727
 728         after_init_bootmem = 1;
 729 }
 730
 731 static void __init find_early_table_space(unsigned long end)
 732 {
 733         unsigned long puds, pmds, ptes, tables, start;
 734
 735         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 736         tables = PAGE_ALIGN(puds * sizeof(pud_t));
 737
 738         pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 739         tables += PAGE_ALIGN(pmds * sizeof(pmd_t));
 740
 741         if (cpu_has_pse) {
 742                 unsigned long extra;
 743
 744                 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
 745                 extra += PMD_SIZE;
 746                 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
 747         } else
 748                 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
 749
 750         tables += PAGE_ALIGN(ptes * sizeof(pte_t));
 751
 752         /* for fixmap */
 753         tables += PAGE_SIZE * 2;
 754
 755         /*
 756          * RED-PEN putting page tables only on node 0 could
 757          * cause a hotspot and fill up ZONE_DMA. The page tables
 758          * need roughly 0.5KB per GB.
 759          */
 760         start = 0x7000;
 761         table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
 762                                         tables, PAGE_SIZE);
 763         if (table_start == -1UL)
 764                 panic("Cannot find space for the kernel page tables");
 765
 766         table_start >>= PAGE_SHIFT;
 767         table_end = table_start;
 768         table_top = table_start + (tables>>PAGE_SHIFT);
 769
 770         printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
 771                 end, table_start << PAGE_SHIFT,
 772                 (table_start << PAGE_SHIFT) + tables);
 773 }
 774
 775 unsigned long __init_refok init_memory_mapping(unsigned long start,
 776                                                 unsigned long end)
 777 {
 778         pgd_t *pgd_base = swapper_pg_dir;
 779         unsigned long start_pfn, end_pfn;
 780         unsigned long big_page_start;
 781
 782         /*
 783          * Find space for the kernel direct mapping tables.
 784          */
 785         if (!after_init_bootmem)
 786                 find_early_table_space(end);
 787
 788 #ifdef CONFIG_X86_PAE
 789         set_nx();
 790         if (nx_enabled)
 791                 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
 792 #endif
 793
 794         /* Enable PSE if available */
 795         if (cpu_has_pse)
 796                 set_in_cr4(X86_CR4_PSE);
 797
 798         /* Enable PGE if available */
 799         if (cpu_has_pge) {
 800                 set_in_cr4(X86_CR4_PGE);
 801                 __PAGE_KERNEL |= _PAGE_GLOBAL;
 802                 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
 803         }
 804
 805         /*
 806          * Don't use a large page for the first 2/4MB of memory
 807          * because there are often fixed size MTRRs in there
 808          * and overlapping MTRRs into large pages can cause
 809          * slowdowns.
 810          */
 811         big_page_start = PMD_SIZE;
 812
 813         if (start < big_page_start) {
 814                 start_pfn = start >> PAGE_SHIFT;
 815                 end_pfn = min(big_page_start>>PAGE_SHIFT, end>>PAGE_SHIFT);
 816         } else {
 817                 /* head is not big page alignment ? */
 818                 start_pfn = start >> PAGE_SHIFT;
 819                 end_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
 820                                  << (PMD_SHIFT - PAGE_SHIFT);
 821         }
 822         if (start_pfn < end_pfn)
 823                 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, 0);
 824
 825         /* big page range */
 826         start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
 827                          << (PMD_SHIFT - PAGE_SHIFT);
 828         if (start_pfn < (big_page_start >> PAGE_SHIFT))
 829                 start_pfn =  big_page_start >> PAGE_SHIFT;
 830         end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 831         if (start_pfn < end_pfn)
 832                 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn,
 833                                                 cpu_has_pse);
 834
 835         /* tail is not big page alignment ? */
 836         start_pfn = end_pfn;
 837         if (start_pfn > (big_page_start>>PAGE_SHIFT)) {
 838                 end_pfn = end >> PAGE_SHIFT;
 839                 if (start_pfn < end_pfn)
 840                         kernel_physical_mapping_init(pgd_base, start_pfn,
 841                                                          end_pfn, 0);
 842         }
 843
 844         early_ioremap_page_table_range_init(pgd_base);
 845
 846         load_cr3(swapper_pg_dir);
 847
 848         __flush_tlb_all();
 849
 850         if (!after_init_bootmem)
 851                 reserve_early(table_start << PAGE_SHIFT,
 852                                  table_end << PAGE_SHIFT, "PGTABLE");
 853
 854         return end >> PAGE_SHIFT;
 855 }
 856
 857
 858 /*
 859  * paging_init() sets up the page tables - note that the first 8MB are
 860  * already mapped by head.S.
 861  *
 862  * This routines also unmaps the page at virtual kernel address 0, so
 863  * that we can trap those pesky NULL-reference errors in the kernel.
 864  */
 865 void __init paging_init(void)
 866 {
 867         pagetable_init();
 868
 869         __flush_tlb_all();
 870
 871         kmap_init();
 872
 873         /*
 874          * NOTE: at this point the bootmem allocator is fully available.
 875          */
 876         sparse_init();
 877         zone_sizes_init();
 878
 879         paravirt_post_allocator_init();
 880 }
 881
 882 /*
 883  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 884  * and also on some strange 486's. All 586+'s are OK. This used to involve
 885  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 886  * switch to using exceptions got rid of all that.
 887  */
 888 static void __init test_wp_bit(void)
 889 {
 890         printk(KERN_INFO
 891   "Checking if this processor honours the WP bit even in supervisor mode...");
 892
 893         /* Any page-aligned address will do, the test is non-destructive */
 894         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 895         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 896         clear_fixmap(FIX_WP_TEST);
 897
 898         if (!boot_cpu_data.wp_works_ok) {
 899                 printk(KERN_CONT "No.\n");
 900 #ifdef CONFIG_X86_WP_WORKS_OK
 901                 panic(
 902   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 903 #endif
 904         } else {
 905                 printk(KERN_CONT "Ok.\n");
 906         }
 907 }
 908
 909 static struct kcore_list kcore_mem, kcore_vmalloc;
 910
 911 void __init mem_init(void)
 912 {
 913         int codesize, reservedpages, datasize, initsize;
 914         int tmp;
 915
 916 #ifdef CONFIG_FLATMEM
 917         BUG_ON(!mem_map);
 918 #endif
 919         /* this will put all low memory onto the freelists */
 920         totalram_pages += free_all_bootmem();
 921
 922         reservedpages = 0;
 923         for (tmp = 0; tmp < max_low_pfn; tmp++)
 924                 /*
 925                  * Only count reserved RAM pages:
 926                  */
 927                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 928                         reservedpages++;
 929
 930         set_highmem_pages_init();
 931
 932         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 933         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 934         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 935
 936         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 937         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 938                    VMALLOC_END-VMALLOC_START);
 939
 940         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 941                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 942                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 943                 num_physpages << (PAGE_SHIFT-10),
 944                 codesize >> 10,
 945                 reservedpages << (PAGE_SHIFT-10),
 946                 datasize >> 10,
 947                 initsize >> 10,
 948                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 949                );
 950
 951         printk(KERN_INFO "virtual kernel memory layout:\n"
 952                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 953 #ifdef CONFIG_HIGHMEM
 954                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 955 #endif
 956                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 957                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 958                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 959                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 960                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 961                 FIXADDR_START, FIXADDR_TOP,
 962                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 963
 964 #ifdef CONFIG_HIGHMEM
 965                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 966                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 967 #endif
 968
 969                 VMALLOC_START, VMALLOC_END,
 970                 (VMALLOC_END - VMALLOC_START) >> 20,
 971
 972                 (unsigned long)__va(0), (unsigned long)high_memory,
 973                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 974
 975                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 976                 ((unsigned long)&__init_end -
 977                  (unsigned long)&__init_begin) >> 10,
 978
 979                 (unsigned long)&_etext, (unsigned long)&_edata,
 980                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 981
 982                 (unsigned long)&_text, (unsigned long)&_etext,
 983                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 984
 985 #ifdef CONFIG_HIGHMEM
 986         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 987         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 988 #endif
 989         BUG_ON(VMALLOC_START                            > VMALLOC_END);
 990         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 991
 992         if (boot_cpu_data.wp_works_ok < 0)
 993                 test_wp_bit();
 994
 995         cpa_init();
 996         save_pg_dir();
 997         zap_low_mappings();
 998 }
 999
1000 #ifdef CONFIG_MEMORY_HOTPLUG
1001 int arch_add_memory(int nid, u64 start, u64 size)
1002 {
1003         struct pglist_data *pgdata = NODE_DATA(nid);
1004         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1005         unsigned long start_pfn = start >> PAGE_SHIFT;
1006         unsigned long nr_pages = size >> PAGE_SHIFT;
1007
1008         return __add_pages(zone, start_pfn, nr_pages);
1009 }
1010 #endif
1011
1012 /*
1013  * This function cannot be __init, since exceptions don't work in that
1014  * section.  Put this after the callers, so that it cannot be inlined.
1015  */
1016 static noinline int do_test_wp_bit(void)
1017 {
1018         char tmp_reg;
1019         int flag;
1020
1021         __asm__ __volatile__(
1022                 "       movb %0, %1     \n"
1023                 "1:     movb %1, %0     \n"
1024                 "       xorl %2, %2     \n"
1025                 "2:                     \n"
1026                 _ASM_EXTABLE(1b,2b)
1027                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1028                  "=q" (tmp_reg),
1029                  "=r" (flag)
1030                 :"2" (1)
1031                 :"memory");
1032
1033         return flag;
1034 }
1035
1036 #ifdef CONFIG_DEBUG_RODATA
1037 const int rodata_test_data = 0xC3;
1038 EXPORT_SYMBOL_GPL(rodata_test_data);
1039
1040 void mark_rodata_ro(void)
1041 {
1042         unsigned long start = PFN_ALIGN(_text);
1043         unsigned long size = PFN_ALIGN(_etext) - start;
1044
1045         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1046         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1047                 size >> 10);
1048
1049 #ifdef CONFIG_CPA_DEBUG
1050         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1051                 start, start+size);
1052         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1053
1054         printk(KERN_INFO "Testing CPA: write protecting again\n");
1055         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1056 #endif
1057         start += size;
1058         size = (unsigned long)__end_rodata - start;
1059         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1060         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1061                 size >> 10);
1062         rodata_test();
1063
1064 #ifdef CONFIG_CPA_DEBUG
1065         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1066         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1067
1068         printk(KERN_INFO "Testing CPA: write protecting again\n");
1069         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1070 #endif
1071 }
1072 #endif
1073
1074 void free_init_pages(char *what, unsigned long begin, unsigned long end)
1075 {
1076 #ifdef CONFIG_DEBUG_PAGEALLOC
1077         /*
1078          * If debugging page accesses then do not free this memory but
1079          * mark them not present - any buggy init-section access will
1080          * create a kernel page fault:
1081          */
1082         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
1083                 begin, PAGE_ALIGN(end));
1084         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
1085 #else
1086         unsigned long addr;
1087
1088         /*
1089          * We just marked the kernel text read only above, now that
1090          * we are going to free part of that, we need to make that
1091          * writeable first.
1092          */
1093         set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
1094
1095         for (addr = begin; addr < end; addr += PAGE_SIZE) {
1096                 ClearPageReserved(virt_to_page(addr));
1097                 init_page_count(virt_to_page(addr));
1098                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
1099                 free_page(addr);
1100                 totalram_pages++;
1101         }
1102         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
1103 #endif
1104 }
1105
1106 void free_initmem(void)
1107 {
1108         free_init_pages("unused kernel memory",
1109                         (unsigned long)(&__init_begin),
1110                         (unsigned long)(&__init_end));
1111 }
1112
1113 #ifdef CONFIG_BLK_DEV_INITRD
1114 void free_initrd_mem(unsigned long start, unsigned long end)
1115 {
1116         free_init_pages("initrd memory", start, end);
1117 }
1118 #endif
1119
1120 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1121                                    int flags)
1122 {
1123         return reserve_bootmem(phys, len, flags);
1124 }