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