SAFE public projects git trees. - safe/jmp/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 void __init native_pagetable_setup_start(pgd_t *base)
 387 {
 388         unsigned long pfn, va;
 389         pgd_t *pgd;
 390         pud_t *pud;
 391         pmd_t *pmd;
 392         pte_t *pte;
 393
 394         /*
 395          * Remove any mappings which extend past the end of physical
 396          * memory from the boot time page table:
 397          */
 398         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 399                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 400                 pgd = base + pgd_index(va);
 401                 if (!pgd_present(*pgd))
 402                         break;
 403
 404                 pud = pud_offset(pgd, va);
 405                 pmd = pmd_offset(pud, va);
 406                 if (!pmd_present(*pmd))
 407                         break;
 408
 409                 pte = pte_offset_kernel(pmd, va);
 410                 if (!pte_present(*pte))
 411                         break;
 412
 413                 pte_clear(NULL, va, pte);
 414         }
 415         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 416 }
 417
 418 void __init native_pagetable_setup_done(pgd_t *base)
 419 {
 420 }
 421
 422 /*
 423  * Build a proper pagetable for the kernel mappings.  Up until this
 424  * point, we've been running on some set of pagetables constructed by
 425  * the boot process.
 426  *
 427  * If we're booting on native hardware, this will be a pagetable
 428  * constructed in arch/x86/kernel/head_32.S.  The root of the
 429  * pagetable will be swapper_pg_dir.
 430  *
 431  * If we're booting paravirtualized under a hypervisor, then there are
 432  * more options: we may already be running PAE, and the pagetable may
 433  * or may not be based in swapper_pg_dir.  In any case,
 434  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 435  * appropriately for the rest of the initialization to work.
 436  *
 437  * In general, pagetable_init() assumes that the pagetable may already
 438  * be partially populated, and so it avoids stomping on any existing
 439  * mappings.
 440  */
 441 static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base)
 442 {
 443         unsigned long vaddr, end;
 444
 445         /*
 446          * Fixed mappings, only the page table structure has to be
 447          * created - mappings will be set by set_fixmap():
 448          */
 449         early_ioremap_clear();
 450         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 451         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 452         page_table_range_init(vaddr, end, pgd_base);
 453         early_ioremap_reset();
 454 }
 455
 456 static void __init pagetable_init(void)
 457 {
 458         pgd_t *pgd_base = swapper_pg_dir;
 459
 460         paravirt_pagetable_setup_start(pgd_base);
 461
 462         permanent_kmaps_init(pgd_base);
 463
 464         paravirt_pagetable_setup_done(pgd_base);
 465 }
 466
 467 #ifdef CONFIG_ACPI_SLEEP
 468 /*
 469  * ACPI suspend needs this for resume, because things like the intel-agp
 470  * driver might have split up a kernel 4MB mapping.
 471  */
 472 char swsusp_pg_dir[PAGE_SIZE]
 473         __attribute__ ((aligned(PAGE_SIZE)));
 474
 475 static inline void save_pg_dir(void)
 476 {
 477         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 478 }
 479 #else /* !CONFIG_ACPI_SLEEP */
 480 static inline void save_pg_dir(void)
 481 {
 482 }
 483 #endif /* !CONFIG_ACPI_SLEEP */
 484
 485 void zap_low_mappings(void)
 486 {
 487         int i;
 488
 489         /*
 490          * Zap initial low-memory mappings.
 491          *
 492          * Note that "pgd_clear()" doesn't do it for
 493          * us, because pgd_clear() is a no-op on i386.
 494          */
 495         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 496 #ifdef CONFIG_X86_PAE
 497                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 498 #else
 499                 set_pgd(swapper_pg_dir+i, __pgd(0));
 500 #endif
 501         }
 502         flush_tlb_all();
 503 }
 504
 505 int nx_enabled;
 506
 507 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL);
 508 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 509
 510 #ifdef CONFIG_X86_PAE
 511
 512 static int disable_nx __initdata;
 513
 514 /*
 515  * noexec = on|off
 516  *
 517  * Control non executable mappings.
 518  *
 519  * on      Enable
 520  * off     Disable
 521  */
 522 static int __init noexec_setup(char *str)
 523 {
 524         if (!str || !strcmp(str, "on")) {
 525                 if (cpu_has_nx) {
 526                         __supported_pte_mask |= _PAGE_NX;
 527                         disable_nx = 0;
 528                 }
 529         } else {
 530                 if (!strcmp(str, "off")) {
 531                         disable_nx = 1;
 532                         __supported_pte_mask &= ~_PAGE_NX;
 533                 } else {
 534                         return -EINVAL;
 535                 }
 536         }
 537
 538         return 0;
 539 }
 540 early_param("noexec", noexec_setup);
 541
 542 static void __init set_nx(void)
 543 {
 544         unsigned int v[4], l, h;
 545
 546         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 547                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 548
 549                 if ((v[3] & (1 << 20)) && !disable_nx) {
 550                         rdmsr(MSR_EFER, l, h);
 551                         l |= EFER_NX;
 552                         wrmsr(MSR_EFER, l, h);
 553                         nx_enabled = 1;
 554                         __supported_pte_mask |= _PAGE_NX;
 555                 }
 556         }
 557 }
 558 #endif
 559
 560 /* user-defined highmem size */
 561 static unsigned int highmem_pages = -1;
 562
 563 /*
 564  * highmem=size forces highmem to be exactly 'size' bytes.
 565  * This works even on boxes that have no highmem otherwise.
 566  * This also works to reduce highmem size on bigger boxes.
 567  */
 568 static int __init parse_highmem(char *arg)
 569 {
 570         if (!arg)
 571                 return -EINVAL;
 572
 573         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 574         return 0;
 575 }
 576 early_param("highmem", parse_highmem);
 577
 578 /*
 579  * Determine low and high memory ranges:
 580  */
 581 void __init find_low_pfn_range(void)
 582 {
 583         /* it could update max_pfn */
 584
 585         /* max_low_pfn is 0, we already have early_res support */
 586
 587         max_low_pfn = max_pfn;
 588         if (max_low_pfn > MAXMEM_PFN) {
 589                 if (highmem_pages == -1)
 590                         highmem_pages = max_pfn - MAXMEM_PFN;
 591                 if (highmem_pages + MAXMEM_PFN < max_pfn)
 592                         max_pfn = MAXMEM_PFN + highmem_pages;
 593                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
 594                         printk(KERN_WARNING "only %luMB highmem pages "
 595                                 "available, ignoring highmem size of %uMB.\n",
 596                                 pages_to_mb(max_pfn - MAXMEM_PFN),
 597                                 pages_to_mb(highmem_pages));
 598                         highmem_pages = 0;
 599                 }
 600                 max_low_pfn = MAXMEM_PFN;
 601 #ifndef CONFIG_HIGHMEM
 602                 /* Maximum memory usable is what is directly addressable */
 603                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
 604                                         MAXMEM>>20);
 605                 if (max_pfn > MAX_NONPAE_PFN)
 606                         printk(KERN_WARNING
 607                                  "Use a HIGHMEM64G enabled kernel.\n");
 608                 else
 609                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 610                 max_pfn = MAXMEM_PFN;
 611 #else /* !CONFIG_HIGHMEM */
 612 #ifndef CONFIG_HIGHMEM64G
 613                 if (max_pfn > MAX_NONPAE_PFN) {
 614                         max_pfn = MAX_NONPAE_PFN;
 615                         printk(KERN_WARNING "Warning only 4GB will be used."
 616                                 "Use a HIGHMEM64G enabled kernel.\n");
 617                 }
 618 #endif /* !CONFIG_HIGHMEM64G */
 619 #endif /* !CONFIG_HIGHMEM */
 620         } else {
 621                 if (highmem_pages == -1)
 622                         highmem_pages = 0;
 623 #ifdef CONFIG_HIGHMEM
 624                 if (highmem_pages >= max_pfn) {
 625                         printk(KERN_ERR "highmem size specified (%uMB) is "
 626                                 "bigger than pages available (%luMB)!.\n",
 627                                 pages_to_mb(highmem_pages),
 628                                 pages_to_mb(max_pfn));
 629                         highmem_pages = 0;
 630                 }
 631                 if (highmem_pages) {
 632                         if (max_low_pfn - highmem_pages <
 633                             64*1024*1024/PAGE_SIZE){
 634                                 printk(KERN_ERR "highmem size %uMB results in "
 635                                 "smaller than 64MB lowmem, ignoring it.\n"
 636                                         , pages_to_mb(highmem_pages));
 637                                 highmem_pages = 0;
 638                         }
 639                         max_low_pfn -= highmem_pages;
 640                 }
 641 #else
 642                 if (highmem_pages)
 643                         printk(KERN_ERR "ignoring highmem size on non-highmem"
 644                                         " kernel!\n");
 645 #endif
 646         }
 647 }
 648
 649 #ifndef CONFIG_NEED_MULTIPLE_NODES
 650 void __init initmem_init(unsigned long start_pfn,
 651                                   unsigned long end_pfn)
 652 {
 653 #ifdef CONFIG_HIGHMEM
 654         highstart_pfn = highend_pfn = max_pfn;
 655         if (max_pfn > max_low_pfn)
 656                 highstart_pfn = max_low_pfn;
 657         memory_present(0, 0, highend_pfn);
 658         e820_register_active_regions(0, 0, highend_pfn);
 659         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 660                 pages_to_mb(highend_pfn - highstart_pfn));
 661         num_physpages = highend_pfn;
 662         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 663 #else
 664         memory_present(0, 0, max_low_pfn);
 665         e820_register_active_regions(0, 0, max_low_pfn);
 666         num_physpages = max_low_pfn;
 667         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 668 #endif
 669 #ifdef CONFIG_FLATMEM
 670         max_mapnr = num_physpages;
 671 #endif
 672         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 673                         pages_to_mb(max_low_pfn));
 674
 675         setup_bootmem_allocator();
 676 }
 677 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 678
 679 static void __init zone_sizes_init(void)
 680 {
 681         unsigned long max_zone_pfns[MAX_NR_ZONES];
 682         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 683         max_zone_pfns[ZONE_DMA] =
 684                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 685         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 686 #ifdef CONFIG_HIGHMEM
 687         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 688 #endif
 689
 690         free_area_init_nodes(max_zone_pfns);
 691 }
 692
 693 void __init setup_bootmem_allocator(void)
 694 {
 695         int i;
 696         unsigned long bootmap_size, bootmap;
 697         /*
 698          * Initialize the boot-time allocator (with low memory only):
 699          */
 700         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 701         bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT,
 702                                  max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 703                                  PAGE_SIZE);
 704         if (bootmap == -1L)
 705                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 706         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 707
 708         /* don't touch min_low_pfn */
 709         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
 710                                          min_low_pfn, max_low_pfn);
 711         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 712                  max_pfn_mapped<<PAGE_SHIFT);
 713         printk(KERN_INFO "  low ram: %08lx - %08lx\n",
 714                  min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT);
 715         printk(KERN_INFO "  bootmap %08lx - %08lx\n",
 716                  bootmap, bootmap + bootmap_size);
 717         for_each_online_node(i)
 718                 free_bootmem_with_active_regions(i, max_low_pfn);
 719         early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
 720
 721         after_init_bootmem = 1;
 722 }
 723
 724 static void __init find_early_table_space(unsigned long end)
 725 {
 726         unsigned long puds, pmds, ptes, tables, start;
 727
 728         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 729         tables = PAGE_ALIGN(puds * sizeof(pud_t));
 730
 731         pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 732         tables += PAGE_ALIGN(pmds * sizeof(pmd_t));
 733
 734         if (cpu_has_pse) {
 735                 unsigned long extra;
 736
 737                 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
 738                 extra += PMD_SIZE;
 739                 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
 740         } else
 741                 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
 742
 743         tables += PAGE_ALIGN(ptes * sizeof(pte_t));
 744
 745         /* for fixmap */
 746         tables += PAGE_SIZE * 2;
 747
 748         /*
 749          * RED-PEN putting page tables only on node 0 could
 750          * cause a hotspot and fill up ZONE_DMA. The page tables
 751          * need roughly 0.5KB per GB.
 752          */
 753         start = 0x7000;
 754         table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
 755                                         tables, PAGE_SIZE);
 756         if (table_start == -1UL)
 757                 panic("Cannot find space for the kernel page tables");
 758
 759         table_start >>= PAGE_SHIFT;
 760         table_end = table_start;
 761         table_top = table_start + (tables>>PAGE_SHIFT);
 762
 763         printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
 764                 end, table_start << PAGE_SHIFT,
 765                 (table_start << PAGE_SHIFT) + tables);
 766 }
 767
 768 unsigned long __init_refok init_memory_mapping(unsigned long start,
 769                                                 unsigned long end)
 770 {
 771         pgd_t *pgd_base = swapper_pg_dir;
 772         unsigned long start_pfn, end_pfn;
 773         unsigned long big_page_start;
 774
 775         /*
 776          * Find space for the kernel direct mapping tables.
 777          */
 778         if (!after_init_bootmem)
 779                 find_early_table_space(end);
 780
 781 #ifdef CONFIG_X86_PAE
 782         set_nx();
 783         if (nx_enabled)
 784                 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
 785 #endif
 786
 787         /* Enable PSE if available */
 788         if (cpu_has_pse)
 789                 set_in_cr4(X86_CR4_PSE);
 790
 791         /* Enable PGE if available */
 792         if (cpu_has_pge) {
 793                 set_in_cr4(X86_CR4_PGE);
 794                 __supported_pte_mask |= _PAGE_GLOBAL;
 795         }
 796
 797         /*
 798          * Don't use a large page for the first 2/4MB of memory
 799          * because there are often fixed size MTRRs in there
 800          * and overlapping MTRRs into large pages can cause
 801          * slowdowns.
 802          */
 803         big_page_start = PMD_SIZE;
 804
 805         if (start < big_page_start) {
 806                 start_pfn = start >> PAGE_SHIFT;
 807                 end_pfn = min(big_page_start>>PAGE_SHIFT, end>>PAGE_SHIFT);
 808         } else {
 809                 /* head is not big page alignment ? */
 810                 start_pfn = start >> PAGE_SHIFT;
 811                 end_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
 812                                  << (PMD_SHIFT - PAGE_SHIFT);
 813         }
 814         if (start_pfn < end_pfn)
 815                 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, 0);
 816
 817         /* big page range */
 818         start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
 819                          << (PMD_SHIFT - PAGE_SHIFT);
 820         if (start_pfn < (big_page_start >> PAGE_SHIFT))
 821                 start_pfn =  big_page_start >> PAGE_SHIFT;
 822         end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 823         if (start_pfn < end_pfn)
 824                 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn,
 825                                                 cpu_has_pse);
 826
 827         /* tail is not big page alignment ? */
 828         start_pfn = end_pfn;
 829         if (start_pfn > (big_page_start>>PAGE_SHIFT)) {
 830                 end_pfn = end >> PAGE_SHIFT;
 831                 if (start_pfn < end_pfn)
 832                         kernel_physical_mapping_init(pgd_base, start_pfn,
 833                                                          end_pfn, 0);
 834         }
 835
 836         early_ioremap_page_table_range_init(pgd_base);
 837
 838         load_cr3(swapper_pg_dir);
 839
 840         __flush_tlb_all();
 841
 842         if (!after_init_bootmem)
 843                 reserve_early(table_start << PAGE_SHIFT,
 844                                  table_end << PAGE_SHIFT, "PGTABLE");
 845
 846         return end >> PAGE_SHIFT;
 847 }
 848
 849
 850 /*
 851  * paging_init() sets up the page tables - note that the first 8MB are
 852  * already mapped by head.S.
 853  *
 854  * This routines also unmaps the page at virtual kernel address 0, so
 855  * that we can trap those pesky NULL-reference errors in the kernel.
 856  */
 857 void __init paging_init(void)
 858 {
 859         pagetable_init();
 860
 861         __flush_tlb_all();
 862
 863         kmap_init();
 864
 865         /*
 866          * NOTE: at this point the bootmem allocator is fully available.
 867          */
 868         sparse_init();
 869         zone_sizes_init();
 870
 871         paravirt_post_allocator_init();
 872 }
 873
 874 /*
 875  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 876  * and also on some strange 486's. All 586+'s are OK. This used to involve
 877  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 878  * switch to using exceptions got rid of all that.
 879  */
 880 static void __init test_wp_bit(void)
 881 {
 882         printk(KERN_INFO
 883   "Checking if this processor honours the WP bit even in supervisor mode...");
 884
 885         /* Any page-aligned address will do, the test is non-destructive */
 886         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 887         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 888         clear_fixmap(FIX_WP_TEST);
 889
 890         if (!boot_cpu_data.wp_works_ok) {
 891                 printk(KERN_CONT "No.\n");
 892 #ifdef CONFIG_X86_WP_WORKS_OK
 893                 panic(
 894   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 895 #endif
 896         } else {
 897                 printk(KERN_CONT "Ok.\n");
 898         }
 899 }
 900
 901 static struct kcore_list kcore_mem, kcore_vmalloc;
 902
 903 void __init mem_init(void)
 904 {
 905         int codesize, reservedpages, datasize, initsize;
 906         int tmp;
 907
 908 #ifdef CONFIG_FLATMEM
 909         BUG_ON(!mem_map);
 910 #endif
 911         /* this will put all low memory onto the freelists */
 912         totalram_pages += free_all_bootmem();
 913
 914         reservedpages = 0;
 915         for (tmp = 0; tmp < max_low_pfn; tmp++)
 916                 /*
 917                  * Only count reserved RAM pages:
 918                  */
 919                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 920                         reservedpages++;
 921
 922         set_highmem_pages_init();
 923
 924         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 925         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 926         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 927
 928         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 929         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 930                    VMALLOC_END-VMALLOC_START);
 931
 932         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 933                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 934                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 935                 num_physpages << (PAGE_SHIFT-10),
 936                 codesize >> 10,
 937                 reservedpages << (PAGE_SHIFT-10),
 938                 datasize >> 10,
 939                 initsize >> 10,
 940                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 941                );
 942
 943         printk(KERN_INFO "virtual kernel memory layout:\n"
 944                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 945 #ifdef CONFIG_HIGHMEM
 946                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 947 #endif
 948                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 949                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 950                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 951                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 952                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 953                 FIXADDR_START, FIXADDR_TOP,
 954                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 955
 956 #ifdef CONFIG_HIGHMEM
 957                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 958                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 959 #endif
 960
 961                 VMALLOC_START, VMALLOC_END,
 962                 (VMALLOC_END - VMALLOC_START) >> 20,
 963
 964                 (unsigned long)__va(0), (unsigned long)high_memory,
 965                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 966
 967                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 968                 ((unsigned long)&__init_end -
 969                  (unsigned long)&__init_begin) >> 10,
 970
 971                 (unsigned long)&_etext, (unsigned long)&_edata,
 972                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 973
 974                 (unsigned long)&_text, (unsigned long)&_etext,
 975                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 976
 977 #ifdef CONFIG_HIGHMEM
 978         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 979         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 980 #endif
 981         BUG_ON(VMALLOC_START                            > VMALLOC_END);
 982         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 983
 984         if (boot_cpu_data.wp_works_ok < 0)
 985                 test_wp_bit();
 986
 987         cpa_init();
 988         save_pg_dir();
 989         zap_low_mappings();
 990 }
 991
 992 #ifdef CONFIG_MEMORY_HOTPLUG
 993 int arch_add_memory(int nid, u64 start, u64 size)
 994 {
 995         struct pglist_data *pgdata = NODE_DATA(nid);
 996         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 997         unsigned long start_pfn = start >> PAGE_SHIFT;
 998         unsigned long nr_pages = size >> PAGE_SHIFT;
 999
1000         return __add_pages(zone, start_pfn, nr_pages);
1001 }
1002 #endif
1003
1004 /*
1005  * This function cannot be __init, since exceptions don't work in that
1006  * section.  Put this after the callers, so that it cannot be inlined.
1007  */
1008 static noinline int do_test_wp_bit(void)
1009 {
1010         char tmp_reg;
1011         int flag;
1012
1013         __asm__ __volatile__(
1014                 "       movb %0, %1     \n"
1015                 "1:     movb %1, %0     \n"
1016                 "       xorl %2, %2     \n"
1017                 "2:                     \n"
1018                 _ASM_EXTABLE(1b,2b)
1019                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1020                  "=q" (tmp_reg),
1021                  "=r" (flag)
1022                 :"2" (1)
1023                 :"memory");
1024
1025         return flag;
1026 }
1027
1028 #ifdef CONFIG_DEBUG_RODATA
1029 const int rodata_test_data = 0xC3;
1030 EXPORT_SYMBOL_GPL(rodata_test_data);
1031
1032 void mark_rodata_ro(void)
1033 {
1034         unsigned long start = PFN_ALIGN(_text);
1035         unsigned long size = PFN_ALIGN(_etext) - start;
1036
1037         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1038         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1039                 size >> 10);
1040
1041 #ifdef CONFIG_CPA_DEBUG
1042         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1043                 start, start+size);
1044         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1045
1046         printk(KERN_INFO "Testing CPA: write protecting again\n");
1047         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1048 #endif
1049         start += size;
1050         size = (unsigned long)__end_rodata - start;
1051         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1052         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1053                 size >> 10);
1054         rodata_test();
1055
1056 #ifdef CONFIG_CPA_DEBUG
1057         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1058         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1059
1060         printk(KERN_INFO "Testing CPA: write protecting again\n");
1061         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1062 #endif
1063 }
1064 #endif
1065
1066 void free_init_pages(char *what, unsigned long begin, unsigned long end)
1067 {
1068 #ifdef CONFIG_DEBUG_PAGEALLOC
1069         /*
1070          * If debugging page accesses then do not free this memory but
1071          * mark them not present - any buggy init-section access will
1072          * create a kernel page fault:
1073          */
1074         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
1075                 begin, PAGE_ALIGN(end));
1076         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
1077 #else
1078         unsigned long addr;
1079
1080         /*
1081          * We just marked the kernel text read only above, now that
1082          * we are going to free part of that, we need to make that
1083          * writeable first.
1084          */
1085         set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
1086
1087         for (addr = begin; addr < end; addr += PAGE_SIZE) {
1088                 ClearPageReserved(virt_to_page(addr));
1089                 init_page_count(virt_to_page(addr));
1090                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
1091                 free_page(addr);
1092                 totalram_pages++;
1093         }
1094         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
1095 #endif
1096 }
1097
1098 void free_initmem(void)
1099 {
1100         free_init_pages("unused kernel memory",
1101                         (unsigned long)(&__init_begin),
1102                         (unsigned long)(&__init_end));
1103 }
1104
1105 #ifdef CONFIG_BLK_DEV_INITRD
1106 void free_initrd_mem(unsigned long start, unsigned long end)
1107 {
1108         free_init_pages("initrd memory", start, end);
1109 }
1110 #endif
1111
1112 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1113                                    int flags)
1114 {
1115         return reserve_bootmem(phys, len, flags);
1116 }