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