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