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/pci.h>
  25 #include <linux/pfn.h>
  26 #include <linux/poison.h>
  27 #include <linux/bootmem.h>
  28 #include <linux/slab.h>
  29 #include <linux/proc_fs.h>
  30 #include <linux/memory_hotplug.h>
  31 #include <linux/initrd.h>
  32 #include <linux/cpumask.h>
  33
  34 #include <asm/asm.h>
  35 #include <asm/bios_ebda.h>
  36 #include <asm/processor.h>
  37 #include <asm/system.h>
  38 #include <asm/uaccess.h>
  39 #include <asm/pgtable.h>
  40 #include <asm/dma.h>
  41 #include <asm/fixmap.h>
  42 #include <asm/e820.h>
  43 #include <asm/apic.h>
  44 #include <asm/bugs.h>
  45 #include <asm/tlb.h>
  46 #include <asm/tlbflush.h>
  47 #include <asm/pgalloc.h>
  48 #include <asm/sections.h>
  49 #include <asm/paravirt.h>
  50 #include <asm/setup.h>
  51 #include <asm/cacheflush.h>
  52
  53 unsigned long max_low_pfn_mapped;
  54 unsigned long max_pfn_mapped;
  55
  56 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  57 unsigned long highstart_pfn, highend_pfn;
  58
  59 static noinline int do_test_wp_bit(void);
  60
  61
  62 extern unsigned long __initdata table_start;
  63 extern unsigned long __meminitdata table_end;
  64 extern unsigned long __meminitdata table_top;
  65
  66 static __init void *alloc_low_page(void)
  67 {
  68         unsigned long pfn = table_end++;
  69         void *adr;
  70
  71         if (pfn >= table_top)
  72                 panic("alloc_low_page: ran out of memory");
  73
  74         adr = __va(pfn * PAGE_SIZE);
  75         memset(adr, 0, PAGE_SIZE);
  76         return adr;
  77 }
  78
  79 /*
  80  * Creates a middle page table and puts a pointer to it in the
  81  * given global directory entry. This only returns the gd entry
  82  * in non-PAE compilation mode, since the middle layer is folded.
  83  */
  84 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  85 {
  86         pud_t *pud;
  87         pmd_t *pmd_table;
  88
  89 #ifdef CONFIG_X86_PAE
  90         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
  91                 if (after_bootmem)
  92                         pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
  93                 else
  94                         pmd_table = (pmd_t *)alloc_low_page();
  95                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
  96                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  97                 pud = pud_offset(pgd, 0);
  98                 BUG_ON(pmd_table != pmd_offset(pud, 0));
  99
 100                 return pmd_table;
 101         }
 102 #endif
 103         pud = pud_offset(pgd, 0);
 104         pmd_table = pmd_offset(pud, 0);
 105
 106         return pmd_table;
 107 }
 108
 109 /*
 110  * Create a page table and place a pointer to it in a middle page
 111  * directory entry:
 112  */
 113 static pte_t * __init one_page_table_init(pmd_t *pmd)
 114 {
 115         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
 116                 pte_t *page_table = NULL;
 117
 118                 if (after_bootmem) {
 119 #ifdef CONFIG_DEBUG_PAGEALLOC
 120                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
 121 #endif
 122                         if (!page_table)
 123                                 page_table =
 124                                 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 125                 } else
 126                         page_table = (pte_t *)alloc_low_page();
 127
 128                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
 129                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
 130                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
 131         }
 132
 133         return pte_offset_kernel(pmd, 0);
 134 }
 135
 136 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
 137                                            unsigned long vaddr, pte_t *lastpte)
 138 {
 139 #ifdef CONFIG_HIGHMEM
 140         /*
 141          * Something (early fixmap) may already have put a pte
 142          * page here, which causes the page table allocation
 143          * to become nonlinear. Attempt to fix it, and if it
 144          * is still nonlinear then we have to bug.
 145          */
 146         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
 147         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
 148
 149         if (pmd_idx_kmap_begin != pmd_idx_kmap_end
 150             && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
 151             && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
 152             && ((__pa(pte) >> PAGE_SHIFT) < table_start
 153                 || (__pa(pte) >> PAGE_SHIFT) >= table_end)) {
 154                 pte_t *newpte;
 155                 int i;
 156
 157                 BUG_ON(after_bootmem);
 158                 newpte = alloc_low_page();
 159                 for (i = 0; i < PTRS_PER_PTE; i++)
 160                         set_pte(newpte + i, pte[i]);
 161
 162                 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
 163                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
 164                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
 165                 __flush_tlb_all();
 166
 167                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
 168                 pte = newpte;
 169         }
 170         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
 171                && vaddr > fix_to_virt(FIX_KMAP_END)
 172                && lastpte && lastpte + PTRS_PER_PTE != pte);
 173 #endif
 174         return pte;
 175 }
 176
 177 /*
 178  * This function initializes a certain range of kernel virtual memory
 179  * with new bootmem page tables, everywhere page tables are missing in
 180  * the given range.
 181  *
 182  * NOTE: The pagetables are allocated contiguous on the physical space
 183  * so we can cache the place of the first one and move around without
 184  * checking the pgd every time.
 185  */
 186 static void __init
 187 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 188 {
 189         int pgd_idx, pmd_idx;
 190         unsigned long vaddr;
 191         pgd_t *pgd;
 192         pmd_t *pmd;
 193         pte_t *pte = NULL;
 194
 195         vaddr = start;
 196         pgd_idx = pgd_index(vaddr);
 197         pmd_idx = pmd_index(vaddr);
 198         pgd = pgd_base + pgd_idx;
 199
 200         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 201                 pmd = one_md_table_init(pgd);
 202                 pmd = pmd + pmd_index(vaddr);
 203                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 204                                                         pmd++, pmd_idx++) {
 205                         pte = page_table_kmap_check(one_page_table_init(pmd),
 206                                                     pmd, vaddr, pte);
 207
 208                         vaddr += PMD_SIZE;
 209                 }
 210                 pmd_idx = 0;
 211         }
 212 }
 213
 214 static inline int is_kernel_text(unsigned long addr)
 215 {
 216         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 217                 return 1;
 218         return 0;
 219 }
 220
 221 /*
 222  * This maps the physical memory to kernel virtual address space, a total
 223  * of max_low_pfn pages, by creating page tables starting from address
 224  * PAGE_OFFSET:
 225  */
 226 void __init kernel_physical_mapping_init(unsigned long start_pfn,
 227                                          unsigned long end_pfn,
 228                                          int use_pse)
 229 {
 230         pgd_t *pgd_base = swapper_pg_dir;
 231         int pgd_idx, pmd_idx, pte_ofs;
 232         unsigned long pfn;
 233         pgd_t *pgd;
 234         pmd_t *pmd;
 235         pte_t *pte;
 236         unsigned pages_2m, pages_4k;
 237         int mapping_iter;
 238
 239         /*
 240          * First iteration will setup identity mapping using large/small pages
 241          * based on use_pse, with other attributes same as set by
 242          * the early code in head_32.S
 243          *
 244          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
 245          * as desired for the kernel identity mapping.
 246          *
 247          * This two pass mechanism conforms to the TLB app note which says:
 248          *
 249          *     "Software should not write to a paging-structure entry in a way
 250          *      that would change, for any linear address, both the page size
 251          *      and either the page frame or attributes."
 252          */
 253         mapping_iter = 1;
 254
 255         if (!cpu_has_pse)
 256                 use_pse = 0;
 257
 258 repeat:
 259         pages_2m = pages_4k = 0;
 260         pfn = start_pfn;
 261         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 262         pgd = pgd_base + pgd_idx;
 263         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 264                 pmd = one_md_table_init(pgd);
 265
 266                 if (pfn >= end_pfn)
 267                         continue;
 268 #ifdef CONFIG_X86_PAE
 269                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 270                 pmd += pmd_idx;
 271 #else
 272                 pmd_idx = 0;
 273 #endif
 274                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
 275                      pmd++, pmd_idx++) {
 276                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 277
 278                         /*
 279                          * Map with big pages if possible, otherwise
 280                          * create normal page tables:
 281                          */
 282                         if (use_pse) {
 283                                 unsigned int addr2;
 284                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 285                                 /*
 286                                  * first pass will use the same initial
 287                                  * identity mapping attribute + _PAGE_PSE.
 288                                  */
 289                                 pgprot_t init_prot =
 290                                         __pgprot(PTE_IDENT_ATTR |
 291                                                  _PAGE_PSE);
 292
 293                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 294                                         PAGE_OFFSET + PAGE_SIZE-1;
 295
 296                                 if (is_kernel_text(addr) ||
 297                                     is_kernel_text(addr2))
 298                                         prot = PAGE_KERNEL_LARGE_EXEC;
 299
 300                                 pages_2m++;
 301                                 if (mapping_iter == 1)
 302                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
 303                                 else
 304                                         set_pmd(pmd, pfn_pmd(pfn, prot));
 305
 306                                 pfn += PTRS_PER_PTE;
 307                                 continue;
 308                         }
 309                         pte = one_page_table_init(pmd);
 310
 311                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 312                         pte += pte_ofs;
 313                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 314                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 315                                 pgprot_t prot = PAGE_KERNEL;
 316                                 /*
 317                                  * first pass will use the same initial
 318                                  * identity mapping attribute.
 319                                  */
 320                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
 321
 322                                 if (is_kernel_text(addr))
 323                                         prot = PAGE_KERNEL_EXEC;
 324
 325                                 pages_4k++;
 326                                 if (mapping_iter == 1)
 327                                         set_pte(pte, pfn_pte(pfn, init_prot));
 328                                 else
 329                                         set_pte(pte, pfn_pte(pfn, prot));
 330                         }
 331                 }
 332         }
 333         if (mapping_iter == 1) {
 334                 /*
 335                  * update direct mapping page count only in the first
 336                  * iteration.
 337                  */
 338                 update_page_count(PG_LEVEL_2M, pages_2m);
 339                 update_page_count(PG_LEVEL_4K, pages_4k);
 340
 341                 /*
 342                  * local global flush tlb, which will flush the previous
 343                  * mappings present in both small and large page TLB's.
 344                  */
 345                 __flush_tlb_all();
 346
 347                 /*
 348                  * Second iteration will set the actual desired PTE attributes.
 349                  */
 350                 mapping_iter = 2;
 351                 goto repeat;
 352         }
 353 }
 354
 355 pte_t *kmap_pte;
 356 pgprot_t kmap_prot;
 357
 358 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 359 {
 360         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 361                         vaddr), vaddr), vaddr);
 362 }
 363
 364 static void __init kmap_init(void)
 365 {
 366         unsigned long kmap_vstart;
 367
 368         /*
 369          * Cache the first kmap pte:
 370          */
 371         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 372         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 373
 374         kmap_prot = PAGE_KERNEL;
 375 }
 376
 377 #ifdef CONFIG_HIGHMEM
 378 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 379 {
 380         unsigned long vaddr;
 381         pgd_t *pgd;
 382         pud_t *pud;
 383         pmd_t *pmd;
 384         pte_t *pte;
 385
 386         vaddr = PKMAP_BASE;
 387         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 388
 389         pgd = swapper_pg_dir + pgd_index(vaddr);
 390         pud = pud_offset(pgd, vaddr);
 391         pmd = pmd_offset(pud, vaddr);
 392         pte = pte_offset_kernel(pmd, vaddr);
 393         pkmap_page_table = pte;
 394 }
 395
 396 static void __init add_one_highpage_init(struct page *page, int pfn)
 397 {
 398         ClearPageReserved(page);
 399         init_page_count(page);
 400         __free_page(page);
 401         totalhigh_pages++;
 402 }
 403
 404 struct add_highpages_data {
 405         unsigned long start_pfn;
 406         unsigned long end_pfn;
 407 };
 408
 409 static int __init add_highpages_work_fn(unsigned long start_pfn,
 410                                          unsigned long end_pfn, void *datax)
 411 {
 412         int node_pfn;
 413         struct page *page;
 414         unsigned long final_start_pfn, final_end_pfn;
 415         struct add_highpages_data *data;
 416
 417         data = (struct add_highpages_data *)datax;
 418
 419         final_start_pfn = max(start_pfn, data->start_pfn);
 420         final_end_pfn = min(end_pfn, data->end_pfn);
 421         if (final_start_pfn >= final_end_pfn)
 422                 return 0;
 423
 424         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
 425              node_pfn++) {
 426                 if (!pfn_valid(node_pfn))
 427                         continue;
 428                 page = pfn_to_page(node_pfn);
 429                 add_one_highpage_init(page, node_pfn);
 430         }
 431
 432         return 0;
 433
 434 }
 435
 436 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
 437                                               unsigned long end_pfn)
 438 {
 439         struct add_highpages_data data;
 440
 441         data.start_pfn = start_pfn;
 442         data.end_pfn = end_pfn;
 443
 444         work_with_active_regions(nid, add_highpages_work_fn, &data);
 445 }
 446
 447 #else
 448 static inline void permanent_kmaps_init(pgd_t *pgd_base)
 449 {
 450 }
 451 #endif /* CONFIG_HIGHMEM */
 452
 453 void __init native_pagetable_setup_start(pgd_t *base)
 454 {
 455         unsigned long pfn, va;
 456         pgd_t *pgd;
 457         pud_t *pud;
 458         pmd_t *pmd;
 459         pte_t *pte;
 460
 461         /*
 462          * Remove any mappings which extend past the end of physical
 463          * memory from the boot time page table:
 464          */
 465         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 466                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 467                 pgd = base + pgd_index(va);
 468                 if (!pgd_present(*pgd))
 469                         break;
 470
 471                 pud = pud_offset(pgd, va);
 472                 pmd = pmd_offset(pud, va);
 473                 if (!pmd_present(*pmd))
 474                         break;
 475
 476                 pte = pte_offset_kernel(pmd, va);
 477                 if (!pte_present(*pte))
 478                         break;
 479
 480                 pte_clear(NULL, va, pte);
 481         }
 482         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 483 }
 484
 485 void __init native_pagetable_setup_done(pgd_t *base)
 486 {
 487 }
 488
 489 /*
 490  * Build a proper pagetable for the kernel mappings.  Up until this
 491  * point, we've been running on some set of pagetables constructed by
 492  * the boot process.
 493  *
 494  * If we're booting on native hardware, this will be a pagetable
 495  * constructed in arch/x86/kernel/head_32.S.  The root of the
 496  * pagetable will be swapper_pg_dir.
 497  *
 498  * If we're booting paravirtualized under a hypervisor, then there are
 499  * more options: we may already be running PAE, and the pagetable may
 500  * or may not be based in swapper_pg_dir.  In any case,
 501  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 502  * appropriately for the rest of the initialization to work.
 503  *
 504  * In general, pagetable_init() assumes that the pagetable may already
 505  * be partially populated, and so it avoids stomping on any existing
 506  * mappings.
 507  */
 508 void __init early_ioremap_page_table_range_init(void)
 509 {
 510         pgd_t *pgd_base = swapper_pg_dir;
 511         unsigned long vaddr, end;
 512
 513         /*
 514          * Fixed mappings, only the page table structure has to be
 515          * created - mappings will be set by set_fixmap():
 516          */
 517         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 518         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 519         page_table_range_init(vaddr, end, pgd_base);
 520         early_ioremap_reset();
 521 }
 522
 523 static void __init pagetable_init(void)
 524 {
 525         pgd_t *pgd_base = swapper_pg_dir;
 526
 527         permanent_kmaps_init(pgd_base);
 528 }
 529
 530 #ifdef CONFIG_ACPI_SLEEP
 531 /*
 532  * ACPI suspend needs this for resume, because things like the intel-agp
 533  * driver might have split up a kernel 4MB mapping.
 534  */
 535 char swsusp_pg_dir[PAGE_SIZE]
 536         __attribute__ ((aligned(PAGE_SIZE)));
 537
 538 static inline void save_pg_dir(void)
 539 {
 540         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 541 }
 542 #else /* !CONFIG_ACPI_SLEEP */
 543 static inline void save_pg_dir(void)
 544 {
 545 }
 546 #endif /* !CONFIG_ACPI_SLEEP */
 547
 548 void zap_low_mappings(void)
 549 {
 550         int i;
 551
 552         /*
 553          * Zap initial low-memory mappings.
 554          *
 555          * Note that "pgd_clear()" doesn't do it for
 556          * us, because pgd_clear() is a no-op on i386.
 557          */
 558         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 559 #ifdef CONFIG_X86_PAE
 560                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 561 #else
 562                 set_pgd(swapper_pg_dir+i, __pgd(0));
 563 #endif
 564         }
 565         flush_tlb_all();
 566 }
 567
 568 int nx_enabled;
 569
 570 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
 571 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 572
 573 #ifdef CONFIG_X86_PAE
 574
 575 static int disable_nx __initdata;
 576
 577 /*
 578  * noexec = on|off
 579  *
 580  * Control non executable mappings.
 581  *
 582  * on      Enable
 583  * off     Disable
 584  */
 585 static int __init noexec_setup(char *str)
 586 {
 587         if (!str || !strcmp(str, "on")) {
 588                 if (cpu_has_nx) {
 589                         __supported_pte_mask |= _PAGE_NX;
 590                         disable_nx = 0;
 591                 }
 592         } else {
 593                 if (!strcmp(str, "off")) {
 594                         disable_nx = 1;
 595                         __supported_pte_mask &= ~_PAGE_NX;
 596                 } else {
 597                         return -EINVAL;
 598                 }
 599         }
 600
 601         return 0;
 602 }
 603 early_param("noexec", noexec_setup);
 604
 605 static void __init set_nx(void)
 606 {
 607         unsigned int v[4], l, h;
 608
 609         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 610                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 611
 612                 if ((v[3] & (1 << 20)) && !disable_nx) {
 613                         rdmsr(MSR_EFER, l, h);
 614                         l |= EFER_NX;
 615                         wrmsr(MSR_EFER, l, h);
 616                         nx_enabled = 1;
 617                         __supported_pte_mask |= _PAGE_NX;
 618                 }
 619         }
 620 }
 621 #endif
 622
 623 /* user-defined highmem size */
 624 static unsigned int highmem_pages = -1;
 625
 626 /*
 627  * highmem=size forces highmem to be exactly 'size' bytes.
 628  * This works even on boxes that have no highmem otherwise.
 629  * This also works to reduce highmem size on bigger boxes.
 630  */
 631 static int __init parse_highmem(char *arg)
 632 {
 633         if (!arg)
 634                 return -EINVAL;
 635
 636         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 637         return 0;
 638 }
 639 early_param("highmem", parse_highmem);
 640
 641 #define MSG_HIGHMEM_TOO_BIG \
 642         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
 643
 644 #define MSG_LOWMEM_TOO_SMALL \
 645         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
 646 /*
 647  * All of RAM fits into lowmem - but if user wants highmem
 648  * artificially via the highmem=x boot parameter then create
 649  * it:
 650  */
 651 void __init lowmem_pfn_init(void)
 652 {
 653         /* max_low_pfn is 0, we already have early_res support */
 654         max_low_pfn = max_pfn;
 655
 656         if (highmem_pages == -1)
 657                 highmem_pages = 0;
 658 #ifdef CONFIG_HIGHMEM
 659         if (highmem_pages >= max_pfn) {
 660                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
 661                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 662                 highmem_pages = 0;
 663         }
 664         if (highmem_pages) {
 665                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
 666                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
 667                                 pages_to_mb(highmem_pages));
 668                         highmem_pages = 0;
 669                 }
 670                 max_low_pfn -= highmem_pages;
 671         }
 672 #else
 673         if (highmem_pages)
 674                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 675 #endif
 676 }
 677
 678 #define MSG_HIGHMEM_TOO_SMALL \
 679         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
 680
 681 #define MSG_HIGHMEM_TRIMMED \
 682         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
 683 /*
 684  * We have more RAM than fits into lowmem - we try to put it into
 685  * highmem, also taking the highmem=x boot parameter into account:
 686  */
 687 void __init highmem_pfn_init(void)
 688 {
 689         max_low_pfn = MAXMEM_PFN;
 690
 691         if (highmem_pages == -1)
 692                 highmem_pages = max_pfn - MAXMEM_PFN;
 693
 694         if (highmem_pages + MAXMEM_PFN < max_pfn)
 695                 max_pfn = MAXMEM_PFN + highmem_pages;
 696
 697         if (highmem_pages + MAXMEM_PFN > max_pfn) {
 698                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
 699                         pages_to_mb(max_pfn - MAXMEM_PFN),
 700                         pages_to_mb(highmem_pages));
 701                 highmem_pages = 0;
 702         }
 703 #ifndef CONFIG_HIGHMEM
 704         /* Maximum memory usable is what is directly addressable */
 705         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
 706         if (max_pfn > MAX_NONPAE_PFN)
 707                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 708         else
 709                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 710         max_pfn = MAXMEM_PFN;
 711 #else /* !CONFIG_HIGHMEM */
 712 #ifndef CONFIG_HIGHMEM64G
 713         if (max_pfn > MAX_NONPAE_PFN) {
 714                 max_pfn = MAX_NONPAE_PFN;
 715                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
 716         }
 717 #endif /* !CONFIG_HIGHMEM64G */
 718 #endif /* !CONFIG_HIGHMEM */
 719 }
 720
 721 /*
 722  * Determine low and high memory ranges:
 723  */
 724 void __init find_low_pfn_range(void)
 725 {
 726         /* it could update max_pfn */
 727
 728         if (max_pfn <= MAXMEM_PFN)
 729                 lowmem_pfn_init();
 730         else
 731                 highmem_pfn_init();
 732 }
 733
 734 #ifndef CONFIG_NEED_MULTIPLE_NODES
 735 void __init initmem_init(unsigned long start_pfn,
 736                                   unsigned long end_pfn)
 737 {
 738 #ifdef CONFIG_HIGHMEM
 739         highstart_pfn = highend_pfn = max_pfn;
 740         if (max_pfn > max_low_pfn)
 741                 highstart_pfn = max_low_pfn;
 742         memory_present(0, 0, highend_pfn);
 743         e820_register_active_regions(0, 0, highend_pfn);
 744         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 745                 pages_to_mb(highend_pfn - highstart_pfn));
 746         num_physpages = highend_pfn;
 747         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 748 #else
 749         memory_present(0, 0, max_low_pfn);
 750         e820_register_active_regions(0, 0, max_low_pfn);
 751         num_physpages = max_low_pfn;
 752         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 753 #endif
 754 #ifdef CONFIG_FLATMEM
 755         max_mapnr = num_physpages;
 756 #endif
 757         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 758                         pages_to_mb(max_low_pfn));
 759
 760         setup_bootmem_allocator();
 761 }
 762 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 763
 764 static void __init zone_sizes_init(void)
 765 {
 766         unsigned long max_zone_pfns[MAX_NR_ZONES];
 767         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 768         max_zone_pfns[ZONE_DMA] =
 769                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 770         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 771 #ifdef CONFIG_HIGHMEM
 772         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 773 #endif
 774
 775         free_area_init_nodes(max_zone_pfns);
 776 }
 777
 778 static unsigned long __init setup_node_bootmem(int nodeid,
 779                                  unsigned long start_pfn,
 780                                  unsigned long end_pfn,
 781                                  unsigned long bootmap)
 782 {
 783         unsigned long bootmap_size;
 784
 785         if (start_pfn > max_low_pfn)
 786                 return bootmap;
 787         if (end_pfn > max_low_pfn)
 788                 end_pfn = max_low_pfn;
 789
 790         /* don't touch min_low_pfn */
 791         bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
 792                                          bootmap >> PAGE_SHIFT,
 793                                          start_pfn, end_pfn);
 794         printk(KERN_INFO "  node %d low ram: %08lx - %08lx\n",
 795                 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 796         printk(KERN_INFO "  node %d bootmap %08lx - %08lx\n",
 797                  nodeid, bootmap, bootmap + bootmap_size);
 798         free_bootmem_with_active_regions(nodeid, end_pfn);
 799         early_res_to_bootmem(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 800
 801         return bootmap + bootmap_size;
 802 }
 803
 804 void __init setup_bootmem_allocator(void)
 805 {
 806         int nodeid;
 807         unsigned long bootmap_size, bootmap;
 808         /*
 809          * Initialize the boot-time allocator (with low memory only):
 810          */
 811         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 812         bootmap = find_e820_area(0, max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 813                                  PAGE_SIZE);
 814         if (bootmap == -1L)
 815                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 816         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 817
 818         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 819                  max_pfn_mapped<<PAGE_SHIFT);
 820         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
 821
 822 #ifdef CONFIG_NEED_MULTIPLE_NODES
 823         for_each_online_node(nodeid)
 824                 bootmap = setup_node_bootmem(nodeid, node_start_pfn[nodeid],
 825                                         node_end_pfn[nodeid], bootmap);
 826 #else
 827         bootmap = setup_node_bootmem(0, 0, max_low_pfn, bootmap);
 828 #endif
 829
 830         after_bootmem = 1;
 831 }
 832
 833 /*
 834  * paging_init() sets up the page tables - note that the first 8MB are
 835  * already mapped by head.S.
 836  *
 837  * This routines also unmaps the page at virtual kernel address 0, so
 838  * that we can trap those pesky NULL-reference errors in the kernel.
 839  */
 840 void __init paging_init(void)
 841 {
 842         pagetable_init();
 843
 844         __flush_tlb_all();
 845
 846         kmap_init();
 847
 848         /*
 849          * NOTE: at this point the bootmem allocator is fully available.
 850          */
 851         sparse_init();
 852         zone_sizes_init();
 853 }
 854
 855 /*
 856  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 857  * and also on some strange 486's. All 586+'s are OK. This used to involve
 858  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 859  * switch to using exceptions got rid of all that.
 860  */
 861 static void __init test_wp_bit(void)
 862 {
 863         printk(KERN_INFO
 864   "Checking if this processor honours the WP bit even in supervisor mode...");
 865
 866         /* Any page-aligned address will do, the test is non-destructive */
 867         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 868         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 869         clear_fixmap(FIX_WP_TEST);
 870
 871         if (!boot_cpu_data.wp_works_ok) {
 872                 printk(KERN_CONT "No.\n");
 873 #ifdef CONFIG_X86_WP_WORKS_OK
 874                 panic(
 875   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 876 #endif
 877         } else {
 878                 printk(KERN_CONT "Ok.\n");
 879         }
 880 }
 881
 882 static struct kcore_list kcore_mem, kcore_vmalloc;
 883
 884 void __init mem_init(void)
 885 {
 886         int codesize, reservedpages, datasize, initsize;
 887         int tmp;
 888
 889         pci_iommu_alloc();
 890
 891 #ifdef CONFIG_FLATMEM
 892         BUG_ON(!mem_map);
 893 #endif
 894         /* this will put all low memory onto the freelists */
 895         totalram_pages += free_all_bootmem();
 896
 897         reservedpages = 0;
 898         for (tmp = 0; tmp < max_low_pfn; tmp++)
 899                 /*
 900                  * Only count reserved RAM pages:
 901                  */
 902                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 903                         reservedpages++;
 904
 905         set_highmem_pages_init();
 906
 907         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 908         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 909         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 910
 911         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 912         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 913                    VMALLOC_END-VMALLOC_START);
 914
 915         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 916                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 917                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 918                 num_physpages << (PAGE_SHIFT-10),
 919                 codesize >> 10,
 920                 reservedpages << (PAGE_SHIFT-10),
 921                 datasize >> 10,
 922                 initsize >> 10,
 923                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 924                );
 925
 926         printk(KERN_INFO "virtual kernel memory layout:\n"
 927                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 928 #ifdef CONFIG_HIGHMEM
 929                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 930 #endif
 931                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 932                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 933                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 934                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 935                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 936                 FIXADDR_START, FIXADDR_TOP,
 937                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 938
 939 #ifdef CONFIG_HIGHMEM
 940                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 941                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 942 #endif
 943
 944                 VMALLOC_START, VMALLOC_END,
 945                 (VMALLOC_END - VMALLOC_START) >> 20,
 946
 947                 (unsigned long)__va(0), (unsigned long)high_memory,
 948                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 949
 950                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 951                 ((unsigned long)&__init_end -
 952                  (unsigned long)&__init_begin) >> 10,
 953
 954                 (unsigned long)&_etext, (unsigned long)&_edata,
 955                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 956
 957                 (unsigned long)&_text, (unsigned long)&_etext,
 958                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 959
 960         /*
 961          * Check boundaries twice: Some fundamental inconsistencies can
 962          * be detected at build time already.
 963          */
 964 #define __FIXADDR_TOP (-PAGE_SIZE)
 965 #ifdef CONFIG_HIGHMEM
 966         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
 967         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
 968 #endif
 969 #define high_memory (-128UL << 20)
 970         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
 971 #undef high_memory
 972 #undef __FIXADDR_TOP
 973
 974 #ifdef CONFIG_HIGHMEM
 975         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 976         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 977 #endif
 978         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
 979         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 980
 981         if (boot_cpu_data.wp_works_ok < 0)
 982                 test_wp_bit();
 983
 984         save_pg_dir();
 985         zap_low_mappings();
 986 }
 987
 988 #ifdef CONFIG_MEMORY_HOTPLUG
 989 int arch_add_memory(int nid, u64 start, u64 size)
 990 {
 991         struct pglist_data *pgdata = NODE_DATA(nid);
 992         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 993         unsigned long start_pfn = start >> PAGE_SHIFT;
 994         unsigned long nr_pages = size >> PAGE_SHIFT;
 995
 996         return __add_pages(nid, zone, start_pfn, nr_pages);
 997 }
 998 #endif
 999
1000 /*
1001  * This function cannot be __init, since exceptions don't work in that
1002  * section.  Put this after the callers, so that it cannot be inlined.
1003  */
1004 static noinline int do_test_wp_bit(void)
1005 {
1006         char tmp_reg;
1007         int flag;
1008
1009         __asm__ __volatile__(
1010                 "       movb %0, %1     \n"
1011                 "1:     movb %1, %0     \n"
1012                 "       xorl %2, %2     \n"
1013                 "2:                     \n"
1014                 _ASM_EXTABLE(1b,2b)
1015                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1016                  "=q" (tmp_reg),
1017                  "=r" (flag)
1018                 :"2" (1)
1019                 :"memory");
1020
1021         return flag;
1022 }
1023
1024 #ifdef CONFIG_DEBUG_RODATA
1025 const int rodata_test_data = 0xC3;
1026 EXPORT_SYMBOL_GPL(rodata_test_data);
1027
1028 void mark_rodata_ro(void)
1029 {
1030         unsigned long start = PFN_ALIGN(_text);
1031         unsigned long size = PFN_ALIGN(_etext) - start;
1032
1033 #ifndef CONFIG_DYNAMIC_FTRACE
1034         /* Dynamic tracing modifies the kernel text section */
1035         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1036         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1037                 size >> 10);
1038
1039 #ifdef CONFIG_CPA_DEBUG
1040         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1041                 start, start+size);
1042         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1043
1044         printk(KERN_INFO "Testing CPA: write protecting again\n");
1045         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1046 #endif
1047 #endif /* CONFIG_DYNAMIC_FTRACE */
1048
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 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1067                                    int flags)
1068 {
1069         return reserve_bootmem(phys, len, flags);
1070 }