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/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 #include <asm/init.h>
  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 bool __read_mostly __vmalloc_start_set = false;
  63
  64 static __init void *alloc_low_page(void)
  65 {
  66         unsigned long pfn = e820_table_end++;
  67         void *adr;
  68
  69         if (pfn >= e820_table_top)
  70                 panic("alloc_low_page: ran out of memory");
  71
  72         adr = __va(pfn * PAGE_SIZE);
  73         memset(adr, 0, PAGE_SIZE);
  74         return adr;
  75 }
  76
  77 /*
  78  * Creates a middle page table and puts a pointer to it in the
  79  * given global directory entry. This only returns the gd entry
  80  * in non-PAE compilation mode, since the middle layer is folded.
  81  */
  82 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  83 {
  84         pud_t *pud;
  85         pmd_t *pmd_table;
  86
  87 #ifdef CONFIG_X86_PAE
  88         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
  89                 if (after_bootmem)
  90                         pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
  91                 else
  92                         pmd_table = (pmd_t *)alloc_low_page();
  93                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
  94                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  95                 pud = pud_offset(pgd, 0);
  96                 BUG_ON(pmd_table != pmd_offset(pud, 0));
  97
  98                 return pmd_table;
  99         }
 100 #endif
 101         pud = pud_offset(pgd, 0);
 102         pmd_table = pmd_offset(pud, 0);
 103
 104         return pmd_table;
 105 }
 106
 107 /*
 108  * Create a page table and place a pointer to it in a middle page
 109  * directory entry:
 110  */
 111 static pte_t * __init one_page_table_init(pmd_t *pmd)
 112 {
 113         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
 114                 pte_t *page_table = NULL;
 115
 116                 if (after_bootmem) {
 117 #ifdef CONFIG_DEBUG_PAGEALLOC
 118                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
 119 #endif
 120                         if (!page_table)
 121                                 page_table =
 122                                 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 123                 } else
 124                         page_table = (pte_t *)alloc_low_page();
 125
 126                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
 127                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
 128                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
 129         }
 130
 131         return pte_offset_kernel(pmd, 0);
 132 }
 133
 134 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
 135                                            unsigned long vaddr, pte_t *lastpte)
 136 {
 137 #ifdef CONFIG_HIGHMEM
 138         /*
 139          * Something (early fixmap) may already have put a pte
 140          * page here, which causes the page table allocation
 141          * to become nonlinear. Attempt to fix it, and if it
 142          * is still nonlinear then we have to bug.
 143          */
 144         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
 145         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
 146
 147         if (pmd_idx_kmap_begin != pmd_idx_kmap_end
 148             && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
 149             && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
 150             && ((__pa(pte) >> PAGE_SHIFT) < e820_table_start
 151                 || (__pa(pte) >> PAGE_SHIFT) >= e820_table_end)) {
 152                 pte_t *newpte;
 153                 int i;
 154
 155                 BUG_ON(after_bootmem);
 156                 newpte = alloc_low_page();
 157                 for (i = 0; i < PTRS_PER_PTE; i++)
 158                         set_pte(newpte + i, pte[i]);
 159
 160                 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
 161                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
 162                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
 163                 __flush_tlb_all();
 164
 165                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
 166                 pte = newpte;
 167         }
 168         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
 169                && vaddr > fix_to_virt(FIX_KMAP_END)
 170                && lastpte && lastpte + PTRS_PER_PTE != pte);
 171 #endif
 172         return pte;
 173 }
 174
 175 /*
 176  * This function initializes a certain range of kernel virtual memory
 177  * with new bootmem page tables, everywhere page tables are missing in
 178  * the given range.
 179  *
 180  * NOTE: The pagetables are allocated contiguous on the physical space
 181  * so we can cache the place of the first one and move around without
 182  * checking the pgd every time.
 183  */
 184 static void __init
 185 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 186 {
 187         int pgd_idx, pmd_idx;
 188         unsigned long vaddr;
 189         pgd_t *pgd;
 190         pmd_t *pmd;
 191         pte_t *pte = NULL;
 192
 193         vaddr = start;
 194         pgd_idx = pgd_index(vaddr);
 195         pmd_idx = pmd_index(vaddr);
 196         pgd = pgd_base + pgd_idx;
 197
 198         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 199                 pmd = one_md_table_init(pgd);
 200                 pmd = pmd + pmd_index(vaddr);
 201                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 202                                                         pmd++, pmd_idx++) {
 203                         pte = page_table_kmap_check(one_page_table_init(pmd),
 204                                                     pmd, vaddr, pte);
 205
 206                         vaddr += PMD_SIZE;
 207                 }
 208                 pmd_idx = 0;
 209         }
 210 }
 211
 212 static inline int is_kernel_text(unsigned long addr)
 213 {
 214         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 215                 return 1;
 216         return 0;
 217 }
 218
 219 /*
 220  * This maps the physical memory to kernel virtual address space, a total
 221  * of max_low_pfn pages, by creating page tables starting from address
 222  * PAGE_OFFSET:
 223  */
 224 unsigned long __init
 225 kernel_physical_mapping_init(unsigned long start,
 226                              unsigned long end,
 227                              unsigned long page_size_mask)
 228 {
 229         int use_pse = page_size_mask == (1<<PG_LEVEL_2M);
 230         unsigned long start_pfn, end_pfn;
 231         pgd_t *pgd_base = swapper_pg_dir;
 232         int pgd_idx, pmd_idx, pte_ofs;
 233         unsigned long pfn;
 234         pgd_t *pgd;
 235         pmd_t *pmd;
 236         pte_t *pte;
 237         unsigned pages_2m, pages_4k;
 238         int mapping_iter;
 239
 240         start_pfn = start >> PAGE_SHIFT;
 241         end_pfn = end >> PAGE_SHIFT;
 242
 243         /*
 244          * First iteration will setup identity mapping using large/small pages
 245          * based on use_pse, with other attributes same as set by
 246          * the early code in head_32.S
 247          *
 248          * Second iteration will setup the appropriate attributes (NX, GLOBAL..)
 249          * as desired for the kernel identity mapping.
 250          *
 251          * This two pass mechanism conforms to the TLB app note which says:
 252          *
 253          *     "Software should not write to a paging-structure entry in a way
 254          *      that would change, for any linear address, both the page size
 255          *      and either the page frame or attributes."
 256          */
 257         mapping_iter = 1;
 258
 259         if (!cpu_has_pse)
 260                 use_pse = 0;
 261
 262 repeat:
 263         pages_2m = pages_4k = 0;
 264         pfn = start_pfn;
 265         pgd_idx = pgd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 266         pgd = pgd_base + pgd_idx;
 267         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 268                 pmd = one_md_table_init(pgd);
 269
 270                 if (pfn >= end_pfn)
 271                         continue;
 272 #ifdef CONFIG_X86_PAE
 273                 pmd_idx = pmd_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 274                 pmd += pmd_idx;
 275 #else
 276                 pmd_idx = 0;
 277 #endif
 278                 for (; pmd_idx < PTRS_PER_PMD && pfn < end_pfn;
 279                      pmd++, pmd_idx++) {
 280                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 281
 282                         /*
 283                          * Map with big pages if possible, otherwise
 284                          * create normal page tables:
 285                          */
 286                         if (use_pse) {
 287                                 unsigned int addr2;
 288                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 289                                 /*
 290                                  * first pass will use the same initial
 291                                  * identity mapping attribute + _PAGE_PSE.
 292                                  */
 293                                 pgprot_t init_prot =
 294                                         __pgprot(PTE_IDENT_ATTR |
 295                                                  _PAGE_PSE);
 296
 297                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 298                                         PAGE_OFFSET + PAGE_SIZE-1;
 299
 300                                 if (is_kernel_text(addr) ||
 301                                     is_kernel_text(addr2))
 302                                         prot = PAGE_KERNEL_LARGE_EXEC;
 303
 304                                 pages_2m++;
 305                                 if (mapping_iter == 1)
 306                                         set_pmd(pmd, pfn_pmd(pfn, init_prot));
 307                                 else
 308                                         set_pmd(pmd, pfn_pmd(pfn, prot));
 309
 310                                 pfn += PTRS_PER_PTE;
 311                                 continue;
 312                         }
 313                         pte = one_page_table_init(pmd);
 314
 315                         pte_ofs = pte_index((pfn<<PAGE_SHIFT) + PAGE_OFFSET);
 316                         pte += pte_ofs;
 317                         for (; pte_ofs < PTRS_PER_PTE && pfn < end_pfn;
 318                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 319                                 pgprot_t prot = PAGE_KERNEL;
 320                                 /*
 321                                  * first pass will use the same initial
 322                                  * identity mapping attribute.
 323                                  */
 324                                 pgprot_t init_prot = __pgprot(PTE_IDENT_ATTR);
 325
 326                                 if (is_kernel_text(addr))
 327                                         prot = PAGE_KERNEL_EXEC;
 328
 329                                 pages_4k++;
 330                                 if (mapping_iter == 1)
 331                                         set_pte(pte, pfn_pte(pfn, init_prot));
 332                                 else
 333                                         set_pte(pte, pfn_pte(pfn, prot));
 334                         }
 335                 }
 336         }
 337         if (mapping_iter == 1) {
 338                 /*
 339                  * update direct mapping page count only in the first
 340                  * iteration.
 341                  */
 342                 update_page_count(PG_LEVEL_2M, pages_2m);
 343                 update_page_count(PG_LEVEL_4K, pages_4k);
 344
 345                 /*
 346                  * local global flush tlb, which will flush the previous
 347                  * mappings present in both small and large page TLB's.
 348                  */
 349                 __flush_tlb_all();
 350
 351                 /*
 352                  * Second iteration will set the actual desired PTE attributes.
 353                  */
 354                 mapping_iter = 2;
 355                 goto repeat;
 356         }
 357         return 0;
 358 }
 359
 360 pte_t *kmap_pte;
 361 pgprot_t kmap_prot;
 362
 363 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 364 {
 365         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 366                         vaddr), vaddr), vaddr);
 367 }
 368
 369 static void __init kmap_init(void)
 370 {
 371         unsigned long kmap_vstart;
 372
 373         /*
 374          * Cache the first kmap pte:
 375          */
 376         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 377         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 378
 379         kmap_prot = PAGE_KERNEL;
 380 }
 381
 382 #ifdef CONFIG_HIGHMEM
 383 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 384 {
 385         unsigned long vaddr;
 386         pgd_t *pgd;
 387         pud_t *pud;
 388         pmd_t *pmd;
 389         pte_t *pte;
 390
 391         vaddr = PKMAP_BASE;
 392         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 393
 394         pgd = swapper_pg_dir + pgd_index(vaddr);
 395         pud = pud_offset(pgd, vaddr);
 396         pmd = pmd_offset(pud, vaddr);
 397         pte = pte_offset_kernel(pmd, vaddr);
 398         pkmap_page_table = pte;
 399 }
 400
 401 static void __init add_one_highpage_init(struct page *page, int pfn)
 402 {
 403         ClearPageReserved(page);
 404         init_page_count(page);
 405         __free_page(page);
 406         totalhigh_pages++;
 407 }
 408
 409 struct add_highpages_data {
 410         unsigned long start_pfn;
 411         unsigned long end_pfn;
 412 };
 413
 414 static int __init add_highpages_work_fn(unsigned long start_pfn,
 415                                          unsigned long end_pfn, void *datax)
 416 {
 417         int node_pfn;
 418         struct page *page;
 419         unsigned long final_start_pfn, final_end_pfn;
 420         struct add_highpages_data *data;
 421
 422         data = (struct add_highpages_data *)datax;
 423
 424         final_start_pfn = max(start_pfn, data->start_pfn);
 425         final_end_pfn = min(end_pfn, data->end_pfn);
 426         if (final_start_pfn >= final_end_pfn)
 427                 return 0;
 428
 429         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
 430              node_pfn++) {
 431                 if (!pfn_valid(node_pfn))
 432                         continue;
 433                 page = pfn_to_page(node_pfn);
 434                 add_one_highpage_init(page, node_pfn);
 435         }
 436
 437         return 0;
 438
 439 }
 440
 441 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
 442                                               unsigned long end_pfn)
 443 {
 444         struct add_highpages_data data;
 445
 446         data.start_pfn = start_pfn;
 447         data.end_pfn = end_pfn;
 448
 449         work_with_active_regions(nid, add_highpages_work_fn, &data);
 450 }
 451
 452 #else
 453 static inline void permanent_kmaps_init(pgd_t *pgd_base)
 454 {
 455 }
 456 #endif /* CONFIG_HIGHMEM */
 457
 458 void __init native_pagetable_setup_start(pgd_t *base)
 459 {
 460         unsigned long pfn, va;
 461         pgd_t *pgd;
 462         pud_t *pud;
 463         pmd_t *pmd;
 464         pte_t *pte;
 465
 466         /*
 467          * Remove any mappings which extend past the end of physical
 468          * memory from the boot time page table:
 469          */
 470         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 471                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 472                 pgd = base + pgd_index(va);
 473                 if (!pgd_present(*pgd))
 474                         break;
 475
 476                 pud = pud_offset(pgd, va);
 477                 pmd = pmd_offset(pud, va);
 478                 if (!pmd_present(*pmd))
 479                         break;
 480
 481                 pte = pte_offset_kernel(pmd, va);
 482                 if (!pte_present(*pte))
 483                         break;
 484
 485                 pte_clear(NULL, va, pte);
 486         }
 487         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 488 }
 489
 490 void __init native_pagetable_setup_done(pgd_t *base)
 491 {
 492 }
 493
 494 /*
 495  * Build a proper pagetable for the kernel mappings.  Up until this
 496  * point, we've been running on some set of pagetables constructed by
 497  * the boot process.
 498  *
 499  * If we're booting on native hardware, this will be a pagetable
 500  * constructed in arch/x86/kernel/head_32.S.  The root of the
 501  * pagetable will be swapper_pg_dir.
 502  *
 503  * If we're booting paravirtualized under a hypervisor, then there are
 504  * more options: we may already be running PAE, and the pagetable may
 505  * or may not be based in swapper_pg_dir.  In any case,
 506  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 507  * appropriately for the rest of the initialization to work.
 508  *
 509  * In general, pagetable_init() assumes that the pagetable may already
 510  * be partially populated, and so it avoids stomping on any existing
 511  * mappings.
 512  */
 513 void __init early_ioremap_page_table_range_init(void)
 514 {
 515         pgd_t *pgd_base = swapper_pg_dir;
 516         unsigned long vaddr, end;
 517
 518         /*
 519          * Fixed mappings, only the page table structure has to be
 520          * created - mappings will be set by set_fixmap():
 521          */
 522         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 523         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 524         page_table_range_init(vaddr, end, pgd_base);
 525         early_ioremap_reset();
 526 }
 527
 528 static void __init pagetable_init(void)
 529 {
 530         pgd_t *pgd_base = swapper_pg_dir;
 531
 532         permanent_kmaps_init(pgd_base);
 533 }
 534
 535 #ifdef CONFIG_ACPI_SLEEP
 536 /*
 537  * ACPI suspend needs this for resume, because things like the intel-agp
 538  * driver might have split up a kernel 4MB mapping.
 539  */
 540 char swsusp_pg_dir[PAGE_SIZE]
 541         __attribute__ ((aligned(PAGE_SIZE)));
 542
 543 static inline void save_pg_dir(void)
 544 {
 545         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 546 }
 547 #else /* !CONFIG_ACPI_SLEEP */
 548 static inline void save_pg_dir(void)
 549 {
 550 }
 551 #endif /* !CONFIG_ACPI_SLEEP */
 552
 553 void zap_low_mappings(void)
 554 {
 555         int i;
 556
 557         /*
 558          * Zap initial low-memory mappings.
 559          *
 560          * Note that "pgd_clear()" doesn't do it for
 561          * us, because pgd_clear() is a no-op on i386.
 562          */
 563         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 564 #ifdef CONFIG_X86_PAE
 565                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 566 #else
 567                 set_pgd(swapper_pg_dir+i, __pgd(0));
 568 #endif
 569         }
 570         flush_tlb_all();
 571 }
 572
 573 int nx_enabled;
 574
 575 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
 576 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 577
 578 #ifdef CONFIG_X86_PAE
 579
 580 static int disable_nx __initdata;
 581
 582 /*
 583  * noexec = on|off
 584  *
 585  * Control non executable mappings.
 586  *
 587  * on      Enable
 588  * off     Disable
 589  */
 590 static int __init noexec_setup(char *str)
 591 {
 592         if (!str || !strcmp(str, "on")) {
 593                 if (cpu_has_nx) {
 594                         __supported_pte_mask |= _PAGE_NX;
 595                         disable_nx = 0;
 596                 }
 597         } else {
 598                 if (!strcmp(str, "off")) {
 599                         disable_nx = 1;
 600                         __supported_pte_mask &= ~_PAGE_NX;
 601                 } else {
 602                         return -EINVAL;
 603                 }
 604         }
 605
 606         return 0;
 607 }
 608 early_param("noexec", noexec_setup);
 609
 610 void __init set_nx(void)
 611 {
 612         unsigned int v[4], l, h;
 613
 614         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 615                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 616
 617                 if ((v[3] & (1 << 20)) && !disable_nx) {
 618                         rdmsr(MSR_EFER, l, h);
 619                         l |= EFER_NX;
 620                         wrmsr(MSR_EFER, l, h);
 621                         nx_enabled = 1;
 622                         __supported_pte_mask |= _PAGE_NX;
 623                 }
 624         }
 625 }
 626 #endif
 627
 628 /* user-defined highmem size */
 629 static unsigned int highmem_pages = -1;
 630
 631 /*
 632  * highmem=size forces highmem to be exactly 'size' bytes.
 633  * This works even on boxes that have no highmem otherwise.
 634  * This also works to reduce highmem size on bigger boxes.
 635  */
 636 static int __init parse_highmem(char *arg)
 637 {
 638         if (!arg)
 639                 return -EINVAL;
 640
 641         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 642         return 0;
 643 }
 644 early_param("highmem", parse_highmem);
 645
 646 #define MSG_HIGHMEM_TOO_BIG \
 647         "highmem size (%luMB) is bigger than pages available (%luMB)!\n"
 648
 649 #define MSG_LOWMEM_TOO_SMALL \
 650         "highmem size (%luMB) results in <64MB lowmem, ignoring it!\n"
 651 /*
 652  * All of RAM fits into lowmem - but if user wants highmem
 653  * artificially via the highmem=x boot parameter then create
 654  * it:
 655  */
 656 void __init lowmem_pfn_init(void)
 657 {
 658         /* max_low_pfn is 0, we already have early_res support */
 659         max_low_pfn = max_pfn;
 660
 661         if (highmem_pages == -1)
 662                 highmem_pages = 0;
 663 #ifdef CONFIG_HIGHMEM
 664         if (highmem_pages >= max_pfn) {
 665                 printk(KERN_ERR MSG_HIGHMEM_TOO_BIG,
 666                         pages_to_mb(highmem_pages), pages_to_mb(max_pfn));
 667                 highmem_pages = 0;
 668         }
 669         if (highmem_pages) {
 670                 if (max_low_pfn - highmem_pages < 64*1024*1024/PAGE_SIZE) {
 671                         printk(KERN_ERR MSG_LOWMEM_TOO_SMALL,
 672                                 pages_to_mb(highmem_pages));
 673                         highmem_pages = 0;
 674                 }
 675                 max_low_pfn -= highmem_pages;
 676         }
 677 #else
 678         if (highmem_pages)
 679                 printk(KERN_ERR "ignoring highmem size on non-highmem kernel!\n");
 680 #endif
 681 }
 682
 683 #define MSG_HIGHMEM_TOO_SMALL \
 684         "only %luMB highmem pages available, ignoring highmem size of %luMB!\n"
 685
 686 #define MSG_HIGHMEM_TRIMMED \
 687         "Warning: only 4GB will be used. Use a HIGHMEM64G enabled kernel!\n"
 688 /*
 689  * We have more RAM than fits into lowmem - we try to put it into
 690  * highmem, also taking the highmem=x boot parameter into account:
 691  */
 692 void __init highmem_pfn_init(void)
 693 {
 694         max_low_pfn = MAXMEM_PFN;
 695
 696         if (highmem_pages == -1)
 697                 highmem_pages = max_pfn - MAXMEM_PFN;
 698
 699         if (highmem_pages + MAXMEM_PFN < max_pfn)
 700                 max_pfn = MAXMEM_PFN + highmem_pages;
 701
 702         if (highmem_pages + MAXMEM_PFN > max_pfn) {
 703                 printk(KERN_WARNING MSG_HIGHMEM_TOO_SMALL,
 704                         pages_to_mb(max_pfn - MAXMEM_PFN),
 705                         pages_to_mb(highmem_pages));
 706                 highmem_pages = 0;
 707         }
 708 #ifndef CONFIG_HIGHMEM
 709         /* Maximum memory usable is what is directly addressable */
 710         printk(KERN_WARNING "Warning only %ldMB will be used.\n", MAXMEM>>20);
 711         if (max_pfn > MAX_NONPAE_PFN)
 712                 printk(KERN_WARNING "Use a HIGHMEM64G enabled kernel.\n");
 713         else
 714                 printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 715         max_pfn = MAXMEM_PFN;
 716 #else /* !CONFIG_HIGHMEM */
 717 #ifndef CONFIG_HIGHMEM64G
 718         if (max_pfn > MAX_NONPAE_PFN) {
 719                 max_pfn = MAX_NONPAE_PFN;
 720                 printk(KERN_WARNING MSG_HIGHMEM_TRIMMED);
 721         }
 722 #endif /* !CONFIG_HIGHMEM64G */
 723 #endif /* !CONFIG_HIGHMEM */
 724 }
 725
 726 /*
 727  * Determine low and high memory ranges:
 728  */
 729 void __init find_low_pfn_range(void)
 730 {
 731         /* it could update max_pfn */
 732
 733         if (max_pfn <= MAXMEM_PFN)
 734                 lowmem_pfn_init();
 735         else
 736                 highmem_pfn_init();
 737 }
 738
 739 #ifndef CONFIG_NEED_MULTIPLE_NODES
 740 void __init initmem_init(unsigned long start_pfn,
 741                                   unsigned long end_pfn)
 742 {
 743 #ifdef CONFIG_HIGHMEM
 744         highstart_pfn = highend_pfn = max_pfn;
 745         if (max_pfn > max_low_pfn)
 746                 highstart_pfn = max_low_pfn;
 747         memory_present(0, 0, highend_pfn);
 748         e820_register_active_regions(0, 0, highend_pfn);
 749         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 750                 pages_to_mb(highend_pfn - highstart_pfn));
 751         num_physpages = highend_pfn;
 752         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 753 #else
 754         memory_present(0, 0, max_low_pfn);
 755         e820_register_active_regions(0, 0, max_low_pfn);
 756         num_physpages = max_low_pfn;
 757         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 758 #endif
 759 #ifdef CONFIG_FLATMEM
 760         max_mapnr = num_physpages;
 761 #endif
 762         __vmalloc_start_set = true;
 763
 764         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 765                         pages_to_mb(max_low_pfn));
 766
 767         setup_bootmem_allocator();
 768 }
 769 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 770
 771 static void __init zone_sizes_init(void)
 772 {
 773         unsigned long max_zone_pfns[MAX_NR_ZONES];
 774         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 775         max_zone_pfns[ZONE_DMA] =
 776                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 777         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 778 #ifdef CONFIG_HIGHMEM
 779         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 780 #endif
 781
 782         free_area_init_nodes(max_zone_pfns);
 783 }
 784
 785 static unsigned long __init setup_node_bootmem(int nodeid,
 786                                  unsigned long start_pfn,
 787                                  unsigned long end_pfn,
 788                                  unsigned long bootmap)
 789 {
 790         unsigned long bootmap_size;
 791
 792         if (start_pfn > max_low_pfn)
 793                 return bootmap;
 794         if (end_pfn > max_low_pfn)
 795                 end_pfn = max_low_pfn;
 796
 797         /* don't touch min_low_pfn */
 798         bootmap_size = init_bootmem_node(NODE_DATA(nodeid),
 799                                          bootmap >> PAGE_SHIFT,
 800                                          start_pfn, end_pfn);
 801         printk(KERN_INFO "  node %d low ram: %08lx - %08lx\n",
 802                 nodeid, start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 803         printk(KERN_INFO "  node %d bootmap %08lx - %08lx\n",
 804                  nodeid, bootmap, bootmap + bootmap_size);
 805         free_bootmem_with_active_regions(nodeid, end_pfn);
 806         early_res_to_bootmem(start_pfn<<PAGE_SHIFT, end_pfn<<PAGE_SHIFT);
 807
 808         return bootmap + bootmap_size;
 809 }
 810
 811 void __init setup_bootmem_allocator(void)
 812 {
 813         int nodeid;
 814         unsigned long bootmap_size, bootmap;
 815         /*
 816          * Initialize the boot-time allocator (with low memory only):
 817          */
 818         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 819         bootmap = find_e820_area(0, max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 820                                  PAGE_SIZE);
 821         if (bootmap == -1L)
 822                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 823         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 824
 825         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 826                  max_pfn_mapped<<PAGE_SHIFT);
 827         printk(KERN_INFO "  low ram: 0 - %08lx\n", max_low_pfn<<PAGE_SHIFT);
 828
 829 #ifdef CONFIG_NEED_MULTIPLE_NODES
 830         for_each_online_node(nodeid)
 831                 bootmap = setup_node_bootmem(nodeid, node_start_pfn[nodeid],
 832                                         node_end_pfn[nodeid], bootmap);
 833 #else
 834         bootmap = setup_node_bootmem(0, 0, max_low_pfn, bootmap);
 835 #endif
 836
 837         after_bootmem = 1;
 838 }
 839
 840 /*
 841  * paging_init() sets up the page tables - note that the first 8MB are
 842  * already mapped by head.S.
 843  *
 844  * This routines also unmaps the page at virtual kernel address 0, so
 845  * that we can trap those pesky NULL-reference errors in the kernel.
 846  */
 847 void __init paging_init(void)
 848 {
 849         pagetable_init();
 850
 851         __flush_tlb_all();
 852
 853         kmap_init();
 854
 855         /*
 856          * NOTE: at this point the bootmem allocator is fully available.
 857          */
 858         sparse_init();
 859         zone_sizes_init();
 860 }
 861
 862 /*
 863  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 864  * and also on some strange 486's. All 586+'s are OK. This used to involve
 865  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 866  * switch to using exceptions got rid of all that.
 867  */
 868 static void __init test_wp_bit(void)
 869 {
 870         printk(KERN_INFO
 871   "Checking if this processor honours the WP bit even in supervisor mode...");
 872
 873         /* Any page-aligned address will do, the test is non-destructive */
 874         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 875         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 876         clear_fixmap(FIX_WP_TEST);
 877
 878         if (!boot_cpu_data.wp_works_ok) {
 879                 printk(KERN_CONT "No.\n");
 880 #ifdef CONFIG_X86_WP_WORKS_OK
 881                 panic(
 882   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 883 #endif
 884         } else {
 885                 printk(KERN_CONT "Ok.\n");
 886         }
 887 }
 888
 889 static struct kcore_list kcore_mem, kcore_vmalloc;
 890
 891 void __init mem_init(void)
 892 {
 893         int codesize, reservedpages, datasize, initsize;
 894         int tmp;
 895
 896         pci_iommu_alloc();
 897
 898 #ifdef CONFIG_FLATMEM
 899         BUG_ON(!mem_map);
 900 #endif
 901         /* this will put all low memory onto the freelists */
 902         totalram_pages += free_all_bootmem();
 903
 904         reservedpages = 0;
 905         for (tmp = 0; tmp < max_low_pfn; tmp++)
 906                 /*
 907                  * Only count reserved RAM pages:
 908                  */
 909                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 910                         reservedpages++;
 911
 912         set_highmem_pages_init();
 913
 914         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 915         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 916         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 917
 918         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 919         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 920                    VMALLOC_END-VMALLOC_START);
 921
 922         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 923                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 924                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 925                 num_physpages << (PAGE_SHIFT-10),
 926                 codesize >> 10,
 927                 reservedpages << (PAGE_SHIFT-10),
 928                 datasize >> 10,
 929                 initsize >> 10,
 930                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 931                );
 932
 933         printk(KERN_INFO "virtual kernel memory layout:\n"
 934                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 935 #ifdef CONFIG_HIGHMEM
 936                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 937 #endif
 938                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 939                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 940                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 941                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 942                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 943                 FIXADDR_START, FIXADDR_TOP,
 944                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 945
 946 #ifdef CONFIG_HIGHMEM
 947                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 948                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 949 #endif
 950
 951                 VMALLOC_START, VMALLOC_END,
 952                 (VMALLOC_END - VMALLOC_START) >> 20,
 953
 954                 (unsigned long)__va(0), (unsigned long)high_memory,
 955                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 956
 957                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 958                 ((unsigned long)&__init_end -
 959                  (unsigned long)&__init_begin) >> 10,
 960
 961                 (unsigned long)&_etext, (unsigned long)&_edata,
 962                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 963
 964                 (unsigned long)&_text, (unsigned long)&_etext,
 965                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 966
 967         /*
 968          * Check boundaries twice: Some fundamental inconsistencies can
 969          * be detected at build time already.
 970          */
 971 #define __FIXADDR_TOP (-PAGE_SIZE)
 972 #ifdef CONFIG_HIGHMEM
 973         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
 974         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
 975 #endif
 976 #define high_memory (-128UL << 20)
 977         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
 978 #undef high_memory
 979 #undef __FIXADDR_TOP
 980
 981 #ifdef CONFIG_HIGHMEM
 982         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 983         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 984 #endif
 985         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
 986         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 987
 988         if (boot_cpu_data.wp_works_ok < 0)
 989                 test_wp_bit();
 990
 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(nid, 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 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1074                                    int flags)
1075 {
1076         return reserve_bootmem(phys, len, flags);
1077 }