git.oblomov.eu Git - 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
  53 unsigned int __VMALLOC_RESERVE = 128 << 20;
  54
  55 unsigned long max_low_pfn_mapped;
  56 unsigned long max_pfn_mapped;
  57
  58 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  59 unsigned long highstart_pfn, highend_pfn;
  60
  61 static noinline int do_test_wp_bit(void);
  62
  63
  64 static unsigned long __initdata table_start;
  65 static unsigned long __meminitdata table_end;
  66 static unsigned long __meminitdata table_top;
  67
  68 static int __initdata after_init_bootmem;
  69
  70 static __init void *alloc_low_page(void)
  71 {
  72         unsigned long pfn = table_end++;
  73         void *adr;
  74
  75         if (pfn >= table_top)
  76                 panic("alloc_low_page: ran out of memory");
  77
  78         adr = __va(pfn * PAGE_SIZE);
  79         memset(adr, 0, PAGE_SIZE);
  80         return adr;
  81 }
  82
  83 /*
  84  * Creates a middle page table and puts a pointer to it in the
  85  * given global directory entry. This only returns the gd entry
  86  * in non-PAE compilation mode, since the middle layer is folded.
  87  */
  88 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  89 {
  90         pud_t *pud;
  91         pmd_t *pmd_table;
  92
  93 #ifdef CONFIG_X86_PAE
  94         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
  95                 if (after_init_bootmem)
  96                         pmd_table = (pmd_t *)alloc_bootmem_low_pages(PAGE_SIZE);
  97                 else
  98                         pmd_table = (pmd_t *)alloc_low_page();
  99                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
 100                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
 101                 pud = pud_offset(pgd, 0);
 102                 BUG_ON(pmd_table != pmd_offset(pud, 0));
 103
 104                 return pmd_table;
 105         }
 106 #endif
 107         pud = pud_offset(pgd, 0);
 108         pmd_table = pmd_offset(pud, 0);
 109
 110         return pmd_table;
 111 }
 112
 113 /*
 114  * Create a page table and place a pointer to it in a middle page
 115  * directory entry:
 116  */
 117 static pte_t * __init one_page_table_init(pmd_t *pmd)
 118 {
 119         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
 120                 pte_t *page_table = NULL;
 121
 122                 if (after_init_bootmem) {
 123 #ifdef CONFIG_DEBUG_PAGEALLOC
 124                         page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
 125 #endif
 126                         if (!page_table)
 127                                 page_table =
 128                                 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 129                 } else
 130                         page_table = (pte_t *)alloc_low_page();
 131
 132                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
 133                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
 134                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
 135         }
 136
 137         return pte_offset_kernel(pmd, 0);
 138 }
 139
 140 static pte_t *__init page_table_kmap_check(pte_t *pte, pmd_t *pmd,
 141                                            unsigned long vaddr, pte_t *lastpte)
 142 {
 143 #ifdef CONFIG_HIGHMEM
 144         /*
 145          * Something (early fixmap) may already have put a pte
 146          * page here, which causes the page table allocation
 147          * to become nonlinear. Attempt to fix it, and if it
 148          * is still nonlinear then we have to bug.
 149          */
 150         int pmd_idx_kmap_begin = fix_to_virt(FIX_KMAP_END) >> PMD_SHIFT;
 151         int pmd_idx_kmap_end = fix_to_virt(FIX_KMAP_BEGIN) >> PMD_SHIFT;
 152
 153         if (pmd_idx_kmap_begin != pmd_idx_kmap_end
 154             && (vaddr >> PMD_SHIFT) >= pmd_idx_kmap_begin
 155             && (vaddr >> PMD_SHIFT) <= pmd_idx_kmap_end
 156             && ((__pa(pte) >> PAGE_SHIFT) < table_start
 157                 || (__pa(pte) >> PAGE_SHIFT) >= table_end)) {
 158                 pte_t *newpte;
 159                 int i;
 160
 161                 BUG_ON(after_init_bootmem);
 162                 newpte = alloc_low_page();
 163                 for (i = 0; i < PTRS_PER_PTE; i++)
 164                         set_pte(newpte + i, pte[i]);
 165
 166                 paravirt_alloc_pte(&init_mm, __pa(newpte) >> PAGE_SHIFT);
 167                 set_pmd(pmd, __pmd(__pa(newpte)|_PAGE_TABLE));
 168                 BUG_ON(newpte != pte_offset_kernel(pmd, 0));
 169                 __flush_tlb_all();
 170
 171                 paravirt_release_pte(__pa(pte) >> PAGE_SHIFT);
 172                 pte = newpte;
 173         }
 174         BUG_ON(vaddr < fix_to_virt(FIX_KMAP_BEGIN - 1)
 175                && vaddr > fix_to_virt(FIX_KMAP_END)
 176                && lastpte && lastpte + PTRS_PER_PTE != pte);
 177 #endif
 178         return pte;
 179 }
 180
 181 /*
 182  * This function initializes a certain range of kernel virtual memory
 183  * with new bootmem page tables, everywhere page tables are missing in
 184  * the given range.
 185  *
 186  * NOTE: The pagetables are allocated contiguous on the physical space
 187  * so we can cache the place of the first one and move around without
 188  * checking the pgd every time.
 189  */
 190 static void __init
 191 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 192 {
 193         int pgd_idx, pmd_idx;
 194         unsigned long vaddr;
 195         pgd_t *pgd;
 196         pmd_t *pmd;
 197         pte_t *pte = NULL;
 198
 199         vaddr = start;
 200         pgd_idx = pgd_index(vaddr);
 201         pmd_idx = pmd_index(vaddr);
 202         pgd = pgd_base + pgd_idx;
 203
 204         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 205                 pmd = one_md_table_init(pgd);
 206                 pmd = pmd + pmd_index(vaddr);
 207                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 208                                                         pmd++, pmd_idx++) {
 209                         pte = page_table_kmap_check(one_page_table_init(pmd),
 210                                                     pmd, vaddr, pte);
 211
 212                         vaddr += PMD_SIZE;
 213                 }
 214                 pmd_idx = 0;
 215         }
 216 }
 217
 218 static inline int is_kernel_text(unsigned long addr)
 219 {
 220         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 221                 return 1;
 222         return 0;
 223 }
 224
 225 /*
 226  * This maps the physical memory to kernel virtual address space, a total
 227  * of max_low_pfn pages, by creating page tables starting from address
 228  * PAGE_OFFSET:
 229  */
 230 static void __init kernel_physical_mapping_init(pgd_t *pgd_base,
 231                                                 unsigned long start_pfn,
 232                                                 unsigned long end_pfn,
 233                                                 int use_pse)
 234 {
 235         int pgd_idx, pmd_idx, pte_ofs;
 236         unsigned long pfn;
 237         pgd_t *pgd;
 238         pmd_t *pmd;
 239         pte_t *pte;
 240         unsigned pages_2m, pages_4k;
 241         int mapping_iter;
 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 }
 358
 359 /*
 360  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
 361  * is valid. The argument is a physical page number.
 362  *
 363  *
 364  * On x86, access has to be given to the first megabyte of ram because that area
 365  * contains bios code and data regions used by X and dosemu and similar apps.
 366  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 367  * mmio resources as well as potential bios/acpi data regions.
 368  */
 369 int devmem_is_allowed(unsigned long pagenr)
 370 {
 371         if (pagenr <= 256)
 372                 return 1;
 373         if (iomem_is_exclusive(pagenr << PAGE_SHIFT))
 374                 return 0;
 375         if (!page_is_ram(pagenr))
 376                 return 1;
 377         return 0;
 378 }
 379
 380 pte_t *kmap_pte;
 381 pgprot_t kmap_prot;
 382
 383 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 384 {
 385         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 386                         vaddr), vaddr), vaddr);
 387 }
 388
 389 static void __init kmap_init(void)
 390 {
 391         unsigned long kmap_vstart;
 392
 393         /*
 394          * Cache the first kmap pte:
 395          */
 396         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 397         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 398
 399         kmap_prot = PAGE_KERNEL;
 400 }
 401
 402 #ifdef CONFIG_HIGHMEM
 403 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 404 {
 405         unsigned long vaddr;
 406         pgd_t *pgd;
 407         pud_t *pud;
 408         pmd_t *pmd;
 409         pte_t *pte;
 410
 411         vaddr = PKMAP_BASE;
 412         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 413
 414         pgd = swapper_pg_dir + pgd_index(vaddr);
 415         pud = pud_offset(pgd, vaddr);
 416         pmd = pmd_offset(pud, vaddr);
 417         pte = pte_offset_kernel(pmd, vaddr);
 418         pkmap_page_table = pte;
 419 }
 420
 421 static void __init add_one_highpage_init(struct page *page, int pfn)
 422 {
 423         ClearPageReserved(page);
 424         init_page_count(page);
 425         __free_page(page);
 426         totalhigh_pages++;
 427 }
 428
 429 struct add_highpages_data {
 430         unsigned long start_pfn;
 431         unsigned long end_pfn;
 432 };
 433
 434 static int __init add_highpages_work_fn(unsigned long start_pfn,
 435                                          unsigned long end_pfn, void *datax)
 436 {
 437         int node_pfn;
 438         struct page *page;
 439         unsigned long final_start_pfn, final_end_pfn;
 440         struct add_highpages_data *data;
 441
 442         data = (struct add_highpages_data *)datax;
 443
 444         final_start_pfn = max(start_pfn, data->start_pfn);
 445         final_end_pfn = min(end_pfn, data->end_pfn);
 446         if (final_start_pfn >= final_end_pfn)
 447                 return 0;
 448
 449         for (node_pfn = final_start_pfn; node_pfn < final_end_pfn;
 450              node_pfn++) {
 451                 if (!pfn_valid(node_pfn))
 452                         continue;
 453                 page = pfn_to_page(node_pfn);
 454                 add_one_highpage_init(page, node_pfn);
 455         }
 456
 457         return 0;
 458
 459 }
 460
 461 void __init add_highpages_with_active_regions(int nid, unsigned long start_pfn,
 462                                               unsigned long end_pfn)
 463 {
 464         struct add_highpages_data data;
 465
 466         data.start_pfn = start_pfn;
 467         data.end_pfn = end_pfn;
 468
 469         work_with_active_regions(nid, add_highpages_work_fn, &data);
 470 }
 471
 472 #ifndef CONFIG_NUMA
 473 static void __init set_highmem_pages_init(void)
 474 {
 475         add_highpages_with_active_regions(0, highstart_pfn, highend_pfn);
 476
 477         totalram_pages += totalhigh_pages;
 478 }
 479 #endif /* !CONFIG_NUMA */
 480
 481 #else
 482 static inline void permanent_kmaps_init(pgd_t *pgd_base)
 483 {
 484 }
 485 static inline void set_highmem_pages_init(void)
 486 {
 487 }
 488 #endif /* CONFIG_HIGHMEM */
 489
 490 void __init native_pagetable_setup_start(pgd_t *base)
 491 {
 492         unsigned long pfn, va;
 493         pgd_t *pgd;
 494         pud_t *pud;
 495         pmd_t *pmd;
 496         pte_t *pte;
 497
 498         /*
 499          * Remove any mappings which extend past the end of physical
 500          * memory from the boot time page table:
 501          */
 502         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 503                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 504                 pgd = base + pgd_index(va);
 505                 if (!pgd_present(*pgd))
 506                         break;
 507
 508                 pud = pud_offset(pgd, va);
 509                 pmd = pmd_offset(pud, va);
 510                 if (!pmd_present(*pmd))
 511                         break;
 512
 513                 pte = pte_offset_kernel(pmd, va);
 514                 if (!pte_present(*pte))
 515                         break;
 516
 517                 pte_clear(NULL, va, pte);
 518         }
 519         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 520 }
 521
 522 void __init native_pagetable_setup_done(pgd_t *base)
 523 {
 524 }
 525
 526 /*
 527  * Build a proper pagetable for the kernel mappings.  Up until this
 528  * point, we've been running on some set of pagetables constructed by
 529  * the boot process.
 530  *
 531  * If we're booting on native hardware, this will be a pagetable
 532  * constructed in arch/x86/kernel/head_32.S.  The root of the
 533  * pagetable will be swapper_pg_dir.
 534  *
 535  * If we're booting paravirtualized under a hypervisor, then there are
 536  * more options: we may already be running PAE, and the pagetable may
 537  * or may not be based in swapper_pg_dir.  In any case,
 538  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 539  * appropriately for the rest of the initialization to work.
 540  *
 541  * In general, pagetable_init() assumes that the pagetable may already
 542  * be partially populated, and so it avoids stomping on any existing
 543  * mappings.
 544  */
 545 static void __init early_ioremap_page_table_range_init(pgd_t *pgd_base)
 546 {
 547         unsigned long vaddr, end;
 548
 549         /*
 550          * Fixed mappings, only the page table structure has to be
 551          * created - mappings will be set by set_fixmap():
 552          */
 553         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 554         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 555         page_table_range_init(vaddr, end, pgd_base);
 556         early_ioremap_reset();
 557 }
 558
 559 static void __init pagetable_init(void)
 560 {
 561         pgd_t *pgd_base = swapper_pg_dir;
 562
 563         permanent_kmaps_init(pgd_base);
 564 }
 565
 566 #ifdef CONFIG_ACPI_SLEEP
 567 /*
 568  * ACPI suspend needs this for resume, because things like the intel-agp
 569  * driver might have split up a kernel 4MB mapping.
 570  */
 571 char swsusp_pg_dir[PAGE_SIZE]
 572         __attribute__ ((aligned(PAGE_SIZE)));
 573
 574 static inline void save_pg_dir(void)
 575 {
 576         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 577 }
 578 #else /* !CONFIG_ACPI_SLEEP */
 579 static inline void save_pg_dir(void)
 580 {
 581 }
 582 #endif /* !CONFIG_ACPI_SLEEP */
 583
 584 void zap_low_mappings(void)
 585 {
 586         int i;
 587
 588         /*
 589          * Zap initial low-memory mappings.
 590          *
 591          * Note that "pgd_clear()" doesn't do it for
 592          * us, because pgd_clear() is a no-op on i386.
 593          */
 594         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 595 #ifdef CONFIG_X86_PAE
 596                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 597 #else
 598                 set_pgd(swapper_pg_dir+i, __pgd(0));
 599 #endif
 600         }
 601         flush_tlb_all();
 602 }
 603
 604 int nx_enabled;
 605
 606 pteval_t __supported_pte_mask __read_mostly = ~(_PAGE_NX | _PAGE_GLOBAL | _PAGE_IOMAP);
 607 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 608
 609 #ifdef CONFIG_X86_PAE
 610
 611 static int disable_nx __initdata;
 612
 613 /*
 614  * noexec = on|off
 615  *
 616  * Control non executable mappings.
 617  *
 618  * on      Enable
 619  * off     Disable
 620  */
 621 static int __init noexec_setup(char *str)
 622 {
 623         if (!str || !strcmp(str, "on")) {
 624                 if (cpu_has_nx) {
 625                         __supported_pte_mask |= _PAGE_NX;
 626                         disable_nx = 0;
 627                 }
 628         } else {
 629                 if (!strcmp(str, "off")) {
 630                         disable_nx = 1;
 631                         __supported_pte_mask &= ~_PAGE_NX;
 632                 } else {
 633                         return -EINVAL;
 634                 }
 635         }
 636
 637         return 0;
 638 }
 639 early_param("noexec", noexec_setup);
 640
 641 static void __init set_nx(void)
 642 {
 643         unsigned int v[4], l, h;
 644
 645         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 646                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 647
 648                 if ((v[3] & (1 << 20)) && !disable_nx) {
 649                         rdmsr(MSR_EFER, l, h);
 650                         l |= EFER_NX;
 651                         wrmsr(MSR_EFER, l, h);
 652                         nx_enabled = 1;
 653                         __supported_pte_mask |= _PAGE_NX;
 654                 }
 655         }
 656 }
 657 #endif
 658
 659 /* user-defined highmem size */
 660 static unsigned int highmem_pages = -1;
 661
 662 /*
 663  * highmem=size forces highmem to be exactly 'size' bytes.
 664  * This works even on boxes that have no highmem otherwise.
 665  * This also works to reduce highmem size on bigger boxes.
 666  */
 667 static int __init parse_highmem(char *arg)
 668 {
 669         if (!arg)
 670                 return -EINVAL;
 671
 672         highmem_pages = memparse(arg, &arg) >> PAGE_SHIFT;
 673         return 0;
 674 }
 675 early_param("highmem", parse_highmem);
 676
 677 /*
 678  * Determine low and high memory ranges:
 679  */
 680 void __init find_low_pfn_range(void)
 681 {
 682         /* it could update max_pfn */
 683
 684         /* max_low_pfn is 0, we already have early_res support */
 685
 686         max_low_pfn = max_pfn;
 687         if (max_low_pfn > MAXMEM_PFN) {
 688                 if (highmem_pages == -1)
 689                         highmem_pages = max_pfn - MAXMEM_PFN;
 690                 if (highmem_pages + MAXMEM_PFN < max_pfn)
 691                         max_pfn = MAXMEM_PFN + highmem_pages;
 692                 if (highmem_pages + MAXMEM_PFN > max_pfn) {
 693                         printk(KERN_WARNING "only %luMB highmem pages "
 694                                 "available, ignoring highmem size of %uMB.\n",
 695                                 pages_to_mb(max_pfn - MAXMEM_PFN),
 696                                 pages_to_mb(highmem_pages));
 697                         highmem_pages = 0;
 698                 }
 699                 max_low_pfn = MAXMEM_PFN;
 700 #ifndef CONFIG_HIGHMEM
 701                 /* Maximum memory usable is what is directly addressable */
 702                 printk(KERN_WARNING "Warning only %ldMB will be used.\n",
 703                                         MAXMEM>>20);
 704                 if (max_pfn > MAX_NONPAE_PFN)
 705                         printk(KERN_WARNING
 706                                  "Use a HIGHMEM64G enabled kernel.\n");
 707                 else
 708                         printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
 709                 max_pfn = MAXMEM_PFN;
 710 #else /* !CONFIG_HIGHMEM */
 711 #ifndef CONFIG_HIGHMEM64G
 712                 if (max_pfn > MAX_NONPAE_PFN) {
 713                         max_pfn = MAX_NONPAE_PFN;
 714                         printk(KERN_WARNING "Warning only 4GB will be used."
 715                                 "Use a HIGHMEM64G enabled kernel.\n");
 716                 }
 717 #endif /* !CONFIG_HIGHMEM64G */
 718 #endif /* !CONFIG_HIGHMEM */
 719         } else {
 720                 if (highmem_pages == -1)
 721                         highmem_pages = 0;
 722 #ifdef CONFIG_HIGHMEM
 723                 if (highmem_pages >= max_pfn) {
 724                         printk(KERN_ERR "highmem size specified (%uMB) is "
 725                                 "bigger than pages available (%luMB)!.\n",
 726                                 pages_to_mb(highmem_pages),
 727                                 pages_to_mb(max_pfn));
 728                         highmem_pages = 0;
 729                 }
 730                 if (highmem_pages) {
 731                         if (max_low_pfn - highmem_pages <
 732                             64*1024*1024/PAGE_SIZE){
 733                                 printk(KERN_ERR "highmem size %uMB results in "
 734                                 "smaller than 64MB lowmem, ignoring it.\n"
 735                                         , pages_to_mb(highmem_pages));
 736                                 highmem_pages = 0;
 737                         }
 738                         max_low_pfn -= highmem_pages;
 739                 }
 740 #else
 741                 if (highmem_pages)
 742                         printk(KERN_ERR "ignoring highmem size on non-highmem"
 743                                         " kernel!\n");
 744 #endif
 745         }
 746 }
 747
 748 #ifndef CONFIG_NEED_MULTIPLE_NODES
 749 void __init initmem_init(unsigned long start_pfn,
 750                                   unsigned long end_pfn)
 751 {
 752 #ifdef CONFIG_HIGHMEM
 753         highstart_pfn = highend_pfn = max_pfn;
 754         if (max_pfn > max_low_pfn)
 755                 highstart_pfn = max_low_pfn;
 756         memory_present(0, 0, highend_pfn);
 757         e820_register_active_regions(0, 0, highend_pfn);
 758         printk(KERN_NOTICE "%ldMB HIGHMEM available.\n",
 759                 pages_to_mb(highend_pfn - highstart_pfn));
 760         num_physpages = highend_pfn;
 761         high_memory = (void *) __va(highstart_pfn * PAGE_SIZE - 1) + 1;
 762 #else
 763         memory_present(0, 0, max_low_pfn);
 764         e820_register_active_regions(0, 0, max_low_pfn);
 765         num_physpages = max_low_pfn;
 766         high_memory = (void *) __va(max_low_pfn * PAGE_SIZE - 1) + 1;
 767 #endif
 768 #ifdef CONFIG_FLATMEM
 769         max_mapnr = num_physpages;
 770 #endif
 771         printk(KERN_NOTICE "%ldMB LOWMEM available.\n",
 772                         pages_to_mb(max_low_pfn));
 773
 774         setup_bootmem_allocator();
 775 }
 776 #endif /* !CONFIG_NEED_MULTIPLE_NODES */
 777
 778 static void __init zone_sizes_init(void)
 779 {
 780         unsigned long max_zone_pfns[MAX_NR_ZONES];
 781         memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
 782         max_zone_pfns[ZONE_DMA] =
 783                 virt_to_phys((char *)MAX_DMA_ADDRESS) >> PAGE_SHIFT;
 784         max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
 785 #ifdef CONFIG_HIGHMEM
 786         max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
 787 #endif
 788
 789         free_area_init_nodes(max_zone_pfns);
 790 }
 791
 792 void __init setup_bootmem_allocator(void)
 793 {
 794         int i;
 795         unsigned long bootmap_size, bootmap;
 796         /*
 797          * Initialize the boot-time allocator (with low memory only):
 798          */
 799         bootmap_size = bootmem_bootmap_pages(max_low_pfn)<<PAGE_SHIFT;
 800         bootmap = find_e820_area(min_low_pfn<<PAGE_SHIFT,
 801                                  max_pfn_mapped<<PAGE_SHIFT, bootmap_size,
 802                                  PAGE_SIZE);
 803         if (bootmap == -1L)
 804                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 805         reserve_early(bootmap, bootmap + bootmap_size, "BOOTMAP");
 806
 807         /* don't touch min_low_pfn */
 808         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap >> PAGE_SHIFT,
 809                                          min_low_pfn, max_low_pfn);
 810         printk(KERN_INFO "  mapped low ram: 0 - %08lx\n",
 811                  max_pfn_mapped<<PAGE_SHIFT);
 812         printk(KERN_INFO "  low ram: %08lx - %08lx\n",
 813                  min_low_pfn<<PAGE_SHIFT, max_low_pfn<<PAGE_SHIFT);
 814         printk(KERN_INFO "  bootmap %08lx - %08lx\n",
 815                  bootmap, bootmap + bootmap_size);
 816         for_each_online_node(i)
 817                 free_bootmem_with_active_regions(i, max_low_pfn);
 818         early_res_to_bootmem(0, max_low_pfn<<PAGE_SHIFT);
 819
 820         after_init_bootmem = 1;
 821 }
 822
 823 static void __init find_early_table_space(unsigned long end, int use_pse)
 824 {
 825         unsigned long puds, pmds, ptes, tables, start;
 826
 827         puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
 828         tables = PAGE_ALIGN(puds * sizeof(pud_t));
 829
 830         pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
 831         tables += PAGE_ALIGN(pmds * sizeof(pmd_t));
 832
 833         if (use_pse) {
 834                 unsigned long extra;
 835
 836                 extra = end - ((end>>PMD_SHIFT) << PMD_SHIFT);
 837                 extra += PMD_SIZE;
 838                 ptes = (extra + PAGE_SIZE - 1) >> PAGE_SHIFT;
 839         } else
 840                 ptes = (end + PAGE_SIZE - 1) >> PAGE_SHIFT;
 841
 842         tables += PAGE_ALIGN(ptes * sizeof(pte_t));
 843
 844         /* for fixmap */
 845         tables += PAGE_ALIGN(__end_of_fixed_addresses * sizeof(pte_t));
 846
 847         /*
 848          * RED-PEN putting page tables only on node 0 could
 849          * cause a hotspot and fill up ZONE_DMA. The page tables
 850          * need roughly 0.5KB per GB.
 851          */
 852         start = 0x7000;
 853         table_start = find_e820_area(start, max_pfn_mapped<<PAGE_SHIFT,
 854                                         tables, PAGE_SIZE);
 855         if (table_start == -1UL)
 856                 panic("Cannot find space for the kernel page tables");
 857
 858         table_start >>= PAGE_SHIFT;
 859         table_end = table_start;
 860         table_top = table_start + (tables>>PAGE_SHIFT);
 861
 862         printk(KERN_DEBUG "kernel direct mapping tables up to %lx @ %lx-%lx\n",
 863                 end, table_start << PAGE_SHIFT,
 864                 (table_start << PAGE_SHIFT) + tables);
 865 }
 866
 867 unsigned long __init_refok init_memory_mapping(unsigned long start,
 868                                                 unsigned long end)
 869 {
 870         pgd_t *pgd_base = swapper_pg_dir;
 871         unsigned long start_pfn, end_pfn;
 872         unsigned long big_page_start;
 873 #ifdef CONFIG_DEBUG_PAGEALLOC
 874         /*
 875          * For CONFIG_DEBUG_PAGEALLOC, identity mapping will use small pages.
 876          * This will simplify cpa(), which otherwise needs to support splitting
 877          * large pages into small in interrupt context, etc.
 878          */
 879         int use_pse = 0;
 880 #else
 881         int use_pse = cpu_has_pse;
 882 #endif
 883
 884         /*
 885          * Find space for the kernel direct mapping tables.
 886          */
 887         if (!after_init_bootmem)
 888                 find_early_table_space(end, use_pse);
 889
 890 #ifdef CONFIG_X86_PAE
 891         set_nx();
 892         if (nx_enabled)
 893                 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
 894 #endif
 895
 896         /* Enable PSE if available */
 897         if (cpu_has_pse)
 898                 set_in_cr4(X86_CR4_PSE);
 899
 900         /* Enable PGE if available */
 901         if (cpu_has_pge) {
 902                 set_in_cr4(X86_CR4_PGE);
 903                 __supported_pte_mask |= _PAGE_GLOBAL;
 904         }
 905
 906         /*
 907          * Don't use a large page for the first 2/4MB of memory
 908          * because there are often fixed size MTRRs in there
 909          * and overlapping MTRRs into large pages can cause
 910          * slowdowns.
 911          */
 912         big_page_start = PMD_SIZE;
 913
 914         if (start < big_page_start) {
 915                 start_pfn = start >> PAGE_SHIFT;
 916                 end_pfn = min(big_page_start>>PAGE_SHIFT, end>>PAGE_SHIFT);
 917         } else {
 918                 /* head is not big page alignment ? */
 919                 start_pfn = start >> PAGE_SHIFT;
 920                 end_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
 921                                  << (PMD_SHIFT - PAGE_SHIFT);
 922         }
 923         if (start_pfn < end_pfn)
 924                 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn, 0);
 925
 926         /* big page range */
 927         start_pfn = ((start + (PMD_SIZE - 1))>>PMD_SHIFT)
 928                          << (PMD_SHIFT - PAGE_SHIFT);
 929         if (start_pfn < (big_page_start >> PAGE_SHIFT))
 930                 start_pfn =  big_page_start >> PAGE_SHIFT;
 931         end_pfn = (end>>PMD_SHIFT) << (PMD_SHIFT - PAGE_SHIFT);
 932         if (start_pfn < end_pfn)
 933                 kernel_physical_mapping_init(pgd_base, start_pfn, end_pfn,
 934                                              use_pse);
 935
 936         /* tail is not big page alignment ? */
 937         start_pfn = end_pfn;
 938         if (start_pfn > (big_page_start>>PAGE_SHIFT)) {
 939                 end_pfn = end >> PAGE_SHIFT;
 940                 if (start_pfn < end_pfn)
 941                         kernel_physical_mapping_init(pgd_base, start_pfn,
 942                                                          end_pfn, 0);
 943         }
 944
 945         early_ioremap_page_table_range_init(pgd_base);
 946
 947         load_cr3(swapper_pg_dir);
 948
 949         __flush_tlb_all();
 950
 951         if (!after_init_bootmem)
 952                 reserve_early(table_start << PAGE_SHIFT,
 953                                  table_end << PAGE_SHIFT, "PGTABLE");
 954
 955         if (!after_init_bootmem)
 956                 early_memtest(start, end);
 957
 958         return end >> PAGE_SHIFT;
 959 }
 960
 961
 962 /*
 963  * paging_init() sets up the page tables - note that the first 8MB are
 964  * already mapped by head.S.
 965  *
 966  * This routines also unmaps the page at virtual kernel address 0, so
 967  * that we can trap those pesky NULL-reference errors in the kernel.
 968  */
 969 void __init paging_init(void)
 970 {
 971         pagetable_init();
 972
 973         __flush_tlb_all();
 974
 975         kmap_init();
 976
 977         /*
 978          * NOTE: at this point the bootmem allocator is fully available.
 979          */
 980         sparse_init();
 981         zone_sizes_init();
 982 }
 983
 984 /*
 985  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 986  * and also on some strange 486's. All 586+'s are OK. This used to involve
 987  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 988  * switch to using exceptions got rid of all that.
 989  */
 990 static void __init test_wp_bit(void)
 991 {
 992         printk(KERN_INFO
 993   "Checking if this processor honours the WP bit even in supervisor mode...");
 994
 995         /* Any page-aligned address will do, the test is non-destructive */
 996         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 997         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 998         clear_fixmap(FIX_WP_TEST);
 999
1000         if (!boot_cpu_data.wp_works_ok) {
1001                 printk(KERN_CONT "No.\n");
1002 #ifdef CONFIG_X86_WP_WORKS_OK
1003                 panic(
1004   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
1005 #endif
1006         } else {
1007                 printk(KERN_CONT "Ok.\n");
1008         }
1009 }
1010
1011 static struct kcore_list kcore_mem, kcore_vmalloc;
1012
1013 void __init mem_init(void)
1014 {
1015         int codesize, reservedpages, datasize, initsize;
1016         int tmp;
1017
1018         pci_iommu_alloc();
1019
1020 #ifdef CONFIG_FLATMEM
1021         BUG_ON(!mem_map);
1022 #endif
1023         /* this will put all low memory onto the freelists */
1024         totalram_pages += free_all_bootmem();
1025
1026         reservedpages = 0;
1027         for (tmp = 0; tmp < max_low_pfn; tmp++)
1028                 /*
1029                  * Only count reserved RAM pages:
1030                  */
1031                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
1032                         reservedpages++;
1033
1034         set_highmem_pages_init();
1035
1036         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
1037         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
1038         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
1039
1040         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
1041         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
1042                    VMALLOC_END-VMALLOC_START);
1043
1044         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
1045                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
1046                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
1047                 num_physpages << (PAGE_SHIFT-10),
1048                 codesize >> 10,
1049                 reservedpages << (PAGE_SHIFT-10),
1050                 datasize >> 10,
1051                 initsize >> 10,
1052                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
1053                );
1054
1055         printk(KERN_INFO "virtual kernel memory layout:\n"
1056                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1057 #ifdef CONFIG_HIGHMEM
1058                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1059 #endif
1060                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
1061                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
1062                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1063                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
1064                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
1065                 FIXADDR_START, FIXADDR_TOP,
1066                 (FIXADDR_TOP - FIXADDR_START) >> 10,
1067
1068 #ifdef CONFIG_HIGHMEM
1069                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
1070                 (LAST_PKMAP*PAGE_SIZE) >> 10,
1071 #endif
1072
1073                 VMALLOC_START, VMALLOC_END,
1074                 (VMALLOC_END - VMALLOC_START) >> 20,
1075
1076                 (unsigned long)__va(0), (unsigned long)high_memory,
1077                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
1078
1079                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
1080                 ((unsigned long)&__init_end -
1081                  (unsigned long)&__init_begin) >> 10,
1082
1083                 (unsigned long)&_etext, (unsigned long)&_edata,
1084                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
1085
1086                 (unsigned long)&_text, (unsigned long)&_etext,
1087                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
1088
1089         /*
1090          * Check boundaries twice: Some fundamental inconsistencies can
1091          * be detected at build time already.
1092          */
1093 #define __FIXADDR_TOP (-PAGE_SIZE)
1094 #ifdef CONFIG_HIGHMEM
1095         BUILD_BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE  > FIXADDR_START);
1096         BUILD_BUG_ON(VMALLOC_END                        > PKMAP_BASE);
1097 #endif
1098 #define high_memory (-128UL << 20)
1099         BUILD_BUG_ON(VMALLOC_START                      >= VMALLOC_END);
1100 #undef high_memory
1101 #undef __FIXADDR_TOP
1102
1103 #ifdef CONFIG_HIGHMEM
1104         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
1105         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
1106 #endif
1107         BUG_ON(VMALLOC_START                            >= VMALLOC_END);
1108         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
1109
1110         if (boot_cpu_data.wp_works_ok < 0)
1111                 test_wp_bit();
1112
1113         save_pg_dir();
1114         zap_low_mappings();
1115 }
1116
1117 #ifdef CONFIG_MEMORY_HOTPLUG
1118 int arch_add_memory(int nid, u64 start, u64 size)
1119 {
1120         struct pglist_data *pgdata = NODE_DATA(nid);
1121         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
1122         unsigned long start_pfn = start >> PAGE_SHIFT;
1123         unsigned long nr_pages = size >> PAGE_SHIFT;
1124
1125         return __add_pages(nid, zone, start_pfn, nr_pages);
1126 }
1127 #endif
1128
1129 /*
1130  * This function cannot be __init, since exceptions don't work in that
1131  * section.  Put this after the callers, so that it cannot be inlined.
1132  */
1133 static noinline int do_test_wp_bit(void)
1134 {
1135         char tmp_reg;
1136         int flag;
1137
1138         __asm__ __volatile__(
1139                 "       movb %0, %1     \n"
1140                 "1:     movb %1, %0     \n"
1141                 "       xorl %2, %2     \n"
1142                 "2:                     \n"
1143                 _ASM_EXTABLE(1b,2b)
1144                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
1145                  "=q" (tmp_reg),
1146                  "=r" (flag)
1147                 :"2" (1)
1148                 :"memory");
1149
1150         return flag;
1151 }
1152
1153 #ifdef CONFIG_DEBUG_RODATA
1154 const int rodata_test_data = 0xC3;
1155 EXPORT_SYMBOL_GPL(rodata_test_data);
1156
1157 void mark_rodata_ro(void)
1158 {
1159         unsigned long start = PFN_ALIGN(_text);
1160         unsigned long size = PFN_ALIGN(_etext) - start;
1161
1162 #ifndef CONFIG_DYNAMIC_FTRACE
1163         /* Dynamic tracing modifies the kernel text section */
1164         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1165         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
1166                 size >> 10);
1167
1168 #ifdef CONFIG_CPA_DEBUG
1169         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
1170                 start, start+size);
1171         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
1172
1173         printk(KERN_INFO "Testing CPA: write protecting again\n");
1174         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
1175 #endif
1176 #endif /* CONFIG_DYNAMIC_FTRACE */
1177
1178         start += size;
1179         size = (unsigned long)__end_rodata - start;
1180         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1181         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
1182                 size >> 10);
1183         rodata_test();
1184
1185 #ifdef CONFIG_CPA_DEBUG
1186         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
1187         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
1188
1189         printk(KERN_INFO "Testing CPA: write protecting again\n");
1190         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
1191 #endif
1192 }
1193 #endif
1194
1195 void free_init_pages(char *what, unsigned long begin, unsigned long end)
1196 {
1197 #ifdef CONFIG_DEBUG_PAGEALLOC
1198         /*
1199          * If debugging page accesses then do not free this memory but
1200          * mark them not present - any buggy init-section access will
1201          * create a kernel page fault:
1202          */
1203         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
1204                 begin, PAGE_ALIGN(end));
1205         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
1206 #else
1207         unsigned long addr;
1208
1209         /*
1210          * We just marked the kernel text read only above, now that
1211          * we are going to free part of that, we need to make that
1212          * writeable first.
1213          */
1214         set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
1215
1216         for (addr = begin; addr < end; addr += PAGE_SIZE) {
1217                 ClearPageReserved(virt_to_page(addr));
1218                 init_page_count(virt_to_page(addr));
1219                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
1220                 free_page(addr);
1221                 totalram_pages++;
1222         }
1223         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
1224 #endif
1225 }
1226
1227 void free_initmem(void)
1228 {
1229         free_init_pages("unused kernel memory",
1230                         (unsigned long)(&__init_begin),
1231                         (unsigned long)(&__init_end));
1232 }
1233
1234 #ifdef CONFIG_BLK_DEV_INITRD
1235 void free_initrd_mem(unsigned long start, unsigned long end)
1236 {
1237         free_init_pages("initrd memory", start, end);
1238 }
1239 #endif
1240
1241 int __init reserve_bootmem_generic(unsigned long phys, unsigned long len,
1242                                    int flags)
1243 {
1244         return reserve_bootmem(phys, len, flags);
1245 }