arch/x86/mm/init_32.c

   1 /*
   2  *
   3  *  Copyright (C) 1995  Linus Torvalds
   4  *
   5  *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/signal.h>
  10 #include <linux/sched.h>
  11 #include <linux/kernel.h>
  12 #include <linux/errno.h>
  13 #include <linux/string.h>
  14 #include <linux/types.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/mman.h>
  17 #include <linux/mm.h>
  18 #include <linux/hugetlb.h>
  19 #include <linux/swap.h>
  20 #include <linux/smp.h>
  21 #include <linux/init.h>
  22 #include <linux/highmem.h>
  23 #include <linux/pagemap.h>
  24 #include <linux/pfn.h>
  25 #include <linux/poison.h>
  26 #include <linux/bootmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/proc_fs.h>
  29 #include <linux/memory_hotplug.h>
  30 #include <linux/initrd.h>
  31 #include <linux/cpumask.h>
  32
  33 #include <asm/asm.h>
  34 #include <asm/processor.h>
  35 #include <asm/system.h>
  36 #include <asm/uaccess.h>
  37 #include <asm/pgtable.h>
  38 #include <asm/dma.h>
  39 #include <asm/fixmap.h>
  40 #include <asm/e820.h>
  41 #include <asm/apic.h>
  42 #include <asm/bugs.h>
  43 #include <asm/tlb.h>
  44 #include <asm/tlbflush.h>
  45 #include <asm/pgalloc.h>
  46 #include <asm/sections.h>
  47 #include <asm/paravirt.h>
  48 #include <asm/setup.h>
  49 #include <asm/cacheflush.h>
  50
  51 unsigned int __VMALLOC_RESERVE = 128 << 20;
  52
  53 unsigned long max_pfn_mapped;
  54
  55 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  56 unsigned long highstart_pfn, highend_pfn;
  57
  58 static noinline int do_test_wp_bit(void);
  59
  60 /*
  61  * Creates a middle page table and puts a pointer to it in the
  62  * given global directory entry. This only returns the gd entry
  63  * in non-PAE compilation mode, since the middle layer is folded.
  64  */
  65 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  66 {
  67         pud_t *pud;
  68         pmd_t *pmd_table;
  69
  70 #ifdef CONFIG_X86_PAE
  71         if (!(pgd_val(*pgd) & _PAGE_PRESENT)) {
  72                 pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  73
  74                 paravirt_alloc_pmd(&init_mm, __pa(pmd_table) >> PAGE_SHIFT);
  75                 set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  76                 pud = pud_offset(pgd, 0);
  77                 BUG_ON(pmd_table != pmd_offset(pud, 0));
  78         }
  79 #endif
  80         pud = pud_offset(pgd, 0);
  81         pmd_table = pmd_offset(pud, 0);
  82
  83         return pmd_table;
  84 }
  85
  86 /*
  87  * Create a page table and place a pointer to it in a middle page
  88  * directory entry:
  89  */
  90 static pte_t * __init one_page_table_init(pmd_t *pmd)
  91 {
  92         if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
  93                 pte_t *page_table = NULL;
  94
  95 #ifdef CONFIG_DEBUG_PAGEALLOC
  96                 page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
  97 #endif
  98                 if (!page_table) {
  99                         page_table =
 100                                 (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
 101                 }
 102
 103                 paravirt_alloc_pte(&init_mm, __pa(page_table) >> PAGE_SHIFT);
 104                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
 105                 BUG_ON(page_table != pte_offset_kernel(pmd, 0));
 106         }
 107
 108         return pte_offset_kernel(pmd, 0);
 109 }
 110
 111 /*
 112  * This function initializes a certain range of kernel virtual memory
 113  * with new bootmem page tables, everywhere page tables are missing in
 114  * the given range.
 115  *
 116  * NOTE: The pagetables are allocated contiguous on the physical space
 117  * so we can cache the place of the first one and move around without
 118  * checking the pgd every time.
 119  */
 120 static void __init
 121 page_table_range_init(unsigned long start, unsigned long end, pgd_t *pgd_base)
 122 {
 123         int pgd_idx, pmd_idx;
 124         unsigned long vaddr;
 125         pgd_t *pgd;
 126         pmd_t *pmd;
 127
 128         vaddr = start;
 129         pgd_idx = pgd_index(vaddr);
 130         pmd_idx = pmd_index(vaddr);
 131         pgd = pgd_base + pgd_idx;
 132
 133         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 134                 pmd = one_md_table_init(pgd);
 135                 pmd = pmd + pmd_index(vaddr);
 136                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end);
 137                                                         pmd++, pmd_idx++) {
 138                         one_page_table_init(pmd);
 139
 140                         vaddr += PMD_SIZE;
 141                 }
 142                 pmd_idx = 0;
 143         }
 144 }
 145
 146 static inline int is_kernel_text(unsigned long addr)
 147 {
 148         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 149                 return 1;
 150         return 0;
 151 }
 152
 153 /*
 154  * This maps the physical memory to kernel virtual address space, a total
 155  * of max_low_pfn pages, by creating page tables starting from address
 156  * PAGE_OFFSET:
 157  */
 158 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
 159 {
 160         int pgd_idx, pmd_idx, pte_ofs;
 161         unsigned long pfn;
 162         pgd_t *pgd;
 163         pmd_t *pmd;
 164         pte_t *pte;
 165
 166         pgd_idx = pgd_index(PAGE_OFFSET);
 167         pgd = pgd_base + pgd_idx;
 168         pfn = 0;
 169
 170         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 171                 pmd = one_md_table_init(pgd);
 172                 if (pfn >= max_low_pfn)
 173                         continue;
 174
 175                 for (pmd_idx = 0;
 176                      pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn;
 177                      pmd++, pmd_idx++) {
 178                         unsigned int addr = pfn * PAGE_SIZE + PAGE_OFFSET;
 179
 180                         /*
 181                          * Map with big pages if possible, otherwise
 182                          * create normal page tables:
 183                          *
 184                          * Don't use a large page for the first 2/4MB of memory
 185                          * because there are often fixed size MTRRs in there
 186                          * and overlapping MTRRs into large pages can cause
 187                          * slowdowns.
 188                          */
 189                         if (cpu_has_pse && !(pgd_idx == 0 && pmd_idx == 0)) {
 190                                 unsigned int addr2;
 191                                 pgprot_t prot = PAGE_KERNEL_LARGE;
 192
 193                                 addr2 = (pfn + PTRS_PER_PTE-1) * PAGE_SIZE +
 194                                         PAGE_OFFSET + PAGE_SIZE-1;
 195
 196                                 if (is_kernel_text(addr) ||
 197                                     is_kernel_text(addr2))
 198                                         prot = PAGE_KERNEL_LARGE_EXEC;
 199
 200                                 set_pmd(pmd, pfn_pmd(pfn, prot));
 201
 202                                 pfn += PTRS_PER_PTE;
 203                                 max_pfn_mapped = pfn;
 204                                 continue;
 205                         }
 206                         pte = one_page_table_init(pmd);
 207
 208                         for (pte_ofs = 0;
 209                              pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn;
 210                              pte++, pfn++, pte_ofs++, addr += PAGE_SIZE) {
 211                                 pgprot_t prot = PAGE_KERNEL;
 212
 213                                 if (is_kernel_text(addr))
 214                                         prot = PAGE_KERNEL_EXEC;
 215
 216                                 set_pte(pte, pfn_pte(pfn, prot));
 217                         }
 218                         max_pfn_mapped = pfn;
 219                 }
 220         }
 221 }
 222
 223 static inline int page_kills_ppro(unsigned long pagenr)
 224 {
 225         if (pagenr >= 0x70000 && pagenr <= 0x7003F)
 226                 return 1;
 227         return 0;
 228 }
 229
 230 #ifdef CONFIG_HIGHMEM
 231 pte_t *kmap_pte;
 232 pgprot_t kmap_prot;
 233
 234 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 235 {
 236         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 237                         vaddr), vaddr), vaddr);
 238 }
 239
 240 static void __init kmap_init(void)
 241 {
 242         unsigned long kmap_vstart;
 243
 244         /*
 245          * Cache the first kmap pte:
 246          */
 247         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 248         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 249
 250         kmap_prot = PAGE_KERNEL;
 251 }
 252
 253 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 254 {
 255         unsigned long vaddr;
 256         pgd_t *pgd;
 257         pud_t *pud;
 258         pmd_t *pmd;
 259         pte_t *pte;
 260
 261         vaddr = PKMAP_BASE;
 262         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 263
 264         pgd = swapper_pg_dir + pgd_index(vaddr);
 265         pud = pud_offset(pgd, vaddr);
 266         pmd = pmd_offset(pud, vaddr);
 267         pte = pte_offset_kernel(pmd, vaddr);
 268         pkmap_page_table = pte;
 269 }
 270
 271 static void __meminit free_new_highpage(struct page *page)
 272 {
 273         init_page_count(page);
 274         __free_page(page);
 275         totalhigh_pages++;
 276 }
 277
 278 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
 279 {
 280         if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
 281                 ClearPageReserved(page);
 282                 free_new_highpage(page);
 283         } else
 284                 SetPageReserved(page);
 285 }
 286
 287 static int __meminit
 288 add_one_highpage_hotplug(struct page *page, unsigned long pfn)
 289 {
 290         free_new_highpage(page);
 291         totalram_pages++;
 292 #ifdef CONFIG_FLATMEM
 293         max_mapnr = max(pfn, max_mapnr);
 294 #endif
 295         num_physpages++;
 296
 297         return 0;
 298 }
 299
 300 /*
 301  * Not currently handling the NUMA case.
 302  * Assuming single node and all memory that
 303  * has been added dynamically that would be
 304  * onlined here is in HIGHMEM.
 305  */
 306 void __meminit online_page(struct page *page)
 307 {
 308         ClearPageReserved(page);
 309         add_one_highpage_hotplug(page, page_to_pfn(page));
 310 }
 311
 312 #ifndef CONFIG_NUMA
 313 static void __init set_highmem_pages_init(int bad_ppro)
 314 {
 315         int pfn;
 316
 317         for (pfn = highstart_pfn; pfn < highend_pfn; pfn++) {
 318                 /*
 319                  * Holes under sparsemem might not have no mem_map[]:
 320                  */
 321                 if (pfn_valid(pfn))
 322                         add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
 323         }
 324         totalram_pages += totalhigh_pages;
 325 }
 326 #endif /* !CONFIG_NUMA */
 327
 328 #else
 329 # define kmap_init()                            do { } while (0)
 330 # define permanent_kmaps_init(pgd_base)         do { } while (0)
 331 # define set_highmem_pages_init(bad_ppro)       do { } while (0)
 332 #endif /* CONFIG_HIGHMEM */
 333
 334 pteval_t __PAGE_KERNEL = _PAGE_KERNEL;
 335 EXPORT_SYMBOL(__PAGE_KERNEL);
 336
 337 pteval_t __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
 338
 339 void __init native_pagetable_setup_start(pgd_t *base)
 340 {
 341         unsigned long pfn, va;
 342         pgd_t *pgd;
 343         pud_t *pud;
 344         pmd_t *pmd;
 345         pte_t *pte;
 346
 347         /*
 348          * Remove any mappings which extend past the end of physical
 349          * memory from the boot time page table:
 350          */
 351         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 352                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 353                 pgd = base + pgd_index(va);
 354                 if (!pgd_present(*pgd))
 355                         break;
 356
 357                 pud = pud_offset(pgd, va);
 358                 pmd = pmd_offset(pud, va);
 359                 if (!pmd_present(*pmd))
 360                         break;
 361
 362                 pte = pte_offset_kernel(pmd, va);
 363                 if (!pte_present(*pte))
 364                         break;
 365
 366                 pte_clear(NULL, va, pte);
 367         }
 368         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 369 }
 370
 371 void __init native_pagetable_setup_done(pgd_t *base)
 372 {
 373 }
 374
 375 /*
 376  * Build a proper pagetable for the kernel mappings.  Up until this
 377  * point, we've been running on some set of pagetables constructed by
 378  * the boot process.
 379  *
 380  * If we're booting on native hardware, this will be a pagetable
 381  * constructed in arch/x86/kernel/head_32.S.  The root of the
 382  * pagetable will be swapper_pg_dir.
 383  *
 384  * If we're booting paravirtualized under a hypervisor, then there are
 385  * more options: we may already be running PAE, and the pagetable may
 386  * or may not be based in swapper_pg_dir.  In any case,
 387  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 388  * appropriately for the rest of the initialization to work.
 389  *
 390  * In general, pagetable_init() assumes that the pagetable may already
 391  * be partially populated, and so it avoids stomping on any existing
 392  * mappings.
 393  */
 394 static void __init pagetable_init(void)
 395 {
 396         pgd_t *pgd_base = swapper_pg_dir;
 397         unsigned long vaddr, end;
 398
 399         paravirt_pagetable_setup_start(pgd_base);
 400
 401         /* Enable PSE if available */
 402         if (cpu_has_pse)
 403                 set_in_cr4(X86_CR4_PSE);
 404
 405         /* Enable PGE if available */
 406         if (cpu_has_pge) {
 407                 set_in_cr4(X86_CR4_PGE);
 408                 __PAGE_KERNEL |= _PAGE_GLOBAL;
 409                 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
 410         }
 411
 412         kernel_physical_mapping_init(pgd_base);
 413         remap_numa_kva();
 414
 415         /*
 416          * Fixed mappings, only the page table structure has to be
 417          * created - mappings will be set by set_fixmap():
 418          */
 419         early_ioremap_clear();
 420         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 421         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 422         page_table_range_init(vaddr, end, pgd_base);
 423         early_ioremap_reset();
 424
 425         permanent_kmaps_init(pgd_base);
 426
 427         paravirt_pagetable_setup_done(pgd_base);
 428 }
 429
 430 #ifdef CONFIG_ACPI_SLEEP
 431 /*
 432  * ACPI suspend needs this for resume, because things like the intel-agp
 433  * driver might have split up a kernel 4MB mapping.
 434  */
 435 char swsusp_pg_dir[PAGE_SIZE]
 436         __attribute__ ((aligned(PAGE_SIZE)));
 437
 438 static inline void save_pg_dir(void)
 439 {
 440         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 441 }
 442 #else /* !CONFIG_ACPI_SLEEP */
 443 static inline void save_pg_dir(void)
 444 {
 445 }
 446 #endif /* !CONFIG_ACPI_SLEEP */
 447
 448 void zap_low_mappings(void)
 449 {
 450         int i;
 451
 452         save_pg_dir();
 453
 454         /*
 455          * Zap initial low-memory mappings.
 456          *
 457          * Note that "pgd_clear()" doesn't do it for
 458          * us, because pgd_clear() is a no-op on i386.
 459          */
 460         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 461 #ifdef CONFIG_X86_PAE
 462                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 463 #else
 464                 set_pgd(swapper_pg_dir+i, __pgd(0));
 465 #endif
 466         }
 467         flush_tlb_all();
 468 }
 469
 470 int nx_enabled;
 471
 472 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_NX;
 473 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 474
 475 #ifdef CONFIG_X86_PAE
 476
 477 static int disable_nx __initdata;
 478
 479 /*
 480  * noexec = on|off
 481  *
 482  * Control non executable mappings.
 483  *
 484  * on      Enable
 485  * off     Disable
 486  */
 487 static int __init noexec_setup(char *str)
 488 {
 489         if (!str || !strcmp(str, "on")) {
 490                 if (cpu_has_nx) {
 491                         __supported_pte_mask |= _PAGE_NX;
 492                         disable_nx = 0;
 493                 }
 494         } else {
 495                 if (!strcmp(str, "off")) {
 496                         disable_nx = 1;
 497                         __supported_pte_mask &= ~_PAGE_NX;
 498                 } else {
 499                         return -EINVAL;
 500                 }
 501         }
 502
 503         return 0;
 504 }
 505 early_param("noexec", noexec_setup);
 506
 507 static void __init set_nx(void)
 508 {
 509         unsigned int v[4], l, h;
 510
 511         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 512                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 513
 514                 if ((v[3] & (1 << 20)) && !disable_nx) {
 515                         rdmsr(MSR_EFER, l, h);
 516                         l |= EFER_NX;
 517                         wrmsr(MSR_EFER, l, h);
 518                         nx_enabled = 1;
 519                         __supported_pte_mask |= _PAGE_NX;
 520                 }
 521         }
 522 }
 523 #endif
 524
 525 /*
 526  * paging_init() sets up the page tables - note that the first 8MB are
 527  * already mapped by head.S.
 528  *
 529  * This routines also unmaps the page at virtual kernel address 0, so
 530  * that we can trap those pesky NULL-reference errors in the kernel.
 531  */
 532 void __init paging_init(void)
 533 {
 534 #ifdef CONFIG_X86_PAE
 535         set_nx();
 536         if (nx_enabled)
 537                 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
 538 #endif
 539         pagetable_init();
 540
 541         load_cr3(swapper_pg_dir);
 542
 543         __flush_tlb_all();
 544
 545         kmap_init();
 546 }
 547
 548 /*
 549  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 550  * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
 551  * used to involve black magic jumps to work around some nasty CPU bugs,
 552  * but fortunately the switch to using exceptions got rid of all that.
 553  */
 554 static void __init test_wp_bit(void)
 555 {
 556         printk(KERN_INFO
 557   "Checking if this processor honours the WP bit even in supervisor mode...");
 558
 559         /* Any page-aligned address will do, the test is non-destructive */
 560         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 561         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 562         clear_fixmap(FIX_WP_TEST);
 563
 564         if (!boot_cpu_data.wp_works_ok) {
 565                 printk(KERN_CONT "No.\n");
 566 #ifdef CONFIG_X86_WP_WORKS_OK
 567                 panic(
 568   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 569 #endif
 570         } else {
 571                 printk(KERN_CONT "Ok.\n");
 572         }
 573 }
 574
 575 static struct kcore_list kcore_mem, kcore_vmalloc;
 576
 577 void __init mem_init(void)
 578 {
 579         int codesize, reservedpages, datasize, initsize;
 580         int tmp, bad_ppro;
 581
 582 #ifdef CONFIG_FLATMEM
 583         BUG_ON(!mem_map);
 584 #endif
 585         bad_ppro = ppro_with_ram_bug();
 586
 587 #ifdef CONFIG_HIGHMEM
 588         /* check that fixmap and pkmap do not overlap */
 589         if (PKMAP_BASE + LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
 590                 printk(KERN_ERR
 591                         "fixmap and kmap areas overlap - this will crash\n");
 592                 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
 593                                 PKMAP_BASE, PKMAP_BASE + LAST_PKMAP*PAGE_SIZE,
 594                                 FIXADDR_START);
 595                 BUG();
 596         }
 597 #endif
 598         /* this will put all low memory onto the freelists */
 599         totalram_pages += free_all_bootmem();
 600
 601         reservedpages = 0;
 602         for (tmp = 0; tmp < max_low_pfn; tmp++)
 603                 /*
 604                  * Only count reserved RAM pages:
 605                  */
 606                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 607                         reservedpages++;
 608
 609         set_highmem_pages_init(bad_ppro);
 610
 611         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 612         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 613         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 614
 615         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 616         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 617                    VMALLOC_END-VMALLOC_START);
 618
 619         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 620                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 621                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 622                 num_physpages << (PAGE_SHIFT-10),
 623                 codesize >> 10,
 624                 reservedpages << (PAGE_SHIFT-10),
 625                 datasize >> 10,
 626                 initsize >> 10,
 627                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 628                );
 629
 630 #if 1 /* double-sanity-check paranoia */
 631         printk(KERN_INFO "virtual kernel memory layout:\n"
 632                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 633 #ifdef CONFIG_HIGHMEM
 634                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 635 #endif
 636                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 637                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 638                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 639                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 640                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 641                 FIXADDR_START, FIXADDR_TOP,
 642                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 643
 644 #ifdef CONFIG_HIGHMEM
 645                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 646                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 647 #endif
 648
 649                 VMALLOC_START, VMALLOC_END,
 650                 (VMALLOC_END - VMALLOC_START) >> 20,
 651
 652                 (unsigned long)__va(0), (unsigned long)high_memory,
 653                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 654
 655                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 656                 ((unsigned long)&__init_end -
 657                  (unsigned long)&__init_begin) >> 10,
 658
 659                 (unsigned long)&_etext, (unsigned long)&_edata,
 660                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 661
 662                 (unsigned long)&_text, (unsigned long)&_etext,
 663                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 664
 665 #ifdef CONFIG_HIGHMEM
 666         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 667         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 668 #endif
 669         BUG_ON(VMALLOC_START                            > VMALLOC_END);
 670         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 671 #endif /* double-sanity-check paranoia */
 672
 673         if (boot_cpu_data.wp_works_ok < 0)
 674                 test_wp_bit();
 675
 676         cpa_init();
 677
 678         /*
 679          * Subtle. SMP is doing it's boot stuff late (because it has to
 680          * fork idle threads) - but it also needs low mappings for the
 681          * protected-mode entry to work. We zap these entries only after
 682          * the WP-bit has been tested.
 683          */
 684 #ifndef CONFIG_SMP
 685         zap_low_mappings();
 686 #endif
 687 }
 688
 689 #ifdef CONFIG_MEMORY_HOTPLUG
 690 int arch_add_memory(int nid, u64 start, u64 size)
 691 {
 692         struct pglist_data *pgdata = NODE_DATA(nid);
 693         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 694         unsigned long start_pfn = start >> PAGE_SHIFT;
 695         unsigned long nr_pages = size >> PAGE_SHIFT;
 696
 697         return __add_pages(zone, start_pfn, nr_pages);
 698 }
 699 #endif
 700
 701 /*
 702  * This function cannot be __init, since exceptions don't work in that
 703  * section.  Put this after the callers, so that it cannot be inlined.
 704  */
 705 static noinline int do_test_wp_bit(void)
 706 {
 707         char tmp_reg;
 708         int flag;
 709
 710         __asm__ __volatile__(
 711                 "       movb %0, %1     \n"
 712                 "1:     movb %1, %0     \n"
 713                 "       xorl %2, %2     \n"
 714                 "2:                     \n"
 715                 _ASM_EXTABLE(1b,2b)
 716                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 717                  "=q" (tmp_reg),
 718                  "=r" (flag)
 719                 :"2" (1)
 720                 :"memory");
 721
 722         return flag;
 723 }
 724
 725 #ifdef CONFIG_DEBUG_RODATA
 726 const int rodata_test_data = 0xC3;
 727 EXPORT_SYMBOL_GPL(rodata_test_data);
 728
 729 void mark_rodata_ro(void)
 730 {
 731         unsigned long start = PFN_ALIGN(_text);
 732         unsigned long size = PFN_ALIGN(_etext) - start;
 733
 734         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 735         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
 736                 size >> 10);
 737
 738 #ifdef CONFIG_CPA_DEBUG
 739         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
 740                 start, start+size);
 741         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
 742
 743         printk(KERN_INFO "Testing CPA: write protecting again\n");
 744         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
 745 #endif
 746         start += size;
 747         size = (unsigned long)__end_rodata - start;
 748         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 749         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 750                 size >> 10);
 751         rodata_test();
 752
 753 #ifdef CONFIG_CPA_DEBUG
 754         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
 755         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
 756
 757         printk(KERN_INFO "Testing CPA: write protecting again\n");
 758         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 759 #endif
 760 }
 761 #endif
 762
 763 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 764 {
 765 #ifdef CONFIG_DEBUG_PAGEALLOC
 766         /*
 767          * If debugging page accesses then do not free this memory but
 768          * mark them not present - any buggy init-section access will
 769          * create a kernel page fault:
 770          */
 771         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 772                 begin, PAGE_ALIGN(end));
 773         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 774 #else
 775         unsigned long addr;
 776
 777         /*
 778          * We just marked the kernel text read only above, now that
 779          * we are going to free part of that, we need to make that
 780          * writeable first.
 781          */
 782         set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
 783
 784         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 785                 ClearPageReserved(virt_to_page(addr));
 786                 init_page_count(virt_to_page(addr));
 787                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
 788                 free_page(addr);
 789                 totalram_pages++;
 790         }
 791         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 792 #endif
 793 }
 794
 795 void free_initmem(void)
 796 {
 797         free_init_pages("unused kernel memory",
 798                         (unsigned long)(&__init_begin),
 799                         (unsigned long)(&__init_end));
 800 }
 801
 802 #ifdef CONFIG_BLK_DEV_INITRD
 803 void free_initrd_mem(unsigned long start, unsigned long end)
 804 {
 805         free_init_pages("initrd memory", start, end);
 806 }
 807 #endif