arch/x86/mm/init_32.c

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