arch/i386/mm/init.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/efi.h>
  31 #include <linux/memory_hotplug.h>
  32 #include <linux/initrd.h>
  33 #include <linux/cpumask.h>
  34
  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/tlb.h>
  44 #include <asm/tlbflush.h>
  45 #include <asm/sections.h>
  46
  47 unsigned int __VMALLOC_RESERVE = 128 << 20;
  48
  49 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
  50 unsigned long highstart_pfn, highend_pfn;
  51
  52 static int noinline do_test_wp_bit(void);
  53
  54 /*
  55  * Creates a middle page table and puts a pointer to it in the
  56  * given global directory entry. This only returns the gd entry
  57  * in non-PAE compilation mode, since the middle layer is folded.
  58  */
  59 static pmd_t * __init one_md_table_init(pgd_t *pgd)
  60 {
  61         pud_t *pud;
  62         pmd_t *pmd_table;
  63
  64 #ifdef CONFIG_X86_PAE
  65         pmd_table = (pmd_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  66         paravirt_alloc_pd(__pa(pmd_table) >> PAGE_SHIFT);
  67         set_pgd(pgd, __pgd(__pa(pmd_table) | _PAGE_PRESENT));
  68         pud = pud_offset(pgd, 0);
  69         if (pmd_table != pmd_offset(pud, 0))
  70                 BUG();
  71 #else
  72         pud = pud_offset(pgd, 0);
  73         pmd_table = pmd_offset(pud, 0);
  74 #endif
  75
  76         return pmd_table;
  77 }
  78
  79 /*
  80  * Create a page table and place a pointer to it in a middle page
  81  * directory entry.
  82  */
  83 static pte_t * __init one_page_table_init(pmd_t *pmd)
  84 {
  85         if (pmd_none(*pmd)) {
  86                 pte_t *page_table = (pte_t *) alloc_bootmem_low_pages(PAGE_SIZE);
  87                 paravirt_alloc_pt(__pa(page_table) >> PAGE_SHIFT);
  88                 set_pmd(pmd, __pmd(__pa(page_table) | _PAGE_TABLE));
  89                 if (page_table != pte_offset_kernel(pmd, 0))
  90                         BUG();
  91
  92                 return page_table;
  93         }
  94
  95         return pte_offset_kernel(pmd, 0);
  96 }
  97
  98 /*
  99  * This function initializes a certain range of kernel virtual memory
 100  * with new bootmem page tables, everywhere page tables are missing in
 101  * the given range.
 102  */
 103
 104 /*
 105  * NOTE: The pagetables are allocated contiguous on the physical space
 106  * so we can cache the place of the first one and move around without
 107  * checking the pgd every time.
 108  */
 109 static void __init page_table_range_init (unsigned long start, unsigned long end, pgd_t *pgd_base)
 110 {
 111         pgd_t *pgd;
 112         pud_t *pud;
 113         pmd_t *pmd;
 114         int pgd_idx, pmd_idx;
 115         unsigned long vaddr;
 116
 117         vaddr = start;
 118         pgd_idx = pgd_index(vaddr);
 119         pmd_idx = pmd_index(vaddr);
 120         pgd = pgd_base + pgd_idx;
 121
 122         for ( ; (pgd_idx < PTRS_PER_PGD) && (vaddr != end); pgd++, pgd_idx++) {
 123                 if (pgd_none(*pgd))
 124                         one_md_table_init(pgd);
 125                 pud = pud_offset(pgd, vaddr);
 126                 pmd = pmd_offset(pud, vaddr);
 127                 for (; (pmd_idx < PTRS_PER_PMD) && (vaddr != end); pmd++, pmd_idx++) {
 128                         if (pmd_none(*pmd))
 129                                 one_page_table_init(pmd);
 130
 131                         vaddr += PMD_SIZE;
 132                 }
 133                 pmd_idx = 0;
 134         }
 135 }
 136
 137 static inline int is_kernel_text(unsigned long addr)
 138 {
 139         if (addr >= PAGE_OFFSET && addr <= (unsigned long)__init_end)
 140                 return 1;
 141         return 0;
 142 }
 143
 144 /*
 145  * This maps the physical memory to kernel virtual address space, a total
 146  * of max_low_pfn pages, by creating page tables starting from address
 147  * PAGE_OFFSET.
 148  */
 149 static void __init kernel_physical_mapping_init(pgd_t *pgd_base)
 150 {
 151         unsigned long pfn;
 152         pgd_t *pgd;
 153         pmd_t *pmd;
 154         pte_t *pte;
 155         int pgd_idx, pmd_idx, pte_ofs;
 156
 157         pgd_idx = pgd_index(PAGE_OFFSET);
 158         pgd = pgd_base + pgd_idx;
 159         pfn = 0;
 160
 161         for (; pgd_idx < PTRS_PER_PGD; pgd++, pgd_idx++) {
 162                 pmd = one_md_table_init(pgd);
 163                 if (pfn >= max_low_pfn)
 164                         continue;
 165                 for (pmd_idx = 0; pmd_idx < PTRS_PER_PMD && pfn < max_low_pfn; pmd++, pmd_idx++) {
 166                         unsigned int address = pfn * PAGE_SIZE + PAGE_OFFSET;
 167
 168                         /* Map with big pages if possible, otherwise create normal page tables. */
 169                         if (cpu_has_pse) {
 170                                 unsigned int address2 = (pfn + PTRS_PER_PTE - 1) * PAGE_SIZE + PAGE_OFFSET + PAGE_SIZE-1;
 171
 172                                 if (is_kernel_text(address) || is_kernel_text(address2))
 173                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE_EXEC));
 174                                 else
 175                                         set_pmd(pmd, pfn_pmd(pfn, PAGE_KERNEL_LARGE));
 176                                 pfn += PTRS_PER_PTE;
 177                         } else {
 178                                 pte = one_page_table_init(pmd);
 179
 180                                 for (pte_ofs = 0; pte_ofs < PTRS_PER_PTE && pfn < max_low_pfn; pte++, pfn++, pte_ofs++) {
 181                                                 if (is_kernel_text(address))
 182                                                         set_pte(pte, pfn_pte(pfn, PAGE_KERNEL_EXEC));
 183                                                 else
 184                                                         set_pte(pte, pfn_pte(pfn, PAGE_KERNEL));
 185                                 }
 186                         }
 187                 }
 188         }
 189 }
 190
 191 static inline int page_kills_ppro(unsigned long pagenr)
 192 {
 193         if (pagenr >= 0x70000 && pagenr <= 0x7003F)
 194                 return 1;
 195         return 0;
 196 }
 197
 198 int page_is_ram(unsigned long pagenr)
 199 {
 200         int i;
 201         unsigned long addr, end;
 202
 203         if (efi_enabled) {
 204                 efi_memory_desc_t *md;
 205                 void *p;
 206
 207                 for (p = memmap.map; p < memmap.map_end; p += memmap.desc_size) {
 208                         md = p;
 209                         if (!is_available_memory(md))
 210                                 continue;
 211                         addr = (md->phys_addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 212                         end = (md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT)) >> PAGE_SHIFT;
 213
 214                         if ((pagenr >= addr) && (pagenr < end))
 215                                 return 1;
 216                 }
 217                 return 0;
 218         }
 219
 220         for (i = 0; i < e820.nr_map; i++) {
 221
 222                 if (e820.map[i].type != E820_RAM)       /* not usable memory */
 223                         continue;
 224                 /*
 225                  *      !!!FIXME!!! Some BIOSen report areas as RAM that
 226                  *      are not. Notably the 640->1Mb area. We need a sanity
 227                  *      check here.
 228                  */
 229                 addr = (e820.map[i].addr+PAGE_SIZE-1) >> PAGE_SHIFT;
 230                 end = (e820.map[i].addr+e820.map[i].size) >> PAGE_SHIFT;
 231                 if  ((pagenr >= addr) && (pagenr < end))
 232                         return 1;
 233         }
 234         return 0;
 235 }
 236
 237 #ifdef CONFIG_HIGHMEM
 238 pte_t *kmap_pte;
 239 pgprot_t kmap_prot;
 240
 241 #define kmap_get_fixmap_pte(vaddr)                                      \
 242         pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr), vaddr), (vaddr)), (vaddr))
 243
 244 static void __init kmap_init(void)
 245 {
 246         unsigned long kmap_vstart;
 247
 248         /* cache the first kmap pte */
 249         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 250         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 251
 252         kmap_prot = PAGE_KERNEL;
 253 }
 254
 255 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 256 {
 257         pgd_t *pgd;
 258         pud_t *pud;
 259         pmd_t *pmd;
 260         pte_t *pte;
 261         unsigned long vaddr;
 262
 263         vaddr = PKMAP_BASE;
 264         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 265
 266         pgd = swapper_pg_dir + pgd_index(vaddr);
 267         pud = pud_offset(pgd, vaddr);
 268         pmd = pmd_offset(pud, vaddr);
 269         pte = pte_offset_kernel(pmd, vaddr);
 270         pkmap_page_table = pte;
 271 }
 272
 273 static void __meminit free_new_highpage(struct page *page)
 274 {
 275         init_page_count(page);
 276         __free_page(page);
 277         totalhigh_pages++;
 278 }
 279
 280 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
 281 {
 282         if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
 283                 ClearPageReserved(page);
 284                 free_new_highpage(page);
 285         } else
 286                 SetPageReserved(page);
 287 }
 288
 289 static int __meminit add_one_highpage_hotplug(struct page *page, unsigned long pfn)
 290 {
 291         free_new_highpage(page);
 292         totalram_pages++;
 293 #ifdef CONFIG_FLATMEM
 294         max_mapnr = max(pfn, max_mapnr);
 295 #endif
 296         num_physpages++;
 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
 313 #ifdef CONFIG_NUMA
 314 extern void set_highmem_pages_init(int);
 315 #else
 316 static void __init set_highmem_pages_init(int bad_ppro)
 317 {
 318         int pfn;
 319         for (pfn = highstart_pfn; pfn < highend_pfn; pfn++)
 320                 add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
 321         totalram_pages += totalhigh_pages;
 322 }
 323 #endif /* CONFIG_FLATMEM */
 324
 325 #else
 326 #define kmap_init() do { } while (0)
 327 #define permanent_kmaps_init(pgd_base) do { } while (0)
 328 #define set_highmem_pages_init(bad_ppro) do { } while (0)
 329 #endif /* CONFIG_HIGHMEM */
 330
 331 unsigned long long __PAGE_KERNEL = _PAGE_KERNEL;
 332 EXPORT_SYMBOL(__PAGE_KERNEL);
 333 unsigned long long __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
 334
 335 #ifdef CONFIG_NUMA
 336 extern void __init remap_numa_kva(void);
 337 #else
 338 #define remap_numa_kva() do {} while (0)
 339 #endif
 340
 341 static void __init pagetable_init (void)
 342 {
 343         unsigned long vaddr;
 344         pgd_t *pgd_base = swapper_pg_dir;
 345
 346 #ifdef CONFIG_X86_PAE
 347         int i;
 348         /* Init entries of the first-level page table to the zero page */
 349         for (i = 0; i < PTRS_PER_PGD; i++)
 350                 set_pgd(pgd_base + i, __pgd(__pa(empty_zero_page) | _PAGE_PRESENT));
 351 #else
 352         paravirt_alloc_pd(__pa(swapper_pg_dir) >> PAGE_SHIFT);
 353 #endif
 354
 355         /* Enable PSE if available */
 356         if (cpu_has_pse) {
 357                 set_in_cr4(X86_CR4_PSE);
 358         }
 359
 360         /* Enable PGE if available */
 361         if (cpu_has_pge) {
 362                 set_in_cr4(X86_CR4_PGE);
 363                 __PAGE_KERNEL |= _PAGE_GLOBAL;
 364                 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
 365         }
 366
 367         kernel_physical_mapping_init(pgd_base);
 368         remap_numa_kva();
 369
 370         /*
 371          * Fixed mappings, only the page table structure has to be
 372          * created - mappings will be set by set_fixmap():
 373          */
 374         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 375         page_table_range_init(vaddr, 0, pgd_base);
 376
 377         permanent_kmaps_init(pgd_base);
 378
 379 #ifdef CONFIG_X86_PAE
 380         /*
 381          * Add low memory identity-mappings - SMP needs it when
 382          * starting up on an AP from real-mode. In the non-PAE
 383          * case we already have these mappings through head.S.
 384          * All user-space mappings are explicitly cleared after
 385          * SMP startup.
 386          */
 387         set_pgd(&pgd_base[0], pgd_base[USER_PTRS_PER_PGD]);
 388 #endif
 389 }
 390
 391 #if defined(CONFIG_SOFTWARE_SUSPEND) || defined(CONFIG_ACPI_SLEEP)
 392 /*
 393  * Swap suspend & friends need this for resume because things like the intel-agp
 394  * driver might have split up a kernel 4MB mapping.
 395  */
 396 char __nosavedata swsusp_pg_dir[PAGE_SIZE]
 397         __attribute__ ((aligned (PAGE_SIZE)));
 398
 399 static inline void save_pg_dir(void)
 400 {
 401         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 402 }
 403 #else
 404 static inline void save_pg_dir(void)
 405 {
 406 }
 407 #endif
 408
 409 void zap_low_mappings (void)
 410 {
 411         int i;
 412
 413         save_pg_dir();
 414
 415         /*
 416          * Zap initial low-memory mappings.
 417          *
 418          * Note that "pgd_clear()" doesn't do it for
 419          * us, because pgd_clear() is a no-op on i386.
 420          */
 421         for (i = 0; i < USER_PTRS_PER_PGD; i++)
 422 #ifdef CONFIG_X86_PAE
 423                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 424 #else
 425                 set_pgd(swapper_pg_dir+i, __pgd(0));
 426 #endif
 427         flush_tlb_all();
 428 }
 429
 430 static int disable_nx __initdata = 0;
 431 u64 __supported_pte_mask __read_mostly = ~_PAGE_NX;
 432
 433 /*
 434  * noexec = on|off
 435  *
 436  * Control non executable mappings.
 437  *
 438  * on      Enable
 439  * off     Disable
 440  */
 441 static int __init noexec_setup(char *str)
 442 {
 443         if (!str || !strcmp(str, "on")) {
 444                 if (cpu_has_nx) {
 445                         __supported_pte_mask |= _PAGE_NX;
 446                         disable_nx = 0;
 447                 }
 448         } else if (!strcmp(str,"off")) {
 449                 disable_nx = 1;
 450                 __supported_pte_mask &= ~_PAGE_NX;
 451         } else
 452                 return -EINVAL;
 453
 454         return 0;
 455 }
 456 early_param("noexec", noexec_setup);
 457
 458 int nx_enabled = 0;
 459 #ifdef CONFIG_X86_PAE
 460
 461 static void __init set_nx(void)
 462 {
 463         unsigned int v[4], l, h;
 464
 465         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 466                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 467                 if ((v[3] & (1 << 20)) && !disable_nx) {
 468                         rdmsr(MSR_EFER, l, h);
 469                         l |= EFER_NX;
 470                         wrmsr(MSR_EFER, l, h);
 471                         nx_enabled = 1;
 472                         __supported_pte_mask |= _PAGE_NX;
 473                 }
 474         }
 475 }
 476
 477 /*
 478  * Enables/disables executability of a given kernel page and
 479  * returns the previous setting.
 480  */
 481 int __init set_kernel_exec(unsigned long vaddr, int enable)
 482 {
 483         pte_t *pte;
 484         int ret = 1;
 485
 486         if (!nx_enabled)
 487                 goto out;
 488
 489         pte = lookup_address(vaddr);
 490         BUG_ON(!pte);
 491
 492         if (!pte_exec_kernel(*pte))
 493                 ret = 0;
 494
 495         if (enable)
 496                 pte->pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
 497         else
 498                 pte->pte_high |= 1 << (_PAGE_BIT_NX - 32);
 499         pte_update_defer(&init_mm, vaddr, pte);
 500         __flush_tlb_all();
 501 out:
 502         return ret;
 503 }
 504
 505 #endif
 506
 507 /*
 508  * paging_init() sets up the page tables - note that the first 8MB are
 509  * already mapped by head.S.
 510  *
 511  * This routines also unmaps the page at virtual kernel address 0, so
 512  * that we can trap those pesky NULL-reference errors in the kernel.
 513  */
 514 void __init paging_init(void)
 515 {
 516 #ifdef CONFIG_X86_PAE
 517         set_nx();
 518         if (nx_enabled)
 519                 printk("NX (Execute Disable) protection: active\n");
 520 #endif
 521
 522         pagetable_init();
 523
 524         load_cr3(swapper_pg_dir);
 525
 526 #ifdef CONFIG_X86_PAE
 527         /*
 528          * We will bail out later - printk doesn't work right now so
 529          * the user would just see a hanging kernel.
 530          */
 531         if (cpu_has_pae)
 532                 set_in_cr4(X86_CR4_PAE);
 533 #endif
 534         __flush_tlb_all();
 535
 536         kmap_init();
 537 }
 538
 539 /*
 540  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 541  * and also on some strange 486's (NexGen etc.). All 586+'s are OK. This
 542  * used to involve black magic jumps to work around some nasty CPU bugs,
 543  * but fortunately the switch to using exceptions got rid of all that.
 544  */
 545
 546 static void __init test_wp_bit(void)
 547 {
 548         printk("Checking if this processor honours the WP bit even in supervisor mode... ");
 549
 550         /* Any page-aligned address will do, the test is non-destructive */
 551         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 552         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 553         clear_fixmap(FIX_WP_TEST);
 554
 555         if (!boot_cpu_data.wp_works_ok) {
 556                 printk("No.\n");
 557 #ifdef CONFIG_X86_WP_WORKS_OK
 558                 panic("This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 559 #endif
 560         } else {
 561                 printk("Ok.\n");
 562         }
 563 }
 564
 565 static struct kcore_list kcore_mem, kcore_vmalloc;
 566
 567 void __init mem_init(void)
 568 {
 569         extern int ppro_with_ram_bug(void);
 570         int codesize, reservedpages, datasize, initsize;
 571         int tmp;
 572         int bad_ppro;
 573
 574 #ifdef CONFIG_FLATMEM
 575         BUG_ON(!mem_map);
 576 #endif
 577
 578         bad_ppro = ppro_with_ram_bug();
 579
 580 #ifdef CONFIG_HIGHMEM
 581         /* check that fixmap and pkmap do not overlap */
 582         if (PKMAP_BASE+LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
 583                 printk(KERN_ERR "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, FIXADDR_START);
 586                 BUG();
 587         }
 588 #endif
 589
 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, %dk reserved, %dk data, %dk init, %ldk highmem)\n",
 612                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 613                 num_physpages << (PAGE_SHIFT-10),
 614                 codesize >> 10,
 615                 reservedpages << (PAGE_SHIFT-10),
 616                 datasize >> 10,
 617                 initsize >> 10,
 618                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 619                );
 620
 621 #if 1 /* double-sanity-check paranoia */
 622         printk("virtual kernel memory layout:\n"
 623                "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 624 #ifdef CONFIG_HIGHMEM
 625                "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 626 #endif
 627                "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 628                "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 629                "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 630                "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 631                "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 632                FIXADDR_START, FIXADDR_TOP,
 633                (FIXADDR_TOP - FIXADDR_START) >> 10,
 634
 635 #ifdef CONFIG_HIGHMEM
 636                PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 637                (LAST_PKMAP*PAGE_SIZE) >> 10,
 638 #endif
 639
 640                VMALLOC_START, VMALLOC_END,
 641                (VMALLOC_END - VMALLOC_START) >> 20,
 642
 643                (unsigned long)__va(0), (unsigned long)high_memory,
 644                ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 645
 646                (unsigned long)&__init_begin, (unsigned long)&__init_end,
 647                ((unsigned long)&__init_end - (unsigned long)&__init_begin) >> 10,
 648
 649                (unsigned long)&_etext, (unsigned long)&_edata,
 650                ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 651
 652                (unsigned long)&_text, (unsigned long)&_etext,
 653                ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 654
 655 #ifdef CONFIG_HIGHMEM
 656         BUG_ON(PKMAP_BASE+LAST_PKMAP*PAGE_SIZE > FIXADDR_START);
 657         BUG_ON(VMALLOC_END                     > PKMAP_BASE);
 658 #endif
 659         BUG_ON(VMALLOC_START                   > VMALLOC_END);
 660         BUG_ON((unsigned long)high_memory      > VMALLOC_START);
 661 #endif /* double-sanity-check paranoia */
 662
 663 #ifdef CONFIG_X86_PAE
 664         if (!cpu_has_pae)
 665                 panic("cannot execute a PAE-enabled kernel on a PAE-less CPU!");
 666 #endif
 667         if (boot_cpu_data.wp_works_ok < 0)
 668                 test_wp_bit();
 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
 692 int remove_memory(u64 start, u64 size)
 693 {
 694         return -EINVAL;
 695 }
 696 EXPORT_SYMBOL_GPL(remove_memory);
 697 #endif
 698
 699 struct kmem_cache *pgd_cache;
 700 struct kmem_cache *pmd_cache;
 701
 702 void __init pgtable_cache_init(void)
 703 {
 704         if (PTRS_PER_PMD > 1) {
 705                 pmd_cache = kmem_cache_create("pmd",
 706                                         PTRS_PER_PMD*sizeof(pmd_t),
 707                                         PTRS_PER_PMD*sizeof(pmd_t),
 708                                         0,
 709                                         pmd_ctor,
 710                                         NULL);
 711                 if (!pmd_cache)
 712                         panic("pgtable_cache_init(): cannot create pmd cache");
 713         }
 714         pgd_cache = kmem_cache_create("pgd",
 715                                 PTRS_PER_PGD*sizeof(pgd_t),
 716                                 PTRS_PER_PGD*sizeof(pgd_t),
 717                                 0,
 718                                 pgd_ctor,
 719                                 PTRS_PER_PMD == 1 ? pgd_dtor : NULL);
 720         if (!pgd_cache)
 721                 panic("pgtable_cache_init(): Cannot create pgd cache");
 722 }
 723
 724 /*
 725  * This function cannot be __init, since exceptions don't work in that
 726  * section.  Put this after the callers, so that it cannot be inlined.
 727  */
 728 static int noinline do_test_wp_bit(void)
 729 {
 730         char tmp_reg;
 731         int flag;
 732
 733         __asm__ __volatile__(
 734                 "       movb %0,%1      \n"
 735                 "1:     movb %1,%0      \n"
 736                 "       xorl %2,%2      \n"
 737                 "2:                     \n"
 738                 ".section __ex_table,\"a\"\n"
 739                 "       .align 4        \n"
 740                 "       .long 1b,2b     \n"
 741                 ".previous              \n"
 742                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 743                  "=q" (tmp_reg),
 744                  "=r" (flag)
 745                 :"2" (1)
 746                 :"memory");
 747
 748         return flag;
 749 }
 750
 751 #ifdef CONFIG_DEBUG_RODATA
 752
 753 void mark_rodata_ro(void)
 754 {
 755         unsigned long start = PFN_ALIGN(_text);
 756         unsigned long size = PFN_ALIGN(_etext) - start;
 757
 758 #ifdef CONFIG_HOTPLUG_CPU
 759         /* It must still be possible to apply SMP alternatives. */
 760         if (num_possible_cpus() <= 1)
 761 #endif
 762         {
 763                 change_page_attr(virt_to_page(start),
 764                                  size >> PAGE_SHIFT, PAGE_KERNEL_RX);
 765                 printk("Write protecting the kernel text: %luk\n", size >> 10);
 766         }
 767
 768         start += size;
 769         size = (unsigned long)__end_rodata - start;
 770         change_page_attr(virt_to_page(start),
 771                          size >> PAGE_SHIFT, PAGE_KERNEL_RO);
 772         printk("Write protecting the kernel read-only data: %luk\n",
 773                size >> 10);
 774
 775         /*
 776          * change_page_attr() requires a global_flush_tlb() call after it.
 777          * We do this after the printk so that if something went wrong in the
 778          * change, the printk gets out at least to give a better debug hint
 779          * of who is the culprit.
 780          */
 781         global_flush_tlb();
 782 }
 783 #endif
 784
 785 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 786 {
 787         unsigned long addr;
 788
 789         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 790                 struct page *page = pfn_to_page(addr >> PAGE_SHIFT);
 791                 ClearPageReserved(page);
 792                 init_page_count(page);
 793                 memset(page_address(page), POISON_FREE_INITMEM, PAGE_SIZE);
 794                 __free_page(page);
 795                 totalram_pages++;
 796         }
 797         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 798 }
 799
 800 void free_initmem(void)
 801 {
 802         free_init_pages("unused kernel memory",
 803                         __pa_symbol(&__init_begin),
 804                         __pa_symbol(&__init_end));
 805 }
 806
 807 #ifdef CONFIG_BLK_DEV_INITRD
 808 void free_initrd_mem(unsigned long start, unsigned long end)
 809 {
 810         free_init_pages("initrd memory", __pa(start), __pa(end));
 811 }
 812 #endif
 813