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 /*
 231  * devmem_is_allowed() checks to see if /dev/mem access to a certain address
 232  * is valid. The argument is a physical page number.
 233  *
 234  *
 235  * On x86, access has to be given to the first megabyte of ram because that area
 236  * contains bios code and data regions used by X and dosemu and similar apps.
 237  * Access has to be given to non-kernel-ram areas as well, these contain the PCI
 238  * mmio resources as well as potential bios/acpi data regions.
 239  */
 240 int devmem_is_allowed(unsigned long pagenr)
 241 {
 242         if (pagenr <= 256)
 243                 return 1;
 244         if (!page_is_ram(pagenr))
 245                 return 1;
 246         return 0;
 247 }
 248
 249 #ifdef CONFIG_HIGHMEM
 250 pte_t *kmap_pte;
 251 pgprot_t kmap_prot;
 252
 253 static inline pte_t *kmap_get_fixmap_pte(unsigned long vaddr)
 254 {
 255         return pte_offset_kernel(pmd_offset(pud_offset(pgd_offset_k(vaddr),
 256                         vaddr), vaddr), vaddr);
 257 }
 258
 259 static void __init kmap_init(void)
 260 {
 261         unsigned long kmap_vstart;
 262
 263         /*
 264          * Cache the first kmap pte:
 265          */
 266         kmap_vstart = __fix_to_virt(FIX_KMAP_BEGIN);
 267         kmap_pte = kmap_get_fixmap_pte(kmap_vstart);
 268
 269         kmap_prot = PAGE_KERNEL;
 270 }
 271
 272 static void __init permanent_kmaps_init(pgd_t *pgd_base)
 273 {
 274         unsigned long vaddr;
 275         pgd_t *pgd;
 276         pud_t *pud;
 277         pmd_t *pmd;
 278         pte_t *pte;
 279
 280         vaddr = PKMAP_BASE;
 281         page_table_range_init(vaddr, vaddr + PAGE_SIZE*LAST_PKMAP, pgd_base);
 282
 283         pgd = swapper_pg_dir + pgd_index(vaddr);
 284         pud = pud_offset(pgd, vaddr);
 285         pmd = pmd_offset(pud, vaddr);
 286         pte = pte_offset_kernel(pmd, vaddr);
 287         pkmap_page_table = pte;
 288 }
 289
 290 static void __meminit free_new_highpage(struct page *page)
 291 {
 292         init_page_count(page);
 293         __free_page(page);
 294         totalhigh_pages++;
 295 }
 296
 297 void __init add_one_highpage_init(struct page *page, int pfn, int bad_ppro)
 298 {
 299         if (page_is_ram(pfn) && !(bad_ppro && page_kills_ppro(pfn))) {
 300                 ClearPageReserved(page);
 301                 free_new_highpage(page);
 302         } else
 303                 SetPageReserved(page);
 304 }
 305
 306 static int __meminit
 307 add_one_highpage_hotplug(struct page *page, unsigned long pfn)
 308 {
 309         free_new_highpage(page);
 310         totalram_pages++;
 311 #ifdef CONFIG_FLATMEM
 312         max_mapnr = max(pfn, max_mapnr);
 313 #endif
 314         num_physpages++;
 315
 316         return 0;
 317 }
 318
 319 /*
 320  * Not currently handling the NUMA case.
 321  * Assuming single node and all memory that
 322  * has been added dynamically that would be
 323  * onlined here is in HIGHMEM.
 324  */
 325 void __meminit online_page(struct page *page)
 326 {
 327         ClearPageReserved(page);
 328         add_one_highpage_hotplug(page, page_to_pfn(page));
 329 }
 330
 331 #ifndef CONFIG_NUMA
 332 static void __init set_highmem_pages_init(int bad_ppro)
 333 {
 334         int pfn;
 335
 336         for (pfn = highstart_pfn; pfn < highend_pfn; pfn++) {
 337                 /*
 338                  * Holes under sparsemem might not have no mem_map[]:
 339                  */
 340                 if (pfn_valid(pfn))
 341                         add_one_highpage_init(pfn_to_page(pfn), pfn, bad_ppro);
 342         }
 343         totalram_pages += totalhigh_pages;
 344 }
 345 #endif /* !CONFIG_NUMA */
 346
 347 #else
 348 # define kmap_init()                            do { } while (0)
 349 # define permanent_kmaps_init(pgd_base)         do { } while (0)
 350 # define set_highmem_pages_init(bad_ppro)       do { } while (0)
 351 #endif /* CONFIG_HIGHMEM */
 352
 353 pteval_t __PAGE_KERNEL = _PAGE_KERNEL;
 354 EXPORT_SYMBOL(__PAGE_KERNEL);
 355
 356 pteval_t __PAGE_KERNEL_EXEC = _PAGE_KERNEL_EXEC;
 357
 358 void __init native_pagetable_setup_start(pgd_t *base)
 359 {
 360         unsigned long pfn, va;
 361         pgd_t *pgd;
 362         pud_t *pud;
 363         pmd_t *pmd;
 364         pte_t *pte;
 365
 366         /*
 367          * Remove any mappings which extend past the end of physical
 368          * memory from the boot time page table:
 369          */
 370         for (pfn = max_low_pfn + 1; pfn < 1<<(32-PAGE_SHIFT); pfn++) {
 371                 va = PAGE_OFFSET + (pfn<<PAGE_SHIFT);
 372                 pgd = base + pgd_index(va);
 373                 if (!pgd_present(*pgd))
 374                         break;
 375
 376                 pud = pud_offset(pgd, va);
 377                 pmd = pmd_offset(pud, va);
 378                 if (!pmd_present(*pmd))
 379                         break;
 380
 381                 pte = pte_offset_kernel(pmd, va);
 382                 if (!pte_present(*pte))
 383                         break;
 384
 385                 pte_clear(NULL, va, pte);
 386         }
 387         paravirt_alloc_pmd(&init_mm, __pa(base) >> PAGE_SHIFT);
 388 }
 389
 390 void __init native_pagetable_setup_done(pgd_t *base)
 391 {
 392 }
 393
 394 /*
 395  * Build a proper pagetable for the kernel mappings.  Up until this
 396  * point, we've been running on some set of pagetables constructed by
 397  * the boot process.
 398  *
 399  * If we're booting on native hardware, this will be a pagetable
 400  * constructed in arch/x86/kernel/head_32.S.  The root of the
 401  * pagetable will be swapper_pg_dir.
 402  *
 403  * If we're booting paravirtualized under a hypervisor, then there are
 404  * more options: we may already be running PAE, and the pagetable may
 405  * or may not be based in swapper_pg_dir.  In any case,
 406  * paravirt_pagetable_setup_start() will set up swapper_pg_dir
 407  * appropriately for the rest of the initialization to work.
 408  *
 409  * In general, pagetable_init() assumes that the pagetable may already
 410  * be partially populated, and so it avoids stomping on any existing
 411  * mappings.
 412  */
 413 static void __init pagetable_init(void)
 414 {
 415         pgd_t *pgd_base = swapper_pg_dir;
 416         unsigned long vaddr, end;
 417
 418         paravirt_pagetable_setup_start(pgd_base);
 419
 420         /* Enable PSE if available */
 421         if (cpu_has_pse)
 422                 set_in_cr4(X86_CR4_PSE);
 423
 424         /* Enable PGE if available */
 425         if (cpu_has_pge) {
 426                 set_in_cr4(X86_CR4_PGE);
 427                 __PAGE_KERNEL |= _PAGE_GLOBAL;
 428                 __PAGE_KERNEL_EXEC |= _PAGE_GLOBAL;
 429         }
 430
 431         kernel_physical_mapping_init(pgd_base);
 432         remap_numa_kva();
 433
 434         /*
 435          * Fixed mappings, only the page table structure has to be
 436          * created - mappings will be set by set_fixmap():
 437          */
 438         early_ioremap_clear();
 439         vaddr = __fix_to_virt(__end_of_fixed_addresses - 1) & PMD_MASK;
 440         end = (FIXADDR_TOP + PMD_SIZE - 1) & PMD_MASK;
 441         page_table_range_init(vaddr, end, pgd_base);
 442         early_ioremap_reset();
 443
 444         permanent_kmaps_init(pgd_base);
 445
 446         paravirt_pagetable_setup_done(pgd_base);
 447 }
 448
 449 #ifdef CONFIG_ACPI_SLEEP
 450 /*
 451  * ACPI suspend needs this for resume, because things like the intel-agp
 452  * driver might have split up a kernel 4MB mapping.
 453  */
 454 char swsusp_pg_dir[PAGE_SIZE]
 455         __attribute__ ((aligned(PAGE_SIZE)));
 456
 457 static inline void save_pg_dir(void)
 458 {
 459         memcpy(swsusp_pg_dir, swapper_pg_dir, PAGE_SIZE);
 460 }
 461 #else /* !CONFIG_ACPI_SLEEP */
 462 static inline void save_pg_dir(void)
 463 {
 464 }
 465 #endif /* !CONFIG_ACPI_SLEEP */
 466
 467 void zap_low_mappings(void)
 468 {
 469         int i;
 470
 471         save_pg_dir();
 472
 473         /*
 474          * Zap initial low-memory mappings.
 475          *
 476          * Note that "pgd_clear()" doesn't do it for
 477          * us, because pgd_clear() is a no-op on i386.
 478          */
 479         for (i = 0; i < KERNEL_PGD_BOUNDARY; i++) {
 480 #ifdef CONFIG_X86_PAE
 481                 set_pgd(swapper_pg_dir+i, __pgd(1 + __pa(empty_zero_page)));
 482 #else
 483                 set_pgd(swapper_pg_dir+i, __pgd(0));
 484 #endif
 485         }
 486         flush_tlb_all();
 487 }
 488
 489 int nx_enabled;
 490
 491 pteval_t __supported_pte_mask __read_mostly = ~_PAGE_NX;
 492 EXPORT_SYMBOL_GPL(__supported_pte_mask);
 493
 494 #ifdef CONFIG_X86_PAE
 495
 496 static int disable_nx __initdata;
 497
 498 /*
 499  * noexec = on|off
 500  *
 501  * Control non executable mappings.
 502  *
 503  * on      Enable
 504  * off     Disable
 505  */
 506 static int __init noexec_setup(char *str)
 507 {
 508         if (!str || !strcmp(str, "on")) {
 509                 if (cpu_has_nx) {
 510                         __supported_pte_mask |= _PAGE_NX;
 511                         disable_nx = 0;
 512                 }
 513         } else {
 514                 if (!strcmp(str, "off")) {
 515                         disable_nx = 1;
 516                         __supported_pte_mask &= ~_PAGE_NX;
 517                 } else {
 518                         return -EINVAL;
 519                 }
 520         }
 521
 522         return 0;
 523 }
 524 early_param("noexec", noexec_setup);
 525
 526 static void __init set_nx(void)
 527 {
 528         unsigned int v[4], l, h;
 529
 530         if (cpu_has_pae && (cpuid_eax(0x80000000) > 0x80000001)) {
 531                 cpuid(0x80000001, &v[0], &v[1], &v[2], &v[3]);
 532
 533                 if ((v[3] & (1 << 20)) && !disable_nx) {
 534                         rdmsr(MSR_EFER, l, h);
 535                         l |= EFER_NX;
 536                         wrmsr(MSR_EFER, l, h);
 537                         nx_enabled = 1;
 538                         __supported_pte_mask |= _PAGE_NX;
 539                 }
 540         }
 541 }
 542 #endif
 543
 544 /*
 545  * paging_init() sets up the page tables - note that the first 8MB are
 546  * already mapped by head.S.
 547  *
 548  * This routines also unmaps the page at virtual kernel address 0, so
 549  * that we can trap those pesky NULL-reference errors in the kernel.
 550  */
 551 void __init paging_init(void)
 552 {
 553 #ifdef CONFIG_X86_PAE
 554         set_nx();
 555         if (nx_enabled)
 556                 printk(KERN_INFO "NX (Execute Disable) protection: active\n");
 557 #endif
 558         pagetable_init();
 559
 560         load_cr3(swapper_pg_dir);
 561
 562         __flush_tlb_all();
 563
 564         kmap_init();
 565 }
 566
 567 /*
 568  * Test if the WP bit works in supervisor mode. It isn't supported on 386's
 569  * and also on some strange 486's. All 586+'s are OK. This used to involve
 570  * black magic jumps to work around some nasty CPU bugs, but fortunately the
 571  * switch to using exceptions got rid of all that.
 572  */
 573 static void __init test_wp_bit(void)
 574 {
 575         printk(KERN_INFO
 576   "Checking if this processor honours the WP bit even in supervisor mode...");
 577
 578         /* Any page-aligned address will do, the test is non-destructive */
 579         __set_fixmap(FIX_WP_TEST, __pa(&swapper_pg_dir), PAGE_READONLY);
 580         boot_cpu_data.wp_works_ok = do_test_wp_bit();
 581         clear_fixmap(FIX_WP_TEST);
 582
 583         if (!boot_cpu_data.wp_works_ok) {
 584                 printk(KERN_CONT "No.\n");
 585 #ifdef CONFIG_X86_WP_WORKS_OK
 586                 panic(
 587   "This kernel doesn't support CPU's with broken WP. Recompile it for a 386!");
 588 #endif
 589         } else {
 590                 printk(KERN_CONT "Ok.\n");
 591         }
 592 }
 593
 594 static struct kcore_list kcore_mem, kcore_vmalloc;
 595
 596 void __init mem_init(void)
 597 {
 598         int codesize, reservedpages, datasize, initsize;
 599         int tmp, bad_ppro;
 600
 601 #ifdef CONFIG_FLATMEM
 602         BUG_ON(!mem_map);
 603 #endif
 604         bad_ppro = ppro_with_ram_bug();
 605
 606 #ifdef CONFIG_HIGHMEM
 607         /* check that fixmap and pkmap do not overlap */
 608         if (PKMAP_BASE + LAST_PKMAP*PAGE_SIZE >= FIXADDR_START) {
 609                 printk(KERN_ERR
 610                         "fixmap and kmap areas overlap - this will crash\n");
 611                 printk(KERN_ERR "pkstart: %lxh pkend: %lxh fixstart %lxh\n",
 612                                 PKMAP_BASE, PKMAP_BASE + LAST_PKMAP*PAGE_SIZE,
 613                                 FIXADDR_START);
 614                 BUG();
 615         }
 616 #endif
 617         /* this will put all low memory onto the freelists */
 618         totalram_pages += free_all_bootmem();
 619
 620         reservedpages = 0;
 621         for (tmp = 0; tmp < max_low_pfn; tmp++)
 622                 /*
 623                  * Only count reserved RAM pages:
 624                  */
 625                 if (page_is_ram(tmp) && PageReserved(pfn_to_page(tmp)))
 626                         reservedpages++;
 627
 628         set_highmem_pages_init(bad_ppro);
 629
 630         codesize =  (unsigned long) &_etext - (unsigned long) &_text;
 631         datasize =  (unsigned long) &_edata - (unsigned long) &_etext;
 632         initsize =  (unsigned long) &__init_end - (unsigned long) &__init_begin;
 633
 634         kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
 635         kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
 636                    VMALLOC_END-VMALLOC_START);
 637
 638         printk(KERN_INFO "Memory: %luk/%luk available (%dk kernel code, "
 639                         "%dk reserved, %dk data, %dk init, %ldk highmem)\n",
 640                 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
 641                 num_physpages << (PAGE_SHIFT-10),
 642                 codesize >> 10,
 643                 reservedpages << (PAGE_SHIFT-10),
 644                 datasize >> 10,
 645                 initsize >> 10,
 646                 (unsigned long) (totalhigh_pages << (PAGE_SHIFT-10))
 647                );
 648
 649 #if 1 /* double-sanity-check paranoia */
 650         printk(KERN_INFO "virtual kernel memory layout:\n"
 651                 "    fixmap  : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 652 #ifdef CONFIG_HIGHMEM
 653                 "    pkmap   : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 654 #endif
 655                 "    vmalloc : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 656                 "    lowmem  : 0x%08lx - 0x%08lx   (%4ld MB)\n"
 657                 "      .init : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 658                 "      .data : 0x%08lx - 0x%08lx   (%4ld kB)\n"
 659                 "      .text : 0x%08lx - 0x%08lx   (%4ld kB)\n",
 660                 FIXADDR_START, FIXADDR_TOP,
 661                 (FIXADDR_TOP - FIXADDR_START) >> 10,
 662
 663 #ifdef CONFIG_HIGHMEM
 664                 PKMAP_BASE, PKMAP_BASE+LAST_PKMAP*PAGE_SIZE,
 665                 (LAST_PKMAP*PAGE_SIZE) >> 10,
 666 #endif
 667
 668                 VMALLOC_START, VMALLOC_END,
 669                 (VMALLOC_END - VMALLOC_START) >> 20,
 670
 671                 (unsigned long)__va(0), (unsigned long)high_memory,
 672                 ((unsigned long)high_memory - (unsigned long)__va(0)) >> 20,
 673
 674                 (unsigned long)&__init_begin, (unsigned long)&__init_end,
 675                 ((unsigned long)&__init_end -
 676                  (unsigned long)&__init_begin) >> 10,
 677
 678                 (unsigned long)&_etext, (unsigned long)&_edata,
 679                 ((unsigned long)&_edata - (unsigned long)&_etext) >> 10,
 680
 681                 (unsigned long)&_text, (unsigned long)&_etext,
 682                 ((unsigned long)&_etext - (unsigned long)&_text) >> 10);
 683
 684 #ifdef CONFIG_HIGHMEM
 685         BUG_ON(PKMAP_BASE + LAST_PKMAP*PAGE_SIZE        > FIXADDR_START);
 686         BUG_ON(VMALLOC_END                              > PKMAP_BASE);
 687 #endif
 688         BUG_ON(VMALLOC_START                            > VMALLOC_END);
 689         BUG_ON((unsigned long)high_memory               > VMALLOC_START);
 690 #endif /* double-sanity-check paranoia */
 691
 692         if (boot_cpu_data.wp_works_ok < 0)
 693                 test_wp_bit();
 694
 695         cpa_init();
 696
 697         /*
 698          * Subtle. SMP is doing it's boot stuff late (because it has to
 699          * fork idle threads) - but it also needs low mappings for the
 700          * protected-mode entry to work. We zap these entries only after
 701          * the WP-bit has been tested.
 702          */
 703 #ifndef CONFIG_SMP
 704         zap_low_mappings();
 705 #endif
 706 }
 707
 708 #ifdef CONFIG_MEMORY_HOTPLUG
 709 int arch_add_memory(int nid, u64 start, u64 size)
 710 {
 711         struct pglist_data *pgdata = NODE_DATA(nid);
 712         struct zone *zone = pgdata->node_zones + ZONE_HIGHMEM;
 713         unsigned long start_pfn = start >> PAGE_SHIFT;
 714         unsigned long nr_pages = size >> PAGE_SHIFT;
 715
 716         return __add_pages(zone, start_pfn, nr_pages);
 717 }
 718 #endif
 719
 720 /*
 721  * This function cannot be __init, since exceptions don't work in that
 722  * section.  Put this after the callers, so that it cannot be inlined.
 723  */
 724 static noinline int do_test_wp_bit(void)
 725 {
 726         char tmp_reg;
 727         int flag;
 728
 729         __asm__ __volatile__(
 730                 "       movb %0, %1     \n"
 731                 "1:     movb %1, %0     \n"
 732                 "       xorl %2, %2     \n"
 733                 "2:                     \n"
 734                 _ASM_EXTABLE(1b,2b)
 735                 :"=m" (*(char *)fix_to_virt(FIX_WP_TEST)),
 736                  "=q" (tmp_reg),
 737                  "=r" (flag)
 738                 :"2" (1)
 739                 :"memory");
 740
 741         return flag;
 742 }
 743
 744 #ifdef CONFIG_DEBUG_RODATA
 745 const int rodata_test_data = 0xC3;
 746 EXPORT_SYMBOL_GPL(rodata_test_data);
 747
 748 void mark_rodata_ro(void)
 749 {
 750         unsigned long start = PFN_ALIGN(_text);
 751         unsigned long size = PFN_ALIGN(_etext) - start;
 752
 753         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 754         printk(KERN_INFO "Write protecting the kernel text: %luk\n",
 755                 size >> 10);
 756
 757 #ifdef CONFIG_CPA_DEBUG
 758         printk(KERN_INFO "Testing CPA: Reverting %lx-%lx\n",
 759                 start, start+size);
 760         set_pages_rw(virt_to_page(start), size>>PAGE_SHIFT);
 761
 762         printk(KERN_INFO "Testing CPA: write protecting again\n");
 763         set_pages_ro(virt_to_page(start), size>>PAGE_SHIFT);
 764 #endif
 765         start += size;
 766         size = (unsigned long)__end_rodata - start;
 767         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 768         printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
 769                 size >> 10);
 770         rodata_test();
 771
 772 #ifdef CONFIG_CPA_DEBUG
 773         printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, start + size);
 774         set_pages_rw(virt_to_page(start), size >> PAGE_SHIFT);
 775
 776         printk(KERN_INFO "Testing CPA: write protecting again\n");
 777         set_pages_ro(virt_to_page(start), size >> PAGE_SHIFT);
 778 #endif
 779 }
 780 #endif
 781
 782 void free_init_pages(char *what, unsigned long begin, unsigned long end)
 783 {
 784 #ifdef CONFIG_DEBUG_PAGEALLOC
 785         /*
 786          * If debugging page accesses then do not free this memory but
 787          * mark them not present - any buggy init-section access will
 788          * create a kernel page fault:
 789          */
 790         printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
 791                 begin, PAGE_ALIGN(end));
 792         set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
 793 #else
 794         unsigned long addr;
 795
 796         /*
 797          * We just marked the kernel text read only above, now that
 798          * we are going to free part of that, we need to make that
 799          * writeable first.
 800          */
 801         set_memory_rw(begin, (end - begin) >> PAGE_SHIFT);
 802
 803         for (addr = begin; addr < end; addr += PAGE_SIZE) {
 804                 ClearPageReserved(virt_to_page(addr));
 805                 init_page_count(virt_to_page(addr));
 806                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
 807                 free_page(addr);
 808                 totalram_pages++;
 809         }
 810         printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
 811 #endif
 812 }
 813
 814 void free_initmem(void)
 815 {
 816         free_init_pages("unused kernel memory",
 817                         (unsigned long)(&__init_begin),
 818                         (unsigned long)(&__init_end));
 819 }
 820
 821 #ifdef CONFIG_BLK_DEV_INITRD
 822 void free_initrd_mem(unsigned long start, unsigned long end)
 823 {
 824         free_init_pages("initrd memory", start, end);
 825 }
 826 #endif