arch/x86_64/kernel/setup.c

   1 /*
   2  *  linux/arch/x86-64/kernel/setup.c
   3  *
   4  *  Copyright (C) 1995  Linus Torvalds
   5  *
   6  *  Nov 2001 Dave Jones <davej@suse.de>
   7  *  Forked from i386 setup code.
   8  */
   9
  10 /*
  11  * This file handles the architecture-dependent parts of initialization
  12  */
  13
  14 #include <linux/errno.h>
  15 #include <linux/sched.h>
  16 #include <linux/kernel.h>
  17 #include <linux/mm.h>
  18 #include <linux/stddef.h>
  19 #include <linux/unistd.h>
  20 #include <linux/ptrace.h>
  21 #include <linux/slab.h>
  22 #include <linux/user.h>
  23 #include <linux/a.out.h>
  24 #include <linux/screen_info.h>
  25 #include <linux/ioport.h>
  26 #include <linux/delay.h>
  27 #include <linux/init.h>
  28 #include <linux/initrd.h>
  29 #include <linux/highmem.h>
  30 #include <linux/bootmem.h>
  31 #include <linux/module.h>
  32 #include <asm/processor.h>
  33 #include <linux/console.h>
  34 #include <linux/seq_file.h>
  35 #include <linux/crash_dump.h>
  36 #include <linux/root_dev.h>
  37 #include <linux/pci.h>
  38 #include <linux/acpi.h>
  39 #include <linux/kallsyms.h>
  40 #include <linux/edd.h>
  41 #include <linux/mmzone.h>
  42 #include <linux/kexec.h>
  43 #include <linux/cpufreq.h>
  44 #include <linux/dmi.h>
  45 #include <linux/dma-mapping.h>
  46 #include <linux/ctype.h>
  47
  48 #include <asm/mtrr.h>
  49 #include <asm/uaccess.h>
  50 #include <asm/system.h>
  51 #include <asm/io.h>
  52 #include <asm/smp.h>
  53 #include <asm/msr.h>
  54 #include <asm/desc.h>
  55 #include <video/edid.h>
  56 #include <asm/e820.h>
  57 #include <asm/dma.h>
  58 #include <asm/mpspec.h>
  59 #include <asm/mmu_context.h>
  60 #include <asm/bootsetup.h>
  61 #include <asm/proto.h>
  62 #include <asm/setup.h>
  63 #include <asm/mach_apic.h>
  64 #include <asm/numa.h>
  65 #include <asm/sections.h>
  66 #include <asm/dmi.h>
  67
  68 /*
  69  * Machine setup..
  70  */
  71
  72 struct cpuinfo_x86 boot_cpu_data __read_mostly;
  73 EXPORT_SYMBOL(boot_cpu_data);
  74
  75 unsigned long mmu_cr4_features;
  76
  77 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
  78 int bootloader_type;
  79
  80 unsigned long saved_video_mode;
  81
  82 /*
  83  * Early DMI memory
  84  */
  85 int dmi_alloc_index;
  86 char dmi_alloc_data[DMI_MAX_DATA];
  87
  88 /*
  89  * Setup options
  90  */
  91 struct screen_info screen_info;
  92 EXPORT_SYMBOL(screen_info);
  93 struct sys_desc_table_struct {
  94         unsigned short length;
  95         unsigned char table[0];
  96 };
  97
  98 struct edid_info edid_info;
  99 EXPORT_SYMBOL_GPL(edid_info);
 100
 101 extern int root_mountflags;
 102
 103 char command_line[COMMAND_LINE_SIZE];
 104
 105 struct resource standard_io_resources[] = {
 106         { .name = "dma1", .start = 0x00, .end = 0x1f,
 107                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 108         { .name = "pic1", .start = 0x20, .end = 0x21,
 109                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 110         { .name = "timer0", .start = 0x40, .end = 0x43,
 111                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 112         { .name = "timer1", .start = 0x50, .end = 0x53,
 113                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 114         { .name = "keyboard", .start = 0x60, .end = 0x6f,
 115                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 116         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 117                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 118         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 119                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 120         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 121                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 122         { .name = "fpu", .start = 0xf0, .end = 0xff,
 123                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 124 };
 125
 126 #define STANDARD_IO_RESOURCES \
 127         (sizeof standard_io_resources / sizeof standard_io_resources[0])
 128
 129 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
 130
 131 struct resource data_resource = {
 132         .name = "Kernel data",
 133         .start = 0,
 134         .end = 0,
 135         .flags = IORESOURCE_RAM,
 136 };
 137 struct resource code_resource = {
 138         .name = "Kernel code",
 139         .start = 0,
 140         .end = 0,
 141         .flags = IORESOURCE_RAM,
 142 };
 143
 144 #define IORESOURCE_ROM (IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM)
 145
 146 static struct resource system_rom_resource = {
 147         .name = "System ROM",
 148         .start = 0xf0000,
 149         .end = 0xfffff,
 150         .flags = IORESOURCE_ROM,
 151 };
 152
 153 static struct resource extension_rom_resource = {
 154         .name = "Extension ROM",
 155         .start = 0xe0000,
 156         .end = 0xeffff,
 157         .flags = IORESOURCE_ROM,
 158 };
 159
 160 static struct resource adapter_rom_resources[] = {
 161         { .name = "Adapter ROM", .start = 0xc8000, .end = 0,
 162                 .flags = IORESOURCE_ROM },
 163         { .name = "Adapter ROM", .start = 0, .end = 0,
 164                 .flags = IORESOURCE_ROM },
 165         { .name = "Adapter ROM", .start = 0, .end = 0,
 166                 .flags = IORESOURCE_ROM },
 167         { .name = "Adapter ROM", .start = 0, .end = 0,
 168                 .flags = IORESOURCE_ROM },
 169         { .name = "Adapter ROM", .start = 0, .end = 0,
 170                 .flags = IORESOURCE_ROM },
 171         { .name = "Adapter ROM", .start = 0, .end = 0,
 172                 .flags = IORESOURCE_ROM }
 173 };
 174
 175 #define ADAPTER_ROM_RESOURCES \
 176         (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
 177
 178 static struct resource video_rom_resource = {
 179         .name = "Video ROM",
 180         .start = 0xc0000,
 181         .end = 0xc7fff,
 182         .flags = IORESOURCE_ROM,
 183 };
 184
 185 static struct resource video_ram_resource = {
 186         .name = "Video RAM area",
 187         .start = 0xa0000,
 188         .end = 0xbffff,
 189         .flags = IORESOURCE_RAM,
 190 };
 191
 192 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
 193
 194 static int __init romchecksum(unsigned char *rom, unsigned long length)
 195 {
 196         unsigned char *p, sum = 0;
 197
 198         for (p = rom; p < rom + length; p++)
 199                 sum += *p;
 200         return sum == 0;
 201 }
 202
 203 static void __init probe_roms(void)
 204 {
 205         unsigned long start, length, upper;
 206         unsigned char *rom;
 207         int           i;
 208
 209         /* video rom */
 210         upper = adapter_rom_resources[0].start;
 211         for (start = video_rom_resource.start; start < upper; start += 2048) {
 212                 rom = isa_bus_to_virt(start);
 213                 if (!romsignature(rom))
 214                         continue;
 215
 216                 video_rom_resource.start = start;
 217
 218                 /* 0 < length <= 0x7f * 512, historically */
 219                 length = rom[2] * 512;
 220
 221                 /* if checksum okay, trust length byte */
 222                 if (length && romchecksum(rom, length))
 223                         video_rom_resource.end = start + length - 1;
 224
 225                 request_resource(&iomem_resource, &video_rom_resource);
 226                 break;
 227                         }
 228
 229         start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
 230         if (start < upper)
 231                 start = upper;
 232
 233         /* system rom */
 234         request_resource(&iomem_resource, &system_rom_resource);
 235         upper = system_rom_resource.start;
 236
 237         /* check for extension rom (ignore length byte!) */
 238         rom = isa_bus_to_virt(extension_rom_resource.start);
 239         if (romsignature(rom)) {
 240                 length = extension_rom_resource.end - extension_rom_resource.start + 1;
 241                 if (romchecksum(rom, length)) {
 242                         request_resource(&iomem_resource, &extension_rom_resource);
 243                         upper = extension_rom_resource.start;
 244                 }
 245         }
 246
 247         /* check for adapter roms on 2k boundaries */
 248         for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) {
 249                 rom = isa_bus_to_virt(start);
 250                 if (!romsignature(rom))
 251                         continue;
 252
 253                 /* 0 < length <= 0x7f * 512, historically */
 254                 length = rom[2] * 512;
 255
 256                 /* but accept any length that fits if checksum okay */
 257                 if (!length || start + length > upper || !romchecksum(rom, length))
 258                         continue;
 259
 260                 adapter_rom_resources[i].start = start;
 261                 adapter_rom_resources[i].end = start + length - 1;
 262                 request_resource(&iomem_resource, &adapter_rom_resources[i]);
 263
 264                 start = adapter_rom_resources[i++].end & ~2047UL;
 265         }
 266 }
 267
 268 #ifdef CONFIG_PROC_VMCORE
 269 /* elfcorehdr= specifies the location of elf core header
 270  * stored by the crashed kernel. This option will be passed
 271  * by kexec loader to the capture kernel.
 272  */
 273 static int __init setup_elfcorehdr(char *arg)
 274 {
 275         char *end;
 276         if (!arg)
 277                 return -EINVAL;
 278         elfcorehdr_addr = memparse(arg, &end);
 279         return end > arg ? 0 : -EINVAL;
 280 }
 281 early_param("elfcorehdr", setup_elfcorehdr);
 282 #endif
 283
 284 #ifndef CONFIG_NUMA
 285 static void __init
 286 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
 287 {
 288         unsigned long bootmap_size, bootmap;
 289
 290         bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
 291         bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size);
 292         if (bootmap == -1L)
 293                 panic("Cannot find bootmem map of size %ld\n",bootmap_size);
 294         bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
 295         e820_bootmem_free(NODE_DATA(0), 0, end_pfn << PAGE_SHIFT);
 296         reserve_bootmem(bootmap, bootmap_size);
 297 }
 298 #endif
 299
 300 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 301 struct edd edd;
 302 #ifdef CONFIG_EDD_MODULE
 303 EXPORT_SYMBOL(edd);
 304 #endif
 305 /**
 306  * copy_edd() - Copy the BIOS EDD information
 307  *              from boot_params into a safe place.
 308  *
 309  */
 310 static inline void copy_edd(void)
 311 {
 312      memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
 313      memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
 314      edd.mbr_signature_nr = EDD_MBR_SIG_NR;
 315      edd.edd_info_nr = EDD_NR;
 316 }
 317 #else
 318 static inline void copy_edd(void)
 319 {
 320 }
 321 #endif
 322
 323 #define EBDA_ADDR_POINTER 0x40E
 324
 325 unsigned __initdata ebda_addr;
 326 unsigned __initdata ebda_size;
 327
 328 static void discover_ebda(void)
 329 {
 330         /*
 331          * there is a real-mode segmented pointer pointing to the
 332          * 4K EBDA area at 0x40E
 333          */
 334         ebda_addr = *(unsigned short *)EBDA_ADDR_POINTER;
 335         ebda_addr <<= 4;
 336
 337         ebda_size = *(unsigned short *)(unsigned long)ebda_addr;
 338
 339         /* Round EBDA up to pages */
 340         if (ebda_size == 0)
 341                 ebda_size = 1;
 342         ebda_size <<= 10;
 343         ebda_size = round_up(ebda_size + (ebda_addr & ~PAGE_MASK), PAGE_SIZE);
 344         if (ebda_size > 64*1024)
 345                 ebda_size = 64*1024;
 346 }
 347
 348 void __init setup_arch(char **cmdline_p)
 349 {
 350         printk(KERN_INFO "Command line: %s\n", saved_command_line);
 351
 352         ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
 353         screen_info = SCREEN_INFO;
 354         edid_info = EDID_INFO;
 355         saved_video_mode = SAVED_VIDEO_MODE;
 356         bootloader_type = LOADER_TYPE;
 357
 358 #ifdef CONFIG_BLK_DEV_RAM
 359         rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
 360         rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
 361         rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
 362 #endif
 363         setup_memory_region();
 364         copy_edd();
 365
 366         if (!MOUNT_ROOT_RDONLY)
 367                 root_mountflags &= ~MS_RDONLY;
 368         init_mm.start_code = (unsigned long) &_text;
 369         init_mm.end_code = (unsigned long) &_etext;
 370         init_mm.end_data = (unsigned long) &_edata;
 371         init_mm.brk = (unsigned long) &_end;
 372
 373         code_resource.start = virt_to_phys(&_text);
 374         code_resource.end = virt_to_phys(&_etext)-1;
 375         data_resource.start = virt_to_phys(&_etext);
 376         data_resource.end = virt_to_phys(&_edata)-1;
 377
 378         early_identify_cpu(&boot_cpu_data);
 379
 380         strlcpy(command_line, saved_command_line, COMMAND_LINE_SIZE);
 381         *cmdline_p = command_line;
 382
 383         parse_early_param();
 384
 385         finish_e820_parsing();
 386
 387         /*
 388          * partially used pages are not usable - thus
 389          * we are rounding upwards:
 390          */
 391         end_pfn = e820_end_of_ram();
 392         num_physpages = end_pfn;
 393
 394         check_efer();
 395
 396         discover_ebda();
 397
 398         init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
 399
 400         dmi_scan_machine();
 401
 402         zap_low_mappings(0);
 403
 404 #ifdef CONFIG_ACPI
 405         /*
 406          * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
 407          * Call this early for SRAT node setup.
 408          */
 409         acpi_boot_table_init();
 410 #endif
 411
 412         /* How many end-of-memory variables you have, grandma! */
 413         max_low_pfn = end_pfn;
 414         max_pfn = end_pfn;
 415         high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
 416
 417 #ifdef CONFIG_ACPI_NUMA
 418         /*
 419          * Parse SRAT to discover nodes.
 420          */
 421         acpi_numa_init();
 422 #endif
 423
 424 #ifdef CONFIG_NUMA
 425         numa_initmem_init(0, end_pfn);
 426 #else
 427         contig_initmem_init(0, end_pfn);
 428 #endif
 429
 430         /* Reserve direct mapping */
 431         reserve_bootmem_generic(table_start << PAGE_SHIFT,
 432                                 (table_end - table_start) << PAGE_SHIFT);
 433
 434         /* reserve kernel */
 435         reserve_bootmem_generic(__pa_symbol(&_text),
 436                                 __pa_symbol(&_end) - __pa_symbol(&_text));
 437
 438         /*
 439          * reserve physical page 0 - it's a special BIOS page on many boxes,
 440          * enabling clean reboots, SMP operation, laptop functions.
 441          */
 442         reserve_bootmem_generic(0, PAGE_SIZE);
 443
 444         /* reserve ebda region */
 445         if (ebda_addr)
 446                 reserve_bootmem_generic(ebda_addr, ebda_size);
 447
 448 #ifdef CONFIG_SMP
 449         /*
 450          * But first pinch a few for the stack/trampoline stuff
 451          * FIXME: Don't need the extra page at 4K, but need to fix
 452          * trampoline before removing it. (see the GDT stuff)
 453          */
 454         reserve_bootmem_generic(PAGE_SIZE, PAGE_SIZE);
 455
 456         /* Reserve SMP trampoline */
 457         reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, PAGE_SIZE);
 458 #endif
 459
 460 #ifdef CONFIG_ACPI_SLEEP
 461        /*
 462         * Reserve low memory region for sleep support.
 463         */
 464        acpi_reserve_bootmem();
 465 #endif
 466         /*
 467          * Find and reserve possible boot-time SMP configuration:
 468          */
 469         find_smp_config();
 470 #ifdef CONFIG_BLK_DEV_INITRD
 471         if (LOADER_TYPE && INITRD_START) {
 472                 if (INITRD_START + INITRD_SIZE <= (end_pfn << PAGE_SHIFT)) {
 473                         reserve_bootmem_generic(INITRD_START, INITRD_SIZE);
 474                         initrd_start =
 475                                 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
 476                         initrd_end = initrd_start+INITRD_SIZE;
 477                 }
 478                 else {
 479                         printk(KERN_ERR "initrd extends beyond end of memory "
 480                             "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
 481                             (unsigned long)(INITRD_START + INITRD_SIZE),
 482                             (unsigned long)(end_pfn << PAGE_SHIFT));
 483                         initrd_start = 0;
 484                 }
 485         }
 486 #endif
 487 #ifdef CONFIG_KEXEC
 488         if (crashk_res.start != crashk_res.end) {
 489                 reserve_bootmem_generic(crashk_res.start,
 490                         crashk_res.end - crashk_res.start + 1);
 491         }
 492 #endif
 493
 494         paging_init();
 495
 496         early_quirks();
 497
 498         /*
 499          * set this early, so we dont allocate cpu0
 500          * if MADT list doesnt list BSP first
 501          * mpparse.c/MP_processor_info() allocates logical cpu numbers.
 502          */
 503         cpu_set(0, cpu_present_map);
 504 #ifdef CONFIG_ACPI
 505         /*
 506          * Read APIC and some other early information from ACPI tables.
 507          */
 508         acpi_boot_init();
 509 #endif
 510
 511         init_cpu_to_node();
 512
 513         /*
 514          * get boot-time SMP configuration:
 515          */
 516         if (smp_found_config)
 517                 get_smp_config();
 518         init_apic_mappings();
 519
 520         /*
 521          * Request address space for all standard RAM and ROM resources
 522          * and also for regions reported as reserved by the e820.
 523          */
 524         probe_roms();
 525         e820_reserve_resources();
 526
 527         request_resource(&iomem_resource, &video_ram_resource);
 528
 529         {
 530         unsigned i;
 531         /* request I/O space for devices used on all i[345]86 PCs */
 532         for (i = 0; i < STANDARD_IO_RESOURCES; i++)
 533                 request_resource(&ioport_resource, &standard_io_resources[i]);
 534         }
 535
 536         e820_setup_gap();
 537
 538 #ifdef CONFIG_VT
 539 #if defined(CONFIG_VGA_CONSOLE)
 540         conswitchp = &vga_con;
 541 #elif defined(CONFIG_DUMMY_CONSOLE)
 542         conswitchp = &dummy_con;
 543 #endif
 544 #endif
 545 }
 546
 547 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
 548 {
 549         unsigned int *v;
 550
 551         if (c->extended_cpuid_level < 0x80000004)
 552                 return 0;
 553
 554         v = (unsigned int *) c->x86_model_id;
 555         cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
 556         cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
 557         cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
 558         c->x86_model_id[48] = 0;
 559         return 1;
 560 }
 561
 562
 563 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
 564 {
 565         unsigned int n, dummy, eax, ebx, ecx, edx;
 566
 567         n = c->extended_cpuid_level;
 568
 569         if (n >= 0x80000005) {
 570                 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
 571                 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
 572                         edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
 573                 c->x86_cache_size=(ecx>>24)+(edx>>24);
 574                 /* On K8 L1 TLB is inclusive, so don't count it */
 575                 c->x86_tlbsize = 0;
 576         }
 577
 578         if (n >= 0x80000006) {
 579                 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
 580                 ecx = cpuid_ecx(0x80000006);
 581                 c->x86_cache_size = ecx >> 16;
 582                 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
 583
 584                 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
 585                 c->x86_cache_size, ecx & 0xFF);
 586         }
 587
 588         if (n >= 0x80000007)
 589                 cpuid(0x80000007, &dummy, &dummy, &dummy, &c->x86_power);
 590         if (n >= 0x80000008) {
 591                 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
 592                 c->x86_virt_bits = (eax >> 8) & 0xff;
 593                 c->x86_phys_bits = eax & 0xff;
 594         }
 595 }
 596
 597 #ifdef CONFIG_NUMA
 598 static int nearby_node(int apicid)
 599 {
 600         int i;
 601         for (i = apicid - 1; i >= 0; i--) {
 602                 int node = apicid_to_node[i];
 603                 if (node != NUMA_NO_NODE && node_online(node))
 604                         return node;
 605         }
 606         for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
 607                 int node = apicid_to_node[i];
 608                 if (node != NUMA_NO_NODE && node_online(node))
 609                         return node;
 610         }
 611         return first_node(node_online_map); /* Shouldn't happen */
 612 }
 613 #endif
 614
 615 /*
 616  * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
 617  * Assumes number of cores is a power of two.
 618  */
 619 static void __init amd_detect_cmp(struct cpuinfo_x86 *c)
 620 {
 621 #ifdef CONFIG_SMP
 622         unsigned bits;
 623 #ifdef CONFIG_NUMA
 624         int cpu = smp_processor_id();
 625         int node = 0;
 626         unsigned apicid = hard_smp_processor_id();
 627 #endif
 628         unsigned ecx = cpuid_ecx(0x80000008);
 629
 630         c->x86_max_cores = (ecx & 0xff) + 1;
 631
 632         /* CPU telling us the core id bits shift? */
 633         bits = (ecx >> 12) & 0xF;
 634
 635         /* Otherwise recompute */
 636         if (bits == 0) {
 637                 while ((1 << bits) < c->x86_max_cores)
 638                         bits++;
 639         }
 640
 641         /* Low order bits define the core id (index of core in socket) */
 642         c->cpu_core_id = c->phys_proc_id & ((1 << bits)-1);
 643         /* Convert the APIC ID into the socket ID */
 644         c->phys_proc_id = phys_pkg_id(bits);
 645
 646 #ifdef CONFIG_NUMA
 647         node = c->phys_proc_id;
 648         if (apicid_to_node[apicid] != NUMA_NO_NODE)
 649                 node = apicid_to_node[apicid];
 650         if (!node_online(node)) {
 651                 /* Two possibilities here:
 652                    - The CPU is missing memory and no node was created.
 653                    In that case try picking one from a nearby CPU
 654                    - The APIC IDs differ from the HyperTransport node IDs
 655                    which the K8 northbridge parsing fills in.
 656                    Assume they are all increased by a constant offset,
 657                    but in the same order as the HT nodeids.
 658                    If that doesn't result in a usable node fall back to the
 659                    path for the previous case.  */
 660                 int ht_nodeid = apicid - (cpu_data[0].phys_proc_id << bits);
 661                 if (ht_nodeid >= 0 &&
 662                     apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
 663                         node = apicid_to_node[ht_nodeid];
 664                 /* Pick a nearby node */
 665                 if (!node_online(node))
 666                         node = nearby_node(apicid);
 667         }
 668         numa_set_node(cpu, node);
 669
 670         printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
 671 #endif
 672 #endif
 673 }
 674
 675 static void __init init_amd(struct cpuinfo_x86 *c)
 676 {
 677         unsigned level;
 678
 679 #ifdef CONFIG_SMP
 680         unsigned long value;
 681
 682         /*
 683          * Disable TLB flush filter by setting HWCR.FFDIS on K8
 684          * bit 6 of msr C001_0015
 685          *
 686          * Errata 63 for SH-B3 steppings
 687          * Errata 122 for all steppings (F+ have it disabled by default)
 688          */
 689         if (c->x86 == 15) {
 690                 rdmsrl(MSR_K8_HWCR, value);
 691                 value |= 1 << 6;
 692                 wrmsrl(MSR_K8_HWCR, value);
 693         }
 694 #endif
 695
 696         /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
 697            3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
 698         clear_bit(0*32+31, &c->x86_capability);
 699
 700         /* On C+ stepping K8 rep microcode works well for copy/memset */
 701         level = cpuid_eax(1);
 702         if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58))
 703                 set_bit(X86_FEATURE_REP_GOOD, &c->x86_capability);
 704
 705         /* Enable workaround for FXSAVE leak */
 706         if (c->x86 >= 6)
 707                 set_bit(X86_FEATURE_FXSAVE_LEAK, &c->x86_capability);
 708
 709         level = get_model_name(c);
 710         if (!level) {
 711                 switch (c->x86) {
 712                 case 15:
 713                         /* Should distinguish Models here, but this is only
 714                            a fallback anyways. */
 715                         strcpy(c->x86_model_id, "Hammer");
 716                         break;
 717                 }
 718         }
 719         display_cacheinfo(c);
 720
 721         /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
 722         if (c->x86_power & (1<<8))
 723                 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
 724
 725         /* Multi core CPU? */
 726         if (c->extended_cpuid_level >= 0x80000008)
 727                 amd_detect_cmp(c);
 728
 729         /* Fix cpuid4 emulation for more */
 730         num_cache_leaves = 3;
 731 }
 732
 733 static void __cpuinit detect_ht(struct cpuinfo_x86 *c)
 734 {
 735 #ifdef CONFIG_SMP
 736         u32     eax, ebx, ecx, edx;
 737         int     index_msb, core_bits;
 738
 739         cpuid(1, &eax, &ebx, &ecx, &edx);
 740
 741
 742         if (!cpu_has(c, X86_FEATURE_HT))
 743                 return;
 744         if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
 745                 goto out;
 746
 747         smp_num_siblings = (ebx & 0xff0000) >> 16;
 748
 749         if (smp_num_siblings == 1) {
 750                 printk(KERN_INFO  "CPU: Hyper-Threading is disabled\n");
 751         } else if (smp_num_siblings > 1 ) {
 752
 753                 if (smp_num_siblings > NR_CPUS) {
 754                         printk(KERN_WARNING "CPU: Unsupported number of the siblings %d", smp_num_siblings);
 755                         smp_num_siblings = 1;
 756                         return;
 757                 }
 758
 759                 index_msb = get_count_order(smp_num_siblings);
 760                 c->phys_proc_id = phys_pkg_id(index_msb);
 761
 762                 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
 763
 764                 index_msb = get_count_order(smp_num_siblings) ;
 765
 766                 core_bits = get_count_order(c->x86_max_cores);
 767
 768                 c->cpu_core_id = phys_pkg_id(index_msb) &
 769                                                ((1 << core_bits) - 1);
 770         }
 771 out:
 772         if ((c->x86_max_cores * smp_num_siblings) > 1) {
 773                 printk(KERN_INFO  "CPU: Physical Processor ID: %d\n", c->phys_proc_id);
 774                 printk(KERN_INFO  "CPU: Processor Core ID: %d\n", c->cpu_core_id);
 775         }
 776
 777 #endif
 778 }
 779
 780 /*
 781  * find out the number of processor cores on the die
 782  */
 783 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
 784 {
 785         unsigned int eax, t;
 786
 787         if (c->cpuid_level < 4)
 788                 return 1;
 789
 790         cpuid_count(4, 0, &eax, &t, &t, &t);
 791
 792         if (eax & 0x1f)
 793                 return ((eax >> 26) + 1);
 794         else
 795                 return 1;
 796 }
 797
 798 static void srat_detect_node(void)
 799 {
 800 #ifdef CONFIG_NUMA
 801         unsigned node;
 802         int cpu = smp_processor_id();
 803         int apicid = hard_smp_processor_id();
 804
 805         /* Don't do the funky fallback heuristics the AMD version employs
 806            for now. */
 807         node = apicid_to_node[apicid];
 808         if (node == NUMA_NO_NODE)
 809                 node = first_node(node_online_map);
 810         numa_set_node(cpu, node);
 811
 812         printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
 813 #endif
 814 }
 815
 816 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
 817 {
 818         /* Cache sizes */
 819         unsigned n;
 820
 821         init_intel_cacheinfo(c);
 822         if (c->cpuid_level > 9 ) {
 823                 unsigned eax = cpuid_eax(10);
 824                 /* Check for version and the number of counters */
 825                 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
 826                         set_bit(X86_FEATURE_ARCH_PERFMON, &c->x86_capability);
 827         }
 828
 829         n = c->extended_cpuid_level;
 830         if (n >= 0x80000008) {
 831                 unsigned eax = cpuid_eax(0x80000008);
 832                 c->x86_virt_bits = (eax >> 8) & 0xff;
 833                 c->x86_phys_bits = eax & 0xff;
 834                 /* CPUID workaround for Intel 0F34 CPU */
 835                 if (c->x86_vendor == X86_VENDOR_INTEL &&
 836                     c->x86 == 0xF && c->x86_model == 0x3 &&
 837                     c->x86_mask == 0x4)
 838                         c->x86_phys_bits = 36;
 839         }
 840
 841         if (c->x86 == 15)
 842                 c->x86_cache_alignment = c->x86_clflush_size * 2;
 843         if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
 844             (c->x86 == 0x6 && c->x86_model >= 0x0e))
 845                 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
 846         set_bit(X86_FEATURE_SYNC_RDTSC, &c->x86_capability);
 847         c->x86_max_cores = intel_num_cpu_cores(c);
 848
 849         srat_detect_node();
 850 }
 851
 852 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
 853 {
 854         char *v = c->x86_vendor_id;
 855
 856         if (!strcmp(v, "AuthenticAMD"))
 857                 c->x86_vendor = X86_VENDOR_AMD;
 858         else if (!strcmp(v, "GenuineIntel"))
 859                 c->x86_vendor = X86_VENDOR_INTEL;
 860         else
 861                 c->x86_vendor = X86_VENDOR_UNKNOWN;
 862 }
 863
 864 struct cpu_model_info {
 865         int vendor;
 866         int family;
 867         char *model_names[16];
 868 };
 869
 870 /* Do some early cpuid on the boot CPU to get some parameter that are
 871    needed before check_bugs. Everything advanced is in identify_cpu
 872    below. */
 873 void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
 874 {
 875         u32 tfms;
 876
 877         c->loops_per_jiffy = loops_per_jiffy;
 878         c->x86_cache_size = -1;
 879         c->x86_vendor = X86_VENDOR_UNKNOWN;
 880         c->x86_model = c->x86_mask = 0; /* So far unknown... */
 881         c->x86_vendor_id[0] = '\0'; /* Unset */
 882         c->x86_model_id[0] = '\0';  /* Unset */
 883         c->x86_clflush_size = 64;
 884         c->x86_cache_alignment = c->x86_clflush_size;
 885         c->x86_max_cores = 1;
 886         c->extended_cpuid_level = 0;
 887         memset(&c->x86_capability, 0, sizeof c->x86_capability);
 888
 889         /* Get vendor name */
 890         cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
 891               (unsigned int *)&c->x86_vendor_id[0],
 892               (unsigned int *)&c->x86_vendor_id[8],
 893               (unsigned int *)&c->x86_vendor_id[4]);
 894
 895         get_cpu_vendor(c);
 896
 897         /* Initialize the standard set of capabilities */
 898         /* Note that the vendor-specific code below might override */
 899
 900         /* Intel-defined flags: level 0x00000001 */
 901         if (c->cpuid_level >= 0x00000001) {
 902                 __u32 misc;
 903                 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
 904                       &c->x86_capability[0]);
 905                 c->x86 = (tfms >> 8) & 0xf;
 906                 c->x86_model = (tfms >> 4) & 0xf;
 907                 c->x86_mask = tfms & 0xf;
 908                 if (c->x86 == 0xf)
 909                         c->x86 += (tfms >> 20) & 0xff;
 910                 if (c->x86 >= 0x6)
 911                         c->x86_model += ((tfms >> 16) & 0xF) << 4;
 912                 if (c->x86_capability[0] & (1<<19))
 913                         c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
 914         } else {
 915                 /* Have CPUID level 0 only - unheard of */
 916                 c->x86 = 4;
 917         }
 918
 919 #ifdef CONFIG_SMP
 920         c->phys_proc_id = (cpuid_ebx(1) >> 24) & 0xff;
 921 #endif
 922 }
 923
 924 /*
 925  * This does the hard work of actually picking apart the CPU stuff...
 926  */
 927 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
 928 {
 929         int i;
 930         u32 xlvl;
 931
 932         early_identify_cpu(c);
 933
 934         /* AMD-defined flags: level 0x80000001 */
 935         xlvl = cpuid_eax(0x80000000);
 936         c->extended_cpuid_level = xlvl;
 937         if ((xlvl & 0xffff0000) == 0x80000000) {
 938                 if (xlvl >= 0x80000001) {
 939                         c->x86_capability[1] = cpuid_edx(0x80000001);
 940                         c->x86_capability[6] = cpuid_ecx(0x80000001);
 941                 }
 942                 if (xlvl >= 0x80000004)
 943                         get_model_name(c); /* Default name */
 944         }
 945
 946         /* Transmeta-defined flags: level 0x80860001 */
 947         xlvl = cpuid_eax(0x80860000);
 948         if ((xlvl & 0xffff0000) == 0x80860000) {
 949                 /* Don't set x86_cpuid_level here for now to not confuse. */
 950                 if (xlvl >= 0x80860001)
 951                         c->x86_capability[2] = cpuid_edx(0x80860001);
 952         }
 953
 954         c->apicid = phys_pkg_id(0);
 955
 956         /*
 957          * Vendor-specific initialization.  In this section we
 958          * canonicalize the feature flags, meaning if there are
 959          * features a certain CPU supports which CPUID doesn't
 960          * tell us, CPUID claiming incorrect flags, or other bugs,
 961          * we handle them here.
 962          *
 963          * At the end of this section, c->x86_capability better
 964          * indicate the features this CPU genuinely supports!
 965          */
 966         switch (c->x86_vendor) {
 967         case X86_VENDOR_AMD:
 968                 init_amd(c);
 969                 break;
 970
 971         case X86_VENDOR_INTEL:
 972                 init_intel(c);
 973                 break;
 974
 975         case X86_VENDOR_UNKNOWN:
 976         default:
 977                 display_cacheinfo(c);
 978                 break;
 979         }
 980
 981         select_idle_routine(c);
 982         detect_ht(c);
 983
 984         /*
 985          * On SMP, boot_cpu_data holds the common feature set between
 986          * all CPUs; so make sure that we indicate which features are
 987          * common between the CPUs.  The first time this routine gets
 988          * executed, c == &boot_cpu_data.
 989          */
 990         if (c != &boot_cpu_data) {
 991                 /* AND the already accumulated flags with these */
 992                 for (i = 0 ; i < NCAPINTS ; i++)
 993                         boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
 994         }
 995
 996 #ifdef CONFIG_X86_MCE
 997         mcheck_init(c);
 998 #endif
 999         if (c == &boot_cpu_data)
1000                 mtrr_bp_init();
1001         else
1002                 mtrr_ap_init();
1003 #ifdef CONFIG_NUMA
1004         numa_add_cpu(smp_processor_id());
1005 #endif
1006 }
1007
1008
1009 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1010 {
1011         if (c->x86_model_id[0])
1012                 printk("%s", c->x86_model_id);
1013
1014         if (c->x86_mask || c->cpuid_level >= 0)
1015                 printk(" stepping %02x\n", c->x86_mask);
1016         else
1017                 printk("\n");
1018 }
1019
1020 /*
1021  *      Get CPU information for use by the procfs.
1022  */
1023
1024 static int show_cpuinfo(struct seq_file *m, void *v)
1025 {
1026         struct cpuinfo_x86 *c = v;
1027
1028         /*
1029          * These flag bits must match the definitions in <asm/cpufeature.h>.
1030          * NULL means this bit is undefined or reserved; either way it doesn't
1031          * have meaning as far as Linux is concerned.  Note that it's important
1032          * to realize there is a difference between this table and CPUID -- if
1033          * applications want to get the raw CPUID data, they should access
1034          * /dev/cpu/<cpu_nr>/cpuid instead.
1035          */
1036         static char *x86_cap_flags[] = {
1037                 /* Intel-defined */
1038                 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1039                 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
1040                 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
1041                 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", NULL,
1042
1043                 /* AMD-defined */
1044                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1045                 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
1046                 NULL, NULL, NULL, NULL, "nx", NULL, "mmxext", NULL,
1047                 NULL, "fxsr_opt", NULL, "rdtscp", NULL, "lm", "3dnowext", "3dnow",
1048
1049                 /* Transmeta-defined */
1050                 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
1051                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1052                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1053                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1054
1055                 /* Other (Linux-defined) */
1056                 "cxmmx", NULL, "cyrix_arr", "centaur_mcr", NULL,
1057                 "constant_tsc", NULL, NULL,
1058                 "up", NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1059                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1060                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1061
1062                 /* Intel-defined (#2) */
1063                 "pni", NULL, NULL, "monitor", "ds_cpl", "vmx", "smx", "est",
1064                 "tm2", NULL, "cid", NULL, NULL, "cx16", "xtpr", NULL,
1065                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1066                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1067
1068                 /* VIA/Cyrix/Centaur-defined */
1069                 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
1070                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1071                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1072                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1073
1074                 /* AMD-defined (#2) */
1075                 "lahf_lm", "cmp_legacy", "svm", NULL, "cr8_legacy", NULL, NULL, NULL,
1076                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1077                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1078                 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1079         };
1080         static char *x86_power_flags[] = {
1081                 "ts",   /* temperature sensor */
1082                 "fid",  /* frequency id control */
1083                 "vid",  /* voltage id control */
1084                 "ttp",  /* thermal trip */
1085                 "tm",
1086                 "stc",
1087                 NULL,
1088                 /* nothing */   /* constant_tsc - moved to flags */
1089         };
1090
1091
1092 #ifdef CONFIG_SMP
1093         if (!cpu_online(c-cpu_data))
1094                 return 0;
1095 #endif
1096
1097         seq_printf(m,"processor\t: %u\n"
1098                      "vendor_id\t: %s\n"
1099                      "cpu family\t: %d\n"
1100                      "model\t\t: %d\n"
1101                      "model name\t: %s\n",
1102                      (unsigned)(c-cpu_data),
1103                      c->x86_vendor_id[0] ? c->x86_vendor_id : "unknown",
1104                      c->x86,
1105                      (int)c->x86_model,
1106                      c->x86_model_id[0] ? c->x86_model_id : "unknown");
1107
1108         if (c->x86_mask || c->cpuid_level >= 0)
1109                 seq_printf(m, "stepping\t: %d\n", c->x86_mask);
1110         else
1111                 seq_printf(m, "stepping\t: unknown\n");
1112
1113         if (cpu_has(c,X86_FEATURE_TSC)) {
1114                 unsigned int freq = cpufreq_quick_get((unsigned)(c-cpu_data));
1115                 if (!freq)
1116                         freq = cpu_khz;
1117                 seq_printf(m, "cpu MHz\t\t: %u.%03u\n",
1118                              freq / 1000, (freq % 1000));
1119         }
1120
1121         /* Cache size */
1122         if (c->x86_cache_size >= 0)
1123                 seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
1124
1125 #ifdef CONFIG_SMP
1126         if (smp_num_siblings * c->x86_max_cores > 1) {
1127                 int cpu = c - cpu_data;
1128                 seq_printf(m, "physical id\t: %d\n", c->phys_proc_id);
1129                 seq_printf(m, "siblings\t: %d\n", cpus_weight(cpu_core_map[cpu]));
1130                 seq_printf(m, "core id\t\t: %d\n", c->cpu_core_id);
1131                 seq_printf(m, "cpu cores\t: %d\n", c->booted_cores);
1132         }
1133 #endif
1134
1135         seq_printf(m,
1136                 "fpu\t\t: yes\n"
1137                 "fpu_exception\t: yes\n"
1138                 "cpuid level\t: %d\n"
1139                 "wp\t\t: yes\n"
1140                 "flags\t\t:",
1141                    c->cpuid_level);
1142
1143         {
1144                 int i;
1145                 for ( i = 0 ; i < 32*NCAPINTS ; i++ )
1146                         if (cpu_has(c, i) && x86_cap_flags[i] != NULL)
1147                                 seq_printf(m, " %s", x86_cap_flags[i]);
1148         }
1149
1150         seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
1151                    c->loops_per_jiffy/(500000/HZ),
1152                    (c->loops_per_jiffy/(5000/HZ)) % 100);
1153
1154         if (c->x86_tlbsize > 0)
1155                 seq_printf(m, "TLB size\t: %d 4K pages\n", c->x86_tlbsize);
1156         seq_printf(m, "clflush size\t: %d\n", c->x86_clflush_size);
1157         seq_printf(m, "cache_alignment\t: %d\n", c->x86_cache_alignment);
1158
1159         seq_printf(m, "address sizes\t: %u bits physical, %u bits virtual\n",
1160                    c->x86_phys_bits, c->x86_virt_bits);
1161
1162         seq_printf(m, "power management:");
1163         {
1164                 unsigned i;
1165                 for (i = 0; i < 32; i++)
1166                         if (c->x86_power & (1 << i)) {
1167                                 if (i < ARRAY_SIZE(x86_power_flags) &&
1168                                         x86_power_flags[i])
1169                                         seq_printf(m, "%s%s",
1170                                                 x86_power_flags[i][0]?" ":"",
1171                                                 x86_power_flags[i]);
1172                                 else
1173                                         seq_printf(m, " [%d]", i);
1174                         }
1175         }
1176
1177         seq_printf(m, "\n\n");
1178
1179         return 0;
1180 }
1181
1182 static void *c_start(struct seq_file *m, loff_t *pos)
1183 {
1184         return *pos < NR_CPUS ? cpu_data + *pos : NULL;
1185 }
1186
1187 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
1188 {
1189         ++*pos;
1190         return c_start(m, pos);
1191 }
1192
1193 static void c_stop(struct seq_file *m, void *v)
1194 {
1195 }
1196
1197 struct seq_operations cpuinfo_op = {
1198         .start =c_start,
1199         .next = c_next,
1200         .stop = c_stop,
1201         .show = show_cpuinfo,
1202 };
1203
1204 #if defined(CONFIG_INPUT_PCSPKR) || defined(CONFIG_INPUT_PCSPKR_MODULE)
1205 #include <linux/platform_device.h>
1206 static __init int add_pcspkr(void)
1207 {
1208         struct platform_device *pd;
1209         int ret;
1210
1211         pd = platform_device_alloc("pcspkr", -1);
1212         if (!pd)
1213                 return -ENOMEM;
1214
1215         ret = platform_device_add(pd);
1216         if (ret)
1217                 platform_device_put(pd);
1218
1219         return ret;
1220 }
1221 device_initcall(add_pcspkr);
1222 #endif