arch/x86/kernel/setup_64.c

   1 /*
   2  *  Copyright (C) 1995  Linus Torvalds
   3  */
   4
   5 /*
   6  * This file handles the architecture-dependent parts of initialization
   7  */
   8
   9 #include <linux/errno.h>
  10 #include <linux/sched.h>
  11 #include <linux/kernel.h>
  12 #include <linux/mm.h>
  13 #include <linux/stddef.h>
  14 #include <linux/unistd.h>
  15 #include <linux/ptrace.h>
  16 #include <linux/slab.h>
  17 #include <linux/user.h>
  18 #include <linux/screen_info.h>
  19 #include <linux/ioport.h>
  20 #include <linux/delay.h>
  21 #include <linux/init.h>
  22 #include <linux/initrd.h>
  23 #include <linux/highmem.h>
  24 #include <linux/bootmem.h>
  25 #include <linux/module.h>
  26 #include <asm/processor.h>
  27 #include <linux/console.h>
  28 #include <linux/seq_file.h>
  29 #include <linux/crash_dump.h>
  30 #include <linux/root_dev.h>
  31 #include <linux/pci.h>
  32 #include <asm/pci-direct.h>
  33 #include <linux/efi.h>
  34 #include <linux/acpi.h>
  35 #include <linux/kallsyms.h>
  36 #include <linux/edd.h>
  37 #include <linux/iscsi_ibft.h>
  38 #include <linux/mmzone.h>
  39 #include <linux/kexec.h>
  40 #include <linux/cpufreq.h>
  41 #include <linux/dmi.h>
  42 #include <linux/dma-mapping.h>
  43 #include <linux/ctype.h>
  44 #include <linux/sort.h>
  45 #include <linux/uaccess.h>
  46 #include <linux/init_ohci1394_dma.h>
  47 #include <linux/kvm_para.h>
  48
  49 #include <asm/mtrr.h>
  50 #include <asm/uaccess.h>
  51 #include <asm/system.h>
  52 #include <asm/vsyscall.h>
  53 #include <asm/io.h>
  54 #include <asm/smp.h>
  55 #include <asm/msr.h>
  56 #include <asm/desc.h>
  57 #include <video/edid.h>
  58 #include <asm/e820.h>
  59 #include <asm/dma.h>
  60 #include <asm/gart.h>
  61 #include <asm/mpspec.h>
  62 #include <asm/mmu_context.h>
  63 #include <asm/proto.h>
  64 #include <asm/setup.h>
  65 #include <asm/numa.h>
  66 #include <asm/sections.h>
  67 #include <asm/dmi.h>
  68 #include <asm/cacheflush.h>
  69 #include <asm/mce.h>
  70 #include <asm/ds.h>
  71 #include <asm/topology.h>
  72 #include <asm/trampoline.h>
  73 #include <asm/pat.h>
  74
  75 #include <mach_apic.h>
  76 #ifdef CONFIG_PARAVIRT
  77 #include <asm/paravirt.h>
  78 #else
  79 #define ARCH_SETUP
  80 #endif
  81
  82 /*
  83  * Machine setup..
  84  */
  85
  86 struct cpuinfo_x86 boot_cpu_data __read_mostly;
  87 EXPORT_SYMBOL(boot_cpu_data);
  88
  89 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
  90
  91 unsigned long mmu_cr4_features;
  92
  93 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
  94 int bootloader_type;
  95
  96 unsigned long saved_video_mode;
  97
  98 int force_mwait __cpuinitdata;
  99
 100 /*
 101  * Early DMI memory
 102  */
 103 int dmi_alloc_index;
 104 char dmi_alloc_data[DMI_MAX_DATA];
 105
 106 /*
 107  * Setup options
 108  */
 109 struct screen_info screen_info;
 110 EXPORT_SYMBOL(screen_info);
 111 struct sys_desc_table_struct {
 112         unsigned short length;
 113         unsigned char table[0];
 114 };
 115
 116 struct edid_info edid_info;
 117 EXPORT_SYMBOL_GPL(edid_info);
 118
 119 extern int root_mountflags;
 120
 121 char __initdata command_line[COMMAND_LINE_SIZE];
 122
 123 static struct resource standard_io_resources[] = {
 124         { .name = "dma1", .start = 0x00, .end = 0x1f,
 125                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 126         { .name = "pic1", .start = 0x20, .end = 0x21,
 127                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 128         { .name = "timer0", .start = 0x40, .end = 0x43,
 129                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 130         { .name = "timer1", .start = 0x50, .end = 0x53,
 131                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 132         { .name = "keyboard", .start = 0x60, .end = 0x60,
 133                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 134         { .name = "keyboard", .start = 0x64, .end = 0x64,
 135                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 136         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 137                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 138         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 139                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 140         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 141                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 142         { .name = "fpu", .start = 0xf0, .end = 0xff,
 143                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 144 };
 145
 146 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
 147
 148 static struct resource data_resource = {
 149         .name = "Kernel data",
 150         .start = 0,
 151         .end = 0,
 152         .flags = IORESOURCE_RAM,
 153 };
 154 static struct resource code_resource = {
 155         .name = "Kernel code",
 156         .start = 0,
 157         .end = 0,
 158         .flags = IORESOURCE_RAM,
 159 };
 160 static struct resource bss_resource = {
 161         .name = "Kernel bss",
 162         .start = 0,
 163         .end = 0,
 164         .flags = IORESOURCE_RAM,
 165 };
 166
 167 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
 168
 169 #ifdef CONFIG_PROC_VMCORE
 170 /* elfcorehdr= specifies the location of elf core header
 171  * stored by the crashed kernel. This option will be passed
 172  * by kexec loader to the capture kernel.
 173  */
 174 static int __init setup_elfcorehdr(char *arg)
 175 {
 176         char *end;
 177         if (!arg)
 178                 return -EINVAL;
 179         elfcorehdr_addr = memparse(arg, &end);
 180         return end > arg ? 0 : -EINVAL;
 181 }
 182 early_param("elfcorehdr", setup_elfcorehdr);
 183 #endif
 184
 185 #ifndef CONFIG_NUMA
 186 static void __init
 187 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
 188 {
 189         unsigned long bootmap_size, bootmap;
 190
 191         bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
 192         bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
 193                                  PAGE_SIZE);
 194         if (bootmap == -1L)
 195                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 196         bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
 197         e820_register_active_regions(0, start_pfn, end_pfn);
 198         free_bootmem_with_active_regions(0, end_pfn);
 199         early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
 200         reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
 201 }
 202 #endif
 203
 204 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 205 struct edd edd;
 206 #ifdef CONFIG_EDD_MODULE
 207 EXPORT_SYMBOL(edd);
 208 #endif
 209 /**
 210  * copy_edd() - Copy the BIOS EDD information
 211  *              from boot_params into a safe place.
 212  *
 213  */
 214 static inline void copy_edd(void)
 215 {
 216      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 217             sizeof(edd.mbr_signature));
 218      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 219      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 220      edd.edd_info_nr = boot_params.eddbuf_entries;
 221 }
 222 #else
 223 static inline void copy_edd(void)
 224 {
 225 }
 226 #endif
 227
 228 #ifdef CONFIG_KEXEC
 229 static void __init reserve_crashkernel(void)
 230 {
 231         unsigned long long total_mem;
 232         unsigned long long crash_size, crash_base;
 233         int ret;
 234
 235         total_mem = ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
 236
 237         ret = parse_crashkernel(boot_command_line, total_mem,
 238                         &crash_size, &crash_base);
 239         if (ret == 0 && crash_size) {
 240                 if (crash_base <= 0) {
 241                         printk(KERN_INFO "crashkernel reservation failed - "
 242                                         "you have to specify a base address\n");
 243                         return;
 244                 }
 245
 246                 if (reserve_bootmem(crash_base, crash_size,
 247                                         BOOTMEM_EXCLUSIVE) < 0) {
 248                         printk(KERN_INFO "crashkernel reservation failed - "
 249                                         "memory is in use\n");
 250                         return;
 251                 }
 252
 253                 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 254                                 "for crashkernel (System RAM: %ldMB)\n",
 255                                 (unsigned long)(crash_size >> 20),
 256                                 (unsigned long)(crash_base >> 20),
 257                                 (unsigned long)(total_mem >> 20));
 258                 crashk_res.start = crash_base;
 259                 crashk_res.end   = crash_base + crash_size - 1;
 260                 insert_resource(&iomem_resource, &crashk_res);
 261         }
 262 }
 263 #else
 264 static inline void __init reserve_crashkernel(void)
 265 {}
 266 #endif
 267
 268 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
 269 void __attribute__((weak)) __init memory_setup(void)
 270 {
 271        machine_specific_memory_setup();
 272 }
 273
 274 static void __init parse_setup_data(void)
 275 {
 276         struct setup_data *data;
 277         unsigned long pa_data;
 278
 279         if (boot_params.hdr.version < 0x0209)
 280                 return;
 281         pa_data = boot_params.hdr.setup_data;
 282         while (pa_data) {
 283                 data = early_ioremap(pa_data, PAGE_SIZE);
 284                 switch (data->type) {
 285                 default:
 286                         break;
 287                 }
 288 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 289                 free_early(pa_data, pa_data+sizeof(*data)+data->len);
 290 #endif
 291                 pa_data = data->next;
 292                 early_iounmap(data, PAGE_SIZE);
 293         }
 294 }
 295
 296 #ifdef CONFIG_PCI_MMCONFIG
 297 extern void __cpuinit fam10h_check_enable_mmcfg(void);
 298 extern void __init check_enable_amd_mmconf_dmi(void);
 299 #else
 300 void __cpuinit fam10h_check_enable_mmcfg(void)
 301 {
 302 }
 303 void __init check_enable_amd_mmconf_dmi(void)
 304 {
 305 }
 306 #endif
 307
 308 /*
 309  * setup_arch - architecture-specific boot-time initializations
 310  *
 311  * Note: On x86_64, fixmaps are ready for use even before this is called.
 312  */
 313 void __init setup_arch(char **cmdline_p)
 314 {
 315         unsigned i;
 316
 317         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 318
 319         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 320         screen_info = boot_params.screen_info;
 321         edid_info = boot_params.edid_info;
 322         saved_video_mode = boot_params.hdr.vid_mode;
 323         bootloader_type = boot_params.hdr.type_of_loader;
 324
 325 #ifdef CONFIG_BLK_DEV_RAM
 326         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 327         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 328         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 329 #endif
 330 #ifdef CONFIG_EFI
 331         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 332                      "EL64", 4))
 333                 efi_enabled = 1;
 334 #endif
 335
 336         ARCH_SETUP
 337
 338         memory_setup();
 339         copy_edd();
 340
 341         if (!boot_params.hdr.root_flags)
 342                 root_mountflags &= ~MS_RDONLY;
 343         init_mm.start_code = (unsigned long) &_text;
 344         init_mm.end_code = (unsigned long) &_etext;
 345         init_mm.end_data = (unsigned long) &_edata;
 346         init_mm.brk = (unsigned long) &_end;
 347
 348         code_resource.start = virt_to_phys(&_text);
 349         code_resource.end = virt_to_phys(&_etext)-1;
 350         data_resource.start = virt_to_phys(&_etext);
 351         data_resource.end = virt_to_phys(&_edata)-1;
 352         bss_resource.start = virt_to_phys(&__bss_start);
 353         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 354
 355         early_identify_cpu(&boot_cpu_data);
 356
 357         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 358         *cmdline_p = command_line;
 359
 360         parse_setup_data();
 361
 362         parse_early_param();
 363
 364 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 365         if (init_ohci1394_dma_early)
 366                 init_ohci1394_dma_on_all_controllers();
 367 #endif
 368
 369         finish_e820_parsing();
 370
 371         /* after parse_early_param, so could debug it */
 372         insert_resource(&iomem_resource, &code_resource);
 373         insert_resource(&iomem_resource, &data_resource);
 374         insert_resource(&iomem_resource, &bss_resource);
 375
 376         early_gart_iommu_check();
 377
 378         e820_register_active_regions(0, 0, -1UL);
 379         /*
 380          * partially used pages are not usable - thus
 381          * we are rounding upwards:
 382          */
 383         end_pfn = e820_end_of_ram();
 384         /* update e820 for memory not covered by WB MTRRs */
 385         mtrr_bp_init();
 386         if (mtrr_trim_uncached_memory(end_pfn)) {
 387                 e820_register_active_regions(0, 0, -1UL);
 388                 end_pfn = e820_end_of_ram();
 389         }
 390
 391         num_physpages = end_pfn;
 392
 393         check_efer();
 394
 395         max_pfn_mapped = init_memory_mapping(0, (max_pfn_mapped << PAGE_SHIFT));
 396         if (efi_enabled)
 397                 efi_init();
 398
 399         vsmp_init();
 400
 401         dmi_scan_machine();
 402
 403         io_delay_init();
 404
 405 #ifdef CONFIG_KVM_CLOCK
 406         kvmclock_init();
 407 #endif
 408
 409 #ifdef CONFIG_SMP
 410         /* setup to use the early static init tables during kernel startup */
 411         x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
 412         x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
 413 #ifdef CONFIG_NUMA
 414         x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
 415 #endif
 416 #endif
 417
 418 #ifdef CONFIG_ACPI
 419         /*
 420          * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
 421          * Call this early for SRAT node setup.
 422          */
 423         acpi_boot_table_init();
 424 #endif
 425
 426         /* How many end-of-memory variables you have, grandma! */
 427         max_low_pfn = end_pfn;
 428         max_pfn = end_pfn;
 429         high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
 430
 431         /* Remove active ranges so rediscovery with NUMA-awareness happens */
 432         remove_all_active_ranges();
 433
 434 #ifdef CONFIG_ACPI_NUMA
 435         /*
 436          * Parse SRAT to discover nodes.
 437          */
 438         acpi_numa_init();
 439 #endif
 440
 441 #ifdef CONFIG_NUMA
 442         numa_initmem_init(0, end_pfn);
 443 #else
 444         contig_initmem_init(0, end_pfn);
 445 #endif
 446
 447         dma32_reserve_bootmem();
 448
 449 #ifdef CONFIG_ACPI_SLEEP
 450         /*
 451          * Reserve low memory region for sleep support.
 452          */
 453        acpi_reserve_bootmem();
 454 #endif
 455
 456         if (efi_enabled)
 457                 efi_reserve_bootmem();
 458
 459        /*
 460         * Find and reserve possible boot-time SMP configuration:
 461         */
 462         find_smp_config();
 463 #ifdef CONFIG_BLK_DEV_INITRD
 464         if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
 465                 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
 466                 unsigned long ramdisk_size  = boot_params.hdr.ramdisk_size;
 467                 unsigned long ramdisk_end   = ramdisk_image + ramdisk_size;
 468                 unsigned long end_of_mem    = end_pfn << PAGE_SHIFT;
 469
 470                 if (ramdisk_end <= end_of_mem) {
 471                         /*
 472                          * don't need to reserve again, already reserved early
 473                          * in x86_64_start_kernel, and early_res_to_bootmem
 474                          * convert that to reserved in bootmem
 475                          */
 476                         initrd_start = ramdisk_image + PAGE_OFFSET;
 477                         initrd_end = initrd_start+ramdisk_size;
 478                 } else {
 479                         free_bootmem(ramdisk_image, ramdisk_size);
 480                         printk(KERN_ERR "initrd extends beyond end of memory "
 481                                "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
 482                                ramdisk_end, end_of_mem);
 483                         initrd_start = 0;
 484                 }
 485         }
 486 #endif
 487         reserve_crashkernel();
 488
 489         reserve_ibft_region();
 490
 491         paging_init();
 492         map_vsyscall();
 493
 494         early_quirks();
 495
 496 #ifdef CONFIG_ACPI
 497         /*
 498          * Read APIC and some other early information from ACPI tables.
 499          */
 500         acpi_boot_init();
 501 #endif
 502
 503         init_cpu_to_node();
 504
 505         /*
 506          * get boot-time SMP configuration:
 507          */
 508         if (smp_found_config)
 509                 get_smp_config();
 510         init_apic_mappings();
 511         ioapic_init_mappings();
 512
 513         kvm_guest_init();
 514
 515         /*
 516          * We trust e820 completely. No explicit ROM probing in memory.
 517          */
 518         e820_reserve_resources();
 519         e820_mark_nosave_regions();
 520
 521         /* request I/O space for devices used on all i[345]86 PCs */
 522         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 523                 request_resource(&ioport_resource, &standard_io_resources[i]);
 524
 525         e820_setup_gap();
 526
 527 #ifdef CONFIG_VT
 528 #if defined(CONFIG_VGA_CONSOLE)
 529         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 530                 conswitchp = &vga_con;
 531 #elif defined(CONFIG_DUMMY_CONSOLE)
 532         conswitchp = &dummy_con;
 533 #endif
 534 #endif
 535
 536         /* do this before identify_cpu for boot cpu */
 537         check_enable_amd_mmconf_dmi();
 538 }
 539
 540 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
 541 {
 542         unsigned int *v;
 543
 544         if (c->extended_cpuid_level < 0x80000004)
 545                 return 0;
 546
 547         v = (unsigned int *) c->x86_model_id;
 548         cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
 549         cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
 550         cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
 551         c->x86_model_id[48] = 0;
 552         return 1;
 553 }
 554
 555
 556 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
 557 {
 558         unsigned int n, dummy, eax, ebx, ecx, edx;
 559
 560         n = c->extended_cpuid_level;
 561
 562         if (n >= 0x80000005) {
 563                 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
 564                 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
 565                        "D cache %dK (%d bytes/line)\n",
 566                        edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
 567                 c->x86_cache_size = (ecx>>24) + (edx>>24);
 568                 /* On K8 L1 TLB is inclusive, so don't count it */
 569                 c->x86_tlbsize = 0;
 570         }
 571
 572         if (n >= 0x80000006) {
 573                 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
 574                 ecx = cpuid_ecx(0x80000006);
 575                 c->x86_cache_size = ecx >> 16;
 576                 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
 577
 578                 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
 579                 c->x86_cache_size, ecx & 0xFF);
 580         }
 581         if (n >= 0x80000008) {
 582                 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
 583                 c->x86_virt_bits = (eax >> 8) & 0xff;
 584                 c->x86_phys_bits = eax & 0xff;
 585         }
 586 }
 587
 588 #ifdef CONFIG_NUMA
 589 static int __cpuinit nearby_node(int apicid)
 590 {
 591         int i, node;
 592
 593         for (i = apicid - 1; i >= 0; i--) {
 594                 node = apicid_to_node[i];
 595                 if (node != NUMA_NO_NODE && node_online(node))
 596                         return node;
 597         }
 598         for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
 599                 node = apicid_to_node[i];
 600                 if (node != NUMA_NO_NODE && node_online(node))
 601                         return node;
 602         }
 603         return first_node(node_online_map); /* Shouldn't happen */
 604 }
 605 #endif
 606
 607 /*
 608  * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
 609  * Assumes number of cores is a power of two.
 610  */
 611 static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
 612 {
 613 #ifdef CONFIG_SMP
 614         unsigned bits;
 615 #ifdef CONFIG_NUMA
 616         int cpu = smp_processor_id();
 617         int node = 0;
 618         unsigned apicid = hard_smp_processor_id();
 619 #endif
 620         bits = c->x86_coreid_bits;
 621
 622         /* Low order bits define the core id (index of core in socket) */
 623         c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
 624         /* Convert the initial APIC ID into the socket ID */
 625         c->phys_proc_id = c->initial_apicid >> bits;
 626
 627 #ifdef CONFIG_NUMA
 628         node = c->phys_proc_id;
 629         if (apicid_to_node[apicid] != NUMA_NO_NODE)
 630                 node = apicid_to_node[apicid];
 631         if (!node_online(node)) {
 632                 /* Two possibilities here:
 633                    - The CPU is missing memory and no node was created.
 634                    In that case try picking one from a nearby CPU
 635                    - The APIC IDs differ from the HyperTransport node IDs
 636                    which the K8 northbridge parsing fills in.
 637                    Assume they are all increased by a constant offset,
 638                    but in the same order as the HT nodeids.
 639                    If that doesn't result in a usable node fall back to the
 640                    path for the previous case.  */
 641
 642                 int ht_nodeid = c->initial_apicid;
 643
 644                 if (ht_nodeid >= 0 &&
 645                     apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
 646                         node = apicid_to_node[ht_nodeid];
 647                 /* Pick a nearby node */
 648                 if (!node_online(node))
 649                         node = nearby_node(apicid);
 650         }
 651         numa_set_node(cpu, node);
 652
 653         printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
 654 #endif
 655 #endif
 656 }
 657
 658 static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
 659 {
 660 #ifdef CONFIG_SMP
 661         unsigned bits, ecx;
 662
 663         /* Multi core CPU? */
 664         if (c->extended_cpuid_level < 0x80000008)
 665                 return;
 666
 667         ecx = cpuid_ecx(0x80000008);
 668
 669         c->x86_max_cores = (ecx & 0xff) + 1;
 670
 671         /* CPU telling us the core id bits shift? */
 672         bits = (ecx >> 12) & 0xF;
 673
 674         /* Otherwise recompute */
 675         if (bits == 0) {
 676                 while ((1 << bits) < c->x86_max_cores)
 677                         bits++;
 678         }
 679
 680         c->x86_coreid_bits = bits;
 681
 682 #endif
 683 }
 684
 685 #define ENABLE_C1E_MASK         0x18000000
 686 #define CPUID_PROCESSOR_SIGNATURE       1
 687 #define CPUID_XFAM              0x0ff00000
 688 #define CPUID_XFAM_K8           0x00000000
 689 #define CPUID_XFAM_10H          0x00100000
 690 #define CPUID_XFAM_11H          0x00200000
 691 #define CPUID_XMOD              0x000f0000
 692 #define CPUID_XMOD_REV_F        0x00040000
 693
 694 /* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
 695 static __cpuinit int amd_apic_timer_broken(void)
 696 {
 697         u32 lo, hi, eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
 698
 699         switch (eax & CPUID_XFAM) {
 700         case CPUID_XFAM_K8:
 701                 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
 702                         break;
 703         case CPUID_XFAM_10H:
 704         case CPUID_XFAM_11H:
 705                 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
 706                 if (lo & ENABLE_C1E_MASK)
 707                         return 1;
 708                 break;
 709         default:
 710                 /* err on the side of caution */
 711                 return 1;
 712         }
 713         return 0;
 714 }
 715
 716 static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
 717 {
 718         early_init_amd_mc(c);
 719
 720         /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
 721         if (c->x86_power & (1<<8))
 722                 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
 723 }
 724
 725 static void __cpuinit init_amd(struct cpuinfo_x86 *c)
 726 {
 727         unsigned level;
 728
 729 #ifdef CONFIG_SMP
 730         unsigned long value;
 731
 732         /*
 733          * Disable TLB flush filter by setting HWCR.FFDIS on K8
 734          * bit 6 of msr C001_0015
 735          *
 736          * Errata 63 for SH-B3 steppings
 737          * Errata 122 for all steppings (F+ have it disabled by default)
 738          */
 739         if (c->x86 == 15) {
 740                 rdmsrl(MSR_K8_HWCR, value);
 741                 value |= 1 << 6;
 742                 wrmsrl(MSR_K8_HWCR, value);
 743         }
 744 #endif
 745
 746         /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
 747            3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
 748         clear_cpu_cap(c, 0*32+31);
 749
 750         /* On C+ stepping K8 rep microcode works well for copy/memset */
 751         level = cpuid_eax(1);
 752         if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) ||
 753                              level >= 0x0f58))
 754                 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
 755         if (c->x86 == 0x10 || c->x86 == 0x11)
 756                 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
 757
 758         /* Enable workaround for FXSAVE leak */
 759         if (c->x86 >= 6)
 760                 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
 761
 762         level = get_model_name(c);
 763         if (!level) {
 764                 switch (c->x86) {
 765                 case 15:
 766                         /* Should distinguish Models here, but this is only
 767                            a fallback anyways. */
 768                         strcpy(c->x86_model_id, "Hammer");
 769                         break;
 770                 }
 771         }
 772         display_cacheinfo(c);
 773
 774         /* Multi core CPU? */
 775         if (c->extended_cpuid_level >= 0x80000008)
 776                 amd_detect_cmp(c);
 777
 778         if (c->extended_cpuid_level >= 0x80000006 &&
 779                 (cpuid_edx(0x80000006) & 0xf000))
 780                 num_cache_leaves = 4;
 781         else
 782                 num_cache_leaves = 3;
 783
 784         if (c->x86 == 0xf || c->x86 == 0x10 || c->x86 == 0x11)
 785                 set_cpu_cap(c, X86_FEATURE_K8);
 786
 787         /* MFENCE stops RDTSC speculation */
 788         set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
 789
 790         if (c->x86 == 0x10)
 791                 fam10h_check_enable_mmcfg();
 792
 793         if (amd_apic_timer_broken())
 794                 disable_apic_timer = 1;
 795
 796         if (c == &boot_cpu_data && c->x86 >= 0xf && c->x86 <= 0x11) {
 797                 unsigned long long tseg;
 798
 799                 /*
 800                  * Split up direct mapping around the TSEG SMM area.
 801                  * Don't do it for gbpages because there seems very little
 802                  * benefit in doing so.
 803                  */
 804                 if (!rdmsrl_safe(MSR_K8_TSEG_ADDR, &tseg) &&
 805                 (tseg >> PMD_SHIFT) < (max_pfn_mapped >> (PMD_SHIFT-PAGE_SHIFT)))
 806                         set_memory_4k((unsigned long)__va(tseg), 1);
 807         }
 808 }
 809
 810 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
 811 {
 812 #ifdef CONFIG_SMP
 813         u32 eax, ebx, ecx, edx;
 814         int index_msb, core_bits;
 815
 816         cpuid(1, &eax, &ebx, &ecx, &edx);
 817
 818
 819         if (!cpu_has(c, X86_FEATURE_HT))
 820                 return;
 821         if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
 822                 goto out;
 823
 824         smp_num_siblings = (ebx & 0xff0000) >> 16;
 825
 826         if (smp_num_siblings == 1) {
 827                 printk(KERN_INFO  "CPU: Hyper-Threading is disabled\n");
 828         } else if (smp_num_siblings > 1) {
 829
 830                 if (smp_num_siblings > NR_CPUS) {
 831                         printk(KERN_WARNING "CPU: Unsupported number of "
 832                                "siblings %d", smp_num_siblings);
 833                         smp_num_siblings = 1;
 834                         return;
 835                 }
 836
 837                 index_msb = get_count_order(smp_num_siblings);
 838                 c->phys_proc_id = phys_pkg_id(index_msb);
 839
 840                 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
 841
 842                 index_msb = get_count_order(smp_num_siblings);
 843
 844                 core_bits = get_count_order(c->x86_max_cores);
 845
 846                 c->cpu_core_id = phys_pkg_id(index_msb) &
 847                                                ((1 << core_bits) - 1);
 848         }
 849 out:
 850         if ((c->x86_max_cores * smp_num_siblings) > 1) {
 851                 printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
 852                        c->phys_proc_id);
 853                 printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
 854                        c->cpu_core_id);
 855         }
 856
 857 #endif
 858 }
 859
 860 /*
 861  * find out the number of processor cores on the die
 862  */
 863 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
 864 {
 865         unsigned int eax, t;
 866
 867         if (c->cpuid_level < 4)
 868                 return 1;
 869
 870         cpuid_count(4, 0, &eax, &t, &t, &t);
 871
 872         if (eax & 0x1f)
 873                 return ((eax >> 26) + 1);
 874         else
 875                 return 1;
 876 }
 877
 878 static void __cpuinit srat_detect_node(void)
 879 {
 880 #ifdef CONFIG_NUMA
 881         unsigned node;
 882         int cpu = smp_processor_id();
 883         int apicid = hard_smp_processor_id();
 884
 885         /* Don't do the funky fallback heuristics the AMD version employs
 886            for now. */
 887         node = apicid_to_node[apicid];
 888         if (node == NUMA_NO_NODE || !node_online(node))
 889                 node = first_node(node_online_map);
 890         numa_set_node(cpu, node);
 891
 892         printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
 893 #endif
 894 }
 895
 896 static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
 897 {
 898         if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
 899             (c->x86 == 0x6 && c->x86_model >= 0x0e))
 900                 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
 901 }
 902
 903 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
 904 {
 905         /* Cache sizes */
 906         unsigned n;
 907
 908         init_intel_cacheinfo(c);
 909         if (c->cpuid_level > 9) {
 910                 unsigned eax = cpuid_eax(10);
 911                 /* Check for version and the number of counters */
 912                 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
 913                         set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
 914         }
 915
 916         if (cpu_has_ds) {
 917                 unsigned int l1, l2;
 918                 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
 919                 if (!(l1 & (1<<11)))
 920                         set_cpu_cap(c, X86_FEATURE_BTS);
 921                 if (!(l1 & (1<<12)))
 922                         set_cpu_cap(c, X86_FEATURE_PEBS);
 923         }
 924
 925
 926         if (cpu_has_bts)
 927                 ds_init_intel(c);
 928
 929         n = c->extended_cpuid_level;
 930         if (n >= 0x80000008) {
 931                 unsigned eax = cpuid_eax(0x80000008);
 932                 c->x86_virt_bits = (eax >> 8) & 0xff;
 933                 c->x86_phys_bits = eax & 0xff;
 934                 /* CPUID workaround for Intel 0F34 CPU */
 935                 if (c->x86_vendor == X86_VENDOR_INTEL &&
 936                     c->x86 == 0xF && c->x86_model == 0x3 &&
 937                     c->x86_mask == 0x4)
 938                         c->x86_phys_bits = 36;
 939         }
 940
 941         if (c->x86 == 15)
 942                 c->x86_cache_alignment = c->x86_clflush_size * 2;
 943         if (c->x86 == 6)
 944                 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
 945         set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
 946         c->x86_max_cores = intel_num_cpu_cores(c);
 947
 948         srat_detect_node();
 949 }
 950
 951 static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
 952 {
 953         if (c->x86 == 0x6 && c->x86_model >= 0xf)
 954                 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
 955 }
 956
 957 static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
 958 {
 959         /* Cache sizes */
 960         unsigned n;
 961
 962         n = c->extended_cpuid_level;
 963         if (n >= 0x80000008) {
 964                 unsigned eax = cpuid_eax(0x80000008);
 965                 c->x86_virt_bits = (eax >> 8) & 0xff;
 966                 c->x86_phys_bits = eax & 0xff;
 967         }
 968
 969         if (c->x86 == 0x6 && c->x86_model >= 0xf) {
 970                 c->x86_cache_alignment = c->x86_clflush_size * 2;
 971                 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
 972                 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
 973         }
 974         set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
 975 }
 976
 977 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
 978 {
 979         char *v = c->x86_vendor_id;
 980
 981         if (!strcmp(v, "AuthenticAMD"))
 982                 c->x86_vendor = X86_VENDOR_AMD;
 983         else if (!strcmp(v, "GenuineIntel"))
 984                 c->x86_vendor = X86_VENDOR_INTEL;
 985         else if (!strcmp(v, "CentaurHauls"))
 986                 c->x86_vendor = X86_VENDOR_CENTAUR;
 987         else
 988                 c->x86_vendor = X86_VENDOR_UNKNOWN;
 989 }
 990
 991 /* Do some early cpuid on the boot CPU to get some parameter that are
 992    needed before check_bugs. Everything advanced is in identify_cpu
 993    below. */
 994 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
 995 {
 996         u32 tfms, xlvl;
 997
 998         c->loops_per_jiffy = loops_per_jiffy;
 999         c->x86_cache_size = -1;
1000         c->x86_vendor = X86_VENDOR_UNKNOWN;
1001         c->x86_model = c->x86_mask = 0; /* So far unknown... */
1002         c->x86_vendor_id[0] = '\0'; /* Unset */
1003         c->x86_model_id[0] = '\0';  /* Unset */
1004         c->x86_clflush_size = 64;
1005         c->x86_cache_alignment = c->x86_clflush_size;
1006         c->x86_max_cores = 1;
1007         c->x86_coreid_bits = 0;
1008         c->extended_cpuid_level = 0;
1009         memset(&c->x86_capability, 0, sizeof c->x86_capability);
1010
1011         /* Get vendor name */
1012         cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
1013               (unsigned int *)&c->x86_vendor_id[0],
1014               (unsigned int *)&c->x86_vendor_id[8],
1015               (unsigned int *)&c->x86_vendor_id[4]);
1016
1017         get_cpu_vendor(c);
1018
1019         /* Initialize the standard set of capabilities */
1020         /* Note that the vendor-specific code below might override */
1021
1022         /* Intel-defined flags: level 0x00000001 */
1023         if (c->cpuid_level >= 0x00000001) {
1024                 __u32 misc;
1025                 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
1026                       &c->x86_capability[0]);
1027                 c->x86 = (tfms >> 8) & 0xf;
1028                 c->x86_model = (tfms >> 4) & 0xf;
1029                 c->x86_mask = tfms & 0xf;
1030                 if (c->x86 == 0xf)
1031                         c->x86 += (tfms >> 20) & 0xff;
1032                 if (c->x86 >= 0x6)
1033                         c->x86_model += ((tfms >> 16) & 0xF) << 4;
1034                 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
1035                         c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1036         } else {
1037                 /* Have CPUID level 0 only - unheard of */
1038                 c->x86 = 4;
1039         }
1040
1041         c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
1042 #ifdef CONFIG_SMP
1043         c->phys_proc_id = c->initial_apicid;
1044 #endif
1045         /* AMD-defined flags: level 0x80000001 */
1046         xlvl = cpuid_eax(0x80000000);
1047         c->extended_cpuid_level = xlvl;
1048         if ((xlvl & 0xffff0000) == 0x80000000) {
1049                 if (xlvl >= 0x80000001) {
1050                         c->x86_capability[1] = cpuid_edx(0x80000001);
1051                         c->x86_capability[6] = cpuid_ecx(0x80000001);
1052                 }
1053                 if (xlvl >= 0x80000004)
1054                         get_model_name(c); /* Default name */
1055         }
1056
1057         /* Transmeta-defined flags: level 0x80860001 */
1058         xlvl = cpuid_eax(0x80860000);
1059         if ((xlvl & 0xffff0000) == 0x80860000) {
1060                 /* Don't set x86_cpuid_level here for now to not confuse. */
1061                 if (xlvl >= 0x80860001)
1062                         c->x86_capability[2] = cpuid_edx(0x80860001);
1063         }
1064
1065         c->extended_cpuid_level = cpuid_eax(0x80000000);
1066         if (c->extended_cpuid_level >= 0x80000007)
1067                 c->x86_power = cpuid_edx(0x80000007);
1068
1069         switch (c->x86_vendor) {
1070         case X86_VENDOR_AMD:
1071                 early_init_amd(c);
1072                 break;
1073         case X86_VENDOR_INTEL:
1074                 early_init_intel(c);
1075                 break;
1076         case X86_VENDOR_CENTAUR:
1077                 early_init_centaur(c);
1078                 break;
1079         }
1080
1081         validate_pat_support(c);
1082 }
1083
1084 /*
1085  * This does the hard work of actually picking apart the CPU stuff...
1086  */
1087 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1088 {
1089         int i;
1090
1091         early_identify_cpu(c);
1092
1093         init_scattered_cpuid_features(c);
1094
1095         c->apicid = phys_pkg_id(0);
1096
1097         /*
1098          * Vendor-specific initialization.  In this section we
1099          * canonicalize the feature flags, meaning if there are
1100          * features a certain CPU supports which CPUID doesn't
1101          * tell us, CPUID claiming incorrect flags, or other bugs,
1102          * we handle them here.
1103          *
1104          * At the end of this section, c->x86_capability better
1105          * indicate the features this CPU genuinely supports!
1106          */
1107         switch (c->x86_vendor) {
1108         case X86_VENDOR_AMD:
1109                 init_amd(c);
1110                 break;
1111
1112         case X86_VENDOR_INTEL:
1113                 init_intel(c);
1114                 break;
1115
1116         case X86_VENDOR_CENTAUR:
1117                 init_centaur(c);
1118                 break;
1119
1120         case X86_VENDOR_UNKNOWN:
1121         default:
1122                 display_cacheinfo(c);
1123                 break;
1124         }
1125
1126         detect_ht(c);
1127
1128         /*
1129          * On SMP, boot_cpu_data holds the common feature set between
1130          * all CPUs; so make sure that we indicate which features are
1131          * common between the CPUs.  The first time this routine gets
1132          * executed, c == &boot_cpu_data.
1133          */
1134         if (c != &boot_cpu_data) {
1135                 /* AND the already accumulated flags with these */
1136                 for (i = 0; i < NCAPINTS; i++)
1137                         boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1138         }
1139
1140         /* Clear all flags overriden by options */
1141         for (i = 0; i < NCAPINTS; i++)
1142                 c->x86_capability[i] &= ~cleared_cpu_caps[i];
1143
1144 #ifdef CONFIG_X86_MCE
1145         mcheck_init(c);
1146 #endif
1147         select_idle_routine(c);
1148
1149 #ifdef CONFIG_NUMA
1150         numa_add_cpu(smp_processor_id());
1151 #endif
1152
1153 }
1154
1155 void __cpuinit identify_boot_cpu(void)
1156 {
1157         identify_cpu(&boot_cpu_data);
1158 }
1159
1160 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
1161 {
1162         BUG_ON(c == &boot_cpu_data);
1163         identify_cpu(c);
1164         mtrr_ap_init();
1165 }
1166
1167 static __init int setup_noclflush(char *arg)
1168 {
1169         setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
1170         return 1;
1171 }
1172 __setup("noclflush", setup_noclflush);
1173
1174 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1175 {
1176         if (c->x86_model_id[0])
1177                 printk(KERN_CONT "%s", c->x86_model_id);
1178
1179         if (c->x86_mask || c->cpuid_level >= 0)
1180                 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
1181         else
1182                 printk(KERN_CONT "\n");
1183 }
1184
1185 static __init int setup_disablecpuid(char *arg)
1186 {
1187         int bit;
1188         if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1189                 setup_clear_cpu_cap(bit);
1190         else
1191                 return 0;
1192         return 1;
1193 }
1194 __setup("clearcpuid=", setup_disablecpuid);