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 #include <asm/mmconfig.h>
  75
  76 #include <mach_apic.h>
  77 #ifdef CONFIG_PARAVIRT
  78 #include <asm/paravirt.h>
  79 #else
  80 #define ARCH_SETUP
  81 #endif
  82
  83 /*
  84  * Machine setup..
  85  */
  86
  87 struct cpuinfo_x86 boot_cpu_data __read_mostly;
  88 EXPORT_SYMBOL(boot_cpu_data);
  89
  90 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
  91
  92 unsigned long mmu_cr4_features;
  93
  94 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
  95 int bootloader_type;
  96
  97 unsigned long saved_video_mode;
  98
  99 /*
 100  * Early DMI memory
 101  */
 102 int dmi_alloc_index;
 103 char dmi_alloc_data[DMI_MAX_DATA];
 104
 105 /*
 106  * Setup options
 107  */
 108 struct screen_info screen_info;
 109 EXPORT_SYMBOL(screen_info);
 110 struct sys_desc_table_struct {
 111         unsigned short length;
 112         unsigned char table[0];
 113 };
 114
 115 struct edid_info edid_info;
 116 EXPORT_SYMBOL_GPL(edid_info);
 117
 118 extern int root_mountflags;
 119
 120 char __initdata command_line[COMMAND_LINE_SIZE];
 121
 122 static struct resource standard_io_resources[] = {
 123         { .name = "dma1", .start = 0x00, .end = 0x1f,
 124                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 125         { .name = "pic1", .start = 0x20, .end = 0x21,
 126                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 127         { .name = "timer0", .start = 0x40, .end = 0x43,
 128                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 129         { .name = "timer1", .start = 0x50, .end = 0x53,
 130                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 131         { .name = "keyboard", .start = 0x60, .end = 0x60,
 132                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 133         { .name = "keyboard", .start = 0x64, .end = 0x64,
 134                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 135         { .name = "dma page reg", .start = 0x80, .end = 0x8f,
 136                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 137         { .name = "pic2", .start = 0xa0, .end = 0xa1,
 138                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 139         { .name = "dma2", .start = 0xc0, .end = 0xdf,
 140                 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
 141         { .name = "fpu", .start = 0xf0, .end = 0xff,
 142                 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
 143 };
 144
 145 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
 146
 147 static struct resource data_resource = {
 148         .name = "Kernel data",
 149         .start = 0,
 150         .end = 0,
 151         .flags = IORESOURCE_RAM,
 152 };
 153 static struct resource code_resource = {
 154         .name = "Kernel code",
 155         .start = 0,
 156         .end = 0,
 157         .flags = IORESOURCE_RAM,
 158 };
 159 static struct resource bss_resource = {
 160         .name = "Kernel bss",
 161         .start = 0,
 162         .end = 0,
 163         .flags = IORESOURCE_RAM,
 164 };
 165
 166 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
 167
 168 #ifdef CONFIG_PROC_VMCORE
 169 /* elfcorehdr= specifies the location of elf core header
 170  * stored by the crashed kernel. This option will be passed
 171  * by kexec loader to the capture kernel.
 172  */
 173 static int __init setup_elfcorehdr(char *arg)
 174 {
 175         char *end;
 176         if (!arg)
 177                 return -EINVAL;
 178         elfcorehdr_addr = memparse(arg, &end);
 179         return end > arg ? 0 : -EINVAL;
 180 }
 181 early_param("elfcorehdr", setup_elfcorehdr);
 182 #endif
 183
 184 #ifndef CONFIG_NUMA
 185 static void __init
 186 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
 187 {
 188         unsigned long bootmap_size, bootmap;
 189
 190         bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
 191         bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
 192                                  PAGE_SIZE);
 193         if (bootmap == -1L)
 194                 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
 195         bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
 196         e820_register_active_regions(0, start_pfn, end_pfn);
 197         free_bootmem_with_active_regions(0, end_pfn);
 198         early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
 199         reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
 200 }
 201 #endif
 202
 203 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
 204 struct edd edd;
 205 #ifdef CONFIG_EDD_MODULE
 206 EXPORT_SYMBOL(edd);
 207 #endif
 208 /**
 209  * copy_edd() - Copy the BIOS EDD information
 210  *              from boot_params into a safe place.
 211  *
 212  */
 213 static inline void copy_edd(void)
 214 {
 215      memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
 216             sizeof(edd.mbr_signature));
 217      memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
 218      edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
 219      edd.edd_info_nr = boot_params.eddbuf_entries;
 220 }
 221 #else
 222 static inline void copy_edd(void)
 223 {
 224 }
 225 #endif
 226
 227 #ifdef CONFIG_KEXEC
 228 static void __init reserve_crashkernel(void)
 229 {
 230         unsigned long long total_mem;
 231         unsigned long long crash_size, crash_base;
 232         int ret;
 233
 234         total_mem = ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
 235
 236         ret = parse_crashkernel(boot_command_line, total_mem,
 237                         &crash_size, &crash_base);
 238         if (ret == 0 && crash_size) {
 239                 if (crash_base <= 0) {
 240                         printk(KERN_INFO "crashkernel reservation failed - "
 241                                         "you have to specify a base address\n");
 242                         return;
 243                 }
 244
 245                 if (reserve_bootmem(crash_base, crash_size,
 246                                         BOOTMEM_EXCLUSIVE) < 0) {
 247                         printk(KERN_INFO "crashkernel reservation failed - "
 248                                         "memory is in use\n");
 249                         return;
 250                 }
 251
 252                 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
 253                                 "for crashkernel (System RAM: %ldMB)\n",
 254                                 (unsigned long)(crash_size >> 20),
 255                                 (unsigned long)(crash_base >> 20),
 256                                 (unsigned long)(total_mem >> 20));
 257                 crashk_res.start = crash_base;
 258                 crashk_res.end   = crash_base + crash_size - 1;
 259                 insert_resource(&iomem_resource, &crashk_res);
 260         }
 261 }
 262 #else
 263 static inline void __init reserve_crashkernel(void)
 264 {}
 265 #endif
 266
 267 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
 268 void __attribute__((weak)) __init memory_setup(void)
 269 {
 270        machine_specific_memory_setup();
 271 }
 272
 273 static void __init parse_setup_data(void)
 274 {
 275         struct setup_data *data;
 276         unsigned long pa_data;
 277
 278         if (boot_params.hdr.version < 0x0209)
 279                 return;
 280         pa_data = boot_params.hdr.setup_data;
 281         while (pa_data) {
 282                 data = early_ioremap(pa_data, PAGE_SIZE);
 283                 switch (data->type) {
 284                 default:
 285                         break;
 286                 }
 287 #ifndef CONFIG_DEBUG_BOOT_PARAMS
 288                 free_early(pa_data, pa_data+sizeof(*data)+data->len);
 289 #endif
 290                 pa_data = data->next;
 291                 early_iounmap(data, PAGE_SIZE);
 292         }
 293 }
 294
 295 /*
 296  * setup_arch - architecture-specific boot-time initializations
 297  *
 298  * Note: On x86_64, fixmaps are ready for use even before this is called.
 299  */
 300 void __init setup_arch(char **cmdline_p)
 301 {
 302         unsigned i;
 303
 304         printk(KERN_INFO "Command line: %s\n", boot_command_line);
 305
 306         ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
 307         screen_info = boot_params.screen_info;
 308         edid_info = boot_params.edid_info;
 309         saved_video_mode = boot_params.hdr.vid_mode;
 310         bootloader_type = boot_params.hdr.type_of_loader;
 311
 312 #ifdef CONFIG_BLK_DEV_RAM
 313         rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
 314         rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
 315         rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
 316 #endif
 317 #ifdef CONFIG_EFI
 318         if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
 319                      "EL64", 4))
 320                 efi_enabled = 1;
 321 #endif
 322
 323         ARCH_SETUP
 324
 325         memory_setup();
 326         copy_edd();
 327
 328         if (!boot_params.hdr.root_flags)
 329                 root_mountflags &= ~MS_RDONLY;
 330         init_mm.start_code = (unsigned long) &_text;
 331         init_mm.end_code = (unsigned long) &_etext;
 332         init_mm.end_data = (unsigned long) &_edata;
 333         init_mm.brk = (unsigned long) &_end;
 334
 335         code_resource.start = virt_to_phys(&_text);
 336         code_resource.end = virt_to_phys(&_etext)-1;
 337         data_resource.start = virt_to_phys(&_etext);
 338         data_resource.end = virt_to_phys(&_edata)-1;
 339         bss_resource.start = virt_to_phys(&__bss_start);
 340         bss_resource.end = virt_to_phys(&__bss_stop)-1;
 341
 342         early_identify_cpu(&boot_cpu_data);
 343
 344         strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
 345         *cmdline_p = command_line;
 346
 347         parse_setup_data();
 348
 349         parse_early_param();
 350
 351 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
 352         if (init_ohci1394_dma_early)
 353                 init_ohci1394_dma_on_all_controllers();
 354 #endif
 355
 356         finish_e820_parsing();
 357
 358         /* after parse_early_param, so could debug it */
 359         insert_resource(&iomem_resource, &code_resource);
 360         insert_resource(&iomem_resource, &data_resource);
 361         insert_resource(&iomem_resource, &bss_resource);
 362
 363         early_gart_iommu_check();
 364
 365         e820_register_active_regions(0, 0, -1UL);
 366         /*
 367          * partially used pages are not usable - thus
 368          * we are rounding upwards:
 369          */
 370         end_pfn = e820_end_of_ram();
 371         /* update e820 for memory not covered by WB MTRRs */
 372         mtrr_bp_init();
 373         if (mtrr_trim_uncached_memory(end_pfn)) {
 374                 e820_register_active_regions(0, 0, -1UL);
 375                 end_pfn = e820_end_of_ram();
 376         }
 377
 378         num_physpages = end_pfn;
 379
 380         check_efer();
 381
 382         max_pfn_mapped = init_memory_mapping(0, (max_pfn_mapped << PAGE_SHIFT));
 383         if (efi_enabled)
 384                 efi_init();
 385
 386         vsmp_init();
 387
 388         dmi_scan_machine();
 389
 390         io_delay_init();
 391
 392 #ifdef CONFIG_KVM_CLOCK
 393         kvmclock_init();
 394 #endif
 395
 396 #ifdef CONFIG_SMP
 397         /* setup to use the early static init tables during kernel startup */
 398         x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
 399         x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
 400 #ifdef CONFIG_NUMA
 401         x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
 402 #endif
 403 #endif
 404
 405 #ifdef CONFIG_ACPI
 406         /*
 407          * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
 408          * Call this early for SRAT node setup.
 409          */
 410         acpi_boot_table_init();
 411 #endif
 412
 413         /* How many end-of-memory variables you have, grandma! */
 414         max_low_pfn = end_pfn;
 415         max_pfn = end_pfn;
 416         high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
 417
 418         /* Remove active ranges so rediscovery with NUMA-awareness happens */
 419         remove_all_active_ranges();
 420
 421 #ifdef CONFIG_ACPI_NUMA
 422         /*
 423          * Parse SRAT to discover nodes.
 424          */
 425         acpi_numa_init();
 426 #endif
 427
 428 #ifdef CONFIG_NUMA
 429         numa_initmem_init(0, end_pfn);
 430 #else
 431         contig_initmem_init(0, end_pfn);
 432 #endif
 433
 434         dma32_reserve_bootmem();
 435
 436 #ifdef CONFIG_ACPI_SLEEP
 437         /*
 438          * Reserve low memory region for sleep support.
 439          */
 440        acpi_reserve_bootmem();
 441 #endif
 442
 443         if (efi_enabled)
 444                 efi_reserve_bootmem();
 445
 446        /*
 447         * Find and reserve possible boot-time SMP configuration:
 448         */
 449         find_smp_config();
 450 #ifdef CONFIG_BLK_DEV_INITRD
 451         if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
 452                 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
 453                 unsigned long ramdisk_size  = boot_params.hdr.ramdisk_size;
 454                 unsigned long ramdisk_end   = ramdisk_image + ramdisk_size;
 455                 unsigned long end_of_mem    = end_pfn << PAGE_SHIFT;
 456
 457                 if (ramdisk_end <= end_of_mem) {
 458                         /*
 459                          * don't need to reserve again, already reserved early
 460                          * in x86_64_start_kernel, and early_res_to_bootmem
 461                          * convert that to reserved in bootmem
 462                          */
 463                         initrd_start = ramdisk_image + PAGE_OFFSET;
 464                         initrd_end = initrd_start+ramdisk_size;
 465                 } else {
 466                         free_bootmem(ramdisk_image, ramdisk_size);
 467                         printk(KERN_ERR "initrd extends beyond end of memory "
 468                                "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
 469                                ramdisk_end, end_of_mem);
 470                         initrd_start = 0;
 471                 }
 472         }
 473 #endif
 474         reserve_crashkernel();
 475
 476         reserve_ibft_region();
 477
 478         paging_init();
 479         map_vsyscall();
 480
 481         early_quirks();
 482
 483 #ifdef CONFIG_ACPI
 484         /*
 485          * Read APIC and some other early information from ACPI tables.
 486          */
 487         acpi_boot_init();
 488 #endif
 489
 490         init_cpu_to_node();
 491
 492         /*
 493          * get boot-time SMP configuration:
 494          */
 495         if (smp_found_config)
 496                 get_smp_config();
 497         init_apic_mappings();
 498         ioapic_init_mappings();
 499
 500         kvm_guest_init();
 501
 502         /*
 503          * We trust e820 completely. No explicit ROM probing in memory.
 504          */
 505         e820_reserve_resources();
 506         e820_mark_nosave_regions();
 507
 508         /* request I/O space for devices used on all i[345]86 PCs */
 509         for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
 510                 request_resource(&ioport_resource, &standard_io_resources[i]);
 511
 512         e820_setup_gap();
 513
 514 #ifdef CONFIG_VT
 515 #if defined(CONFIG_VGA_CONSOLE)
 516         if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
 517                 conswitchp = &vga_con;
 518 #elif defined(CONFIG_DUMMY_CONSOLE)
 519         conswitchp = &dummy_con;
 520 #endif
 521 #endif
 522
 523         /* do this before identify_cpu for boot cpu */
 524         check_enable_amd_mmconf_dmi();
 525 }
 526
 527 int __cpuinit get_model_name(struct cpuinfo_x86 *c)
 528 {
 529         unsigned int *v;
 530
 531         if (c->extended_cpuid_level < 0x80000004)
 532                 return 0;
 533
 534         v = (unsigned int *) c->x86_model_id;
 535         cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
 536         cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
 537         cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
 538         c->x86_model_id[48] = 0;
 539         return 1;
 540 }
 541
 542
 543 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
 544 {
 545         unsigned int n, dummy, eax, ebx, ecx, edx;
 546
 547         n = c->extended_cpuid_level;
 548
 549         if (n >= 0x80000005) {
 550                 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
 551                 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
 552                        "D cache %dK (%d bytes/line)\n",
 553                        edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
 554                 c->x86_cache_size = (ecx>>24) + (edx>>24);
 555                 /* On K8 L1 TLB is inclusive, so don't count it */
 556                 c->x86_tlbsize = 0;
 557         }
 558
 559         if (n >= 0x80000006) {
 560                 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
 561                 ecx = cpuid_ecx(0x80000006);
 562                 c->x86_cache_size = ecx >> 16;
 563                 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
 564
 565                 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
 566                 c->x86_cache_size, ecx & 0xFF);
 567         }
 568         if (n >= 0x80000008) {
 569                 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
 570                 c->x86_virt_bits = (eax >> 8) & 0xff;
 571                 c->x86_phys_bits = eax & 0xff;
 572         }
 573 }
 574
 575 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
 576 {
 577 #ifdef CONFIG_SMP
 578         u32 eax, ebx, ecx, edx;
 579         int index_msb, core_bits;
 580
 581         cpuid(1, &eax, &ebx, &ecx, &edx);
 582
 583
 584         if (!cpu_has(c, X86_FEATURE_HT))
 585                 return;
 586         if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
 587                 goto out;
 588
 589         smp_num_siblings = (ebx & 0xff0000) >> 16;
 590
 591         if (smp_num_siblings == 1) {
 592                 printk(KERN_INFO  "CPU: Hyper-Threading is disabled\n");
 593         } else if (smp_num_siblings > 1) {
 594
 595                 if (smp_num_siblings > NR_CPUS) {
 596                         printk(KERN_WARNING "CPU: Unsupported number of "
 597                                "siblings %d", smp_num_siblings);
 598                         smp_num_siblings = 1;
 599                         return;
 600                 }
 601
 602                 index_msb = get_count_order(smp_num_siblings);
 603                 c->phys_proc_id = phys_pkg_id(index_msb);
 604
 605                 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
 606
 607                 index_msb = get_count_order(smp_num_siblings);
 608
 609                 core_bits = get_count_order(c->x86_max_cores);
 610
 611                 c->cpu_core_id = phys_pkg_id(index_msb) &
 612                                                ((1 << core_bits) - 1);
 613         }
 614 out:
 615         if ((c->x86_max_cores * smp_num_siblings) > 1) {
 616                 printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
 617                        c->phys_proc_id);
 618                 printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
 619                        c->cpu_core_id);
 620         }
 621
 622 #endif
 623 }
 624
 625 /*
 626  * find out the number of processor cores on the die
 627  */
 628 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
 629 {
 630         unsigned int eax, t;
 631
 632         if (c->cpuid_level < 4)
 633                 return 1;
 634
 635         cpuid_count(4, 0, &eax, &t, &t, &t);
 636
 637         if (eax & 0x1f)
 638                 return ((eax >> 26) + 1);
 639         else
 640                 return 1;
 641 }
 642
 643 static void __cpuinit srat_detect_node(void)
 644 {
 645 #ifdef CONFIG_NUMA
 646         unsigned node;
 647         int cpu = smp_processor_id();
 648         int apicid = hard_smp_processor_id();
 649
 650         /* Don't do the funky fallback heuristics the AMD version employs
 651            for now. */
 652         node = apicid_to_node[apicid];
 653         if (node == NUMA_NO_NODE || !node_online(node))
 654                 node = first_node(node_online_map);
 655         numa_set_node(cpu, node);
 656
 657         printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
 658 #endif
 659 }
 660
 661 static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
 662 {
 663         if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
 664             (c->x86 == 0x6 && c->x86_model >= 0x0e))
 665                 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
 666 }
 667
 668 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
 669 {
 670         /* Cache sizes */
 671         unsigned n;
 672
 673         init_intel_cacheinfo(c);
 674         if (c->cpuid_level > 9) {
 675                 unsigned eax = cpuid_eax(10);
 676                 /* Check for version and the number of counters */
 677                 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
 678                         set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
 679         }
 680
 681         if (cpu_has_ds) {
 682                 unsigned int l1, l2;
 683                 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
 684                 if (!(l1 & (1<<11)))
 685                         set_cpu_cap(c, X86_FEATURE_BTS);
 686                 if (!(l1 & (1<<12)))
 687                         set_cpu_cap(c, X86_FEATURE_PEBS);
 688         }
 689
 690
 691         if (cpu_has_bts)
 692                 ds_init_intel(c);
 693
 694         n = c->extended_cpuid_level;
 695         if (n >= 0x80000008) {
 696                 unsigned eax = cpuid_eax(0x80000008);
 697                 c->x86_virt_bits = (eax >> 8) & 0xff;
 698                 c->x86_phys_bits = eax & 0xff;
 699                 /* CPUID workaround for Intel 0F34 CPU */
 700                 if (c->x86_vendor == X86_VENDOR_INTEL &&
 701                     c->x86 == 0xF && c->x86_model == 0x3 &&
 702                     c->x86_mask == 0x4)
 703                         c->x86_phys_bits = 36;
 704         }
 705
 706         if (c->x86 == 15)
 707                 c->x86_cache_alignment = c->x86_clflush_size * 2;
 708         if (c->x86 == 6)
 709                 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
 710         set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
 711         c->x86_max_cores = intel_num_cpu_cores(c);
 712
 713         srat_detect_node();
 714 }
 715
 716 static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
 717 {
 718         if (c->x86 == 0x6 && c->x86_model >= 0xf)
 719                 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
 720 }
 721
 722 static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
 723 {
 724         /* Cache sizes */
 725         unsigned n;
 726
 727         n = c->extended_cpuid_level;
 728         if (n >= 0x80000008) {
 729                 unsigned eax = cpuid_eax(0x80000008);
 730                 c->x86_virt_bits = (eax >> 8) & 0xff;
 731                 c->x86_phys_bits = eax & 0xff;
 732         }
 733
 734         if (c->x86 == 0x6 && c->x86_model >= 0xf) {
 735                 c->x86_cache_alignment = c->x86_clflush_size * 2;
 736                 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
 737                 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
 738         }
 739         set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
 740 }
 741
 742 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
 743 {
 744         char *v = c->x86_vendor_id;
 745
 746         if (!strcmp(v, "AuthenticAMD"))
 747                 c->x86_vendor = X86_VENDOR_AMD;
 748         else if (!strcmp(v, "GenuineIntel"))
 749                 c->x86_vendor = X86_VENDOR_INTEL;
 750         else if (!strcmp(v, "CentaurHauls"))
 751                 c->x86_vendor = X86_VENDOR_CENTAUR;
 752         else
 753                 c->x86_vendor = X86_VENDOR_UNKNOWN;
 754 }
 755
 756 // FIXME: Needs to use cpu_vendor_dev_register
 757 extern void __cpuinit early_init_amd(struct cpuinfo_x86 *c);
 758 extern void __cpuinit init_amd(struct cpuinfo_x86 *c);
 759
 760 /* Do some early cpuid on the boot CPU to get some parameter that are
 761    needed before check_bugs. Everything advanced is in identify_cpu
 762    below. */
 763 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
 764 {
 765         u32 tfms, xlvl;
 766
 767         c->loops_per_jiffy = loops_per_jiffy;
 768         c->x86_cache_size = -1;
 769         c->x86_vendor = X86_VENDOR_UNKNOWN;
 770         c->x86_model = c->x86_mask = 0; /* So far unknown... */
 771         c->x86_vendor_id[0] = '\0'; /* Unset */
 772         c->x86_model_id[0] = '\0';  /* Unset */
 773         c->x86_clflush_size = 64;
 774         c->x86_cache_alignment = c->x86_clflush_size;
 775         c->x86_max_cores = 1;
 776         c->x86_coreid_bits = 0;
 777         c->extended_cpuid_level = 0;
 778         memset(&c->x86_capability, 0, sizeof c->x86_capability);
 779
 780         /* Get vendor name */
 781         cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
 782               (unsigned int *)&c->x86_vendor_id[0],
 783               (unsigned int *)&c->x86_vendor_id[8],
 784               (unsigned int *)&c->x86_vendor_id[4]);
 785
 786         get_cpu_vendor(c);
 787
 788         /* Initialize the standard set of capabilities */
 789         /* Note that the vendor-specific code below might override */
 790
 791         /* Intel-defined flags: level 0x00000001 */
 792         if (c->cpuid_level >= 0x00000001) {
 793                 __u32 misc;
 794                 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
 795                       &c->x86_capability[0]);
 796                 c->x86 = (tfms >> 8) & 0xf;
 797                 c->x86_model = (tfms >> 4) & 0xf;
 798                 c->x86_mask = tfms & 0xf;
 799                 if (c->x86 == 0xf)
 800                         c->x86 += (tfms >> 20) & 0xff;
 801                 if (c->x86 >= 0x6)
 802                         c->x86_model += ((tfms >> 16) & 0xF) << 4;
 803                 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
 804                         c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
 805         } else {
 806                 /* Have CPUID level 0 only - unheard of */
 807                 c->x86 = 4;
 808         }
 809
 810         c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
 811 #ifdef CONFIG_SMP
 812         c->phys_proc_id = c->initial_apicid;
 813 #endif
 814         /* AMD-defined flags: level 0x80000001 */
 815         xlvl = cpuid_eax(0x80000000);
 816         c->extended_cpuid_level = xlvl;
 817         if ((xlvl & 0xffff0000) == 0x80000000) {
 818                 if (xlvl >= 0x80000001) {
 819                         c->x86_capability[1] = cpuid_edx(0x80000001);
 820                         c->x86_capability[6] = cpuid_ecx(0x80000001);
 821                 }
 822                 if (xlvl >= 0x80000004)
 823                         get_model_name(c); /* Default name */
 824         }
 825
 826         /* Transmeta-defined flags: level 0x80860001 */
 827         xlvl = cpuid_eax(0x80860000);
 828         if ((xlvl & 0xffff0000) == 0x80860000) {
 829                 /* Don't set x86_cpuid_level here for now to not confuse. */
 830                 if (xlvl >= 0x80860001)
 831                         c->x86_capability[2] = cpuid_edx(0x80860001);
 832         }
 833
 834         c->extended_cpuid_level = cpuid_eax(0x80000000);
 835         if (c->extended_cpuid_level >= 0x80000007)
 836                 c->x86_power = cpuid_edx(0x80000007);
 837
 838         switch (c->x86_vendor) {
 839         case X86_VENDOR_AMD:
 840                 early_init_amd(c);
 841                 break;
 842         case X86_VENDOR_INTEL:
 843                 early_init_intel(c);
 844                 break;
 845         case X86_VENDOR_CENTAUR:
 846                 early_init_centaur(c);
 847                 break;
 848         }
 849
 850         validate_pat_support(c);
 851 }
 852
 853 /*
 854  * This does the hard work of actually picking apart the CPU stuff...
 855  */
 856 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
 857 {
 858         int i;
 859
 860         early_identify_cpu(c);
 861
 862         init_scattered_cpuid_features(c);
 863
 864         c->apicid = phys_pkg_id(0);
 865
 866         /*
 867          * Vendor-specific initialization.  In this section we
 868          * canonicalize the feature flags, meaning if there are
 869          * features a certain CPU supports which CPUID doesn't
 870          * tell us, CPUID claiming incorrect flags, or other bugs,
 871          * we handle them here.
 872          *
 873          * At the end of this section, c->x86_capability better
 874          * indicate the features this CPU genuinely supports!
 875          */
 876         switch (c->x86_vendor) {
 877         case X86_VENDOR_AMD:
 878                 init_amd(c);
 879                 break;
 880
 881         case X86_VENDOR_INTEL:
 882                 init_intel(c);
 883                 break;
 884
 885         case X86_VENDOR_CENTAUR:
 886                 init_centaur(c);
 887                 break;
 888
 889         case X86_VENDOR_UNKNOWN:
 890         default:
 891                 display_cacheinfo(c);
 892                 break;
 893         }
 894
 895         detect_ht(c);
 896
 897         /*
 898          * On SMP, boot_cpu_data holds the common feature set between
 899          * all CPUs; so make sure that we indicate which features are
 900          * common between the CPUs.  The first time this routine gets
 901          * executed, c == &boot_cpu_data.
 902          */
 903         if (c != &boot_cpu_data) {
 904                 /* AND the already accumulated flags with these */
 905                 for (i = 0; i < NCAPINTS; i++)
 906                         boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
 907         }
 908
 909         /* Clear all flags overriden by options */
 910         for (i = 0; i < NCAPINTS; i++)
 911                 c->x86_capability[i] &= ~cleared_cpu_caps[i];
 912
 913 #ifdef CONFIG_X86_MCE
 914         mcheck_init(c);
 915 #endif
 916         select_idle_routine(c);
 917
 918 #ifdef CONFIG_NUMA
 919         numa_add_cpu(smp_processor_id());
 920 #endif
 921
 922 }
 923
 924 void __cpuinit identify_boot_cpu(void)
 925 {
 926         identify_cpu(&boot_cpu_data);
 927 }
 928
 929 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
 930 {
 931         BUG_ON(c == &boot_cpu_data);
 932         identify_cpu(c);
 933         mtrr_ap_init();
 934 }
 935
 936 static __init int setup_noclflush(char *arg)
 937 {
 938         setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
 939         return 1;
 940 }
 941 __setup("noclflush", setup_noclflush);
 942
 943 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
 944 {
 945         if (c->x86_model_id[0])
 946                 printk(KERN_CONT "%s", c->x86_model_id);
 947
 948         if (c->x86_mask || c->cpuid_level >= 0)
 949                 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
 950         else
 951                 printk(KERN_CONT "\n");
 952 }
 953
 954 static __init int setup_disablecpuid(char *arg)
 955 {
 956         int bit;
 957         if (get_option(&arg, &bit) && bit < NCAPINTS*32)
 958                 setup_clear_cpu_cap(bit);
 959         else
 960                 return 0;
 961         return 1;
 962 }
 963 __setup("clearcpuid=", setup_disablecpuid);