2 * Copyright (C) 1995 Linus Torvalds
6 * This file handles the architecture-dependent parts of initialization
9 #include <linux/errno.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.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 <linux/efi.h>
33 #include <linux/acpi.h>
34 #include <linux/kallsyms.h>
35 #include <linux/edd.h>
36 #include <linux/iscsi_ibft.h>
37 #include <linux/mmzone.h>
38 #include <linux/kexec.h>
39 #include <linux/cpufreq.h>
40 #include <linux/dmi.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/ctype.h>
43 #include <linux/uaccess.h>
44 #include <linux/init_ohci1394_dma.h>
45 #include <linux/kvm_para.h>
48 #include <asm/uaccess.h>
49 #include <asm/system.h>
50 #include <asm/vsyscall.h>
55 #include <video/edid.h>
59 #include <asm/mpspec.h>
60 #include <asm/mmu_context.h>
61 #include <asm/proto.h>
62 #include <asm/setup.h>
64 #include <asm/sections.h>
66 #include <asm/cacheflush.h>
69 #include <asm/topology.h>
70 #include <asm/trampoline.h>
72 #include <mach_apic.h>
73 #ifdef CONFIG_PARAVIRT
74 #include <asm/paravirt.h>
83 struct cpuinfo_x86 boot_cpu_data __read_mostly;
84 EXPORT_SYMBOL(boot_cpu_data);
86 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
88 unsigned long mmu_cr4_features;
90 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
93 unsigned long saved_video_mode;
95 int force_mwait __cpuinitdata;
101 char dmi_alloc_data[DMI_MAX_DATA];
106 struct screen_info screen_info;
107 EXPORT_SYMBOL(screen_info);
108 struct sys_desc_table_struct {
109 unsigned short length;
110 unsigned char table[0];
113 struct edid_info edid_info;
114 EXPORT_SYMBOL_GPL(edid_info);
116 extern int root_mountflags;
118 char __initdata command_line[COMMAND_LINE_SIZE];
120 static struct resource standard_io_resources[] = {
121 { .name = "dma1", .start = 0x00, .end = 0x1f,
122 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
123 { .name = "pic1", .start = 0x20, .end = 0x21,
124 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
125 { .name = "timer0", .start = 0x40, .end = 0x43,
126 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
127 { .name = "timer1", .start = 0x50, .end = 0x53,
128 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
129 { .name = "keyboard", .start = 0x60, .end = 0x6f,
130 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
131 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
132 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
133 { .name = "pic2", .start = 0xa0, .end = 0xa1,
134 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
135 { .name = "dma2", .start = 0xc0, .end = 0xdf,
136 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
137 { .name = "fpu", .start = 0xf0, .end = 0xff,
138 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
141 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
143 static struct resource data_resource = {
144 .name = "Kernel data",
147 .flags = IORESOURCE_RAM,
149 static struct resource code_resource = {
150 .name = "Kernel code",
153 .flags = IORESOURCE_RAM,
155 static struct resource bss_resource = {
156 .name = "Kernel bss",
159 .flags = IORESOURCE_RAM,
162 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c);
164 #ifdef CONFIG_PROC_VMCORE
165 /* elfcorehdr= specifies the location of elf core header
166 * stored by the crashed kernel. This option will be passed
167 * by kexec loader to the capture kernel.
169 static int __init setup_elfcorehdr(char *arg)
174 elfcorehdr_addr = memparse(arg, &end);
175 return end > arg ? 0 : -EINVAL;
177 early_param("elfcorehdr", setup_elfcorehdr);
182 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
184 unsigned long bootmap_size, bootmap;
186 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
187 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size,
190 panic("Cannot find bootmem map of size %ld\n", bootmap_size);
191 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
192 e820_register_active_regions(0, start_pfn, end_pfn);
193 free_bootmem_with_active_regions(0, end_pfn);
194 early_res_to_bootmem(0, end_pfn<<PAGE_SHIFT);
195 reserve_bootmem(bootmap, bootmap_size, BOOTMEM_DEFAULT);
199 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
201 #ifdef CONFIG_EDD_MODULE
205 * copy_edd() - Copy the BIOS EDD information
206 * from boot_params into a safe place.
209 static inline void copy_edd(void)
211 memcpy(edd.mbr_signature, boot_params.edd_mbr_sig_buffer,
212 sizeof(edd.mbr_signature));
213 memcpy(edd.edd_info, boot_params.eddbuf, sizeof(edd.edd_info));
214 edd.mbr_signature_nr = boot_params.edd_mbr_sig_buf_entries;
215 edd.edd_info_nr = boot_params.eddbuf_entries;
218 static inline void copy_edd(void)
224 static void __init reserve_crashkernel(void)
226 unsigned long long total_mem;
227 unsigned long long crash_size, crash_base;
230 total_mem = ((unsigned long long)max_low_pfn - min_low_pfn) << PAGE_SHIFT;
232 ret = parse_crashkernel(boot_command_line, total_mem,
233 &crash_size, &crash_base);
234 if (ret == 0 && crash_size) {
235 if (crash_base <= 0) {
236 printk(KERN_INFO "crashkernel reservation failed - "
237 "you have to specify a base address\n");
241 if (reserve_bootmem(crash_base, crash_size,
242 BOOTMEM_EXCLUSIVE) < 0) {
243 printk(KERN_INFO "crashkernel reservation failed - "
244 "memory is in use\n");
248 printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
249 "for crashkernel (System RAM: %ldMB)\n",
250 (unsigned long)(crash_size >> 20),
251 (unsigned long)(crash_base >> 20),
252 (unsigned long)(total_mem >> 20));
253 crashk_res.start = crash_base;
254 crashk_res.end = crash_base + crash_size - 1;
255 insert_resource(&iomem_resource, &crashk_res);
259 static inline void __init reserve_crashkernel(void)
263 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
264 void __attribute__((weak)) __init memory_setup(void)
266 machine_specific_memory_setup();
269 static void __init parse_setup_data(void)
271 struct setup_data *data;
272 unsigned long pa_data;
274 if (boot_params.hdr.version < 0x0209)
276 pa_data = boot_params.hdr.setup_data;
278 data = early_ioremap(pa_data, PAGE_SIZE);
279 switch (data->type) {
283 #ifndef CONFIG_DEBUG_BOOT_PARAMS
284 free_early(pa_data, pa_data+sizeof(*data)+data->len);
286 pa_data = data->next;
287 early_iounmap(data, PAGE_SIZE);
292 * setup_arch - architecture-specific boot-time initializations
294 * Note: On x86_64, fixmaps are ready for use even before this is called.
296 void __init setup_arch(char **cmdline_p)
300 printk(KERN_INFO "Command line: %s\n", boot_command_line);
302 ROOT_DEV = old_decode_dev(boot_params.hdr.root_dev);
303 screen_info = boot_params.screen_info;
304 edid_info = boot_params.edid_info;
305 saved_video_mode = boot_params.hdr.vid_mode;
306 bootloader_type = boot_params.hdr.type_of_loader;
308 #ifdef CONFIG_BLK_DEV_RAM
309 rd_image_start = boot_params.hdr.ram_size & RAMDISK_IMAGE_START_MASK;
310 rd_prompt = ((boot_params.hdr.ram_size & RAMDISK_PROMPT_FLAG) != 0);
311 rd_doload = ((boot_params.hdr.ram_size & RAMDISK_LOAD_FLAG) != 0);
314 if (!strncmp((char *)&boot_params.efi_info.efi_loader_signature,
324 if (!boot_params.hdr.root_flags)
325 root_mountflags &= ~MS_RDONLY;
326 init_mm.start_code = (unsigned long) &_text;
327 init_mm.end_code = (unsigned long) &_etext;
328 init_mm.end_data = (unsigned long) &_edata;
329 init_mm.brk = (unsigned long) &_end;
331 code_resource.start = virt_to_phys(&_text);
332 code_resource.end = virt_to_phys(&_etext)-1;
333 data_resource.start = virt_to_phys(&_etext);
334 data_resource.end = virt_to_phys(&_edata)-1;
335 bss_resource.start = virt_to_phys(&__bss_start);
336 bss_resource.end = virt_to_phys(&__bss_stop)-1;
338 early_identify_cpu(&boot_cpu_data);
340 strlcpy(command_line, boot_command_line, COMMAND_LINE_SIZE);
341 *cmdline_p = command_line;
347 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
348 if (init_ohci1394_dma_early)
349 init_ohci1394_dma_on_all_controllers();
352 finish_e820_parsing();
354 /* after parse_early_param, so could debug it */
355 insert_resource(&iomem_resource, &code_resource);
356 insert_resource(&iomem_resource, &data_resource);
357 insert_resource(&iomem_resource, &bss_resource);
359 early_gart_iommu_check();
361 e820_register_active_regions(0, 0, -1UL);
363 * partially used pages are not usable - thus
364 * we are rounding upwards:
366 end_pfn = e820_end_of_ram();
367 /* update e820 for memory not covered by WB MTRRs */
369 if (mtrr_trim_uncached_memory(end_pfn)) {
370 e820_register_active_regions(0, 0, -1UL);
371 end_pfn = e820_end_of_ram();
374 num_physpages = end_pfn;
378 max_pfn_mapped = init_memory_mapping(0, (max_pfn_mapped << PAGE_SHIFT));
388 #ifdef CONFIG_KVM_CLOCK
393 /* setup to use the early static init tables during kernel startup */
394 x86_cpu_to_apicid_early_ptr = (void *)x86_cpu_to_apicid_init;
395 x86_bios_cpu_apicid_early_ptr = (void *)x86_bios_cpu_apicid_init;
397 x86_cpu_to_node_map_early_ptr = (void *)x86_cpu_to_node_map_init;
403 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
404 * Call this early for SRAT node setup.
406 acpi_boot_table_init();
409 /* How many end-of-memory variables you have, grandma! */
410 max_low_pfn = end_pfn;
412 high_memory = (void *)__va(end_pfn * PAGE_SIZE - 1) + 1;
414 /* Remove active ranges so rediscovery with NUMA-awareness happens */
415 remove_all_active_ranges();
417 #ifdef CONFIG_ACPI_NUMA
419 * Parse SRAT to discover nodes.
425 numa_initmem_init(0, end_pfn);
427 contig_initmem_init(0, end_pfn);
430 dma32_reserve_bootmem();
432 #ifdef CONFIG_ACPI_SLEEP
434 * Reserve low memory region for sleep support.
436 acpi_reserve_bootmem();
440 efi_reserve_bootmem();
443 * Find and reserve possible boot-time SMP configuration:
446 #ifdef CONFIG_BLK_DEV_INITRD
447 if (boot_params.hdr.type_of_loader && boot_params.hdr.ramdisk_image) {
448 unsigned long ramdisk_image = boot_params.hdr.ramdisk_image;
449 unsigned long ramdisk_size = boot_params.hdr.ramdisk_size;
450 unsigned long ramdisk_end = ramdisk_image + ramdisk_size;
451 unsigned long end_of_mem = end_pfn << PAGE_SHIFT;
453 if (ramdisk_end <= end_of_mem) {
455 * don't need to reserve again, already reserved early
456 * in x86_64_start_kernel, and early_res_to_bootmem
457 * convert that to reserved in bootmem
459 initrd_start = ramdisk_image + PAGE_OFFSET;
460 initrd_end = initrd_start+ramdisk_size;
462 free_bootmem(ramdisk_image, ramdisk_size);
463 printk(KERN_ERR "initrd extends beyond end of memory "
464 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
465 ramdisk_end, end_of_mem);
470 reserve_crashkernel();
472 reserve_ibft_region();
481 * Read APIC and some other early information from ACPI tables.
489 * get boot-time SMP configuration:
491 if (smp_found_config)
493 init_apic_mappings();
494 ioapic_init_mappings();
499 * We trust e820 completely. No explicit ROM probing in memory.
501 e820_reserve_resources();
502 e820_mark_nosave_regions();
504 /* request I/O space for devices used on all i[345]86 PCs */
505 for (i = 0; i < ARRAY_SIZE(standard_io_resources); i++)
506 request_resource(&ioport_resource, &standard_io_resources[i]);
511 #if defined(CONFIG_VGA_CONSOLE)
512 if (!efi_enabled || (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY))
513 conswitchp = &vga_con;
514 #elif defined(CONFIG_DUMMY_CONSOLE)
515 conswitchp = &dummy_con;
520 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
524 if (c->extended_cpuid_level < 0x80000004)
527 v = (unsigned int *) c->x86_model_id;
528 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
529 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
530 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
531 c->x86_model_id[48] = 0;
536 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
538 unsigned int n, dummy, eax, ebx, ecx, edx;
540 n = c->extended_cpuid_level;
542 if (n >= 0x80000005) {
543 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
544 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), "
545 "D cache %dK (%d bytes/line)\n",
546 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
547 c->x86_cache_size = (ecx>>24) + (edx>>24);
548 /* On K8 L1 TLB is inclusive, so don't count it */
552 if (n >= 0x80000006) {
553 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
554 ecx = cpuid_ecx(0x80000006);
555 c->x86_cache_size = ecx >> 16;
556 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
558 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
559 c->x86_cache_size, ecx & 0xFF);
561 if (n >= 0x80000008) {
562 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
563 c->x86_virt_bits = (eax >> 8) & 0xff;
564 c->x86_phys_bits = eax & 0xff;
569 static int __cpuinit nearby_node(int apicid)
573 for (i = apicid - 1; i >= 0; i--) {
574 node = apicid_to_node[i];
575 if (node != NUMA_NO_NODE && node_online(node))
578 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
579 node = apicid_to_node[i];
580 if (node != NUMA_NO_NODE && node_online(node))
583 return first_node(node_online_map); /* Shouldn't happen */
588 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
589 * Assumes number of cores is a power of two.
591 static void __cpuinit amd_detect_cmp(struct cpuinfo_x86 *c)
596 int cpu = smp_processor_id();
598 unsigned apicid = hard_smp_processor_id();
600 bits = c->x86_coreid_bits;
602 /* Low order bits define the core id (index of core in socket) */
603 c->cpu_core_id = c->initial_apicid & ((1 << bits)-1);
604 /* Convert the initial APIC ID into the socket ID */
605 c->phys_proc_id = c->initial_apicid >> bits;
608 node = c->phys_proc_id;
609 if (apicid_to_node[apicid] != NUMA_NO_NODE)
610 node = apicid_to_node[apicid];
611 if (!node_online(node)) {
612 /* Two possibilities here:
613 - The CPU is missing memory and no node was created.
614 In that case try picking one from a nearby CPU
615 - The APIC IDs differ from the HyperTransport node IDs
616 which the K8 northbridge parsing fills in.
617 Assume they are all increased by a constant offset,
618 but in the same order as the HT nodeids.
619 If that doesn't result in a usable node fall back to the
620 path for the previous case. */
622 int ht_nodeid = c->initial_apicid;
624 if (ht_nodeid >= 0 &&
625 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
626 node = apicid_to_node[ht_nodeid];
627 /* Pick a nearby node */
628 if (!node_online(node))
629 node = nearby_node(apicid);
631 numa_set_node(cpu, node);
633 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
638 static void __cpuinit early_init_amd_mc(struct cpuinfo_x86 *c)
643 /* Multi core CPU? */
644 if (c->extended_cpuid_level < 0x80000008)
647 ecx = cpuid_ecx(0x80000008);
649 c->x86_max_cores = (ecx & 0xff) + 1;
651 /* CPU telling us the core id bits shift? */
652 bits = (ecx >> 12) & 0xF;
654 /* Otherwise recompute */
656 while ((1 << bits) < c->x86_max_cores)
660 c->x86_coreid_bits = bits;
665 #define ENABLE_C1E_MASK 0x18000000
666 #define CPUID_PROCESSOR_SIGNATURE 1
667 #define CPUID_XFAM 0x0ff00000
668 #define CPUID_XFAM_K8 0x00000000
669 #define CPUID_XFAM_10H 0x00100000
670 #define CPUID_XFAM_11H 0x00200000
671 #define CPUID_XMOD 0x000f0000
672 #define CPUID_XMOD_REV_F 0x00040000
674 /* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
675 static __cpuinit int amd_apic_timer_broken(void)
677 u32 lo, hi, eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE);
679 switch (eax & CPUID_XFAM) {
681 if ((eax & CPUID_XMOD) < CPUID_XMOD_REV_F)
685 rdmsr(MSR_K8_ENABLE_C1E, lo, hi);
686 if (lo & ENABLE_C1E_MASK)
690 /* err on the side of caution */
696 static void __cpuinit early_init_amd(struct cpuinfo_x86 *c)
698 early_init_amd_mc(c);
700 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
701 if (c->x86_power & (1<<8))
702 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
705 static void __cpuinit init_amd(struct cpuinfo_x86 *c)
713 * Disable TLB flush filter by setting HWCR.FFDIS on K8
714 * bit 6 of msr C001_0015
716 * Errata 63 for SH-B3 steppings
717 * Errata 122 for all steppings (F+ have it disabled by default)
720 rdmsrl(MSR_K8_HWCR, value);
722 wrmsrl(MSR_K8_HWCR, value);
726 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
727 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
728 clear_cpu_cap(c, 0*32+31);
730 /* On C+ stepping K8 rep microcode works well for copy/memset */
731 level = cpuid_eax(1);
732 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) ||
734 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
735 if (c->x86 == 0x10 || c->x86 == 0x11)
736 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
738 /* Enable workaround for FXSAVE leak */
740 set_cpu_cap(c, X86_FEATURE_FXSAVE_LEAK);
742 level = get_model_name(c);
746 /* Should distinguish Models here, but this is only
747 a fallback anyways. */
748 strcpy(c->x86_model_id, "Hammer");
752 display_cacheinfo(c);
754 /* Multi core CPU? */
755 if (c->extended_cpuid_level >= 0x80000008)
758 if (c->extended_cpuid_level >= 0x80000006 &&
759 (cpuid_edx(0x80000006) & 0xf000))
760 num_cache_leaves = 4;
762 num_cache_leaves = 3;
764 if (c->x86 == 0xf || c->x86 == 0x10 || c->x86 == 0x11)
765 set_cpu_cap(c, X86_FEATURE_K8);
767 /* MFENCE stops RDTSC speculation */
768 set_cpu_cap(c, X86_FEATURE_MFENCE_RDTSC);
770 if (amd_apic_timer_broken())
771 disable_apic_timer = 1;
773 if (c == &boot_cpu_data && c->x86 >= 0xf && c->x86 <= 0x11) {
774 unsigned long long tseg;
777 * Split up direct mapping around the TSEG SMM area.
778 * Don't do it for gbpages because there seems very little
779 * benefit in doing so.
781 if (!rdmsrl_safe(MSR_K8_TSEG_ADDR, &tseg) &&
782 (tseg >> PMD_SHIFT) < (max_pfn_mapped >> (PMD_SHIFT-PAGE_SHIFT)))
783 set_memory_4k((unsigned long)__va(tseg), 1);
787 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
790 u32 eax, ebx, ecx, edx;
791 int index_msb, core_bits;
793 cpuid(1, &eax, &ebx, &ecx, &edx);
796 if (!cpu_has(c, X86_FEATURE_HT))
798 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
801 smp_num_siblings = (ebx & 0xff0000) >> 16;
803 if (smp_num_siblings == 1) {
804 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
805 } else if (smp_num_siblings > 1) {
807 if (smp_num_siblings > NR_CPUS) {
808 printk(KERN_WARNING "CPU: Unsupported number of "
809 "siblings %d", smp_num_siblings);
810 smp_num_siblings = 1;
814 index_msb = get_count_order(smp_num_siblings);
815 c->phys_proc_id = phys_pkg_id(index_msb);
817 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
819 index_msb = get_count_order(smp_num_siblings);
821 core_bits = get_count_order(c->x86_max_cores);
823 c->cpu_core_id = phys_pkg_id(index_msb) &
824 ((1 << core_bits) - 1);
827 if ((c->x86_max_cores * smp_num_siblings) > 1) {
828 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
830 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
838 * find out the number of processor cores on the die
840 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
844 if (c->cpuid_level < 4)
847 cpuid_count(4, 0, &eax, &t, &t, &t);
850 return ((eax >> 26) + 1);
855 static void __cpuinit srat_detect_node(void)
859 int cpu = smp_processor_id();
860 int apicid = hard_smp_processor_id();
862 /* Don't do the funky fallback heuristics the AMD version employs
864 node = apicid_to_node[apicid];
865 if (node == NUMA_NO_NODE || !node_online(node))
866 node = first_node(node_online_map);
867 numa_set_node(cpu, node);
869 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
873 static void __cpuinit early_init_intel(struct cpuinfo_x86 *c)
875 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
876 (c->x86 == 0x6 && c->x86_model >= 0x0e))
877 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
880 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
885 init_intel_cacheinfo(c);
886 if (c->cpuid_level > 9) {
887 unsigned eax = cpuid_eax(10);
888 /* Check for version and the number of counters */
889 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
890 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
895 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
897 set_cpu_cap(c, X86_FEATURE_BTS);
899 set_cpu_cap(c, X86_FEATURE_PEBS);
906 n = c->extended_cpuid_level;
907 if (n >= 0x80000008) {
908 unsigned eax = cpuid_eax(0x80000008);
909 c->x86_virt_bits = (eax >> 8) & 0xff;
910 c->x86_phys_bits = eax & 0xff;
911 /* CPUID workaround for Intel 0F34 CPU */
912 if (c->x86_vendor == X86_VENDOR_INTEL &&
913 c->x86 == 0xF && c->x86_model == 0x3 &&
915 c->x86_phys_bits = 36;
919 c->x86_cache_alignment = c->x86_clflush_size * 2;
921 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
922 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
923 c->x86_max_cores = intel_num_cpu_cores(c);
928 static void __cpuinit early_init_centaur(struct cpuinfo_x86 *c)
930 if (c->x86 == 0x6 && c->x86_model >= 0xf)
931 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
934 static void __cpuinit init_centaur(struct cpuinfo_x86 *c)
939 n = c->extended_cpuid_level;
940 if (n >= 0x80000008) {
941 unsigned eax = cpuid_eax(0x80000008);
942 c->x86_virt_bits = (eax >> 8) & 0xff;
943 c->x86_phys_bits = eax & 0xff;
946 if (c->x86 == 0x6 && c->x86_model >= 0xf) {
947 c->x86_cache_alignment = c->x86_clflush_size * 2;
948 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
949 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
951 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
954 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
956 char *v = c->x86_vendor_id;
958 if (!strcmp(v, "AuthenticAMD"))
959 c->x86_vendor = X86_VENDOR_AMD;
960 else if (!strcmp(v, "GenuineIntel"))
961 c->x86_vendor = X86_VENDOR_INTEL;
962 else if (!strcmp(v, "CentaurHauls"))
963 c->x86_vendor = X86_VENDOR_CENTAUR;
965 c->x86_vendor = X86_VENDOR_UNKNOWN;
968 /* Do some early cpuid on the boot CPU to get some parameter that are
969 needed before check_bugs. Everything advanced is in identify_cpu
971 static void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
975 c->loops_per_jiffy = loops_per_jiffy;
976 c->x86_cache_size = -1;
977 c->x86_vendor = X86_VENDOR_UNKNOWN;
978 c->x86_model = c->x86_mask = 0; /* So far unknown... */
979 c->x86_vendor_id[0] = '\0'; /* Unset */
980 c->x86_model_id[0] = '\0'; /* Unset */
981 c->x86_clflush_size = 64;
982 c->x86_cache_alignment = c->x86_clflush_size;
983 c->x86_max_cores = 1;
984 c->x86_coreid_bits = 0;
985 c->extended_cpuid_level = 0;
986 memset(&c->x86_capability, 0, sizeof c->x86_capability);
988 /* Get vendor name */
989 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
990 (unsigned int *)&c->x86_vendor_id[0],
991 (unsigned int *)&c->x86_vendor_id[8],
992 (unsigned int *)&c->x86_vendor_id[4]);
996 /* Initialize the standard set of capabilities */
997 /* Note that the vendor-specific code below might override */
999 /* Intel-defined flags: level 0x00000001 */
1000 if (c->cpuid_level >= 0x00000001) {
1002 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
1003 &c->x86_capability[0]);
1004 c->x86 = (tfms >> 8) & 0xf;
1005 c->x86_model = (tfms >> 4) & 0xf;
1006 c->x86_mask = tfms & 0xf;
1008 c->x86 += (tfms >> 20) & 0xff;
1010 c->x86_model += ((tfms >> 16) & 0xF) << 4;
1011 if (test_cpu_cap(c, X86_FEATURE_CLFLSH))
1012 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1014 /* Have CPUID level 0 only - unheard of */
1018 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xff;
1020 c->phys_proc_id = c->initial_apicid;
1022 /* AMD-defined flags: level 0x80000001 */
1023 xlvl = cpuid_eax(0x80000000);
1024 c->extended_cpuid_level = xlvl;
1025 if ((xlvl & 0xffff0000) == 0x80000000) {
1026 if (xlvl >= 0x80000001) {
1027 c->x86_capability[1] = cpuid_edx(0x80000001);
1028 c->x86_capability[6] = cpuid_ecx(0x80000001);
1030 if (xlvl >= 0x80000004)
1031 get_model_name(c); /* Default name */
1034 /* Transmeta-defined flags: level 0x80860001 */
1035 xlvl = cpuid_eax(0x80860000);
1036 if ((xlvl & 0xffff0000) == 0x80860000) {
1037 /* Don't set x86_cpuid_level here for now to not confuse. */
1038 if (xlvl >= 0x80860001)
1039 c->x86_capability[2] = cpuid_edx(0x80860001);
1042 c->extended_cpuid_level = cpuid_eax(0x80000000);
1043 if (c->extended_cpuid_level >= 0x80000007)
1044 c->x86_power = cpuid_edx(0x80000007);
1047 clear_cpu_cap(c, X86_FEATURE_PAT);
1049 switch (c->x86_vendor) {
1050 case X86_VENDOR_AMD:
1052 if (c->x86 >= 0xf && c->x86 <= 0x11)
1053 set_cpu_cap(c, X86_FEATURE_PAT);
1055 case X86_VENDOR_INTEL:
1056 early_init_intel(c);
1057 if (c->x86 == 0xF || (c->x86 == 6 && c->x86_model >= 15))
1058 set_cpu_cap(c, X86_FEATURE_PAT);
1060 case X86_VENDOR_CENTAUR:
1061 early_init_centaur(c);
1068 * This does the hard work of actually picking apart the CPU stuff...
1070 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1074 early_identify_cpu(c);
1076 init_scattered_cpuid_features(c);
1078 c->apicid = phys_pkg_id(0);
1081 * Vendor-specific initialization. In this section we
1082 * canonicalize the feature flags, meaning if there are
1083 * features a certain CPU supports which CPUID doesn't
1084 * tell us, CPUID claiming incorrect flags, or other bugs,
1085 * we handle them here.
1087 * At the end of this section, c->x86_capability better
1088 * indicate the features this CPU genuinely supports!
1090 switch (c->x86_vendor) {
1091 case X86_VENDOR_AMD:
1095 case X86_VENDOR_INTEL:
1099 case X86_VENDOR_CENTAUR:
1103 case X86_VENDOR_UNKNOWN:
1105 display_cacheinfo(c);
1112 * On SMP, boot_cpu_data holds the common feature set between
1113 * all CPUs; so make sure that we indicate which features are
1114 * common between the CPUs. The first time this routine gets
1115 * executed, c == &boot_cpu_data.
1117 if (c != &boot_cpu_data) {
1118 /* AND the already accumulated flags with these */
1119 for (i = 0; i < NCAPINTS; i++)
1120 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1123 /* Clear all flags overriden by options */
1124 for (i = 0; i < NCAPINTS; i++)
1125 c->x86_capability[i] &= ~cleared_cpu_caps[i];
1127 #ifdef CONFIG_X86_MCE
1130 select_idle_routine(c);
1133 numa_add_cpu(smp_processor_id());
1138 void __cpuinit identify_boot_cpu(void)
1140 identify_cpu(&boot_cpu_data);
1143 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
1145 BUG_ON(c == &boot_cpu_data);
1150 static __init int setup_noclflush(char *arg)
1152 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
1155 __setup("noclflush", setup_noclflush);
1157 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1159 if (c->x86_model_id[0])
1160 printk(KERN_CONT "%s", c->x86_model_id);
1162 if (c->x86_mask || c->cpuid_level >= 0)
1163 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
1165 printk(KERN_CONT "\n");
1168 static __init int setup_disablecpuid(char *arg)
1171 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1172 setup_clear_cpu_cap(bit);
1177 __setup("clearcpuid=", setup_disablecpuid);