2 * linux/arch/x86-64/kernel/setup.c
4 * Copyright (C) 1995 Linus Torvalds
6 * Nov 2001 Dave Jones <davej@suse.de>
7 * Forked from i386 setup code.
13 * This file handles the architecture-dependent parts of initialization
16 #include <linux/errno.h>
17 #include <linux/sched.h>
18 #include <linux/kernel.h>
20 #include <linux/stddef.h>
21 #include <linux/unistd.h>
22 #include <linux/ptrace.h>
23 #include <linux/slab.h>
24 #include <linux/user.h>
25 #include <linux/a.out.h>
26 #include <linux/tty.h>
27 #include <linux/ioport.h>
28 #include <linux/delay.h>
29 #include <linux/config.h>
30 #include <linux/init.h>
31 #include <linux/initrd.h>
32 #include <linux/highmem.h>
33 #include <linux/bootmem.h>
34 #include <linux/module.h>
35 #include <asm/processor.h>
36 #include <linux/console.h>
37 #include <linux/seq_file.h>
38 #include <linux/crash_dump.h>
39 #include <linux/root_dev.h>
40 #include <linux/pci.h>
41 #include <linux/acpi.h>
42 #include <linux/kallsyms.h>
43 #include <linux/edd.h>
44 #include <linux/mmzone.h>
45 #include <linux/kexec.h>
46 #include <linux/cpufreq.h>
47 #include <linux/dmi.h>
48 #include <linux/dma-mapping.h>
51 #include <asm/uaccess.h>
52 #include <asm/system.h>
57 #include <video/edid.h>
60 #include <asm/mpspec.h>
61 #include <asm/mmu_context.h>
62 #include <asm/bootsetup.h>
63 #include <asm/proto.h>
64 #include <asm/setup.h>
65 #include <asm/mach_apic.h>
67 #include <asm/swiotlb.h>
68 #include <asm/sections.h>
69 #include <asm/gart-mapping.h>
75 struct cpuinfo_x86 boot_cpu_data __read_mostly;
77 unsigned long mmu_cr4_features;
80 EXPORT_SYMBOL(acpi_disabled);
82 extern int __initdata acpi_ht;
83 extern acpi_interrupt_flags acpi_sci_flags;
84 int __initdata acpi_force = 0;
87 int acpi_numa __initdata;
89 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
92 unsigned long saved_video_mode;
97 struct screen_info screen_info;
98 struct sys_desc_table_struct {
99 unsigned short length;
100 unsigned char table[0];
103 struct edid_info edid_info;
106 extern int root_mountflags;
108 char command_line[COMMAND_LINE_SIZE];
110 struct resource standard_io_resources[] = {
111 { .name = "dma1", .start = 0x00, .end = 0x1f,
112 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
113 { .name = "pic1", .start = 0x20, .end = 0x21,
114 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
115 { .name = "timer0", .start = 0x40, .end = 0x43,
116 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
117 { .name = "timer1", .start = 0x50, .end = 0x53,
118 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
119 { .name = "keyboard", .start = 0x60, .end = 0x6f,
120 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
121 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
122 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
123 { .name = "pic2", .start = 0xa0, .end = 0xa1,
124 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
125 { .name = "dma2", .start = 0xc0, .end = 0xdf,
126 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
127 { .name = "fpu", .start = 0xf0, .end = 0xff,
128 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
131 #define STANDARD_IO_RESOURCES \
132 (sizeof standard_io_resources / sizeof standard_io_resources[0])
134 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
136 struct resource data_resource = {
137 .name = "Kernel data",
140 .flags = IORESOURCE_RAM,
142 struct resource code_resource = {
143 .name = "Kernel code",
146 .flags = IORESOURCE_RAM,
149 #define IORESOURCE_ROM (IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM)
151 static struct resource system_rom_resource = {
152 .name = "System ROM",
155 .flags = IORESOURCE_ROM,
158 static struct resource extension_rom_resource = {
159 .name = "Extension ROM",
162 .flags = IORESOURCE_ROM,
165 static struct resource adapter_rom_resources[] = {
166 { .name = "Adapter ROM", .start = 0xc8000, .end = 0,
167 .flags = IORESOURCE_ROM },
168 { .name = "Adapter ROM", .start = 0, .end = 0,
169 .flags = IORESOURCE_ROM },
170 { .name = "Adapter ROM", .start = 0, .end = 0,
171 .flags = IORESOURCE_ROM },
172 { .name = "Adapter ROM", .start = 0, .end = 0,
173 .flags = IORESOURCE_ROM },
174 { .name = "Adapter ROM", .start = 0, .end = 0,
175 .flags = IORESOURCE_ROM },
176 { .name = "Adapter ROM", .start = 0, .end = 0,
177 .flags = IORESOURCE_ROM }
180 #define ADAPTER_ROM_RESOURCES \
181 (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
183 static struct resource video_rom_resource = {
187 .flags = IORESOURCE_ROM,
190 static struct resource video_ram_resource = {
191 .name = "Video RAM area",
194 .flags = IORESOURCE_RAM,
197 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
199 static int __init romchecksum(unsigned char *rom, unsigned long length)
201 unsigned char *p, sum = 0;
203 for (p = rom; p < rom + length; p++)
208 static void __init probe_roms(void)
210 unsigned long start, length, upper;
215 upper = adapter_rom_resources[0].start;
216 for (start = video_rom_resource.start; start < upper; start += 2048) {
217 rom = isa_bus_to_virt(start);
218 if (!romsignature(rom))
221 video_rom_resource.start = start;
223 /* 0 < length <= 0x7f * 512, historically */
224 length = rom[2] * 512;
226 /* if checksum okay, trust length byte */
227 if (length && romchecksum(rom, length))
228 video_rom_resource.end = start + length - 1;
230 request_resource(&iomem_resource, &video_rom_resource);
234 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
239 request_resource(&iomem_resource, &system_rom_resource);
240 upper = system_rom_resource.start;
242 /* check for extension rom (ignore length byte!) */
243 rom = isa_bus_to_virt(extension_rom_resource.start);
244 if (romsignature(rom)) {
245 length = extension_rom_resource.end - extension_rom_resource.start + 1;
246 if (romchecksum(rom, length)) {
247 request_resource(&iomem_resource, &extension_rom_resource);
248 upper = extension_rom_resource.start;
252 /* check for adapter roms on 2k boundaries */
253 for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) {
254 rom = isa_bus_to_virt(start);
255 if (!romsignature(rom))
258 /* 0 < length <= 0x7f * 512, historically */
259 length = rom[2] * 512;
261 /* but accept any length that fits if checksum okay */
262 if (!length || start + length > upper || !romchecksum(rom, length))
265 adapter_rom_resources[i].start = start;
266 adapter_rom_resources[i].end = start + length - 1;
267 request_resource(&iomem_resource, &adapter_rom_resources[i]);
269 start = adapter_rom_resources[i++].end & ~2047UL;
273 static __init void parse_cmdline_early (char ** cmdline_p)
275 char c = ' ', *to = command_line, *from = COMMAND_LINE;
285 * If the BIOS enumerates physical processors before logical,
286 * maxcpus=N at enumeration-time can be used to disable HT.
288 else if (!memcmp(from, "maxcpus=", 8)) {
289 extern unsigned int maxcpus;
291 maxcpus = simple_strtoul(from + 8, NULL, 0);
295 /* "acpi=off" disables both ACPI table parsing and interpreter init */
296 if (!memcmp(from, "acpi=off", 8))
299 if (!memcmp(from, "acpi=force", 10)) {
300 /* add later when we do DMI horrors: */
305 /* acpi=ht just means: do ACPI MADT parsing
306 at bootup, but don't enable the full ACPI interpreter */
307 if (!memcmp(from, "acpi=ht", 7)) {
312 else if (!memcmp(from, "pci=noacpi", 10))
314 else if (!memcmp(from, "acpi=noirq", 10))
317 else if (!memcmp(from, "acpi_sci=edge", 13))
318 acpi_sci_flags.trigger = 1;
319 else if (!memcmp(from, "acpi_sci=level", 14))
320 acpi_sci_flags.trigger = 3;
321 else if (!memcmp(from, "acpi_sci=high", 13))
322 acpi_sci_flags.polarity = 1;
323 else if (!memcmp(from, "acpi_sci=low", 12))
324 acpi_sci_flags.polarity = 3;
326 /* acpi=strict disables out-of-spec workarounds */
327 else if (!memcmp(from, "acpi=strict", 11)) {
330 #ifdef CONFIG_X86_IO_APIC
331 else if (!memcmp(from, "acpi_skip_timer_override", 24))
332 acpi_skip_timer_override = 1;
336 if (!memcmp(from, "disable_timer_pin_1", 19))
337 disable_timer_pin_1 = 1;
338 if (!memcmp(from, "enable_timer_pin_1", 18))
339 disable_timer_pin_1 = -1;
341 if (!memcmp(from, "nolapic", 7) ||
342 !memcmp(from, "disableapic", 11))
345 /* Don't confuse with noapictimer */
346 if (!memcmp(from, "noapic", 6) &&
347 (from[6] == ' ' || from[6] == 0))
348 skip_ioapic_setup = 1;
350 /* Make sure to not confuse with apic= */
351 if (!memcmp(from, "apic", 4) &&
352 (from[4] == ' ' || from[4] == 0)) {
353 skip_ioapic_setup = 0;
357 if (!memcmp(from, "mem=", 4))
358 parse_memopt(from+4, &from);
360 if (!memcmp(from, "memmap=", 7)) {
361 /* exactmap option is for used defined memory */
362 if (!memcmp(from+7, "exactmap", 8)) {
363 #ifdef CONFIG_CRASH_DUMP
364 /* If we are doing a crash dump, we
365 * still need to know the real mem
366 * size before original memory map is
369 saved_max_pfn = e820_end_of_ram();
377 parse_memmapopt(from+7, &from);
383 if (!memcmp(from, "numa=", 5))
387 if (!memcmp(from,"iommu=",6)) {
391 if (!memcmp(from,"oops=panic", 10))
394 if (!memcmp(from, "noexec=", 7))
395 nonx_setup(from + 7);
398 /* crashkernel=size@addr specifies the location to reserve for
399 * a crash kernel. By reserving this memory we guarantee
400 * that linux never set's it up as a DMA target.
401 * Useful for holding code to do something appropriate
402 * after a kernel panic.
404 else if (!memcmp(from, "crashkernel=", 12)) {
405 unsigned long size, base;
406 size = memparse(from+12, &from);
408 base = memparse(from+1, &from);
409 /* FIXME: Do I want a sanity check
410 * to validate the memory range?
412 crashk_res.start = base;
413 crashk_res.end = base + size - 1;
418 #ifdef CONFIG_PROC_VMCORE
419 /* elfcorehdr= specifies the location of elf core header
420 * stored by the crashed kernel. This option will be passed
421 * by kexec loader to the capture kernel.
423 else if(!memcmp(from, "elfcorehdr=", 11))
424 elfcorehdr_addr = memparse(from+11, &from);
430 if (COMMAND_LINE_SIZE <= ++len)
435 printk(KERN_INFO "user-defined physical RAM map:\n");
436 e820_print_map("user");
439 *cmdline_p = command_line;
444 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
446 unsigned long bootmap_size, bootmap;
448 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
449 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size);
451 panic("Cannot find bootmem map of size %ld\n",bootmap_size);
452 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
453 e820_bootmem_free(NODE_DATA(0), 0, end_pfn << PAGE_SHIFT);
454 reserve_bootmem(bootmap, bootmap_size);
458 /* Use inline assembly to define this because the nops are defined
459 as inline assembly strings in the include files and we cannot
460 get them easily into strings. */
461 asm("\t.data\nk8nops: "
462 K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6
465 extern unsigned char k8nops[];
466 static unsigned char *k8_nops[ASM_NOP_MAX+1] = {
472 k8nops + 1 + 2 + 3 + 4,
473 k8nops + 1 + 2 + 3 + 4 + 5,
474 k8nops + 1 + 2 + 3 + 4 + 5 + 6,
475 k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
478 extern char __vsyscall_0;
480 /* Replace instructions with better alternatives for this CPU type.
482 This runs before SMP is initialized to avoid SMP problems with
483 self modifying code. This implies that assymetric systems where
484 APs have less capabilities than the boot processor are not handled.
485 In this case boot with "noreplacement". */
486 void apply_alternatives(void *start, void *end)
490 for (a = start; (void *)a < end; a++) {
493 if (!boot_cpu_has(a->cpuid))
496 BUG_ON(a->replacementlen > a->instrlen);
498 /* vsyscall code is not mapped yet. resolve it manually. */
499 if (instr >= (u8 *)VSYSCALL_START && instr < (u8*)VSYSCALL_END)
500 instr = __va(instr - (u8*)VSYSCALL_START + (u8*)__pa_symbol(&__vsyscall_0));
501 __inline_memcpy(instr, a->replacement, a->replacementlen);
502 diff = a->instrlen - a->replacementlen;
504 /* Pad the rest with nops */
505 for (i = a->replacementlen; diff > 0; diff -= k, i += k) {
509 __inline_memcpy(instr + i, k8_nops[k], k);
514 static int no_replacement __initdata = 0;
516 void __init alternative_instructions(void)
518 extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
521 apply_alternatives(__alt_instructions, __alt_instructions_end);
524 static int __init noreplacement_setup(char *s)
530 __setup("noreplacement", noreplacement_setup);
532 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
534 #ifdef CONFIG_EDD_MODULE
538 * copy_edd() - Copy the BIOS EDD information
539 * from boot_params into a safe place.
542 static inline void copy_edd(void)
544 memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
545 memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
546 edd.mbr_signature_nr = EDD_MBR_SIG_NR;
547 edd.edd_info_nr = EDD_NR;
550 static inline void copy_edd(void)
555 #define EBDA_ADDR_POINTER 0x40E
556 static void __init reserve_ebda_region(void)
560 * there is a real-mode segmented pointer pointing to the
561 * 4K EBDA area at 0x40E
563 addr = *(unsigned short *)phys_to_virt(EBDA_ADDR_POINTER);
566 reserve_bootmem_generic(addr, PAGE_SIZE);
569 void __init setup_arch(char **cmdline_p)
571 unsigned long kernel_end;
573 ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
574 screen_info = SCREEN_INFO;
575 edid_info = EDID_INFO;
576 saved_video_mode = SAVED_VIDEO_MODE;
577 bootloader_type = LOADER_TYPE;
579 #ifdef CONFIG_BLK_DEV_RAM
580 rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
581 rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
582 rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
584 setup_memory_region();
587 if (!MOUNT_ROOT_RDONLY)
588 root_mountflags &= ~MS_RDONLY;
589 init_mm.start_code = (unsigned long) &_text;
590 init_mm.end_code = (unsigned long) &_etext;
591 init_mm.end_data = (unsigned long) &_edata;
592 init_mm.brk = (unsigned long) &_end;
594 code_resource.start = virt_to_phys(&_text);
595 code_resource.end = virt_to_phys(&_etext)-1;
596 data_resource.start = virt_to_phys(&_etext);
597 data_resource.end = virt_to_phys(&_edata)-1;
599 parse_cmdline_early(cmdline_p);
601 early_identify_cpu(&boot_cpu_data);
604 * partially used pages are not usable - thus
605 * we are rounding upwards:
607 end_pfn = e820_end_of_ram();
611 init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
617 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
618 * Call this early for SRAT node setup.
620 acpi_boot_table_init();
623 #ifdef CONFIG_ACPI_NUMA
625 * Parse SRAT to discover nodes.
631 numa_initmem_init(0, end_pfn);
633 contig_initmem_init(0, end_pfn);
636 /* Reserve direct mapping */
637 reserve_bootmem_generic(table_start << PAGE_SHIFT,
638 (table_end - table_start) << PAGE_SHIFT);
641 kernel_end = round_up(__pa_symbol(&_end),PAGE_SIZE);
642 reserve_bootmem_generic(HIGH_MEMORY, kernel_end - HIGH_MEMORY);
645 * reserve physical page 0 - it's a special BIOS page on many boxes,
646 * enabling clean reboots, SMP operation, laptop functions.
648 reserve_bootmem_generic(0, PAGE_SIZE);
650 /* reserve ebda region */
651 reserve_ebda_region();
655 * But first pinch a few for the stack/trampoline stuff
656 * FIXME: Don't need the extra page at 4K, but need to fix
657 * trampoline before removing it. (see the GDT stuff)
659 reserve_bootmem_generic(PAGE_SIZE, PAGE_SIZE);
661 /* Reserve SMP trampoline */
662 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, PAGE_SIZE);
665 #ifdef CONFIG_ACPI_SLEEP
667 * Reserve low memory region for sleep support.
669 acpi_reserve_bootmem();
671 #ifdef CONFIG_X86_LOCAL_APIC
673 * Find and reserve possible boot-time SMP configuration:
677 #ifdef CONFIG_BLK_DEV_INITRD
678 if (LOADER_TYPE && INITRD_START) {
679 if (INITRD_START + INITRD_SIZE <= (end_pfn << PAGE_SHIFT)) {
680 reserve_bootmem_generic(INITRD_START, INITRD_SIZE);
682 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
683 initrd_end = initrd_start+INITRD_SIZE;
686 printk(KERN_ERR "initrd extends beyond end of memory "
687 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
688 (unsigned long)(INITRD_START + INITRD_SIZE),
689 (unsigned long)(end_pfn << PAGE_SHIFT));
695 if (crashk_res.start != crashk_res.end) {
696 reserve_bootmem(crashk_res.start,
697 crashk_res.end - crashk_res.start + 1);
707 * Read APIC and some other early information from ACPI tables.
714 #ifdef CONFIG_X86_LOCAL_APIC
716 * get boot-time SMP configuration:
718 if (smp_found_config)
720 init_apic_mappings();
724 * Request address space for all standard RAM and ROM resources
725 * and also for regions reported as reserved by the e820.
728 e820_reserve_resources();
730 request_resource(&iomem_resource, &video_ram_resource);
734 /* request I/O space for devices used on all i[345]86 PCs */
735 for (i = 0; i < STANDARD_IO_RESOURCES; i++)
736 request_resource(&ioport_resource, &standard_io_resources[i]);
741 #ifdef CONFIG_GART_IOMMU
746 #if defined(CONFIG_VGA_CONSOLE)
747 conswitchp = &vga_con;
748 #elif defined(CONFIG_DUMMY_CONSOLE)
749 conswitchp = &dummy_con;
754 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
758 if (c->extended_cpuid_level < 0x80000004)
761 v = (unsigned int *) c->x86_model_id;
762 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
763 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
764 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
765 c->x86_model_id[48] = 0;
770 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
772 unsigned int n, dummy, eax, ebx, ecx, edx;
774 n = c->extended_cpuid_level;
776 if (n >= 0x80000005) {
777 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
778 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
779 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
780 c->x86_cache_size=(ecx>>24)+(edx>>24);
781 /* On K8 L1 TLB is inclusive, so don't count it */
785 if (n >= 0x80000006) {
786 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
787 ecx = cpuid_ecx(0x80000006);
788 c->x86_cache_size = ecx >> 16;
789 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
791 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
792 c->x86_cache_size, ecx & 0xFF);
796 cpuid(0x80000007, &dummy, &dummy, &dummy, &c->x86_power);
797 if (n >= 0x80000008) {
798 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
799 c->x86_virt_bits = (eax >> 8) & 0xff;
800 c->x86_phys_bits = eax & 0xff;
805 static int nearby_node(int apicid)
808 for (i = apicid - 1; i >= 0; i--) {
809 int node = apicid_to_node[i];
810 if (node != NUMA_NO_NODE && node_online(node))
813 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
814 int node = apicid_to_node[i];
815 if (node != NUMA_NO_NODE && node_online(node))
818 return first_node(node_online_map); /* Shouldn't happen */
823 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
824 * Assumes number of cores is a power of two.
826 static void __init amd_detect_cmp(struct cpuinfo_x86 *c)
829 int cpu = smp_processor_id();
833 unsigned apicid = phys_proc_id[cpu];
837 while ((1 << bits) < c->x86_max_cores)
840 /* Low order bits define the core id (index of core in socket) */
841 cpu_core_id[cpu] = phys_proc_id[cpu] & ((1 << bits)-1);
842 /* Convert the APIC ID into the socket ID */
843 phys_proc_id[cpu] >>= bits;
846 node = phys_proc_id[cpu];
847 if (apicid_to_node[apicid] != NUMA_NO_NODE)
848 node = apicid_to_node[apicid];
849 if (!node_online(node)) {
850 /* Two possibilities here:
851 - The CPU is missing memory and no node was created.
852 In that case try picking one from a nearby CPU
853 - The APIC IDs differ from the HyperTransport node IDs
854 which the K8 northbridge parsing fills in.
855 Assume they are all increased by a constant offset,
856 but in the same order as the HT nodeids.
857 If that doesn't result in a usable node fall back to the
858 path for the previous case. */
859 int ht_nodeid = apicid - (phys_proc_id[0] << bits);
860 if (ht_nodeid >= 0 &&
861 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
862 node = apicid_to_node[ht_nodeid];
863 /* Pick a nearby node */
864 if (!node_online(node))
865 node = nearby_node(apicid);
867 numa_set_node(cpu, node);
869 printk(KERN_INFO "CPU %d(%d) -> Node %d -> Core %d\n",
870 cpu, c->x86_max_cores, node, cpu_core_id[cpu]);
875 static int __init init_amd(struct cpuinfo_x86 *c)
884 * Disable TLB flush filter by setting HWCR.FFDIS on K8
885 * bit 6 of msr C001_0015
887 * Errata 63 for SH-B3 steppings
888 * Errata 122 for all steppings (F+ have it disabled by default)
891 rdmsrl(MSR_K8_HWCR, value);
893 wrmsrl(MSR_K8_HWCR, value);
897 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
898 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
899 clear_bit(0*32+31, &c->x86_capability);
901 /* On C+ stepping K8 rep microcode works well for copy/memset */
902 level = cpuid_eax(1);
903 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58))
904 set_bit(X86_FEATURE_REP_GOOD, &c->x86_capability);
906 r = get_model_name(c);
910 /* Should distinguish Models here, but this is only
911 a fallback anyways. */
912 strcpy(c->x86_model_id, "Hammer");
916 display_cacheinfo(c);
918 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
919 if (c->x86_power & (1<<8))
920 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
922 if (c->extended_cpuid_level >= 0x80000008) {
923 c->x86_max_cores = (cpuid_ecx(0x80000008) & 0xff) + 1;
924 if (c->x86_max_cores & (c->x86_max_cores - 1))
925 c->x86_max_cores = 1;
933 static void __cpuinit detect_ht(struct cpuinfo_x86 *c)
936 u32 eax, ebx, ecx, edx;
937 int index_msb, core_bits;
938 int cpu = smp_processor_id();
940 cpuid(1, &eax, &ebx, &ecx, &edx);
942 c->apicid = phys_pkg_id(0);
944 if (!cpu_has(c, X86_FEATURE_HT) || cpu_has(c, X86_FEATURE_CMP_LEGACY))
947 smp_num_siblings = (ebx & 0xff0000) >> 16;
949 if (smp_num_siblings == 1) {
950 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
951 } else if (smp_num_siblings > 1 ) {
953 if (smp_num_siblings > NR_CPUS) {
954 printk(KERN_WARNING "CPU: Unsupported number of the siblings %d", smp_num_siblings);
955 smp_num_siblings = 1;
959 index_msb = get_count_order(smp_num_siblings);
960 phys_proc_id[cpu] = phys_pkg_id(index_msb);
962 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
965 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
967 index_msb = get_count_order(smp_num_siblings) ;
969 core_bits = get_count_order(c->x86_max_cores);
971 cpu_core_id[cpu] = phys_pkg_id(index_msb) &
972 ((1 << core_bits) - 1);
974 if (c->x86_max_cores > 1)
975 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
982 * find out the number of processor cores on the die
984 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
988 if (c->cpuid_level < 4)
997 return ((eax >> 26) + 1);
1002 static void srat_detect_node(void)
1006 int cpu = smp_processor_id();
1008 /* Don't do the funky fallback heuristics the AMD version employs
1010 node = apicid_to_node[hard_smp_processor_id()];
1011 if (node == NUMA_NO_NODE)
1013 numa_set_node(cpu, node);
1016 printk(KERN_INFO "CPU %d -> Node %d\n", cpu, node);
1020 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
1025 init_intel_cacheinfo(c);
1026 n = c->extended_cpuid_level;
1027 if (n >= 0x80000008) {
1028 unsigned eax = cpuid_eax(0x80000008);
1029 c->x86_virt_bits = (eax >> 8) & 0xff;
1030 c->x86_phys_bits = eax & 0xff;
1031 /* CPUID workaround for Intel 0F34 CPU */
1032 if (c->x86_vendor == X86_VENDOR_INTEL &&
1033 c->x86 == 0xF && c->x86_model == 0x3 &&
1035 c->x86_phys_bits = 36;
1039 c->x86_cache_alignment = c->x86_clflush_size * 2;
1040 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
1041 (c->x86 == 0x6 && c->x86_model >= 0x0e))
1042 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
1043 set_bit(X86_FEATURE_SYNC_RDTSC, &c->x86_capability);
1044 c->x86_max_cores = intel_num_cpu_cores(c);
1049 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
1051 char *v = c->x86_vendor_id;
1053 if (!strcmp(v, "AuthenticAMD"))
1054 c->x86_vendor = X86_VENDOR_AMD;
1055 else if (!strcmp(v, "GenuineIntel"))
1056 c->x86_vendor = X86_VENDOR_INTEL;
1058 c->x86_vendor = X86_VENDOR_UNKNOWN;
1061 struct cpu_model_info {
1064 char *model_names[16];
1067 /* Do some early cpuid on the boot CPU to get some parameter that are
1068 needed before check_bugs. Everything advanced is in identify_cpu
1070 void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
1074 c->loops_per_jiffy = loops_per_jiffy;
1075 c->x86_cache_size = -1;
1076 c->x86_vendor = X86_VENDOR_UNKNOWN;
1077 c->x86_model = c->x86_mask = 0; /* So far unknown... */
1078 c->x86_vendor_id[0] = '\0'; /* Unset */
1079 c->x86_model_id[0] = '\0'; /* Unset */
1080 c->x86_clflush_size = 64;
1081 c->x86_cache_alignment = c->x86_clflush_size;
1082 c->x86_max_cores = 1;
1083 c->extended_cpuid_level = 0;
1084 memset(&c->x86_capability, 0, sizeof c->x86_capability);
1086 /* Get vendor name */
1087 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
1088 (unsigned int *)&c->x86_vendor_id[0],
1089 (unsigned int *)&c->x86_vendor_id[8],
1090 (unsigned int *)&c->x86_vendor_id[4]);
1094 /* Initialize the standard set of capabilities */
1095 /* Note that the vendor-specific code below might override */
1097 /* Intel-defined flags: level 0x00000001 */
1098 if (c->cpuid_level >= 0x00000001) {
1100 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
1101 &c->x86_capability[0]);
1102 c->x86 = (tfms >> 8) & 0xf;
1103 c->x86_model = (tfms >> 4) & 0xf;
1104 c->x86_mask = tfms & 0xf;
1106 c->x86 += (tfms >> 20) & 0xff;
1108 c->x86_model += ((tfms >> 16) & 0xF) << 4;
1109 if (c->x86_capability[0] & (1<<19))
1110 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1112 /* Have CPUID level 0 only - unheard of */
1117 phys_proc_id[smp_processor_id()] = (cpuid_ebx(1) >> 24) & 0xff;
1122 * This does the hard work of actually picking apart the CPU stuff...
1124 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1129 early_identify_cpu(c);
1131 /* AMD-defined flags: level 0x80000001 */
1132 xlvl = cpuid_eax(0x80000000);
1133 c->extended_cpuid_level = xlvl;
1134 if ((xlvl & 0xffff0000) == 0x80000000) {
1135 if (xlvl >= 0x80000001) {
1136 c->x86_capability[1] = cpuid_edx(0x80000001);
1137 c->x86_capability[6] = cpuid_ecx(0x80000001);
1139 if (xlvl >= 0x80000004)
1140 get_model_name(c); /* Default name */
1143 /* Transmeta-defined flags: level 0x80860001 */
1144 xlvl = cpuid_eax(0x80860000);
1145 if ((xlvl & 0xffff0000) == 0x80860000) {
1146 /* Don't set x86_cpuid_level here for now to not confuse. */
1147 if (xlvl >= 0x80860001)
1148 c->x86_capability[2] = cpuid_edx(0x80860001);
1152 * Vendor-specific initialization. In this section we
1153 * canonicalize the feature flags, meaning if there are
1154 * features a certain CPU supports which CPUID doesn't
1155 * tell us, CPUID claiming incorrect flags, or other bugs,
1156 * we handle them here.
1158 * At the end of this section, c->x86_capability better
1159 * indicate the features this CPU genuinely supports!
1161 switch (c->x86_vendor) {
1162 case X86_VENDOR_AMD:
1166 case X86_VENDOR_INTEL:
1170 case X86_VENDOR_UNKNOWN:
1172 display_cacheinfo(c);
1176 select_idle_routine(c);
1180 * On SMP, boot_cpu_data holds the common feature set between
1181 * all CPUs; so make sure that we indicate which features are
1182 * common between the CPUs. The first time this routine gets
1183 * executed, c == &boot_cpu_data.
1185 if (c != &boot_cpu_data) {
1186 /* AND the already accumulated flags with these */
1187 for (i = 0 ; i < NCAPINTS ; i++)
1188 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1191 #ifdef CONFIG_X86_MCE
1194 if (c == &boot_cpu_data)
1199 numa_add_cpu(smp_processor_id());
1204 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1206 if (c->x86_model_id[0])
1207 printk("%s", c->x86_model_id);
1209 if (c->x86_mask || c->cpuid_level >= 0)
1210 printk(" stepping %02x\n", c->x86_mask);
1216 * Get CPU information for use by the procfs.
1219 static int show_cpuinfo(struct seq_file *m, void *v)
1221 struct cpuinfo_x86 *c = v;
1224 * These flag bits must match the definitions in <asm/cpufeature.h>.
1225 * NULL means this bit is undefined or reserved; either way it doesn't
1226 * have meaning as far as Linux is concerned. Note that it's important
1227 * to realize there is a difference between this table and CPUID -- if
1228 * applications want to get the raw CPUID data, they should access
1229 * /dev/cpu/<cpu_nr>/cpuid instead.
1231 static char *x86_cap_flags[] = {
1233 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1234 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
1235 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
1236 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", NULL,
1239 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1240 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
1241 NULL, NULL, NULL, NULL, "nx", NULL, "mmxext", NULL,
1242 NULL, "fxsr_opt", "rdtscp", NULL, NULL, "lm", "3dnowext", "3dnow",
1244 /* Transmeta-defined */
1245 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
1246 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1247 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1248 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1250 /* Other (Linux-defined) */
1251 "cxmmx", NULL, "cyrix_arr", "centaur_mcr", NULL,
1252 "constant_tsc", NULL, NULL,
1253 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1254 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1255 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1257 /* Intel-defined (#2) */
1258 "pni", NULL, NULL, "monitor", "ds_cpl", "vmx", NULL, "est",
1259 "tm2", NULL, "cid", NULL, NULL, "cx16", "xtpr", NULL,
1260 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1261 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1263 /* VIA/Cyrix/Centaur-defined */
1264 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
1265 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1266 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1267 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1269 /* AMD-defined (#2) */
1270 "lahf_lm", "cmp_legacy", "svm", NULL, "cr8_legacy", NULL, NULL, NULL,
1271 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1272 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1273 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1275 static char *x86_power_flags[] = {
1276 "ts", /* temperature sensor */
1277 "fid", /* frequency id control */
1278 "vid", /* voltage id control */
1279 "ttp", /* thermal trip */
1283 /* nothing */ /* constant_tsc - moved to flags */
1288 if (!cpu_online(c-cpu_data))
1292 seq_printf(m,"processor\t: %u\n"
1294 "cpu family\t: %d\n"
1296 "model name\t: %s\n",
1297 (unsigned)(c-cpu_data),
1298 c->x86_vendor_id[0] ? c->x86_vendor_id : "unknown",
1301 c->x86_model_id[0] ? c->x86_model_id : "unknown");
1303 if (c->x86_mask || c->cpuid_level >= 0)
1304 seq_printf(m, "stepping\t: %d\n", c->x86_mask);
1306 seq_printf(m, "stepping\t: unknown\n");
1308 if (cpu_has(c,X86_FEATURE_TSC)) {
1309 unsigned int freq = cpufreq_quick_get((unsigned)(c-cpu_data));
1312 seq_printf(m, "cpu MHz\t\t: %u.%03u\n",
1313 freq / 1000, (freq % 1000));
1317 if (c->x86_cache_size >= 0)
1318 seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
1321 if (smp_num_siblings * c->x86_max_cores > 1) {
1322 int cpu = c - cpu_data;
1323 seq_printf(m, "physical id\t: %d\n", phys_proc_id[cpu]);
1324 seq_printf(m, "siblings\t: %d\n", cpus_weight(cpu_core_map[cpu]));
1325 seq_printf(m, "core id\t\t: %d\n", cpu_core_id[cpu]);
1326 seq_printf(m, "cpu cores\t: %d\n", c->booted_cores);
1332 "fpu_exception\t: yes\n"
1333 "cpuid level\t: %d\n"
1340 for ( i = 0 ; i < 32*NCAPINTS ; i++ )
1341 if ( test_bit(i, &c->x86_capability) &&
1342 x86_cap_flags[i] != NULL )
1343 seq_printf(m, " %s", x86_cap_flags[i]);
1346 seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
1347 c->loops_per_jiffy/(500000/HZ),
1348 (c->loops_per_jiffy/(5000/HZ)) % 100);
1350 if (c->x86_tlbsize > 0)
1351 seq_printf(m, "TLB size\t: %d 4K pages\n", c->x86_tlbsize);
1352 seq_printf(m, "clflush size\t: %d\n", c->x86_clflush_size);
1353 seq_printf(m, "cache_alignment\t: %d\n", c->x86_cache_alignment);
1355 seq_printf(m, "address sizes\t: %u bits physical, %u bits virtual\n",
1356 c->x86_phys_bits, c->x86_virt_bits);
1358 seq_printf(m, "power management:");
1361 for (i = 0; i < 32; i++)
1362 if (c->x86_power & (1 << i)) {
1363 if (i < ARRAY_SIZE(x86_power_flags) &&
1365 seq_printf(m, "%s%s",
1366 x86_power_flags[i][0]?" ":"",
1367 x86_power_flags[i]);
1369 seq_printf(m, " [%d]", i);
1373 seq_printf(m, "\n\n");
1378 static void *c_start(struct seq_file *m, loff_t *pos)
1380 return *pos < NR_CPUS ? cpu_data + *pos : NULL;
1383 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
1386 return c_start(m, pos);
1389 static void c_stop(struct seq_file *m, void *v)
1393 struct seq_operations cpuinfo_op = {
1397 .show = show_cpuinfo,
1400 static int __init run_dmi_scan(void)
1405 core_initcall(run_dmi_scan);