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/root_dev.h>
39 #include <linux/pci.h>
40 #include <linux/acpi.h>
41 #include <linux/kallsyms.h>
42 #include <linux/edd.h>
43 #include <linux/mmzone.h>
44 #include <linux/kexec.h>
47 #include <asm/uaccess.h>
48 #include <asm/system.h>
53 #include <video/edid.h>
56 #include <asm/mpspec.h>
57 #include <asm/mmu_context.h>
58 #include <asm/bootsetup.h>
59 #include <asm/proto.h>
60 #include <asm/setup.h>
61 #include <asm/mach_apic.h>
63 #include <asm/sections.h>
69 struct cpuinfo_x86 boot_cpu_data __read_mostly;
71 unsigned long mmu_cr4_features;
74 EXPORT_SYMBOL(acpi_disabled);
76 extern int __initdata acpi_ht;
77 extern acpi_interrupt_flags acpi_sci_flags;
78 int __initdata acpi_force = 0;
81 int acpi_numa __initdata;
83 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
86 unsigned long saved_video_mode;
90 EXPORT_SYMBOL(swiotlb);
96 struct drive_info_struct { char dummy[32]; } drive_info;
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;
278 /* Save unparsed command line copy for /proc/cmdline */
279 memcpy(saved_command_line, COMMAND_LINE, COMMAND_LINE_SIZE);
280 saved_command_line[COMMAND_LINE_SIZE-1] = '\0';
288 * If the BIOS enumerates physical processors before logical,
289 * maxcpus=N at enumeration-time can be used to disable HT.
291 else if (!memcmp(from, "maxcpus=", 8)) {
292 extern unsigned int maxcpus;
294 maxcpus = simple_strtoul(from + 8, NULL, 0);
298 /* "acpi=off" disables both ACPI table parsing and interpreter init */
299 if (!memcmp(from, "acpi=off", 8))
302 if (!memcmp(from, "acpi=force", 10)) {
303 /* add later when we do DMI horrors: */
308 /* acpi=ht just means: do ACPI MADT parsing
309 at bootup, but don't enable the full ACPI interpreter */
310 if (!memcmp(from, "acpi=ht", 7)) {
315 else if (!memcmp(from, "pci=noacpi", 10))
317 else if (!memcmp(from, "acpi=noirq", 10))
320 else if (!memcmp(from, "acpi_sci=edge", 13))
321 acpi_sci_flags.trigger = 1;
322 else if (!memcmp(from, "acpi_sci=level", 14))
323 acpi_sci_flags.trigger = 3;
324 else if (!memcmp(from, "acpi_sci=high", 13))
325 acpi_sci_flags.polarity = 1;
326 else if (!memcmp(from, "acpi_sci=low", 12))
327 acpi_sci_flags.polarity = 3;
329 /* acpi=strict disables out-of-spec workarounds */
330 else if (!memcmp(from, "acpi=strict", 11)) {
333 #ifdef CONFIG_X86_IO_APIC
334 else if (!memcmp(from, "acpi_skip_timer_override", 24))
335 acpi_skip_timer_override = 1;
339 if (!memcmp(from, "disable_timer_pin_1", 19))
340 disable_timer_pin_1 = 1;
341 if (!memcmp(from, "enable_timer_pin_1", 18))
342 disable_timer_pin_1 = -1;
344 if (!memcmp(from, "nolapic", 7) ||
345 !memcmp(from, "disableapic", 11))
348 if (!memcmp(from, "noapic", 6))
349 skip_ioapic_setup = 1;
351 if (!memcmp(from, "apic", 4)) {
352 skip_ioapic_setup = 0;
356 if (!memcmp(from, "mem=", 4))
357 parse_memopt(from+4, &from);
360 if (!memcmp(from, "numa=", 5))
364 #ifdef CONFIG_GART_IOMMU
365 if (!memcmp(from,"iommu=",6)) {
370 if (!memcmp(from,"oops=panic", 10))
373 if (!memcmp(from, "noexec=", 7))
374 nonx_setup(from + 7);
377 /* crashkernel=size@addr specifies the location to reserve for
378 * a crash kernel. By reserving this memory we guarantee
379 * that linux never set's it up as a DMA target.
380 * Useful for holding code to do something appropriate
381 * after a kernel panic.
383 else if (!memcmp(from, "crashkernel=", 12)) {
384 unsigned long size, base;
385 size = memparse(from+12, &from);
387 base = memparse(from+1, &from);
388 /* FIXME: Do I want a sanity check
389 * to validate the memory range?
391 crashk_res.start = base;
392 crashk_res.end = base + size - 1;
401 if (COMMAND_LINE_SIZE <= ++len)
406 *cmdline_p = command_line;
411 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
413 unsigned long bootmap_size, bootmap;
415 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
416 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size);
418 panic("Cannot find bootmem map of size %ld\n",bootmap_size);
419 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
420 e820_bootmem_free(NODE_DATA(0), 0, end_pfn << PAGE_SHIFT);
421 reserve_bootmem(bootmap, bootmap_size);
425 /* Use inline assembly to define this because the nops are defined
426 as inline assembly strings in the include files and we cannot
427 get them easily into strings. */
428 asm("\t.data\nk8nops: "
429 K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6
432 extern unsigned char k8nops[];
433 static unsigned char *k8_nops[ASM_NOP_MAX+1] = {
439 k8nops + 1 + 2 + 3 + 4,
440 k8nops + 1 + 2 + 3 + 4 + 5,
441 k8nops + 1 + 2 + 3 + 4 + 5 + 6,
442 k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
445 /* Replace instructions with better alternatives for this CPU type.
447 This runs before SMP is initialized to avoid SMP problems with
448 self modifying code. This implies that assymetric systems where
449 APs have less capabilities than the boot processor are not handled.
450 In this case boot with "noreplacement". */
451 void apply_alternatives(void *start, void *end)
455 for (a = start; (void *)a < end; a++) {
456 if (!boot_cpu_has(a->cpuid))
459 BUG_ON(a->replacementlen > a->instrlen);
460 __inline_memcpy(a->instr, a->replacement, a->replacementlen);
461 diff = a->instrlen - a->replacementlen;
463 /* Pad the rest with nops */
464 for (i = a->replacementlen; diff > 0; diff -= k, i += k) {
468 __inline_memcpy(a->instr + i, k8_nops[k], k);
473 static int no_replacement __initdata = 0;
475 void __init alternative_instructions(void)
477 extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
480 apply_alternatives(__alt_instructions, __alt_instructions_end);
483 static int __init noreplacement_setup(char *s)
489 __setup("noreplacement", noreplacement_setup);
491 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
493 #ifdef CONFIG_EDD_MODULE
497 * copy_edd() - Copy the BIOS EDD information
498 * from boot_params into a safe place.
501 static inline void copy_edd(void)
503 memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
504 memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
505 edd.mbr_signature_nr = EDD_MBR_SIG_NR;
506 edd.edd_info_nr = EDD_NR;
509 static inline void copy_edd(void)
514 #define EBDA_ADDR_POINTER 0x40E
515 static void __init reserve_ebda_region(void)
519 * there is a real-mode segmented pointer pointing to the
520 * 4K EBDA area at 0x40E
522 addr = *(unsigned short *)phys_to_virt(EBDA_ADDR_POINTER);
525 reserve_bootmem_generic(addr, PAGE_SIZE);
528 void __init setup_arch(char **cmdline_p)
530 unsigned long kernel_end;
532 ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
533 drive_info = DRIVE_INFO;
534 screen_info = SCREEN_INFO;
535 edid_info = EDID_INFO;
536 saved_video_mode = SAVED_VIDEO_MODE;
537 bootloader_type = LOADER_TYPE;
539 #ifdef CONFIG_BLK_DEV_RAM
540 rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
541 rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
542 rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
544 setup_memory_region();
547 if (!MOUNT_ROOT_RDONLY)
548 root_mountflags &= ~MS_RDONLY;
549 init_mm.start_code = (unsigned long) &_text;
550 init_mm.end_code = (unsigned long) &_etext;
551 init_mm.end_data = (unsigned long) &_edata;
552 init_mm.brk = (unsigned long) &_end;
554 code_resource.start = virt_to_phys(&_text);
555 code_resource.end = virt_to_phys(&_etext)-1;
556 data_resource.start = virt_to_phys(&_etext);
557 data_resource.end = virt_to_phys(&_edata)-1;
559 parse_cmdline_early(cmdline_p);
561 early_identify_cpu(&boot_cpu_data);
564 * partially used pages are not usable - thus
565 * we are rounding upwards:
567 end_pfn = e820_end_of_ram();
571 init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
577 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
578 * Call this early for SRAT node setup.
580 acpi_boot_table_init();
583 #ifdef CONFIG_ACPI_NUMA
585 * Parse SRAT to discover nodes.
591 numa_initmem_init(0, end_pfn);
593 contig_initmem_init(0, end_pfn);
596 /* Reserve direct mapping */
597 reserve_bootmem_generic(table_start << PAGE_SHIFT,
598 (table_end - table_start) << PAGE_SHIFT);
601 kernel_end = round_up(__pa_symbol(&_end),PAGE_SIZE);
602 reserve_bootmem_generic(HIGH_MEMORY, kernel_end - HIGH_MEMORY);
605 * reserve physical page 0 - it's a special BIOS page on many boxes,
606 * enabling clean reboots, SMP operation, laptop functions.
608 reserve_bootmem_generic(0, PAGE_SIZE);
610 /* reserve ebda region */
611 reserve_ebda_region();
615 * But first pinch a few for the stack/trampoline stuff
616 * FIXME: Don't need the extra page at 4K, but need to fix
617 * trampoline before removing it. (see the GDT stuff)
619 reserve_bootmem_generic(PAGE_SIZE, PAGE_SIZE);
621 /* Reserve SMP trampoline */
622 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, PAGE_SIZE);
625 #ifdef CONFIG_ACPI_SLEEP
627 * Reserve low memory region for sleep support.
629 acpi_reserve_bootmem();
631 #ifdef CONFIG_X86_LOCAL_APIC
633 * Find and reserve possible boot-time SMP configuration:
637 #ifdef CONFIG_BLK_DEV_INITRD
638 if (LOADER_TYPE && INITRD_START) {
639 if (INITRD_START + INITRD_SIZE <= (end_pfn << PAGE_SHIFT)) {
640 reserve_bootmem_generic(INITRD_START, INITRD_SIZE);
642 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
643 initrd_end = initrd_start+INITRD_SIZE;
646 printk(KERN_ERR "initrd extends beyond end of memory "
647 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
648 (unsigned long)(INITRD_START + INITRD_SIZE),
649 (unsigned long)(end_pfn << PAGE_SHIFT));
655 if (crashk_res.start != crashk_res.end) {
656 reserve_bootmem(crashk_res.start,
657 crashk_res.end - crashk_res.start + 1);
667 * Read APIC and some other early information from ACPI tables.
672 #ifdef CONFIG_X86_LOCAL_APIC
674 * get boot-time SMP configuration:
676 if (smp_found_config)
678 init_apic_mappings();
682 * Request address space for all standard RAM and ROM resources
683 * and also for regions reported as reserved by the e820.
686 e820_reserve_resources();
688 request_resource(&iomem_resource, &video_ram_resource);
692 /* request I/O space for devices used on all i[345]86 PCs */
693 for (i = 0; i < STANDARD_IO_RESOURCES; i++)
694 request_resource(&ioport_resource, &standard_io_resources[i]);
699 #ifdef CONFIG_GART_IOMMU
704 #if defined(CONFIG_VGA_CONSOLE)
705 conswitchp = &vga_con;
706 #elif defined(CONFIG_DUMMY_CONSOLE)
707 conswitchp = &dummy_con;
712 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
716 if (c->extended_cpuid_level < 0x80000004)
719 v = (unsigned int *) c->x86_model_id;
720 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
721 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
722 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
723 c->x86_model_id[48] = 0;
728 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
730 unsigned int n, dummy, eax, ebx, ecx, edx;
732 n = c->extended_cpuid_level;
734 if (n >= 0x80000005) {
735 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
736 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
737 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
738 c->x86_cache_size=(ecx>>24)+(edx>>24);
739 /* On K8 L1 TLB is inclusive, so don't count it */
743 if (n >= 0x80000006) {
744 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
745 ecx = cpuid_ecx(0x80000006);
746 c->x86_cache_size = ecx >> 16;
747 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
749 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
750 c->x86_cache_size, ecx & 0xFF);
754 cpuid(0x80000007, &dummy, &dummy, &dummy, &c->x86_power);
755 if (n >= 0x80000008) {
756 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
757 c->x86_virt_bits = (eax >> 8) & 0xff;
758 c->x86_phys_bits = eax & 0xff;
763 static int nearby_node(int apicid)
766 for (i = apicid - 1; i >= 0; i--) {
767 int node = apicid_to_node[i];
768 if (node != NUMA_NO_NODE && node_online(node))
771 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
772 int node = apicid_to_node[i];
773 if (node != NUMA_NO_NODE && node_online(node))
776 return first_node(node_online_map); /* Shouldn't happen */
781 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
782 * Assumes number of cores is a power of two.
784 static void __init amd_detect_cmp(struct cpuinfo_x86 *c)
787 int cpu = smp_processor_id();
791 unsigned apicid = phys_proc_id[cpu];
795 while ((1 << bits) < c->x86_max_cores)
798 /* Low order bits define the core id (index of core in socket) */
799 cpu_core_id[cpu] = phys_proc_id[cpu] & ((1 << bits)-1);
800 /* Convert the APIC ID into the socket ID */
801 phys_proc_id[cpu] >>= bits;
804 node = phys_proc_id[cpu];
805 if (apicid_to_node[apicid] != NUMA_NO_NODE)
806 node = apicid_to_node[apicid];
807 if (!node_online(node)) {
808 /* Two possibilities here:
809 - The CPU is missing memory and no node was created.
810 In that case try picking one from a nearby CPU
811 - The APIC IDs differ from the HyperTransport node IDs
812 which the K8 northbridge parsing fills in.
813 Assume they are all increased by a constant offset,
814 but in the same order as the HT nodeids.
815 If that doesn't result in a usable node fall back to the
816 path for the previous case. */
817 int ht_nodeid = apicid - (phys_proc_id[0] << bits);
818 if (ht_nodeid >= 0 &&
819 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
820 node = apicid_to_node[ht_nodeid];
821 /* Pick a nearby node */
822 if (!node_online(node))
823 node = nearby_node(apicid);
825 numa_set_node(cpu, node);
827 printk(KERN_INFO "CPU %d(%d) -> Node %d -> Core %d\n",
828 cpu, c->x86_max_cores, node, cpu_core_id[cpu]);
833 static int __init init_amd(struct cpuinfo_x86 *c)
842 * Disable TLB flush filter by setting HWCR.FFDIS on K8
843 * bit 6 of msr C001_0015
845 * Errata 63 for SH-B3 steppings
846 * Errata 122 for all steppings (F+ have it disabled by default)
849 rdmsrl(MSR_K8_HWCR, value);
851 wrmsrl(MSR_K8_HWCR, value);
855 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
856 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
857 clear_bit(0*32+31, &c->x86_capability);
860 level = cpuid_eax(1);
861 if ((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58)
862 set_bit(X86_FEATURE_K8_C, &c->x86_capability);
864 r = get_model_name(c);
868 /* Should distinguish Models here, but this is only
869 a fallback anyways. */
870 strcpy(c->x86_model_id, "Hammer");
874 display_cacheinfo(c);
876 if (c->extended_cpuid_level >= 0x80000008) {
877 c->x86_max_cores = (cpuid_ecx(0x80000008) & 0xff) + 1;
878 if (c->x86_max_cores & (c->x86_max_cores - 1))
879 c->x86_max_cores = 1;
887 static void __cpuinit detect_ht(struct cpuinfo_x86 *c)
890 u32 eax, ebx, ecx, edx;
891 int index_msb, core_bits;
892 int cpu = smp_processor_id();
894 cpuid(1, &eax, &ebx, &ecx, &edx);
896 c->apicid = phys_pkg_id(0);
898 if (!cpu_has(c, X86_FEATURE_HT) || cpu_has(c, X86_FEATURE_CMP_LEGACY))
901 smp_num_siblings = (ebx & 0xff0000) >> 16;
903 if (smp_num_siblings == 1) {
904 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
905 } else if (smp_num_siblings > 1 ) {
907 if (smp_num_siblings > NR_CPUS) {
908 printk(KERN_WARNING "CPU: Unsupported number of the siblings %d", smp_num_siblings);
909 smp_num_siblings = 1;
913 index_msb = get_count_order(smp_num_siblings);
914 phys_proc_id[cpu] = phys_pkg_id(index_msb);
916 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
919 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
921 index_msb = get_count_order(smp_num_siblings) ;
923 core_bits = get_count_order(c->x86_max_cores);
925 cpu_core_id[cpu] = phys_pkg_id(index_msb) &
926 ((1 << core_bits) - 1);
928 if (c->x86_max_cores > 1)
929 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
936 * find out the number of processor cores on the die
938 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
942 if (c->cpuid_level < 4)
951 return ((eax >> 26) + 1);
956 static void srat_detect_node(void)
960 int cpu = smp_processor_id();
962 /* Don't do the funky fallback heuristics the AMD version employs
964 node = apicid_to_node[hard_smp_processor_id()];
965 if (node == NUMA_NO_NODE)
967 numa_set_node(cpu, node);
970 printk(KERN_INFO "CPU %d -> Node %d\n", cpu, node);
974 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
979 init_intel_cacheinfo(c);
980 n = c->extended_cpuid_level;
981 if (n >= 0x80000008) {
982 unsigned eax = cpuid_eax(0x80000008);
983 c->x86_virt_bits = (eax >> 8) & 0xff;
984 c->x86_phys_bits = eax & 0xff;
985 /* CPUID workaround for Intel 0F34 CPU */
986 if (c->x86_vendor == X86_VENDOR_INTEL &&
987 c->x86 == 0xF && c->x86_model == 0x3 &&
989 c->x86_phys_bits = 36;
993 c->x86_cache_alignment = c->x86_clflush_size * 2;
995 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
996 c->x86_max_cores = intel_num_cpu_cores(c);
1001 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
1003 char *v = c->x86_vendor_id;
1005 if (!strcmp(v, "AuthenticAMD"))
1006 c->x86_vendor = X86_VENDOR_AMD;
1007 else if (!strcmp(v, "GenuineIntel"))
1008 c->x86_vendor = X86_VENDOR_INTEL;
1010 c->x86_vendor = X86_VENDOR_UNKNOWN;
1013 struct cpu_model_info {
1016 char *model_names[16];
1019 /* Do some early cpuid on the boot CPU to get some parameter that are
1020 needed before check_bugs. Everything advanced is in identify_cpu
1022 void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
1026 c->loops_per_jiffy = loops_per_jiffy;
1027 c->x86_cache_size = -1;
1028 c->x86_vendor = X86_VENDOR_UNKNOWN;
1029 c->x86_model = c->x86_mask = 0; /* So far unknown... */
1030 c->x86_vendor_id[0] = '\0'; /* Unset */
1031 c->x86_model_id[0] = '\0'; /* Unset */
1032 c->x86_clflush_size = 64;
1033 c->x86_cache_alignment = c->x86_clflush_size;
1034 c->x86_max_cores = 1;
1035 c->extended_cpuid_level = 0;
1036 memset(&c->x86_capability, 0, sizeof c->x86_capability);
1038 /* Get vendor name */
1039 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
1040 (unsigned int *)&c->x86_vendor_id[0],
1041 (unsigned int *)&c->x86_vendor_id[8],
1042 (unsigned int *)&c->x86_vendor_id[4]);
1046 /* Initialize the standard set of capabilities */
1047 /* Note that the vendor-specific code below might override */
1049 /* Intel-defined flags: level 0x00000001 */
1050 if (c->cpuid_level >= 0x00000001) {
1052 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
1053 &c->x86_capability[0]);
1054 c->x86 = (tfms >> 8) & 0xf;
1055 c->x86_model = (tfms >> 4) & 0xf;
1056 c->x86_mask = tfms & 0xf;
1058 c->x86 += (tfms >> 20) & 0xff;
1060 c->x86_model += ((tfms >> 16) & 0xF) << 4;
1061 if (c->x86_capability[0] & (1<<19))
1062 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1064 /* Have CPUID level 0 only - unheard of */
1069 phys_proc_id[smp_processor_id()] = (cpuid_ebx(1) >> 24) & 0xff;
1074 * This does the hard work of actually picking apart the CPU stuff...
1076 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1081 early_identify_cpu(c);
1083 /* AMD-defined flags: level 0x80000001 */
1084 xlvl = cpuid_eax(0x80000000);
1085 c->extended_cpuid_level = xlvl;
1086 if ((xlvl & 0xffff0000) == 0x80000000) {
1087 if (xlvl >= 0x80000001) {
1088 c->x86_capability[1] = cpuid_edx(0x80000001);
1089 c->x86_capability[6] = cpuid_ecx(0x80000001);
1091 if (xlvl >= 0x80000004)
1092 get_model_name(c); /* Default name */
1095 /* Transmeta-defined flags: level 0x80860001 */
1096 xlvl = cpuid_eax(0x80860000);
1097 if ((xlvl & 0xffff0000) == 0x80860000) {
1098 /* Don't set x86_cpuid_level here for now to not confuse. */
1099 if (xlvl >= 0x80860001)
1100 c->x86_capability[2] = cpuid_edx(0x80860001);
1104 * Vendor-specific initialization. In this section we
1105 * canonicalize the feature flags, meaning if there are
1106 * features a certain CPU supports which CPUID doesn't
1107 * tell us, CPUID claiming incorrect flags, or other bugs,
1108 * we handle them here.
1110 * At the end of this section, c->x86_capability better
1111 * indicate the features this CPU genuinely supports!
1113 switch (c->x86_vendor) {
1114 case X86_VENDOR_AMD:
1118 case X86_VENDOR_INTEL:
1122 case X86_VENDOR_UNKNOWN:
1124 display_cacheinfo(c);
1128 select_idle_routine(c);
1132 * On SMP, boot_cpu_data holds the common feature set between
1133 * all CPUs; so make sure that we indicate which features are
1134 * common between the CPUs. The first time this routine gets
1135 * executed, c == &boot_cpu_data.
1137 if (c != &boot_cpu_data) {
1138 /* AND the already accumulated flags with these */
1139 for (i = 0 ; i < NCAPINTS ; i++)
1140 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1143 #ifdef CONFIG_X86_MCE
1146 if (c == &boot_cpu_data)
1151 numa_add_cpu(smp_processor_id());
1156 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1158 if (c->x86_model_id[0])
1159 printk("%s", c->x86_model_id);
1161 if (c->x86_mask || c->cpuid_level >= 0)
1162 printk(" stepping %02x\n", c->x86_mask);
1168 * Get CPU information for use by the procfs.
1171 static int show_cpuinfo(struct seq_file *m, void *v)
1173 struct cpuinfo_x86 *c = v;
1176 * These flag bits must match the definitions in <asm/cpufeature.h>.
1177 * NULL means this bit is undefined or reserved; either way it doesn't
1178 * have meaning as far as Linux is concerned. Note that it's important
1179 * to realize there is a difference between this table and CPUID -- if
1180 * applications want to get the raw CPUID data, they should access
1181 * /dev/cpu/<cpu_nr>/cpuid instead.
1183 static char *x86_cap_flags[] = {
1185 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1186 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
1187 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
1188 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", NULL,
1191 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1192 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
1193 NULL, NULL, NULL, NULL, "nx", NULL, "mmxext", NULL,
1194 NULL, "fxsr_opt", NULL, NULL, NULL, "lm", "3dnowext", "3dnow",
1196 /* Transmeta-defined */
1197 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
1198 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1199 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1200 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1202 /* Other (Linux-defined) */
1203 "cxmmx", NULL, "cyrix_arr", "centaur_mcr", NULL,
1204 "constant_tsc", NULL, NULL,
1205 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1206 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1207 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1209 /* Intel-defined (#2) */
1210 "pni", NULL, NULL, "monitor", "ds_cpl", "vmx", NULL, "est",
1211 "tm2", NULL, "cid", NULL, NULL, "cx16", "xtpr", NULL,
1212 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1213 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1215 /* VIA/Cyrix/Centaur-defined */
1216 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
1217 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1218 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1219 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1221 /* AMD-defined (#2) */
1222 "lahf_lm", "cmp_legacy", NULL, NULL, NULL, NULL, NULL, NULL,
1223 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1224 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1225 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1227 static char *x86_power_flags[] = {
1228 "ts", /* temperature sensor */
1229 "fid", /* frequency id control */
1230 "vid", /* voltage id control */
1231 "ttp", /* thermal trip */
1238 if (!cpu_online(c-cpu_data))
1242 seq_printf(m,"processor\t: %u\n"
1244 "cpu family\t: %d\n"
1246 "model name\t: %s\n",
1247 (unsigned)(c-cpu_data),
1248 c->x86_vendor_id[0] ? c->x86_vendor_id : "unknown",
1251 c->x86_model_id[0] ? c->x86_model_id : "unknown");
1253 if (c->x86_mask || c->cpuid_level >= 0)
1254 seq_printf(m, "stepping\t: %d\n", c->x86_mask);
1256 seq_printf(m, "stepping\t: unknown\n");
1258 if (cpu_has(c,X86_FEATURE_TSC)) {
1259 seq_printf(m, "cpu MHz\t\t: %u.%03u\n",
1260 cpu_khz / 1000, (cpu_khz % 1000));
1264 if (c->x86_cache_size >= 0)
1265 seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
1268 if (smp_num_siblings * c->x86_max_cores > 1) {
1269 int cpu = c - cpu_data;
1270 seq_printf(m, "physical id\t: %d\n", phys_proc_id[cpu]);
1271 seq_printf(m, "siblings\t: %d\n", cpus_weight(cpu_core_map[cpu]));
1272 seq_printf(m, "core id\t\t: %d\n", cpu_core_id[cpu]);
1273 seq_printf(m, "cpu cores\t: %d\n", c->booted_cores);
1279 "fpu_exception\t: yes\n"
1280 "cpuid level\t: %d\n"
1287 for ( i = 0 ; i < 32*NCAPINTS ; i++ )
1288 if ( test_bit(i, &c->x86_capability) &&
1289 x86_cap_flags[i] != NULL )
1290 seq_printf(m, " %s", x86_cap_flags[i]);
1293 seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
1294 c->loops_per_jiffy/(500000/HZ),
1295 (c->loops_per_jiffy/(5000/HZ)) % 100);
1297 if (c->x86_tlbsize > 0)
1298 seq_printf(m, "TLB size\t: %d 4K pages\n", c->x86_tlbsize);
1299 seq_printf(m, "clflush size\t: %d\n", c->x86_clflush_size);
1300 seq_printf(m, "cache_alignment\t: %d\n", c->x86_cache_alignment);
1302 seq_printf(m, "address sizes\t: %u bits physical, %u bits virtual\n",
1303 c->x86_phys_bits, c->x86_virt_bits);
1305 seq_printf(m, "power management:");
1308 for (i = 0; i < 32; i++)
1309 if (c->x86_power & (1 << i)) {
1310 if (i < ARRAY_SIZE(x86_power_flags))
1311 seq_printf(m, " %s", x86_power_flags[i]);
1313 seq_printf(m, " [%d]", i);
1317 seq_printf(m, "\n\n");
1322 static void *c_start(struct seq_file *m, loff_t *pos)
1324 return *pos < NR_CPUS ? cpu_data + *pos : NULL;
1327 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
1330 return c_start(m, pos);
1333 static void c_stop(struct seq_file *m, void *v)
1337 struct seq_operations cpuinfo_op = {
1341 .show = show_cpuinfo,