[PATCH] x86-64/i386: Intel HT, Multi core detection fixes
[linux-2.6/mini2440.git] / arch / x86_64 / kernel / setup.c
blob40c77f6fe4b0d246b105d4bad17db0cbd110f96a
1 /*
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.
9 * $Id$
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>
19 #include <linux/mm.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>
46 #include <asm/mtrr.h>
47 #include <asm/uaccess.h>
48 #include <asm/system.h>
49 #include <asm/io.h>
50 #include <asm/smp.h>
51 #include <asm/msr.h>
52 #include <asm/desc.h>
53 #include <video/edid.h>
54 #include <asm/e820.h>
55 #include <asm/dma.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>
62 #include <asm/numa.h>
63 #include <asm/sections.h>
66 * Machine setup..
69 struct cpuinfo_x86 boot_cpu_data __read_mostly;
71 unsigned long mmu_cr4_features;
73 int acpi_disabled;
74 EXPORT_SYMBOL(acpi_disabled);
75 #ifdef CONFIG_ACPI
76 extern int __initdata acpi_ht;
77 extern acpi_interrupt_flags acpi_sci_flags;
78 int __initdata acpi_force = 0;
79 #endif
81 int acpi_numa __initdata;
83 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
84 int bootloader_type;
86 unsigned long saved_video_mode;
88 #ifdef CONFIG_SWIOTLB
89 int swiotlb;
90 EXPORT_SYMBOL(swiotlb);
91 #endif
94 * Setup options
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;
104 struct e820map e820;
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",
138 .start = 0,
139 .end = 0,
140 .flags = IORESOURCE_RAM,
142 struct resource code_resource = {
143 .name = "Kernel code",
144 .start = 0,
145 .end = 0,
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",
153 .start = 0xf0000,
154 .end = 0xfffff,
155 .flags = IORESOURCE_ROM,
158 static struct resource extension_rom_resource = {
159 .name = "Extension ROM",
160 .start = 0xe0000,
161 .end = 0xeffff,
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 = {
184 .name = "Video ROM",
185 .start = 0xc0000,
186 .end = 0xc7fff,
187 .flags = IORESOURCE_ROM,
190 static struct resource video_ram_resource = {
191 .name = "Video RAM area",
192 .start = 0xa0000,
193 .end = 0xbffff,
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++)
204 sum += *p;
205 return sum == 0;
208 static void __init probe_roms(void)
210 unsigned long start, length, upper;
211 unsigned char *rom;
212 int i;
214 /* video rom */
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))
219 continue;
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);
231 break;
234 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
235 if (start < upper)
236 start = upper;
238 /* system rom */
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))
256 continue;
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))
263 continue;
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;
276 int len = 0;
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';
282 for (;;) {
283 if (c != ' ')
284 goto next_char;
286 #ifdef CONFIG_SMP
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);
296 #endif
297 #ifdef CONFIG_ACPI
298 /* "acpi=off" disables both ACPI table parsing and interpreter init */
299 if (!memcmp(from, "acpi=off", 8))
300 disable_acpi();
302 if (!memcmp(from, "acpi=force", 10)) {
303 /* add later when we do DMI horrors: */
304 acpi_force = 1;
305 acpi_disabled = 0;
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)) {
311 if (!acpi_force)
312 disable_acpi();
313 acpi_ht = 1;
315 else if (!memcmp(from, "pci=noacpi", 10))
316 acpi_disable_pci();
317 else if (!memcmp(from, "acpi=noirq", 10))
318 acpi_noirq_set();
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)) {
331 acpi_strict = 1;
333 #ifdef CONFIG_X86_IO_APIC
334 else if (!memcmp(from, "acpi_skip_timer_override", 24))
335 acpi_skip_timer_override = 1;
336 #endif
337 #endif
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))
346 disable_apic = 1;
348 if (!memcmp(from, "noapic", 6))
349 skip_ioapic_setup = 1;
351 if (!memcmp(from, "apic", 4)) {
352 skip_ioapic_setup = 0;
353 ioapic_force = 1;
356 if (!memcmp(from, "mem=", 4))
357 parse_memopt(from+4, &from);
359 #ifdef CONFIG_NUMA
360 if (!memcmp(from, "numa=", 5))
361 numa_setup(from+5);
362 #endif
364 #ifdef CONFIG_GART_IOMMU
365 if (!memcmp(from,"iommu=",6)) {
366 iommu_setup(from+6);
368 #endif
370 if (!memcmp(from,"oops=panic", 10))
371 panic_on_oops = 1;
373 if (!memcmp(from, "noexec=", 7))
374 nonx_setup(from + 7);
376 #ifdef CONFIG_KEXEC
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);
386 if (*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;
395 #endif
397 next_char:
398 c = *(from++);
399 if (!c)
400 break;
401 if (COMMAND_LINE_SIZE <= ++len)
402 break;
403 *(to++) = c;
405 *to = '\0';
406 *cmdline_p = command_line;
409 #ifndef CONFIG_NUMA
410 static void __init
411 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
413 unsigned long bootmap_size, bootmap;
415 memory_present(0, start_pfn, end_pfn);
416 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
417 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size);
418 if (bootmap == -1L)
419 panic("Cannot find bootmem map of size %ld\n",bootmap_size);
420 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
421 e820_bootmem_free(NODE_DATA(0), 0, end_pfn << PAGE_SHIFT);
422 reserve_bootmem(bootmap, bootmap_size);
424 #endif
426 /* Use inline assembly to define this because the nops are defined
427 as inline assembly strings in the include files and we cannot
428 get them easily into strings. */
429 asm("\t.data\nk8nops: "
430 K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6
431 K8_NOP7 K8_NOP8);
433 extern unsigned char k8nops[];
434 static unsigned char *k8_nops[ASM_NOP_MAX+1] = {
435 NULL,
436 k8nops,
437 k8nops + 1,
438 k8nops + 1 + 2,
439 k8nops + 1 + 2 + 3,
440 k8nops + 1 + 2 + 3 + 4,
441 k8nops + 1 + 2 + 3 + 4 + 5,
442 k8nops + 1 + 2 + 3 + 4 + 5 + 6,
443 k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
446 /* Replace instructions with better alternatives for this CPU type.
448 This runs before SMP is initialized to avoid SMP problems with
449 self modifying code. This implies that assymetric systems where
450 APs have less capabilities than the boot processor are not handled.
451 In this case boot with "noreplacement". */
452 void apply_alternatives(void *start, void *end)
454 struct alt_instr *a;
455 int diff, i, k;
456 for (a = start; (void *)a < end; a++) {
457 if (!boot_cpu_has(a->cpuid))
458 continue;
460 BUG_ON(a->replacementlen > a->instrlen);
461 __inline_memcpy(a->instr, a->replacement, a->replacementlen);
462 diff = a->instrlen - a->replacementlen;
464 /* Pad the rest with nops */
465 for (i = a->replacementlen; diff > 0; diff -= k, i += k) {
466 k = diff;
467 if (k > ASM_NOP_MAX)
468 k = ASM_NOP_MAX;
469 __inline_memcpy(a->instr + i, k8_nops[k], k);
474 static int no_replacement __initdata = 0;
476 void __init alternative_instructions(void)
478 extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
479 if (no_replacement)
480 return;
481 apply_alternatives(__alt_instructions, __alt_instructions_end);
484 static int __init noreplacement_setup(char *s)
486 no_replacement = 1;
487 return 0;
490 __setup("noreplacement", noreplacement_setup);
492 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
493 struct edd edd;
494 #ifdef CONFIG_EDD_MODULE
495 EXPORT_SYMBOL(edd);
496 #endif
498 * copy_edd() - Copy the BIOS EDD information
499 * from boot_params into a safe place.
502 static inline void copy_edd(void)
504 memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
505 memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
506 edd.mbr_signature_nr = EDD_MBR_SIG_NR;
507 edd.edd_info_nr = EDD_NR;
509 #else
510 static inline void copy_edd(void)
513 #endif
515 #define EBDA_ADDR_POINTER 0x40E
516 static void __init reserve_ebda_region(void)
518 unsigned int addr;
519 /**
520 * there is a real-mode segmented pointer pointing to the
521 * 4K EBDA area at 0x40E
523 addr = *(unsigned short *)phys_to_virt(EBDA_ADDR_POINTER);
524 addr <<= 4;
525 if (addr)
526 reserve_bootmem_generic(addr, PAGE_SIZE);
529 void __init setup_arch(char **cmdline_p)
531 unsigned long kernel_end;
533 ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
534 drive_info = DRIVE_INFO;
535 screen_info = SCREEN_INFO;
536 edid_info = EDID_INFO;
537 saved_video_mode = SAVED_VIDEO_MODE;
538 bootloader_type = LOADER_TYPE;
540 #ifdef CONFIG_BLK_DEV_RAM
541 rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
542 rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
543 rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
544 #endif
545 setup_memory_region();
546 copy_edd();
548 if (!MOUNT_ROOT_RDONLY)
549 root_mountflags &= ~MS_RDONLY;
550 init_mm.start_code = (unsigned long) &_text;
551 init_mm.end_code = (unsigned long) &_etext;
552 init_mm.end_data = (unsigned long) &_edata;
553 init_mm.brk = (unsigned long) &_end;
555 code_resource.start = virt_to_phys(&_text);
556 code_resource.end = virt_to_phys(&_etext)-1;
557 data_resource.start = virt_to_phys(&_etext);
558 data_resource.end = virt_to_phys(&_edata)-1;
560 parse_cmdline_early(cmdline_p);
562 early_identify_cpu(&boot_cpu_data);
565 * partially used pages are not usable - thus
566 * we are rounding upwards:
568 end_pfn = e820_end_of_ram();
570 check_efer();
572 init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
574 zap_low_mappings(0);
576 #ifdef CONFIG_ACPI
578 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
579 * Call this early for SRAT node setup.
581 acpi_boot_table_init();
582 #endif
584 #ifdef CONFIG_ACPI_NUMA
586 * Parse SRAT to discover nodes.
588 acpi_numa_init();
589 #endif
591 #ifdef CONFIG_NUMA
592 numa_initmem_init(0, end_pfn);
593 #else
594 contig_initmem_init(0, end_pfn);
595 #endif
597 /* Reserve direct mapping */
598 reserve_bootmem_generic(table_start << PAGE_SHIFT,
599 (table_end - table_start) << PAGE_SHIFT);
601 /* reserve kernel */
602 kernel_end = round_up(__pa_symbol(&_end),PAGE_SIZE);
603 reserve_bootmem_generic(HIGH_MEMORY, kernel_end - HIGH_MEMORY);
606 * reserve physical page 0 - it's a special BIOS page on many boxes,
607 * enabling clean reboots, SMP operation, laptop functions.
609 reserve_bootmem_generic(0, PAGE_SIZE);
611 /* reserve ebda region */
612 reserve_ebda_region();
614 #ifdef CONFIG_SMP
616 * But first pinch a few for the stack/trampoline stuff
617 * FIXME: Don't need the extra page at 4K, but need to fix
618 * trampoline before removing it. (see the GDT stuff)
620 reserve_bootmem_generic(PAGE_SIZE, PAGE_SIZE);
622 /* Reserve SMP trampoline */
623 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, PAGE_SIZE);
624 #endif
626 #ifdef CONFIG_ACPI_SLEEP
628 * Reserve low memory region for sleep support.
630 acpi_reserve_bootmem();
631 #endif
632 #ifdef CONFIG_X86_LOCAL_APIC
634 * Find and reserve possible boot-time SMP configuration:
636 find_smp_config();
637 #endif
638 #ifdef CONFIG_BLK_DEV_INITRD
639 if (LOADER_TYPE && INITRD_START) {
640 if (INITRD_START + INITRD_SIZE <= (end_pfn << PAGE_SHIFT)) {
641 reserve_bootmem_generic(INITRD_START, INITRD_SIZE);
642 initrd_start =
643 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
644 initrd_end = initrd_start+INITRD_SIZE;
646 else {
647 printk(KERN_ERR "initrd extends beyond end of memory "
648 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
649 (unsigned long)(INITRD_START + INITRD_SIZE),
650 (unsigned long)(end_pfn << PAGE_SHIFT));
651 initrd_start = 0;
654 #endif
655 #ifdef CONFIG_KEXEC
656 if (crashk_res.start != crashk_res.end) {
657 reserve_bootmem(crashk_res.start,
658 crashk_res.end - crashk_res.start + 1);
660 #endif
662 sparse_init();
664 paging_init();
666 check_ioapic();
668 #ifdef CONFIG_ACPI
670 * Read APIC and some other early information from ACPI tables.
672 acpi_boot_init();
673 #endif
675 #ifdef CONFIG_X86_LOCAL_APIC
677 * get boot-time SMP configuration:
679 if (smp_found_config)
680 get_smp_config();
681 init_apic_mappings();
682 #endif
685 * Request address space for all standard RAM and ROM resources
686 * and also for regions reported as reserved by the e820.
688 probe_roms();
689 e820_reserve_resources();
691 request_resource(&iomem_resource, &video_ram_resource);
694 unsigned i;
695 /* request I/O space for devices used on all i[345]86 PCs */
696 for (i = 0; i < STANDARD_IO_RESOURCES; i++)
697 request_resource(&ioport_resource, &standard_io_resources[i]);
700 e820_setup_gap();
702 #ifdef CONFIG_GART_IOMMU
703 iommu_hole_init();
704 #endif
706 #ifdef CONFIG_VT
707 #if defined(CONFIG_VGA_CONSOLE)
708 conswitchp = &vga_con;
709 #elif defined(CONFIG_DUMMY_CONSOLE)
710 conswitchp = &dummy_con;
711 #endif
712 #endif
715 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
717 unsigned int *v;
719 if (c->extended_cpuid_level < 0x80000004)
720 return 0;
722 v = (unsigned int *) c->x86_model_id;
723 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
724 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
725 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
726 c->x86_model_id[48] = 0;
727 return 1;
731 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
733 unsigned int n, dummy, eax, ebx, ecx, edx;
735 n = c->extended_cpuid_level;
737 if (n >= 0x80000005) {
738 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
739 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
740 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
741 c->x86_cache_size=(ecx>>24)+(edx>>24);
742 /* On K8 L1 TLB is inclusive, so don't count it */
743 c->x86_tlbsize = 0;
746 if (n >= 0x80000006) {
747 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
748 ecx = cpuid_ecx(0x80000006);
749 c->x86_cache_size = ecx >> 16;
750 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
752 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
753 c->x86_cache_size, ecx & 0xFF);
756 if (n >= 0x80000007)
757 cpuid(0x80000007, &dummy, &dummy, &dummy, &c->x86_power);
758 if (n >= 0x80000008) {
759 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
760 c->x86_virt_bits = (eax >> 8) & 0xff;
761 c->x86_phys_bits = eax & 0xff;
765 #ifdef CONFIG_NUMA
766 static int nearby_node(int apicid)
768 int i;
769 for (i = apicid - 1; i >= 0; i--) {
770 int node = apicid_to_node[i];
771 if (node != NUMA_NO_NODE && node_online(node))
772 return node;
774 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
775 int node = apicid_to_node[i];
776 if (node != NUMA_NO_NODE && node_online(node))
777 return node;
779 return first_node(node_online_map); /* Shouldn't happen */
781 #endif
784 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
785 * Assumes number of cores is a power of two.
787 static void __init amd_detect_cmp(struct cpuinfo_x86 *c)
789 #ifdef CONFIG_SMP
790 int cpu = smp_processor_id();
791 unsigned bits;
792 #ifdef CONFIG_NUMA
793 int node = 0;
794 unsigned apicid = phys_proc_id[cpu];
795 #endif
797 bits = 0;
798 while ((1 << bits) < c->x86_max_cores)
799 bits++;
801 /* Low order bits define the core id (index of core in socket) */
802 cpu_core_id[cpu] = phys_proc_id[cpu] & ((1 << bits)-1);
803 /* Convert the APIC ID into the socket ID */
804 phys_proc_id[cpu] >>= bits;
806 #ifdef CONFIG_NUMA
807 node = phys_proc_id[cpu];
808 if (apicid_to_node[apicid] != NUMA_NO_NODE)
809 node = apicid_to_node[apicid];
810 if (!node_online(node)) {
811 /* Two possibilities here:
812 - The CPU is missing memory and no node was created.
813 In that case try picking one from a nearby CPU
814 - The APIC IDs differ from the HyperTransport node IDs
815 which the K8 northbridge parsing fills in.
816 Assume they are all increased by a constant offset,
817 but in the same order as the HT nodeids.
818 If that doesn't result in a usable node fall back to the
819 path for the previous case. */
820 int ht_nodeid = apicid - (phys_proc_id[0] << bits);
821 if (ht_nodeid >= 0 &&
822 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
823 node = apicid_to_node[ht_nodeid];
824 /* Pick a nearby node */
825 if (!node_online(node))
826 node = nearby_node(apicid);
828 numa_set_node(cpu, node);
830 printk(KERN_INFO "CPU %d(%d) -> Node %d -> Core %d\n",
831 cpu, c->x86_max_cores, node, cpu_core_id[cpu]);
832 #endif
833 #endif
836 static int __init init_amd(struct cpuinfo_x86 *c)
838 int r;
839 int level;
841 #ifdef CONFIG_SMP
842 unsigned long value;
845 * Disable TLB flush filter by setting HWCR.FFDIS on K8
846 * bit 6 of msr C001_0015
848 * Errata 63 for SH-B3 steppings
849 * Errata 122 for all steppings (F+ have it disabled by default)
851 if (c->x86 == 15) {
852 rdmsrl(MSR_K8_HWCR, value);
853 value |= 1 << 6;
854 wrmsrl(MSR_K8_HWCR, value);
856 #endif
858 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
859 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
860 clear_bit(0*32+31, &c->x86_capability);
862 /* C-stepping K8? */
863 level = cpuid_eax(1);
864 if ((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58)
865 set_bit(X86_FEATURE_K8_C, &c->x86_capability);
867 r = get_model_name(c);
868 if (!r) {
869 switch (c->x86) {
870 case 15:
871 /* Should distinguish Models here, but this is only
872 a fallback anyways. */
873 strcpy(c->x86_model_id, "Hammer");
874 break;
877 display_cacheinfo(c);
879 if (c->extended_cpuid_level >= 0x80000008) {
880 c->x86_max_cores = (cpuid_ecx(0x80000008) & 0xff) + 1;
881 if (c->x86_max_cores & (c->x86_max_cores - 1))
882 c->x86_max_cores = 1;
884 amd_detect_cmp(c);
887 return r;
890 static void __cpuinit detect_ht(struct cpuinfo_x86 *c)
892 #ifdef CONFIG_SMP
893 u32 eax, ebx, ecx, edx;
894 int index_msb, core_bits;
895 int cpu = smp_processor_id();
897 cpuid(1, &eax, &ebx, &ecx, &edx);
899 c->apicid = phys_pkg_id(0);
901 if (!cpu_has(c, X86_FEATURE_HT) || cpu_has(c, X86_FEATURE_CMP_LEGACY))
902 return;
904 smp_num_siblings = (ebx & 0xff0000) >> 16;
906 if (smp_num_siblings == 1) {
907 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
908 } else if (smp_num_siblings > 1 ) {
910 if (smp_num_siblings > NR_CPUS) {
911 printk(KERN_WARNING "CPU: Unsupported number of the siblings %d", smp_num_siblings);
912 smp_num_siblings = 1;
913 return;
916 index_msb = get_count_order(smp_num_siblings);
917 phys_proc_id[cpu] = phys_pkg_id(index_msb);
919 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
920 phys_proc_id[cpu]);
922 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
924 index_msb = get_count_order(smp_num_siblings) ;
926 core_bits = get_count_order(c->x86_max_cores);
928 cpu_core_id[cpu] = phys_pkg_id(index_msb) &
929 ((1 << core_bits) - 1);
931 if (c->x86_max_cores > 1)
932 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
933 cpu_core_id[cpu]);
935 #endif
939 * find out the number of processor cores on the die
941 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
943 unsigned int eax;
945 if (c->cpuid_level < 4)
946 return 1;
948 __asm__("cpuid"
949 : "=a" (eax)
950 : "0" (4), "c" (0)
951 : "bx", "dx");
953 if (eax & 0x1f)
954 return ((eax >> 26) + 1);
955 else
956 return 1;
959 static void srat_detect_node(void)
961 #ifdef CONFIG_NUMA
962 unsigned node;
963 int cpu = smp_processor_id();
965 /* Don't do the funky fallback heuristics the AMD version employs
966 for now. */
967 node = apicid_to_node[hard_smp_processor_id()];
968 if (node == NUMA_NO_NODE)
969 node = 0;
970 numa_set_node(cpu, node);
972 if (acpi_numa > 0)
973 printk(KERN_INFO "CPU %d -> Node %d\n", cpu, node);
974 #endif
977 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
979 /* Cache sizes */
980 unsigned n;
982 init_intel_cacheinfo(c);
983 n = c->extended_cpuid_level;
984 if (n >= 0x80000008) {
985 unsigned eax = cpuid_eax(0x80000008);
986 c->x86_virt_bits = (eax >> 8) & 0xff;
987 c->x86_phys_bits = eax & 0xff;
988 /* CPUID workaround for Intel 0F34 CPU */
989 if (c->x86_vendor == X86_VENDOR_INTEL &&
990 c->x86 == 0xF && c->x86_model == 0x3 &&
991 c->x86_mask == 0x4)
992 c->x86_phys_bits = 36;
995 if (c->x86 == 15)
996 c->x86_cache_alignment = c->x86_clflush_size * 2;
997 if (c->x86 >= 15)
998 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
999 c->x86_max_cores = intel_num_cpu_cores(c);
1001 srat_detect_node();
1004 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
1006 char *v = c->x86_vendor_id;
1008 if (!strcmp(v, "AuthenticAMD"))
1009 c->x86_vendor = X86_VENDOR_AMD;
1010 else if (!strcmp(v, "GenuineIntel"))
1011 c->x86_vendor = X86_VENDOR_INTEL;
1012 else
1013 c->x86_vendor = X86_VENDOR_UNKNOWN;
1016 struct cpu_model_info {
1017 int vendor;
1018 int family;
1019 char *model_names[16];
1022 /* Do some early cpuid on the boot CPU to get some parameter that are
1023 needed before check_bugs. Everything advanced is in identify_cpu
1024 below. */
1025 void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
1027 u32 tfms;
1029 c->loops_per_jiffy = loops_per_jiffy;
1030 c->x86_cache_size = -1;
1031 c->x86_vendor = X86_VENDOR_UNKNOWN;
1032 c->x86_model = c->x86_mask = 0; /* So far unknown... */
1033 c->x86_vendor_id[0] = '\0'; /* Unset */
1034 c->x86_model_id[0] = '\0'; /* Unset */
1035 c->x86_clflush_size = 64;
1036 c->x86_cache_alignment = c->x86_clflush_size;
1037 c->x86_max_cores = 1;
1038 c->extended_cpuid_level = 0;
1039 memset(&c->x86_capability, 0, sizeof c->x86_capability);
1041 /* Get vendor name */
1042 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
1043 (unsigned int *)&c->x86_vendor_id[0],
1044 (unsigned int *)&c->x86_vendor_id[8],
1045 (unsigned int *)&c->x86_vendor_id[4]);
1047 get_cpu_vendor(c);
1049 /* Initialize the standard set of capabilities */
1050 /* Note that the vendor-specific code below might override */
1052 /* Intel-defined flags: level 0x00000001 */
1053 if (c->cpuid_level >= 0x00000001) {
1054 __u32 misc;
1055 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
1056 &c->x86_capability[0]);
1057 c->x86 = (tfms >> 8) & 0xf;
1058 c->x86_model = (tfms >> 4) & 0xf;
1059 c->x86_mask = tfms & 0xf;
1060 if (c->x86 == 0xf)
1061 c->x86 += (tfms >> 20) & 0xff;
1062 if (c->x86 >= 0x6)
1063 c->x86_model += ((tfms >> 16) & 0xF) << 4;
1064 if (c->x86_capability[0] & (1<<19))
1065 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1066 } else {
1067 /* Have CPUID level 0 only - unheard of */
1068 c->x86 = 4;
1071 #ifdef CONFIG_SMP
1072 phys_proc_id[smp_processor_id()] = (cpuid_ebx(1) >> 24) & 0xff;
1073 #endif
1077 * This does the hard work of actually picking apart the CPU stuff...
1079 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1081 int i;
1082 u32 xlvl;
1084 early_identify_cpu(c);
1086 /* AMD-defined flags: level 0x80000001 */
1087 xlvl = cpuid_eax(0x80000000);
1088 c->extended_cpuid_level = xlvl;
1089 if ((xlvl & 0xffff0000) == 0x80000000) {
1090 if (xlvl >= 0x80000001) {
1091 c->x86_capability[1] = cpuid_edx(0x80000001);
1092 c->x86_capability[6] = cpuid_ecx(0x80000001);
1094 if (xlvl >= 0x80000004)
1095 get_model_name(c); /* Default name */
1098 /* Transmeta-defined flags: level 0x80860001 */
1099 xlvl = cpuid_eax(0x80860000);
1100 if ((xlvl & 0xffff0000) == 0x80860000) {
1101 /* Don't set x86_cpuid_level here for now to not confuse. */
1102 if (xlvl >= 0x80860001)
1103 c->x86_capability[2] = cpuid_edx(0x80860001);
1107 * Vendor-specific initialization. In this section we
1108 * canonicalize the feature flags, meaning if there are
1109 * features a certain CPU supports which CPUID doesn't
1110 * tell us, CPUID claiming incorrect flags, or other bugs,
1111 * we handle them here.
1113 * At the end of this section, c->x86_capability better
1114 * indicate the features this CPU genuinely supports!
1116 switch (c->x86_vendor) {
1117 case X86_VENDOR_AMD:
1118 init_amd(c);
1119 break;
1121 case X86_VENDOR_INTEL:
1122 init_intel(c);
1123 break;
1125 case X86_VENDOR_UNKNOWN:
1126 default:
1127 display_cacheinfo(c);
1128 break;
1131 select_idle_routine(c);
1132 detect_ht(c);
1135 * On SMP, boot_cpu_data holds the common feature set between
1136 * all CPUs; so make sure that we indicate which features are
1137 * common between the CPUs. The first time this routine gets
1138 * executed, c == &boot_cpu_data.
1140 if (c != &boot_cpu_data) {
1141 /* AND the already accumulated flags with these */
1142 for (i = 0 ; i < NCAPINTS ; i++)
1143 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1146 #ifdef CONFIG_X86_MCE
1147 mcheck_init(c);
1148 #endif
1149 if (c == &boot_cpu_data)
1150 mtrr_bp_init();
1151 else
1152 mtrr_ap_init();
1153 #ifdef CONFIG_NUMA
1154 numa_add_cpu(smp_processor_id());
1155 #endif
1159 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1161 if (c->x86_model_id[0])
1162 printk("%s", c->x86_model_id);
1164 if (c->x86_mask || c->cpuid_level >= 0)
1165 printk(" stepping %02x\n", c->x86_mask);
1166 else
1167 printk("\n");
1171 * Get CPU information for use by the procfs.
1174 static int show_cpuinfo(struct seq_file *m, void *v)
1176 struct cpuinfo_x86 *c = v;
1179 * These flag bits must match the definitions in <asm/cpufeature.h>.
1180 * NULL means this bit is undefined or reserved; either way it doesn't
1181 * have meaning as far as Linux is concerned. Note that it's important
1182 * to realize there is a difference between this table and CPUID -- if
1183 * applications want to get the raw CPUID data, they should access
1184 * /dev/cpu/<cpu_nr>/cpuid instead.
1186 static char *x86_cap_flags[] = {
1187 /* Intel-defined */
1188 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1189 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
1190 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
1191 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", NULL,
1193 /* AMD-defined */
1194 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1195 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
1196 NULL, NULL, NULL, NULL, "nx", NULL, "mmxext", NULL,
1197 NULL, "fxsr_opt", NULL, NULL, NULL, "lm", "3dnowext", "3dnow",
1199 /* Transmeta-defined */
1200 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
1201 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1202 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1203 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1205 /* Other (Linux-defined) */
1206 "cxmmx", NULL, "cyrix_arr", "centaur_mcr", NULL,
1207 "constant_tsc", NULL, NULL,
1208 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1209 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1210 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1212 /* Intel-defined (#2) */
1213 "pni", NULL, NULL, "monitor", "ds_cpl", "vmx", NULL, "est",
1214 "tm2", NULL, "cid", NULL, NULL, "cx16", "xtpr", NULL,
1215 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1216 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1218 /* VIA/Cyrix/Centaur-defined */
1219 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
1220 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1221 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1222 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1224 /* AMD-defined (#2) */
1225 "lahf_lm", "cmp_legacy", NULL, NULL, NULL, NULL, NULL, NULL,
1226 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1227 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1228 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1230 static char *x86_power_flags[] = {
1231 "ts", /* temperature sensor */
1232 "fid", /* frequency id control */
1233 "vid", /* voltage id control */
1234 "ttp", /* thermal trip */
1235 "tm",
1236 "stc"
1240 #ifdef CONFIG_SMP
1241 if (!cpu_online(c-cpu_data))
1242 return 0;
1243 #endif
1245 seq_printf(m,"processor\t: %u\n"
1246 "vendor_id\t: %s\n"
1247 "cpu family\t: %d\n"
1248 "model\t\t: %d\n"
1249 "model name\t: %s\n",
1250 (unsigned)(c-cpu_data),
1251 c->x86_vendor_id[0] ? c->x86_vendor_id : "unknown",
1252 c->x86,
1253 (int)c->x86_model,
1254 c->x86_model_id[0] ? c->x86_model_id : "unknown");
1256 if (c->x86_mask || c->cpuid_level >= 0)
1257 seq_printf(m, "stepping\t: %d\n", c->x86_mask);
1258 else
1259 seq_printf(m, "stepping\t: unknown\n");
1261 if (cpu_has(c,X86_FEATURE_TSC)) {
1262 seq_printf(m, "cpu MHz\t\t: %u.%03u\n",
1263 cpu_khz / 1000, (cpu_khz % 1000));
1266 /* Cache size */
1267 if (c->x86_cache_size >= 0)
1268 seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
1270 #ifdef CONFIG_SMP
1271 if (smp_num_siblings * c->x86_max_cores > 1) {
1272 int cpu = c - cpu_data;
1273 seq_printf(m, "physical id\t: %d\n", phys_proc_id[cpu]);
1274 seq_printf(m, "siblings\t: %d\n", cpus_weight(cpu_core_map[cpu]));
1275 seq_printf(m, "core id\t\t: %d\n", cpu_core_id[cpu]);
1276 seq_printf(m, "cpu cores\t: %d\n", c->booted_cores);
1278 #endif
1280 seq_printf(m,
1281 "fpu\t\t: yes\n"
1282 "fpu_exception\t: yes\n"
1283 "cpuid level\t: %d\n"
1284 "wp\t\t: yes\n"
1285 "flags\t\t:",
1286 c->cpuid_level);
1289 int i;
1290 for ( i = 0 ; i < 32*NCAPINTS ; i++ )
1291 if ( test_bit(i, &c->x86_capability) &&
1292 x86_cap_flags[i] != NULL )
1293 seq_printf(m, " %s", x86_cap_flags[i]);
1296 seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
1297 c->loops_per_jiffy/(500000/HZ),
1298 (c->loops_per_jiffy/(5000/HZ)) % 100);
1300 if (c->x86_tlbsize > 0)
1301 seq_printf(m, "TLB size\t: %d 4K pages\n", c->x86_tlbsize);
1302 seq_printf(m, "clflush size\t: %d\n", c->x86_clflush_size);
1303 seq_printf(m, "cache_alignment\t: %d\n", c->x86_cache_alignment);
1305 seq_printf(m, "address sizes\t: %u bits physical, %u bits virtual\n",
1306 c->x86_phys_bits, c->x86_virt_bits);
1308 seq_printf(m, "power management:");
1310 unsigned i;
1311 for (i = 0; i < 32; i++)
1312 if (c->x86_power & (1 << i)) {
1313 if (i < ARRAY_SIZE(x86_power_flags))
1314 seq_printf(m, " %s", x86_power_flags[i]);
1315 else
1316 seq_printf(m, " [%d]", i);
1320 seq_printf(m, "\n\n");
1322 return 0;
1325 static void *c_start(struct seq_file *m, loff_t *pos)
1327 return *pos < NR_CPUS ? cpu_data + *pos : NULL;
1330 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
1332 ++*pos;
1333 return c_start(m, pos);
1336 static void c_stop(struct seq_file *m, void *v)
1340 struct seq_operations cpuinfo_op = {
1341 .start =c_start,
1342 .next = c_next,
1343 .stop = c_stop,
1344 .show = show_cpuinfo,