NPTL, round one.
[linux-2.6/verdex.git] / arch / x86_64 / kernel / setup.c
blobcb28df14ff6fea87991daba07241b86835db30ee
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>
65 * Machine setup..
68 struct cpuinfo_x86 boot_cpu_data __read_mostly;
70 unsigned long mmu_cr4_features;
72 int acpi_disabled;
73 EXPORT_SYMBOL(acpi_disabled);
74 #ifdef CONFIG_ACPI
75 extern int __initdata acpi_ht;
76 extern acpi_interrupt_flags acpi_sci_flags;
77 int __initdata acpi_force = 0;
78 #endif
80 int acpi_numa __initdata;
82 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
83 int bootloader_type;
85 unsigned long saved_video_mode;
87 #ifdef CONFIG_SWIOTLB
88 int swiotlb;
89 EXPORT_SYMBOL(swiotlb);
90 #endif
93 * Setup options
95 struct drive_info_struct { char dummy[32]; } drive_info;
96 struct screen_info screen_info;
97 struct sys_desc_table_struct {
98 unsigned short length;
99 unsigned char table[0];
102 struct edid_info edid_info;
103 struct e820map e820;
105 extern int root_mountflags;
106 extern char _text, _etext, _edata, _end;
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 #ifdef CONFIG_ACPI
576 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
577 * Call this early for SRAT node setup.
579 acpi_boot_table_init();
580 #endif
582 #ifdef CONFIG_ACPI_NUMA
584 * Parse SRAT to discover nodes.
586 acpi_numa_init();
587 #endif
589 #ifdef CONFIG_NUMA
590 numa_initmem_init(0, end_pfn);
591 #else
592 contig_initmem_init(0, end_pfn);
593 #endif
595 /* Reserve direct mapping */
596 reserve_bootmem_generic(table_start << PAGE_SHIFT,
597 (table_end - table_start) << PAGE_SHIFT);
599 /* reserve kernel */
600 kernel_end = round_up(__pa_symbol(&_end),PAGE_SIZE);
601 reserve_bootmem_generic(HIGH_MEMORY, kernel_end - HIGH_MEMORY);
604 * reserve physical page 0 - it's a special BIOS page on many boxes,
605 * enabling clean reboots, SMP operation, laptop functions.
607 reserve_bootmem_generic(0, PAGE_SIZE);
609 /* reserve ebda region */
610 reserve_ebda_region();
612 #ifdef CONFIG_SMP
614 * But first pinch a few for the stack/trampoline stuff
615 * FIXME: Don't need the extra page at 4K, but need to fix
616 * trampoline before removing it. (see the GDT stuff)
618 reserve_bootmem_generic(PAGE_SIZE, PAGE_SIZE);
620 /* Reserve SMP trampoline */
621 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, PAGE_SIZE);
622 #endif
624 #ifdef CONFIG_ACPI_SLEEP
626 * Reserve low memory region for sleep support.
628 acpi_reserve_bootmem();
629 #endif
630 #ifdef CONFIG_X86_LOCAL_APIC
632 * Find and reserve possible boot-time SMP configuration:
634 find_smp_config();
635 #endif
636 #ifdef CONFIG_BLK_DEV_INITRD
637 if (LOADER_TYPE && INITRD_START) {
638 if (INITRD_START + INITRD_SIZE <= (end_pfn << PAGE_SHIFT)) {
639 reserve_bootmem_generic(INITRD_START, INITRD_SIZE);
640 initrd_start =
641 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
642 initrd_end = initrd_start+INITRD_SIZE;
644 else {
645 printk(KERN_ERR "initrd extends beyond end of memory "
646 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
647 (unsigned long)(INITRD_START + INITRD_SIZE),
648 (unsigned long)(end_pfn << PAGE_SHIFT));
649 initrd_start = 0;
652 #endif
653 #ifdef CONFIG_KEXEC
654 if (crashk_res.start != crashk_res.end) {
655 reserve_bootmem(crashk_res.start,
656 crashk_res.end - crashk_res.start + 1);
658 #endif
660 sparse_init();
662 paging_init();
664 check_ioapic();
666 #ifdef CONFIG_ACPI
668 * Read APIC and some other early information from ACPI tables.
670 acpi_boot_init();
671 #endif
673 #ifdef CONFIG_X86_LOCAL_APIC
675 * get boot-time SMP configuration:
677 if (smp_found_config)
678 get_smp_config();
679 init_apic_mappings();
680 #endif
683 * Request address space for all standard RAM and ROM resources
684 * and also for regions reported as reserved by the e820.
686 probe_roms();
687 e820_reserve_resources();
689 request_resource(&iomem_resource, &video_ram_resource);
692 unsigned i;
693 /* request I/O space for devices used on all i[345]86 PCs */
694 for (i = 0; i < STANDARD_IO_RESOURCES; i++)
695 request_resource(&ioport_resource, &standard_io_resources[i]);
698 e820_setup_gap();
700 #ifdef CONFIG_GART_IOMMU
701 iommu_hole_init();
702 #endif
704 #ifdef CONFIG_VT
705 #if defined(CONFIG_VGA_CONSOLE)
706 conswitchp = &vga_con;
707 #elif defined(CONFIG_DUMMY_CONSOLE)
708 conswitchp = &dummy_con;
709 #endif
710 #endif
713 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
715 unsigned int *v;
717 if (c->extended_cpuid_level < 0x80000004)
718 return 0;
720 v = (unsigned int *) c->x86_model_id;
721 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
722 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
723 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
724 c->x86_model_id[48] = 0;
725 return 1;
729 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
731 unsigned int n, dummy, eax, ebx, ecx, edx;
733 n = c->extended_cpuid_level;
735 if (n >= 0x80000005) {
736 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
737 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
738 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
739 c->x86_cache_size=(ecx>>24)+(edx>>24);
740 /* On K8 L1 TLB is inclusive, so don't count it */
741 c->x86_tlbsize = 0;
744 if (n >= 0x80000006) {
745 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
746 ecx = cpuid_ecx(0x80000006);
747 c->x86_cache_size = ecx >> 16;
748 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
750 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
751 c->x86_cache_size, ecx & 0xFF);
754 if (n >= 0x80000007)
755 cpuid(0x80000007, &dummy, &dummy, &dummy, &c->x86_power);
756 if (n >= 0x80000008) {
757 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
758 c->x86_virt_bits = (eax >> 8) & 0xff;
759 c->x86_phys_bits = eax & 0xff;
763 #ifdef CONFIG_NUMA
764 static int nearby_node(int apicid)
766 int i;
767 for (i = apicid - 1; i >= 0; i--) {
768 int node = apicid_to_node[i];
769 if (node != NUMA_NO_NODE && node_online(node))
770 return node;
772 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
773 int node = apicid_to_node[i];
774 if (node != NUMA_NO_NODE && node_online(node))
775 return node;
777 return first_node(node_online_map); /* Shouldn't happen */
779 #endif
782 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
783 * Assumes number of cores is a power of two.
785 static void __init amd_detect_cmp(struct cpuinfo_x86 *c)
787 #ifdef CONFIG_SMP
788 int cpu = smp_processor_id();
789 unsigned bits;
790 #ifdef CONFIG_NUMA
791 int node = 0;
792 unsigned apicid = phys_proc_id[cpu];
793 #endif
795 bits = 0;
796 while ((1 << bits) < c->x86_num_cores)
797 bits++;
799 /* Low order bits define the core id (index of core in socket) */
800 cpu_core_id[cpu] = phys_proc_id[cpu] & ((1 << bits)-1);
801 /* Convert the APIC ID into the socket ID */
802 phys_proc_id[cpu] >>= bits;
804 #ifdef CONFIG_NUMA
805 node = phys_proc_id[cpu];
806 if (apicid_to_node[apicid] != NUMA_NO_NODE)
807 node = apicid_to_node[apicid];
808 if (!node_online(node)) {
809 /* Two possibilities here:
810 - The CPU is missing memory and no node was created.
811 In that case try picking one from a nearby CPU
812 - The APIC IDs differ from the HyperTransport node IDs
813 which the K8 northbridge parsing fills in.
814 Assume they are all increased by a constant offset,
815 but in the same order as the HT nodeids.
816 If that doesn't result in a usable node fall back to the
817 path for the previous case. */
818 int ht_nodeid = apicid - (phys_proc_id[0] << bits);
819 if (ht_nodeid >= 0 &&
820 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
821 node = apicid_to_node[ht_nodeid];
822 /* Pick a nearby node */
823 if (!node_online(node))
824 node = nearby_node(apicid);
826 cpu_to_node[cpu] = node;
828 printk(KERN_INFO "CPU %d(%d) -> Node %d -> Core %d\n",
829 cpu, c->x86_num_cores, node, cpu_core_id[cpu]);
830 #endif
831 #endif
834 static int __init init_amd(struct cpuinfo_x86 *c)
836 int r;
837 int level;
839 #ifdef CONFIG_SMP
840 unsigned long value;
843 * Disable TLB flush filter by setting HWCR.FFDIS on K8
844 * bit 6 of msr C001_0015
846 * Errata 63 for SH-B3 steppings
847 * Errata 122 for all steppings (F+ have it disabled by default)
849 if (c->x86 == 15) {
850 rdmsrl(MSR_K8_HWCR, value);
851 value |= 1 << 6;
852 wrmsrl(MSR_K8_HWCR, value);
854 #endif
856 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
857 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
858 clear_bit(0*32+31, &c->x86_capability);
860 /* C-stepping K8? */
861 level = cpuid_eax(1);
862 if ((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58)
863 set_bit(X86_FEATURE_K8_C, &c->x86_capability);
865 r = get_model_name(c);
866 if (!r) {
867 switch (c->x86) {
868 case 15:
869 /* Should distinguish Models here, but this is only
870 a fallback anyways. */
871 strcpy(c->x86_model_id, "Hammer");
872 break;
875 display_cacheinfo(c);
877 if (c->extended_cpuid_level >= 0x80000008) {
878 c->x86_num_cores = (cpuid_ecx(0x80000008) & 0xff) + 1;
879 if (c->x86_num_cores & (c->x86_num_cores - 1))
880 c->x86_num_cores = 1;
882 amd_detect_cmp(c);
885 return r;
888 static void __cpuinit detect_ht(struct cpuinfo_x86 *c)
890 #ifdef CONFIG_SMP
891 u32 eax, ebx, ecx, edx;
892 int index_msb, tmp;
893 int cpu = smp_processor_id();
895 if (!cpu_has(c, X86_FEATURE_HT) || cpu_has(c, X86_FEATURE_CMP_LEGACY))
896 return;
898 cpuid(1, &eax, &ebx, &ecx, &edx);
899 smp_num_siblings = (ebx & 0xff0000) >> 16;
901 if (smp_num_siblings == 1) {
902 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
903 } else if (smp_num_siblings > 1) {
904 index_msb = 31;
906 * At this point we only support two siblings per
907 * processor package.
909 if (smp_num_siblings > NR_CPUS) {
910 printk(KERN_WARNING "CPU: Unsupported number of the siblings %d", smp_num_siblings);
911 smp_num_siblings = 1;
912 return;
914 tmp = smp_num_siblings;
915 while ((tmp & 0x80000000 ) == 0) {
916 tmp <<=1 ;
917 index_msb--;
919 if (smp_num_siblings & (smp_num_siblings - 1))
920 index_msb++;
921 phys_proc_id[cpu] = phys_pkg_id(index_msb);
923 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
924 phys_proc_id[cpu]);
926 smp_num_siblings = smp_num_siblings / c->x86_num_cores;
928 tmp = smp_num_siblings;
929 index_msb = 31;
930 while ((tmp & 0x80000000) == 0) {
931 tmp <<=1 ;
932 index_msb--;
934 if (smp_num_siblings & (smp_num_siblings - 1))
935 index_msb++;
937 cpu_core_id[cpu] = phys_pkg_id(index_msb);
939 if (c->x86_num_cores > 1)
940 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
941 cpu_core_id[cpu]);
943 #endif
947 * find out the number of processor cores on the die
949 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
951 unsigned int eax;
953 if (c->cpuid_level < 4)
954 return 1;
956 __asm__("cpuid"
957 : "=a" (eax)
958 : "0" (4), "c" (0)
959 : "bx", "dx");
961 if (eax & 0x1f)
962 return ((eax >> 26) + 1);
963 else
964 return 1;
967 static void srat_detect_node(void)
969 #ifdef CONFIG_NUMA
970 unsigned node;
971 int cpu = smp_processor_id();
973 /* Don't do the funky fallback heuristics the AMD version employs
974 for now. */
975 node = apicid_to_node[hard_smp_processor_id()];
976 if (node == NUMA_NO_NODE)
977 node = 0;
978 cpu_to_node[cpu] = node;
980 if (acpi_numa > 0)
981 printk(KERN_INFO "CPU %d -> Node %d\n", cpu, node);
982 #endif
985 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
987 /* Cache sizes */
988 unsigned n;
990 init_intel_cacheinfo(c);
991 n = c->extended_cpuid_level;
992 if (n >= 0x80000008) {
993 unsigned eax = cpuid_eax(0x80000008);
994 c->x86_virt_bits = (eax >> 8) & 0xff;
995 c->x86_phys_bits = eax & 0xff;
998 if (c->x86 == 15)
999 c->x86_cache_alignment = c->x86_clflush_size * 2;
1000 if (c->x86 >= 15)
1001 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
1002 c->x86_num_cores = intel_num_cpu_cores(c);
1004 srat_detect_node();
1007 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
1009 char *v = c->x86_vendor_id;
1011 if (!strcmp(v, "AuthenticAMD"))
1012 c->x86_vendor = X86_VENDOR_AMD;
1013 else if (!strcmp(v, "GenuineIntel"))
1014 c->x86_vendor = X86_VENDOR_INTEL;
1015 else
1016 c->x86_vendor = X86_VENDOR_UNKNOWN;
1019 struct cpu_model_info {
1020 int vendor;
1021 int family;
1022 char *model_names[16];
1025 /* Do some early cpuid on the boot CPU to get some parameter that are
1026 needed before check_bugs. Everything advanced is in identify_cpu
1027 below. */
1028 void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
1030 u32 tfms;
1032 c->loops_per_jiffy = loops_per_jiffy;
1033 c->x86_cache_size = -1;
1034 c->x86_vendor = X86_VENDOR_UNKNOWN;
1035 c->x86_model = c->x86_mask = 0; /* So far unknown... */
1036 c->x86_vendor_id[0] = '\0'; /* Unset */
1037 c->x86_model_id[0] = '\0'; /* Unset */
1038 c->x86_clflush_size = 64;
1039 c->x86_cache_alignment = c->x86_clflush_size;
1040 c->x86_num_cores = 1;
1041 c->extended_cpuid_level = 0;
1042 memset(&c->x86_capability, 0, sizeof c->x86_capability);
1044 /* Get vendor name */
1045 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
1046 (unsigned int *)&c->x86_vendor_id[0],
1047 (unsigned int *)&c->x86_vendor_id[8],
1048 (unsigned int *)&c->x86_vendor_id[4]);
1050 get_cpu_vendor(c);
1052 /* Initialize the standard set of capabilities */
1053 /* Note that the vendor-specific code below might override */
1055 /* Intel-defined flags: level 0x00000001 */
1056 if (c->cpuid_level >= 0x00000001) {
1057 __u32 misc;
1058 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
1059 &c->x86_capability[0]);
1060 c->x86 = (tfms >> 8) & 0xf;
1061 c->x86_model = (tfms >> 4) & 0xf;
1062 c->x86_mask = tfms & 0xf;
1063 if (c->x86 == 0xf) {
1064 c->x86 += (tfms >> 20) & 0xff;
1065 c->x86_model += ((tfms >> 16) & 0xF) << 4;
1067 if (c->x86_capability[0] & (1<<19))
1068 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1069 } else {
1070 /* Have CPUID level 0 only - unheard of */
1071 c->x86 = 4;
1074 #ifdef CONFIG_SMP
1075 phys_proc_id[smp_processor_id()] = (cpuid_ebx(1) >> 24) & 0xff;
1076 #endif
1080 * This does the hard work of actually picking apart the CPU stuff...
1082 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1084 int i;
1085 u32 xlvl;
1087 early_identify_cpu(c);
1089 /* AMD-defined flags: level 0x80000001 */
1090 xlvl = cpuid_eax(0x80000000);
1091 c->extended_cpuid_level = xlvl;
1092 if ((xlvl & 0xffff0000) == 0x80000000) {
1093 if (xlvl >= 0x80000001) {
1094 c->x86_capability[1] = cpuid_edx(0x80000001);
1095 c->x86_capability[6] = cpuid_ecx(0x80000001);
1097 if (xlvl >= 0x80000004)
1098 get_model_name(c); /* Default name */
1101 /* Transmeta-defined flags: level 0x80860001 */
1102 xlvl = cpuid_eax(0x80860000);
1103 if ((xlvl & 0xffff0000) == 0x80860000) {
1104 /* Don't set x86_cpuid_level here for now to not confuse. */
1105 if (xlvl >= 0x80860001)
1106 c->x86_capability[2] = cpuid_edx(0x80860001);
1110 * Vendor-specific initialization. In this section we
1111 * canonicalize the feature flags, meaning if there are
1112 * features a certain CPU supports which CPUID doesn't
1113 * tell us, CPUID claiming incorrect flags, or other bugs,
1114 * we handle them here.
1116 * At the end of this section, c->x86_capability better
1117 * indicate the features this CPU genuinely supports!
1119 switch (c->x86_vendor) {
1120 case X86_VENDOR_AMD:
1121 init_amd(c);
1122 break;
1124 case X86_VENDOR_INTEL:
1125 init_intel(c);
1126 break;
1128 case X86_VENDOR_UNKNOWN:
1129 default:
1130 display_cacheinfo(c);
1131 break;
1134 select_idle_routine(c);
1135 detect_ht(c);
1138 * On SMP, boot_cpu_data holds the common feature set between
1139 * all CPUs; so make sure that we indicate which features are
1140 * common between the CPUs. The first time this routine gets
1141 * executed, c == &boot_cpu_data.
1143 if (c != &boot_cpu_data) {
1144 /* AND the already accumulated flags with these */
1145 for (i = 0 ; i < NCAPINTS ; i++)
1146 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1149 #ifdef CONFIG_X86_MCE
1150 mcheck_init(c);
1151 #endif
1152 if (c == &boot_cpu_data)
1153 mtrr_bp_init();
1154 else
1155 mtrr_ap_init();
1156 #ifdef CONFIG_NUMA
1157 numa_add_cpu(smp_processor_id());
1158 #endif
1162 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1164 if (c->x86_model_id[0])
1165 printk("%s", c->x86_model_id);
1167 if (c->x86_mask || c->cpuid_level >= 0)
1168 printk(" stepping %02x\n", c->x86_mask);
1169 else
1170 printk("\n");
1174 * Get CPU information for use by the procfs.
1177 static int show_cpuinfo(struct seq_file *m, void *v)
1179 struct cpuinfo_x86 *c = v;
1182 * These flag bits must match the definitions in <asm/cpufeature.h>.
1183 * NULL means this bit is undefined or reserved; either way it doesn't
1184 * have meaning as far as Linux is concerned. Note that it's important
1185 * to realize there is a difference between this table and CPUID -- if
1186 * applications want to get the raw CPUID data, they should access
1187 * /dev/cpu/<cpu_nr>/cpuid instead.
1189 static char *x86_cap_flags[] = {
1190 /* Intel-defined */
1191 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1192 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
1193 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
1194 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", NULL,
1196 /* AMD-defined */
1197 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1198 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
1199 NULL, NULL, NULL, NULL, "nx", NULL, "mmxext", NULL,
1200 NULL, "fxsr_opt", NULL, NULL, NULL, "lm", "3dnowext", "3dnow",
1202 /* Transmeta-defined */
1203 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
1204 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1205 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1206 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1208 /* Other (Linux-defined) */
1209 "cxmmx", NULL, "cyrix_arr", "centaur_mcr", NULL,
1210 "constant_tsc", NULL, NULL,
1211 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1212 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1213 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1215 /* Intel-defined (#2) */
1216 "pni", NULL, NULL, "monitor", "ds_cpl", NULL, NULL, "est",
1217 "tm2", NULL, "cid", NULL, NULL, "cx16", "xtpr", NULL,
1218 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1219 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1221 /* VIA/Cyrix/Centaur-defined */
1222 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
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 /* AMD-defined (#2) */
1228 "lahf_lm", "cmp_legacy", NULL, NULL, NULL, NULL, NULL, NULL,
1229 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1230 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1231 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1233 static char *x86_power_flags[] = {
1234 "ts", /* temperature sensor */
1235 "fid", /* frequency id control */
1236 "vid", /* voltage id control */
1237 "ttp", /* thermal trip */
1238 "tm",
1239 "stc"
1243 #ifdef CONFIG_SMP
1244 if (!cpu_online(c-cpu_data))
1245 return 0;
1246 #endif
1248 seq_printf(m,"processor\t: %u\n"
1249 "vendor_id\t: %s\n"
1250 "cpu family\t: %d\n"
1251 "model\t\t: %d\n"
1252 "model name\t: %s\n",
1253 (unsigned)(c-cpu_data),
1254 c->x86_vendor_id[0] ? c->x86_vendor_id : "unknown",
1255 c->x86,
1256 (int)c->x86_model,
1257 c->x86_model_id[0] ? c->x86_model_id : "unknown");
1259 if (c->x86_mask || c->cpuid_level >= 0)
1260 seq_printf(m, "stepping\t: %d\n", c->x86_mask);
1261 else
1262 seq_printf(m, "stepping\t: unknown\n");
1264 if (cpu_has(c,X86_FEATURE_TSC)) {
1265 seq_printf(m, "cpu MHz\t\t: %u.%03u\n",
1266 cpu_khz / 1000, (cpu_khz % 1000));
1269 /* Cache size */
1270 if (c->x86_cache_size >= 0)
1271 seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
1273 #ifdef CONFIG_SMP
1274 if (smp_num_siblings * c->x86_num_cores > 1) {
1275 int cpu = c - cpu_data;
1276 seq_printf(m, "physical id\t: %d\n", phys_proc_id[cpu]);
1277 seq_printf(m, "siblings\t: %d\n",
1278 c->x86_num_cores * smp_num_siblings);
1279 seq_printf(m, "core id\t\t: %d\n", cpu_core_id[cpu]);
1280 seq_printf(m, "cpu cores\t: %d\n", c->x86_num_cores);
1282 #endif
1284 seq_printf(m,
1285 "fpu\t\t: yes\n"
1286 "fpu_exception\t: yes\n"
1287 "cpuid level\t: %d\n"
1288 "wp\t\t: yes\n"
1289 "flags\t\t:",
1290 c->cpuid_level);
1293 int i;
1294 for ( i = 0 ; i < 32*NCAPINTS ; i++ )
1295 if ( test_bit(i, &c->x86_capability) &&
1296 x86_cap_flags[i] != NULL )
1297 seq_printf(m, " %s", x86_cap_flags[i]);
1300 seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
1301 c->loops_per_jiffy/(500000/HZ),
1302 (c->loops_per_jiffy/(5000/HZ)) % 100);
1304 if (c->x86_tlbsize > 0)
1305 seq_printf(m, "TLB size\t: %d 4K pages\n", c->x86_tlbsize);
1306 seq_printf(m, "clflush size\t: %d\n", c->x86_clflush_size);
1307 seq_printf(m, "cache_alignment\t: %d\n", c->x86_cache_alignment);
1309 seq_printf(m, "address sizes\t: %u bits physical, %u bits virtual\n",
1310 c->x86_phys_bits, c->x86_virt_bits);
1312 seq_printf(m, "power management:");
1314 unsigned i;
1315 for (i = 0; i < 32; i++)
1316 if (c->x86_power & (1 << i)) {
1317 if (i < ARRAY_SIZE(x86_power_flags))
1318 seq_printf(m, " %s", x86_power_flags[i]);
1319 else
1320 seq_printf(m, " [%d]", i);
1324 seq_printf(m, "\n\n");
1326 return 0;
1329 static void *c_start(struct seq_file *m, loff_t *pos)
1331 return *pos < NR_CPUS ? cpu_data + *pos : NULL;
1334 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
1336 ++*pos;
1337 return c_start(m, pos);
1340 static void c_stop(struct seq_file *m, void *v)
1344 struct seq_operations cpuinfo_op = {
1345 .start =c_start,
1346 .next = c_next,
1347 .stop = c_stop,
1348 .show = show_cpuinfo,