[PATCH] i386: add memory clobbers to syscall macros
[linux-2.6/verdex.git] / arch / x86_64 / kernel / setup.c
blob869770da2933194d7fe5a3e2b4bc6084f2bff397
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, "nolapic", 7) ||
340 !memcmp(from, "disableapic", 11))
341 disable_apic = 1;
343 if (!memcmp(from, "noapic", 6))
344 skip_ioapic_setup = 1;
346 if (!memcmp(from, "apic", 4)) {
347 skip_ioapic_setup = 0;
348 ioapic_force = 1;
351 if (!memcmp(from, "mem=", 4))
352 parse_memopt(from+4, &from);
354 #ifdef CONFIG_NUMA
355 if (!memcmp(from, "numa=", 5))
356 numa_setup(from+5);
357 #endif
359 #ifdef CONFIG_GART_IOMMU
360 if (!memcmp(from,"iommu=",6)) {
361 iommu_setup(from+6);
363 #endif
365 if (!memcmp(from,"oops=panic", 10))
366 panic_on_oops = 1;
368 if (!memcmp(from, "noexec=", 7))
369 nonx_setup(from + 7);
371 #ifdef CONFIG_KEXEC
372 /* crashkernel=size@addr specifies the location to reserve for
373 * a crash kernel. By reserving this memory we guarantee
374 * that linux never set's it up as a DMA target.
375 * Useful for holding code to do something appropriate
376 * after a kernel panic.
378 else if (!memcmp(from, "crashkernel=", 12)) {
379 unsigned long size, base;
380 size = memparse(from+12, &from);
381 if (*from == '@') {
382 base = memparse(from+1, &from);
383 /* FIXME: Do I want a sanity check
384 * to validate the memory range?
386 crashk_res.start = base;
387 crashk_res.end = base + size - 1;
390 #endif
392 next_char:
393 c = *(from++);
394 if (!c)
395 break;
396 if (COMMAND_LINE_SIZE <= ++len)
397 break;
398 *(to++) = c;
400 *to = '\0';
401 *cmdline_p = command_line;
404 #ifndef CONFIG_NUMA
405 static void __init
406 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
408 unsigned long bootmap_size, bootmap;
410 memory_present(0, start_pfn, end_pfn);
411 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
412 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size);
413 if (bootmap == -1L)
414 panic("Cannot find bootmem map of size %ld\n",bootmap_size);
415 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
416 e820_bootmem_free(NODE_DATA(0), 0, end_pfn << PAGE_SHIFT);
417 reserve_bootmem(bootmap, bootmap_size);
419 #endif
421 /* Use inline assembly to define this because the nops are defined
422 as inline assembly strings in the include files and we cannot
423 get them easily into strings. */
424 asm("\t.data\nk8nops: "
425 K8_NOP1 K8_NOP2 K8_NOP3 K8_NOP4 K8_NOP5 K8_NOP6
426 K8_NOP7 K8_NOP8);
428 extern unsigned char k8nops[];
429 static unsigned char *k8_nops[ASM_NOP_MAX+1] = {
430 NULL,
431 k8nops,
432 k8nops + 1,
433 k8nops + 1 + 2,
434 k8nops + 1 + 2 + 3,
435 k8nops + 1 + 2 + 3 + 4,
436 k8nops + 1 + 2 + 3 + 4 + 5,
437 k8nops + 1 + 2 + 3 + 4 + 5 + 6,
438 k8nops + 1 + 2 + 3 + 4 + 5 + 6 + 7,
441 /* Replace instructions with better alternatives for this CPU type.
443 This runs before SMP is initialized to avoid SMP problems with
444 self modifying code. This implies that assymetric systems where
445 APs have less capabilities than the boot processor are not handled.
446 In this case boot with "noreplacement". */
447 void apply_alternatives(void *start, void *end)
449 struct alt_instr *a;
450 int diff, i, k;
451 for (a = start; (void *)a < end; a++) {
452 if (!boot_cpu_has(a->cpuid))
453 continue;
455 BUG_ON(a->replacementlen > a->instrlen);
456 __inline_memcpy(a->instr, a->replacement, a->replacementlen);
457 diff = a->instrlen - a->replacementlen;
459 /* Pad the rest with nops */
460 for (i = a->replacementlen; diff > 0; diff -= k, i += k) {
461 k = diff;
462 if (k > ASM_NOP_MAX)
463 k = ASM_NOP_MAX;
464 __inline_memcpy(a->instr + i, k8_nops[k], k);
469 static int no_replacement __initdata = 0;
471 void __init alternative_instructions(void)
473 extern struct alt_instr __alt_instructions[], __alt_instructions_end[];
474 if (no_replacement)
475 return;
476 apply_alternatives(__alt_instructions, __alt_instructions_end);
479 static int __init noreplacement_setup(char *s)
481 no_replacement = 1;
482 return 0;
485 __setup("noreplacement", noreplacement_setup);
487 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
488 struct edd edd;
489 #ifdef CONFIG_EDD_MODULE
490 EXPORT_SYMBOL(edd);
491 #endif
493 * copy_edd() - Copy the BIOS EDD information
494 * from boot_params into a safe place.
497 static inline void copy_edd(void)
499 memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
500 memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
501 edd.mbr_signature_nr = EDD_MBR_SIG_NR;
502 edd.edd_info_nr = EDD_NR;
504 #else
505 static inline void copy_edd(void)
508 #endif
510 #define EBDA_ADDR_POINTER 0x40E
511 static void __init reserve_ebda_region(void)
513 unsigned int addr;
514 /**
515 * there is a real-mode segmented pointer pointing to the
516 * 4K EBDA area at 0x40E
518 addr = *(unsigned short *)phys_to_virt(EBDA_ADDR_POINTER);
519 addr <<= 4;
520 if (addr)
521 reserve_bootmem_generic(addr, PAGE_SIZE);
524 void __init setup_arch(char **cmdline_p)
526 unsigned long kernel_end;
528 ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
529 drive_info = DRIVE_INFO;
530 screen_info = SCREEN_INFO;
531 edid_info = EDID_INFO;
532 saved_video_mode = SAVED_VIDEO_MODE;
533 bootloader_type = LOADER_TYPE;
535 #ifdef CONFIG_BLK_DEV_RAM
536 rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
537 rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
538 rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
539 #endif
540 setup_memory_region();
541 copy_edd();
543 if (!MOUNT_ROOT_RDONLY)
544 root_mountflags &= ~MS_RDONLY;
545 init_mm.start_code = (unsigned long) &_text;
546 init_mm.end_code = (unsigned long) &_etext;
547 init_mm.end_data = (unsigned long) &_edata;
548 init_mm.brk = (unsigned long) &_end;
550 code_resource.start = virt_to_phys(&_text);
551 code_resource.end = virt_to_phys(&_etext)-1;
552 data_resource.start = virt_to_phys(&_etext);
553 data_resource.end = virt_to_phys(&_edata)-1;
555 parse_cmdline_early(cmdline_p);
557 early_identify_cpu(&boot_cpu_data);
560 * partially used pages are not usable - thus
561 * we are rounding upwards:
563 end_pfn = e820_end_of_ram();
565 check_efer();
567 init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
569 #ifdef CONFIG_ACPI
571 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
572 * Call this early for SRAT node setup.
574 acpi_boot_table_init();
575 #endif
577 #ifdef CONFIG_ACPI_NUMA
579 * Parse SRAT to discover nodes.
581 acpi_numa_init();
582 #endif
584 #ifdef CONFIG_NUMA
585 numa_initmem_init(0, end_pfn);
586 #else
587 contig_initmem_init(0, end_pfn);
588 #endif
590 /* Reserve direct mapping */
591 reserve_bootmem_generic(table_start << PAGE_SHIFT,
592 (table_end - table_start) << PAGE_SHIFT);
594 /* reserve kernel */
595 kernel_end = round_up(__pa_symbol(&_end),PAGE_SIZE);
596 reserve_bootmem_generic(HIGH_MEMORY, kernel_end - HIGH_MEMORY);
599 * reserve physical page 0 - it's a special BIOS page on many boxes,
600 * enabling clean reboots, SMP operation, laptop functions.
602 reserve_bootmem_generic(0, PAGE_SIZE);
604 /* reserve ebda region */
605 reserve_ebda_region();
607 #ifdef CONFIG_SMP
609 * But first pinch a few for the stack/trampoline stuff
610 * FIXME: Don't need the extra page at 4K, but need to fix
611 * trampoline before removing it. (see the GDT stuff)
613 reserve_bootmem_generic(PAGE_SIZE, PAGE_SIZE);
615 /* Reserve SMP trampoline */
616 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, PAGE_SIZE);
617 #endif
619 #ifdef CONFIG_ACPI_SLEEP
621 * Reserve low memory region for sleep support.
623 acpi_reserve_bootmem();
624 #endif
625 #ifdef CONFIG_X86_LOCAL_APIC
627 * Find and reserve possible boot-time SMP configuration:
629 find_smp_config();
630 #endif
631 #ifdef CONFIG_BLK_DEV_INITRD
632 if (LOADER_TYPE && INITRD_START) {
633 if (INITRD_START + INITRD_SIZE <= (end_pfn << PAGE_SHIFT)) {
634 reserve_bootmem_generic(INITRD_START, INITRD_SIZE);
635 initrd_start =
636 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
637 initrd_end = initrd_start+INITRD_SIZE;
639 else {
640 printk(KERN_ERR "initrd extends beyond end of memory "
641 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
642 (unsigned long)(INITRD_START + INITRD_SIZE),
643 (unsigned long)(end_pfn << PAGE_SHIFT));
644 initrd_start = 0;
647 #endif
648 #ifdef CONFIG_KEXEC
649 if (crashk_res.start != crashk_res.end) {
650 reserve_bootmem(crashk_res.start,
651 crashk_res.end - crashk_res.start + 1);
653 #endif
655 sparse_init();
657 paging_init();
659 check_ioapic();
661 #ifdef CONFIG_ACPI
663 * Read APIC and some other early information from ACPI tables.
665 acpi_boot_init();
666 #endif
668 #ifdef CONFIG_X86_LOCAL_APIC
670 * get boot-time SMP configuration:
672 if (smp_found_config)
673 get_smp_config();
674 init_apic_mappings();
675 #endif
678 * Request address space for all standard RAM and ROM resources
679 * and also for regions reported as reserved by the e820.
681 probe_roms();
682 e820_reserve_resources();
684 request_resource(&iomem_resource, &video_ram_resource);
687 unsigned i;
688 /* request I/O space for devices used on all i[345]86 PCs */
689 for (i = 0; i < STANDARD_IO_RESOURCES; i++)
690 request_resource(&ioport_resource, &standard_io_resources[i]);
693 e820_setup_gap();
695 #ifdef CONFIG_GART_IOMMU
696 iommu_hole_init();
697 #endif
699 #ifdef CONFIG_VT
700 #if defined(CONFIG_VGA_CONSOLE)
701 conswitchp = &vga_con;
702 #elif defined(CONFIG_DUMMY_CONSOLE)
703 conswitchp = &dummy_con;
704 #endif
705 #endif
708 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
710 unsigned int *v;
712 if (c->extended_cpuid_level < 0x80000004)
713 return 0;
715 v = (unsigned int *) c->x86_model_id;
716 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
717 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
718 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
719 c->x86_model_id[48] = 0;
720 return 1;
724 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
726 unsigned int n, dummy, eax, ebx, ecx, edx;
728 n = c->extended_cpuid_level;
730 if (n >= 0x80000005) {
731 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
732 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
733 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
734 c->x86_cache_size=(ecx>>24)+(edx>>24);
735 /* On K8 L1 TLB is inclusive, so don't count it */
736 c->x86_tlbsize = 0;
739 if (n >= 0x80000006) {
740 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
741 ecx = cpuid_ecx(0x80000006);
742 c->x86_cache_size = ecx >> 16;
743 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
745 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
746 c->x86_cache_size, ecx & 0xFF);
749 if (n >= 0x80000007)
750 cpuid(0x80000007, &dummy, &dummy, &dummy, &c->x86_power);
751 if (n >= 0x80000008) {
752 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
753 c->x86_virt_bits = (eax >> 8) & 0xff;
754 c->x86_phys_bits = eax & 0xff;
758 #ifdef CONFIG_NUMA
759 static int nearby_node(int apicid)
761 int i;
762 for (i = apicid - 1; i >= 0; i--) {
763 int node = apicid_to_node[i];
764 if (node != NUMA_NO_NODE && node_online(node))
765 return node;
767 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
768 int node = apicid_to_node[i];
769 if (node != NUMA_NO_NODE && node_online(node))
770 return node;
772 return first_node(node_online_map); /* Shouldn't happen */
774 #endif
777 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
778 * Assumes number of cores is a power of two.
780 static void __init amd_detect_cmp(struct cpuinfo_x86 *c)
782 #ifdef CONFIG_SMP
783 int cpu = smp_processor_id();
784 unsigned bits;
785 #ifdef CONFIG_NUMA
786 int node = 0;
787 unsigned apicid = phys_proc_id[cpu];
788 #endif
790 bits = 0;
791 while ((1 << bits) < c->x86_num_cores)
792 bits++;
794 /* Low order bits define the core id (index of core in socket) */
795 cpu_core_id[cpu] = phys_proc_id[cpu] & ((1 << bits)-1);
796 /* Convert the APIC ID into the socket ID */
797 phys_proc_id[cpu] >>= bits;
799 #ifdef CONFIG_NUMA
800 node = phys_proc_id[cpu];
801 if (apicid_to_node[apicid] != NUMA_NO_NODE)
802 node = apicid_to_node[apicid];
803 if (!node_online(node)) {
804 /* Two possibilities here:
805 - The CPU is missing memory and no node was created.
806 In that case try picking one from a nearby CPU
807 - The APIC IDs differ from the HyperTransport node IDs
808 which the K8 northbridge parsing fills in.
809 Assume they are all increased by a constant offset,
810 but in the same order as the HT nodeids.
811 If that doesn't result in a usable node fall back to the
812 path for the previous case. */
813 int ht_nodeid = apicid - (phys_proc_id[0] << bits);
814 if (ht_nodeid >= 0 &&
815 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
816 node = apicid_to_node[ht_nodeid];
817 /* Pick a nearby node */
818 if (!node_online(node))
819 node = nearby_node(apicid);
821 cpu_to_node[cpu] = node;
823 printk(KERN_INFO "CPU %d(%d) -> Node %d -> Core %d\n",
824 cpu, c->x86_num_cores, node, cpu_core_id[cpu]);
825 #endif
826 #endif
829 static int __init init_amd(struct cpuinfo_x86 *c)
831 int r;
832 int level;
834 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
835 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
836 clear_bit(0*32+31, &c->x86_capability);
838 /* C-stepping K8? */
839 level = cpuid_eax(1);
840 if ((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58)
841 set_bit(X86_FEATURE_K8_C, &c->x86_capability);
843 r = get_model_name(c);
844 if (!r) {
845 switch (c->x86) {
846 case 15:
847 /* Should distinguish Models here, but this is only
848 a fallback anyways. */
849 strcpy(c->x86_model_id, "Hammer");
850 break;
853 display_cacheinfo(c);
855 if (c->extended_cpuid_level >= 0x80000008) {
856 c->x86_num_cores = (cpuid_ecx(0x80000008) & 0xff) + 1;
857 if (c->x86_num_cores & (c->x86_num_cores - 1))
858 c->x86_num_cores = 1;
860 amd_detect_cmp(c);
863 return r;
866 static void __cpuinit detect_ht(struct cpuinfo_x86 *c)
868 #ifdef CONFIG_SMP
869 u32 eax, ebx, ecx, edx;
870 int index_msb, tmp;
871 int cpu = smp_processor_id();
873 if (!cpu_has(c, X86_FEATURE_HT) || cpu_has(c, X86_FEATURE_CMP_LEGACY))
874 return;
876 cpuid(1, &eax, &ebx, &ecx, &edx);
877 smp_num_siblings = (ebx & 0xff0000) >> 16;
879 if (smp_num_siblings == 1) {
880 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
881 } else if (smp_num_siblings > 1) {
882 index_msb = 31;
884 * At this point we only support two siblings per
885 * processor package.
887 if (smp_num_siblings > NR_CPUS) {
888 printk(KERN_WARNING "CPU: Unsupported number of the siblings %d", smp_num_siblings);
889 smp_num_siblings = 1;
890 return;
892 tmp = smp_num_siblings;
893 while ((tmp & 0x80000000 ) == 0) {
894 tmp <<=1 ;
895 index_msb--;
897 if (smp_num_siblings & (smp_num_siblings - 1))
898 index_msb++;
899 phys_proc_id[cpu] = phys_pkg_id(index_msb);
901 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
902 phys_proc_id[cpu]);
904 smp_num_siblings = smp_num_siblings / c->x86_num_cores;
906 tmp = smp_num_siblings;
907 index_msb = 31;
908 while ((tmp & 0x80000000) == 0) {
909 tmp <<=1 ;
910 index_msb--;
912 if (smp_num_siblings & (smp_num_siblings - 1))
913 index_msb++;
915 cpu_core_id[cpu] = phys_pkg_id(index_msb);
917 if (c->x86_num_cores > 1)
918 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
919 cpu_core_id[cpu]);
921 #endif
925 * find out the number of processor cores on the die
927 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
929 unsigned int eax;
931 if (c->cpuid_level < 4)
932 return 1;
934 __asm__("cpuid"
935 : "=a" (eax)
936 : "0" (4), "c" (0)
937 : "bx", "dx");
939 if (eax & 0x1f)
940 return ((eax >> 26) + 1);
941 else
942 return 1;
945 static void srat_detect_node(void)
947 #ifdef CONFIG_NUMA
948 unsigned apicid, node;
949 int cpu = smp_processor_id();
951 /* Don't do the funky fallback heuristics the AMD version employs
952 for now. */
953 apicid = phys_proc_id[cpu];
954 node = apicid_to_node[apicid];
955 if (node == NUMA_NO_NODE)
956 node = 0;
957 cpu_to_node[cpu] = node;
959 if (acpi_numa > 0)
960 printk(KERN_INFO "CPU %d -> Node %d\n", cpu, node);
961 #endif
964 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
966 /* Cache sizes */
967 unsigned n;
969 init_intel_cacheinfo(c);
970 n = c->extended_cpuid_level;
971 if (n >= 0x80000008) {
972 unsigned eax = cpuid_eax(0x80000008);
973 c->x86_virt_bits = (eax >> 8) & 0xff;
974 c->x86_phys_bits = eax & 0xff;
977 if (c->x86 == 15)
978 c->x86_cache_alignment = c->x86_clflush_size * 2;
979 if (c->x86 >= 15)
980 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
981 c->x86_num_cores = intel_num_cpu_cores(c);
983 srat_detect_node();
986 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
988 char *v = c->x86_vendor_id;
990 if (!strcmp(v, "AuthenticAMD"))
991 c->x86_vendor = X86_VENDOR_AMD;
992 else if (!strcmp(v, "GenuineIntel"))
993 c->x86_vendor = X86_VENDOR_INTEL;
994 else
995 c->x86_vendor = X86_VENDOR_UNKNOWN;
998 struct cpu_model_info {
999 int vendor;
1000 int family;
1001 char *model_names[16];
1004 /* Do some early cpuid on the boot CPU to get some parameter that are
1005 needed before check_bugs. Everything advanced is in identify_cpu
1006 below. */
1007 void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
1009 u32 tfms;
1011 c->loops_per_jiffy = loops_per_jiffy;
1012 c->x86_cache_size = -1;
1013 c->x86_vendor = X86_VENDOR_UNKNOWN;
1014 c->x86_model = c->x86_mask = 0; /* So far unknown... */
1015 c->x86_vendor_id[0] = '\0'; /* Unset */
1016 c->x86_model_id[0] = '\0'; /* Unset */
1017 c->x86_clflush_size = 64;
1018 c->x86_cache_alignment = c->x86_clflush_size;
1019 c->x86_num_cores = 1;
1020 c->extended_cpuid_level = 0;
1021 memset(&c->x86_capability, 0, sizeof c->x86_capability);
1023 /* Get vendor name */
1024 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
1025 (unsigned int *)&c->x86_vendor_id[0],
1026 (unsigned int *)&c->x86_vendor_id[8],
1027 (unsigned int *)&c->x86_vendor_id[4]);
1029 get_cpu_vendor(c);
1031 /* Initialize the standard set of capabilities */
1032 /* Note that the vendor-specific code below might override */
1034 /* Intel-defined flags: level 0x00000001 */
1035 if (c->cpuid_level >= 0x00000001) {
1036 __u32 misc;
1037 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
1038 &c->x86_capability[0]);
1039 c->x86 = (tfms >> 8) & 0xf;
1040 c->x86_model = (tfms >> 4) & 0xf;
1041 c->x86_mask = tfms & 0xf;
1042 if (c->x86 == 0xf) {
1043 c->x86 += (tfms >> 20) & 0xff;
1044 c->x86_model += ((tfms >> 16) & 0xF) << 4;
1046 if (c->x86_capability[0] & (1<<19))
1047 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1048 } else {
1049 /* Have CPUID level 0 only - unheard of */
1050 c->x86 = 4;
1053 #ifdef CONFIG_SMP
1054 phys_proc_id[smp_processor_id()] = (cpuid_ebx(1) >> 24) & 0xff;
1055 #endif
1059 * This does the hard work of actually picking apart the CPU stuff...
1061 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1063 int i;
1064 u32 xlvl;
1066 early_identify_cpu(c);
1068 /* AMD-defined flags: level 0x80000001 */
1069 xlvl = cpuid_eax(0x80000000);
1070 c->extended_cpuid_level = xlvl;
1071 if ((xlvl & 0xffff0000) == 0x80000000) {
1072 if (xlvl >= 0x80000001) {
1073 c->x86_capability[1] = cpuid_edx(0x80000001);
1074 c->x86_capability[6] = cpuid_ecx(0x80000001);
1076 if (xlvl >= 0x80000004)
1077 get_model_name(c); /* Default name */
1080 /* Transmeta-defined flags: level 0x80860001 */
1081 xlvl = cpuid_eax(0x80860000);
1082 if ((xlvl & 0xffff0000) == 0x80860000) {
1083 /* Don't set x86_cpuid_level here for now to not confuse. */
1084 if (xlvl >= 0x80860001)
1085 c->x86_capability[2] = cpuid_edx(0x80860001);
1089 * Vendor-specific initialization. In this section we
1090 * canonicalize the feature flags, meaning if there are
1091 * features a certain CPU supports which CPUID doesn't
1092 * tell us, CPUID claiming incorrect flags, or other bugs,
1093 * we handle them here.
1095 * At the end of this section, c->x86_capability better
1096 * indicate the features this CPU genuinely supports!
1098 switch (c->x86_vendor) {
1099 case X86_VENDOR_AMD:
1100 init_amd(c);
1101 break;
1103 case X86_VENDOR_INTEL:
1104 init_intel(c);
1105 break;
1107 case X86_VENDOR_UNKNOWN:
1108 default:
1109 display_cacheinfo(c);
1110 break;
1113 select_idle_routine(c);
1114 detect_ht(c);
1117 * On SMP, boot_cpu_data holds the common feature set between
1118 * all CPUs; so make sure that we indicate which features are
1119 * common between the CPUs. The first time this routine gets
1120 * executed, c == &boot_cpu_data.
1122 if (c != &boot_cpu_data) {
1123 /* AND the already accumulated flags with these */
1124 for (i = 0 ; i < NCAPINTS ; i++)
1125 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1128 #ifdef CONFIG_X86_MCE
1129 mcheck_init(c);
1130 #endif
1131 if (c == &boot_cpu_data)
1132 mtrr_bp_init();
1133 else
1134 mtrr_ap_init();
1135 #ifdef CONFIG_NUMA
1136 numa_add_cpu(smp_processor_id());
1137 #endif
1141 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1143 if (c->x86_model_id[0])
1144 printk("%s", c->x86_model_id);
1146 if (c->x86_mask || c->cpuid_level >= 0)
1147 printk(" stepping %02x\n", c->x86_mask);
1148 else
1149 printk("\n");
1153 * Get CPU information for use by the procfs.
1156 static int show_cpuinfo(struct seq_file *m, void *v)
1158 struct cpuinfo_x86 *c = v;
1161 * These flag bits must match the definitions in <asm/cpufeature.h>.
1162 * NULL means this bit is undefined or reserved; either way it doesn't
1163 * have meaning as far as Linux is concerned. Note that it's important
1164 * to realize there is a difference between this table and CPUID -- if
1165 * applications want to get the raw CPUID data, they should access
1166 * /dev/cpu/<cpu_nr>/cpuid instead.
1168 static char *x86_cap_flags[] = {
1169 /* Intel-defined */
1170 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1171 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
1172 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
1173 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", NULL,
1175 /* AMD-defined */
1176 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1177 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
1178 NULL, NULL, NULL, NULL, "nx", NULL, "mmxext", NULL,
1179 NULL, "fxsr_opt", NULL, NULL, NULL, "lm", "3dnowext", "3dnow",
1181 /* Transmeta-defined */
1182 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
1183 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1184 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1185 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1187 /* Other (Linux-defined) */
1188 "cxmmx", NULL, "cyrix_arr", "centaur_mcr", NULL,
1189 "constant_tsc", NULL, NULL,
1190 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1191 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1192 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1194 /* Intel-defined (#2) */
1195 "pni", NULL, NULL, "monitor", "ds_cpl", NULL, NULL, "est",
1196 "tm2", NULL, "cid", NULL, NULL, "cx16", "xtpr", NULL,
1197 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1198 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1200 /* VIA/Cyrix/Centaur-defined */
1201 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
1202 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1203 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1204 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1206 /* AMD-defined (#2) */
1207 "lahf_lm", "cmp_legacy", NULL, NULL, NULL, NULL, 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 static char *x86_power_flags[] = {
1213 "ts", /* temperature sensor */
1214 "fid", /* frequency id control */
1215 "vid", /* voltage id control */
1216 "ttp", /* thermal trip */
1217 "tm",
1218 "stc"
1222 #ifdef CONFIG_SMP
1223 if (!cpu_online(c-cpu_data))
1224 return 0;
1225 #endif
1227 seq_printf(m,"processor\t: %u\n"
1228 "vendor_id\t: %s\n"
1229 "cpu family\t: %d\n"
1230 "model\t\t: %d\n"
1231 "model name\t: %s\n",
1232 (unsigned)(c-cpu_data),
1233 c->x86_vendor_id[0] ? c->x86_vendor_id : "unknown",
1234 c->x86,
1235 (int)c->x86_model,
1236 c->x86_model_id[0] ? c->x86_model_id : "unknown");
1238 if (c->x86_mask || c->cpuid_level >= 0)
1239 seq_printf(m, "stepping\t: %d\n", c->x86_mask);
1240 else
1241 seq_printf(m, "stepping\t: unknown\n");
1243 if (cpu_has(c,X86_FEATURE_TSC)) {
1244 seq_printf(m, "cpu MHz\t\t: %u.%03u\n",
1245 cpu_khz / 1000, (cpu_khz % 1000));
1248 /* Cache size */
1249 if (c->x86_cache_size >= 0)
1250 seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
1252 #ifdef CONFIG_SMP
1253 if (smp_num_siblings * c->x86_num_cores > 1) {
1254 int cpu = c - cpu_data;
1255 seq_printf(m, "physical id\t: %d\n", phys_proc_id[cpu]);
1256 seq_printf(m, "siblings\t: %d\n",
1257 c->x86_num_cores * smp_num_siblings);
1258 seq_printf(m, "core id\t\t: %d\n", cpu_core_id[cpu]);
1259 seq_printf(m, "cpu cores\t: %d\n", c->x86_num_cores);
1261 #endif
1263 seq_printf(m,
1264 "fpu\t\t: yes\n"
1265 "fpu_exception\t: yes\n"
1266 "cpuid level\t: %d\n"
1267 "wp\t\t: yes\n"
1268 "flags\t\t:",
1269 c->cpuid_level);
1272 int i;
1273 for ( i = 0 ; i < 32*NCAPINTS ; i++ )
1274 if ( test_bit(i, &c->x86_capability) &&
1275 x86_cap_flags[i] != NULL )
1276 seq_printf(m, " %s", x86_cap_flags[i]);
1279 seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
1280 c->loops_per_jiffy/(500000/HZ),
1281 (c->loops_per_jiffy/(5000/HZ)) % 100);
1283 if (c->x86_tlbsize > 0)
1284 seq_printf(m, "TLB size\t: %d 4K pages\n", c->x86_tlbsize);
1285 seq_printf(m, "clflush size\t: %d\n", c->x86_clflush_size);
1286 seq_printf(m, "cache_alignment\t: %d\n", c->x86_cache_alignment);
1288 seq_printf(m, "address sizes\t: %u bits physical, %u bits virtual\n",
1289 c->x86_phys_bits, c->x86_virt_bits);
1291 seq_printf(m, "power management:");
1293 unsigned i;
1294 for (i = 0; i < 32; i++)
1295 if (c->x86_power & (1 << i)) {
1296 if (i < ARRAY_SIZE(x86_power_flags))
1297 seq_printf(m, " %s", x86_power_flags[i]);
1298 else
1299 seq_printf(m, " [%d]", i);
1303 seq_printf(m, "\n\n");
1305 return 0;
1308 static void *c_start(struct seq_file *m, loff_t *pos)
1310 return *pos < NR_CPUS ? cpu_data + *pos : NULL;
1313 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
1315 ++*pos;
1316 return c_start(m, pos);
1319 static void c_stop(struct seq_file *m, void *v)
1323 struct seq_operations cpuinfo_op = {
1324 .start =c_start,
1325 .next = c_next,
1326 .stop = c_stop,
1327 .show = show_cpuinfo,