[PATCH] Make swsusp avoid memory holes and reserved memory regions on x86_64
[linux-2.6/mini2440.git] / arch / x86_64 / kernel / setup.c
blob4b39f0da17f3d3f56795233aa4745b42f652ca79
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.
8 */
11 * This file handles the architecture-dependent parts of initialization
14 #include <linux/errno.h>
15 #include <linux/sched.h>
16 #include <linux/kernel.h>
17 #include <linux/mm.h>
18 #include <linux/stddef.h>
19 #include <linux/unistd.h>
20 #include <linux/ptrace.h>
21 #include <linux/slab.h>
22 #include <linux/user.h>
23 #include <linux/a.out.h>
24 #include <linux/screen_info.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/init.h>
28 #include <linux/initrd.h>
29 #include <linux/highmem.h>
30 #include <linux/bootmem.h>
31 #include <linux/module.h>
32 #include <asm/processor.h>
33 #include <linux/console.h>
34 #include <linux/seq_file.h>
35 #include <linux/crash_dump.h>
36 #include <linux/root_dev.h>
37 #include <linux/pci.h>
38 #include <linux/acpi.h>
39 #include <linux/kallsyms.h>
40 #include <linux/edd.h>
41 #include <linux/mmzone.h>
42 #include <linux/kexec.h>
43 #include <linux/cpufreq.h>
44 #include <linux/dmi.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/ctype.h>
48 #include <asm/mtrr.h>
49 #include <asm/uaccess.h>
50 #include <asm/system.h>
51 #include <asm/io.h>
52 #include <asm/smp.h>
53 #include <asm/msr.h>
54 #include <asm/desc.h>
55 #include <video/edid.h>
56 #include <asm/e820.h>
57 #include <asm/dma.h>
58 #include <asm/mpspec.h>
59 #include <asm/mmu_context.h>
60 #include <asm/bootsetup.h>
61 #include <asm/proto.h>
62 #include <asm/setup.h>
63 #include <asm/mach_apic.h>
64 #include <asm/numa.h>
65 #include <asm/sections.h>
66 #include <asm/dmi.h>
69 * Machine setup..
72 struct cpuinfo_x86 boot_cpu_data __read_mostly;
73 EXPORT_SYMBOL(boot_cpu_data);
75 unsigned long mmu_cr4_features;
77 int acpi_disabled;
78 EXPORT_SYMBOL(acpi_disabled);
79 #ifdef CONFIG_ACPI
80 extern int __initdata acpi_ht;
81 extern acpi_interrupt_flags acpi_sci_flags;
82 int __initdata acpi_force = 0;
83 #endif
85 int acpi_numa __initdata;
87 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
88 int bootloader_type;
90 unsigned long saved_video_mode;
92 /*
93 * Early DMI memory
95 int dmi_alloc_index;
96 char dmi_alloc_data[DMI_MAX_DATA];
99 * Setup options
101 struct screen_info screen_info;
102 EXPORT_SYMBOL(screen_info);
103 struct sys_desc_table_struct {
104 unsigned short length;
105 unsigned char table[0];
108 struct edid_info edid_info;
109 EXPORT_SYMBOL_GPL(edid_info);
110 struct e820map e820;
112 extern int root_mountflags;
114 char command_line[COMMAND_LINE_SIZE];
116 struct resource standard_io_resources[] = {
117 { .name = "dma1", .start = 0x00, .end = 0x1f,
118 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
119 { .name = "pic1", .start = 0x20, .end = 0x21,
120 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
121 { .name = "timer0", .start = 0x40, .end = 0x43,
122 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
123 { .name = "timer1", .start = 0x50, .end = 0x53,
124 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
125 { .name = "keyboard", .start = 0x60, .end = 0x6f,
126 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
127 { .name = "dma page reg", .start = 0x80, .end = 0x8f,
128 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
129 { .name = "pic2", .start = 0xa0, .end = 0xa1,
130 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
131 { .name = "dma2", .start = 0xc0, .end = 0xdf,
132 .flags = IORESOURCE_BUSY | IORESOURCE_IO },
133 { .name = "fpu", .start = 0xf0, .end = 0xff,
134 .flags = IORESOURCE_BUSY | IORESOURCE_IO }
137 #define STANDARD_IO_RESOURCES \
138 (sizeof standard_io_resources / sizeof standard_io_resources[0])
140 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
142 struct resource data_resource = {
143 .name = "Kernel data",
144 .start = 0,
145 .end = 0,
146 .flags = IORESOURCE_RAM,
148 struct resource code_resource = {
149 .name = "Kernel code",
150 .start = 0,
151 .end = 0,
152 .flags = IORESOURCE_RAM,
155 #define IORESOURCE_ROM (IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM)
157 static struct resource system_rom_resource = {
158 .name = "System ROM",
159 .start = 0xf0000,
160 .end = 0xfffff,
161 .flags = IORESOURCE_ROM,
164 static struct resource extension_rom_resource = {
165 .name = "Extension ROM",
166 .start = 0xe0000,
167 .end = 0xeffff,
168 .flags = IORESOURCE_ROM,
171 static struct resource adapter_rom_resources[] = {
172 { .name = "Adapter ROM", .start = 0xc8000, .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 },
178 { .name = "Adapter ROM", .start = 0, .end = 0,
179 .flags = IORESOURCE_ROM },
180 { .name = "Adapter ROM", .start = 0, .end = 0,
181 .flags = IORESOURCE_ROM },
182 { .name = "Adapter ROM", .start = 0, .end = 0,
183 .flags = IORESOURCE_ROM }
186 #define ADAPTER_ROM_RESOURCES \
187 (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
189 static struct resource video_rom_resource = {
190 .name = "Video ROM",
191 .start = 0xc0000,
192 .end = 0xc7fff,
193 .flags = IORESOURCE_ROM,
196 static struct resource video_ram_resource = {
197 .name = "Video RAM area",
198 .start = 0xa0000,
199 .end = 0xbffff,
200 .flags = IORESOURCE_RAM,
203 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
205 static int __init romchecksum(unsigned char *rom, unsigned long length)
207 unsigned char *p, sum = 0;
209 for (p = rom; p < rom + length; p++)
210 sum += *p;
211 return sum == 0;
214 static void __init probe_roms(void)
216 unsigned long start, length, upper;
217 unsigned char *rom;
218 int i;
220 /* video rom */
221 upper = adapter_rom_resources[0].start;
222 for (start = video_rom_resource.start; start < upper; start += 2048) {
223 rom = isa_bus_to_virt(start);
224 if (!romsignature(rom))
225 continue;
227 video_rom_resource.start = start;
229 /* 0 < length <= 0x7f * 512, historically */
230 length = rom[2] * 512;
232 /* if checksum okay, trust length byte */
233 if (length && romchecksum(rom, length))
234 video_rom_resource.end = start + length - 1;
236 request_resource(&iomem_resource, &video_rom_resource);
237 break;
240 start = (video_rom_resource.end + 1 + 2047) & ~2047UL;
241 if (start < upper)
242 start = upper;
244 /* system rom */
245 request_resource(&iomem_resource, &system_rom_resource);
246 upper = system_rom_resource.start;
248 /* check for extension rom (ignore length byte!) */
249 rom = isa_bus_to_virt(extension_rom_resource.start);
250 if (romsignature(rom)) {
251 length = extension_rom_resource.end - extension_rom_resource.start + 1;
252 if (romchecksum(rom, length)) {
253 request_resource(&iomem_resource, &extension_rom_resource);
254 upper = extension_rom_resource.start;
258 /* check for adapter roms on 2k boundaries */
259 for (i = 0; i < ADAPTER_ROM_RESOURCES && start < upper; start += 2048) {
260 rom = isa_bus_to_virt(start);
261 if (!romsignature(rom))
262 continue;
264 /* 0 < length <= 0x7f * 512, historically */
265 length = rom[2] * 512;
267 /* but accept any length that fits if checksum okay */
268 if (!length || start + length > upper || !romchecksum(rom, length))
269 continue;
271 adapter_rom_resources[i].start = start;
272 adapter_rom_resources[i].end = start + length - 1;
273 request_resource(&iomem_resource, &adapter_rom_resources[i]);
275 start = adapter_rom_resources[i++].end & ~2047UL;
279 /* Check for full argument with no trailing characters */
280 static int fullarg(char *p, char *arg)
282 int l = strlen(arg);
283 return !memcmp(p, arg, l) && (p[l] == 0 || isspace(p[l]));
286 static __init void parse_cmdline_early (char ** cmdline_p)
288 char c = ' ', *to = command_line, *from = COMMAND_LINE;
289 int len = 0;
290 int userdef = 0;
292 for (;;) {
293 if (c != ' ')
294 goto next_char;
296 #ifdef CONFIG_SMP
298 * If the BIOS enumerates physical processors before logical,
299 * maxcpus=N at enumeration-time can be used to disable HT.
301 else if (!memcmp(from, "maxcpus=", 8)) {
302 extern unsigned int maxcpus;
304 maxcpus = simple_strtoul(from + 8, NULL, 0);
306 #endif
307 #ifdef CONFIG_ACPI
308 /* "acpi=off" disables both ACPI table parsing and interpreter init */
309 if (fullarg(from,"acpi=off"))
310 disable_acpi();
312 if (fullarg(from, "acpi=force")) {
313 /* add later when we do DMI horrors: */
314 acpi_force = 1;
315 acpi_disabled = 0;
318 /* acpi=ht just means: do ACPI MADT parsing
319 at bootup, but don't enable the full ACPI interpreter */
320 if (fullarg(from, "acpi=ht")) {
321 if (!acpi_force)
322 disable_acpi();
323 acpi_ht = 1;
325 else if (fullarg(from, "pci=noacpi"))
326 acpi_disable_pci();
327 else if (fullarg(from, "acpi=noirq"))
328 acpi_noirq_set();
330 else if (fullarg(from, "acpi_sci=edge"))
331 acpi_sci_flags.trigger = 1;
332 else if (fullarg(from, "acpi_sci=level"))
333 acpi_sci_flags.trigger = 3;
334 else if (fullarg(from, "acpi_sci=high"))
335 acpi_sci_flags.polarity = 1;
336 else if (fullarg(from, "acpi_sci=low"))
337 acpi_sci_flags.polarity = 3;
339 /* acpi=strict disables out-of-spec workarounds */
340 else if (fullarg(from, "acpi=strict")) {
341 acpi_strict = 1;
343 #ifdef CONFIG_X86_IO_APIC
344 else if (fullarg(from, "acpi_skip_timer_override"))
345 acpi_skip_timer_override = 1;
346 #endif
347 #endif
349 if (fullarg(from, "disable_timer_pin_1"))
350 disable_timer_pin_1 = 1;
351 if (fullarg(from, "enable_timer_pin_1"))
352 disable_timer_pin_1 = -1;
354 if (fullarg(from, "nolapic") || fullarg(from, "disableapic")) {
355 clear_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability);
356 disable_apic = 1;
359 if (fullarg(from, "noapic"))
360 skip_ioapic_setup = 1;
362 if (fullarg(from,"apic")) {
363 skip_ioapic_setup = 0;
364 ioapic_force = 1;
367 if (!memcmp(from, "mem=", 4))
368 parse_memopt(from+4, &from);
370 if (!memcmp(from, "memmap=", 7)) {
371 /* exactmap option is for used defined memory */
372 if (!memcmp(from+7, "exactmap", 8)) {
373 #ifdef CONFIG_CRASH_DUMP
374 /* If we are doing a crash dump, we
375 * still need to know the real mem
376 * size before original memory map is
377 * reset.
379 saved_max_pfn = e820_end_of_ram();
380 #endif
381 from += 8+7;
382 end_pfn_map = 0;
383 e820.nr_map = 0;
384 userdef = 1;
386 else {
387 parse_memmapopt(from+7, &from);
388 userdef = 1;
392 #ifdef CONFIG_NUMA
393 if (!memcmp(from, "numa=", 5))
394 numa_setup(from+5);
395 #endif
397 if (!memcmp(from,"iommu=",6)) {
398 iommu_setup(from+6);
401 if (fullarg(from,"oops=panic"))
402 panic_on_oops = 1;
404 if (!memcmp(from, "noexec=", 7))
405 nonx_setup(from + 7);
407 #ifdef CONFIG_KEXEC
408 /* crashkernel=size@addr specifies the location to reserve for
409 * a crash kernel. By reserving this memory we guarantee
410 * that linux never set's it up as a DMA target.
411 * Useful for holding code to do something appropriate
412 * after a kernel panic.
414 else if (!memcmp(from, "crashkernel=", 12)) {
415 unsigned long size, base;
416 size = memparse(from+12, &from);
417 if (*from == '@') {
418 base = memparse(from+1, &from);
419 /* FIXME: Do I want a sanity check
420 * to validate the memory range?
422 crashk_res.start = base;
423 crashk_res.end = base + size - 1;
426 #endif
428 #ifdef CONFIG_PROC_VMCORE
429 /* elfcorehdr= specifies the location of elf core header
430 * stored by the crashed kernel. This option will be passed
431 * by kexec loader to the capture kernel.
433 else if(!memcmp(from, "elfcorehdr=", 11))
434 elfcorehdr_addr = memparse(from+11, &from);
435 #endif
437 #ifdef CONFIG_HOTPLUG_CPU
438 else if (!memcmp(from, "additional_cpus=", 16))
439 setup_additional_cpus(from+16);
440 #endif
442 next_char:
443 c = *(from++);
444 if (!c)
445 break;
446 if (COMMAND_LINE_SIZE <= ++len)
447 break;
448 *(to++) = c;
450 if (userdef) {
451 printk(KERN_INFO "user-defined physical RAM map:\n");
452 e820_print_map("user");
454 *to = '\0';
455 *cmdline_p = command_line;
458 #ifndef CONFIG_NUMA
459 static void __init
460 contig_initmem_init(unsigned long start_pfn, unsigned long end_pfn)
462 unsigned long bootmap_size, bootmap;
464 bootmap_size = bootmem_bootmap_pages(end_pfn)<<PAGE_SHIFT;
465 bootmap = find_e820_area(0, end_pfn<<PAGE_SHIFT, bootmap_size);
466 if (bootmap == -1L)
467 panic("Cannot find bootmem map of size %ld\n",bootmap_size);
468 bootmap_size = init_bootmem(bootmap >> PAGE_SHIFT, end_pfn);
469 e820_bootmem_free(NODE_DATA(0), 0, end_pfn << PAGE_SHIFT);
470 reserve_bootmem(bootmap, bootmap_size);
472 #endif
474 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
475 struct edd edd;
476 #ifdef CONFIG_EDD_MODULE
477 EXPORT_SYMBOL(edd);
478 #endif
480 * copy_edd() - Copy the BIOS EDD information
481 * from boot_params into a safe place.
484 static inline void copy_edd(void)
486 memcpy(edd.mbr_signature, EDD_MBR_SIGNATURE, sizeof(edd.mbr_signature));
487 memcpy(edd.edd_info, EDD_BUF, sizeof(edd.edd_info));
488 edd.mbr_signature_nr = EDD_MBR_SIG_NR;
489 edd.edd_info_nr = EDD_NR;
491 #else
492 static inline void copy_edd(void)
495 #endif
497 #define EBDA_ADDR_POINTER 0x40E
499 unsigned __initdata ebda_addr;
500 unsigned __initdata ebda_size;
502 static void discover_ebda(void)
505 * there is a real-mode segmented pointer pointing to the
506 * 4K EBDA area at 0x40E
508 ebda_addr = *(unsigned short *)EBDA_ADDR_POINTER;
509 ebda_addr <<= 4;
511 ebda_size = *(unsigned short *)(unsigned long)ebda_addr;
513 /* Round EBDA up to pages */
514 if (ebda_size == 0)
515 ebda_size = 1;
516 ebda_size <<= 10;
517 ebda_size = round_up(ebda_size + (ebda_addr & ~PAGE_MASK), PAGE_SIZE);
518 if (ebda_size > 64*1024)
519 ebda_size = 64*1024;
522 void __init setup_arch(char **cmdline_p)
524 ROOT_DEV = old_decode_dev(ORIG_ROOT_DEV);
525 screen_info = SCREEN_INFO;
526 edid_info = EDID_INFO;
527 saved_video_mode = SAVED_VIDEO_MODE;
528 bootloader_type = LOADER_TYPE;
530 #ifdef CONFIG_BLK_DEV_RAM
531 rd_image_start = RAMDISK_FLAGS & RAMDISK_IMAGE_START_MASK;
532 rd_prompt = ((RAMDISK_FLAGS & RAMDISK_PROMPT_FLAG) != 0);
533 rd_doload = ((RAMDISK_FLAGS & RAMDISK_LOAD_FLAG) != 0);
534 #endif
535 setup_memory_region();
536 copy_edd();
538 if (!MOUNT_ROOT_RDONLY)
539 root_mountflags &= ~MS_RDONLY;
540 init_mm.start_code = (unsigned long) &_text;
541 init_mm.end_code = (unsigned long) &_etext;
542 init_mm.end_data = (unsigned long) &_edata;
543 init_mm.brk = (unsigned long) &_end;
545 code_resource.start = virt_to_phys(&_text);
546 code_resource.end = virt_to_phys(&_etext)-1;
547 data_resource.start = virt_to_phys(&_etext);
548 data_resource.end = virt_to_phys(&_edata)-1;
550 parse_cmdline_early(cmdline_p);
552 early_identify_cpu(&boot_cpu_data);
555 * partially used pages are not usable - thus
556 * we are rounding upwards:
558 end_pfn = e820_end_of_ram();
559 num_physpages = end_pfn; /* for pfn_valid */
561 check_efer();
563 discover_ebda();
565 init_memory_mapping(0, (end_pfn_map << PAGE_SHIFT));
567 dmi_scan_machine();
569 zap_low_mappings(0);
571 #ifdef CONFIG_ACPI
573 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
574 * Call this early for SRAT node setup.
576 acpi_boot_table_init();
577 #endif
579 #ifdef CONFIG_ACPI_NUMA
581 * Parse SRAT to discover nodes.
583 acpi_numa_init();
584 #endif
586 #ifdef CONFIG_NUMA
587 numa_initmem_init(0, end_pfn);
588 #else
589 contig_initmem_init(0, end_pfn);
590 #endif
592 /* Reserve direct mapping */
593 reserve_bootmem_generic(table_start << PAGE_SHIFT,
594 (table_end - table_start) << PAGE_SHIFT);
596 /* reserve kernel */
597 reserve_bootmem_generic(__pa_symbol(&_text),
598 __pa_symbol(&_end) - __pa_symbol(&_text));
601 * reserve physical page 0 - it's a special BIOS page on many boxes,
602 * enabling clean reboots, SMP operation, laptop functions.
604 reserve_bootmem_generic(0, PAGE_SIZE);
606 /* reserve ebda region */
607 if (ebda_addr)
608 reserve_bootmem_generic(ebda_addr, ebda_size);
610 #ifdef CONFIG_SMP
612 * But first pinch a few for the stack/trampoline stuff
613 * FIXME: Don't need the extra page at 4K, but need to fix
614 * trampoline before removing it. (see the GDT stuff)
616 reserve_bootmem_generic(PAGE_SIZE, PAGE_SIZE);
618 /* Reserve SMP trampoline */
619 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE, PAGE_SIZE);
620 #endif
622 #ifdef CONFIG_ACPI_SLEEP
624 * Reserve low memory region for sleep support.
626 acpi_reserve_bootmem();
627 #endif
628 #ifdef CONFIG_X86_LOCAL_APIC
630 * Find and reserve possible boot-time SMP configuration:
632 find_smp_config();
633 #endif
634 #ifdef CONFIG_BLK_DEV_INITRD
635 if (LOADER_TYPE && INITRD_START) {
636 if (INITRD_START + INITRD_SIZE <= (end_pfn << PAGE_SHIFT)) {
637 reserve_bootmem_generic(INITRD_START, INITRD_SIZE);
638 initrd_start =
639 INITRD_START ? INITRD_START + PAGE_OFFSET : 0;
640 initrd_end = initrd_start+INITRD_SIZE;
642 else {
643 printk(KERN_ERR "initrd extends beyond end of memory "
644 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
645 (unsigned long)(INITRD_START + INITRD_SIZE),
646 (unsigned long)(end_pfn << PAGE_SHIFT));
647 initrd_start = 0;
650 #endif
651 #ifdef CONFIG_KEXEC
652 if (crashk_res.start != crashk_res.end) {
653 reserve_bootmem_generic(crashk_res.start,
654 crashk_res.end - crashk_res.start + 1);
656 #endif
658 paging_init();
660 check_ioapic();
663 * set this early, so we dont allocate cpu0
664 * if MADT list doesnt list BSP first
665 * mpparse.c/MP_processor_info() allocates logical cpu numbers.
667 cpu_set(0, cpu_present_map);
668 #ifdef CONFIG_ACPI
670 * Read APIC and some other early information from ACPI tables.
672 acpi_boot_init();
673 #endif
675 init_cpu_to_node();
677 #ifdef CONFIG_X86_LOCAL_APIC
679 * get boot-time SMP configuration:
681 if (smp_found_config)
682 get_smp_config();
683 init_apic_mappings();
684 #endif
687 * Request address space for all standard RAM and ROM resources
688 * and also for regions reported as reserved by the e820.
690 probe_roms();
691 e820_reserve_resources();
692 e820_mark_nosave_regions();
694 request_resource(&iomem_resource, &video_ram_resource);
697 unsigned i;
698 /* request I/O space for devices used on all i[345]86 PCs */
699 for (i = 0; i < STANDARD_IO_RESOURCES; i++)
700 request_resource(&ioport_resource, &standard_io_resources[i]);
703 e820_setup_gap();
705 #ifdef CONFIG_VT
706 #if defined(CONFIG_VGA_CONSOLE)
707 conswitchp = &vga_con;
708 #elif defined(CONFIG_DUMMY_CONSOLE)
709 conswitchp = &dummy_con;
710 #endif
711 #endif
714 static int __cpuinit get_model_name(struct cpuinfo_x86 *c)
716 unsigned int *v;
718 if (c->extended_cpuid_level < 0x80000004)
719 return 0;
721 v = (unsigned int *) c->x86_model_id;
722 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
723 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
724 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
725 c->x86_model_id[48] = 0;
726 return 1;
730 static void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
732 unsigned int n, dummy, eax, ebx, ecx, edx;
734 n = c->extended_cpuid_level;
736 if (n >= 0x80000005) {
737 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
738 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
739 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
740 c->x86_cache_size=(ecx>>24)+(edx>>24);
741 /* On K8 L1 TLB is inclusive, so don't count it */
742 c->x86_tlbsize = 0;
745 if (n >= 0x80000006) {
746 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
747 ecx = cpuid_ecx(0x80000006);
748 c->x86_cache_size = ecx >> 16;
749 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
751 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
752 c->x86_cache_size, ecx & 0xFF);
755 if (n >= 0x80000007)
756 cpuid(0x80000007, &dummy, &dummy, &dummy, &c->x86_power);
757 if (n >= 0x80000008) {
758 cpuid(0x80000008, &eax, &dummy, &dummy, &dummy);
759 c->x86_virt_bits = (eax >> 8) & 0xff;
760 c->x86_phys_bits = eax & 0xff;
764 #ifdef CONFIG_NUMA
765 static int nearby_node(int apicid)
767 int i;
768 for (i = apicid - 1; i >= 0; i--) {
769 int node = apicid_to_node[i];
770 if (node != NUMA_NO_NODE && node_online(node))
771 return node;
773 for (i = apicid + 1; i < MAX_LOCAL_APIC; i++) {
774 int node = apicid_to_node[i];
775 if (node != NUMA_NO_NODE && node_online(node))
776 return node;
778 return first_node(node_online_map); /* Shouldn't happen */
780 #endif
783 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
784 * Assumes number of cores is a power of two.
786 static void __init amd_detect_cmp(struct cpuinfo_x86 *c)
788 #ifdef CONFIG_SMP
789 unsigned bits;
790 #ifdef CONFIG_NUMA
791 int cpu = smp_processor_id();
792 int node = 0;
793 unsigned apicid = hard_smp_processor_id();
794 #endif
795 unsigned ecx = cpuid_ecx(0x80000008);
797 c->x86_max_cores = (ecx & 0xff) + 1;
799 /* CPU telling us the core id bits shift? */
800 bits = (ecx >> 12) & 0xF;
802 /* Otherwise recompute */
803 if (bits == 0) {
804 while ((1 << bits) < c->x86_max_cores)
805 bits++;
808 /* Low order bits define the core id (index of core in socket) */
809 c->cpu_core_id = c->phys_proc_id & ((1 << bits)-1);
810 /* Convert the APIC ID into the socket ID */
811 c->phys_proc_id = phys_pkg_id(bits);
813 #ifdef CONFIG_NUMA
814 node = c->phys_proc_id;
815 if (apicid_to_node[apicid] != NUMA_NO_NODE)
816 node = apicid_to_node[apicid];
817 if (!node_online(node)) {
818 /* Two possibilities here:
819 - The CPU is missing memory and no node was created.
820 In that case try picking one from a nearby CPU
821 - The APIC IDs differ from the HyperTransport node IDs
822 which the K8 northbridge parsing fills in.
823 Assume they are all increased by a constant offset,
824 but in the same order as the HT nodeids.
825 If that doesn't result in a usable node fall back to the
826 path for the previous case. */
827 int ht_nodeid = apicid - (cpu_data[0].phys_proc_id << bits);
828 if (ht_nodeid >= 0 &&
829 apicid_to_node[ht_nodeid] != NUMA_NO_NODE)
830 node = apicid_to_node[ht_nodeid];
831 /* Pick a nearby node */
832 if (!node_online(node))
833 node = nearby_node(apicid);
835 numa_set_node(cpu, node);
837 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
838 #endif
839 #endif
842 static void __init init_amd(struct cpuinfo_x86 *c)
844 unsigned level;
846 #ifdef CONFIG_SMP
847 unsigned long value;
850 * Disable TLB flush filter by setting HWCR.FFDIS on K8
851 * bit 6 of msr C001_0015
853 * Errata 63 for SH-B3 steppings
854 * Errata 122 for all steppings (F+ have it disabled by default)
856 if (c->x86 == 15) {
857 rdmsrl(MSR_K8_HWCR, value);
858 value |= 1 << 6;
859 wrmsrl(MSR_K8_HWCR, value);
861 #endif
863 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
864 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
865 clear_bit(0*32+31, &c->x86_capability);
867 /* On C+ stepping K8 rep microcode works well for copy/memset */
868 level = cpuid_eax(1);
869 if (c->x86 == 15 && ((level >= 0x0f48 && level < 0x0f50) || level >= 0x0f58))
870 set_bit(X86_FEATURE_REP_GOOD, &c->x86_capability);
872 /* Enable workaround for FXSAVE leak */
873 if (c->x86 >= 6)
874 set_bit(X86_FEATURE_FXSAVE_LEAK, &c->x86_capability);
876 level = get_model_name(c);
877 if (!level) {
878 switch (c->x86) {
879 case 15:
880 /* Should distinguish Models here, but this is only
881 a fallback anyways. */
882 strcpy(c->x86_model_id, "Hammer");
883 break;
886 display_cacheinfo(c);
888 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
889 if (c->x86_power & (1<<8))
890 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
892 /* Multi core CPU? */
893 if (c->extended_cpuid_level >= 0x80000008)
894 amd_detect_cmp(c);
896 /* Fix cpuid4 emulation for more */
897 num_cache_leaves = 3;
900 static void __cpuinit detect_ht(struct cpuinfo_x86 *c)
902 #ifdef CONFIG_SMP
903 u32 eax, ebx, ecx, edx;
904 int index_msb, core_bits;
906 cpuid(1, &eax, &ebx, &ecx, &edx);
909 if (!cpu_has(c, X86_FEATURE_HT))
910 return;
911 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
912 goto out;
914 smp_num_siblings = (ebx & 0xff0000) >> 16;
916 if (smp_num_siblings == 1) {
917 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
918 } else if (smp_num_siblings > 1 ) {
920 if (smp_num_siblings > NR_CPUS) {
921 printk(KERN_WARNING "CPU: Unsupported number of the siblings %d", smp_num_siblings);
922 smp_num_siblings = 1;
923 return;
926 index_msb = get_count_order(smp_num_siblings);
927 c->phys_proc_id = phys_pkg_id(index_msb);
929 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
931 index_msb = get_count_order(smp_num_siblings) ;
933 core_bits = get_count_order(c->x86_max_cores);
935 c->cpu_core_id = phys_pkg_id(index_msb) &
936 ((1 << core_bits) - 1);
938 out:
939 if ((c->x86_max_cores * smp_num_siblings) > 1) {
940 printk(KERN_INFO "CPU: Physical Processor ID: %d\n", c->phys_proc_id);
941 printk(KERN_INFO "CPU: Processor Core ID: %d\n", c->cpu_core_id);
944 #endif
948 * find out the number of processor cores on the die
950 static int __cpuinit intel_num_cpu_cores(struct cpuinfo_x86 *c)
952 unsigned int eax, t;
954 if (c->cpuid_level < 4)
955 return 1;
957 cpuid_count(4, 0, &eax, &t, &t, &t);
959 if (eax & 0x1f)
960 return ((eax >> 26) + 1);
961 else
962 return 1;
965 static void srat_detect_node(void)
967 #ifdef CONFIG_NUMA
968 unsigned node;
969 int cpu = smp_processor_id();
970 int apicid = hard_smp_processor_id();
972 /* Don't do the funky fallback heuristics the AMD version employs
973 for now. */
974 node = apicid_to_node[apicid];
975 if (node == NUMA_NO_NODE)
976 node = first_node(node_online_map);
977 numa_set_node(cpu, node);
979 if (acpi_numa > 0)
980 printk(KERN_INFO "CPU %d/%x -> Node %d\n", cpu, apicid, node);
981 #endif
984 static void __cpuinit init_intel(struct cpuinfo_x86 *c)
986 /* Cache sizes */
987 unsigned n;
989 init_intel_cacheinfo(c);
990 if (c->cpuid_level > 9 ) {
991 unsigned eax = cpuid_eax(10);
992 /* Check for version and the number of counters */
993 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
994 set_bit(X86_FEATURE_ARCH_PERFMON, &c->x86_capability);
997 n = c->extended_cpuid_level;
998 if (n >= 0x80000008) {
999 unsigned eax = cpuid_eax(0x80000008);
1000 c->x86_virt_bits = (eax >> 8) & 0xff;
1001 c->x86_phys_bits = eax & 0xff;
1002 /* CPUID workaround for Intel 0F34 CPU */
1003 if (c->x86_vendor == X86_VENDOR_INTEL &&
1004 c->x86 == 0xF && c->x86_model == 0x3 &&
1005 c->x86_mask == 0x4)
1006 c->x86_phys_bits = 36;
1009 if (c->x86 == 15)
1010 c->x86_cache_alignment = c->x86_clflush_size * 2;
1011 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
1012 (c->x86 == 0x6 && c->x86_model >= 0x0e))
1013 set_bit(X86_FEATURE_CONSTANT_TSC, &c->x86_capability);
1014 set_bit(X86_FEATURE_SYNC_RDTSC, &c->x86_capability);
1015 c->x86_max_cores = intel_num_cpu_cores(c);
1017 srat_detect_node();
1020 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
1022 char *v = c->x86_vendor_id;
1024 if (!strcmp(v, "AuthenticAMD"))
1025 c->x86_vendor = X86_VENDOR_AMD;
1026 else if (!strcmp(v, "GenuineIntel"))
1027 c->x86_vendor = X86_VENDOR_INTEL;
1028 else
1029 c->x86_vendor = X86_VENDOR_UNKNOWN;
1032 struct cpu_model_info {
1033 int vendor;
1034 int family;
1035 char *model_names[16];
1038 /* Do some early cpuid on the boot CPU to get some parameter that are
1039 needed before check_bugs. Everything advanced is in identify_cpu
1040 below. */
1041 void __cpuinit early_identify_cpu(struct cpuinfo_x86 *c)
1043 u32 tfms;
1045 c->loops_per_jiffy = loops_per_jiffy;
1046 c->x86_cache_size = -1;
1047 c->x86_vendor = X86_VENDOR_UNKNOWN;
1048 c->x86_model = c->x86_mask = 0; /* So far unknown... */
1049 c->x86_vendor_id[0] = '\0'; /* Unset */
1050 c->x86_model_id[0] = '\0'; /* Unset */
1051 c->x86_clflush_size = 64;
1052 c->x86_cache_alignment = c->x86_clflush_size;
1053 c->x86_max_cores = 1;
1054 c->extended_cpuid_level = 0;
1055 memset(&c->x86_capability, 0, sizeof c->x86_capability);
1057 /* Get vendor name */
1058 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
1059 (unsigned int *)&c->x86_vendor_id[0],
1060 (unsigned int *)&c->x86_vendor_id[8],
1061 (unsigned int *)&c->x86_vendor_id[4]);
1063 get_cpu_vendor(c);
1065 /* Initialize the standard set of capabilities */
1066 /* Note that the vendor-specific code below might override */
1068 /* Intel-defined flags: level 0x00000001 */
1069 if (c->cpuid_level >= 0x00000001) {
1070 __u32 misc;
1071 cpuid(0x00000001, &tfms, &misc, &c->x86_capability[4],
1072 &c->x86_capability[0]);
1073 c->x86 = (tfms >> 8) & 0xf;
1074 c->x86_model = (tfms >> 4) & 0xf;
1075 c->x86_mask = tfms & 0xf;
1076 if (c->x86 == 0xf)
1077 c->x86 += (tfms >> 20) & 0xff;
1078 if (c->x86 >= 0x6)
1079 c->x86_model += ((tfms >> 16) & 0xF) << 4;
1080 if (c->x86_capability[0] & (1<<19))
1081 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
1082 } else {
1083 /* Have CPUID level 0 only - unheard of */
1084 c->x86 = 4;
1087 #ifdef CONFIG_SMP
1088 c->phys_proc_id = (cpuid_ebx(1) >> 24) & 0xff;
1089 #endif
1093 * This does the hard work of actually picking apart the CPU stuff...
1095 void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
1097 int i;
1098 u32 xlvl;
1100 early_identify_cpu(c);
1102 /* AMD-defined flags: level 0x80000001 */
1103 xlvl = cpuid_eax(0x80000000);
1104 c->extended_cpuid_level = xlvl;
1105 if ((xlvl & 0xffff0000) == 0x80000000) {
1106 if (xlvl >= 0x80000001) {
1107 c->x86_capability[1] = cpuid_edx(0x80000001);
1108 c->x86_capability[6] = cpuid_ecx(0x80000001);
1110 if (xlvl >= 0x80000004)
1111 get_model_name(c); /* Default name */
1114 /* Transmeta-defined flags: level 0x80860001 */
1115 xlvl = cpuid_eax(0x80860000);
1116 if ((xlvl & 0xffff0000) == 0x80860000) {
1117 /* Don't set x86_cpuid_level here for now to not confuse. */
1118 if (xlvl >= 0x80860001)
1119 c->x86_capability[2] = cpuid_edx(0x80860001);
1122 c->apicid = phys_pkg_id(0);
1125 * Vendor-specific initialization. In this section we
1126 * canonicalize the feature flags, meaning if there are
1127 * features a certain CPU supports which CPUID doesn't
1128 * tell us, CPUID claiming incorrect flags, or other bugs,
1129 * we handle them here.
1131 * At the end of this section, c->x86_capability better
1132 * indicate the features this CPU genuinely supports!
1134 switch (c->x86_vendor) {
1135 case X86_VENDOR_AMD:
1136 init_amd(c);
1137 break;
1139 case X86_VENDOR_INTEL:
1140 init_intel(c);
1141 break;
1143 case X86_VENDOR_UNKNOWN:
1144 default:
1145 display_cacheinfo(c);
1146 break;
1149 select_idle_routine(c);
1150 detect_ht(c);
1153 * On SMP, boot_cpu_data holds the common feature set between
1154 * all CPUs; so make sure that we indicate which features are
1155 * common between the CPUs. The first time this routine gets
1156 * executed, c == &boot_cpu_data.
1158 if (c != &boot_cpu_data) {
1159 /* AND the already accumulated flags with these */
1160 for (i = 0 ; i < NCAPINTS ; i++)
1161 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1164 #ifdef CONFIG_X86_MCE
1165 mcheck_init(c);
1166 #endif
1167 if (c == &boot_cpu_data)
1168 mtrr_bp_init();
1169 else
1170 mtrr_ap_init();
1171 #ifdef CONFIG_NUMA
1172 numa_add_cpu(smp_processor_id());
1173 #endif
1177 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1179 if (c->x86_model_id[0])
1180 printk("%s", c->x86_model_id);
1182 if (c->x86_mask || c->cpuid_level >= 0)
1183 printk(" stepping %02x\n", c->x86_mask);
1184 else
1185 printk("\n");
1189 * Get CPU information for use by the procfs.
1192 static int show_cpuinfo(struct seq_file *m, void *v)
1194 struct cpuinfo_x86 *c = v;
1197 * These flag bits must match the definitions in <asm/cpufeature.h>.
1198 * NULL means this bit is undefined or reserved; either way it doesn't
1199 * have meaning as far as Linux is concerned. Note that it's important
1200 * to realize there is a difference between this table and CPUID -- if
1201 * applications want to get the raw CPUID data, they should access
1202 * /dev/cpu/<cpu_nr>/cpuid instead.
1204 static char *x86_cap_flags[] = {
1205 /* Intel-defined */
1206 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1207 "cx8", "apic", NULL, "sep", "mtrr", "pge", "mca", "cmov",
1208 "pat", "pse36", "pn", "clflush", NULL, "dts", "acpi", "mmx",
1209 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", NULL,
1211 /* AMD-defined */
1212 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1213 NULL, NULL, NULL, "syscall", NULL, NULL, NULL, NULL,
1214 NULL, NULL, NULL, NULL, "nx", NULL, "mmxext", NULL,
1215 NULL, "fxsr_opt", NULL, "rdtscp", NULL, "lm", "3dnowext", "3dnow",
1217 /* Transmeta-defined */
1218 "recovery", "longrun", NULL, "lrti", NULL, NULL, NULL, NULL,
1219 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1220 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1221 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1223 /* Other (Linux-defined) */
1224 "cxmmx", NULL, "cyrix_arr", "centaur_mcr", NULL,
1225 "constant_tsc", NULL, NULL,
1226 "up", 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 /* Intel-defined (#2) */
1231 "pni", NULL, NULL, "monitor", "ds_cpl", "vmx", "smx", "est",
1232 "tm2", NULL, "cid", NULL, NULL, "cx16", "xtpr", NULL,
1233 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1234 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1236 /* VIA/Cyrix/Centaur-defined */
1237 NULL, NULL, "rng", "rng_en", NULL, NULL, "ace", "ace_en",
1238 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1239 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1240 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1242 /* AMD-defined (#2) */
1243 "lahf_lm", "cmp_legacy", "svm", NULL, "cr8_legacy", NULL, NULL, NULL,
1244 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1245 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1246 NULL, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
1248 static char *x86_power_flags[] = {
1249 "ts", /* temperature sensor */
1250 "fid", /* frequency id control */
1251 "vid", /* voltage id control */
1252 "ttp", /* thermal trip */
1253 "tm",
1254 "stc",
1255 NULL,
1256 /* nothing */ /* constant_tsc - moved to flags */
1260 #ifdef CONFIG_SMP
1261 if (!cpu_online(c-cpu_data))
1262 return 0;
1263 #endif
1265 seq_printf(m,"processor\t: %u\n"
1266 "vendor_id\t: %s\n"
1267 "cpu family\t: %d\n"
1268 "model\t\t: %d\n"
1269 "model name\t: %s\n",
1270 (unsigned)(c-cpu_data),
1271 c->x86_vendor_id[0] ? c->x86_vendor_id : "unknown",
1272 c->x86,
1273 (int)c->x86_model,
1274 c->x86_model_id[0] ? c->x86_model_id : "unknown");
1276 if (c->x86_mask || c->cpuid_level >= 0)
1277 seq_printf(m, "stepping\t: %d\n", c->x86_mask);
1278 else
1279 seq_printf(m, "stepping\t: unknown\n");
1281 if (cpu_has(c,X86_FEATURE_TSC)) {
1282 unsigned int freq = cpufreq_quick_get((unsigned)(c-cpu_data));
1283 if (!freq)
1284 freq = cpu_khz;
1285 seq_printf(m, "cpu MHz\t\t: %u.%03u\n",
1286 freq / 1000, (freq % 1000));
1289 /* Cache size */
1290 if (c->x86_cache_size >= 0)
1291 seq_printf(m, "cache size\t: %d KB\n", c->x86_cache_size);
1293 #ifdef CONFIG_SMP
1294 if (smp_num_siblings * c->x86_max_cores > 1) {
1295 int cpu = c - cpu_data;
1296 seq_printf(m, "physical id\t: %d\n", c->phys_proc_id);
1297 seq_printf(m, "siblings\t: %d\n", cpus_weight(cpu_core_map[cpu]));
1298 seq_printf(m, "core id\t\t: %d\n", c->cpu_core_id);
1299 seq_printf(m, "cpu cores\t: %d\n", c->booted_cores);
1301 #endif
1303 seq_printf(m,
1304 "fpu\t\t: yes\n"
1305 "fpu_exception\t: yes\n"
1306 "cpuid level\t: %d\n"
1307 "wp\t\t: yes\n"
1308 "flags\t\t:",
1309 c->cpuid_level);
1312 int i;
1313 for ( i = 0 ; i < 32*NCAPINTS ; i++ )
1314 if (cpu_has(c, i) && x86_cap_flags[i] != NULL)
1315 seq_printf(m, " %s", x86_cap_flags[i]);
1318 seq_printf(m, "\nbogomips\t: %lu.%02lu\n",
1319 c->loops_per_jiffy/(500000/HZ),
1320 (c->loops_per_jiffy/(5000/HZ)) % 100);
1322 if (c->x86_tlbsize > 0)
1323 seq_printf(m, "TLB size\t: %d 4K pages\n", c->x86_tlbsize);
1324 seq_printf(m, "clflush size\t: %d\n", c->x86_clflush_size);
1325 seq_printf(m, "cache_alignment\t: %d\n", c->x86_cache_alignment);
1327 seq_printf(m, "address sizes\t: %u bits physical, %u bits virtual\n",
1328 c->x86_phys_bits, c->x86_virt_bits);
1330 seq_printf(m, "power management:");
1332 unsigned i;
1333 for (i = 0; i < 32; i++)
1334 if (c->x86_power & (1 << i)) {
1335 if (i < ARRAY_SIZE(x86_power_flags) &&
1336 x86_power_flags[i])
1337 seq_printf(m, "%s%s",
1338 x86_power_flags[i][0]?" ":"",
1339 x86_power_flags[i]);
1340 else
1341 seq_printf(m, " [%d]", i);
1345 seq_printf(m, "\n\n");
1347 return 0;
1350 static void *c_start(struct seq_file *m, loff_t *pos)
1352 return *pos < NR_CPUS ? cpu_data + *pos : NULL;
1355 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
1357 ++*pos;
1358 return c_start(m, pos);
1361 static void c_stop(struct seq_file *m, void *v)
1365 struct seq_operations cpuinfo_op = {
1366 .start =c_start,
1367 .next = c_next,
1368 .stop = c_stop,
1369 .show = show_cpuinfo,
1372 #if defined(CONFIG_INPUT_PCSPKR) || defined(CONFIG_INPUT_PCSPKR_MODULE)
1373 #include <linux/platform_device.h>
1374 static __init int add_pcspkr(void)
1376 struct platform_device *pd;
1377 int ret;
1379 pd = platform_device_alloc("pcspkr", -1);
1380 if (!pd)
1381 return -ENOMEM;
1383 ret = platform_device_add(pd);
1384 if (ret)
1385 platform_device_put(pd);
1387 return ret;
1389 device_initcall(add_pcspkr);
1390 #endif