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.
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>
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/tty.h>
25 #include <linux/ioport.h>
26 #include <linux/delay.h>
27 #include <linux/config.h>
28 #include <linux/init.h>
29 #include <linux/initrd.h>
30 #include <linux/highmem.h>
31 #include <linux/bootmem.h>
32 #include <linux/module.h>
33 #include <asm/processor.h>
34 #include <linux/console.h>
35 #include <linux/seq_file.h>
36 #include <linux/crash_dump.h>
37 #include <linux/root_dev.h>
38 #include <linux/pci.h>
39 #include <linux/acpi.h>
40 #include <linux/kallsyms.h>
41 #include <linux/edd.h>
42 #include <linux/mmzone.h>
43 #include <linux/kexec.h>
44 #include <linux/cpufreq.h>
45 #include <linux/dmi.h>
46 #include <linux/dma-mapping.h>
47 #include <linux/ctype.h>
50 #include <asm/uaccess.h>
51 #include <asm/system.h>
56 #include <video/edid.h>
59 #include <asm/mpspec.h>
60 #include <asm/mmu_context.h>
61 #include <asm/bootsetup.h>
62 #include <asm/proto.h>
63 #include <asm/setup.h>
64 #include <asm/mach_apic.h>
66 #include <asm/sections.h>
73 struct cpuinfo_x86 boot_cpu_data __read_mostly
;
75 unsigned long mmu_cr4_features
;
78 EXPORT_SYMBOL(acpi_disabled
);
80 extern int __initdata acpi_ht
;
81 extern acpi_interrupt_flags acpi_sci_flags
;
82 int __initdata acpi_force
= 0;
85 int acpi_numa __initdata
;
87 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
90 unsigned long saved_video_mode
;
96 char dmi_alloc_data
[DMI_MAX_DATA
];
101 struct screen_info screen_info
;
102 struct sys_desc_table_struct
{
103 unsigned short length
;
104 unsigned char table
[0];
107 struct edid_info edid_info
;
110 extern int root_mountflags
;
112 char command_line
[COMMAND_LINE_SIZE
];
114 struct resource standard_io_resources
[] = {
115 { .name
= "dma1", .start
= 0x00, .end
= 0x1f,
116 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
117 { .name
= "pic1", .start
= 0x20, .end
= 0x21,
118 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
119 { .name
= "timer0", .start
= 0x40, .end
= 0x43,
120 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
121 { .name
= "timer1", .start
= 0x50, .end
= 0x53,
122 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
123 { .name
= "keyboard", .start
= 0x60, .end
= 0x6f,
124 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
125 { .name
= "dma page reg", .start
= 0x80, .end
= 0x8f,
126 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
127 { .name
= "pic2", .start
= 0xa0, .end
= 0xa1,
128 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
129 { .name
= "dma2", .start
= 0xc0, .end
= 0xdf,
130 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
131 { .name
= "fpu", .start
= 0xf0, .end
= 0xff,
132 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
}
135 #define STANDARD_IO_RESOURCES \
136 (sizeof standard_io_resources / sizeof standard_io_resources[0])
138 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
140 struct resource data_resource
= {
141 .name
= "Kernel data",
144 .flags
= IORESOURCE_RAM
,
146 struct resource code_resource
= {
147 .name
= "Kernel code",
150 .flags
= IORESOURCE_RAM
,
153 #define IORESOURCE_ROM (IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM)
155 static struct resource system_rom_resource
= {
156 .name
= "System ROM",
159 .flags
= IORESOURCE_ROM
,
162 static struct resource extension_rom_resource
= {
163 .name
= "Extension ROM",
166 .flags
= IORESOURCE_ROM
,
169 static struct resource adapter_rom_resources
[] = {
170 { .name
= "Adapter ROM", .start
= 0xc8000, .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
},
178 { .name
= "Adapter ROM", .start
= 0, .end
= 0,
179 .flags
= IORESOURCE_ROM
},
180 { .name
= "Adapter ROM", .start
= 0, .end
= 0,
181 .flags
= IORESOURCE_ROM
}
184 #define ADAPTER_ROM_RESOURCES \
185 (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
187 static struct resource video_rom_resource
= {
191 .flags
= IORESOURCE_ROM
,
194 static struct resource video_ram_resource
= {
195 .name
= "Video RAM area",
198 .flags
= IORESOURCE_RAM
,
201 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
203 static int __init
romchecksum(unsigned char *rom
, unsigned long length
)
205 unsigned char *p
, sum
= 0;
207 for (p
= rom
; p
< rom
+ length
; p
++)
212 static void __init
probe_roms(void)
214 unsigned long start
, length
, upper
;
219 upper
= adapter_rom_resources
[0].start
;
220 for (start
= video_rom_resource
.start
; start
< upper
; start
+= 2048) {
221 rom
= isa_bus_to_virt(start
);
222 if (!romsignature(rom
))
225 video_rom_resource
.start
= start
;
227 /* 0 < length <= 0x7f * 512, historically */
228 length
= rom
[2] * 512;
230 /* if checksum okay, trust length byte */
231 if (length
&& romchecksum(rom
, length
))
232 video_rom_resource
.end
= start
+ length
- 1;
234 request_resource(&iomem_resource
, &video_rom_resource
);
238 start
= (video_rom_resource
.end
+ 1 + 2047) & ~2047UL;
243 request_resource(&iomem_resource
, &system_rom_resource
);
244 upper
= system_rom_resource
.start
;
246 /* check for extension rom (ignore length byte!) */
247 rom
= isa_bus_to_virt(extension_rom_resource
.start
);
248 if (romsignature(rom
)) {
249 length
= extension_rom_resource
.end
- extension_rom_resource
.start
+ 1;
250 if (romchecksum(rom
, length
)) {
251 request_resource(&iomem_resource
, &extension_rom_resource
);
252 upper
= extension_rom_resource
.start
;
256 /* check for adapter roms on 2k boundaries */
257 for (i
= 0; i
< ADAPTER_ROM_RESOURCES
&& start
< upper
; start
+= 2048) {
258 rom
= isa_bus_to_virt(start
);
259 if (!romsignature(rom
))
262 /* 0 < length <= 0x7f * 512, historically */
263 length
= rom
[2] * 512;
265 /* but accept any length that fits if checksum okay */
266 if (!length
|| start
+ length
> upper
|| !romchecksum(rom
, length
))
269 adapter_rom_resources
[i
].start
= start
;
270 adapter_rom_resources
[i
].end
= start
+ length
- 1;
271 request_resource(&iomem_resource
, &adapter_rom_resources
[i
]);
273 start
= adapter_rom_resources
[i
++].end
& ~2047UL;
277 /* Check for full argument with no trailing characters */
278 static int fullarg(char *p
, char *arg
)
281 return !memcmp(p
, arg
, l
) && (p
[l
] == 0 || isspace(p
[l
]));
284 static __init
void parse_cmdline_early (char ** cmdline_p
)
286 char c
= ' ', *to
= command_line
, *from
= COMMAND_LINE
;
296 * If the BIOS enumerates physical processors before logical,
297 * maxcpus=N at enumeration-time can be used to disable HT.
299 else if (!memcmp(from
, "maxcpus=", 8)) {
300 extern unsigned int maxcpus
;
302 maxcpus
= simple_strtoul(from
+ 8, NULL
, 0);
306 /* "acpi=off" disables both ACPI table parsing and interpreter init */
307 if (fullarg(from
,"acpi=off"))
310 if (fullarg(from
, "acpi=force")) {
311 /* add later when we do DMI horrors: */
316 /* acpi=ht just means: do ACPI MADT parsing
317 at bootup, but don't enable the full ACPI interpreter */
318 if (fullarg(from
, "acpi=ht")) {
323 else if (fullarg(from
, "pci=noacpi"))
325 else if (fullarg(from
, "acpi=noirq"))
328 else if (fullarg(from
, "acpi_sci=edge"))
329 acpi_sci_flags
.trigger
= 1;
330 else if (fullarg(from
, "acpi_sci=level"))
331 acpi_sci_flags
.trigger
= 3;
332 else if (fullarg(from
, "acpi_sci=high"))
333 acpi_sci_flags
.polarity
= 1;
334 else if (fullarg(from
, "acpi_sci=low"))
335 acpi_sci_flags
.polarity
= 3;
337 /* acpi=strict disables out-of-spec workarounds */
338 else if (fullarg(from
, "acpi=strict")) {
341 #ifdef CONFIG_X86_IO_APIC
342 else if (fullarg(from
, "acpi_skip_timer_override"))
343 acpi_skip_timer_override
= 1;
347 if (fullarg(from
, "disable_timer_pin_1"))
348 disable_timer_pin_1
= 1;
349 if (fullarg(from
, "enable_timer_pin_1"))
350 disable_timer_pin_1
= -1;
352 if (fullarg(from
, "nolapic") || fullarg(from
, "disableapic")) {
353 clear_bit(X86_FEATURE_APIC
, boot_cpu_data
.x86_capability
);
357 if (fullarg(from
, "noapic"))
358 skip_ioapic_setup
= 1;
360 if (fullarg(from
,"apic")) {
361 skip_ioapic_setup
= 0;
365 if (!memcmp(from
, "mem=", 4))
366 parse_memopt(from
+4, &from
);
368 if (!memcmp(from
, "memmap=", 7)) {
369 /* exactmap option is for used defined memory */
370 if (!memcmp(from
+7, "exactmap", 8)) {
371 #ifdef CONFIG_CRASH_DUMP
372 /* If we are doing a crash dump, we
373 * still need to know the real mem
374 * size before original memory map is
377 saved_max_pfn
= e820_end_of_ram();
385 parse_memmapopt(from
+7, &from
);
391 if (!memcmp(from
, "numa=", 5))
395 if (!memcmp(from
,"iommu=",6)) {
399 if (fullarg(from
,"oops=panic"))
402 if (!memcmp(from
, "noexec=", 7))
403 nonx_setup(from
+ 7);
406 /* crashkernel=size@addr specifies the location to reserve for
407 * a crash kernel. By reserving this memory we guarantee
408 * that linux never set's it up as a DMA target.
409 * Useful for holding code to do something appropriate
410 * after a kernel panic.
412 else if (!memcmp(from
, "crashkernel=", 12)) {
413 unsigned long size
, base
;
414 size
= memparse(from
+12, &from
);
416 base
= memparse(from
+1, &from
);
417 /* FIXME: Do I want a sanity check
418 * to validate the memory range?
420 crashk_res
.start
= base
;
421 crashk_res
.end
= base
+ size
- 1;
426 #ifdef CONFIG_PROC_VMCORE
427 /* elfcorehdr= specifies the location of elf core header
428 * stored by the crashed kernel. This option will be passed
429 * by kexec loader to the capture kernel.
431 else if(!memcmp(from
, "elfcorehdr=", 11))
432 elfcorehdr_addr
= memparse(from
+11, &from
);
435 #ifdef CONFIG_HOTPLUG_CPU
436 else if (!memcmp(from
, "additional_cpus=", 16))
437 setup_additional_cpus(from
+16);
444 if (COMMAND_LINE_SIZE
<= ++len
)
449 printk(KERN_INFO
"user-defined physical RAM map:\n");
450 e820_print_map("user");
453 *cmdline_p
= command_line
;
458 contig_initmem_init(unsigned long start_pfn
, unsigned long end_pfn
)
460 unsigned long bootmap_size
, bootmap
;
462 bootmap_size
= bootmem_bootmap_pages(end_pfn
)<<PAGE_SHIFT
;
463 bootmap
= find_e820_area(0, end_pfn
<<PAGE_SHIFT
, bootmap_size
);
465 panic("Cannot find bootmem map of size %ld\n",bootmap_size
);
466 bootmap_size
= init_bootmem(bootmap
>> PAGE_SHIFT
, end_pfn
);
467 e820_bootmem_free(NODE_DATA(0), 0, end_pfn
<< PAGE_SHIFT
);
468 reserve_bootmem(bootmap
, bootmap_size
);
472 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
474 #ifdef CONFIG_EDD_MODULE
478 * copy_edd() - Copy the BIOS EDD information
479 * from boot_params into a safe place.
482 static inline void copy_edd(void)
484 memcpy(edd
.mbr_signature
, EDD_MBR_SIGNATURE
, sizeof(edd
.mbr_signature
));
485 memcpy(edd
.edd_info
, EDD_BUF
, sizeof(edd
.edd_info
));
486 edd
.mbr_signature_nr
= EDD_MBR_SIG_NR
;
487 edd
.edd_info_nr
= EDD_NR
;
490 static inline void copy_edd(void)
495 #define EBDA_ADDR_POINTER 0x40E
497 unsigned __initdata ebda_addr
;
498 unsigned __initdata ebda_size
;
500 static void discover_ebda(void)
503 * there is a real-mode segmented pointer pointing to the
504 * 4K EBDA area at 0x40E
506 ebda_addr
= *(unsigned short *)EBDA_ADDR_POINTER
;
509 ebda_size
= *(unsigned short *)(unsigned long)ebda_addr
;
511 /* Round EBDA up to pages */
515 ebda_size
= round_up(ebda_size
+ (ebda_addr
& ~PAGE_MASK
), PAGE_SIZE
);
516 if (ebda_size
> 64*1024)
520 void __init
setup_arch(char **cmdline_p
)
522 unsigned long kernel_end
;
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);
535 setup_memory_region();
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 */
565 init_memory_mapping(0, (end_pfn_map
<< PAGE_SHIFT
));
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();
579 #ifdef CONFIG_ACPI_NUMA
581 * Parse SRAT to discover nodes.
587 numa_initmem_init(0, end_pfn
);
589 contig_initmem_init(0, end_pfn
);
592 /* Reserve direct mapping */
593 reserve_bootmem_generic(table_start
<< PAGE_SHIFT
,
594 (table_end
- table_start
) << PAGE_SHIFT
);
597 kernel_end
= round_up(__pa_symbol(&_end
),PAGE_SIZE
);
598 reserve_bootmem_generic(HIGH_MEMORY
, kernel_end
- HIGH_MEMORY
);
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 */
608 reserve_bootmem_generic(ebda_addr
, ebda_size
);
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
);
622 #ifdef CONFIG_ACPI_SLEEP
624 * Reserve low memory region for sleep support.
626 acpi_reserve_bootmem();
628 #ifdef CONFIG_X86_LOCAL_APIC
630 * Find and reserve possible boot-time SMP configuration:
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
);
639 INITRD_START
? INITRD_START
+ PAGE_OFFSET
: 0;
640 initrd_end
= initrd_start
+INITRD_SIZE
;
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
));
652 if (crashk_res
.start
!= crashk_res
.end
) {
653 reserve_bootmem_generic(crashk_res
.start
,
654 crashk_res
.end
- crashk_res
.start
+ 1);
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
);
670 * Read APIC and some other early information from ACPI tables.
677 #ifdef CONFIG_X86_LOCAL_APIC
679 * get boot-time SMP configuration:
681 if (smp_found_config
)
683 init_apic_mappings();
687 * Request address space for all standard RAM and ROM resources
688 * and also for regions reported as reserved by the e820.
691 e820_reserve_resources();
693 request_resource(&iomem_resource
, &video_ram_resource
);
697 /* request I/O space for devices used on all i[345]86 PCs */
698 for (i
= 0; i
< STANDARD_IO_RESOURCES
; i
++)
699 request_resource(&ioport_resource
, &standard_io_resources
[i
]);
705 #if defined(CONFIG_VGA_CONSOLE)
706 conswitchp
= &vga_con
;
707 #elif defined(CONFIG_DUMMY_CONSOLE)
708 conswitchp
= &dummy_con
;
713 static int __cpuinit
get_model_name(struct cpuinfo_x86
*c
)
717 if (c
->extended_cpuid_level
< 0x80000004)
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;
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 */
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);
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;
764 static int nearby_node(int apicid
)
767 for (i
= apicid
- 1; i
>= 0; i
--) {
768 int node
= apicid_to_node
[i
];
769 if (node
!= NUMA_NO_NODE
&& node_online(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
))
777 return first_node(node_online_map
); /* Shouldn't happen */
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
)
790 int cpu
= smp_processor_id();
792 unsigned apicid
= hard_smp_processor_id();
794 unsigned ecx
= cpuid_ecx(0x80000008);
796 c
->x86_max_cores
= (ecx
& 0xff) + 1;
798 /* CPU telling us the core id bits shift? */
799 bits
= (ecx
>> 12) & 0xF;
801 /* Otherwise recompute */
803 while ((1 << bits
) < c
->x86_max_cores
)
807 /* Low order bits define the core id (index of core in socket) */
808 c
->cpu_core_id
= c
->phys_proc_id
& ((1 << bits
)-1);
809 /* Convert the APIC ID into the socket ID */
810 c
->phys_proc_id
= phys_pkg_id(bits
);
813 node
= c
->phys_proc_id
;
814 if (apicid_to_node
[apicid
] != NUMA_NO_NODE
)
815 node
= apicid_to_node
[apicid
];
816 if (!node_online(node
)) {
817 /* Two possibilities here:
818 - The CPU is missing memory and no node was created.
819 In that case try picking one from a nearby CPU
820 - The APIC IDs differ from the HyperTransport node IDs
821 which the K8 northbridge parsing fills in.
822 Assume they are all increased by a constant offset,
823 but in the same order as the HT nodeids.
824 If that doesn't result in a usable node fall back to the
825 path for the previous case. */
826 int ht_nodeid
= apicid
- (cpu_data
[0].phys_proc_id
<< bits
);
827 if (ht_nodeid
>= 0 &&
828 apicid_to_node
[ht_nodeid
] != NUMA_NO_NODE
)
829 node
= apicid_to_node
[ht_nodeid
];
830 /* Pick a nearby node */
831 if (!node_online(node
))
832 node
= nearby_node(apicid
);
834 numa_set_node(cpu
, node
);
836 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
841 static void __init
init_amd(struct cpuinfo_x86
*c
)
849 * Disable TLB flush filter by setting HWCR.FFDIS on K8
850 * bit 6 of msr C001_0015
852 * Errata 63 for SH-B3 steppings
853 * Errata 122 for all steppings (F+ have it disabled by default)
856 rdmsrl(MSR_K8_HWCR
, value
);
858 wrmsrl(MSR_K8_HWCR
, value
);
862 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
863 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
864 clear_bit(0*32+31, &c
->x86_capability
);
866 /* On C+ stepping K8 rep microcode works well for copy/memset */
867 level
= cpuid_eax(1);
868 if (c
->x86
== 15 && ((level
>= 0x0f48 && level
< 0x0f50) || level
>= 0x0f58))
869 set_bit(X86_FEATURE_REP_GOOD
, &c
->x86_capability
);
871 /* Enable workaround for FXSAVE leak */
873 set_bit(X86_FEATURE_FXSAVE_LEAK
, &c
->x86_capability
);
875 level
= get_model_name(c
);
879 /* Should distinguish Models here, but this is only
880 a fallback anyways. */
881 strcpy(c
->x86_model_id
, "Hammer");
885 display_cacheinfo(c
);
887 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
888 if (c
->x86_power
& (1<<8))
889 set_bit(X86_FEATURE_CONSTANT_TSC
, &c
->x86_capability
);
891 /* Multi core CPU? */
892 if (c
->extended_cpuid_level
>= 0x80000008)
895 /* Fix cpuid4 emulation for more */
896 num_cache_leaves
= 3;
899 static void __cpuinit
detect_ht(struct cpuinfo_x86
*c
)
902 u32 eax
, ebx
, ecx
, edx
;
903 int index_msb
, core_bits
;
905 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
908 if (!cpu_has(c
, X86_FEATURE_HT
))
910 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
913 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
915 if (smp_num_siblings
== 1) {
916 printk(KERN_INFO
"CPU: Hyper-Threading is disabled\n");
917 } else if (smp_num_siblings
> 1 ) {
919 if (smp_num_siblings
> NR_CPUS
) {
920 printk(KERN_WARNING
"CPU: Unsupported number of the siblings %d", smp_num_siblings
);
921 smp_num_siblings
= 1;
925 index_msb
= get_count_order(smp_num_siblings
);
926 c
->phys_proc_id
= phys_pkg_id(index_msb
);
928 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
930 index_msb
= get_count_order(smp_num_siblings
) ;
932 core_bits
= get_count_order(c
->x86_max_cores
);
934 c
->cpu_core_id
= phys_pkg_id(index_msb
) &
935 ((1 << core_bits
) - 1);
938 if ((c
->x86_max_cores
* smp_num_siblings
) > 1) {
939 printk(KERN_INFO
"CPU: Physical Processor ID: %d\n", c
->phys_proc_id
);
940 printk(KERN_INFO
"CPU: Processor Core ID: %d\n", c
->cpu_core_id
);
947 * find out the number of processor cores on the die
949 static int __cpuinit
intel_num_cpu_cores(struct cpuinfo_x86
*c
)
953 if (c
->cpuid_level
< 4)
956 cpuid_count(4, 0, &eax
, &t
, &t
, &t
);
959 return ((eax
>> 26) + 1);
964 static void srat_detect_node(void)
968 int cpu
= smp_processor_id();
969 int apicid
= hard_smp_processor_id();
971 /* Don't do the funky fallback heuristics the AMD version employs
973 node
= apicid_to_node
[apicid
];
974 if (node
== NUMA_NO_NODE
)
975 node
= first_node(node_online_map
);
976 numa_set_node(cpu
, node
);
979 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
983 static void __cpuinit
init_intel(struct cpuinfo_x86
*c
)
988 init_intel_cacheinfo(c
);
989 n
= c
->extended_cpuid_level
;
990 if (n
>= 0x80000008) {
991 unsigned eax
= cpuid_eax(0x80000008);
992 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
993 c
->x86_phys_bits
= eax
& 0xff;
994 /* CPUID workaround for Intel 0F34 CPU */
995 if (c
->x86_vendor
== X86_VENDOR_INTEL
&&
996 c
->x86
== 0xF && c
->x86_model
== 0x3 &&
998 c
->x86_phys_bits
= 36;
1002 c
->x86_cache_alignment
= c
->x86_clflush_size
* 2;
1003 if ((c
->x86
== 0xf && c
->x86_model
>= 0x03) ||
1004 (c
->x86
== 0x6 && c
->x86_model
>= 0x0e))
1005 set_bit(X86_FEATURE_CONSTANT_TSC
, &c
->x86_capability
);
1006 set_bit(X86_FEATURE_SYNC_RDTSC
, &c
->x86_capability
);
1007 c
->x86_max_cores
= intel_num_cpu_cores(c
);
1012 static void __cpuinit
get_cpu_vendor(struct cpuinfo_x86
*c
)
1014 char *v
= c
->x86_vendor_id
;
1016 if (!strcmp(v
, "AuthenticAMD"))
1017 c
->x86_vendor
= X86_VENDOR_AMD
;
1018 else if (!strcmp(v
, "GenuineIntel"))
1019 c
->x86_vendor
= X86_VENDOR_INTEL
;
1021 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
1024 struct cpu_model_info
{
1027 char *model_names
[16];
1030 /* Do some early cpuid on the boot CPU to get some parameter that are
1031 needed before check_bugs. Everything advanced is in identify_cpu
1033 void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
)
1037 c
->loops_per_jiffy
= loops_per_jiffy
;
1038 c
->x86_cache_size
= -1;
1039 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
1040 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
1041 c
->x86_vendor_id
[0] = '\0'; /* Unset */
1042 c
->x86_model_id
[0] = '\0'; /* Unset */
1043 c
->x86_clflush_size
= 64;
1044 c
->x86_cache_alignment
= c
->x86_clflush_size
;
1045 c
->x86_max_cores
= 1;
1046 c
->extended_cpuid_level
= 0;
1047 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
1049 /* Get vendor name */
1050 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
1051 (unsigned int *)&c
->x86_vendor_id
[0],
1052 (unsigned int *)&c
->x86_vendor_id
[8],
1053 (unsigned int *)&c
->x86_vendor_id
[4]);
1057 /* Initialize the standard set of capabilities */
1058 /* Note that the vendor-specific code below might override */
1060 /* Intel-defined flags: level 0x00000001 */
1061 if (c
->cpuid_level
>= 0x00000001) {
1063 cpuid(0x00000001, &tfms
, &misc
, &c
->x86_capability
[4],
1064 &c
->x86_capability
[0]);
1065 c
->x86
= (tfms
>> 8) & 0xf;
1066 c
->x86_model
= (tfms
>> 4) & 0xf;
1067 c
->x86_mask
= tfms
& 0xf;
1069 c
->x86
+= (tfms
>> 20) & 0xff;
1071 c
->x86_model
+= ((tfms
>> 16) & 0xF) << 4;
1072 if (c
->x86_capability
[0] & (1<<19))
1073 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
1075 /* Have CPUID level 0 only - unheard of */
1080 c
->phys_proc_id
= (cpuid_ebx(1) >> 24) & 0xff;
1085 * This does the hard work of actually picking apart the CPU stuff...
1087 void __cpuinit
identify_cpu(struct cpuinfo_x86
*c
)
1092 early_identify_cpu(c
);
1094 /* AMD-defined flags: level 0x80000001 */
1095 xlvl
= cpuid_eax(0x80000000);
1096 c
->extended_cpuid_level
= xlvl
;
1097 if ((xlvl
& 0xffff0000) == 0x80000000) {
1098 if (xlvl
>= 0x80000001) {
1099 c
->x86_capability
[1] = cpuid_edx(0x80000001);
1100 c
->x86_capability
[6] = cpuid_ecx(0x80000001);
1102 if (xlvl
>= 0x80000004)
1103 get_model_name(c
); /* Default name */
1106 /* Transmeta-defined flags: level 0x80860001 */
1107 xlvl
= cpuid_eax(0x80860000);
1108 if ((xlvl
& 0xffff0000) == 0x80860000) {
1109 /* Don't set x86_cpuid_level here for now to not confuse. */
1110 if (xlvl
>= 0x80860001)
1111 c
->x86_capability
[2] = cpuid_edx(0x80860001);
1114 c
->apicid
= phys_pkg_id(0);
1117 * Vendor-specific initialization. In this section we
1118 * canonicalize the feature flags, meaning if there are
1119 * features a certain CPU supports which CPUID doesn't
1120 * tell us, CPUID claiming incorrect flags, or other bugs,
1121 * we handle them here.
1123 * At the end of this section, c->x86_capability better
1124 * indicate the features this CPU genuinely supports!
1126 switch (c
->x86_vendor
) {
1127 case X86_VENDOR_AMD
:
1131 case X86_VENDOR_INTEL
:
1135 case X86_VENDOR_UNKNOWN
:
1137 display_cacheinfo(c
);
1141 select_idle_routine(c
);
1145 * On SMP, boot_cpu_data holds the common feature set between
1146 * all CPUs; so make sure that we indicate which features are
1147 * common between the CPUs. The first time this routine gets
1148 * executed, c == &boot_cpu_data.
1150 if (c
!= &boot_cpu_data
) {
1151 /* AND the already accumulated flags with these */
1152 for (i
= 0 ; i
< NCAPINTS
; i
++)
1153 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
1156 #ifdef CONFIG_X86_MCE
1159 if (c
== &boot_cpu_data
)
1164 numa_add_cpu(smp_processor_id());
1169 void __cpuinit
print_cpu_info(struct cpuinfo_x86
*c
)
1171 if (c
->x86_model_id
[0])
1172 printk("%s", c
->x86_model_id
);
1174 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1175 printk(" stepping %02x\n", c
->x86_mask
);
1181 * Get CPU information for use by the procfs.
1184 static int show_cpuinfo(struct seq_file
*m
, void *v
)
1186 struct cpuinfo_x86
*c
= v
;
1189 * These flag bits must match the definitions in <asm/cpufeature.h>.
1190 * NULL means this bit is undefined or reserved; either way it doesn't
1191 * have meaning as far as Linux is concerned. Note that it's important
1192 * to realize there is a difference between this table and CPUID -- if
1193 * applications want to get the raw CPUID data, they should access
1194 * /dev/cpu/<cpu_nr>/cpuid instead.
1196 static char *x86_cap_flags
[] = {
1198 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1199 "cx8", "apic", NULL
, "sep", "mtrr", "pge", "mca", "cmov",
1200 "pat", "pse36", "pn", "clflush", NULL
, "dts", "acpi", "mmx",
1201 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", NULL
,
1204 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1205 NULL
, NULL
, NULL
, "syscall", NULL
, NULL
, NULL
, NULL
,
1206 NULL
, NULL
, NULL
, NULL
, "nx", NULL
, "mmxext", NULL
,
1207 NULL
, "fxsr_opt", NULL
, "rdtscp", NULL
, "lm", "3dnowext", "3dnow",
1209 /* Transmeta-defined */
1210 "recovery", "longrun", NULL
, "lrti", NULL
, NULL
, 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 /* Other (Linux-defined) */
1216 "cxmmx", NULL
, "cyrix_arr", "centaur_mcr", NULL
,
1217 "constant_tsc", NULL
, NULL
,
1218 "up", NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1219 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1220 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1222 /* Intel-defined (#2) */
1223 "pni", NULL
, NULL
, "monitor", "ds_cpl", "vmx", "smx", "est",
1224 "tm2", NULL
, "cid", NULL
, NULL
, "cx16", "xtpr", NULL
,
1225 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1226 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1228 /* VIA/Cyrix/Centaur-defined */
1229 NULL
, NULL
, "rng", "rng_en", NULL
, NULL
, "ace", "ace_en",
1230 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1231 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1232 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1234 /* AMD-defined (#2) */
1235 "lahf_lm", "cmp_legacy", "svm", NULL
, "cr8_legacy", NULL
, NULL
, NULL
,
1236 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1237 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1238 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1240 static char *x86_power_flags
[] = {
1241 "ts", /* temperature sensor */
1242 "fid", /* frequency id control */
1243 "vid", /* voltage id control */
1244 "ttp", /* thermal trip */
1248 /* nothing */ /* constant_tsc - moved to flags */
1253 if (!cpu_online(c
-cpu_data
))
1257 seq_printf(m
,"processor\t: %u\n"
1259 "cpu family\t: %d\n"
1261 "model name\t: %s\n",
1262 (unsigned)(c
-cpu_data
),
1263 c
->x86_vendor_id
[0] ? c
->x86_vendor_id
: "unknown",
1266 c
->x86_model_id
[0] ? c
->x86_model_id
: "unknown");
1268 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1269 seq_printf(m
, "stepping\t: %d\n", c
->x86_mask
);
1271 seq_printf(m
, "stepping\t: unknown\n");
1273 if (cpu_has(c
,X86_FEATURE_TSC
)) {
1274 unsigned int freq
= cpufreq_quick_get((unsigned)(c
-cpu_data
));
1277 seq_printf(m
, "cpu MHz\t\t: %u.%03u\n",
1278 freq
/ 1000, (freq
% 1000));
1282 if (c
->x86_cache_size
>= 0)
1283 seq_printf(m
, "cache size\t: %d KB\n", c
->x86_cache_size
);
1286 if (smp_num_siblings
* c
->x86_max_cores
> 1) {
1287 int cpu
= c
- cpu_data
;
1288 seq_printf(m
, "physical id\t: %d\n", c
->phys_proc_id
);
1289 seq_printf(m
, "siblings\t: %d\n", cpus_weight(cpu_core_map
[cpu
]));
1290 seq_printf(m
, "core id\t\t: %d\n", c
->cpu_core_id
);
1291 seq_printf(m
, "cpu cores\t: %d\n", c
->booted_cores
);
1297 "fpu_exception\t: yes\n"
1298 "cpuid level\t: %d\n"
1305 for ( i
= 0 ; i
< 32*NCAPINTS
; i
++ )
1306 if (cpu_has(c
, i
) && x86_cap_flags
[i
] != NULL
)
1307 seq_printf(m
, " %s", x86_cap_flags
[i
]);
1310 seq_printf(m
, "\nbogomips\t: %lu.%02lu\n",
1311 c
->loops_per_jiffy
/(500000/HZ
),
1312 (c
->loops_per_jiffy
/(5000/HZ
)) % 100);
1314 if (c
->x86_tlbsize
> 0)
1315 seq_printf(m
, "TLB size\t: %d 4K pages\n", c
->x86_tlbsize
);
1316 seq_printf(m
, "clflush size\t: %d\n", c
->x86_clflush_size
);
1317 seq_printf(m
, "cache_alignment\t: %d\n", c
->x86_cache_alignment
);
1319 seq_printf(m
, "address sizes\t: %u bits physical, %u bits virtual\n",
1320 c
->x86_phys_bits
, c
->x86_virt_bits
);
1322 seq_printf(m
, "power management:");
1325 for (i
= 0; i
< 32; i
++)
1326 if (c
->x86_power
& (1 << i
)) {
1327 if (i
< ARRAY_SIZE(x86_power_flags
) &&
1329 seq_printf(m
, "%s%s",
1330 x86_power_flags
[i
][0]?" ":"",
1331 x86_power_flags
[i
]);
1333 seq_printf(m
, " [%d]", i
);
1337 seq_printf(m
, "\n\n");
1342 static void *c_start(struct seq_file
*m
, loff_t
*pos
)
1344 return *pos
< NR_CPUS
? cpu_data
+ *pos
: NULL
;
1347 static void *c_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1350 return c_start(m
, pos
);
1353 static void c_stop(struct seq_file
*m
, void *v
)
1357 struct seq_operations cpuinfo_op
= {
1361 .show
= show_cpuinfo
,
1364 #if defined(CONFIG_INPUT_PCSPKR) || defined(CONFIG_INPUT_PCSPKR_MODULE)
1365 #include <linux/platform_device.h>
1366 static __init
int add_pcspkr(void)
1368 struct platform_device
*pd
;
1371 pd
= platform_device_alloc("pcspkr", -1);
1375 ret
= platform_device_add(pd
);
1377 platform_device_put(pd
);
1381 device_initcall(add_pcspkr
);