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/swiotlb.h>
67 #include <asm/sections.h>
68 #include <asm/gart-mapping.h>
75 struct cpuinfo_x86 boot_cpu_data __read_mostly
;
77 unsigned long mmu_cr4_features
;
80 EXPORT_SYMBOL(acpi_disabled
);
82 extern int __initdata acpi_ht
;
83 extern acpi_interrupt_flags acpi_sci_flags
;
84 int __initdata acpi_force
= 0;
87 int acpi_numa __initdata
;
89 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
92 unsigned long saved_video_mode
;
98 char dmi_alloc_data
[DMI_MAX_DATA
];
103 struct screen_info screen_info
;
104 struct sys_desc_table_struct
{
105 unsigned short length
;
106 unsigned char table
[0];
109 struct edid_info edid_info
;
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",
146 .flags
= IORESOURCE_RAM
,
148 struct resource code_resource
= {
149 .name
= "Kernel code",
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",
161 .flags
= IORESOURCE_ROM
,
164 static struct resource extension_rom_resource
= {
165 .name
= "Extension ROM",
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
= {
193 .flags
= IORESOURCE_ROM
,
196 static struct resource video_ram_resource
= {
197 .name
= "Video RAM area",
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
++)
214 static void __init
probe_roms(void)
216 unsigned long start
, length
, upper
;
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
))
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
);
240 start
= (video_rom_resource
.end
+ 1 + 2047) & ~2047UL;
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
))
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
))
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
)
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
;
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);
308 /* "acpi=off" disables both ACPI table parsing and interpreter init */
309 if (fullarg(from
,"acpi=off"))
312 if (fullarg(from
, "acpi=force")) {
313 /* add later when we do DMI horrors: */
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")) {
325 else if (fullarg(from
, "pci=noacpi"))
327 else if (fullarg(from
, "acpi=noirq"))
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")) {
343 #ifdef CONFIG_X86_IO_APIC
344 else if (fullarg(from
, "acpi_skip_timer_override"))
345 acpi_skip_timer_override
= 1;
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
);
359 if (fullarg(from
, "noapic"))
360 skip_ioapic_setup
= 1;
362 if (fullarg(from
,"apic")) {
363 skip_ioapic_setup
= 0;
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
379 saved_max_pfn
= e820_end_of_ram();
387 parse_memmapopt(from
+7, &from
);
393 if (!memcmp(from
, "numa=", 5))
397 if (!memcmp(from
,"iommu=",6)) {
401 if (fullarg(from
,"oops=panic"))
404 if (!memcmp(from
, "noexec=", 7))
405 nonx_setup(from
+ 7);
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
);
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;
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
);
437 #ifdef CONFIG_HOTPLUG_CPU
438 else if (!memcmp(from
, "additional_cpus=", 16))
439 setup_additional_cpus(from
+16);
446 if (COMMAND_LINE_SIZE
<= ++len
)
451 printk(KERN_INFO
"user-defined physical RAM map:\n");
452 e820_print_map("user");
455 *cmdline_p
= command_line
;
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
);
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
);
474 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
476 #ifdef CONFIG_EDD_MODULE
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
;
492 static inline void copy_edd(void)
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
;
511 ebda_size
= *(unsigned short *)(unsigned long)ebda_addr
;
513 /* Round EBDA up to pages */
517 ebda_size
= round_up(ebda_size
+ (ebda_addr
& ~PAGE_MASK
), PAGE_SIZE
);
518 if (ebda_size
> 64*1024)
522 void __init
setup_arch(char **cmdline_p
)
524 unsigned long kernel_end
;
526 ROOT_DEV
= old_decode_dev(ORIG_ROOT_DEV
);
527 screen_info
= SCREEN_INFO
;
528 edid_info
= EDID_INFO
;
529 saved_video_mode
= SAVED_VIDEO_MODE
;
530 bootloader_type
= LOADER_TYPE
;
532 #ifdef CONFIG_BLK_DEV_RAM
533 rd_image_start
= RAMDISK_FLAGS
& RAMDISK_IMAGE_START_MASK
;
534 rd_prompt
= ((RAMDISK_FLAGS
& RAMDISK_PROMPT_FLAG
) != 0);
535 rd_doload
= ((RAMDISK_FLAGS
& RAMDISK_LOAD_FLAG
) != 0);
537 setup_memory_region();
540 if (!MOUNT_ROOT_RDONLY
)
541 root_mountflags
&= ~MS_RDONLY
;
542 init_mm
.start_code
= (unsigned long) &_text
;
543 init_mm
.end_code
= (unsigned long) &_etext
;
544 init_mm
.end_data
= (unsigned long) &_edata
;
545 init_mm
.brk
= (unsigned long) &_end
;
547 code_resource
.start
= virt_to_phys(&_text
);
548 code_resource
.end
= virt_to_phys(&_etext
)-1;
549 data_resource
.start
= virt_to_phys(&_etext
);
550 data_resource
.end
= virt_to_phys(&_edata
)-1;
552 parse_cmdline_early(cmdline_p
);
554 early_identify_cpu(&boot_cpu_data
);
557 * partially used pages are not usable - thus
558 * we are rounding upwards:
560 end_pfn
= e820_end_of_ram();
561 num_physpages
= end_pfn
; /* for pfn_valid */
567 init_memory_mapping(0, (end_pfn_map
<< PAGE_SHIFT
));
575 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
576 * Call this early for SRAT node setup.
578 acpi_boot_table_init();
581 #ifdef CONFIG_ACPI_NUMA
583 * Parse SRAT to discover nodes.
589 numa_initmem_init(0, end_pfn
);
591 contig_initmem_init(0, end_pfn
);
594 /* Reserve direct mapping */
595 reserve_bootmem_generic(table_start
<< PAGE_SHIFT
,
596 (table_end
- table_start
) << PAGE_SHIFT
);
599 kernel_end
= round_up(__pa_symbol(&_end
),PAGE_SIZE
);
600 reserve_bootmem_generic(HIGH_MEMORY
, kernel_end
- HIGH_MEMORY
);
603 * reserve physical page 0 - it's a special BIOS page on many boxes,
604 * enabling clean reboots, SMP operation, laptop functions.
606 reserve_bootmem_generic(0, PAGE_SIZE
);
608 /* reserve ebda region */
610 reserve_bootmem_generic(ebda_addr
, ebda_size
);
614 * But first pinch a few for the stack/trampoline stuff
615 * FIXME: Don't need the extra page at 4K, but need to fix
616 * trampoline before removing it. (see the GDT stuff)
618 reserve_bootmem_generic(PAGE_SIZE
, PAGE_SIZE
);
620 /* Reserve SMP trampoline */
621 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE
, PAGE_SIZE
);
624 #ifdef CONFIG_ACPI_SLEEP
626 * Reserve low memory region for sleep support.
628 acpi_reserve_bootmem();
630 #ifdef CONFIG_X86_LOCAL_APIC
632 * Find and reserve possible boot-time SMP configuration:
636 #ifdef CONFIG_BLK_DEV_INITRD
637 if (LOADER_TYPE
&& INITRD_START
) {
638 if (INITRD_START
+ INITRD_SIZE
<= (end_pfn
<< PAGE_SHIFT
)) {
639 reserve_bootmem_generic(INITRD_START
, INITRD_SIZE
);
641 INITRD_START
? INITRD_START
+ PAGE_OFFSET
: 0;
642 initrd_end
= initrd_start
+INITRD_SIZE
;
645 printk(KERN_ERR
"initrd extends beyond end of memory "
646 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
647 (unsigned long)(INITRD_START
+ INITRD_SIZE
),
648 (unsigned long)(end_pfn
<< PAGE_SHIFT
));
654 if (crashk_res
.start
!= crashk_res
.end
) {
655 reserve_bootmem_generic(crashk_res
.start
,
656 crashk_res
.end
- crashk_res
.start
+ 1);
665 * set this early, so we dont allocate cpu0
666 * if MADT list doesnt list BSP first
667 * mpparse.c/MP_processor_info() allocates logical cpu numbers.
669 cpu_set(0, cpu_present_map
);
672 * Read APIC and some other early information from ACPI tables.
679 #ifdef CONFIG_X86_LOCAL_APIC
681 * get boot-time SMP configuration:
683 if (smp_found_config
)
685 init_apic_mappings();
689 * Request address space for all standard RAM and ROM resources
690 * and also for regions reported as reserved by the e820.
693 e820_reserve_resources();
695 request_resource(&iomem_resource
, &video_ram_resource
);
699 /* request I/O space for devices used on all i[345]86 PCs */
700 for (i
= 0; i
< STANDARD_IO_RESOURCES
; i
++)
701 request_resource(&ioport_resource
, &standard_io_resources
[i
]);
706 #ifdef CONFIG_GART_IOMMU
711 #if defined(CONFIG_VGA_CONSOLE)
712 conswitchp
= &vga_con
;
713 #elif defined(CONFIG_DUMMY_CONSOLE)
714 conswitchp
= &dummy_con
;
719 static int __cpuinit
get_model_name(struct cpuinfo_x86
*c
)
723 if (c
->extended_cpuid_level
< 0x80000004)
726 v
= (unsigned int *) c
->x86_model_id
;
727 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
728 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
729 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
730 c
->x86_model_id
[48] = 0;
735 static void __cpuinit
display_cacheinfo(struct cpuinfo_x86
*c
)
737 unsigned int n
, dummy
, eax
, ebx
, ecx
, edx
;
739 n
= c
->extended_cpuid_level
;
741 if (n
>= 0x80000005) {
742 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
743 printk(KERN_INFO
"CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
744 edx
>>24, edx
&0xFF, ecx
>>24, ecx
&0xFF);
745 c
->x86_cache_size
=(ecx
>>24)+(edx
>>24);
746 /* On K8 L1 TLB is inclusive, so don't count it */
750 if (n
>= 0x80000006) {
751 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
752 ecx
= cpuid_ecx(0x80000006);
753 c
->x86_cache_size
= ecx
>> 16;
754 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
756 printk(KERN_INFO
"CPU: L2 Cache: %dK (%d bytes/line)\n",
757 c
->x86_cache_size
, ecx
& 0xFF);
761 cpuid(0x80000007, &dummy
, &dummy
, &dummy
, &c
->x86_power
);
762 if (n
>= 0x80000008) {
763 cpuid(0x80000008, &eax
, &dummy
, &dummy
, &dummy
);
764 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
765 c
->x86_phys_bits
= eax
& 0xff;
770 static int nearby_node(int apicid
)
773 for (i
= apicid
- 1; i
>= 0; i
--) {
774 int node
= apicid_to_node
[i
];
775 if (node
!= NUMA_NO_NODE
&& node_online(node
))
778 for (i
= apicid
+ 1; i
< MAX_LOCAL_APIC
; i
++) {
779 int node
= apicid_to_node
[i
];
780 if (node
!= NUMA_NO_NODE
&& node_online(node
))
783 return first_node(node_online_map
); /* Shouldn't happen */
788 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
789 * Assumes number of cores is a power of two.
791 static void __init
amd_detect_cmp(struct cpuinfo_x86
*c
)
794 int cpu
= smp_processor_id();
798 unsigned apicid
= hard_smp_processor_id();
800 unsigned ecx
= cpuid_ecx(0x80000008);
802 c
->x86_max_cores
= (ecx
& 0xff) + 1;
804 /* CPU telling us the core id bits shift? */
805 bits
= (ecx
>> 12) & 0xF;
807 /* Otherwise recompute */
809 while ((1 << bits
) < c
->x86_max_cores
)
813 /* Low order bits define the core id (index of core in socket) */
814 cpu_core_id
[cpu
] = phys_proc_id
[cpu
] & ((1 << bits
)-1);
815 /* Convert the APIC ID into the socket ID */
816 phys_proc_id
[cpu
] = phys_pkg_id(bits
);
819 node
= phys_proc_id
[cpu
];
820 if (apicid_to_node
[apicid
] != NUMA_NO_NODE
)
821 node
= apicid_to_node
[apicid
];
822 if (!node_online(node
)) {
823 /* Two possibilities here:
824 - The CPU is missing memory and no node was created.
825 In that case try picking one from a nearby CPU
826 - The APIC IDs differ from the HyperTransport node IDs
827 which the K8 northbridge parsing fills in.
828 Assume they are all increased by a constant offset,
829 but in the same order as the HT nodeids.
830 If that doesn't result in a usable node fall back to the
831 path for the previous case. */
832 int ht_nodeid
= apicid
- (phys_proc_id
[0] << bits
);
833 if (ht_nodeid
>= 0 &&
834 apicid_to_node
[ht_nodeid
] != NUMA_NO_NODE
)
835 node
= apicid_to_node
[ht_nodeid
];
836 /* Pick a nearby node */
837 if (!node_online(node
))
838 node
= nearby_node(apicid
);
840 numa_set_node(cpu
, node
);
842 printk(KERN_INFO
"CPU %d/%x(%d) -> Node %d -> Core %d\n",
843 cpu
, apicid
, c
->x86_max_cores
, node
, cpu_core_id
[cpu
]);
848 static int __init
init_amd(struct cpuinfo_x86
*c
)
857 * Disable TLB flush filter by setting HWCR.FFDIS on K8
858 * bit 6 of msr C001_0015
860 * Errata 63 for SH-B3 steppings
861 * Errata 122 for all steppings (F+ have it disabled by default)
864 rdmsrl(MSR_K8_HWCR
, value
);
866 wrmsrl(MSR_K8_HWCR
, value
);
870 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
871 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
872 clear_bit(0*32+31, &c
->x86_capability
);
874 /* On C+ stepping K8 rep microcode works well for copy/memset */
875 level
= cpuid_eax(1);
876 if (c
->x86
== 15 && ((level
>= 0x0f48 && level
< 0x0f50) || level
>= 0x0f58))
877 set_bit(X86_FEATURE_REP_GOOD
, &c
->x86_capability
);
879 /* Enable workaround for FXSAVE leak */
881 set_bit(X86_FEATURE_FXSAVE_LEAK
, &c
->x86_capability
);
883 r
= get_model_name(c
);
887 /* Should distinguish Models here, but this is only
888 a fallback anyways. */
889 strcpy(c
->x86_model_id
, "Hammer");
893 display_cacheinfo(c
);
895 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
896 if (c
->x86_power
& (1<<8))
897 set_bit(X86_FEATURE_CONSTANT_TSC
, &c
->x86_capability
);
899 /* Multi core CPU? */
900 if (c
->extended_cpuid_level
>= 0x80000008)
903 /* Fix cpuid4 emulation for more */
904 num_cache_leaves
= 3;
909 static void __cpuinit
detect_ht(struct cpuinfo_x86
*c
)
912 u32 eax
, ebx
, ecx
, edx
;
913 int index_msb
, core_bits
;
914 int cpu
= smp_processor_id();
916 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
919 if (!cpu_has(c
, X86_FEATURE_HT
) || cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
922 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
924 if (smp_num_siblings
== 1) {
925 printk(KERN_INFO
"CPU: Hyper-Threading is disabled\n");
926 } else if (smp_num_siblings
> 1 ) {
928 if (smp_num_siblings
> NR_CPUS
) {
929 printk(KERN_WARNING
"CPU: Unsupported number of the siblings %d", smp_num_siblings
);
930 smp_num_siblings
= 1;
934 index_msb
= get_count_order(smp_num_siblings
);
935 phys_proc_id
[cpu
] = phys_pkg_id(index_msb
);
937 printk(KERN_INFO
"CPU: Physical Processor ID: %d\n",
940 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
942 index_msb
= get_count_order(smp_num_siblings
) ;
944 core_bits
= get_count_order(c
->x86_max_cores
);
946 cpu_core_id
[cpu
] = phys_pkg_id(index_msb
) &
947 ((1 << core_bits
) - 1);
949 if (c
->x86_max_cores
> 1)
950 printk(KERN_INFO
"CPU: Processor Core ID: %d\n",
957 * find out the number of processor cores on the die
959 static int __cpuinit
intel_num_cpu_cores(struct cpuinfo_x86
*c
)
963 if (c
->cpuid_level
< 4)
972 return ((eax
>> 26) + 1);
977 static void srat_detect_node(void)
981 int cpu
= smp_processor_id();
983 /* Don't do the funky fallback heuristics the AMD version employs
985 node
= apicid_to_node
[hard_smp_processor_id()];
986 if (node
== NUMA_NO_NODE
)
987 node
= first_node(node_online_map
);
988 numa_set_node(cpu
, node
);
991 printk(KERN_INFO
"CPU %d -> Node %d\n", cpu
, node
);
995 static void __cpuinit
init_intel(struct cpuinfo_x86
*c
)
1000 init_intel_cacheinfo(c
);
1001 n
= c
->extended_cpuid_level
;
1002 if (n
>= 0x80000008) {
1003 unsigned eax
= cpuid_eax(0x80000008);
1004 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
1005 c
->x86_phys_bits
= eax
& 0xff;
1006 /* CPUID workaround for Intel 0F34 CPU */
1007 if (c
->x86_vendor
== X86_VENDOR_INTEL
&&
1008 c
->x86
== 0xF && c
->x86_model
== 0x3 &&
1010 c
->x86_phys_bits
= 36;
1014 c
->x86_cache_alignment
= c
->x86_clflush_size
* 2;
1015 if ((c
->x86
== 0xf && c
->x86_model
>= 0x03) ||
1016 (c
->x86
== 0x6 && c
->x86_model
>= 0x0e))
1017 set_bit(X86_FEATURE_CONSTANT_TSC
, &c
->x86_capability
);
1018 set_bit(X86_FEATURE_SYNC_RDTSC
, &c
->x86_capability
);
1019 c
->x86_max_cores
= intel_num_cpu_cores(c
);
1024 static void __cpuinit
get_cpu_vendor(struct cpuinfo_x86
*c
)
1026 char *v
= c
->x86_vendor_id
;
1028 if (!strcmp(v
, "AuthenticAMD"))
1029 c
->x86_vendor
= X86_VENDOR_AMD
;
1030 else if (!strcmp(v
, "GenuineIntel"))
1031 c
->x86_vendor
= X86_VENDOR_INTEL
;
1033 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
1036 struct cpu_model_info
{
1039 char *model_names
[16];
1042 /* Do some early cpuid on the boot CPU to get some parameter that are
1043 needed before check_bugs. Everything advanced is in identify_cpu
1045 void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
)
1049 c
->loops_per_jiffy
= loops_per_jiffy
;
1050 c
->x86_cache_size
= -1;
1051 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
1052 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
1053 c
->x86_vendor_id
[0] = '\0'; /* Unset */
1054 c
->x86_model_id
[0] = '\0'; /* Unset */
1055 c
->x86_clflush_size
= 64;
1056 c
->x86_cache_alignment
= c
->x86_clflush_size
;
1057 c
->x86_max_cores
= 1;
1058 c
->extended_cpuid_level
= 0;
1059 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
1061 /* Get vendor name */
1062 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
1063 (unsigned int *)&c
->x86_vendor_id
[0],
1064 (unsigned int *)&c
->x86_vendor_id
[8],
1065 (unsigned int *)&c
->x86_vendor_id
[4]);
1069 /* Initialize the standard set of capabilities */
1070 /* Note that the vendor-specific code below might override */
1072 /* Intel-defined flags: level 0x00000001 */
1073 if (c
->cpuid_level
>= 0x00000001) {
1075 cpuid(0x00000001, &tfms
, &misc
, &c
->x86_capability
[4],
1076 &c
->x86_capability
[0]);
1077 c
->x86
= (tfms
>> 8) & 0xf;
1078 c
->x86_model
= (tfms
>> 4) & 0xf;
1079 c
->x86_mask
= tfms
& 0xf;
1081 c
->x86
+= (tfms
>> 20) & 0xff;
1083 c
->x86_model
+= ((tfms
>> 16) & 0xF) << 4;
1084 if (c
->x86_capability
[0] & (1<<19))
1085 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
1087 /* Have CPUID level 0 only - unheard of */
1092 phys_proc_id
[smp_processor_id()] = (cpuid_ebx(1) >> 24) & 0xff;
1097 * This does the hard work of actually picking apart the CPU stuff...
1099 void __cpuinit
identify_cpu(struct cpuinfo_x86
*c
)
1104 early_identify_cpu(c
);
1106 /* AMD-defined flags: level 0x80000001 */
1107 xlvl
= cpuid_eax(0x80000000);
1108 c
->extended_cpuid_level
= xlvl
;
1109 if ((xlvl
& 0xffff0000) == 0x80000000) {
1110 if (xlvl
>= 0x80000001) {
1111 c
->x86_capability
[1] = cpuid_edx(0x80000001);
1112 c
->x86_capability
[6] = cpuid_ecx(0x80000001);
1114 if (xlvl
>= 0x80000004)
1115 get_model_name(c
); /* Default name */
1118 /* Transmeta-defined flags: level 0x80860001 */
1119 xlvl
= cpuid_eax(0x80860000);
1120 if ((xlvl
& 0xffff0000) == 0x80860000) {
1121 /* Don't set x86_cpuid_level here for now to not confuse. */
1122 if (xlvl
>= 0x80860001)
1123 c
->x86_capability
[2] = cpuid_edx(0x80860001);
1126 c
->apicid
= phys_pkg_id(0);
1129 * Vendor-specific initialization. In this section we
1130 * canonicalize the feature flags, meaning if there are
1131 * features a certain CPU supports which CPUID doesn't
1132 * tell us, CPUID claiming incorrect flags, or other bugs,
1133 * we handle them here.
1135 * At the end of this section, c->x86_capability better
1136 * indicate the features this CPU genuinely supports!
1138 switch (c
->x86_vendor
) {
1139 case X86_VENDOR_AMD
:
1143 case X86_VENDOR_INTEL
:
1147 case X86_VENDOR_UNKNOWN
:
1149 display_cacheinfo(c
);
1153 select_idle_routine(c
);
1157 * On SMP, boot_cpu_data holds the common feature set between
1158 * all CPUs; so make sure that we indicate which features are
1159 * common between the CPUs. The first time this routine gets
1160 * executed, c == &boot_cpu_data.
1162 if (c
!= &boot_cpu_data
) {
1163 /* AND the already accumulated flags with these */
1164 for (i
= 0 ; i
< NCAPINTS
; i
++)
1165 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
1168 #ifdef CONFIG_X86_MCE
1171 if (c
== &boot_cpu_data
)
1176 numa_add_cpu(smp_processor_id());
1181 void __cpuinit
print_cpu_info(struct cpuinfo_x86
*c
)
1183 if (c
->x86_model_id
[0])
1184 printk("%s", c
->x86_model_id
);
1186 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1187 printk(" stepping %02x\n", c
->x86_mask
);
1193 * Get CPU information for use by the procfs.
1196 static int show_cpuinfo(struct seq_file
*m
, void *v
)
1198 struct cpuinfo_x86
*c
= v
;
1201 * These flag bits must match the definitions in <asm/cpufeature.h>.
1202 * NULL means this bit is undefined or reserved; either way it doesn't
1203 * have meaning as far as Linux is concerned. Note that it's important
1204 * to realize there is a difference between this table and CPUID -- if
1205 * applications want to get the raw CPUID data, they should access
1206 * /dev/cpu/<cpu_nr>/cpuid instead.
1208 static char *x86_cap_flags
[] = {
1210 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1211 "cx8", "apic", NULL
, "sep", "mtrr", "pge", "mca", "cmov",
1212 "pat", "pse36", "pn", "clflush", NULL
, "dts", "acpi", "mmx",
1213 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", NULL
,
1216 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1217 NULL
, NULL
, NULL
, "syscall", NULL
, NULL
, NULL
, NULL
,
1218 NULL
, NULL
, NULL
, NULL
, "nx", NULL
, "mmxext", NULL
,
1219 NULL
, "fxsr_opt", "rdtscp", NULL
, NULL
, "lm", "3dnowext", "3dnow",
1221 /* Transmeta-defined */
1222 "recovery", "longrun", NULL
, "lrti", NULL
, NULL
, NULL
, NULL
,
1223 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1224 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1225 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1227 /* Other (Linux-defined) */
1228 "cxmmx", NULL
, "cyrix_arr", "centaur_mcr", NULL
,
1229 "constant_tsc", NULL
, NULL
,
1230 "up", 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 /* Intel-defined (#2) */
1235 "pni", NULL
, NULL
, "monitor", "ds_cpl", "vmx", "smx", "est",
1236 "tm2", NULL
, "cid", NULL
, NULL
, "cx16", "xtpr", NULL
,
1237 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1238 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1240 /* VIA/Cyrix/Centaur-defined */
1241 NULL
, NULL
, "rng", "rng_en", NULL
, NULL
, "ace", "ace_en",
1242 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1243 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1244 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1246 /* AMD-defined (#2) */
1247 "lahf_lm", "cmp_legacy", "svm", NULL
, "cr8_legacy", NULL
, NULL
, NULL
,
1248 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1249 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1250 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1252 static char *x86_power_flags
[] = {
1253 "ts", /* temperature sensor */
1254 "fid", /* frequency id control */
1255 "vid", /* voltage id control */
1256 "ttp", /* thermal trip */
1260 /* nothing */ /* constant_tsc - moved to flags */
1265 if (!cpu_online(c
-cpu_data
))
1269 seq_printf(m
,"processor\t: %u\n"
1271 "cpu family\t: %d\n"
1273 "model name\t: %s\n",
1274 (unsigned)(c
-cpu_data
),
1275 c
->x86_vendor_id
[0] ? c
->x86_vendor_id
: "unknown",
1278 c
->x86_model_id
[0] ? c
->x86_model_id
: "unknown");
1280 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1281 seq_printf(m
, "stepping\t: %d\n", c
->x86_mask
);
1283 seq_printf(m
, "stepping\t: unknown\n");
1285 if (cpu_has(c
,X86_FEATURE_TSC
)) {
1286 unsigned int freq
= cpufreq_quick_get((unsigned)(c
-cpu_data
));
1289 seq_printf(m
, "cpu MHz\t\t: %u.%03u\n",
1290 freq
/ 1000, (freq
% 1000));
1294 if (c
->x86_cache_size
>= 0)
1295 seq_printf(m
, "cache size\t: %d KB\n", c
->x86_cache_size
);
1298 if (smp_num_siblings
* c
->x86_max_cores
> 1) {
1299 int cpu
= c
- cpu_data
;
1300 seq_printf(m
, "physical id\t: %d\n", phys_proc_id
[cpu
]);
1301 seq_printf(m
, "siblings\t: %d\n", cpus_weight(cpu_core_map
[cpu
]));
1302 seq_printf(m
, "core id\t\t: %d\n", cpu_core_id
[cpu
]);
1303 seq_printf(m
, "cpu cores\t: %d\n", c
->booted_cores
);
1309 "fpu_exception\t: yes\n"
1310 "cpuid level\t: %d\n"
1317 for ( i
= 0 ; i
< 32*NCAPINTS
; i
++ )
1318 if (cpu_has(c
, i
) && x86_cap_flags
[i
] != NULL
)
1319 seq_printf(m
, " %s", x86_cap_flags
[i
]);
1322 seq_printf(m
, "\nbogomips\t: %lu.%02lu\n",
1323 c
->loops_per_jiffy
/(500000/HZ
),
1324 (c
->loops_per_jiffy
/(5000/HZ
)) % 100);
1326 if (c
->x86_tlbsize
> 0)
1327 seq_printf(m
, "TLB size\t: %d 4K pages\n", c
->x86_tlbsize
);
1328 seq_printf(m
, "clflush size\t: %d\n", c
->x86_clflush_size
);
1329 seq_printf(m
, "cache_alignment\t: %d\n", c
->x86_cache_alignment
);
1331 seq_printf(m
, "address sizes\t: %u bits physical, %u bits virtual\n",
1332 c
->x86_phys_bits
, c
->x86_virt_bits
);
1334 seq_printf(m
, "power management:");
1337 for (i
= 0; i
< 32; i
++)
1338 if (c
->x86_power
& (1 << i
)) {
1339 if (i
< ARRAY_SIZE(x86_power_flags
) &&
1341 seq_printf(m
, "%s%s",
1342 x86_power_flags
[i
][0]?" ":"",
1343 x86_power_flags
[i
]);
1345 seq_printf(m
, " [%d]", i
);
1349 seq_printf(m
, "\n\n");
1354 static void *c_start(struct seq_file
*m
, loff_t
*pos
)
1356 return *pos
< NR_CPUS
? cpu_data
+ *pos
: NULL
;
1359 static void *c_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1362 return c_start(m
, pos
);
1365 static void c_stop(struct seq_file
*m
, void *v
)
1369 struct seq_operations cpuinfo_op
= {
1373 .show
= show_cpuinfo
,
1376 #ifdef CONFIG_INPUT_PCSPKR
1377 #include <linux/platform_device.h>
1378 static __init
int add_pcspkr(void)
1380 struct platform_device
*pd
;
1383 pd
= platform_device_alloc("pcspkr", -1);
1387 ret
= platform_device_add(pd
);
1389 platform_device_put(pd
);
1393 device_initcall(add_pcspkr
);