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
;
74 EXPORT_SYMBOL(boot_cpu_data
);
76 unsigned long mmu_cr4_features
;
79 EXPORT_SYMBOL(acpi_disabled
);
81 extern int __initdata acpi_ht
;
82 extern acpi_interrupt_flags acpi_sci_flags
;
83 int __initdata acpi_force
= 0;
86 int acpi_numa __initdata
;
88 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
91 unsigned long saved_video_mode
;
97 char dmi_alloc_data
[DMI_MAX_DATA
];
102 struct screen_info screen_info
;
103 EXPORT_SYMBOL(screen_info
);
104 struct sys_desc_table_struct
{
105 unsigned short length
;
106 unsigned char table
[0];
109 struct edid_info edid_info
;
110 EXPORT_SYMBOL_GPL(edid_info
);
113 extern int root_mountflags
;
115 char command_line
[COMMAND_LINE_SIZE
];
117 struct resource standard_io_resources
[] = {
118 { .name
= "dma1", .start
= 0x00, .end
= 0x1f,
119 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
120 { .name
= "pic1", .start
= 0x20, .end
= 0x21,
121 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
122 { .name
= "timer0", .start
= 0x40, .end
= 0x43,
123 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
124 { .name
= "timer1", .start
= 0x50, .end
= 0x53,
125 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
126 { .name
= "keyboard", .start
= 0x60, .end
= 0x6f,
127 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
128 { .name
= "dma page reg", .start
= 0x80, .end
= 0x8f,
129 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
130 { .name
= "pic2", .start
= 0xa0, .end
= 0xa1,
131 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
132 { .name
= "dma2", .start
= 0xc0, .end
= 0xdf,
133 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
134 { .name
= "fpu", .start
= 0xf0, .end
= 0xff,
135 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
}
138 #define STANDARD_IO_RESOURCES \
139 (sizeof standard_io_resources / sizeof standard_io_resources[0])
141 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
143 struct resource data_resource
= {
144 .name
= "Kernel data",
147 .flags
= IORESOURCE_RAM
,
149 struct resource code_resource
= {
150 .name
= "Kernel code",
153 .flags
= IORESOURCE_RAM
,
156 #define IORESOURCE_ROM (IORESOURCE_BUSY | IORESOURCE_READONLY | IORESOURCE_MEM)
158 static struct resource system_rom_resource
= {
159 .name
= "System ROM",
162 .flags
= IORESOURCE_ROM
,
165 static struct resource extension_rom_resource
= {
166 .name
= "Extension ROM",
169 .flags
= IORESOURCE_ROM
,
172 static struct resource adapter_rom_resources
[] = {
173 { .name
= "Adapter ROM", .start
= 0xc8000, .end
= 0,
174 .flags
= IORESOURCE_ROM
},
175 { .name
= "Adapter ROM", .start
= 0, .end
= 0,
176 .flags
= IORESOURCE_ROM
},
177 { .name
= "Adapter ROM", .start
= 0, .end
= 0,
178 .flags
= IORESOURCE_ROM
},
179 { .name
= "Adapter ROM", .start
= 0, .end
= 0,
180 .flags
= IORESOURCE_ROM
},
181 { .name
= "Adapter ROM", .start
= 0, .end
= 0,
182 .flags
= IORESOURCE_ROM
},
183 { .name
= "Adapter ROM", .start
= 0, .end
= 0,
184 .flags
= IORESOURCE_ROM
}
187 #define ADAPTER_ROM_RESOURCES \
188 (sizeof adapter_rom_resources / sizeof adapter_rom_resources[0])
190 static struct resource video_rom_resource
= {
194 .flags
= IORESOURCE_ROM
,
197 static struct resource video_ram_resource
= {
198 .name
= "Video RAM area",
201 .flags
= IORESOURCE_RAM
,
204 #define romsignature(x) (*(unsigned short *)(x) == 0xaa55)
206 static int __init
romchecksum(unsigned char *rom
, unsigned long length
)
208 unsigned char *p
, sum
= 0;
210 for (p
= rom
; p
< rom
+ length
; p
++)
215 static void __init
probe_roms(void)
217 unsigned long start
, length
, upper
;
222 upper
= adapter_rom_resources
[0].start
;
223 for (start
= video_rom_resource
.start
; start
< upper
; start
+= 2048) {
224 rom
= isa_bus_to_virt(start
);
225 if (!romsignature(rom
))
228 video_rom_resource
.start
= start
;
230 /* 0 < length <= 0x7f * 512, historically */
231 length
= rom
[2] * 512;
233 /* if checksum okay, trust length byte */
234 if (length
&& romchecksum(rom
, length
))
235 video_rom_resource
.end
= start
+ length
- 1;
237 request_resource(&iomem_resource
, &video_rom_resource
);
241 start
= (video_rom_resource
.end
+ 1 + 2047) & ~2047UL;
246 request_resource(&iomem_resource
, &system_rom_resource
);
247 upper
= system_rom_resource
.start
;
249 /* check for extension rom (ignore length byte!) */
250 rom
= isa_bus_to_virt(extension_rom_resource
.start
);
251 if (romsignature(rom
)) {
252 length
= extension_rom_resource
.end
- extension_rom_resource
.start
+ 1;
253 if (romchecksum(rom
, length
)) {
254 request_resource(&iomem_resource
, &extension_rom_resource
);
255 upper
= extension_rom_resource
.start
;
259 /* check for adapter roms on 2k boundaries */
260 for (i
= 0; i
< ADAPTER_ROM_RESOURCES
&& start
< upper
; start
+= 2048) {
261 rom
= isa_bus_to_virt(start
);
262 if (!romsignature(rom
))
265 /* 0 < length <= 0x7f * 512, historically */
266 length
= rom
[2] * 512;
268 /* but accept any length that fits if checksum okay */
269 if (!length
|| start
+ length
> upper
|| !romchecksum(rom
, length
))
272 adapter_rom_resources
[i
].start
= start
;
273 adapter_rom_resources
[i
].end
= start
+ length
- 1;
274 request_resource(&iomem_resource
, &adapter_rom_resources
[i
]);
276 start
= adapter_rom_resources
[i
++].end
& ~2047UL;
280 /* Check for full argument with no trailing characters */
281 static int fullarg(char *p
, char *arg
)
284 return !memcmp(p
, arg
, l
) && (p
[l
] == 0 || isspace(p
[l
]));
287 static __init
void parse_cmdline_early (char ** cmdline_p
)
289 char c
= ' ', *to
= command_line
, *from
= COMMAND_LINE
;
299 * If the BIOS enumerates physical processors before logical,
300 * maxcpus=N at enumeration-time can be used to disable HT.
302 else if (!memcmp(from
, "maxcpus=", 8)) {
303 extern unsigned int maxcpus
;
305 maxcpus
= simple_strtoul(from
+ 8, NULL
, 0);
309 /* "acpi=off" disables both ACPI table parsing and interpreter init */
310 if (fullarg(from
,"acpi=off"))
313 if (fullarg(from
, "acpi=force")) {
314 /* add later when we do DMI horrors: */
319 /* acpi=ht just means: do ACPI MADT parsing
320 at bootup, but don't enable the full ACPI interpreter */
321 if (fullarg(from
, "acpi=ht")) {
326 else if (fullarg(from
, "pci=noacpi"))
328 else if (fullarg(from
, "acpi=noirq"))
331 else if (fullarg(from
, "acpi_sci=edge"))
332 acpi_sci_flags
.trigger
= 1;
333 else if (fullarg(from
, "acpi_sci=level"))
334 acpi_sci_flags
.trigger
= 3;
335 else if (fullarg(from
, "acpi_sci=high"))
336 acpi_sci_flags
.polarity
= 1;
337 else if (fullarg(from
, "acpi_sci=low"))
338 acpi_sci_flags
.polarity
= 3;
340 /* acpi=strict disables out-of-spec workarounds */
341 else if (fullarg(from
, "acpi=strict")) {
344 #ifdef CONFIG_X86_IO_APIC
345 else if (fullarg(from
, "acpi_skip_timer_override"))
346 acpi_skip_timer_override
= 1;
350 if (fullarg(from
, "disable_timer_pin_1"))
351 disable_timer_pin_1
= 1;
352 if (fullarg(from
, "enable_timer_pin_1"))
353 disable_timer_pin_1
= -1;
355 if (fullarg(from
, "nolapic") || fullarg(from
, "disableapic")) {
356 clear_bit(X86_FEATURE_APIC
, boot_cpu_data
.x86_capability
);
360 if (fullarg(from
, "noapic"))
361 skip_ioapic_setup
= 1;
363 if (fullarg(from
,"apic")) {
364 skip_ioapic_setup
= 0;
368 if (!memcmp(from
, "mem=", 4))
369 parse_memopt(from
+4, &from
);
371 if (!memcmp(from
, "memmap=", 7)) {
372 /* exactmap option is for used defined memory */
373 if (!memcmp(from
+7, "exactmap", 8)) {
374 #ifdef CONFIG_CRASH_DUMP
375 /* If we are doing a crash dump, we
376 * still need to know the real mem
377 * size before original memory map is
380 saved_max_pfn
= e820_end_of_ram();
388 parse_memmapopt(from
+7, &from
);
394 if (!memcmp(from
, "numa=", 5))
398 if (!memcmp(from
,"iommu=",6)) {
402 if (fullarg(from
,"oops=panic"))
405 if (!memcmp(from
, "noexec=", 7))
406 nonx_setup(from
+ 7);
409 /* crashkernel=size@addr specifies the location to reserve for
410 * a crash kernel. By reserving this memory we guarantee
411 * that linux never set's it up as a DMA target.
412 * Useful for holding code to do something appropriate
413 * after a kernel panic.
415 else if (!memcmp(from
, "crashkernel=", 12)) {
416 unsigned long size
, base
;
417 size
= memparse(from
+12, &from
);
419 base
= memparse(from
+1, &from
);
420 /* FIXME: Do I want a sanity check
421 * to validate the memory range?
423 crashk_res
.start
= base
;
424 crashk_res
.end
= base
+ size
- 1;
429 #ifdef CONFIG_PROC_VMCORE
430 /* elfcorehdr= specifies the location of elf core header
431 * stored by the crashed kernel. This option will be passed
432 * by kexec loader to the capture kernel.
434 else if(!memcmp(from
, "elfcorehdr=", 11))
435 elfcorehdr_addr
= memparse(from
+11, &from
);
438 #ifdef CONFIG_HOTPLUG_CPU
439 else if (!memcmp(from
, "additional_cpus=", 16))
440 setup_additional_cpus(from
+16);
447 if (COMMAND_LINE_SIZE
<= ++len
)
452 printk(KERN_INFO
"user-defined physical RAM map:\n");
453 e820_print_map("user");
456 *cmdline_p
= command_line
;
461 contig_initmem_init(unsigned long start_pfn
, unsigned long end_pfn
)
463 unsigned long bootmap_size
, bootmap
;
465 bootmap_size
= bootmem_bootmap_pages(end_pfn
)<<PAGE_SHIFT
;
466 bootmap
= find_e820_area(0, end_pfn
<<PAGE_SHIFT
, bootmap_size
);
468 panic("Cannot find bootmem map of size %ld\n",bootmap_size
);
469 bootmap_size
= init_bootmem(bootmap
>> PAGE_SHIFT
, end_pfn
);
470 e820_bootmem_free(NODE_DATA(0), 0, end_pfn
<< PAGE_SHIFT
);
471 reserve_bootmem(bootmap
, bootmap_size
);
475 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
477 #ifdef CONFIG_EDD_MODULE
481 * copy_edd() - Copy the BIOS EDD information
482 * from boot_params into a safe place.
485 static inline void copy_edd(void)
487 memcpy(edd
.mbr_signature
, EDD_MBR_SIGNATURE
, sizeof(edd
.mbr_signature
));
488 memcpy(edd
.edd_info
, EDD_BUF
, sizeof(edd
.edd_info
));
489 edd
.mbr_signature_nr
= EDD_MBR_SIG_NR
;
490 edd
.edd_info_nr
= EDD_NR
;
493 static inline void copy_edd(void)
498 #define EBDA_ADDR_POINTER 0x40E
500 unsigned __initdata ebda_addr
;
501 unsigned __initdata ebda_size
;
503 static void discover_ebda(void)
506 * there is a real-mode segmented pointer pointing to the
507 * 4K EBDA area at 0x40E
509 ebda_addr
= *(unsigned short *)EBDA_ADDR_POINTER
;
512 ebda_size
= *(unsigned short *)(unsigned long)ebda_addr
;
514 /* Round EBDA up to pages */
518 ebda_size
= round_up(ebda_size
+ (ebda_addr
& ~PAGE_MASK
), PAGE_SIZE
);
519 if (ebda_size
> 64*1024)
523 void __init
setup_arch(char **cmdline_p
)
525 unsigned long kernel_end
;
527 ROOT_DEV
= old_decode_dev(ORIG_ROOT_DEV
);
528 screen_info
= SCREEN_INFO
;
529 edid_info
= EDID_INFO
;
530 saved_video_mode
= SAVED_VIDEO_MODE
;
531 bootloader_type
= LOADER_TYPE
;
533 #ifdef CONFIG_BLK_DEV_RAM
534 rd_image_start
= RAMDISK_FLAGS
& RAMDISK_IMAGE_START_MASK
;
535 rd_prompt
= ((RAMDISK_FLAGS
& RAMDISK_PROMPT_FLAG
) != 0);
536 rd_doload
= ((RAMDISK_FLAGS
& RAMDISK_LOAD_FLAG
) != 0);
538 setup_memory_region();
541 if (!MOUNT_ROOT_RDONLY
)
542 root_mountflags
&= ~MS_RDONLY
;
543 init_mm
.start_code
= (unsigned long) &_text
;
544 init_mm
.end_code
= (unsigned long) &_etext
;
545 init_mm
.end_data
= (unsigned long) &_edata
;
546 init_mm
.brk
= (unsigned long) &_end
;
548 code_resource
.start
= virt_to_phys(&_text
);
549 code_resource
.end
= virt_to_phys(&_etext
)-1;
550 data_resource
.start
= virt_to_phys(&_etext
);
551 data_resource
.end
= virt_to_phys(&_edata
)-1;
553 parse_cmdline_early(cmdline_p
);
555 early_identify_cpu(&boot_cpu_data
);
558 * partially used pages are not usable - thus
559 * we are rounding upwards:
561 end_pfn
= e820_end_of_ram();
562 num_physpages
= end_pfn
; /* for pfn_valid */
568 init_memory_mapping(0, (end_pfn_map
<< PAGE_SHIFT
));
576 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
577 * Call this early for SRAT node setup.
579 acpi_boot_table_init();
582 #ifdef CONFIG_ACPI_NUMA
584 * Parse SRAT to discover nodes.
590 numa_initmem_init(0, end_pfn
);
592 contig_initmem_init(0, end_pfn
);
595 /* Reserve direct mapping */
596 reserve_bootmem_generic(table_start
<< PAGE_SHIFT
,
597 (table_end
- table_start
) << PAGE_SHIFT
);
600 kernel_end
= round_up(__pa_symbol(&_end
),PAGE_SIZE
);
601 reserve_bootmem_generic(HIGH_MEMORY
, kernel_end
- HIGH_MEMORY
);
604 * reserve physical page 0 - it's a special BIOS page on many boxes,
605 * enabling clean reboots, SMP operation, laptop functions.
607 reserve_bootmem_generic(0, PAGE_SIZE
);
609 /* reserve ebda region */
611 reserve_bootmem_generic(ebda_addr
, ebda_size
);
615 * But first pinch a few for the stack/trampoline stuff
616 * FIXME: Don't need the extra page at 4K, but need to fix
617 * trampoline before removing it. (see the GDT stuff)
619 reserve_bootmem_generic(PAGE_SIZE
, PAGE_SIZE
);
621 /* Reserve SMP trampoline */
622 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE
, PAGE_SIZE
);
625 #ifdef CONFIG_ACPI_SLEEP
627 * Reserve low memory region for sleep support.
629 acpi_reserve_bootmem();
631 #ifdef CONFIG_X86_LOCAL_APIC
633 * Find and reserve possible boot-time SMP configuration:
637 #ifdef CONFIG_BLK_DEV_INITRD
638 if (LOADER_TYPE
&& INITRD_START
) {
639 if (INITRD_START
+ INITRD_SIZE
<= (end_pfn
<< PAGE_SHIFT
)) {
640 reserve_bootmem_generic(INITRD_START
, INITRD_SIZE
);
642 INITRD_START
? INITRD_START
+ PAGE_OFFSET
: 0;
643 initrd_end
= initrd_start
+INITRD_SIZE
;
646 printk(KERN_ERR
"initrd extends beyond end of memory "
647 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
648 (unsigned long)(INITRD_START
+ INITRD_SIZE
),
649 (unsigned long)(end_pfn
<< PAGE_SHIFT
));
655 if (crashk_res
.start
!= crashk_res
.end
) {
656 reserve_bootmem_generic(crashk_res
.start
,
657 crashk_res
.end
- crashk_res
.start
+ 1);
666 * set this early, so we dont allocate cpu0
667 * if MADT list doesnt list BSP first
668 * mpparse.c/MP_processor_info() allocates logical cpu numbers.
670 cpu_set(0, cpu_present_map
);
673 * Read APIC and some other early information from ACPI tables.
680 #ifdef CONFIG_X86_LOCAL_APIC
682 * get boot-time SMP configuration:
684 if (smp_found_config
)
686 init_apic_mappings();
690 * Request address space for all standard RAM and ROM resources
691 * and also for regions reported as reserved by the e820.
694 e820_reserve_resources();
696 request_resource(&iomem_resource
, &video_ram_resource
);
700 /* request I/O space for devices used on all i[345]86 PCs */
701 for (i
= 0; i
< STANDARD_IO_RESOURCES
; i
++)
702 request_resource(&ioport_resource
, &standard_io_resources
[i
]);
708 #if defined(CONFIG_VGA_CONSOLE)
709 conswitchp
= &vga_con
;
710 #elif defined(CONFIG_DUMMY_CONSOLE)
711 conswitchp
= &dummy_con
;
716 static int __cpuinit
get_model_name(struct cpuinfo_x86
*c
)
720 if (c
->extended_cpuid_level
< 0x80000004)
723 v
= (unsigned int *) c
->x86_model_id
;
724 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
725 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
726 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
727 c
->x86_model_id
[48] = 0;
732 static void __cpuinit
display_cacheinfo(struct cpuinfo_x86
*c
)
734 unsigned int n
, dummy
, eax
, ebx
, ecx
, edx
;
736 n
= c
->extended_cpuid_level
;
738 if (n
>= 0x80000005) {
739 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
740 printk(KERN_INFO
"CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
741 edx
>>24, edx
&0xFF, ecx
>>24, ecx
&0xFF);
742 c
->x86_cache_size
=(ecx
>>24)+(edx
>>24);
743 /* On K8 L1 TLB is inclusive, so don't count it */
747 if (n
>= 0x80000006) {
748 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
749 ecx
= cpuid_ecx(0x80000006);
750 c
->x86_cache_size
= ecx
>> 16;
751 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
753 printk(KERN_INFO
"CPU: L2 Cache: %dK (%d bytes/line)\n",
754 c
->x86_cache_size
, ecx
& 0xFF);
758 cpuid(0x80000007, &dummy
, &dummy
, &dummy
, &c
->x86_power
);
759 if (n
>= 0x80000008) {
760 cpuid(0x80000008, &eax
, &dummy
, &dummy
, &dummy
);
761 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
762 c
->x86_phys_bits
= eax
& 0xff;
767 static int nearby_node(int apicid
)
770 for (i
= apicid
- 1; i
>= 0; i
--) {
771 int node
= apicid_to_node
[i
];
772 if (node
!= NUMA_NO_NODE
&& node_online(node
))
775 for (i
= apicid
+ 1; i
< MAX_LOCAL_APIC
; i
++) {
776 int node
= apicid_to_node
[i
];
777 if (node
!= NUMA_NO_NODE
&& node_online(node
))
780 return first_node(node_online_map
); /* Shouldn't happen */
785 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
786 * Assumes number of cores is a power of two.
788 static void __init
amd_detect_cmp(struct cpuinfo_x86
*c
)
793 int cpu
= smp_processor_id();
795 unsigned apicid
= hard_smp_processor_id();
797 unsigned ecx
= cpuid_ecx(0x80000008);
799 c
->x86_max_cores
= (ecx
& 0xff) + 1;
801 /* CPU telling us the core id bits shift? */
802 bits
= (ecx
>> 12) & 0xF;
804 /* Otherwise recompute */
806 while ((1 << bits
) < c
->x86_max_cores
)
810 /* Low order bits define the core id (index of core in socket) */
811 c
->cpu_core_id
= c
->phys_proc_id
& ((1 << bits
)-1);
812 /* Convert the APIC ID into the socket ID */
813 c
->phys_proc_id
= phys_pkg_id(bits
);
816 node
= c
->phys_proc_id
;
817 if (apicid_to_node
[apicid
] != NUMA_NO_NODE
)
818 node
= apicid_to_node
[apicid
];
819 if (!node_online(node
)) {
820 /* Two possibilities here:
821 - The CPU is missing memory and no node was created.
822 In that case try picking one from a nearby CPU
823 - The APIC IDs differ from the HyperTransport node IDs
824 which the K8 northbridge parsing fills in.
825 Assume they are all increased by a constant offset,
826 but in the same order as the HT nodeids.
827 If that doesn't result in a usable node fall back to the
828 path for the previous case. */
829 int ht_nodeid
= apicid
- (cpu_data
[0].phys_proc_id
<< bits
);
830 if (ht_nodeid
>= 0 &&
831 apicid_to_node
[ht_nodeid
] != NUMA_NO_NODE
)
832 node
= apicid_to_node
[ht_nodeid
];
833 /* Pick a nearby node */
834 if (!node_online(node
))
835 node
= nearby_node(apicid
);
837 numa_set_node(cpu
, node
);
839 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
844 static void __init
init_amd(struct cpuinfo_x86
*c
)
852 * Disable TLB flush filter by setting HWCR.FFDIS on K8
853 * bit 6 of msr C001_0015
855 * Errata 63 for SH-B3 steppings
856 * Errata 122 for all steppings (F+ have it disabled by default)
859 rdmsrl(MSR_K8_HWCR
, value
);
861 wrmsrl(MSR_K8_HWCR
, value
);
865 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
866 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
867 clear_bit(0*32+31, &c
->x86_capability
);
869 /* On C+ stepping K8 rep microcode works well for copy/memset */
870 level
= cpuid_eax(1);
871 if (c
->x86
== 15 && ((level
>= 0x0f48 && level
< 0x0f50) || level
>= 0x0f58))
872 set_bit(X86_FEATURE_REP_GOOD
, &c
->x86_capability
);
874 /* Enable workaround for FXSAVE leak */
876 set_bit(X86_FEATURE_FXSAVE_LEAK
, &c
->x86_capability
);
878 level
= get_model_name(c
);
882 /* Should distinguish Models here, but this is only
883 a fallback anyways. */
884 strcpy(c
->x86_model_id
, "Hammer");
888 display_cacheinfo(c
);
890 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
891 if (c
->x86_power
& (1<<8))
892 set_bit(X86_FEATURE_CONSTANT_TSC
, &c
->x86_capability
);
894 /* Multi core CPU? */
895 if (c
->extended_cpuid_level
>= 0x80000008)
898 /* Fix cpuid4 emulation for more */
899 num_cache_leaves
= 3;
902 static void __cpuinit
detect_ht(struct cpuinfo_x86
*c
)
905 u32 eax
, ebx
, ecx
, edx
;
906 int index_msb
, core_bits
;
908 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
911 if (!cpu_has(c
, X86_FEATURE_HT
))
913 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
916 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
918 if (smp_num_siblings
== 1) {
919 printk(KERN_INFO
"CPU: Hyper-Threading is disabled\n");
920 } else if (smp_num_siblings
> 1 ) {
922 if (smp_num_siblings
> NR_CPUS
) {
923 printk(KERN_WARNING
"CPU: Unsupported number of the siblings %d", smp_num_siblings
);
924 smp_num_siblings
= 1;
928 index_msb
= get_count_order(smp_num_siblings
);
929 c
->phys_proc_id
= phys_pkg_id(index_msb
);
931 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
933 index_msb
= get_count_order(smp_num_siblings
) ;
935 core_bits
= get_count_order(c
->x86_max_cores
);
937 c
->cpu_core_id
= phys_pkg_id(index_msb
) &
938 ((1 << core_bits
) - 1);
941 if ((c
->x86_max_cores
* smp_num_siblings
) > 1) {
942 printk(KERN_INFO
"CPU: Physical Processor ID: %d\n", c
->phys_proc_id
);
943 printk(KERN_INFO
"CPU: Processor Core ID: %d\n", c
->cpu_core_id
);
950 * find out the number of processor cores on the die
952 static int __cpuinit
intel_num_cpu_cores(struct cpuinfo_x86
*c
)
956 if (c
->cpuid_level
< 4)
959 cpuid_count(4, 0, &eax
, &t
, &t
, &t
);
962 return ((eax
>> 26) + 1);
967 static void srat_detect_node(void)
971 int cpu
= smp_processor_id();
972 int apicid
= hard_smp_processor_id();
974 /* Don't do the funky fallback heuristics the AMD version employs
976 node
= apicid_to_node
[apicid
];
977 if (node
== NUMA_NO_NODE
)
978 node
= first_node(node_online_map
);
979 numa_set_node(cpu
, node
);
982 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
986 static void __cpuinit
init_intel(struct cpuinfo_x86
*c
)
991 init_intel_cacheinfo(c
);
992 if (c
->cpuid_level
> 9 ) {
993 unsigned eax
= cpuid_eax(10);
994 /* Check for version and the number of counters */
995 if ((eax
& 0xff) && (((eax
>>8) & 0xff) > 1))
996 set_bit(X86_FEATURE_ARCH_PERFMON
, &c
->x86_capability
);
999 n
= c
->extended_cpuid_level
;
1000 if (n
>= 0x80000008) {
1001 unsigned eax
= cpuid_eax(0x80000008);
1002 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
1003 c
->x86_phys_bits
= eax
& 0xff;
1004 /* CPUID workaround for Intel 0F34 CPU */
1005 if (c
->x86_vendor
== X86_VENDOR_INTEL
&&
1006 c
->x86
== 0xF && c
->x86_model
== 0x3 &&
1008 c
->x86_phys_bits
= 36;
1012 c
->x86_cache_alignment
= c
->x86_clflush_size
* 2;
1013 if ((c
->x86
== 0xf && c
->x86_model
>= 0x03) ||
1014 (c
->x86
== 0x6 && c
->x86_model
>= 0x0e))
1015 set_bit(X86_FEATURE_CONSTANT_TSC
, &c
->x86_capability
);
1016 set_bit(X86_FEATURE_SYNC_RDTSC
, &c
->x86_capability
);
1017 c
->x86_max_cores
= intel_num_cpu_cores(c
);
1022 static void __cpuinit
get_cpu_vendor(struct cpuinfo_x86
*c
)
1024 char *v
= c
->x86_vendor_id
;
1026 if (!strcmp(v
, "AuthenticAMD"))
1027 c
->x86_vendor
= X86_VENDOR_AMD
;
1028 else if (!strcmp(v
, "GenuineIntel"))
1029 c
->x86_vendor
= X86_VENDOR_INTEL
;
1031 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
1034 struct cpu_model_info
{
1037 char *model_names
[16];
1040 /* Do some early cpuid on the boot CPU to get some parameter that are
1041 needed before check_bugs. Everything advanced is in identify_cpu
1043 void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
)
1047 c
->loops_per_jiffy
= loops_per_jiffy
;
1048 c
->x86_cache_size
= -1;
1049 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
1050 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
1051 c
->x86_vendor_id
[0] = '\0'; /* Unset */
1052 c
->x86_model_id
[0] = '\0'; /* Unset */
1053 c
->x86_clflush_size
= 64;
1054 c
->x86_cache_alignment
= c
->x86_clflush_size
;
1055 c
->x86_max_cores
= 1;
1056 c
->extended_cpuid_level
= 0;
1057 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
1059 /* Get vendor name */
1060 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
1061 (unsigned int *)&c
->x86_vendor_id
[0],
1062 (unsigned int *)&c
->x86_vendor_id
[8],
1063 (unsigned int *)&c
->x86_vendor_id
[4]);
1067 /* Initialize the standard set of capabilities */
1068 /* Note that the vendor-specific code below might override */
1070 /* Intel-defined flags: level 0x00000001 */
1071 if (c
->cpuid_level
>= 0x00000001) {
1073 cpuid(0x00000001, &tfms
, &misc
, &c
->x86_capability
[4],
1074 &c
->x86_capability
[0]);
1075 c
->x86
= (tfms
>> 8) & 0xf;
1076 c
->x86_model
= (tfms
>> 4) & 0xf;
1077 c
->x86_mask
= tfms
& 0xf;
1079 c
->x86
+= (tfms
>> 20) & 0xff;
1081 c
->x86_model
+= ((tfms
>> 16) & 0xF) << 4;
1082 if (c
->x86_capability
[0] & (1<<19))
1083 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
1085 /* Have CPUID level 0 only - unheard of */
1090 c
->phys_proc_id
= (cpuid_ebx(1) >> 24) & 0xff;
1095 * This does the hard work of actually picking apart the CPU stuff...
1097 void __cpuinit
identify_cpu(struct cpuinfo_x86
*c
)
1102 early_identify_cpu(c
);
1104 /* AMD-defined flags: level 0x80000001 */
1105 xlvl
= cpuid_eax(0x80000000);
1106 c
->extended_cpuid_level
= xlvl
;
1107 if ((xlvl
& 0xffff0000) == 0x80000000) {
1108 if (xlvl
>= 0x80000001) {
1109 c
->x86_capability
[1] = cpuid_edx(0x80000001);
1110 c
->x86_capability
[6] = cpuid_ecx(0x80000001);
1112 if (xlvl
>= 0x80000004)
1113 get_model_name(c
); /* Default name */
1116 /* Transmeta-defined flags: level 0x80860001 */
1117 xlvl
= cpuid_eax(0x80860000);
1118 if ((xlvl
& 0xffff0000) == 0x80860000) {
1119 /* Don't set x86_cpuid_level here for now to not confuse. */
1120 if (xlvl
>= 0x80860001)
1121 c
->x86_capability
[2] = cpuid_edx(0x80860001);
1124 c
->apicid
= phys_pkg_id(0);
1127 * Vendor-specific initialization. In this section we
1128 * canonicalize the feature flags, meaning if there are
1129 * features a certain CPU supports which CPUID doesn't
1130 * tell us, CPUID claiming incorrect flags, or other bugs,
1131 * we handle them here.
1133 * At the end of this section, c->x86_capability better
1134 * indicate the features this CPU genuinely supports!
1136 switch (c
->x86_vendor
) {
1137 case X86_VENDOR_AMD
:
1141 case X86_VENDOR_INTEL
:
1145 case X86_VENDOR_UNKNOWN
:
1147 display_cacheinfo(c
);
1151 select_idle_routine(c
);
1155 * On SMP, boot_cpu_data holds the common feature set between
1156 * all CPUs; so make sure that we indicate which features are
1157 * common between the CPUs. The first time this routine gets
1158 * executed, c == &boot_cpu_data.
1160 if (c
!= &boot_cpu_data
) {
1161 /* AND the already accumulated flags with these */
1162 for (i
= 0 ; i
< NCAPINTS
; i
++)
1163 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
1166 #ifdef CONFIG_X86_MCE
1169 if (c
== &boot_cpu_data
)
1174 numa_add_cpu(smp_processor_id());
1179 void __cpuinit
print_cpu_info(struct cpuinfo_x86
*c
)
1181 if (c
->x86_model_id
[0])
1182 printk("%s", c
->x86_model_id
);
1184 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1185 printk(" stepping %02x\n", c
->x86_mask
);
1191 * Get CPU information for use by the procfs.
1194 static int show_cpuinfo(struct seq_file
*m
, void *v
)
1196 struct cpuinfo_x86
*c
= v
;
1199 * These flag bits must match the definitions in <asm/cpufeature.h>.
1200 * NULL means this bit is undefined or reserved; either way it doesn't
1201 * have meaning as far as Linux is concerned. Note that it's important
1202 * to realize there is a difference between this table and CPUID -- if
1203 * applications want to get the raw CPUID data, they should access
1204 * /dev/cpu/<cpu_nr>/cpuid instead.
1206 static char *x86_cap_flags
[] = {
1208 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1209 "cx8", "apic", NULL
, "sep", "mtrr", "pge", "mca", "cmov",
1210 "pat", "pse36", "pn", "clflush", NULL
, "dts", "acpi", "mmx",
1211 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", NULL
,
1214 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1215 NULL
, NULL
, NULL
, "syscall", NULL
, NULL
, NULL
, NULL
,
1216 NULL
, NULL
, NULL
, NULL
, "nx", NULL
, "mmxext", NULL
,
1217 NULL
, "fxsr_opt", NULL
, "rdtscp", NULL
, "lm", "3dnowext", "3dnow",
1219 /* Transmeta-defined */
1220 "recovery", "longrun", NULL
, "lrti", NULL
, NULL
, NULL
, NULL
,
1221 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1222 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1223 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1225 /* Other (Linux-defined) */
1226 "cxmmx", NULL
, "cyrix_arr", "centaur_mcr", NULL
,
1227 "constant_tsc", NULL
, NULL
,
1228 "up", NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1229 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1230 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1232 /* Intel-defined (#2) */
1233 "pni", NULL
, NULL
, "monitor", "ds_cpl", "vmx", "smx", "est",
1234 "tm2", NULL
, "cid", NULL
, NULL
, "cx16", "xtpr", NULL
,
1235 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1236 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1238 /* VIA/Cyrix/Centaur-defined */
1239 NULL
, NULL
, "rng", "rng_en", NULL
, NULL
, "ace", "ace_en",
1240 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1241 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1242 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1244 /* AMD-defined (#2) */
1245 "lahf_lm", "cmp_legacy", "svm", NULL
, "cr8_legacy", NULL
, NULL
, NULL
,
1246 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1247 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1248 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1250 static char *x86_power_flags
[] = {
1251 "ts", /* temperature sensor */
1252 "fid", /* frequency id control */
1253 "vid", /* voltage id control */
1254 "ttp", /* thermal trip */
1258 /* nothing */ /* constant_tsc - moved to flags */
1263 if (!cpu_online(c
-cpu_data
))
1267 seq_printf(m
,"processor\t: %u\n"
1269 "cpu family\t: %d\n"
1271 "model name\t: %s\n",
1272 (unsigned)(c
-cpu_data
),
1273 c
->x86_vendor_id
[0] ? c
->x86_vendor_id
: "unknown",
1276 c
->x86_model_id
[0] ? c
->x86_model_id
: "unknown");
1278 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1279 seq_printf(m
, "stepping\t: %d\n", c
->x86_mask
);
1281 seq_printf(m
, "stepping\t: unknown\n");
1283 if (cpu_has(c
,X86_FEATURE_TSC
)) {
1284 unsigned int freq
= cpufreq_quick_get((unsigned)(c
-cpu_data
));
1287 seq_printf(m
, "cpu MHz\t\t: %u.%03u\n",
1288 freq
/ 1000, (freq
% 1000));
1292 if (c
->x86_cache_size
>= 0)
1293 seq_printf(m
, "cache size\t: %d KB\n", c
->x86_cache_size
);
1296 if (smp_num_siblings
* c
->x86_max_cores
> 1) {
1297 int cpu
= c
- cpu_data
;
1298 seq_printf(m
, "physical id\t: %d\n", c
->phys_proc_id
);
1299 seq_printf(m
, "siblings\t: %d\n", cpus_weight(cpu_core_map
[cpu
]));
1300 seq_printf(m
, "core id\t\t: %d\n", c
->cpu_core_id
);
1301 seq_printf(m
, "cpu cores\t: %d\n", c
->booted_cores
);
1307 "fpu_exception\t: yes\n"
1308 "cpuid level\t: %d\n"
1315 for ( i
= 0 ; i
< 32*NCAPINTS
; i
++ )
1316 if (cpu_has(c
, i
) && x86_cap_flags
[i
] != NULL
)
1317 seq_printf(m
, " %s", x86_cap_flags
[i
]);
1320 seq_printf(m
, "\nbogomips\t: %lu.%02lu\n",
1321 c
->loops_per_jiffy
/(500000/HZ
),
1322 (c
->loops_per_jiffy
/(5000/HZ
)) % 100);
1324 if (c
->x86_tlbsize
> 0)
1325 seq_printf(m
, "TLB size\t: %d 4K pages\n", c
->x86_tlbsize
);
1326 seq_printf(m
, "clflush size\t: %d\n", c
->x86_clflush_size
);
1327 seq_printf(m
, "cache_alignment\t: %d\n", c
->x86_cache_alignment
);
1329 seq_printf(m
, "address sizes\t: %u bits physical, %u bits virtual\n",
1330 c
->x86_phys_bits
, c
->x86_virt_bits
);
1332 seq_printf(m
, "power management:");
1335 for (i
= 0; i
< 32; i
++)
1336 if (c
->x86_power
& (1 << i
)) {
1337 if (i
< ARRAY_SIZE(x86_power_flags
) &&
1339 seq_printf(m
, "%s%s",
1340 x86_power_flags
[i
][0]?" ":"",
1341 x86_power_flags
[i
]);
1343 seq_printf(m
, " [%d]", i
);
1347 seq_printf(m
, "\n\n");
1352 static void *c_start(struct seq_file
*m
, loff_t
*pos
)
1354 return *pos
< NR_CPUS
? cpu_data
+ *pos
: NULL
;
1357 static void *c_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1360 return c_start(m
, pos
);
1363 static void c_stop(struct seq_file
*m
, void *v
)
1367 struct seq_operations cpuinfo_op
= {
1371 .show
= show_cpuinfo
,
1374 #if defined(CONFIG_INPUT_PCSPKR) || defined(CONFIG_INPUT_PCSPKR_MODULE)
1375 #include <linux/platform_device.h>
1376 static __init
int add_pcspkr(void)
1378 struct platform_device
*pd
;
1381 pd
= platform_device_alloc("pcspkr", -1);
1385 ret
= platform_device_add(pd
);
1387 platform_device_put(pd
);
1391 device_initcall(add_pcspkr
);