2 * Copyright (C) 1995 Linus Torvalds
6 * This file handles the architecture-dependent parts of initialization
9 #include <linux/errno.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
13 #include <linux/stddef.h>
14 #include <linux/unistd.h>
15 #include <linux/ptrace.h>
16 #include <linux/slab.h>
17 #include <linux/user.h>
18 #include <linux/a.out.h>
19 #include <linux/screen_info.h>
20 #include <linux/ioport.h>
21 #include <linux/delay.h>
22 #include <linux/init.h>
23 #include <linux/initrd.h>
24 #include <linux/highmem.h>
25 #include <linux/bootmem.h>
26 #include <linux/module.h>
27 #include <asm/processor.h>
28 #include <linux/console.h>
29 #include <linux/seq_file.h>
30 #include <linux/crash_dump.h>
31 #include <linux/root_dev.h>
32 #include <linux/pci.h>
33 #include <linux/efi.h>
34 #include <linux/acpi.h>
35 #include <linux/kallsyms.h>
36 #include <linux/edd.h>
37 #include <linux/mmzone.h>
38 #include <linux/kexec.h>
39 #include <linux/cpufreq.h>
40 #include <linux/dmi.h>
41 #include <linux/dma-mapping.h>
42 #include <linux/ctype.h>
43 #include <linux/uaccess.h>
46 #include <asm/uaccess.h>
47 #include <asm/system.h>
48 #include <asm/vsyscall.h>
53 #include <video/edid.h>
57 #include <asm/mpspec.h>
58 #include <asm/mmu_context.h>
59 #include <asm/proto.h>
60 #include <asm/setup.h>
61 #include <asm/mach_apic.h>
63 #include <asm/sections.h>
65 #include <asm/cacheflush.h>
68 #include <asm/topology.h>
70 #ifdef CONFIG_PARAVIRT
71 #include <asm/paravirt.h>
80 struct cpuinfo_x86 boot_cpu_data __read_mostly
;
81 EXPORT_SYMBOL(boot_cpu_data
);
83 unsigned long mmu_cr4_features
;
85 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
88 unsigned long saved_video_mode
;
90 int force_mwait __cpuinitdata
;
96 char dmi_alloc_data
[DMI_MAX_DATA
];
101 struct screen_info screen_info
;
102 EXPORT_SYMBOL(screen_info
);
103 struct sys_desc_table_struct
{
104 unsigned short length
;
105 unsigned char table
[0];
108 struct edid_info edid_info
;
109 EXPORT_SYMBOL_GPL(edid_info
);
111 extern int root_mountflags
;
113 char __initdata command_line
[COMMAND_LINE_SIZE
];
115 struct resource standard_io_resources
[] = {
116 { .name
= "dma1", .start
= 0x00, .end
= 0x1f,
117 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
118 { .name
= "pic1", .start
= 0x20, .end
= 0x21,
119 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
120 { .name
= "timer0", .start
= 0x40, .end
= 0x43,
121 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
122 { .name
= "timer1", .start
= 0x50, .end
= 0x53,
123 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
124 { .name
= "keyboard", .start
= 0x60, .end
= 0x6f,
125 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
126 { .name
= "dma page reg", .start
= 0x80, .end
= 0x8f,
127 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
128 { .name
= "pic2", .start
= 0xa0, .end
= 0xa1,
129 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
130 { .name
= "dma2", .start
= 0xc0, .end
= 0xdf,
131 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
132 { .name
= "fpu", .start
= 0xf0, .end
= 0xff,
133 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
}
136 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
138 static struct resource data_resource
= {
139 .name
= "Kernel data",
142 .flags
= IORESOURCE_RAM
,
144 static struct resource code_resource
= {
145 .name
= "Kernel code",
148 .flags
= IORESOURCE_RAM
,
150 static struct resource bss_resource
= {
151 .name
= "Kernel bss",
154 .flags
= IORESOURCE_RAM
,
157 static void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
);
159 #ifdef CONFIG_PROC_VMCORE
160 /* elfcorehdr= specifies the location of elf core header
161 * stored by the crashed kernel. This option will be passed
162 * by kexec loader to the capture kernel.
164 static int __init
setup_elfcorehdr(char *arg
)
169 elfcorehdr_addr
= memparse(arg
, &end
);
170 return end
> arg
? 0 : -EINVAL
;
172 early_param("elfcorehdr", setup_elfcorehdr
);
177 contig_initmem_init(unsigned long start_pfn
, unsigned long end_pfn
)
179 unsigned long bootmap_size
, bootmap
;
181 bootmap_size
= bootmem_bootmap_pages(end_pfn
)<<PAGE_SHIFT
;
182 bootmap
= find_e820_area(0, end_pfn
<<PAGE_SHIFT
, bootmap_size
);
184 panic("Cannot find bootmem map of size %ld\n", bootmap_size
);
185 bootmap_size
= init_bootmem(bootmap
>> PAGE_SHIFT
, end_pfn
);
186 e820_register_active_regions(0, start_pfn
, end_pfn
);
187 free_bootmem_with_active_regions(0, end_pfn
);
188 reserve_bootmem(bootmap
, bootmap_size
);
192 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
194 #ifdef CONFIG_EDD_MODULE
198 * copy_edd() - Copy the BIOS EDD information
199 * from boot_params into a safe place.
202 static inline void copy_edd(void)
204 memcpy(edd
.mbr_signature
, boot_params
.edd_mbr_sig_buffer
,
205 sizeof(edd
.mbr_signature
));
206 memcpy(edd
.edd_info
, boot_params
.eddbuf
, sizeof(edd
.edd_info
));
207 edd
.mbr_signature_nr
= boot_params
.edd_mbr_sig_buf_entries
;
208 edd
.edd_info_nr
= boot_params
.eddbuf_entries
;
211 static inline void copy_edd(void)
217 static void __init
reserve_crashkernel(void)
219 unsigned long long free_mem
;
220 unsigned long long crash_size
, crash_base
;
224 ((unsigned long long)max_low_pfn
- min_low_pfn
) << PAGE_SHIFT
;
226 ret
= parse_crashkernel(boot_command_line
, free_mem
,
227 &crash_size
, &crash_base
);
228 if (ret
== 0 && crash_size
) {
229 if (crash_base
> 0) {
230 printk(KERN_INFO
"Reserving %ldMB of memory at %ldMB "
231 "for crashkernel (System RAM: %ldMB)\n",
232 (unsigned long)(crash_size
>> 20),
233 (unsigned long)(crash_base
>> 20),
234 (unsigned long)(free_mem
>> 20));
235 crashk_res
.start
= crash_base
;
236 crashk_res
.end
= crash_base
+ crash_size
- 1;
237 reserve_bootmem(crash_base
, crash_size
);
239 printk(KERN_INFO
"crashkernel reservation failed - "
240 "you have to specify a base address\n");
244 static inline void __init
reserve_crashkernel(void)
248 #define EBDA_ADDR_POINTER 0x40E
250 unsigned __initdata ebda_addr
;
251 unsigned __initdata ebda_size
;
253 static void __init
discover_ebda(void)
256 * there is a real-mode segmented pointer pointing to the
257 * 4K EBDA area at 0x40E
259 ebda_addr
= *(unsigned short *)__va(EBDA_ADDR_POINTER
);
261 * There can be some situations, like paravirtualized guests,
262 * in which there is no available ebda information. In such
272 ebda_size
= *(unsigned short *)__va(ebda_addr
);
274 /* Round EBDA up to pages */
278 ebda_size
= round_up(ebda_size
+ (ebda_addr
& ~PAGE_MASK
), PAGE_SIZE
);
279 if (ebda_size
> 64*1024)
283 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
284 void __attribute__((weak
)) __init
memory_setup(void)
286 machine_specific_memory_setup();
289 void __init
setup_arch(char **cmdline_p
)
293 printk(KERN_INFO
"Command line: %s\n", boot_command_line
);
295 ROOT_DEV
= old_decode_dev(boot_params
.hdr
.root_dev
);
296 screen_info
= boot_params
.screen_info
;
297 edid_info
= boot_params
.edid_info
;
298 saved_video_mode
= boot_params
.hdr
.vid_mode
;
299 bootloader_type
= boot_params
.hdr
.type_of_loader
;
301 #ifdef CONFIG_BLK_DEV_RAM
302 rd_image_start
= boot_params
.hdr
.ram_size
& RAMDISK_IMAGE_START_MASK
;
303 rd_prompt
= ((boot_params
.hdr
.ram_size
& RAMDISK_PROMPT_FLAG
) != 0);
304 rd_doload
= ((boot_params
.hdr
.ram_size
& RAMDISK_LOAD_FLAG
) != 0);
307 if (!strncmp((char *)&boot_params
.efi_info
.efi_loader_signature
,
317 if (!boot_params
.hdr
.root_flags
)
318 root_mountflags
&= ~MS_RDONLY
;
319 init_mm
.start_code
= (unsigned long) &_text
;
320 init_mm
.end_code
= (unsigned long) &_etext
;
321 init_mm
.end_data
= (unsigned long) &_edata
;
322 init_mm
.brk
= (unsigned long) &_end
;
324 code_resource
.start
= virt_to_phys(&_text
);
325 code_resource
.end
= virt_to_phys(&_etext
)-1;
326 data_resource
.start
= virt_to_phys(&_etext
);
327 data_resource
.end
= virt_to_phys(&_edata
)-1;
328 bss_resource
.start
= virt_to_phys(&__bss_start
);
329 bss_resource
.end
= virt_to_phys(&__bss_stop
)-1;
331 early_identify_cpu(&boot_cpu_data
);
333 strlcpy(command_line
, boot_command_line
, COMMAND_LINE_SIZE
);
334 *cmdline_p
= command_line
;
338 finish_e820_parsing();
340 early_gart_iommu_check();
342 e820_register_active_regions(0, 0, -1UL);
344 * partially used pages are not usable - thus
345 * we are rounding upwards:
347 end_pfn
= e820_end_of_ram();
348 num_physpages
= end_pfn
;
354 init_memory_mapping(0, (end_pfn_map
<< PAGE_SHIFT
));
363 /* setup to use the early static init tables during kernel startup */
364 x86_cpu_to_apicid_early_ptr
= (void *)&x86_cpu_to_apicid_init
;
365 x86_cpu_to_node_map_early_ptr
= (void *)&x86_cpu_to_node_map_init
;
370 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
371 * Call this early for SRAT node setup.
373 acpi_boot_table_init();
376 /* How many end-of-memory variables you have, grandma! */
377 max_low_pfn
= end_pfn
;
379 high_memory
= (void *)__va(end_pfn
* PAGE_SIZE
- 1) + 1;
381 /* Remove active ranges so rediscovery with NUMA-awareness happens */
382 remove_all_active_ranges();
384 #ifdef CONFIG_ACPI_NUMA
386 * Parse SRAT to discover nodes.
392 numa_initmem_init(0, end_pfn
);
394 contig_initmem_init(0, end_pfn
);
397 /* Reserve direct mapping */
398 reserve_bootmem_generic(table_start
<< PAGE_SHIFT
,
399 (table_end
- table_start
) << PAGE_SHIFT
);
402 reserve_bootmem_generic(__pa_symbol(&_text
),
403 __pa_symbol(&_end
) - __pa_symbol(&_text
));
406 * reserve physical page 0 - it's a special BIOS page on many boxes,
407 * enabling clean reboots, SMP operation, laptop functions.
409 reserve_bootmem_generic(0, PAGE_SIZE
);
411 /* reserve ebda region */
413 reserve_bootmem_generic(ebda_addr
, ebda_size
);
415 /* reserve nodemap region */
417 reserve_bootmem_generic(nodemap_addr
, nodemap_size
);
421 /* Reserve SMP trampoline */
422 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE
, 2*PAGE_SIZE
);
425 #ifdef CONFIG_ACPI_SLEEP
427 * Reserve low memory region for sleep support.
429 acpi_reserve_bootmem();
434 efi_reserve_bootmem();
438 * Find and reserve possible boot-time SMP configuration:
441 #ifdef CONFIG_BLK_DEV_INITRD
442 if (boot_params
.hdr
.type_of_loader
&& boot_params
.hdr
.ramdisk_image
) {
443 unsigned long ramdisk_image
= boot_params
.hdr
.ramdisk_image
;
444 unsigned long ramdisk_size
= boot_params
.hdr
.ramdisk_size
;
445 unsigned long ramdisk_end
= ramdisk_image
+ ramdisk_size
;
446 unsigned long end_of_mem
= end_pfn
<< PAGE_SHIFT
;
448 if (ramdisk_end
<= end_of_mem
) {
449 reserve_bootmem_generic(ramdisk_image
, ramdisk_size
);
450 initrd_start
= ramdisk_image
+ PAGE_OFFSET
;
451 initrd_end
= initrd_start
+ramdisk_size
;
453 printk(KERN_ERR
"initrd extends beyond end of memory "
454 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
455 ramdisk_end
, end_of_mem
);
460 reserve_crashkernel();
467 * set this early, so we dont allocate cpu0
468 * if MADT list doesnt list BSP first
469 * mpparse.c/MP_processor_info() allocates logical cpu numbers.
471 cpu_set(0, cpu_present_map
);
474 * Read APIC and some other early information from ACPI tables.
482 * get boot-time SMP configuration:
484 if (smp_found_config
)
486 init_apic_mappings();
487 ioapic_init_mappings();
490 * We trust e820 completely. No explicit ROM probing in memory.
492 e820_reserve_resources(&code_resource
, &data_resource
, &bss_resource
);
493 e820_mark_nosave_regions();
495 /* request I/O space for devices used on all i[345]86 PCs */
496 for (i
= 0; i
< ARRAY_SIZE(standard_io_resources
); i
++)
497 request_resource(&ioport_resource
, &standard_io_resources
[i
]);
502 #if defined(CONFIG_VGA_CONSOLE)
503 if (!efi_enabled
|| (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY
))
504 conswitchp
= &vga_con
;
505 #elif defined(CONFIG_DUMMY_CONSOLE)
506 conswitchp
= &dummy_con
;
511 static int __cpuinit
get_model_name(struct cpuinfo_x86
*c
)
515 if (c
->extended_cpuid_level
< 0x80000004)
518 v
= (unsigned int *) c
->x86_model_id
;
519 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
520 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
521 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
522 c
->x86_model_id
[48] = 0;
527 static void __cpuinit
display_cacheinfo(struct cpuinfo_x86
*c
)
529 unsigned int n
, dummy
, eax
, ebx
, ecx
, edx
;
531 n
= c
->extended_cpuid_level
;
533 if (n
>= 0x80000005) {
534 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
535 printk(KERN_INFO
"CPU: L1 I Cache: %dK (%d bytes/line), "
536 "D cache %dK (%d bytes/line)\n",
537 edx
>>24, edx
&0xFF, ecx
>>24, ecx
&0xFF);
538 c
->x86_cache_size
= (ecx
>>24) + (edx
>>24);
539 /* On K8 L1 TLB is inclusive, so don't count it */
543 if (n
>= 0x80000006) {
544 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
545 ecx
= cpuid_ecx(0x80000006);
546 c
->x86_cache_size
= ecx
>> 16;
547 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
549 printk(KERN_INFO
"CPU: L2 Cache: %dK (%d bytes/line)\n",
550 c
->x86_cache_size
, ecx
& 0xFF);
552 if (n
>= 0x80000008) {
553 cpuid(0x80000008, &eax
, &dummy
, &dummy
, &dummy
);
554 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
555 c
->x86_phys_bits
= eax
& 0xff;
560 static int nearby_node(int apicid
)
564 for (i
= apicid
- 1; i
>= 0; i
--) {
565 node
= apicid_to_node
[i
];
566 if (node
!= NUMA_NO_NODE
&& node_online(node
))
569 for (i
= apicid
+ 1; i
< MAX_LOCAL_APIC
; i
++) {
570 node
= apicid_to_node
[i
];
571 if (node
!= NUMA_NO_NODE
&& node_online(node
))
574 return first_node(node_online_map
); /* Shouldn't happen */
579 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
580 * Assumes number of cores is a power of two.
582 static void __init
amd_detect_cmp(struct cpuinfo_x86
*c
)
587 int cpu
= smp_processor_id();
589 unsigned apicid
= hard_smp_processor_id();
591 bits
= c
->x86_coreid_bits
;
593 /* Low order bits define the core id (index of core in socket) */
594 c
->cpu_core_id
= c
->phys_proc_id
& ((1 << bits
)-1);
595 /* Convert the APIC ID into the socket ID */
596 c
->phys_proc_id
= phys_pkg_id(bits
);
599 node
= c
->phys_proc_id
;
600 if (apicid_to_node
[apicid
] != NUMA_NO_NODE
)
601 node
= apicid_to_node
[apicid
];
602 if (!node_online(node
)) {
603 /* Two possibilities here:
604 - The CPU is missing memory and no node was created.
605 In that case try picking one from a nearby CPU
606 - The APIC IDs differ from the HyperTransport node IDs
607 which the K8 northbridge parsing fills in.
608 Assume they are all increased by a constant offset,
609 but in the same order as the HT nodeids.
610 If that doesn't result in a usable node fall back to the
611 path for the previous case. */
613 int ht_nodeid
= apicid
- (cpu_data(0).phys_proc_id
<< bits
);
615 if (ht_nodeid
>= 0 &&
616 apicid_to_node
[ht_nodeid
] != NUMA_NO_NODE
)
617 node
= apicid_to_node
[ht_nodeid
];
618 /* Pick a nearby node */
619 if (!node_online(node
))
620 node
= nearby_node(apicid
);
622 numa_set_node(cpu
, node
);
624 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
629 static void __cpuinit
early_init_amd_mc(struct cpuinfo_x86
*c
)
634 /* Multi core CPU? */
635 if (c
->extended_cpuid_level
< 0x80000008)
638 ecx
= cpuid_ecx(0x80000008);
640 c
->x86_max_cores
= (ecx
& 0xff) + 1;
642 /* CPU telling us the core id bits shift? */
643 bits
= (ecx
>> 12) & 0xF;
645 /* Otherwise recompute */
647 while ((1 << bits
) < c
->x86_max_cores
)
651 c
->x86_coreid_bits
= bits
;
656 #define ENABLE_C1E_MASK 0x18000000
657 #define CPUID_PROCESSOR_SIGNATURE 1
658 #define CPUID_XFAM 0x0ff00000
659 #define CPUID_XFAM_K8 0x00000000
660 #define CPUID_XFAM_10H 0x00100000
661 #define CPUID_XFAM_11H 0x00200000
662 #define CPUID_XMOD 0x000f0000
663 #define CPUID_XMOD_REV_F 0x00040000
665 /* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
666 static __cpuinit
int amd_apic_timer_broken(void)
668 u32 lo
, hi
, eax
= cpuid_eax(CPUID_PROCESSOR_SIGNATURE
);
670 switch (eax
& CPUID_XFAM
) {
672 if ((eax
& CPUID_XMOD
) < CPUID_XMOD_REV_F
)
676 rdmsr(MSR_K8_ENABLE_C1E
, lo
, hi
);
677 if (lo
& ENABLE_C1E_MASK
)
681 /* err on the side of caution */
687 static void __cpuinit
early_init_amd(struct cpuinfo_x86
*c
)
689 early_init_amd_mc(c
);
691 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
692 if (c
->x86_power
& (1<<8))
693 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
696 static void __cpuinit
init_amd(struct cpuinfo_x86
*c
)
704 * Disable TLB flush filter by setting HWCR.FFDIS on K8
705 * bit 6 of msr C001_0015
707 * Errata 63 for SH-B3 steppings
708 * Errata 122 for all steppings (F+ have it disabled by default)
711 rdmsrl(MSR_K8_HWCR
, value
);
713 wrmsrl(MSR_K8_HWCR
, value
);
717 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
718 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
719 clear_bit(0*32+31, (unsigned long *)&c
->x86_capability
);
721 /* On C+ stepping K8 rep microcode works well for copy/memset */
722 level
= cpuid_eax(1);
723 if (c
->x86
== 15 && ((level
>= 0x0f48 && level
< 0x0f50) ||
725 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
726 if (c
->x86
== 0x10 || c
->x86
== 0x11)
727 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
729 /* Enable workaround for FXSAVE leak */
731 set_cpu_cap(c
, X86_FEATURE_FXSAVE_LEAK
);
733 level
= get_model_name(c
);
737 /* Should distinguish Models here, but this is only
738 a fallback anyways. */
739 strcpy(c
->x86_model_id
, "Hammer");
743 display_cacheinfo(c
);
745 /* Multi core CPU? */
746 if (c
->extended_cpuid_level
>= 0x80000008)
749 if (c
->extended_cpuid_level
>= 0x80000006 &&
750 (cpuid_edx(0x80000006) & 0xf000))
751 num_cache_leaves
= 4;
753 num_cache_leaves
= 3;
755 if (c
->x86
== 0xf || c
->x86
== 0x10 || c
->x86
== 0x11)
756 set_cpu_cap(c
, X86_FEATURE_K8
);
758 /* MFENCE stops RDTSC speculation */
759 set_cpu_cap(c
, X86_FEATURE_MFENCE_RDTSC
);
761 /* Family 10 doesn't support C states in MWAIT so don't use it */
762 if (c
->x86
== 0x10 && !force_mwait
)
763 clear_cpu_cap(c
, X86_FEATURE_MWAIT
);
765 if (amd_apic_timer_broken())
766 disable_apic_timer
= 1;
769 void __cpuinit
detect_ht(struct cpuinfo_x86
*c
)
772 u32 eax
, ebx
, ecx
, edx
;
773 int index_msb
, core_bits
;
775 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
778 if (!cpu_has(c
, X86_FEATURE_HT
))
780 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
783 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
785 if (smp_num_siblings
== 1) {
786 printk(KERN_INFO
"CPU: Hyper-Threading is disabled\n");
787 } else if (smp_num_siblings
> 1) {
789 if (smp_num_siblings
> NR_CPUS
) {
790 printk(KERN_WARNING
"CPU: Unsupported number of "
791 "siblings %d", smp_num_siblings
);
792 smp_num_siblings
= 1;
796 index_msb
= get_count_order(smp_num_siblings
);
797 c
->phys_proc_id
= phys_pkg_id(index_msb
);
799 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
801 index_msb
= get_count_order(smp_num_siblings
);
803 core_bits
= get_count_order(c
->x86_max_cores
);
805 c
->cpu_core_id
= phys_pkg_id(index_msb
) &
806 ((1 << core_bits
) - 1);
809 if ((c
->x86_max_cores
* smp_num_siblings
) > 1) {
810 printk(KERN_INFO
"CPU: Physical Processor ID: %d\n",
812 printk(KERN_INFO
"CPU: Processor Core ID: %d\n",
820 * find out the number of processor cores on the die
822 static int __cpuinit
intel_num_cpu_cores(struct cpuinfo_x86
*c
)
826 if (c
->cpuid_level
< 4)
829 cpuid_count(4, 0, &eax
, &t
, &t
, &t
);
832 return ((eax
>> 26) + 1);
837 static void srat_detect_node(void)
841 int cpu
= smp_processor_id();
842 int apicid
= hard_smp_processor_id();
844 /* Don't do the funky fallback heuristics the AMD version employs
846 node
= apicid_to_node
[apicid
];
847 if (node
== NUMA_NO_NODE
)
848 node
= first_node(node_online_map
);
849 numa_set_node(cpu
, node
);
851 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
855 static void __cpuinit
early_init_intel(struct cpuinfo_x86
*c
)
857 if ((c
->x86
== 0xf && c
->x86_model
>= 0x03) ||
858 (c
->x86
== 0x6 && c
->x86_model
>= 0x0e))
859 set_bit(X86_FEATURE_CONSTANT_TSC
, &c
->x86_capability
);
862 static void __cpuinit
init_intel(struct cpuinfo_x86
*c
)
867 init_intel_cacheinfo(c
);
868 if (c
->cpuid_level
> 9) {
869 unsigned eax
= cpuid_eax(10);
870 /* Check for version and the number of counters */
871 if ((eax
& 0xff) && (((eax
>>8) & 0xff) > 1))
872 set_cpu_cap(c
, X86_FEATURE_ARCH_PERFMON
);
877 rdmsr(MSR_IA32_MISC_ENABLE
, l1
, l2
);
879 set_cpu_cap(c
, X86_FEATURE_BTS
);
881 set_cpu_cap(c
, X86_FEATURE_PEBS
);
888 n
= c
->extended_cpuid_level
;
889 if (n
>= 0x80000008) {
890 unsigned eax
= cpuid_eax(0x80000008);
891 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
892 c
->x86_phys_bits
= eax
& 0xff;
893 /* CPUID workaround for Intel 0F34 CPU */
894 if (c
->x86_vendor
== X86_VENDOR_INTEL
&&
895 c
->x86
== 0xF && c
->x86_model
== 0x3 &&
897 c
->x86_phys_bits
= 36;
901 c
->x86_cache_alignment
= c
->x86_clflush_size
* 2;
902 if ((c
->x86
== 0xf && c
->x86_model
>= 0x03) ||
903 (c
->x86
== 0x6 && c
->x86_model
>= 0x0e))
904 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
906 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
907 set_cpu_cap(c
, X86_FEATURE_LFENCE_RDTSC
);
908 c
->x86_max_cores
= intel_num_cpu_cores(c
);
913 static void __cpuinit
get_cpu_vendor(struct cpuinfo_x86
*c
)
915 char *v
= c
->x86_vendor_id
;
917 if (!strcmp(v
, "AuthenticAMD"))
918 c
->x86_vendor
= X86_VENDOR_AMD
;
919 else if (!strcmp(v
, "GenuineIntel"))
920 c
->x86_vendor
= X86_VENDOR_INTEL
;
922 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
925 struct cpu_model_info
{
928 char *model_names
[16];
931 /* Do some early cpuid on the boot CPU to get some parameter that are
932 needed before check_bugs. Everything advanced is in identify_cpu
934 static void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
)
938 c
->loops_per_jiffy
= loops_per_jiffy
;
939 c
->x86_cache_size
= -1;
940 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
941 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
942 c
->x86_vendor_id
[0] = '\0'; /* Unset */
943 c
->x86_model_id
[0] = '\0'; /* Unset */
944 c
->x86_clflush_size
= 64;
945 c
->x86_cache_alignment
= c
->x86_clflush_size
;
946 c
->x86_max_cores
= 1;
947 c
->x86_coreid_bits
= 0;
948 c
->extended_cpuid_level
= 0;
949 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
951 /* Get vendor name */
952 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
953 (unsigned int *)&c
->x86_vendor_id
[0],
954 (unsigned int *)&c
->x86_vendor_id
[8],
955 (unsigned int *)&c
->x86_vendor_id
[4]);
959 /* Initialize the standard set of capabilities */
960 /* Note that the vendor-specific code below might override */
962 /* Intel-defined flags: level 0x00000001 */
963 if (c
->cpuid_level
>= 0x00000001) {
965 cpuid(0x00000001, &tfms
, &misc
, &c
->x86_capability
[4],
966 &c
->x86_capability
[0]);
967 c
->x86
= (tfms
>> 8) & 0xf;
968 c
->x86_model
= (tfms
>> 4) & 0xf;
969 c
->x86_mask
= tfms
& 0xf;
971 c
->x86
+= (tfms
>> 20) & 0xff;
973 c
->x86_model
+= ((tfms
>> 16) & 0xF) << 4;
974 if (c
->x86_capability
[0] & (1<<19))
975 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
977 /* Have CPUID level 0 only - unheard of */
982 c
->phys_proc_id
= (cpuid_ebx(1) >> 24) & 0xff;
984 /* AMD-defined flags: level 0x80000001 */
985 xlvl
= cpuid_eax(0x80000000);
986 c
->extended_cpuid_level
= xlvl
;
987 if ((xlvl
& 0xffff0000) == 0x80000000) {
988 if (xlvl
>= 0x80000001) {
989 c
->x86_capability
[1] = cpuid_edx(0x80000001);
990 c
->x86_capability
[6] = cpuid_ecx(0x80000001);
992 if (xlvl
>= 0x80000004)
993 get_model_name(c
); /* Default name */
996 /* Transmeta-defined flags: level 0x80860001 */
997 xlvl
= cpuid_eax(0x80860000);
998 if ((xlvl
& 0xffff0000) == 0x80860000) {
999 /* Don't set x86_cpuid_level here for now to not confuse. */
1000 if (xlvl
>= 0x80860001)
1001 c
->x86_capability
[2] = cpuid_edx(0x80860001);
1004 c
->extended_cpuid_level
= cpuid_eax(0x80000000);
1005 if (c
->extended_cpuid_level
>= 0x80000007)
1006 c
->x86_power
= cpuid_edx(0x80000007);
1008 switch (c
->x86_vendor
) {
1009 case X86_VENDOR_AMD
:
1017 * This does the hard work of actually picking apart the CPU stuff...
1019 void __cpuinit
identify_cpu(struct cpuinfo_x86
*c
)
1023 early_identify_cpu(c
);
1025 init_scattered_cpuid_features(c
);
1027 c
->apicid
= phys_pkg_id(0);
1030 * Vendor-specific initialization. In this section we
1031 * canonicalize the feature flags, meaning if there are
1032 * features a certain CPU supports which CPUID doesn't
1033 * tell us, CPUID claiming incorrect flags, or other bugs,
1034 * we handle them here.
1036 * At the end of this section, c->x86_capability better
1037 * indicate the features this CPU genuinely supports!
1039 switch (c
->x86_vendor
) {
1040 case X86_VENDOR_AMD
:
1044 case X86_VENDOR_INTEL
:
1048 case X86_VENDOR_UNKNOWN
:
1050 display_cacheinfo(c
);
1054 select_idle_routine(c
);
1058 * On SMP, boot_cpu_data holds the common feature set between
1059 * all CPUs; so make sure that we indicate which features are
1060 * common between the CPUs. The first time this routine gets
1061 * executed, c == &boot_cpu_data.
1063 if (c
!= &boot_cpu_data
) {
1064 /* AND the already accumulated flags with these */
1065 for (i
= 0; i
< NCAPINTS
; i
++)
1066 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
1069 #ifdef CONFIG_X86_MCE
1072 if (c
!= &boot_cpu_data
)
1075 numa_add_cpu(smp_processor_id());
1078 switch (c
->x86_vendor
) {
1079 case X86_VENDOR_AMD
:
1082 case X86_VENDOR_INTEL
:
1083 early_init_intel(c
);
1088 void __cpuinit
print_cpu_info(struct cpuinfo_x86
*c
)
1090 if (c
->x86_model_id
[0])
1091 printk(KERN_INFO
"%s", c
->x86_model_id
);
1093 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1094 printk(KERN_CONT
" stepping %02x\n", c
->x86_mask
);
1096 printk(KERN_CONT
"\n");
1100 * Get CPU information for use by the procfs.
1103 static int show_cpuinfo(struct seq_file
*m
, void *v
)
1105 struct cpuinfo_x86
*c
= v
;
1109 * These flag bits must match the definitions in <asm/cpufeature.h>.
1110 * NULL means this bit is undefined or reserved; either way it doesn't
1111 * have meaning as far as Linux is concerned. Note that it's important
1112 * to realize there is a difference between this table and CPUID -- if
1113 * applications want to get the raw CPUID data, they should access
1114 * /dev/cpu/<cpu_nr>/cpuid instead.
1116 static const char *const x86_cap_flags
[] = {
1118 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1119 "cx8", "apic", NULL
, "sep", "mtrr", "pge", "mca", "cmov",
1120 "pat", "pse36", "pn", "clflush", NULL
, "dts", "acpi", "mmx",
1121 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", "pbe",
1124 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1125 NULL
, NULL
, NULL
, "syscall", NULL
, NULL
, NULL
, NULL
,
1126 NULL
, NULL
, NULL
, NULL
, "nx", NULL
, "mmxext", NULL
,
1127 NULL
, "fxsr_opt", "pdpe1gb", "rdtscp", NULL
, "lm",
1128 "3dnowext", "3dnow",
1130 /* Transmeta-defined */
1131 "recovery", "longrun", NULL
, "lrti", NULL
, NULL
, NULL
, NULL
,
1132 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1133 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1134 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1136 /* Other (Linux-defined) */
1137 "cxmmx", "k6_mtrr", "cyrix_arr", "centaur_mcr",
1138 NULL
, NULL
, NULL
, NULL
,
1139 "constant_tsc", "up", NULL
, "arch_perfmon",
1140 "pebs", "bts", NULL
, "sync_rdtsc",
1141 "rep_good", NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1142 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1144 /* Intel-defined (#2) */
1145 "pni", NULL
, NULL
, "monitor", "ds_cpl", "vmx", "smx", "est",
1146 "tm2", "ssse3", "cid", NULL
, NULL
, "cx16", "xtpr", NULL
,
1147 NULL
, NULL
, "dca", "sse4_1", "sse4_2", NULL
, NULL
, "popcnt",
1148 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1150 /* VIA/Cyrix/Centaur-defined */
1151 NULL
, NULL
, "rng", "rng_en", NULL
, NULL
, "ace", "ace_en",
1152 "ace2", "ace2_en", "phe", "phe_en", "pmm", "pmm_en", NULL
, NULL
,
1153 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1154 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1156 /* AMD-defined (#2) */
1157 "lahf_lm", "cmp_legacy", "svm", "extapic",
1158 "cr8_legacy", "abm", "sse4a", "misalignsse",
1159 "3dnowprefetch", "osvw", "ibs", "sse5",
1160 "skinit", "wdt", NULL
, NULL
,
1161 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1162 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1164 /* Auxiliary (Linux-defined) */
1165 "ida", NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1166 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1167 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1168 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1170 static const char *const x86_power_flags
[] = {
1171 "ts", /* temperature sensor */
1172 "fid", /* frequency id control */
1173 "vid", /* voltage id control */
1174 "ttp", /* thermal trip */
1179 "", /* tsc invariant mapped to constant_tsc */
1188 seq_printf(m
, "processor\t: %u\n"
1190 "cpu family\t: %d\n"
1192 "model name\t: %s\n",
1194 c
->x86_vendor_id
[0] ? c
->x86_vendor_id
: "unknown",
1197 c
->x86_model_id
[0] ? c
->x86_model_id
: "unknown");
1199 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1200 seq_printf(m
, "stepping\t: %d\n", c
->x86_mask
);
1202 seq_printf(m
, "stepping\t: unknown\n");
1204 if (cpu_has(c
, X86_FEATURE_TSC
)) {
1205 unsigned int freq
= cpufreq_quick_get((unsigned)cpu
);
1209 seq_printf(m
, "cpu MHz\t\t: %u.%03u\n",
1210 freq
/ 1000, (freq
% 1000));
1214 if (c
->x86_cache_size
>= 0)
1215 seq_printf(m
, "cache size\t: %d KB\n", c
->x86_cache_size
);
1218 if (smp_num_siblings
* c
->x86_max_cores
> 1) {
1219 seq_printf(m
, "physical id\t: %d\n", c
->phys_proc_id
);
1220 seq_printf(m
, "siblings\t: %d\n",
1221 cpus_weight(per_cpu(cpu_core_map
, cpu
)));
1222 seq_printf(m
, "core id\t\t: %d\n", c
->cpu_core_id
);
1223 seq_printf(m
, "cpu cores\t: %d\n", c
->booted_cores
);
1229 "fpu_exception\t: yes\n"
1230 "cpuid level\t: %d\n"
1235 for (i
= 0; i
< 32*NCAPINTS
; i
++)
1236 if (cpu_has(c
, i
) && x86_cap_flags
[i
] != NULL
)
1237 seq_printf(m
, " %s", x86_cap_flags
[i
]);
1239 seq_printf(m
, "\nbogomips\t: %lu.%02lu\n",
1240 c
->loops_per_jiffy
/(500000/HZ
),
1241 (c
->loops_per_jiffy
/(5000/HZ
)) % 100);
1243 if (c
->x86_tlbsize
> 0)
1244 seq_printf(m
, "TLB size\t: %d 4K pages\n", c
->x86_tlbsize
);
1245 seq_printf(m
, "clflush size\t: %d\n", c
->x86_clflush_size
);
1246 seq_printf(m
, "cache_alignment\t: %d\n", c
->x86_cache_alignment
);
1248 seq_printf(m
, "address sizes\t: %u bits physical, %u bits virtual\n",
1249 c
->x86_phys_bits
, c
->x86_virt_bits
);
1251 seq_printf(m
, "power management:");
1252 for (i
= 0; i
< 32; i
++) {
1253 if (c
->x86_power
& (1 << i
)) {
1254 if (i
< ARRAY_SIZE(x86_power_flags
) &&
1256 seq_printf(m
, "%s%s",
1257 x86_power_flags
[i
][0]?" ":"",
1258 x86_power_flags
[i
]);
1260 seq_printf(m
, " [%d]", i
);
1264 seq_printf(m
, "\n\n");
1269 static void *c_start(struct seq_file
*m
, loff_t
*pos
)
1271 if (*pos
== 0) /* just in case, cpu 0 is not the first */
1272 *pos
= first_cpu(cpu_online_map
);
1273 if ((*pos
) < NR_CPUS
&& cpu_online(*pos
))
1274 return &cpu_data(*pos
);
1278 static void *c_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1280 *pos
= next_cpu(*pos
, cpu_online_map
);
1281 return c_start(m
, pos
);
1284 static void c_stop(struct seq_file
*m
, void *v
)
1288 struct seq_operations cpuinfo_op
= {
1292 .show
= show_cpuinfo
,