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 __u32 cleared_cpu_caps
[NCAPINTS
] __cpuinitdata
;
85 unsigned long mmu_cr4_features
;
87 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
90 unsigned long saved_video_mode
;
92 int force_mwait __cpuinitdata
;
98 char dmi_alloc_data
[DMI_MAX_DATA
];
103 struct screen_info screen_info
;
104 EXPORT_SYMBOL(screen_info
);
105 struct sys_desc_table_struct
{
106 unsigned short length
;
107 unsigned char table
[0];
110 struct edid_info edid_info
;
111 EXPORT_SYMBOL_GPL(edid_info
);
113 extern int root_mountflags
;
115 char __initdata 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 IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
140 static struct resource data_resource
= {
141 .name
= "Kernel data",
144 .flags
= IORESOURCE_RAM
,
146 static struct resource code_resource
= {
147 .name
= "Kernel code",
150 .flags
= IORESOURCE_RAM
,
152 static struct resource bss_resource
= {
153 .name
= "Kernel bss",
156 .flags
= IORESOURCE_RAM
,
159 static void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
);
161 #ifdef CONFIG_PROC_VMCORE
162 /* elfcorehdr= specifies the location of elf core header
163 * stored by the crashed kernel. This option will be passed
164 * by kexec loader to the capture kernel.
166 static int __init
setup_elfcorehdr(char *arg
)
171 elfcorehdr_addr
= memparse(arg
, &end
);
172 return end
> arg
? 0 : -EINVAL
;
174 early_param("elfcorehdr", setup_elfcorehdr
);
179 contig_initmem_init(unsigned long start_pfn
, unsigned long end_pfn
)
181 unsigned long bootmap_size
, bootmap
;
183 bootmap_size
= bootmem_bootmap_pages(end_pfn
)<<PAGE_SHIFT
;
184 bootmap
= find_e820_area(0, end_pfn
<<PAGE_SHIFT
, bootmap_size
);
186 panic("Cannot find bootmem map of size %ld\n", bootmap_size
);
187 bootmap_size
= init_bootmem(bootmap
>> PAGE_SHIFT
, end_pfn
);
188 e820_register_active_regions(0, start_pfn
, end_pfn
);
189 free_bootmem_with_active_regions(0, end_pfn
);
190 reserve_bootmem(bootmap
, bootmap_size
);
194 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
196 #ifdef CONFIG_EDD_MODULE
200 * copy_edd() - Copy the BIOS EDD information
201 * from boot_params into a safe place.
204 static inline void copy_edd(void)
206 memcpy(edd
.mbr_signature
, boot_params
.edd_mbr_sig_buffer
,
207 sizeof(edd
.mbr_signature
));
208 memcpy(edd
.edd_info
, boot_params
.eddbuf
, sizeof(edd
.edd_info
));
209 edd
.mbr_signature_nr
= boot_params
.edd_mbr_sig_buf_entries
;
210 edd
.edd_info_nr
= boot_params
.eddbuf_entries
;
213 static inline void copy_edd(void)
219 static void __init
reserve_crashkernel(void)
221 unsigned long long free_mem
;
222 unsigned long long crash_size
, crash_base
;
226 ((unsigned long long)max_low_pfn
- min_low_pfn
) << PAGE_SHIFT
;
228 ret
= parse_crashkernel(boot_command_line
, free_mem
,
229 &crash_size
, &crash_base
);
230 if (ret
== 0 && crash_size
) {
231 if (crash_base
> 0) {
232 printk(KERN_INFO
"Reserving %ldMB of memory at %ldMB "
233 "for crashkernel (System RAM: %ldMB)\n",
234 (unsigned long)(crash_size
>> 20),
235 (unsigned long)(crash_base
>> 20),
236 (unsigned long)(free_mem
>> 20));
237 crashk_res
.start
= crash_base
;
238 crashk_res
.end
= crash_base
+ crash_size
- 1;
239 reserve_bootmem(crash_base
, crash_size
);
241 printk(KERN_INFO
"crashkernel reservation failed - "
242 "you have to specify a base address\n");
246 static inline void __init
reserve_crashkernel(void)
250 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
251 void __attribute__((weak
)) __init
memory_setup(void)
253 machine_specific_memory_setup();
256 void __init
setup_arch(char **cmdline_p
)
260 printk(KERN_INFO
"Command line: %s\n", boot_command_line
);
262 ROOT_DEV
= old_decode_dev(boot_params
.hdr
.root_dev
);
263 screen_info
= boot_params
.screen_info
;
264 edid_info
= boot_params
.edid_info
;
265 saved_video_mode
= boot_params
.hdr
.vid_mode
;
266 bootloader_type
= boot_params
.hdr
.type_of_loader
;
268 #ifdef CONFIG_BLK_DEV_RAM
269 rd_image_start
= boot_params
.hdr
.ram_size
& RAMDISK_IMAGE_START_MASK
;
270 rd_prompt
= ((boot_params
.hdr
.ram_size
& RAMDISK_PROMPT_FLAG
) != 0);
271 rd_doload
= ((boot_params
.hdr
.ram_size
& RAMDISK_LOAD_FLAG
) != 0);
274 if (!strncmp((char *)&boot_params
.efi_info
.efi_loader_signature
,
284 if (!boot_params
.hdr
.root_flags
)
285 root_mountflags
&= ~MS_RDONLY
;
286 init_mm
.start_code
= (unsigned long) &_text
;
287 init_mm
.end_code
= (unsigned long) &_etext
;
288 init_mm
.end_data
= (unsigned long) &_edata
;
289 init_mm
.brk
= (unsigned long) &_end
;
291 code_resource
.start
= virt_to_phys(&_text
);
292 code_resource
.end
= virt_to_phys(&_etext
)-1;
293 data_resource
.start
= virt_to_phys(&_etext
);
294 data_resource
.end
= virt_to_phys(&_edata
)-1;
295 bss_resource
.start
= virt_to_phys(&__bss_start
);
296 bss_resource
.end
= virt_to_phys(&__bss_stop
)-1;
298 early_identify_cpu(&boot_cpu_data
);
300 strlcpy(command_line
, boot_command_line
, COMMAND_LINE_SIZE
);
301 *cmdline_p
= command_line
;
305 finish_e820_parsing();
307 early_gart_iommu_check();
309 e820_register_active_regions(0, 0, -1UL);
311 * partially used pages are not usable - thus
312 * we are rounding upwards:
314 end_pfn
= e820_end_of_ram();
315 /* update e820 for memory not covered by WB MTRRs */
317 if (mtrr_trim_uncached_memory(end_pfn
)) {
318 e820_register_active_regions(0, 0, -1UL);
319 end_pfn
= e820_end_of_ram();
322 num_physpages
= end_pfn
;
326 init_memory_mapping(0, (end_pfn_map
<< PAGE_SHIFT
));
335 /* setup to use the early static init tables during kernel startup */
336 x86_cpu_to_apicid_early_ptr
= (void *)&x86_cpu_to_apicid_init
;
337 x86_bios_cpu_apicid_early_ptr
= (void *)&x86_bios_cpu_apicid_init
;
339 x86_cpu_to_node_map_early_ptr
= (void *)&x86_cpu_to_node_map_init
;
341 x86_bios_cpu_apicid_early_ptr
= (void *)&x86_bios_cpu_apicid_init
;
346 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
347 * Call this early for SRAT node setup.
349 acpi_boot_table_init();
352 /* How many end-of-memory variables you have, grandma! */
353 max_low_pfn
= end_pfn
;
355 high_memory
= (void *)__va(end_pfn
* PAGE_SIZE
- 1) + 1;
357 /* Remove active ranges so rediscovery with NUMA-awareness happens */
358 remove_all_active_ranges();
360 #ifdef CONFIG_ACPI_NUMA
362 * Parse SRAT to discover nodes.
368 numa_initmem_init(0, end_pfn
);
370 contig_initmem_init(0, end_pfn
);
373 early_res_to_bootmem();
375 #ifdef CONFIG_ACPI_SLEEP
377 * Reserve low memory region for sleep support.
379 acpi_reserve_bootmem();
384 efi_reserve_bootmem();
388 * Find and reserve possible boot-time SMP configuration:
391 #ifdef CONFIG_BLK_DEV_INITRD
392 if (boot_params
.hdr
.type_of_loader
&& boot_params
.hdr
.ramdisk_image
) {
393 unsigned long ramdisk_image
= boot_params
.hdr
.ramdisk_image
;
394 unsigned long ramdisk_size
= boot_params
.hdr
.ramdisk_size
;
395 unsigned long ramdisk_end
= ramdisk_image
+ ramdisk_size
;
396 unsigned long end_of_mem
= end_pfn
<< PAGE_SHIFT
;
398 if (ramdisk_end
<= end_of_mem
) {
399 reserve_bootmem_generic(ramdisk_image
, ramdisk_size
);
400 initrd_start
= ramdisk_image
+ PAGE_OFFSET
;
401 initrd_end
= initrd_start
+ramdisk_size
;
403 /* Assumes everything on node 0 */
404 free_bootmem(ramdisk_image
, ramdisk_size
);
405 printk(KERN_ERR
"initrd extends beyond end of memory "
406 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
407 ramdisk_end
, end_of_mem
);
412 reserve_crashkernel();
419 * set this early, so we dont allocate cpu0
420 * if MADT list doesnt list BSP first
421 * mpparse.c/MP_processor_info() allocates logical cpu numbers.
423 cpu_set(0, cpu_present_map
);
426 * Read APIC and some other early information from ACPI tables.
434 * get boot-time SMP configuration:
436 if (smp_found_config
)
438 init_apic_mappings();
439 ioapic_init_mappings();
442 * We trust e820 completely. No explicit ROM probing in memory.
444 e820_reserve_resources(&code_resource
, &data_resource
, &bss_resource
);
445 e820_mark_nosave_regions();
447 /* request I/O space for devices used on all i[345]86 PCs */
448 for (i
= 0; i
< ARRAY_SIZE(standard_io_resources
); i
++)
449 request_resource(&ioport_resource
, &standard_io_resources
[i
]);
454 #if defined(CONFIG_VGA_CONSOLE)
455 if (!efi_enabled
|| (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY
))
456 conswitchp
= &vga_con
;
457 #elif defined(CONFIG_DUMMY_CONSOLE)
458 conswitchp
= &dummy_con
;
463 static int __cpuinit
get_model_name(struct cpuinfo_x86
*c
)
467 if (c
->extended_cpuid_level
< 0x80000004)
470 v
= (unsigned int *) c
->x86_model_id
;
471 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
472 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
473 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
474 c
->x86_model_id
[48] = 0;
479 static void __cpuinit
display_cacheinfo(struct cpuinfo_x86
*c
)
481 unsigned int n
, dummy
, eax
, ebx
, ecx
, edx
;
483 n
= c
->extended_cpuid_level
;
485 if (n
>= 0x80000005) {
486 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
487 printk(KERN_INFO
"CPU: L1 I Cache: %dK (%d bytes/line), "
488 "D cache %dK (%d bytes/line)\n",
489 edx
>>24, edx
&0xFF, ecx
>>24, ecx
&0xFF);
490 c
->x86_cache_size
= (ecx
>>24) + (edx
>>24);
491 /* On K8 L1 TLB is inclusive, so don't count it */
495 if (n
>= 0x80000006) {
496 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
497 ecx
= cpuid_ecx(0x80000006);
498 c
->x86_cache_size
= ecx
>> 16;
499 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
501 printk(KERN_INFO
"CPU: L2 Cache: %dK (%d bytes/line)\n",
502 c
->x86_cache_size
, ecx
& 0xFF);
504 if (n
>= 0x80000008) {
505 cpuid(0x80000008, &eax
, &dummy
, &dummy
, &dummy
);
506 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
507 c
->x86_phys_bits
= eax
& 0xff;
512 static int nearby_node(int apicid
)
516 for (i
= apicid
- 1; i
>= 0; i
--) {
517 node
= apicid_to_node
[i
];
518 if (node
!= NUMA_NO_NODE
&& node_online(node
))
521 for (i
= apicid
+ 1; i
< MAX_LOCAL_APIC
; i
++) {
522 node
= apicid_to_node
[i
];
523 if (node
!= NUMA_NO_NODE
&& node_online(node
))
526 return first_node(node_online_map
); /* Shouldn't happen */
531 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
532 * Assumes number of cores is a power of two.
534 static void __init
amd_detect_cmp(struct cpuinfo_x86
*c
)
539 int cpu
= smp_processor_id();
541 unsigned apicid
= hard_smp_processor_id();
543 bits
= c
->x86_coreid_bits
;
545 /* Low order bits define the core id (index of core in socket) */
546 c
->cpu_core_id
= c
->phys_proc_id
& ((1 << bits
)-1);
547 /* Convert the APIC ID into the socket ID */
548 c
->phys_proc_id
= phys_pkg_id(bits
);
551 node
= c
->phys_proc_id
;
552 if (apicid_to_node
[apicid
] != NUMA_NO_NODE
)
553 node
= apicid_to_node
[apicid
];
554 if (!node_online(node
)) {
555 /* Two possibilities here:
556 - The CPU is missing memory and no node was created.
557 In that case try picking one from a nearby CPU
558 - The APIC IDs differ from the HyperTransport node IDs
559 which the K8 northbridge parsing fills in.
560 Assume they are all increased by a constant offset,
561 but in the same order as the HT nodeids.
562 If that doesn't result in a usable node fall back to the
563 path for the previous case. */
565 int ht_nodeid
= apicid
- (cpu_data(0).phys_proc_id
<< bits
);
567 if (ht_nodeid
>= 0 &&
568 apicid_to_node
[ht_nodeid
] != NUMA_NO_NODE
)
569 node
= apicid_to_node
[ht_nodeid
];
570 /* Pick a nearby node */
571 if (!node_online(node
))
572 node
= nearby_node(apicid
);
574 numa_set_node(cpu
, node
);
576 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
581 static void __cpuinit
early_init_amd_mc(struct cpuinfo_x86
*c
)
586 /* Multi core CPU? */
587 if (c
->extended_cpuid_level
< 0x80000008)
590 ecx
= cpuid_ecx(0x80000008);
592 c
->x86_max_cores
= (ecx
& 0xff) + 1;
594 /* CPU telling us the core id bits shift? */
595 bits
= (ecx
>> 12) & 0xF;
597 /* Otherwise recompute */
599 while ((1 << bits
) < c
->x86_max_cores
)
603 c
->x86_coreid_bits
= bits
;
608 #define ENABLE_C1E_MASK 0x18000000
609 #define CPUID_PROCESSOR_SIGNATURE 1
610 #define CPUID_XFAM 0x0ff00000
611 #define CPUID_XFAM_K8 0x00000000
612 #define CPUID_XFAM_10H 0x00100000
613 #define CPUID_XFAM_11H 0x00200000
614 #define CPUID_XMOD 0x000f0000
615 #define CPUID_XMOD_REV_F 0x00040000
617 /* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
618 static __cpuinit
int amd_apic_timer_broken(void)
620 u32 lo
, hi
, eax
= cpuid_eax(CPUID_PROCESSOR_SIGNATURE
);
622 switch (eax
& CPUID_XFAM
) {
624 if ((eax
& CPUID_XMOD
) < CPUID_XMOD_REV_F
)
628 rdmsr(MSR_K8_ENABLE_C1E
, lo
, hi
);
629 if (lo
& ENABLE_C1E_MASK
)
633 /* err on the side of caution */
639 static void __cpuinit
early_init_amd(struct cpuinfo_x86
*c
)
641 early_init_amd_mc(c
);
643 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
644 if (c
->x86_power
& (1<<8))
645 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
648 static void __cpuinit
init_amd(struct cpuinfo_x86
*c
)
656 * Disable TLB flush filter by setting HWCR.FFDIS on K8
657 * bit 6 of msr C001_0015
659 * Errata 63 for SH-B3 steppings
660 * Errata 122 for all steppings (F+ have it disabled by default)
663 rdmsrl(MSR_K8_HWCR
, value
);
665 wrmsrl(MSR_K8_HWCR
, value
);
669 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
670 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
671 clear_bit(0*32+31, (unsigned long *)&c
->x86_capability
);
673 /* On C+ stepping K8 rep microcode works well for copy/memset */
674 level
= cpuid_eax(1);
675 if (c
->x86
== 15 && ((level
>= 0x0f48 && level
< 0x0f50) ||
677 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
678 if (c
->x86
== 0x10 || c
->x86
== 0x11)
679 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
681 /* Enable workaround for FXSAVE leak */
683 set_cpu_cap(c
, X86_FEATURE_FXSAVE_LEAK
);
685 level
= get_model_name(c
);
689 /* Should distinguish Models here, but this is only
690 a fallback anyways. */
691 strcpy(c
->x86_model_id
, "Hammer");
695 display_cacheinfo(c
);
697 /* Multi core CPU? */
698 if (c
->extended_cpuid_level
>= 0x80000008)
701 if (c
->extended_cpuid_level
>= 0x80000006 &&
702 (cpuid_edx(0x80000006) & 0xf000))
703 num_cache_leaves
= 4;
705 num_cache_leaves
= 3;
707 if (c
->x86
== 0xf || c
->x86
== 0x10 || c
->x86
== 0x11)
708 set_cpu_cap(c
, X86_FEATURE_K8
);
710 /* MFENCE stops RDTSC speculation */
711 set_cpu_cap(c
, X86_FEATURE_MFENCE_RDTSC
);
713 if (amd_apic_timer_broken())
714 disable_apic_timer
= 1;
717 void __cpuinit
detect_ht(struct cpuinfo_x86
*c
)
720 u32 eax
, ebx
, ecx
, edx
;
721 int index_msb
, core_bits
;
723 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
726 if (!cpu_has(c
, X86_FEATURE_HT
))
728 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
731 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
733 if (smp_num_siblings
== 1) {
734 printk(KERN_INFO
"CPU: Hyper-Threading is disabled\n");
735 } else if (smp_num_siblings
> 1) {
737 if (smp_num_siblings
> NR_CPUS
) {
738 printk(KERN_WARNING
"CPU: Unsupported number of "
739 "siblings %d", smp_num_siblings
);
740 smp_num_siblings
= 1;
744 index_msb
= get_count_order(smp_num_siblings
);
745 c
->phys_proc_id
= phys_pkg_id(index_msb
);
747 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
749 index_msb
= get_count_order(smp_num_siblings
);
751 core_bits
= get_count_order(c
->x86_max_cores
);
753 c
->cpu_core_id
= phys_pkg_id(index_msb
) &
754 ((1 << core_bits
) - 1);
757 if ((c
->x86_max_cores
* smp_num_siblings
) > 1) {
758 printk(KERN_INFO
"CPU: Physical Processor ID: %d\n",
760 printk(KERN_INFO
"CPU: Processor Core ID: %d\n",
768 * find out the number of processor cores on the die
770 static int __cpuinit
intel_num_cpu_cores(struct cpuinfo_x86
*c
)
774 if (c
->cpuid_level
< 4)
777 cpuid_count(4, 0, &eax
, &t
, &t
, &t
);
780 return ((eax
>> 26) + 1);
785 static void srat_detect_node(void)
789 int cpu
= smp_processor_id();
790 int apicid
= hard_smp_processor_id();
792 /* Don't do the funky fallback heuristics the AMD version employs
794 node
= apicid_to_node
[apicid
];
795 if (node
== NUMA_NO_NODE
)
796 node
= first_node(node_online_map
);
797 numa_set_node(cpu
, node
);
799 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
803 static void __cpuinit
early_init_intel(struct cpuinfo_x86
*c
)
805 if ((c
->x86
== 0xf && c
->x86_model
>= 0x03) ||
806 (c
->x86
== 0x6 && c
->x86_model
>= 0x0e))
807 set_bit(X86_FEATURE_CONSTANT_TSC
, &c
->x86_capability
);
810 static void __cpuinit
init_intel(struct cpuinfo_x86
*c
)
815 init_intel_cacheinfo(c
);
816 if (c
->cpuid_level
> 9) {
817 unsigned eax
= cpuid_eax(10);
818 /* Check for version and the number of counters */
819 if ((eax
& 0xff) && (((eax
>>8) & 0xff) > 1))
820 set_cpu_cap(c
, X86_FEATURE_ARCH_PERFMON
);
825 rdmsr(MSR_IA32_MISC_ENABLE
, l1
, l2
);
827 set_cpu_cap(c
, X86_FEATURE_BTS
);
829 set_cpu_cap(c
, X86_FEATURE_PEBS
);
836 n
= c
->extended_cpuid_level
;
837 if (n
>= 0x80000008) {
838 unsigned eax
= cpuid_eax(0x80000008);
839 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
840 c
->x86_phys_bits
= eax
& 0xff;
841 /* CPUID workaround for Intel 0F34 CPU */
842 if (c
->x86_vendor
== X86_VENDOR_INTEL
&&
843 c
->x86
== 0xF && c
->x86_model
== 0x3 &&
845 c
->x86_phys_bits
= 36;
849 c
->x86_cache_alignment
= c
->x86_clflush_size
* 2;
850 if ((c
->x86
== 0xf && c
->x86_model
>= 0x03) ||
851 (c
->x86
== 0x6 && c
->x86_model
>= 0x0e))
852 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
854 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
855 set_cpu_cap(c
, X86_FEATURE_LFENCE_RDTSC
);
856 c
->x86_max_cores
= intel_num_cpu_cores(c
);
861 static void __cpuinit
get_cpu_vendor(struct cpuinfo_x86
*c
)
863 char *v
= c
->x86_vendor_id
;
865 if (!strcmp(v
, "AuthenticAMD"))
866 c
->x86_vendor
= X86_VENDOR_AMD
;
867 else if (!strcmp(v
, "GenuineIntel"))
868 c
->x86_vendor
= X86_VENDOR_INTEL
;
870 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
873 struct cpu_model_info
{
876 char *model_names
[16];
879 /* Do some early cpuid on the boot CPU to get some parameter that are
880 needed before check_bugs. Everything advanced is in identify_cpu
882 static void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
)
886 c
->loops_per_jiffy
= loops_per_jiffy
;
887 c
->x86_cache_size
= -1;
888 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
889 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
890 c
->x86_vendor_id
[0] = '\0'; /* Unset */
891 c
->x86_model_id
[0] = '\0'; /* Unset */
892 c
->x86_clflush_size
= 64;
893 c
->x86_cache_alignment
= c
->x86_clflush_size
;
894 c
->x86_max_cores
= 1;
895 c
->x86_coreid_bits
= 0;
896 c
->extended_cpuid_level
= 0;
897 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
899 /* Get vendor name */
900 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
901 (unsigned int *)&c
->x86_vendor_id
[0],
902 (unsigned int *)&c
->x86_vendor_id
[8],
903 (unsigned int *)&c
->x86_vendor_id
[4]);
907 /* Initialize the standard set of capabilities */
908 /* Note that the vendor-specific code below might override */
910 /* Intel-defined flags: level 0x00000001 */
911 if (c
->cpuid_level
>= 0x00000001) {
913 cpuid(0x00000001, &tfms
, &misc
, &c
->x86_capability
[4],
914 &c
->x86_capability
[0]);
915 c
->x86
= (tfms
>> 8) & 0xf;
916 c
->x86_model
= (tfms
>> 4) & 0xf;
917 c
->x86_mask
= tfms
& 0xf;
919 c
->x86
+= (tfms
>> 20) & 0xff;
921 c
->x86_model
+= ((tfms
>> 16) & 0xF) << 4;
922 if (c
->x86_capability
[0] & (1<<19))
923 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
925 /* Have CPUID level 0 only - unheard of */
930 c
->phys_proc_id
= (cpuid_ebx(1) >> 24) & 0xff;
932 /* AMD-defined flags: level 0x80000001 */
933 xlvl
= cpuid_eax(0x80000000);
934 c
->extended_cpuid_level
= xlvl
;
935 if ((xlvl
& 0xffff0000) == 0x80000000) {
936 if (xlvl
>= 0x80000001) {
937 c
->x86_capability
[1] = cpuid_edx(0x80000001);
938 c
->x86_capability
[6] = cpuid_ecx(0x80000001);
940 if (xlvl
>= 0x80000004)
941 get_model_name(c
); /* Default name */
944 /* Transmeta-defined flags: level 0x80860001 */
945 xlvl
= cpuid_eax(0x80860000);
946 if ((xlvl
& 0xffff0000) == 0x80860000) {
947 /* Don't set x86_cpuid_level here for now to not confuse. */
948 if (xlvl
>= 0x80860001)
949 c
->x86_capability
[2] = cpuid_edx(0x80860001);
952 c
->extended_cpuid_level
= cpuid_eax(0x80000000);
953 if (c
->extended_cpuid_level
>= 0x80000007)
954 c
->x86_power
= cpuid_edx(0x80000007);
956 switch (c
->x86_vendor
) {
960 case X86_VENDOR_INTEL
:
968 * This does the hard work of actually picking apart the CPU stuff...
970 void __cpuinit
identify_cpu(struct cpuinfo_x86
*c
)
974 early_identify_cpu(c
);
976 init_scattered_cpuid_features(c
);
978 c
->apicid
= phys_pkg_id(0);
981 * Vendor-specific initialization. In this section we
982 * canonicalize the feature flags, meaning if there are
983 * features a certain CPU supports which CPUID doesn't
984 * tell us, CPUID claiming incorrect flags, or other bugs,
985 * we handle them here.
987 * At the end of this section, c->x86_capability better
988 * indicate the features this CPU genuinely supports!
990 switch (c
->x86_vendor
) {
995 case X86_VENDOR_INTEL
:
999 case X86_VENDOR_UNKNOWN
:
1001 display_cacheinfo(c
);
1008 * On SMP, boot_cpu_data holds the common feature set between
1009 * all CPUs; so make sure that we indicate which features are
1010 * common between the CPUs. The first time this routine gets
1011 * executed, c == &boot_cpu_data.
1013 if (c
!= &boot_cpu_data
) {
1014 /* AND the already accumulated flags with these */
1015 for (i
= 0; i
< NCAPINTS
; i
++)
1016 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
1019 /* Clear all flags overriden by options */
1020 for (i
= 0; i
< NCAPINTS
; i
++)
1021 c
->x86_capability
[i
] ^= cleared_cpu_caps
[i
];
1023 #ifdef CONFIG_X86_MCE
1026 select_idle_routine(c
);
1028 if (c
!= &boot_cpu_data
)
1031 numa_add_cpu(smp_processor_id());
1036 static __init
int setup_noclflush(char *arg
)
1038 setup_clear_cpu_cap(X86_FEATURE_CLFLSH
);
1041 __setup("noclflush", setup_noclflush
);
1043 void __cpuinit
print_cpu_info(struct cpuinfo_x86
*c
)
1045 if (c
->x86_model_id
[0])
1046 printk(KERN_INFO
"%s", c
->x86_model_id
);
1048 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1049 printk(KERN_CONT
" stepping %02x\n", c
->x86_mask
);
1051 printk(KERN_CONT
"\n");
1054 static __init
int setup_disablecpuid(char *arg
)
1057 if (get_option(&arg
, &bit
) && bit
< NCAPINTS
*32)
1058 setup_clear_cpu_cap(bit
);
1063 __setup("clearcpuid=", setup_disablecpuid
);
1066 * Get CPU information for use by the procfs.
1069 static int show_cpuinfo(struct seq_file
*m
, void *v
)
1071 struct cpuinfo_x86
*c
= v
;
1075 * These flag bits must match the definitions in <asm/cpufeature.h>.
1076 * NULL means this bit is undefined or reserved; either way it doesn't
1077 * have meaning as far as Linux is concerned. Note that it's important
1078 * to realize there is a difference between this table and CPUID -- if
1079 * applications want to get the raw CPUID data, they should access
1080 * /dev/cpu/<cpu_nr>/cpuid instead.
1082 static const char *const x86_cap_flags
[] = {
1084 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1085 "cx8", "apic", NULL
, "sep", "mtrr", "pge", "mca", "cmov",
1086 "pat", "pse36", "pn", "clflush", NULL
, "dts", "acpi", "mmx",
1087 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", "pbe",
1090 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1091 NULL
, NULL
, NULL
, "syscall", NULL
, NULL
, NULL
, NULL
,
1092 NULL
, NULL
, NULL
, NULL
, "nx", NULL
, "mmxext", NULL
,
1093 NULL
, "fxsr_opt", "pdpe1gb", "rdtscp", NULL
, "lm",
1094 "3dnowext", "3dnow",
1096 /* Transmeta-defined */
1097 "recovery", "longrun", NULL
, "lrti", NULL
, NULL
, NULL
, NULL
,
1098 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1099 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1100 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1102 /* Other (Linux-defined) */
1103 "cxmmx", "k6_mtrr", "cyrix_arr", "centaur_mcr",
1104 NULL
, NULL
, NULL
, NULL
,
1105 "constant_tsc", "up", NULL
, "arch_perfmon",
1106 "pebs", "bts", NULL
, "sync_rdtsc",
1107 "rep_good", NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1108 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1110 /* Intel-defined (#2) */
1111 "pni", NULL
, NULL
, "monitor", "ds_cpl", "vmx", "smx", "est",
1112 "tm2", "ssse3", "cid", NULL
, NULL
, "cx16", "xtpr", NULL
,
1113 NULL
, NULL
, "dca", "sse4_1", "sse4_2", NULL
, NULL
, "popcnt",
1114 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1116 /* VIA/Cyrix/Centaur-defined */
1117 NULL
, NULL
, "rng", "rng_en", NULL
, NULL
, "ace", "ace_en",
1118 "ace2", "ace2_en", "phe", "phe_en", "pmm", "pmm_en", NULL
, NULL
,
1119 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1120 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1122 /* AMD-defined (#2) */
1123 "lahf_lm", "cmp_legacy", "svm", "extapic",
1124 "cr8_legacy", "abm", "sse4a", "misalignsse",
1125 "3dnowprefetch", "osvw", "ibs", "sse5",
1126 "skinit", "wdt", NULL
, NULL
,
1127 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1128 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1130 /* Auxiliary (Linux-defined) */
1131 "ida", NULL
, NULL
, NULL
, 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 static const char *const x86_power_flags
[] = {
1137 "ts", /* temperature sensor */
1138 "fid", /* frequency id control */
1139 "vid", /* voltage id control */
1140 "ttp", /* thermal trip */
1145 "", /* tsc invariant mapped to constant_tsc */
1154 seq_printf(m
, "processor\t: %u\n"
1156 "cpu family\t: %d\n"
1158 "model name\t: %s\n",
1160 c
->x86_vendor_id
[0] ? c
->x86_vendor_id
: "unknown",
1163 c
->x86_model_id
[0] ? c
->x86_model_id
: "unknown");
1165 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1166 seq_printf(m
, "stepping\t: %d\n", c
->x86_mask
);
1168 seq_printf(m
, "stepping\t: unknown\n");
1170 if (cpu_has(c
, X86_FEATURE_TSC
)) {
1171 unsigned int freq
= cpufreq_quick_get((unsigned)cpu
);
1175 seq_printf(m
, "cpu MHz\t\t: %u.%03u\n",
1176 freq
/ 1000, (freq
% 1000));
1180 if (c
->x86_cache_size
>= 0)
1181 seq_printf(m
, "cache size\t: %d KB\n", c
->x86_cache_size
);
1184 if (smp_num_siblings
* c
->x86_max_cores
> 1) {
1185 seq_printf(m
, "physical id\t: %d\n", c
->phys_proc_id
);
1186 seq_printf(m
, "siblings\t: %d\n",
1187 cpus_weight(per_cpu(cpu_core_map
, cpu
)));
1188 seq_printf(m
, "core id\t\t: %d\n", c
->cpu_core_id
);
1189 seq_printf(m
, "cpu cores\t: %d\n", c
->booted_cores
);
1195 "fpu_exception\t: yes\n"
1196 "cpuid level\t: %d\n"
1201 for (i
= 0; i
< 32*NCAPINTS
; i
++)
1202 if (cpu_has(c
, i
) && x86_cap_flags
[i
] != NULL
)
1203 seq_printf(m
, " %s", x86_cap_flags
[i
]);
1205 seq_printf(m
, "\nbogomips\t: %lu.%02lu\n",
1206 c
->loops_per_jiffy
/(500000/HZ
),
1207 (c
->loops_per_jiffy
/(5000/HZ
)) % 100);
1209 if (c
->x86_tlbsize
> 0)
1210 seq_printf(m
, "TLB size\t: %d 4K pages\n", c
->x86_tlbsize
);
1211 seq_printf(m
, "clflush size\t: %d\n", c
->x86_clflush_size
);
1212 seq_printf(m
, "cache_alignment\t: %d\n", c
->x86_cache_alignment
);
1214 seq_printf(m
, "address sizes\t: %u bits physical, %u bits virtual\n",
1215 c
->x86_phys_bits
, c
->x86_virt_bits
);
1217 seq_printf(m
, "power management:");
1218 for (i
= 0; i
< 32; i
++) {
1219 if (c
->x86_power
& (1 << i
)) {
1220 if (i
< ARRAY_SIZE(x86_power_flags
) &&
1222 seq_printf(m
, "%s%s",
1223 x86_power_flags
[i
][0]?" ":"",
1224 x86_power_flags
[i
]);
1226 seq_printf(m
, " [%d]", i
);
1230 seq_printf(m
, "\n\n");
1235 static void *c_start(struct seq_file
*m
, loff_t
*pos
)
1237 if (*pos
== 0) /* just in case, cpu 0 is not the first */
1238 *pos
= first_cpu(cpu_online_map
);
1239 if ((*pos
) < NR_CPUS
&& cpu_online(*pos
))
1240 return &cpu_data(*pos
);
1244 static void *c_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1246 *pos
= next_cpu(*pos
, cpu_online_map
);
1247 return c_start(m
, pos
);
1250 static void c_stop(struct seq_file
*m
, void *v
)
1254 const struct seq_operations cpuinfo_op
= {
1258 .show
= show_cpuinfo
,