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>
52 #include <video/edid.h>
55 #include <asm/mpspec.h>
56 #include <asm/mmu_context.h>
57 #include <asm/proto.h>
58 #include <asm/setup.h>
59 #include <asm/mach_apic.h>
61 #include <asm/sections.h>
63 #include <asm/cacheflush.h>
67 #ifdef CONFIG_PARAVIRT
68 #include <asm/paravirt.h>
77 struct cpuinfo_x86 boot_cpu_data __read_mostly
;
78 EXPORT_SYMBOL(boot_cpu_data
);
80 unsigned long mmu_cr4_features
;
82 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
85 unsigned long saved_video_mode
;
87 int force_mwait __cpuinitdata
;
93 char dmi_alloc_data
[DMI_MAX_DATA
];
98 struct screen_info screen_info
;
99 EXPORT_SYMBOL(screen_info
);
100 struct sys_desc_table_struct
{
101 unsigned short length
;
102 unsigned char table
[0];
105 struct edid_info edid_info
;
106 EXPORT_SYMBOL_GPL(edid_info
);
108 extern int root_mountflags
;
110 char __initdata command_line
[COMMAND_LINE_SIZE
];
112 struct resource standard_io_resources
[] = {
113 { .name
= "dma1", .start
= 0x00, .end
= 0x1f,
114 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
115 { .name
= "pic1", .start
= 0x20, .end
= 0x21,
116 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
117 { .name
= "timer0", .start
= 0x40, .end
= 0x43,
118 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
119 { .name
= "timer1", .start
= 0x50, .end
= 0x53,
120 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
121 { .name
= "keyboard", .start
= 0x60, .end
= 0x6f,
122 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
123 { .name
= "dma page reg", .start
= 0x80, .end
= 0x8f,
124 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
125 { .name
= "pic2", .start
= 0xa0, .end
= 0xa1,
126 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
127 { .name
= "dma2", .start
= 0xc0, .end
= 0xdf,
128 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
129 { .name
= "fpu", .start
= 0xf0, .end
= 0xff,
130 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
}
133 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
135 static struct resource data_resource
= {
136 .name
= "Kernel data",
139 .flags
= IORESOURCE_RAM
,
141 static struct resource code_resource
= {
142 .name
= "Kernel code",
145 .flags
= IORESOURCE_RAM
,
147 static struct resource bss_resource
= {
148 .name
= "Kernel bss",
151 .flags
= IORESOURCE_RAM
,
154 static void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
);
156 #ifdef CONFIG_PROC_VMCORE
157 /* elfcorehdr= specifies the location of elf core header
158 * stored by the crashed kernel. This option will be passed
159 * by kexec loader to the capture kernel.
161 static int __init
setup_elfcorehdr(char *arg
)
166 elfcorehdr_addr
= memparse(arg
, &end
);
167 return end
> arg
? 0 : -EINVAL
;
169 early_param("elfcorehdr", setup_elfcorehdr
);
174 contig_initmem_init(unsigned long start_pfn
, unsigned long end_pfn
)
176 unsigned long bootmap_size
, bootmap
;
178 bootmap_size
= bootmem_bootmap_pages(end_pfn
)<<PAGE_SHIFT
;
179 bootmap
= find_e820_area(0, end_pfn
<<PAGE_SHIFT
, bootmap_size
);
181 panic("Cannot find bootmem map of size %ld\n", bootmap_size
);
182 bootmap_size
= init_bootmem(bootmap
>> PAGE_SHIFT
, end_pfn
);
183 e820_register_active_regions(0, start_pfn
, end_pfn
);
184 free_bootmem_with_active_regions(0, end_pfn
);
185 reserve_bootmem(bootmap
, bootmap_size
);
189 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
191 #ifdef CONFIG_EDD_MODULE
195 * copy_edd() - Copy the BIOS EDD information
196 * from boot_params into a safe place.
199 static inline void copy_edd(void)
201 memcpy(edd
.mbr_signature
, boot_params
.edd_mbr_sig_buffer
,
202 sizeof(edd
.mbr_signature
));
203 memcpy(edd
.edd_info
, boot_params
.eddbuf
, sizeof(edd
.edd_info
));
204 edd
.mbr_signature_nr
= boot_params
.edd_mbr_sig_buf_entries
;
205 edd
.edd_info_nr
= boot_params
.eddbuf_entries
;
208 static inline void copy_edd(void)
214 static void __init
reserve_crashkernel(void)
216 unsigned long long free_mem
;
217 unsigned long long crash_size
, crash_base
;
221 ((unsigned long long)max_low_pfn
- min_low_pfn
) << PAGE_SHIFT
;
223 ret
= parse_crashkernel(boot_command_line
, free_mem
,
224 &crash_size
, &crash_base
);
225 if (ret
== 0 && crash_size
) {
226 if (crash_base
> 0) {
227 printk(KERN_INFO
"Reserving %ldMB of memory at %ldMB "
228 "for crashkernel (System RAM: %ldMB)\n",
229 (unsigned long)(crash_size
>> 20),
230 (unsigned long)(crash_base
>> 20),
231 (unsigned long)(free_mem
>> 20));
232 crashk_res
.start
= crash_base
;
233 crashk_res
.end
= crash_base
+ crash_size
- 1;
234 reserve_bootmem(crash_base
, crash_size
);
236 printk(KERN_INFO
"crashkernel reservation failed - "
237 "you have to specify a base address\n");
241 static inline void __init
reserve_crashkernel(void)
245 #define EBDA_ADDR_POINTER 0x40E
247 unsigned __initdata ebda_addr
;
248 unsigned __initdata ebda_size
;
250 static void discover_ebda(void)
253 * there is a real-mode segmented pointer pointing to the
254 * 4K EBDA area at 0x40E
256 ebda_addr
= *(unsigned short *)__va(EBDA_ADDR_POINTER
);
258 * There can be some situations, like paravirtualized guests,
259 * in which there is no available ebda information. In such
269 ebda_size
= *(unsigned short *)__va(ebda_addr
);
271 /* Round EBDA up to pages */
275 ebda_size
= round_up(ebda_size
+ (ebda_addr
& ~PAGE_MASK
), PAGE_SIZE
);
276 if (ebda_size
> 64*1024)
280 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
281 void __attribute__((weak
)) memory_setup(void)
283 machine_specific_memory_setup();
286 void __init
setup_arch(char **cmdline_p
)
290 printk(KERN_INFO
"Command line: %s\n", boot_command_line
);
292 ROOT_DEV
= old_decode_dev(boot_params
.hdr
.root_dev
);
293 screen_info
= boot_params
.screen_info
;
294 edid_info
= boot_params
.edid_info
;
295 saved_video_mode
= boot_params
.hdr
.vid_mode
;
296 bootloader_type
= boot_params
.hdr
.type_of_loader
;
298 #ifdef CONFIG_BLK_DEV_RAM
299 rd_image_start
= boot_params
.hdr
.ram_size
& RAMDISK_IMAGE_START_MASK
;
300 rd_prompt
= ((boot_params
.hdr
.ram_size
& RAMDISK_PROMPT_FLAG
) != 0);
301 rd_doload
= ((boot_params
.hdr
.ram_size
& RAMDISK_LOAD_FLAG
) != 0);
304 if (!strncmp((char *)&boot_params
.efi_info
.efi_loader_signature
,
314 if (!boot_params
.hdr
.root_flags
)
315 root_mountflags
&= ~MS_RDONLY
;
316 init_mm
.start_code
= (unsigned long) &_text
;
317 init_mm
.end_code
= (unsigned long) &_etext
;
318 init_mm
.end_data
= (unsigned long) &_edata
;
319 init_mm
.brk
= (unsigned long) &_end
;
321 code_resource
.start
= virt_to_phys(&_text
);
322 code_resource
.end
= virt_to_phys(&_etext
)-1;
323 data_resource
.start
= virt_to_phys(&_etext
);
324 data_resource
.end
= virt_to_phys(&_edata
)-1;
325 bss_resource
.start
= virt_to_phys(&__bss_start
);
326 bss_resource
.end
= virt_to_phys(&__bss_stop
)-1;
328 early_identify_cpu(&boot_cpu_data
);
330 strlcpy(command_line
, boot_command_line
, COMMAND_LINE_SIZE
);
331 *cmdline_p
= command_line
;
335 finish_e820_parsing();
337 e820_register_active_regions(0, 0, -1UL);
339 * partially used pages are not usable - thus
340 * we are rounding upwards:
342 end_pfn
= e820_end_of_ram();
343 num_physpages
= end_pfn
;
349 init_memory_mapping(0, (end_pfn_map
<< PAGE_SHIFT
));
358 /* setup to use the static apicid table during kernel startup */
359 x86_cpu_to_apicid_ptr
= (void *)&x86_cpu_to_apicid_init
;
364 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
365 * Call this early for SRAT node setup.
367 acpi_boot_table_init();
370 /* How many end-of-memory variables you have, grandma! */
371 max_low_pfn
= end_pfn
;
373 high_memory
= (void *)__va(end_pfn
* PAGE_SIZE
- 1) + 1;
375 /* Remove active ranges so rediscovery with NUMA-awareness happens */
376 remove_all_active_ranges();
378 #ifdef CONFIG_ACPI_NUMA
380 * Parse SRAT to discover nodes.
386 numa_initmem_init(0, end_pfn
);
388 contig_initmem_init(0, end_pfn
);
391 /* Reserve direct mapping */
392 reserve_bootmem_generic(table_start
<< PAGE_SHIFT
,
393 (table_end
- table_start
) << PAGE_SHIFT
);
396 reserve_bootmem_generic(__pa_symbol(&_text
),
397 __pa_symbol(&_end
) - __pa_symbol(&_text
));
400 * reserve physical page 0 - it's a special BIOS page on many boxes,
401 * enabling clean reboots, SMP operation, laptop functions.
403 reserve_bootmem_generic(0, PAGE_SIZE
);
405 /* reserve ebda region */
407 reserve_bootmem_generic(ebda_addr
, ebda_size
);
409 /* reserve nodemap region */
411 reserve_bootmem_generic(nodemap_addr
, nodemap_size
);
415 /* Reserve SMP trampoline */
416 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE
, 2*PAGE_SIZE
);
419 #ifdef CONFIG_ACPI_SLEEP
421 * Reserve low memory region for sleep support.
423 acpi_reserve_bootmem();
428 efi_reserve_bootmem();
432 * Find and reserve possible boot-time SMP configuration:
435 #ifdef CONFIG_BLK_DEV_INITRD
436 if (boot_params
.hdr
.type_of_loader
&& boot_params
.hdr
.ramdisk_image
) {
437 unsigned long ramdisk_image
= boot_params
.hdr
.ramdisk_image
;
438 unsigned long ramdisk_size
= boot_params
.hdr
.ramdisk_size
;
439 unsigned long ramdisk_end
= ramdisk_image
+ ramdisk_size
;
440 unsigned long end_of_mem
= end_pfn
<< PAGE_SHIFT
;
442 if (ramdisk_end
<= end_of_mem
) {
443 reserve_bootmem_generic(ramdisk_image
, ramdisk_size
);
444 initrd_start
= ramdisk_image
+ PAGE_OFFSET
;
445 initrd_end
= initrd_start
+ramdisk_size
;
447 printk(KERN_ERR
"initrd extends beyond end of memory "
448 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
449 ramdisk_end
, end_of_mem
);
454 reserve_crashkernel();
460 * set this early, so we dont allocate cpu0
461 * if MADT list doesnt list BSP first
462 * mpparse.c/MP_processor_info() allocates logical cpu numbers.
464 cpu_set(0, cpu_present_map
);
467 * Read APIC and some other early information from ACPI tables.
475 * get boot-time SMP configuration:
477 if (smp_found_config
)
479 init_apic_mappings();
480 ioapic_init_mappings();
483 * We trust e820 completely. No explicit ROM probing in memory.
485 e820_reserve_resources(&code_resource
, &data_resource
, &bss_resource
);
486 e820_mark_nosave_regions();
488 /* request I/O space for devices used on all i[345]86 PCs */
489 for (i
= 0; i
< ARRAY_SIZE(standard_io_resources
); i
++)
490 request_resource(&ioport_resource
, &standard_io_resources
[i
]);
495 #if defined(CONFIG_VGA_CONSOLE)
496 if (!efi_enabled
|| (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY
))
497 conswitchp
= &vga_con
;
498 #elif defined(CONFIG_DUMMY_CONSOLE)
499 conswitchp
= &dummy_con
;
504 static int __cpuinit
get_model_name(struct cpuinfo_x86
*c
)
508 if (c
->extended_cpuid_level
< 0x80000004)
511 v
= (unsigned int *) c
->x86_model_id
;
512 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
513 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
514 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
515 c
->x86_model_id
[48] = 0;
520 static void __cpuinit
display_cacheinfo(struct cpuinfo_x86
*c
)
522 unsigned int n
, dummy
, eax
, ebx
, ecx
, edx
;
524 n
= c
->extended_cpuid_level
;
526 if (n
>= 0x80000005) {
527 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
528 printk(KERN_INFO
"CPU: L1 I Cache: %dK (%d bytes/line), "
529 "D cache %dK (%d bytes/line)\n",
530 edx
>>24, edx
&0xFF, ecx
>>24, ecx
&0xFF);
531 c
->x86_cache_size
= (ecx
>>24) + (edx
>>24);
532 /* On K8 L1 TLB is inclusive, so don't count it */
536 if (n
>= 0x80000006) {
537 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
538 ecx
= cpuid_ecx(0x80000006);
539 c
->x86_cache_size
= ecx
>> 16;
540 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
542 printk(KERN_INFO
"CPU: L2 Cache: %dK (%d bytes/line)\n",
543 c
->x86_cache_size
, ecx
& 0xFF);
547 cpuid(0x80000007, &dummy
, &dummy
, &dummy
, &c
->x86_power
);
548 if (n
>= 0x80000008) {
549 cpuid(0x80000008, &eax
, &dummy
, &dummy
, &dummy
);
550 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
551 c
->x86_phys_bits
= eax
& 0xff;
556 static int nearby_node(int apicid
)
560 for (i
= apicid
- 1; i
>= 0; i
--) {
561 node
= apicid_to_node
[i
];
562 if (node
!= NUMA_NO_NODE
&& node_online(node
))
565 for (i
= apicid
+ 1; i
< MAX_LOCAL_APIC
; i
++) {
566 node
= apicid_to_node
[i
];
567 if (node
!= NUMA_NO_NODE
&& node_online(node
))
570 return first_node(node_online_map
); /* Shouldn't happen */
575 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
576 * Assumes number of cores is a power of two.
578 static void __init
amd_detect_cmp(struct cpuinfo_x86
*c
)
583 int cpu
= smp_processor_id();
585 unsigned apicid
= hard_smp_processor_id();
587 bits
= c
->x86_coreid_bits
;
589 /* Low order bits define the core id (index of core in socket) */
590 c
->cpu_core_id
= c
->phys_proc_id
& ((1 << bits
)-1);
591 /* Convert the APIC ID into the socket ID */
592 c
->phys_proc_id
= phys_pkg_id(bits
);
595 node
= c
->phys_proc_id
;
596 if (apicid_to_node
[apicid
] != NUMA_NO_NODE
)
597 node
= apicid_to_node
[apicid
];
598 if (!node_online(node
)) {
599 /* Two possibilities here:
600 - The CPU is missing memory and no node was created.
601 In that case try picking one from a nearby CPU
602 - The APIC IDs differ from the HyperTransport node IDs
603 which the K8 northbridge parsing fills in.
604 Assume they are all increased by a constant offset,
605 but in the same order as the HT nodeids.
606 If that doesn't result in a usable node fall back to the
607 path for the previous case. */
609 int ht_nodeid
= apicid
- (cpu_data(0).phys_proc_id
<< bits
);
611 if (ht_nodeid
>= 0 &&
612 apicid_to_node
[ht_nodeid
] != NUMA_NO_NODE
)
613 node
= apicid_to_node
[ht_nodeid
];
614 /* Pick a nearby node */
615 if (!node_online(node
))
616 node
= nearby_node(apicid
);
618 numa_set_node(cpu
, node
);
620 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
625 static void __cpuinit
early_init_amd(struct cpuinfo_x86
*c
)
630 /* Multi core CPU? */
631 if (c
->extended_cpuid_level
< 0x80000008)
634 ecx
= cpuid_ecx(0x80000008);
636 c
->x86_max_cores
= (ecx
& 0xff) + 1;
638 /* CPU telling us the core id bits shift? */
639 bits
= (ecx
>> 12) & 0xF;
641 /* Otherwise recompute */
643 while ((1 << bits
) < c
->x86_max_cores
)
647 c
->x86_coreid_bits
= bits
;
652 #define ENABLE_C1E_MASK 0x18000000
653 #define CPUID_PROCESSOR_SIGNATURE 1
654 #define CPUID_XFAM 0x0ff00000
655 #define CPUID_XFAM_K8 0x00000000
656 #define CPUID_XFAM_10H 0x00100000
657 #define CPUID_XFAM_11H 0x00200000
658 #define CPUID_XMOD 0x000f0000
659 #define CPUID_XMOD_REV_F 0x00040000
661 /* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
662 static __cpuinit
int amd_apic_timer_broken(void)
664 u32 lo
, hi
, eax
= cpuid_eax(CPUID_PROCESSOR_SIGNATURE
);
666 switch (eax
& CPUID_XFAM
) {
668 if ((eax
& CPUID_XMOD
) < CPUID_XMOD_REV_F
)
672 rdmsr(MSR_K8_ENABLE_C1E
, lo
, hi
);
673 if (lo
& ENABLE_C1E_MASK
)
677 /* err on the side of caution */
683 static void __cpuinit
init_amd(struct cpuinfo_x86
*c
)
691 * Disable TLB flush filter by setting HWCR.FFDIS on K8
692 * bit 6 of msr C001_0015
694 * Errata 63 for SH-B3 steppings
695 * Errata 122 for all steppings (F+ have it disabled by default)
698 rdmsrl(MSR_K8_HWCR
, value
);
700 wrmsrl(MSR_K8_HWCR
, value
);
704 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
705 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
706 clear_bit(0*32+31, (unsigned long *)&c
->x86_capability
);
708 /* On C+ stepping K8 rep microcode works well for copy/memset */
709 level
= cpuid_eax(1);
710 if (c
->x86
== 15 && ((level
>= 0x0f48 && level
< 0x0f50) ||
712 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
713 if (c
->x86
== 0x10 || c
->x86
== 0x11)
714 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
716 /* Enable workaround for FXSAVE leak */
718 set_cpu_cap(c
, X86_FEATURE_FXSAVE_LEAK
);
720 level
= get_model_name(c
);
724 /* Should distinguish Models here, but this is only
725 a fallback anyways. */
726 strcpy(c
->x86_model_id
, "Hammer");
730 display_cacheinfo(c
);
732 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
733 if (c
->x86_power
& (1<<8))
734 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
736 /* Multi core CPU? */
737 if (c
->extended_cpuid_level
>= 0x80000008)
740 if (c
->extended_cpuid_level
>= 0x80000006 &&
741 (cpuid_edx(0x80000006) & 0xf000))
742 num_cache_leaves
= 4;
744 num_cache_leaves
= 3;
746 if (c
->x86
== 0xf || c
->x86
== 0x10 || c
->x86
== 0x11)
747 set_cpu_cap(c
, X86_FEATURE_K8
);
749 /* RDTSC can be speculated around */
750 clear_cpu_cap(c
, X86_FEATURE_SYNC_RDTSC
);
752 /* Family 10 doesn't support C states in MWAIT so don't use it */
753 if (c
->x86
== 0x10 && !force_mwait
)
754 clear_cpu_cap(c
, X86_FEATURE_MWAIT
);
756 if (amd_apic_timer_broken())
757 disable_apic_timer
= 1;
760 void __cpuinit
detect_ht(struct cpuinfo_x86
*c
)
763 u32 eax
, ebx
, ecx
, edx
;
764 int index_msb
, core_bits
;
766 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
769 if (!cpu_has(c
, X86_FEATURE_HT
))
771 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
774 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
776 if (smp_num_siblings
== 1) {
777 printk(KERN_INFO
"CPU: Hyper-Threading is disabled\n");
778 } else if (smp_num_siblings
> 1) {
780 if (smp_num_siblings
> NR_CPUS
) {
781 printk(KERN_WARNING
"CPU: Unsupported number of "
782 "siblings %d", smp_num_siblings
);
783 smp_num_siblings
= 1;
787 index_msb
= get_count_order(smp_num_siblings
);
788 c
->phys_proc_id
= phys_pkg_id(index_msb
);
790 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
792 index_msb
= get_count_order(smp_num_siblings
);
794 core_bits
= get_count_order(c
->x86_max_cores
);
796 c
->cpu_core_id
= phys_pkg_id(index_msb
) &
797 ((1 << core_bits
) - 1);
800 if ((c
->x86_max_cores
* smp_num_siblings
) > 1) {
801 printk(KERN_INFO
"CPU: Physical Processor ID: %d\n",
803 printk(KERN_INFO
"CPU: Processor Core ID: %d\n",
811 * find out the number of processor cores on the die
813 static int __cpuinit
intel_num_cpu_cores(struct cpuinfo_x86
*c
)
817 if (c
->cpuid_level
< 4)
820 cpuid_count(4, 0, &eax
, &t
, &t
, &t
);
823 return ((eax
>> 26) + 1);
828 static void srat_detect_node(void)
832 int cpu
= smp_processor_id();
833 int apicid
= hard_smp_processor_id();
835 /* Don't do the funky fallback heuristics the AMD version employs
837 node
= apicid_to_node
[apicid
];
838 if (node
== NUMA_NO_NODE
)
839 node
= first_node(node_online_map
);
840 numa_set_node(cpu
, node
);
842 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
846 static void __cpuinit
init_intel(struct cpuinfo_x86
*c
)
851 init_intel_cacheinfo(c
);
852 if (c
->cpuid_level
> 9) {
853 unsigned eax
= cpuid_eax(10);
854 /* Check for version and the number of counters */
855 if ((eax
& 0xff) && (((eax
>>8) & 0xff) > 1))
856 set_cpu_cap(c
, X86_FEATURE_ARCH_PERFMON
);
861 rdmsr(MSR_IA32_MISC_ENABLE
, l1
, l2
);
863 set_cpu_cap(c
, X86_FEATURE_BTS
);
865 set_cpu_cap(c
, X86_FEATURE_PEBS
);
872 n
= c
->extended_cpuid_level
;
873 if (n
>= 0x80000008) {
874 unsigned eax
= cpuid_eax(0x80000008);
875 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
876 c
->x86_phys_bits
= eax
& 0xff;
877 /* CPUID workaround for Intel 0F34 CPU */
878 if (c
->x86_vendor
== X86_VENDOR_INTEL
&&
879 c
->x86
== 0xF && c
->x86_model
== 0x3 &&
881 c
->x86_phys_bits
= 36;
885 c
->x86_cache_alignment
= c
->x86_clflush_size
* 2;
886 if ((c
->x86
== 0xf && c
->x86_model
>= 0x03) ||
887 (c
->x86
== 0x6 && c
->x86_model
>= 0x0e))
888 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
890 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
892 set_cpu_cap(c
, X86_FEATURE_SYNC_RDTSC
);
894 clear_cpu_cap(c
, X86_FEATURE_SYNC_RDTSC
);
895 c
->x86_max_cores
= intel_num_cpu_cores(c
);
900 static void __cpuinit
get_cpu_vendor(struct cpuinfo_x86
*c
)
902 char *v
= c
->x86_vendor_id
;
904 if (!strcmp(v
, "AuthenticAMD"))
905 c
->x86_vendor
= X86_VENDOR_AMD
;
906 else if (!strcmp(v
, "GenuineIntel"))
907 c
->x86_vendor
= X86_VENDOR_INTEL
;
909 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
912 struct cpu_model_info
{
915 char *model_names
[16];
918 /* Do some early cpuid on the boot CPU to get some parameter that are
919 needed before check_bugs. Everything advanced is in identify_cpu
921 static void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
)
925 c
->loops_per_jiffy
= loops_per_jiffy
;
926 c
->x86_cache_size
= -1;
927 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
928 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
929 c
->x86_vendor_id
[0] = '\0'; /* Unset */
930 c
->x86_model_id
[0] = '\0'; /* Unset */
931 c
->x86_clflush_size
= 64;
932 c
->x86_cache_alignment
= c
->x86_clflush_size
;
933 c
->x86_max_cores
= 1;
934 c
->x86_coreid_bits
= 0;
935 c
->extended_cpuid_level
= 0;
936 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
938 /* Get vendor name */
939 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
940 (unsigned int *)&c
->x86_vendor_id
[0],
941 (unsigned int *)&c
->x86_vendor_id
[8],
942 (unsigned int *)&c
->x86_vendor_id
[4]);
946 /* Initialize the standard set of capabilities */
947 /* Note that the vendor-specific code below might override */
949 /* Intel-defined flags: level 0x00000001 */
950 if (c
->cpuid_level
>= 0x00000001) {
952 cpuid(0x00000001, &tfms
, &misc
, &c
->x86_capability
[4],
953 &c
->x86_capability
[0]);
954 c
->x86
= (tfms
>> 8) & 0xf;
955 c
->x86_model
= (tfms
>> 4) & 0xf;
956 c
->x86_mask
= tfms
& 0xf;
958 c
->x86
+= (tfms
>> 20) & 0xff;
960 c
->x86_model
+= ((tfms
>> 16) & 0xF) << 4;
961 if (c
->x86_capability
[0] & (1<<19))
962 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
964 /* Have CPUID level 0 only - unheard of */
969 c
->phys_proc_id
= (cpuid_ebx(1) >> 24) & 0xff;
971 /* AMD-defined flags: level 0x80000001 */
972 xlvl
= cpuid_eax(0x80000000);
973 c
->extended_cpuid_level
= xlvl
;
974 if ((xlvl
& 0xffff0000) == 0x80000000) {
975 if (xlvl
>= 0x80000001) {
976 c
->x86_capability
[1] = cpuid_edx(0x80000001);
977 c
->x86_capability
[6] = cpuid_ecx(0x80000001);
979 if (xlvl
>= 0x80000004)
980 get_model_name(c
); /* Default name */
983 /* Transmeta-defined flags: level 0x80860001 */
984 xlvl
= cpuid_eax(0x80860000);
985 if ((xlvl
& 0xffff0000) == 0x80860000) {
986 /* Don't set x86_cpuid_level here for now to not confuse. */
987 if (xlvl
>= 0x80860001)
988 c
->x86_capability
[2] = cpuid_edx(0x80860001);
991 switch (c
->x86_vendor
) {
1000 * This does the hard work of actually picking apart the CPU stuff...
1002 void __cpuinit
identify_cpu(struct cpuinfo_x86
*c
)
1006 early_identify_cpu(c
);
1008 init_scattered_cpuid_features(c
);
1010 c
->apicid
= phys_pkg_id(0);
1013 * Vendor-specific initialization. In this section we
1014 * canonicalize the feature flags, meaning if there are
1015 * features a certain CPU supports which CPUID doesn't
1016 * tell us, CPUID claiming incorrect flags, or other bugs,
1017 * we handle them here.
1019 * At the end of this section, c->x86_capability better
1020 * indicate the features this CPU genuinely supports!
1022 switch (c
->x86_vendor
) {
1023 case X86_VENDOR_AMD
:
1027 case X86_VENDOR_INTEL
:
1031 case X86_VENDOR_UNKNOWN
:
1033 display_cacheinfo(c
);
1037 select_idle_routine(c
);
1041 * On SMP, boot_cpu_data holds the common feature set between
1042 * all CPUs; so make sure that we indicate which features are
1043 * common between the CPUs. The first time this routine gets
1044 * executed, c == &boot_cpu_data.
1046 if (c
!= &boot_cpu_data
) {
1047 /* AND the already accumulated flags with these */
1048 for (i
= 0; i
< NCAPINTS
; i
++)
1049 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
1052 #ifdef CONFIG_X86_MCE
1055 if (c
!= &boot_cpu_data
)
1058 numa_add_cpu(smp_processor_id());
1062 void __cpuinit
print_cpu_info(struct cpuinfo_x86
*c
)
1064 if (c
->x86_model_id
[0])
1065 printk(KERN_INFO
"%s", c
->x86_model_id
);
1067 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1068 printk(KERN_CONT
" stepping %02x\n", c
->x86_mask
);
1070 printk(KERN_CONT
"\n");
1074 * Get CPU information for use by the procfs.
1077 static int show_cpuinfo(struct seq_file
*m
, void *v
)
1079 struct cpuinfo_x86
*c
= v
;
1083 * These flag bits must match the definitions in <asm/cpufeature.h>.
1084 * NULL means this bit is undefined or reserved; either way it doesn't
1085 * have meaning as far as Linux is concerned. Note that it's important
1086 * to realize there is a difference between this table and CPUID -- if
1087 * applications want to get the raw CPUID data, they should access
1088 * /dev/cpu/<cpu_nr>/cpuid instead.
1090 static const char *const x86_cap_flags
[] = {
1092 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1093 "cx8", "apic", NULL
, "sep", "mtrr", "pge", "mca", "cmov",
1094 "pat", "pse36", "pn", "clflush", NULL
, "dts", "acpi", "mmx",
1095 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", "pbe",
1098 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1099 NULL
, NULL
, NULL
, "syscall", NULL
, NULL
, NULL
, NULL
,
1100 NULL
, NULL
, NULL
, NULL
, "nx", NULL
, "mmxext", NULL
,
1101 NULL
, "fxsr_opt", "pdpe1gb", "rdtscp", NULL
, "lm",
1102 "3dnowext", "3dnow",
1104 /* Transmeta-defined */
1105 "recovery", "longrun", NULL
, "lrti", NULL
, NULL
, NULL
, NULL
,
1106 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1107 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1108 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1110 /* Other (Linux-defined) */
1111 "cxmmx", "k6_mtrr", "cyrix_arr", "centaur_mcr",
1112 NULL
, NULL
, NULL
, NULL
,
1113 "constant_tsc", "up", NULL
, "arch_perfmon",
1114 "pebs", "bts", NULL
, "sync_rdtsc",
1115 "rep_good", NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1116 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1118 /* Intel-defined (#2) */
1119 "pni", NULL
, NULL
, "monitor", "ds_cpl", "vmx", "smx", "est",
1120 "tm2", "ssse3", "cid", NULL
, NULL
, "cx16", "xtpr", NULL
,
1121 NULL
, NULL
, "dca", "sse4_1", "sse4_2", NULL
, NULL
, "popcnt",
1122 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1124 /* VIA/Cyrix/Centaur-defined */
1125 NULL
, NULL
, "rng", "rng_en", NULL
, NULL
, "ace", "ace_en",
1126 "ace2", "ace2_en", "phe", "phe_en", "pmm", "pmm_en", NULL
, NULL
,
1127 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1128 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1130 /* AMD-defined (#2) */
1131 "lahf_lm", "cmp_legacy", "svm", "extapic",
1132 "cr8_legacy", "abm", "sse4a", "misalignsse",
1133 "3dnowprefetch", "osvw", "ibs", "sse5",
1134 "skinit", "wdt", NULL
, NULL
,
1135 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1136 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1138 /* Auxiliary (Linux-defined) */
1139 "ida", NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1140 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1141 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1142 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1144 static const char *const x86_power_flags
[] = {
1145 "ts", /* temperature sensor */
1146 "fid", /* frequency id control */
1147 "vid", /* voltage id control */
1148 "ttp", /* thermal trip */
1153 "", /* tsc invariant mapped to constant_tsc */
1162 seq_printf(m
, "processor\t: %u\n"
1164 "cpu family\t: %d\n"
1166 "model name\t: %s\n",
1168 c
->x86_vendor_id
[0] ? c
->x86_vendor_id
: "unknown",
1171 c
->x86_model_id
[0] ? c
->x86_model_id
: "unknown");
1173 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1174 seq_printf(m
, "stepping\t: %d\n", c
->x86_mask
);
1176 seq_printf(m
, "stepping\t: unknown\n");
1178 if (cpu_has(c
, X86_FEATURE_TSC
)) {
1179 unsigned int freq
= cpufreq_quick_get((unsigned)cpu
);
1183 seq_printf(m
, "cpu MHz\t\t: %u.%03u\n",
1184 freq
/ 1000, (freq
% 1000));
1188 if (c
->x86_cache_size
>= 0)
1189 seq_printf(m
, "cache size\t: %d KB\n", c
->x86_cache_size
);
1192 if (smp_num_siblings
* c
->x86_max_cores
> 1) {
1193 seq_printf(m
, "physical id\t: %d\n", c
->phys_proc_id
);
1194 seq_printf(m
, "siblings\t: %d\n",
1195 cpus_weight(per_cpu(cpu_core_map
, cpu
)));
1196 seq_printf(m
, "core id\t\t: %d\n", c
->cpu_core_id
);
1197 seq_printf(m
, "cpu cores\t: %d\n", c
->booted_cores
);
1203 "fpu_exception\t: yes\n"
1204 "cpuid level\t: %d\n"
1209 for (i
= 0; i
< 32*NCAPINTS
; i
++)
1210 if (cpu_has(c
, i
) && x86_cap_flags
[i
] != NULL
)
1211 seq_printf(m
, " %s", x86_cap_flags
[i
]);
1213 seq_printf(m
, "\nbogomips\t: %lu.%02lu\n",
1214 c
->loops_per_jiffy
/(500000/HZ
),
1215 (c
->loops_per_jiffy
/(5000/HZ
)) % 100);
1217 if (c
->x86_tlbsize
> 0)
1218 seq_printf(m
, "TLB size\t: %d 4K pages\n", c
->x86_tlbsize
);
1219 seq_printf(m
, "clflush size\t: %d\n", c
->x86_clflush_size
);
1220 seq_printf(m
, "cache_alignment\t: %d\n", c
->x86_cache_alignment
);
1222 seq_printf(m
, "address sizes\t: %u bits physical, %u bits virtual\n",
1223 c
->x86_phys_bits
, c
->x86_virt_bits
);
1225 seq_printf(m
, "power management:");
1226 for (i
= 0; i
< 32; i
++) {
1227 if (c
->x86_power
& (1 << i
)) {
1228 if (i
< ARRAY_SIZE(x86_power_flags
) &&
1230 seq_printf(m
, "%s%s",
1231 x86_power_flags
[i
][0]?" ":"",
1232 x86_power_flags
[i
]);
1234 seq_printf(m
, " [%d]", i
);
1238 seq_printf(m
, "\n\n");
1243 static void *c_start(struct seq_file
*m
, loff_t
*pos
)
1245 if (*pos
== 0) /* just in case, cpu 0 is not the first */
1246 *pos
= first_cpu(cpu_online_map
);
1247 if ((*pos
) < NR_CPUS
&& cpu_online(*pos
))
1248 return &cpu_data(*pos
);
1252 static void *c_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1254 *pos
= next_cpu(*pos
, cpu_online_map
);
1255 return c_start(m
, pos
);
1258 static void c_stop(struct seq_file
*m
, void *v
)
1262 struct seq_operations cpuinfo_op
= {
1266 .show
= show_cpuinfo
,