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/acpi.h>
34 #include <linux/kallsyms.h>
35 #include <linux/edd.h>
36 #include <linux/mmzone.h>
37 #include <linux/kexec.h>
38 #include <linux/cpufreq.h>
39 #include <linux/dmi.h>
40 #include <linux/dma-mapping.h>
41 #include <linux/ctype.h>
44 #include <asm/uaccess.h>
45 #include <asm/system.h>
50 #include <video/edid.h>
53 #include <asm/mpspec.h>
54 #include <asm/mmu_context.h>
55 #include <asm/proto.h>
56 #include <asm/setup.h>
57 #include <asm/mach_apic.h>
59 #include <asm/sections.h>
61 #include <asm/cacheflush.h>
68 struct cpuinfo_x86 boot_cpu_data __read_mostly
;
69 EXPORT_SYMBOL(boot_cpu_data
);
71 unsigned long mmu_cr4_features
;
73 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
76 unsigned long saved_video_mode
;
78 int force_mwait __cpuinitdata
;
84 char dmi_alloc_data
[DMI_MAX_DATA
];
89 struct screen_info screen_info
;
90 EXPORT_SYMBOL(screen_info
);
91 struct sys_desc_table_struct
{
92 unsigned short length
;
93 unsigned char table
[0];
96 struct edid_info edid_info
;
97 EXPORT_SYMBOL_GPL(edid_info
);
99 extern int root_mountflags
;
101 char __initdata command_line
[COMMAND_LINE_SIZE
];
103 struct resource standard_io_resources
[] = {
104 { .name
= "dma1", .start
= 0x00, .end
= 0x1f,
105 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
106 { .name
= "pic1", .start
= 0x20, .end
= 0x21,
107 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
108 { .name
= "timer0", .start
= 0x40, .end
= 0x43,
109 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
110 { .name
= "timer1", .start
= 0x50, .end
= 0x53,
111 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
112 { .name
= "keyboard", .start
= 0x60, .end
= 0x6f,
113 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
114 { .name
= "dma page reg", .start
= 0x80, .end
= 0x8f,
115 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
116 { .name
= "pic2", .start
= 0xa0, .end
= 0xa1,
117 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
118 { .name
= "dma2", .start
= 0xc0, .end
= 0xdf,
119 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
120 { .name
= "fpu", .start
= 0xf0, .end
= 0xff,
121 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
}
124 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
126 static struct resource data_resource
= {
127 .name
= "Kernel data",
130 .flags
= IORESOURCE_RAM
,
132 static struct resource code_resource
= {
133 .name
= "Kernel code",
136 .flags
= IORESOURCE_RAM
,
138 static struct resource bss_resource
= {
139 .name
= "Kernel bss",
142 .flags
= IORESOURCE_RAM
,
145 static void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
);
147 #ifdef CONFIG_PROC_VMCORE
148 /* elfcorehdr= specifies the location of elf core header
149 * stored by the crashed kernel. This option will be passed
150 * by kexec loader to the capture kernel.
152 static int __init
setup_elfcorehdr(char *arg
)
157 elfcorehdr_addr
= memparse(arg
, &end
);
158 return end
> arg
? 0 : -EINVAL
;
160 early_param("elfcorehdr", setup_elfcorehdr
);
165 contig_initmem_init(unsigned long start_pfn
, unsigned long end_pfn
)
167 unsigned long bootmap_size
, bootmap
;
169 bootmap_size
= bootmem_bootmap_pages(end_pfn
)<<PAGE_SHIFT
;
170 bootmap
= find_e820_area(0, end_pfn
<<PAGE_SHIFT
, bootmap_size
);
172 panic("Cannot find bootmem map of size %ld\n", bootmap_size
);
173 bootmap_size
= init_bootmem(bootmap
>> PAGE_SHIFT
, end_pfn
);
174 e820_register_active_regions(0, start_pfn
, end_pfn
);
175 free_bootmem_with_active_regions(0, end_pfn
);
176 reserve_bootmem(bootmap
, bootmap_size
);
180 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
182 #ifdef CONFIG_EDD_MODULE
186 * copy_edd() - Copy the BIOS EDD information
187 * from boot_params into a safe place.
190 static inline void copy_edd(void)
192 memcpy(edd
.mbr_signature
, boot_params
.edd_mbr_sig_buffer
,
193 sizeof(edd
.mbr_signature
));
194 memcpy(edd
.edd_info
, boot_params
.eddbuf
, sizeof(edd
.edd_info
));
195 edd
.mbr_signature_nr
= boot_params
.edd_mbr_sig_buf_entries
;
196 edd
.edd_info_nr
= boot_params
.eddbuf_entries
;
199 static inline void copy_edd(void)
205 static void __init
reserve_crashkernel(void)
207 unsigned long long free_mem
;
208 unsigned long long crash_size
, crash_base
;
212 ((unsigned long long)max_low_pfn
- min_low_pfn
) << PAGE_SHIFT
;
214 ret
= parse_crashkernel(boot_command_line
, free_mem
,
215 &crash_size
, &crash_base
);
216 if (ret
== 0 && crash_size
) {
217 if (crash_base
> 0) {
218 printk(KERN_INFO
"Reserving %ldMB of memory at %ldMB "
219 "for crashkernel (System RAM: %ldMB)\n",
220 (unsigned long)(crash_size
>> 20),
221 (unsigned long)(crash_base
>> 20),
222 (unsigned long)(free_mem
>> 20));
223 crashk_res
.start
= crash_base
;
224 crashk_res
.end
= crash_base
+ crash_size
- 1;
225 reserve_bootmem(crash_base
, crash_size
);
227 printk(KERN_INFO
"crashkernel reservation failed - "
228 "you have to specify a base address\n");
232 static inline void __init
reserve_crashkernel(void)
236 #define EBDA_ADDR_POINTER 0x40E
238 unsigned __initdata ebda_addr
;
239 unsigned __initdata ebda_size
;
241 static void discover_ebda(void)
244 * there is a real-mode segmented pointer pointing to the
245 * 4K EBDA area at 0x40E
247 ebda_addr
= *(unsigned short *)__va(EBDA_ADDR_POINTER
);
250 ebda_size
= *(unsigned short *)__va(ebda_addr
);
252 /* Round EBDA up to pages */
256 ebda_size
= round_up(ebda_size
+ (ebda_addr
& ~PAGE_MASK
), PAGE_SIZE
);
257 if (ebda_size
> 64*1024)
261 void __init
setup_arch(char **cmdline_p
)
265 printk(KERN_INFO
"Command line: %s\n", boot_command_line
);
267 ROOT_DEV
= old_decode_dev(boot_params
.hdr
.root_dev
);
268 screen_info
= boot_params
.screen_info
;
269 edid_info
= boot_params
.edid_info
;
270 saved_video_mode
= boot_params
.hdr
.vid_mode
;
271 bootloader_type
= boot_params
.hdr
.type_of_loader
;
273 #ifdef CONFIG_BLK_DEV_RAM
274 rd_image_start
= boot_params
.hdr
.ram_size
& RAMDISK_IMAGE_START_MASK
;
275 rd_prompt
= ((boot_params
.hdr
.ram_size
& RAMDISK_PROMPT_FLAG
) != 0);
276 rd_doload
= ((boot_params
.hdr
.ram_size
& RAMDISK_LOAD_FLAG
) != 0);
278 setup_memory_region();
281 if (!boot_params
.hdr
.root_flags
)
282 root_mountflags
&= ~MS_RDONLY
;
283 init_mm
.start_code
= (unsigned long) &_text
;
284 init_mm
.end_code
= (unsigned long) &_etext
;
285 init_mm
.end_data
= (unsigned long) &_edata
;
286 init_mm
.brk
= (unsigned long) &_end
;
288 code_resource
.start
= virt_to_phys(&_text
);
289 code_resource
.end
= virt_to_phys(&_etext
)-1;
290 data_resource
.start
= virt_to_phys(&_etext
);
291 data_resource
.end
= virt_to_phys(&_edata
)-1;
292 bss_resource
.start
= virt_to_phys(&__bss_start
);
293 bss_resource
.end
= virt_to_phys(&__bss_stop
)-1;
295 early_identify_cpu(&boot_cpu_data
);
297 strlcpy(command_line
, boot_command_line
, COMMAND_LINE_SIZE
);
298 *cmdline_p
= command_line
;
302 finish_e820_parsing();
304 e820_register_active_regions(0, 0, -1UL);
306 * partially used pages are not usable - thus
307 * we are rounding upwards:
309 end_pfn
= e820_end_of_ram();
310 num_physpages
= end_pfn
;
316 init_memory_mapping(0, (end_pfn_map
<< PAGE_SHIFT
));
323 /* setup to use the static apicid table during kernel startup */
324 x86_cpu_to_apicid_ptr
= (void *)&x86_cpu_to_apicid_init
;
329 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
330 * Call this early for SRAT node setup.
332 acpi_boot_table_init();
335 /* How many end-of-memory variables you have, grandma! */
336 max_low_pfn
= end_pfn
;
338 high_memory
= (void *)__va(end_pfn
* PAGE_SIZE
- 1) + 1;
340 /* Remove active ranges so rediscovery with NUMA-awareness happens */
341 remove_all_active_ranges();
343 #ifdef CONFIG_ACPI_NUMA
345 * Parse SRAT to discover nodes.
351 numa_initmem_init(0, end_pfn
);
353 contig_initmem_init(0, end_pfn
);
356 /* Reserve direct mapping */
357 reserve_bootmem_generic(table_start
<< PAGE_SHIFT
,
358 (table_end
- table_start
) << PAGE_SHIFT
);
361 reserve_bootmem_generic(__pa_symbol(&_text
),
362 __pa_symbol(&_end
) - __pa_symbol(&_text
));
365 * reserve physical page 0 - it's a special BIOS page on many boxes,
366 * enabling clean reboots, SMP operation, laptop functions.
368 reserve_bootmem_generic(0, PAGE_SIZE
);
370 /* reserve ebda region */
372 reserve_bootmem_generic(ebda_addr
, ebda_size
);
374 /* reserve nodemap region */
376 reserve_bootmem_generic(nodemap_addr
, nodemap_size
);
380 /* Reserve SMP trampoline */
381 reserve_bootmem_generic(SMP_TRAMPOLINE_BASE
, 2*PAGE_SIZE
);
384 #ifdef CONFIG_ACPI_SLEEP
386 * Reserve low memory region for sleep support.
388 acpi_reserve_bootmem();
391 * Find and reserve possible boot-time SMP configuration:
394 #ifdef CONFIG_BLK_DEV_INITRD
395 if (boot_params
.hdr
.type_of_loader
&& boot_params
.hdr
.ramdisk_image
) {
396 unsigned long ramdisk_image
= boot_params
.hdr
.ramdisk_image
;
397 unsigned long ramdisk_size
= boot_params
.hdr
.ramdisk_size
;
398 unsigned long ramdisk_end
= ramdisk_image
+ ramdisk_size
;
399 unsigned long end_of_mem
= end_pfn
<< PAGE_SHIFT
;
401 if (ramdisk_end
<= end_of_mem
) {
402 reserve_bootmem_generic(ramdisk_image
, ramdisk_size
);
403 initrd_start
= ramdisk_image
+ PAGE_OFFSET
;
404 initrd_end
= initrd_start
+ramdisk_size
;
406 printk(KERN_ERR
"initrd extends beyond end of memory "
407 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
408 ramdisk_end
, end_of_mem
);
413 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 conswitchp
= &vga_con
;
456 #elif defined(CONFIG_DUMMY_CONSOLE)
457 conswitchp
= &dummy_con
;
462 static int __cpuinit
get_model_name(struct cpuinfo_x86
*c
)
466 if (c
->extended_cpuid_level
< 0x80000004)
469 v
= (unsigned int *) c
->x86_model_id
;
470 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
471 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
472 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
473 c
->x86_model_id
[48] = 0;
478 static void __cpuinit
display_cacheinfo(struct cpuinfo_x86
*c
)
480 unsigned int n
, dummy
, eax
, ebx
, ecx
, edx
;
482 n
= c
->extended_cpuid_level
;
484 if (n
>= 0x80000005) {
485 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
486 printk(KERN_INFO
"CPU: L1 I Cache: %dK (%d bytes/line), "
487 "D cache %dK (%d bytes/line)\n",
488 edx
>>24, edx
&0xFF, ecx
>>24, ecx
&0xFF);
489 c
->x86_cache_size
= (ecx
>>24) + (edx
>>24);
490 /* On K8 L1 TLB is inclusive, so don't count it */
494 if (n
>= 0x80000006) {
495 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
496 ecx
= cpuid_ecx(0x80000006);
497 c
->x86_cache_size
= ecx
>> 16;
498 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
500 printk(KERN_INFO
"CPU: L2 Cache: %dK (%d bytes/line)\n",
501 c
->x86_cache_size
, ecx
& 0xFF);
505 cpuid(0x80000007, &dummy
, &dummy
, &dummy
, &c
->x86_power
);
506 if (n
>= 0x80000008) {
507 cpuid(0x80000008, &eax
, &dummy
, &dummy
, &dummy
);
508 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
509 c
->x86_phys_bits
= eax
& 0xff;
514 static int nearby_node(int apicid
)
518 for (i
= apicid
- 1; i
>= 0; i
--) {
519 node
= apicid_to_node
[i
];
520 if (node
!= NUMA_NO_NODE
&& node_online(node
))
523 for (i
= apicid
+ 1; i
< MAX_LOCAL_APIC
; i
++) {
524 node
= apicid_to_node
[i
];
525 if (node
!= NUMA_NO_NODE
&& node_online(node
))
528 return first_node(node_online_map
); /* Shouldn't happen */
533 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
534 * Assumes number of cores is a power of two.
536 static void __init
amd_detect_cmp(struct cpuinfo_x86
*c
)
541 int cpu
= smp_processor_id();
543 unsigned apicid
= hard_smp_processor_id();
545 bits
= c
->x86_coreid_bits
;
547 /* Low order bits define the core id (index of core in socket) */
548 c
->cpu_core_id
= c
->phys_proc_id
& ((1 << bits
)-1);
549 /* Convert the APIC ID into the socket ID */
550 c
->phys_proc_id
= phys_pkg_id(bits
);
553 node
= c
->phys_proc_id
;
554 if (apicid_to_node
[apicid
] != NUMA_NO_NODE
)
555 node
= apicid_to_node
[apicid
];
556 if (!node_online(node
)) {
557 /* Two possibilities here:
558 - The CPU is missing memory and no node was created.
559 In that case try picking one from a nearby CPU
560 - The APIC IDs differ from the HyperTransport node IDs
561 which the K8 northbridge parsing fills in.
562 Assume they are all increased by a constant offset,
563 but in the same order as the HT nodeids.
564 If that doesn't result in a usable node fall back to the
565 path for the previous case. */
567 int ht_nodeid
= apicid
- (cpu_data(0).phys_proc_id
<< bits
);
569 if (ht_nodeid
>= 0 &&
570 apicid_to_node
[ht_nodeid
] != NUMA_NO_NODE
)
571 node
= apicid_to_node
[ht_nodeid
];
572 /* Pick a nearby node */
573 if (!node_online(node
))
574 node
= nearby_node(apicid
);
576 numa_set_node(cpu
, node
);
578 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
583 static void __cpuinit
early_init_amd(struct cpuinfo_x86
*c
)
588 /* Multi core CPU? */
589 if (c
->extended_cpuid_level
< 0x80000008)
592 ecx
= cpuid_ecx(0x80000008);
594 c
->x86_max_cores
= (ecx
& 0xff) + 1;
596 /* CPU telling us the core id bits shift? */
597 bits
= (ecx
>> 12) & 0xF;
599 /* Otherwise recompute */
601 while ((1 << bits
) < c
->x86_max_cores
)
605 c
->x86_coreid_bits
= bits
;
610 #define ENABLE_C1E_MASK 0x18000000
611 #define CPUID_PROCESSOR_SIGNATURE 1
612 #define CPUID_XFAM 0x0ff00000
613 #define CPUID_XFAM_K8 0x00000000
614 #define CPUID_XFAM_10H 0x00100000
615 #define CPUID_XFAM_11H 0x00200000
616 #define CPUID_XMOD 0x000f0000
617 #define CPUID_XMOD_REV_F 0x00040000
619 /* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
620 static __cpuinit
int amd_apic_timer_broken(void)
622 u32 lo
, hi
, eax
= cpuid_eax(CPUID_PROCESSOR_SIGNATURE
);
624 switch (eax
& CPUID_XFAM
) {
626 if ((eax
& CPUID_XMOD
) < CPUID_XMOD_REV_F
)
630 rdmsr(MSR_K8_ENABLE_C1E
, lo
, hi
);
631 if (lo
& ENABLE_C1E_MASK
)
635 /* err on the side of caution */
641 static void __cpuinit
init_amd(struct cpuinfo_x86
*c
)
649 * Disable TLB flush filter by setting HWCR.FFDIS on K8
650 * bit 6 of msr C001_0015
652 * Errata 63 for SH-B3 steppings
653 * Errata 122 for all steppings (F+ have it disabled by default)
656 rdmsrl(MSR_K8_HWCR
, value
);
658 wrmsrl(MSR_K8_HWCR
, value
);
662 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
663 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
664 clear_bit(0*32+31, (unsigned long *)&c
->x86_capability
);
666 /* On C+ stepping K8 rep microcode works well for copy/memset */
667 level
= cpuid_eax(1);
668 if (c
->x86
== 15 && ((level
>= 0x0f48 && level
< 0x0f50) ||
670 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
671 if (c
->x86
== 0x10 || c
->x86
== 0x11)
672 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
674 /* Enable workaround for FXSAVE leak */
676 set_cpu_cap(c
, X86_FEATURE_FXSAVE_LEAK
);
678 level
= get_model_name(c
);
682 /* Should distinguish Models here, but this is only
683 a fallback anyways. */
684 strcpy(c
->x86_model_id
, "Hammer");
688 display_cacheinfo(c
);
690 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
691 if (c
->x86_power
& (1<<8))
692 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
694 /* Multi core CPU? */
695 if (c
->extended_cpuid_level
>= 0x80000008)
698 if (c
->extended_cpuid_level
>= 0x80000006 &&
699 (cpuid_edx(0x80000006) & 0xf000))
700 num_cache_leaves
= 4;
702 num_cache_leaves
= 3;
704 if (c
->x86
== 0xf || c
->x86
== 0x10 || c
->x86
== 0x11)
705 set_cpu_cap(c
, X86_FEATURE_K8
);
707 /* RDTSC can be speculated around */
708 clear_cpu_cap(c
, X86_FEATURE_SYNC_RDTSC
);
710 /* Family 10 doesn't support C states in MWAIT so don't use it */
711 if (c
->x86
== 0x10 && !force_mwait
)
712 clear_cpu_cap(c
, X86_FEATURE_MWAIT
);
714 if (amd_apic_timer_broken())
715 disable_apic_timer
= 1;
718 static void __cpuinit
detect_ht(struct cpuinfo_x86
*c
)
721 u32 eax
, ebx
, ecx
, edx
;
722 int index_msb
, core_bits
;
724 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
727 if (!cpu_has(c
, X86_FEATURE_HT
))
729 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
732 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
734 if (smp_num_siblings
== 1) {
735 printk(KERN_INFO
"CPU: Hyper-Threading is disabled\n");
736 } else if (smp_num_siblings
> 1) {
738 if (smp_num_siblings
> NR_CPUS
) {
739 printk(KERN_WARNING
"CPU: Unsupported number of "
740 "siblings %d", smp_num_siblings
);
741 smp_num_siblings
= 1;
745 index_msb
= get_count_order(smp_num_siblings
);
746 c
->phys_proc_id
= phys_pkg_id(index_msb
);
748 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
750 index_msb
= get_count_order(smp_num_siblings
);
752 core_bits
= get_count_order(c
->x86_max_cores
);
754 c
->cpu_core_id
= phys_pkg_id(index_msb
) &
755 ((1 << core_bits
) - 1);
758 if ((c
->x86_max_cores
* smp_num_siblings
) > 1) {
759 printk(KERN_INFO
"CPU: Physical Processor ID: %d\n",
761 printk(KERN_INFO
"CPU: Processor Core ID: %d\n",
769 * find out the number of processor cores on the die
771 static int __cpuinit
intel_num_cpu_cores(struct cpuinfo_x86
*c
)
775 if (c
->cpuid_level
< 4)
778 cpuid_count(4, 0, &eax
, &t
, &t
, &t
);
781 return ((eax
>> 26) + 1);
786 static void srat_detect_node(void)
790 int cpu
= smp_processor_id();
791 int apicid
= hard_smp_processor_id();
793 /* Don't do the funky fallback heuristics the AMD version employs
795 node
= apicid_to_node
[apicid
];
796 if (node
== NUMA_NO_NODE
)
797 node
= first_node(node_online_map
);
798 numa_set_node(cpu
, node
);
800 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
804 static void __cpuinit
init_intel(struct cpuinfo_x86
*c
)
809 init_intel_cacheinfo(c
);
810 if (c
->cpuid_level
> 9) {
811 unsigned eax
= cpuid_eax(10);
812 /* Check for version and the number of counters */
813 if ((eax
& 0xff) && (((eax
>>8) & 0xff) > 1))
814 set_cpu_cap(c
, X86_FEATURE_ARCH_PERFMON
);
819 rdmsr(MSR_IA32_MISC_ENABLE
, l1
, l2
);
821 set_cpu_cap(c
, X86_FEATURE_BTS
);
823 set_cpu_cap(c
, X86_FEATURE_PEBS
);
826 n
= c
->extended_cpuid_level
;
827 if (n
>= 0x80000008) {
828 unsigned eax
= cpuid_eax(0x80000008);
829 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
830 c
->x86_phys_bits
= eax
& 0xff;
831 /* CPUID workaround for Intel 0F34 CPU */
832 if (c
->x86_vendor
== X86_VENDOR_INTEL
&&
833 c
->x86
== 0xF && c
->x86_model
== 0x3 &&
835 c
->x86_phys_bits
= 36;
839 c
->x86_cache_alignment
= c
->x86_clflush_size
* 2;
840 if ((c
->x86
== 0xf && c
->x86_model
>= 0x03) ||
841 (c
->x86
== 0x6 && c
->x86_model
>= 0x0e))
842 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
844 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
846 set_cpu_cap(c
, X86_FEATURE_SYNC_RDTSC
);
848 clear_cpu_cap(c
, X86_FEATURE_SYNC_RDTSC
);
849 c
->x86_max_cores
= intel_num_cpu_cores(c
);
854 static void __cpuinit
get_cpu_vendor(struct cpuinfo_x86
*c
)
856 char *v
= c
->x86_vendor_id
;
858 if (!strcmp(v
, "AuthenticAMD"))
859 c
->x86_vendor
= X86_VENDOR_AMD
;
860 else if (!strcmp(v
, "GenuineIntel"))
861 c
->x86_vendor
= X86_VENDOR_INTEL
;
863 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
866 struct cpu_model_info
{
869 char *model_names
[16];
872 /* Do some early cpuid on the boot CPU to get some parameter that are
873 needed before check_bugs. Everything advanced is in identify_cpu
875 static void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
)
879 c
->loops_per_jiffy
= loops_per_jiffy
;
880 c
->x86_cache_size
= -1;
881 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
882 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
883 c
->x86_vendor_id
[0] = '\0'; /* Unset */
884 c
->x86_model_id
[0] = '\0'; /* Unset */
885 c
->x86_clflush_size
= 64;
886 c
->x86_cache_alignment
= c
->x86_clflush_size
;
887 c
->x86_max_cores
= 1;
888 c
->x86_coreid_bits
= 0;
889 c
->extended_cpuid_level
= 0;
890 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
892 /* Get vendor name */
893 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
894 (unsigned int *)&c
->x86_vendor_id
[0],
895 (unsigned int *)&c
->x86_vendor_id
[8],
896 (unsigned int *)&c
->x86_vendor_id
[4]);
900 /* Initialize the standard set of capabilities */
901 /* Note that the vendor-specific code below might override */
903 /* Intel-defined flags: level 0x00000001 */
904 if (c
->cpuid_level
>= 0x00000001) {
906 cpuid(0x00000001, &tfms
, &misc
, &c
->x86_capability
[4],
907 &c
->x86_capability
[0]);
908 c
->x86
= (tfms
>> 8) & 0xf;
909 c
->x86_model
= (tfms
>> 4) & 0xf;
910 c
->x86_mask
= tfms
& 0xf;
912 c
->x86
+= (tfms
>> 20) & 0xff;
914 c
->x86_model
+= ((tfms
>> 16) & 0xF) << 4;
915 if (c
->x86_capability
[0] & (1<<19))
916 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
918 /* Have CPUID level 0 only - unheard of */
923 c
->phys_proc_id
= (cpuid_ebx(1) >> 24) & 0xff;
925 /* AMD-defined flags: level 0x80000001 */
926 xlvl
= cpuid_eax(0x80000000);
927 c
->extended_cpuid_level
= xlvl
;
928 if ((xlvl
& 0xffff0000) == 0x80000000) {
929 if (xlvl
>= 0x80000001) {
930 c
->x86_capability
[1] = cpuid_edx(0x80000001);
931 c
->x86_capability
[6] = cpuid_ecx(0x80000001);
933 if (xlvl
>= 0x80000004)
934 get_model_name(c
); /* Default name */
937 /* Transmeta-defined flags: level 0x80860001 */
938 xlvl
= cpuid_eax(0x80860000);
939 if ((xlvl
& 0xffff0000) == 0x80860000) {
940 /* Don't set x86_cpuid_level here for now to not confuse. */
941 if (xlvl
>= 0x80860001)
942 c
->x86_capability
[2] = cpuid_edx(0x80860001);
945 switch (c
->x86_vendor
) {
954 * This does the hard work of actually picking apart the CPU stuff...
956 void __cpuinit
identify_cpu(struct cpuinfo_x86
*c
)
960 early_identify_cpu(c
);
962 init_scattered_cpuid_features(c
);
964 c
->apicid
= phys_pkg_id(0);
967 * Vendor-specific initialization. In this section we
968 * canonicalize the feature flags, meaning if there are
969 * features a certain CPU supports which CPUID doesn't
970 * tell us, CPUID claiming incorrect flags, or other bugs,
971 * we handle them here.
973 * At the end of this section, c->x86_capability better
974 * indicate the features this CPU genuinely supports!
976 switch (c
->x86_vendor
) {
981 case X86_VENDOR_INTEL
:
985 case X86_VENDOR_UNKNOWN
:
987 display_cacheinfo(c
);
991 select_idle_routine(c
);
995 * On SMP, boot_cpu_data holds the common feature set between
996 * all CPUs; so make sure that we indicate which features are
997 * common between the CPUs. The first time this routine gets
998 * executed, c == &boot_cpu_data.
1000 if (c
!= &boot_cpu_data
) {
1001 /* AND the already accumulated flags with these */
1002 for (i
= 0; i
< NCAPINTS
; i
++)
1003 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
1006 #ifdef CONFIG_X86_MCE
1009 if (c
!= &boot_cpu_data
)
1012 numa_add_cpu(smp_processor_id());
1016 void __cpuinit
print_cpu_info(struct cpuinfo_x86
*c
)
1018 if (c
->x86_model_id
[0])
1019 printk(KERN_INFO
"%s", c
->x86_model_id
);
1021 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1022 printk(KERN_CONT
" stepping %02x\n", c
->x86_mask
);
1024 printk(KERN_CONT
"\n");
1028 * Get CPU information for use by the procfs.
1031 static int show_cpuinfo(struct seq_file
*m
, void *v
)
1033 struct cpuinfo_x86
*c
= v
;
1037 * These flag bits must match the definitions in <asm/cpufeature.h>.
1038 * NULL means this bit is undefined or reserved; either way it doesn't
1039 * have meaning as far as Linux is concerned. Note that it's important
1040 * to realize there is a difference between this table and CPUID -- if
1041 * applications want to get the raw CPUID data, they should access
1042 * /dev/cpu/<cpu_nr>/cpuid instead.
1044 static const char *const x86_cap_flags
[] = {
1046 "fpu", "vme", "de", "pse", "tsc", "msr", "pae", "mce",
1047 "cx8", "apic", NULL
, "sep", "mtrr", "pge", "mca", "cmov",
1048 "pat", "pse36", "pn", "clflush", NULL
, "dts", "acpi", "mmx",
1049 "fxsr", "sse", "sse2", "ss", "ht", "tm", "ia64", "pbe",
1052 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1053 NULL
, NULL
, NULL
, "syscall", NULL
, NULL
, NULL
, NULL
,
1054 NULL
, NULL
, NULL
, NULL
, "nx", NULL
, "mmxext", NULL
,
1055 NULL
, "fxsr_opt", "pdpe1gb", "rdtscp", NULL
, "lm",
1056 "3dnowext", "3dnow",
1058 /* Transmeta-defined */
1059 "recovery", "longrun", NULL
, "lrti", NULL
, NULL
, NULL
, NULL
,
1060 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1061 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1062 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1064 /* Other (Linux-defined) */
1065 "cxmmx", "k6_mtrr", "cyrix_arr", "centaur_mcr",
1066 NULL
, NULL
, NULL
, NULL
,
1067 "constant_tsc", "up", NULL
, "arch_perfmon",
1068 "pebs", "bts", NULL
, "sync_rdtsc",
1069 "rep_good", NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1070 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1072 /* Intel-defined (#2) */
1073 "pni", NULL
, NULL
, "monitor", "ds_cpl", "vmx", "smx", "est",
1074 "tm2", "ssse3", "cid", NULL
, NULL
, "cx16", "xtpr", NULL
,
1075 NULL
, NULL
, "dca", "sse4_1", "sse4_2", NULL
, NULL
, "popcnt",
1076 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1078 /* VIA/Cyrix/Centaur-defined */
1079 NULL
, NULL
, "rng", "rng_en", NULL
, NULL
, "ace", "ace_en",
1080 "ace2", "ace2_en", "phe", "phe_en", "pmm", "pmm_en", NULL
, NULL
,
1081 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1082 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1084 /* AMD-defined (#2) */
1085 "lahf_lm", "cmp_legacy", "svm", "extapic",
1086 "cr8_legacy", "abm", "sse4a", "misalignsse",
1087 "3dnowprefetch", "osvw", "ibs", "sse5",
1088 "skinit", "wdt", NULL
, NULL
,
1089 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1090 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1092 /* Auxiliary (Linux-defined) */
1093 "ida", NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1094 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1095 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1096 NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
, NULL
,
1098 static const char *const x86_power_flags
[] = {
1099 "ts", /* temperature sensor */
1100 "fid", /* frequency id control */
1101 "vid", /* voltage id control */
1102 "ttp", /* thermal trip */
1107 "", /* tsc invariant mapped to constant_tsc */
1116 seq_printf(m
, "processor\t: %u\n"
1118 "cpu family\t: %d\n"
1120 "model name\t: %s\n",
1122 c
->x86_vendor_id
[0] ? c
->x86_vendor_id
: "unknown",
1125 c
->x86_model_id
[0] ? c
->x86_model_id
: "unknown");
1127 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1128 seq_printf(m
, "stepping\t: %d\n", c
->x86_mask
);
1130 seq_printf(m
, "stepping\t: unknown\n");
1132 if (cpu_has(c
, X86_FEATURE_TSC
)) {
1133 unsigned int freq
= cpufreq_quick_get((unsigned)cpu
);
1137 seq_printf(m
, "cpu MHz\t\t: %u.%03u\n",
1138 freq
/ 1000, (freq
% 1000));
1142 if (c
->x86_cache_size
>= 0)
1143 seq_printf(m
, "cache size\t: %d KB\n", c
->x86_cache_size
);
1146 if (smp_num_siblings
* c
->x86_max_cores
> 1) {
1147 seq_printf(m
, "physical id\t: %d\n", c
->phys_proc_id
);
1148 seq_printf(m
, "siblings\t: %d\n",
1149 cpus_weight(per_cpu(cpu_core_map
, cpu
)));
1150 seq_printf(m
, "core id\t\t: %d\n", c
->cpu_core_id
);
1151 seq_printf(m
, "cpu cores\t: %d\n", c
->booted_cores
);
1157 "fpu_exception\t: yes\n"
1158 "cpuid level\t: %d\n"
1163 for (i
= 0; i
< 32*NCAPINTS
; i
++)
1164 if (cpu_has(c
, i
) && x86_cap_flags
[i
] != NULL
)
1165 seq_printf(m
, " %s", x86_cap_flags
[i
]);
1167 seq_printf(m
, "\nbogomips\t: %lu.%02lu\n",
1168 c
->loops_per_jiffy
/(500000/HZ
),
1169 (c
->loops_per_jiffy
/(5000/HZ
)) % 100);
1171 if (c
->x86_tlbsize
> 0)
1172 seq_printf(m
, "TLB size\t: %d 4K pages\n", c
->x86_tlbsize
);
1173 seq_printf(m
, "clflush size\t: %d\n", c
->x86_clflush_size
);
1174 seq_printf(m
, "cache_alignment\t: %d\n", c
->x86_cache_alignment
);
1176 seq_printf(m
, "address sizes\t: %u bits physical, %u bits virtual\n",
1177 c
->x86_phys_bits
, c
->x86_virt_bits
);
1179 seq_printf(m
, "power management:");
1180 for (i
= 0; i
< 32; i
++) {
1181 if (c
->x86_power
& (1 << i
)) {
1182 if (i
< ARRAY_SIZE(x86_power_flags
) &&
1184 seq_printf(m
, "%s%s",
1185 x86_power_flags
[i
][0]?" ":"",
1186 x86_power_flags
[i
]);
1188 seq_printf(m
, " [%d]", i
);
1192 seq_printf(m
, "\n\n");
1197 static void *c_start(struct seq_file
*m
, loff_t
*pos
)
1199 if (*pos
== 0) /* just in case, cpu 0 is not the first */
1200 *pos
= first_cpu(cpu_online_map
);
1201 if ((*pos
) < NR_CPUS
&& cpu_online(*pos
))
1202 return &cpu_data(*pos
);
1206 static void *c_next(struct seq_file
*m
, void *v
, loff_t
*pos
)
1208 *pos
= next_cpu(*pos
, cpu_online_map
);
1209 return c_start(m
, pos
);
1212 static void c_stop(struct seq_file
*m
, void *v
)
1216 struct seq_operations cpuinfo_op
= {
1220 .show
= show_cpuinfo
,