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/screen_info.h>
19 #include <linux/ioport.h>
20 #include <linux/delay.h>
21 #include <linux/init.h>
22 #include <linux/initrd.h>
23 #include <linux/highmem.h>
24 #include <linux/bootmem.h>
25 #include <linux/module.h>
26 #include <asm/processor.h>
27 #include <linux/console.h>
28 #include <linux/seq_file.h>
29 #include <linux/crash_dump.h>
30 #include <linux/root_dev.h>
31 #include <linux/pci.h>
32 #include <asm/pci-direct.h>
33 #include <linux/efi.h>
34 #include <linux/acpi.h>
35 #include <linux/kallsyms.h>
36 #include <linux/edd.h>
37 #include <linux/iscsi_ibft.h>
38 #include <linux/mmzone.h>
39 #include <linux/kexec.h>
40 #include <linux/cpufreq.h>
41 #include <linux/dmi.h>
42 #include <linux/dma-mapping.h>
43 #include <linux/ctype.h>
44 #include <linux/sort.h>
45 #include <linux/uaccess.h>
46 #include <linux/init_ohci1394_dma.h>
47 #include <linux/kvm_para.h>
50 #include <asm/uaccess.h>
51 #include <asm/system.h>
52 #include <asm/vsyscall.h>
57 #include <video/edid.h>
61 #include <asm/mpspec.h>
62 #include <asm/mmu_context.h>
63 #include <asm/proto.h>
64 #include <asm/setup.h>
66 #include <asm/sections.h>
68 #include <asm/cacheflush.h>
71 #include <asm/topology.h>
72 #include <asm/trampoline.h>
74 #include <asm/mmconfig.h>
76 #include <mach_apic.h>
77 #ifdef CONFIG_PARAVIRT
78 #include <asm/paravirt.h>
87 struct cpuinfo_x86 boot_cpu_data __read_mostly
;
88 EXPORT_SYMBOL(boot_cpu_data
);
90 __u32 cleared_cpu_caps
[NCAPINTS
] __cpuinitdata
;
92 unsigned long mmu_cr4_features
;
94 /* Boot loader ID as an integer, for the benefit of proc_dointvec */
97 unsigned long saved_video_mode
;
99 int force_mwait __cpuinitdata
;
105 char dmi_alloc_data
[DMI_MAX_DATA
];
110 struct screen_info screen_info
;
111 EXPORT_SYMBOL(screen_info
);
112 struct sys_desc_table_struct
{
113 unsigned short length
;
114 unsigned char table
[0];
117 struct edid_info edid_info
;
118 EXPORT_SYMBOL_GPL(edid_info
);
120 extern int root_mountflags
;
122 char __initdata command_line
[COMMAND_LINE_SIZE
];
124 static struct resource standard_io_resources
[] = {
125 { .name
= "dma1", .start
= 0x00, .end
= 0x1f,
126 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
127 { .name
= "pic1", .start
= 0x20, .end
= 0x21,
128 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
129 { .name
= "timer0", .start
= 0x40, .end
= 0x43,
130 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
131 { .name
= "timer1", .start
= 0x50, .end
= 0x53,
132 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
133 { .name
= "keyboard", .start
= 0x60, .end
= 0x60,
134 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
135 { .name
= "keyboard", .start
= 0x64, .end
= 0x64,
136 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
137 { .name
= "dma page reg", .start
= 0x80, .end
= 0x8f,
138 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
139 { .name
= "pic2", .start
= 0xa0, .end
= 0xa1,
140 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
141 { .name
= "dma2", .start
= 0xc0, .end
= 0xdf,
142 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
},
143 { .name
= "fpu", .start
= 0xf0, .end
= 0xff,
144 .flags
= IORESOURCE_BUSY
| IORESOURCE_IO
}
147 #define IORESOURCE_RAM (IORESOURCE_BUSY | IORESOURCE_MEM)
149 static struct resource data_resource
= {
150 .name
= "Kernel data",
153 .flags
= IORESOURCE_RAM
,
155 static struct resource code_resource
= {
156 .name
= "Kernel code",
159 .flags
= IORESOURCE_RAM
,
161 static struct resource bss_resource
= {
162 .name
= "Kernel bss",
165 .flags
= IORESOURCE_RAM
,
168 static void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
);
170 #ifdef CONFIG_PROC_VMCORE
171 /* elfcorehdr= specifies the location of elf core header
172 * stored by the crashed kernel. This option will be passed
173 * by kexec loader to the capture kernel.
175 static int __init
setup_elfcorehdr(char *arg
)
180 elfcorehdr_addr
= memparse(arg
, &end
);
181 return end
> arg
? 0 : -EINVAL
;
183 early_param("elfcorehdr", setup_elfcorehdr
);
188 contig_initmem_init(unsigned long start_pfn
, unsigned long end_pfn
)
190 unsigned long bootmap_size
, bootmap
;
192 bootmap_size
= bootmem_bootmap_pages(end_pfn
)<<PAGE_SHIFT
;
193 bootmap
= find_e820_area(0, end_pfn
<<PAGE_SHIFT
, bootmap_size
,
196 panic("Cannot find bootmem map of size %ld\n", bootmap_size
);
197 bootmap_size
= init_bootmem(bootmap
>> PAGE_SHIFT
, end_pfn
);
198 e820_register_active_regions(0, start_pfn
, end_pfn
);
199 free_bootmem_with_active_regions(0, end_pfn
);
200 early_res_to_bootmem(0, end_pfn
<<PAGE_SHIFT
);
201 reserve_bootmem(bootmap
, bootmap_size
, BOOTMEM_DEFAULT
);
205 #if defined(CONFIG_EDD) || defined(CONFIG_EDD_MODULE)
207 #ifdef CONFIG_EDD_MODULE
211 * copy_edd() - Copy the BIOS EDD information
212 * from boot_params into a safe place.
215 static inline void copy_edd(void)
217 memcpy(edd
.mbr_signature
, boot_params
.edd_mbr_sig_buffer
,
218 sizeof(edd
.mbr_signature
));
219 memcpy(edd
.edd_info
, boot_params
.eddbuf
, sizeof(edd
.edd_info
));
220 edd
.mbr_signature_nr
= boot_params
.edd_mbr_sig_buf_entries
;
221 edd
.edd_info_nr
= boot_params
.eddbuf_entries
;
224 static inline void copy_edd(void)
230 static void __init
reserve_crashkernel(void)
232 unsigned long long total_mem
;
233 unsigned long long crash_size
, crash_base
;
236 total_mem
= ((unsigned long long)max_low_pfn
- min_low_pfn
) << PAGE_SHIFT
;
238 ret
= parse_crashkernel(boot_command_line
, total_mem
,
239 &crash_size
, &crash_base
);
240 if (ret
== 0 && crash_size
) {
241 if (crash_base
<= 0) {
242 printk(KERN_INFO
"crashkernel reservation failed - "
243 "you have to specify a base address\n");
247 if (reserve_bootmem(crash_base
, crash_size
,
248 BOOTMEM_EXCLUSIVE
) < 0) {
249 printk(KERN_INFO
"crashkernel reservation failed - "
250 "memory is in use\n");
254 printk(KERN_INFO
"Reserving %ldMB of memory at %ldMB "
255 "for crashkernel (System RAM: %ldMB)\n",
256 (unsigned long)(crash_size
>> 20),
257 (unsigned long)(crash_base
>> 20),
258 (unsigned long)(total_mem
>> 20));
259 crashk_res
.start
= crash_base
;
260 crashk_res
.end
= crash_base
+ crash_size
- 1;
261 insert_resource(&iomem_resource
, &crashk_res
);
265 static inline void __init
reserve_crashkernel(void)
269 /* Overridden in paravirt.c if CONFIG_PARAVIRT */
270 void __attribute__((weak
)) __init
memory_setup(void)
272 machine_specific_memory_setup();
275 static void __init
parse_setup_data(void)
277 struct setup_data
*data
;
278 unsigned long pa_data
;
280 if (boot_params
.hdr
.version
< 0x0209)
282 pa_data
= boot_params
.hdr
.setup_data
;
284 data
= early_ioremap(pa_data
, PAGE_SIZE
);
285 switch (data
->type
) {
289 #ifndef CONFIG_DEBUG_BOOT_PARAMS
290 free_early(pa_data
, pa_data
+sizeof(*data
)+data
->len
);
292 pa_data
= data
->next
;
293 early_iounmap(data
, PAGE_SIZE
);
298 * setup_arch - architecture-specific boot-time initializations
300 * Note: On x86_64, fixmaps are ready for use even before this is called.
302 void __init
setup_arch(char **cmdline_p
)
306 printk(KERN_INFO
"Command line: %s\n", boot_command_line
);
308 ROOT_DEV
= old_decode_dev(boot_params
.hdr
.root_dev
);
309 screen_info
= boot_params
.screen_info
;
310 edid_info
= boot_params
.edid_info
;
311 saved_video_mode
= boot_params
.hdr
.vid_mode
;
312 bootloader_type
= boot_params
.hdr
.type_of_loader
;
314 #ifdef CONFIG_BLK_DEV_RAM
315 rd_image_start
= boot_params
.hdr
.ram_size
& RAMDISK_IMAGE_START_MASK
;
316 rd_prompt
= ((boot_params
.hdr
.ram_size
& RAMDISK_PROMPT_FLAG
) != 0);
317 rd_doload
= ((boot_params
.hdr
.ram_size
& RAMDISK_LOAD_FLAG
) != 0);
320 if (!strncmp((char *)&boot_params
.efi_info
.efi_loader_signature
,
330 if (!boot_params
.hdr
.root_flags
)
331 root_mountflags
&= ~MS_RDONLY
;
332 init_mm
.start_code
= (unsigned long) &_text
;
333 init_mm
.end_code
= (unsigned long) &_etext
;
334 init_mm
.end_data
= (unsigned long) &_edata
;
335 init_mm
.brk
= (unsigned long) &_end
;
337 code_resource
.start
= virt_to_phys(&_text
);
338 code_resource
.end
= virt_to_phys(&_etext
)-1;
339 data_resource
.start
= virt_to_phys(&_etext
);
340 data_resource
.end
= virt_to_phys(&_edata
)-1;
341 bss_resource
.start
= virt_to_phys(&__bss_start
);
342 bss_resource
.end
= virt_to_phys(&__bss_stop
)-1;
344 early_identify_cpu(&boot_cpu_data
);
346 strlcpy(command_line
, boot_command_line
, COMMAND_LINE_SIZE
);
347 *cmdline_p
= command_line
;
353 #ifdef CONFIG_PROVIDE_OHCI1394_DMA_INIT
354 if (init_ohci1394_dma_early
)
355 init_ohci1394_dma_on_all_controllers();
358 finish_e820_parsing();
360 /* after parse_early_param, so could debug it */
361 insert_resource(&iomem_resource
, &code_resource
);
362 insert_resource(&iomem_resource
, &data_resource
);
363 insert_resource(&iomem_resource
, &bss_resource
);
365 early_gart_iommu_check();
367 e820_register_active_regions(0, 0, -1UL);
369 * partially used pages are not usable - thus
370 * we are rounding upwards:
372 end_pfn
= e820_end_of_ram();
373 /* update e820 for memory not covered by WB MTRRs */
375 if (mtrr_trim_uncached_memory(end_pfn
)) {
376 e820_register_active_regions(0, 0, -1UL);
377 end_pfn
= e820_end_of_ram();
380 num_physpages
= end_pfn
;
384 max_pfn_mapped
= init_memory_mapping(0, (max_pfn_mapped
<< PAGE_SHIFT
));
394 #ifdef CONFIG_KVM_CLOCK
399 /* setup to use the early static init tables during kernel startup */
400 x86_cpu_to_apicid_early_ptr
= (void *)x86_cpu_to_apicid_init
;
401 x86_bios_cpu_apicid_early_ptr
= (void *)x86_bios_cpu_apicid_init
;
403 x86_cpu_to_node_map_early_ptr
= (void *)x86_cpu_to_node_map_init
;
409 * Initialize the ACPI boot-time table parser (gets the RSDP and SDT).
410 * Call this early for SRAT node setup.
412 acpi_boot_table_init();
415 /* How many end-of-memory variables you have, grandma! */
416 max_low_pfn
= end_pfn
;
418 high_memory
= (void *)__va(end_pfn
* PAGE_SIZE
- 1) + 1;
420 /* Remove active ranges so rediscovery with NUMA-awareness happens */
421 remove_all_active_ranges();
423 #ifdef CONFIG_ACPI_NUMA
425 * Parse SRAT to discover nodes.
431 numa_initmem_init(0, end_pfn
);
433 contig_initmem_init(0, end_pfn
);
436 dma32_reserve_bootmem();
438 #ifdef CONFIG_ACPI_SLEEP
440 * Reserve low memory region for sleep support.
442 acpi_reserve_bootmem();
446 efi_reserve_bootmem();
449 * Find and reserve possible boot-time SMP configuration:
452 #ifdef CONFIG_BLK_DEV_INITRD
453 if (boot_params
.hdr
.type_of_loader
&& boot_params
.hdr
.ramdisk_image
) {
454 unsigned long ramdisk_image
= boot_params
.hdr
.ramdisk_image
;
455 unsigned long ramdisk_size
= boot_params
.hdr
.ramdisk_size
;
456 unsigned long ramdisk_end
= ramdisk_image
+ ramdisk_size
;
457 unsigned long end_of_mem
= end_pfn
<< PAGE_SHIFT
;
459 if (ramdisk_end
<= end_of_mem
) {
461 * don't need to reserve again, already reserved early
462 * in x86_64_start_kernel, and early_res_to_bootmem
463 * convert that to reserved in bootmem
465 initrd_start
= ramdisk_image
+ PAGE_OFFSET
;
466 initrd_end
= initrd_start
+ramdisk_size
;
468 free_bootmem(ramdisk_image
, ramdisk_size
);
469 printk(KERN_ERR
"initrd extends beyond end of memory "
470 "(0x%08lx > 0x%08lx)\ndisabling initrd\n",
471 ramdisk_end
, end_of_mem
);
476 reserve_crashkernel();
478 reserve_ibft_region();
487 * Read APIC and some other early information from ACPI tables.
495 * get boot-time SMP configuration:
497 if (smp_found_config
)
499 init_apic_mappings();
500 ioapic_init_mappings();
505 * We trust e820 completely. No explicit ROM probing in memory.
507 e820_reserve_resources();
508 e820_mark_nosave_regions();
510 /* request I/O space for devices used on all i[345]86 PCs */
511 for (i
= 0; i
< ARRAY_SIZE(standard_io_resources
); i
++)
512 request_resource(&ioport_resource
, &standard_io_resources
[i
]);
517 #if defined(CONFIG_VGA_CONSOLE)
518 if (!efi_enabled
|| (efi_mem_type(0xa0000) != EFI_CONVENTIONAL_MEMORY
))
519 conswitchp
= &vga_con
;
520 #elif defined(CONFIG_DUMMY_CONSOLE)
521 conswitchp
= &dummy_con
;
525 /* do this before identify_cpu for boot cpu */
526 check_enable_amd_mmconf_dmi();
529 static int __cpuinit
get_model_name(struct cpuinfo_x86
*c
)
533 if (c
->extended_cpuid_level
< 0x80000004)
536 v
= (unsigned int *) c
->x86_model_id
;
537 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
538 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
539 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
540 c
->x86_model_id
[48] = 0;
545 static void __cpuinit
display_cacheinfo(struct cpuinfo_x86
*c
)
547 unsigned int n
, dummy
, eax
, ebx
, ecx
, edx
;
549 n
= c
->extended_cpuid_level
;
551 if (n
>= 0x80000005) {
552 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
553 printk(KERN_INFO
"CPU: L1 I Cache: %dK (%d bytes/line), "
554 "D cache %dK (%d bytes/line)\n",
555 edx
>>24, edx
&0xFF, ecx
>>24, ecx
&0xFF);
556 c
->x86_cache_size
= (ecx
>>24) + (edx
>>24);
557 /* On K8 L1 TLB is inclusive, so don't count it */
561 if (n
>= 0x80000006) {
562 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
563 ecx
= cpuid_ecx(0x80000006);
564 c
->x86_cache_size
= ecx
>> 16;
565 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
567 printk(KERN_INFO
"CPU: L2 Cache: %dK (%d bytes/line)\n",
568 c
->x86_cache_size
, ecx
& 0xFF);
570 if (n
>= 0x80000008) {
571 cpuid(0x80000008, &eax
, &dummy
, &dummy
, &dummy
);
572 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
573 c
->x86_phys_bits
= eax
& 0xff;
578 static int __cpuinit
nearby_node(int apicid
)
582 for (i
= apicid
- 1; i
>= 0; i
--) {
583 node
= apicid_to_node
[i
];
584 if (node
!= NUMA_NO_NODE
&& node_online(node
))
587 for (i
= apicid
+ 1; i
< MAX_LOCAL_APIC
; i
++) {
588 node
= apicid_to_node
[i
];
589 if (node
!= NUMA_NO_NODE
&& node_online(node
))
592 return first_node(node_online_map
); /* Shouldn't happen */
597 * On a AMD dual core setup the lower bits of the APIC id distingush the cores.
598 * Assumes number of cores is a power of two.
600 static void __cpuinit
amd_detect_cmp(struct cpuinfo_x86
*c
)
605 int cpu
= smp_processor_id();
607 unsigned apicid
= hard_smp_processor_id();
609 bits
= c
->x86_coreid_bits
;
611 /* Low order bits define the core id (index of core in socket) */
612 c
->cpu_core_id
= c
->initial_apicid
& ((1 << bits
)-1);
613 /* Convert the initial APIC ID into the socket ID */
614 c
->phys_proc_id
= c
->initial_apicid
>> bits
;
617 node
= c
->phys_proc_id
;
618 if (apicid_to_node
[apicid
] != NUMA_NO_NODE
)
619 node
= apicid_to_node
[apicid
];
620 if (!node_online(node
)) {
621 /* Two possibilities here:
622 - The CPU is missing memory and no node was created.
623 In that case try picking one from a nearby CPU
624 - The APIC IDs differ from the HyperTransport node IDs
625 which the K8 northbridge parsing fills in.
626 Assume they are all increased by a constant offset,
627 but in the same order as the HT nodeids.
628 If that doesn't result in a usable node fall back to the
629 path for the previous case. */
631 int ht_nodeid
= c
->initial_apicid
;
633 if (ht_nodeid
>= 0 &&
634 apicid_to_node
[ht_nodeid
] != NUMA_NO_NODE
)
635 node
= apicid_to_node
[ht_nodeid
];
636 /* Pick a nearby node */
637 if (!node_online(node
))
638 node
= nearby_node(apicid
);
640 numa_set_node(cpu
, node
);
642 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
647 static void __cpuinit
early_init_amd_mc(struct cpuinfo_x86
*c
)
652 /* Multi core CPU? */
653 if (c
->extended_cpuid_level
< 0x80000008)
656 ecx
= cpuid_ecx(0x80000008);
658 c
->x86_max_cores
= (ecx
& 0xff) + 1;
660 /* CPU telling us the core id bits shift? */
661 bits
= (ecx
>> 12) & 0xF;
663 /* Otherwise recompute */
665 while ((1 << bits
) < c
->x86_max_cores
)
669 c
->x86_coreid_bits
= bits
;
674 #define ENABLE_C1E_MASK 0x18000000
675 #define CPUID_PROCESSOR_SIGNATURE 1
676 #define CPUID_XFAM 0x0ff00000
677 #define CPUID_XFAM_K8 0x00000000
678 #define CPUID_XFAM_10H 0x00100000
679 #define CPUID_XFAM_11H 0x00200000
680 #define CPUID_XMOD 0x000f0000
681 #define CPUID_XMOD_REV_F 0x00040000
683 /* AMD systems with C1E don't have a working lAPIC timer. Check for that. */
684 static __cpuinit
int amd_apic_timer_broken(void)
686 u32 lo
, hi
, eax
= cpuid_eax(CPUID_PROCESSOR_SIGNATURE
);
688 switch (eax
& CPUID_XFAM
) {
690 if ((eax
& CPUID_XMOD
) < CPUID_XMOD_REV_F
)
694 rdmsr(MSR_K8_ENABLE_C1E
, lo
, hi
);
695 if (lo
& ENABLE_C1E_MASK
)
699 /* err on the side of caution */
705 static void __cpuinit
early_init_amd(struct cpuinfo_x86
*c
)
707 early_init_amd_mc(c
);
709 /* c->x86_power is 8000_0007 edx. Bit 8 is constant TSC */
710 if (c
->x86_power
& (1<<8))
711 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
714 static void __cpuinit
init_amd(struct cpuinfo_x86
*c
)
722 * Disable TLB flush filter by setting HWCR.FFDIS on K8
723 * bit 6 of msr C001_0015
725 * Errata 63 for SH-B3 steppings
726 * Errata 122 for all steppings (F+ have it disabled by default)
729 rdmsrl(MSR_K8_HWCR
, value
);
731 wrmsrl(MSR_K8_HWCR
, value
);
735 /* Bit 31 in normal CPUID used for nonstandard 3DNow ID;
736 3DNow is IDd by bit 31 in extended CPUID (1*32+31) anyway */
737 clear_cpu_cap(c
, 0*32+31);
739 /* On C+ stepping K8 rep microcode works well for copy/memset */
740 level
= cpuid_eax(1);
741 if (c
->x86
== 15 && ((level
>= 0x0f48 && level
< 0x0f50) ||
743 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
744 if (c
->x86
== 0x10 || c
->x86
== 0x11)
745 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
747 /* Enable workaround for FXSAVE leak */
749 set_cpu_cap(c
, X86_FEATURE_FXSAVE_LEAK
);
751 level
= get_model_name(c
);
755 /* Should distinguish Models here, but this is only
756 a fallback anyways. */
757 strcpy(c
->x86_model_id
, "Hammer");
761 display_cacheinfo(c
);
763 /* Multi core CPU? */
764 if (c
->extended_cpuid_level
>= 0x80000008)
767 if (c
->extended_cpuid_level
>= 0x80000006 &&
768 (cpuid_edx(0x80000006) & 0xf000))
769 num_cache_leaves
= 4;
771 num_cache_leaves
= 3;
773 if (c
->x86
== 0xf || c
->x86
== 0x10 || c
->x86
== 0x11)
774 set_cpu_cap(c
, X86_FEATURE_K8
);
776 /* MFENCE stops RDTSC speculation */
777 set_cpu_cap(c
, X86_FEATURE_MFENCE_RDTSC
);
780 fam10h_check_enable_mmcfg();
782 if (amd_apic_timer_broken())
783 disable_apic_timer
= 1;
785 if (c
== &boot_cpu_data
&& c
->x86
>= 0xf && c
->x86
<= 0x11) {
786 unsigned long long tseg
;
789 * Split up direct mapping around the TSEG SMM area.
790 * Don't do it for gbpages because there seems very little
791 * benefit in doing so.
793 if (!rdmsrl_safe(MSR_K8_TSEG_ADDR
, &tseg
) &&
794 (tseg
>> PMD_SHIFT
) < (max_pfn_mapped
>> (PMD_SHIFT
-PAGE_SHIFT
)))
795 set_memory_4k((unsigned long)__va(tseg
), 1);
799 void __cpuinit
detect_ht(struct cpuinfo_x86
*c
)
802 u32 eax
, ebx
, ecx
, edx
;
803 int index_msb
, core_bits
;
805 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
808 if (!cpu_has(c
, X86_FEATURE_HT
))
810 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
813 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
815 if (smp_num_siblings
== 1) {
816 printk(KERN_INFO
"CPU: Hyper-Threading is disabled\n");
817 } else if (smp_num_siblings
> 1) {
819 if (smp_num_siblings
> NR_CPUS
) {
820 printk(KERN_WARNING
"CPU: Unsupported number of "
821 "siblings %d", smp_num_siblings
);
822 smp_num_siblings
= 1;
826 index_msb
= get_count_order(smp_num_siblings
);
827 c
->phys_proc_id
= phys_pkg_id(index_msb
);
829 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
831 index_msb
= get_count_order(smp_num_siblings
);
833 core_bits
= get_count_order(c
->x86_max_cores
);
835 c
->cpu_core_id
= phys_pkg_id(index_msb
) &
836 ((1 << core_bits
) - 1);
839 if ((c
->x86_max_cores
* smp_num_siblings
) > 1) {
840 printk(KERN_INFO
"CPU: Physical Processor ID: %d\n",
842 printk(KERN_INFO
"CPU: Processor Core ID: %d\n",
850 * find out the number of processor cores on the die
852 static int __cpuinit
intel_num_cpu_cores(struct cpuinfo_x86
*c
)
856 if (c
->cpuid_level
< 4)
859 cpuid_count(4, 0, &eax
, &t
, &t
, &t
);
862 return ((eax
>> 26) + 1);
867 static void __cpuinit
srat_detect_node(void)
871 int cpu
= smp_processor_id();
872 int apicid
= hard_smp_processor_id();
874 /* Don't do the funky fallback heuristics the AMD version employs
876 node
= apicid_to_node
[apicid
];
877 if (node
== NUMA_NO_NODE
|| !node_online(node
))
878 node
= first_node(node_online_map
);
879 numa_set_node(cpu
, node
);
881 printk(KERN_INFO
"CPU %d/%x -> Node %d\n", cpu
, apicid
, node
);
885 static void __cpuinit
early_init_intel(struct cpuinfo_x86
*c
)
887 if ((c
->x86
== 0xf && c
->x86_model
>= 0x03) ||
888 (c
->x86
== 0x6 && c
->x86_model
>= 0x0e))
889 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
892 static void __cpuinit
init_intel(struct cpuinfo_x86
*c
)
897 init_intel_cacheinfo(c
);
898 if (c
->cpuid_level
> 9) {
899 unsigned eax
= cpuid_eax(10);
900 /* Check for version and the number of counters */
901 if ((eax
& 0xff) && (((eax
>>8) & 0xff) > 1))
902 set_cpu_cap(c
, X86_FEATURE_ARCH_PERFMON
);
907 rdmsr(MSR_IA32_MISC_ENABLE
, l1
, l2
);
909 set_cpu_cap(c
, X86_FEATURE_BTS
);
911 set_cpu_cap(c
, X86_FEATURE_PEBS
);
918 n
= c
->extended_cpuid_level
;
919 if (n
>= 0x80000008) {
920 unsigned eax
= cpuid_eax(0x80000008);
921 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
922 c
->x86_phys_bits
= eax
& 0xff;
923 /* CPUID workaround for Intel 0F34 CPU */
924 if (c
->x86_vendor
== X86_VENDOR_INTEL
&&
925 c
->x86
== 0xF && c
->x86_model
== 0x3 &&
927 c
->x86_phys_bits
= 36;
931 c
->x86_cache_alignment
= c
->x86_clflush_size
* 2;
933 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
934 set_cpu_cap(c
, X86_FEATURE_LFENCE_RDTSC
);
935 c
->x86_max_cores
= intel_num_cpu_cores(c
);
940 static void __cpuinit
early_init_centaur(struct cpuinfo_x86
*c
)
942 if (c
->x86
== 0x6 && c
->x86_model
>= 0xf)
943 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
946 static void __cpuinit
init_centaur(struct cpuinfo_x86
*c
)
951 n
= c
->extended_cpuid_level
;
952 if (n
>= 0x80000008) {
953 unsigned eax
= cpuid_eax(0x80000008);
954 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
955 c
->x86_phys_bits
= eax
& 0xff;
958 if (c
->x86
== 0x6 && c
->x86_model
>= 0xf) {
959 c
->x86_cache_alignment
= c
->x86_clflush_size
* 2;
960 set_cpu_cap(c
, X86_FEATURE_CONSTANT_TSC
);
961 set_cpu_cap(c
, X86_FEATURE_REP_GOOD
);
963 set_cpu_cap(c
, X86_FEATURE_LFENCE_RDTSC
);
966 static void __cpuinit
get_cpu_vendor(struct cpuinfo_x86
*c
)
968 char *v
= c
->x86_vendor_id
;
970 if (!strcmp(v
, "AuthenticAMD"))
971 c
->x86_vendor
= X86_VENDOR_AMD
;
972 else if (!strcmp(v
, "GenuineIntel"))
973 c
->x86_vendor
= X86_VENDOR_INTEL
;
974 else if (!strcmp(v
, "CentaurHauls"))
975 c
->x86_vendor
= X86_VENDOR_CENTAUR
;
977 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
980 /* Do some early cpuid on the boot CPU to get some parameter that are
981 needed before check_bugs. Everything advanced is in identify_cpu
983 static void __cpuinit
early_identify_cpu(struct cpuinfo_x86
*c
)
987 c
->loops_per_jiffy
= loops_per_jiffy
;
988 c
->x86_cache_size
= -1;
989 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
990 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
991 c
->x86_vendor_id
[0] = '\0'; /* Unset */
992 c
->x86_model_id
[0] = '\0'; /* Unset */
993 c
->x86_clflush_size
= 64;
994 c
->x86_cache_alignment
= c
->x86_clflush_size
;
995 c
->x86_max_cores
= 1;
996 c
->x86_coreid_bits
= 0;
997 c
->extended_cpuid_level
= 0;
998 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
1000 /* Get vendor name */
1001 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
1002 (unsigned int *)&c
->x86_vendor_id
[0],
1003 (unsigned int *)&c
->x86_vendor_id
[8],
1004 (unsigned int *)&c
->x86_vendor_id
[4]);
1008 /* Initialize the standard set of capabilities */
1009 /* Note that the vendor-specific code below might override */
1011 /* Intel-defined flags: level 0x00000001 */
1012 if (c
->cpuid_level
>= 0x00000001) {
1014 cpuid(0x00000001, &tfms
, &misc
, &c
->x86_capability
[4],
1015 &c
->x86_capability
[0]);
1016 c
->x86
= (tfms
>> 8) & 0xf;
1017 c
->x86_model
= (tfms
>> 4) & 0xf;
1018 c
->x86_mask
= tfms
& 0xf;
1020 c
->x86
+= (tfms
>> 20) & 0xff;
1022 c
->x86_model
+= ((tfms
>> 16) & 0xF) << 4;
1023 if (test_cpu_cap(c
, X86_FEATURE_CLFLSH
))
1024 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
1026 /* Have CPUID level 0 only - unheard of */
1030 c
->initial_apicid
= (cpuid_ebx(1) >> 24) & 0xff;
1032 c
->phys_proc_id
= c
->initial_apicid
;
1034 /* AMD-defined flags: level 0x80000001 */
1035 xlvl
= cpuid_eax(0x80000000);
1036 c
->extended_cpuid_level
= xlvl
;
1037 if ((xlvl
& 0xffff0000) == 0x80000000) {
1038 if (xlvl
>= 0x80000001) {
1039 c
->x86_capability
[1] = cpuid_edx(0x80000001);
1040 c
->x86_capability
[6] = cpuid_ecx(0x80000001);
1042 if (xlvl
>= 0x80000004)
1043 get_model_name(c
); /* Default name */
1046 /* Transmeta-defined flags: level 0x80860001 */
1047 xlvl
= cpuid_eax(0x80860000);
1048 if ((xlvl
& 0xffff0000) == 0x80860000) {
1049 /* Don't set x86_cpuid_level here for now to not confuse. */
1050 if (xlvl
>= 0x80860001)
1051 c
->x86_capability
[2] = cpuid_edx(0x80860001);
1054 c
->extended_cpuid_level
= cpuid_eax(0x80000000);
1055 if (c
->extended_cpuid_level
>= 0x80000007)
1056 c
->x86_power
= cpuid_edx(0x80000007);
1058 switch (c
->x86_vendor
) {
1059 case X86_VENDOR_AMD
:
1062 case X86_VENDOR_INTEL
:
1063 early_init_intel(c
);
1065 case X86_VENDOR_CENTAUR
:
1066 early_init_centaur(c
);
1070 validate_pat_support(c
);
1074 * This does the hard work of actually picking apart the CPU stuff...
1076 void __cpuinit
identify_cpu(struct cpuinfo_x86
*c
)
1080 early_identify_cpu(c
);
1082 init_scattered_cpuid_features(c
);
1084 c
->apicid
= phys_pkg_id(0);
1087 * Vendor-specific initialization. In this section we
1088 * canonicalize the feature flags, meaning if there are
1089 * features a certain CPU supports which CPUID doesn't
1090 * tell us, CPUID claiming incorrect flags, or other bugs,
1091 * we handle them here.
1093 * At the end of this section, c->x86_capability better
1094 * indicate the features this CPU genuinely supports!
1096 switch (c
->x86_vendor
) {
1097 case X86_VENDOR_AMD
:
1101 case X86_VENDOR_INTEL
:
1105 case X86_VENDOR_CENTAUR
:
1109 case X86_VENDOR_UNKNOWN
:
1111 display_cacheinfo(c
);
1118 * On SMP, boot_cpu_data holds the common feature set between
1119 * all CPUs; so make sure that we indicate which features are
1120 * common between the CPUs. The first time this routine gets
1121 * executed, c == &boot_cpu_data.
1123 if (c
!= &boot_cpu_data
) {
1124 /* AND the already accumulated flags with these */
1125 for (i
= 0; i
< NCAPINTS
; i
++)
1126 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
1129 /* Clear all flags overriden by options */
1130 for (i
= 0; i
< NCAPINTS
; i
++)
1131 c
->x86_capability
[i
] &= ~cleared_cpu_caps
[i
];
1133 #ifdef CONFIG_X86_MCE
1136 select_idle_routine(c
);
1139 numa_add_cpu(smp_processor_id());
1144 void __cpuinit
identify_boot_cpu(void)
1146 identify_cpu(&boot_cpu_data
);
1149 void __cpuinit
identify_secondary_cpu(struct cpuinfo_x86
*c
)
1151 BUG_ON(c
== &boot_cpu_data
);
1156 static __init
int setup_noclflush(char *arg
)
1158 setup_clear_cpu_cap(X86_FEATURE_CLFLSH
);
1161 __setup("noclflush", setup_noclflush
);
1163 void __cpuinit
print_cpu_info(struct cpuinfo_x86
*c
)
1165 if (c
->x86_model_id
[0])
1166 printk(KERN_CONT
"%s", c
->x86_model_id
);
1168 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1169 printk(KERN_CONT
" stepping %02x\n", c
->x86_mask
);
1171 printk(KERN_CONT
"\n");
1174 static __init
int setup_disablecpuid(char *arg
)
1177 if (get_option(&arg
, &bit
) && bit
< NCAPINTS
*32)
1178 setup_clear_cpu_cap(bit
);
1183 __setup("clearcpuid=", setup_disablecpuid
);