1 #include <linux/init.h>
2 #include <linux/kernel.h>
3 #include <linux/sched.h>
4 #include <linux/string.h>
5 #include <linux/bootmem.h>
6 #include <linux/bitops.h>
7 #include <linux/module.h>
8 #include <linux/kgdb.h>
9 #include <linux/topology.h>
10 #include <linux/delay.h>
11 #include <linux/smp.h>
12 #include <linux/percpu.h>
16 #include <asm/linkage.h>
17 #include <asm/mmu_context.h>
23 #ifdef CONFIG_X86_LOCAL_APIC
24 #include <asm/mpspec.h>
26 #include <mach_apic.h>
27 #include <asm/genapic.h>
30 #include <asm/pgtable.h>
31 #include <asm/processor.h>
33 #include <asm/atomic.h>
34 #include <asm/proto.h>
35 #include <asm/sections.h>
36 #include <asm/setup.h>
40 static struct cpu_dev
*this_cpu __cpuinitdata
;
43 /* We need valid kernel segments for data and code in long mode too
44 * IRET will check the segment types kkeil 2000/10/28
45 * Also sysret mandates a special GDT layout
47 /* The TLS descriptors are currently at a different place compared to i386.
48 Hopefully nobody expects them at a fixed place (Wine?) */
49 DEFINE_PER_CPU(struct gdt_page
, gdt_page
) = { .gdt
= {
50 [GDT_ENTRY_KERNEL32_CS
] = { { { 0x0000ffff, 0x00cf9b00 } } },
51 [GDT_ENTRY_KERNEL_CS
] = { { { 0x0000ffff, 0x00af9b00 } } },
52 [GDT_ENTRY_KERNEL_DS
] = { { { 0x0000ffff, 0x00cf9300 } } },
53 [GDT_ENTRY_DEFAULT_USER32_CS
] = { { { 0x0000ffff, 0x00cffb00 } } },
54 [GDT_ENTRY_DEFAULT_USER_DS
] = { { { 0x0000ffff, 0x00cff300 } } },
55 [GDT_ENTRY_DEFAULT_USER_CS
] = { { { 0x0000ffff, 0x00affb00 } } },
58 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page
, gdt_page
) = { .gdt
= {
59 [GDT_ENTRY_KERNEL_CS
] = { { { 0x0000ffff, 0x00cf9a00 } } },
60 [GDT_ENTRY_KERNEL_DS
] = { { { 0x0000ffff, 0x00cf9200 } } },
61 [GDT_ENTRY_DEFAULT_USER_CS
] = { { { 0x0000ffff, 0x00cffa00 } } },
62 [GDT_ENTRY_DEFAULT_USER_DS
] = { { { 0x0000ffff, 0x00cff200 } } },
64 * Segments used for calling PnP BIOS have byte granularity.
65 * They code segments and data segments have fixed 64k limits,
66 * the transfer segment sizes are set at run time.
69 [GDT_ENTRY_PNPBIOS_CS32
] = { { { 0x0000ffff, 0x00409a00 } } },
71 [GDT_ENTRY_PNPBIOS_CS16
] = { { { 0x0000ffff, 0x00009a00 } } },
73 [GDT_ENTRY_PNPBIOS_DS
] = { { { 0x0000ffff, 0x00009200 } } },
75 [GDT_ENTRY_PNPBIOS_TS1
] = { { { 0x00000000, 0x00009200 } } },
77 [GDT_ENTRY_PNPBIOS_TS2
] = { { { 0x00000000, 0x00009200 } } },
79 * The APM segments have byte granularity and their bases
80 * are set at run time. All have 64k limits.
83 [GDT_ENTRY_APMBIOS_BASE
] = { { { 0x0000ffff, 0x00409a00 } } },
85 [GDT_ENTRY_APMBIOS_BASE
+1] = { { { 0x0000ffff, 0x00009a00 } } },
87 [GDT_ENTRY_APMBIOS_BASE
+2] = { { { 0x0000ffff, 0x00409200 } } },
89 [GDT_ENTRY_ESPFIX_SS
] = { { { 0x00000000, 0x00c09200 } } },
90 [GDT_ENTRY_PERCPU
] = { { { 0x00000000, 0x00000000 } } },
93 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page
);
96 static int cachesize_override __cpuinitdata
= -1;
97 static int disable_x86_serial_nr __cpuinitdata
= 1;
99 static int __init
cachesize_setup(char *str
)
101 get_option(&str
, &cachesize_override
);
104 __setup("cachesize=", cachesize_setup
);
107 * Naming convention should be: <Name> [(<Codename>)]
108 * This table only is used unless init_<vendor>() below doesn't set it;
109 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
113 /* Look up CPU names by table lookup. */
114 static char __cpuinit
*table_lookup_model(struct cpuinfo_x86
*c
)
116 struct cpu_model_info
*info
;
118 if (c
->x86_model
>= 16)
119 return NULL
; /* Range check */
124 info
= this_cpu
->c_models
;
126 while (info
&& info
->family
) {
127 if (info
->family
== c
->x86
)
128 return info
->model_names
[c
->x86_model
];
131 return NULL
; /* Not found */
134 static int __init
x86_fxsr_setup(char *s
)
136 setup_clear_cpu_cap(X86_FEATURE_FXSR
);
137 setup_clear_cpu_cap(X86_FEATURE_XMM
);
140 __setup("nofxsr", x86_fxsr_setup
);
142 static int __init
x86_sep_setup(char *s
)
144 setup_clear_cpu_cap(X86_FEATURE_SEP
);
147 __setup("nosep", x86_sep_setup
);
149 /* Standard macro to see if a specific flag is changeable */
150 static inline int flag_is_changeable_p(u32 flag
)
164 : "=&r" (f1
), "=&r" (f2
)
167 return ((f1
^f2
) & flag
) != 0;
170 /* Probe for the CPUID instruction */
171 static int __cpuinit
have_cpuid_p(void)
173 return flag_is_changeable_p(X86_EFLAGS_ID
);
176 static void __cpuinit
squash_the_stupid_serial_number(struct cpuinfo_x86
*c
)
178 if (cpu_has(c
, X86_FEATURE_PN
) && disable_x86_serial_nr
) {
179 /* Disable processor serial number */
180 unsigned long lo
, hi
;
181 rdmsr(MSR_IA32_BBL_CR_CTL
, lo
, hi
);
183 wrmsr(MSR_IA32_BBL_CR_CTL
, lo
, hi
);
184 printk(KERN_NOTICE
"CPU serial number disabled.\n");
185 clear_cpu_cap(c
, X86_FEATURE_PN
);
187 /* Disabling the serial number may affect the cpuid level */
188 c
->cpuid_level
= cpuid_eax(0);
192 static int __init
x86_serial_nr_setup(char *s
)
194 disable_x86_serial_nr
= 0;
197 __setup("serialnumber", x86_serial_nr_setup
);
199 /* Probe for the CPUID instruction */
200 static inline int have_cpuid_p(void)
206 __u32 cleared_cpu_caps
[NCAPINTS
] __cpuinitdata
;
208 /* Current gdt points %fs at the "master" per-cpu area: after this,
209 * it's on the real one. */
210 void switch_to_new_gdt(void)
212 struct desc_ptr gdt_descr
;
214 gdt_descr
.address
= (long)get_cpu_gdt_table(smp_processor_id());
215 gdt_descr
.size
= GDT_SIZE
- 1;
216 load_gdt(&gdt_descr
);
218 asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU
) : "memory");
222 static struct cpu_dev
*cpu_devs
[X86_VENDOR_NUM
] = {};
224 static void __cpuinit
default_init(struct cpuinfo_x86
*c
)
227 display_cacheinfo(c
);
229 /* Not much we can do here... */
230 /* Check if at least it has cpuid */
231 if (c
->cpuid_level
== -1) {
232 /* No cpuid. It must be an ancient CPU */
234 strcpy(c
->x86_model_id
, "486");
235 else if (c
->x86
== 3)
236 strcpy(c
->x86_model_id
, "386");
241 static struct cpu_dev __cpuinitdata default_cpu
= {
242 .c_init
= default_init
,
243 .c_vendor
= "Unknown",
244 .c_x86_vendor
= X86_VENDOR_UNKNOWN
,
247 int __cpuinit
get_model_name(struct cpuinfo_x86
*c
)
252 if (c
->extended_cpuid_level
< 0x80000004)
255 v
= (unsigned int *) c
->x86_model_id
;
256 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
257 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
258 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
259 c
->x86_model_id
[48] = 0;
261 /* Intel chips right-justify this string for some dumb reason;
262 undo that brain damage */
263 p
= q
= &c
->x86_model_id
[0];
269 while (q
<= &c
->x86_model_id
[48])
270 *q
++ = '\0'; /* Zero-pad the rest */
277 void __cpuinit
display_cacheinfo(struct cpuinfo_x86
*c
)
279 unsigned int n
, dummy
, ebx
, ecx
, edx
, l2size
;
281 n
= c
->extended_cpuid_level
;
283 if (n
>= 0x80000005) {
284 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
285 printk(KERN_INFO
"CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
286 edx
>>24, edx
&0xFF, ecx
>>24, ecx
&0xFF);
287 c
->x86_cache_size
= (ecx
>>24) + (edx
>>24);
289 /* On K8 L1 TLB is inclusive, so don't count it */
294 if (n
< 0x80000006) /* Some chips just has a large L1. */
297 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
301 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
304 /* do processor-specific cache resizing */
305 if (this_cpu
->c_size_cache
)
306 l2size
= this_cpu
->c_size_cache(c
, l2size
);
308 /* Allow user to override all this if necessary. */
309 if (cachesize_override
!= -1)
310 l2size
= cachesize_override
;
313 return; /* Again, no L2 cache is possible */
316 c
->x86_cache_size
= l2size
;
318 printk(KERN_INFO
"CPU: L2 Cache: %dK (%d bytes/line)\n",
322 void __cpuinit
detect_ht(struct cpuinfo_x86
*c
)
325 u32 eax
, ebx
, ecx
, edx
;
326 int index_msb
, core_bits
;
328 if (!cpu_has(c
, X86_FEATURE_HT
))
330 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
333 if (cpu_has(c
, X86_FEATURE_XTOPOLOGY
))
336 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
338 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
340 if (smp_num_siblings
== 1) {
341 printk(KERN_INFO
"CPU: Hyper-Threading is disabled\n");
342 } else if (smp_num_siblings
> 1) {
344 if (smp_num_siblings
> NR_CPUS
) {
345 printk(KERN_WARNING
"CPU: Unsupported number of siblings %d",
347 smp_num_siblings
= 1;
351 index_msb
= get_count_order(smp_num_siblings
);
352 c
->phys_proc_id
= phys_pkg_id(index_msb
);
354 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
356 index_msb
= get_count_order(smp_num_siblings
);
358 core_bits
= get_count_order(c
->x86_max_cores
);
360 c
->cpu_core_id
= phys_pkg_id(index_msb
) &
361 ((1 << core_bits
) - 1);
365 if ((c
->x86_max_cores
* smp_num_siblings
) > 1) {
366 printk(KERN_INFO
"CPU: Physical Processor ID: %d\n",
368 printk(KERN_INFO
"CPU: Processor Core ID: %d\n",
374 static void __cpuinit
get_cpu_vendor(struct cpuinfo_x86
*c
)
376 char *v
= c
->x86_vendor_id
;
380 for (i
= 0; i
< X86_VENDOR_NUM
; i
++) {
384 if (!strcmp(v
, cpu_devs
[i
]->c_ident
[0]) ||
385 (cpu_devs
[i
]->c_ident
[1] &&
386 !strcmp(v
, cpu_devs
[i
]->c_ident
[1]))) {
387 this_cpu
= cpu_devs
[i
];
388 c
->x86_vendor
= this_cpu
->c_x86_vendor
;
395 printk(KERN_ERR
"CPU: Vendor unknown, using generic init.\n");
396 printk(KERN_ERR
"CPU: Your system may be unstable.\n");
399 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
400 this_cpu
= &default_cpu
;
403 void __cpuinit
cpu_detect(struct cpuinfo_x86
*c
)
405 /* Get vendor name */
406 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
407 (unsigned int *)&c
->x86_vendor_id
[0],
408 (unsigned int *)&c
->x86_vendor_id
[8],
409 (unsigned int *)&c
->x86_vendor_id
[4]);
412 /* Intel-defined flags: level 0x00000001 */
413 if (c
->cpuid_level
>= 0x00000001) {
414 u32 junk
, tfms
, cap0
, misc
;
415 cpuid(0x00000001, &tfms
, &misc
, &junk
, &cap0
);
416 c
->x86
= (tfms
>> 8) & 0xf;
417 c
->x86_model
= (tfms
>> 4) & 0xf;
418 c
->x86_mask
= tfms
& 0xf;
420 c
->x86
+= (tfms
>> 20) & 0xff;
422 c
->x86_model
+= ((tfms
>> 16) & 0xf) << 4;
423 if (cap0
& (1<<19)) {
424 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
425 c
->x86_cache_alignment
= c
->x86_clflush_size
;
431 static void __cpuinit
get_cpu_cap(struct cpuinfo_x86
*c
)
436 /* Intel-defined flags: level 0x00000001 */
437 if (c
->cpuid_level
>= 0x00000001) {
438 u32 capability
, excap
;
440 cpuid(0x00000001, &tfms
, &ebx
, &excap
, &capability
);
441 c
->x86_capability
[0] = capability
;
442 c
->x86_capability
[4] = excap
;
445 /* AMD-defined flags: level 0x80000001 */
446 xlvl
= cpuid_eax(0x80000000);
447 c
->extended_cpuid_level
= xlvl
;
448 if ((xlvl
& 0xffff0000) == 0x80000000) {
449 if (xlvl
>= 0x80000001) {
450 c
->x86_capability
[1] = cpuid_edx(0x80000001);
451 c
->x86_capability
[6] = cpuid_ecx(0x80000001);
455 /* Transmeta-defined flags: level 0x80860001 */
456 xlvl
= cpuid_eax(0x80860000);
457 if ((xlvl
& 0xffff0000) == 0x80860000) {
458 /* Don't set x86_cpuid_level here for now to not confuse. */
459 if (xlvl
>= 0x80860001)
460 c
->x86_capability
[2] = cpuid_edx(0x80860001);
463 if (c
->extended_cpuid_level
>= 0x80000007)
464 c
->x86_power
= cpuid_edx(0x80000007);
466 if (c
->extended_cpuid_level
>= 0x80000008) {
467 u32 eax
= cpuid_eax(0x80000008);
469 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
470 c
->x86_phys_bits
= eax
& 0xff;
474 /* Do some early cpuid on the boot CPU to get some parameter that are
475 needed before check_bugs. Everything advanced is in identify_cpu
477 static void __init
early_identify_cpu(struct cpuinfo_x86
*c
)
480 c
->x86_clflush_size
= 64;
481 c
->x86_cache_alignment
= c
->x86_clflush_size
;
483 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
485 c
->extended_cpuid_level
= 0;
493 if (this_cpu
->c_early_init
)
494 this_cpu
->c_early_init(c
);
496 validate_pat_support(c
);
499 void __init
early_cpu_init(void)
501 struct cpu_dev
**cdev
;
504 printk("KERNEL supported cpus:\n");
505 for (cdev
= __x86_cpu_dev_start
; cdev
< __x86_cpu_dev_end
; cdev
++) {
506 struct cpu_dev
*cpudev
= *cdev
;
509 if (count
>= X86_VENDOR_NUM
)
511 cpu_devs
[count
] = cpudev
;
514 for (j
= 0; j
< 2; j
++) {
515 if (!cpudev
->c_ident
[j
])
517 printk(" %s %s\n", cpudev
->c_vendor
,
522 early_identify_cpu(&boot_cpu_data
);
526 * The NOPL instruction is supposed to exist on all CPUs with
527 * family >= 6, unfortunately, that's not true in practice because
528 * of early VIA chips and (more importantly) broken virtualizers that
529 * are not easy to detect. Hence, probe for it based on first
532 * Note: no 64-bit chip is known to lack these, but put the code here
533 * for consistency with 32 bits, and to make it utterly trivial to
534 * diagnose the problem should it ever surface.
536 static void __cpuinit
detect_nopl(struct cpuinfo_x86
*c
)
538 const u32 nopl_signature
= 0x888c53b1; /* Random number */
539 u32 has_nopl
= nopl_signature
;
541 clear_cpu_cap(c
, X86_FEATURE_NOPL
);
544 "1: .byte 0x0f,0x1f,0xc0\n" /* nopl %eax */
546 " .section .fixup,\"ax\"\n"
553 if (has_nopl
== nopl_signature
)
554 set_cpu_cap(c
, X86_FEATURE_NOPL
);
558 static void __cpuinit
generic_identify(struct cpuinfo_x86
*c
)
560 c
->extended_cpuid_level
= 0;
568 c
->initial_apicid
= (cpuid_ebx(1) >> 24) & 0xff;
570 c
->phys_proc_id
= c
->initial_apicid
;
573 if (c
->extended_cpuid_level
>= 0x80000004)
574 get_model_name(c
); /* Default name */
576 init_scattered_cpuid_features(c
);
581 * This does the hard work of actually picking apart the CPU stuff...
583 static void __cpuinit
identify_cpu(struct cpuinfo_x86
*c
)
587 c
->loops_per_jiffy
= loops_per_jiffy
;
588 c
->x86_cache_size
= -1;
589 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
590 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
591 c
->x86_vendor_id
[0] = '\0'; /* Unset */
592 c
->x86_model_id
[0] = '\0'; /* Unset */
593 c
->x86_max_cores
= 1;
594 c
->x86_coreid_bits
= 0;
595 c
->x86_clflush_size
= 64;
596 c
->x86_cache_alignment
= c
->x86_clflush_size
;
597 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
601 c
->apicid
= phys_pkg_id(0);
604 * Vendor-specific initialization. In this section we
605 * canonicalize the feature flags, meaning if there are
606 * features a certain CPU supports which CPUID doesn't
607 * tell us, CPUID claiming incorrect flags, or other bugs,
608 * we handle them here.
610 * At the end of this section, c->x86_capability better
611 * indicate the features this CPU genuinely supports!
613 if (this_cpu
->c_init
)
619 * On SMP, boot_cpu_data holds the common feature set between
620 * all CPUs; so make sure that we indicate which features are
621 * common between the CPUs. The first time this routine gets
622 * executed, c == &boot_cpu_data.
624 if (c
!= &boot_cpu_data
) {
625 /* AND the already accumulated flags with these */
626 for (i
= 0; i
< NCAPINTS
; i
++)
627 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
630 /* Clear all flags overriden by options */
631 for (i
= 0; i
< NCAPINTS
; i
++)
632 c
->x86_capability
[i
] &= ~cleared_cpu_caps
[i
];
634 #ifdef CONFIG_X86_MCE
637 select_idle_routine(c
);
640 numa_add_cpu(smp_processor_id());
645 void __init
identify_boot_cpu(void)
647 identify_cpu(&boot_cpu_data
);
650 void __cpuinit
identify_secondary_cpu(struct cpuinfo_x86
*c
)
652 BUG_ON(c
== &boot_cpu_data
);
662 static struct msr_range msr_range_array
[] __cpuinitdata
= {
663 { 0x00000000, 0x00000418},
664 { 0xc0000000, 0xc000040b},
665 { 0xc0010000, 0xc0010142},
666 { 0xc0011000, 0xc001103b},
669 static void __cpuinit
print_cpu_msr(void)
674 unsigned index_min
, index_max
;
676 for (i
= 0; i
< ARRAY_SIZE(msr_range_array
); i
++) {
677 index_min
= msr_range_array
[i
].min
;
678 index_max
= msr_range_array
[i
].max
;
679 for (index
= index_min
; index
< index_max
; index
++) {
680 if (rdmsrl_amd_safe(index
, &val
))
682 printk(KERN_INFO
" MSR%08x: %016llx\n", index
, val
);
687 static int show_msr __cpuinitdata
;
688 static __init
int setup_show_msr(char *arg
)
692 get_option(&arg
, &num
);
698 __setup("show_msr=", setup_show_msr
);
700 static __init
int setup_noclflush(char *arg
)
702 setup_clear_cpu_cap(X86_FEATURE_CLFLSH
);
705 __setup("noclflush", setup_noclflush
);
707 void __cpuinit
print_cpu_info(struct cpuinfo_x86
*c
)
709 if (c
->x86_model_id
[0])
710 printk(KERN_CONT
"%s", c
->x86_model_id
);
712 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
713 printk(KERN_CONT
" stepping %02x\n", c
->x86_mask
);
715 printk(KERN_CONT
"\n");
718 if (c
->cpu_index
< show_msr
)
726 static __init
int setup_disablecpuid(char *arg
)
729 if (get_option(&arg
, &bit
) && bit
< NCAPINTS
*32)
730 setup_clear_cpu_cap(bit
);
735 __setup("clearcpuid=", setup_disablecpuid
);
737 cpumask_t cpu_initialized __cpuinitdata
= CPU_MASK_NONE
;
740 struct x8664_pda
**_cpu_pda __read_mostly
;
741 EXPORT_SYMBOL(_cpu_pda
);
743 struct desc_ptr idt_descr
= { 256 * 16 - 1, (unsigned long) idt_table
};
745 char boot_cpu_stack
[IRQSTACKSIZE
] __page_aligned_bss
;
747 unsigned long __supported_pte_mask __read_mostly
= ~0UL;
748 EXPORT_SYMBOL_GPL(__supported_pte_mask
);
750 static int do_not_nx __cpuinitdata
;
753 Control non executable mappings for 64bit processes.
758 static int __init
nonx_setup(char *str
)
762 if (!strncmp(str
, "on", 2)) {
763 __supported_pte_mask
|= _PAGE_NX
;
765 } else if (!strncmp(str
, "off", 3)) {
767 __supported_pte_mask
&= ~_PAGE_NX
;
771 early_param("noexec", nonx_setup
);
773 int force_personality32
;
776 Control non executable heap for 32bit processes.
777 To control the stack too use noexec=off
779 on PROT_READ does not imply PROT_EXEC for 32bit processes (default)
780 off PROT_READ implies PROT_EXEC
782 static int __init
nonx32_setup(char *str
)
784 if (!strcmp(str
, "on"))
785 force_personality32
&= ~READ_IMPLIES_EXEC
;
786 else if (!strcmp(str
, "off"))
787 force_personality32
|= READ_IMPLIES_EXEC
;
790 __setup("noexec32=", nonx32_setup
);
792 void pda_init(int cpu
)
794 struct x8664_pda
*pda
= cpu_pda(cpu
);
796 /* Setup up data that may be needed in __get_free_pages early */
799 /* Memory clobbers used to order PDA accessed */
801 wrmsrl(MSR_GS_BASE
, pda
);
804 pda
->cpunumber
= cpu
;
806 pda
->kernelstack
= (unsigned long)stack_thread_info() -
807 PDA_STACKOFFSET
+ THREAD_SIZE
;
808 pda
->active_mm
= &init_mm
;
812 /* others are initialized in smpboot.c */
813 pda
->pcurrent
= &init_task
;
814 pda
->irqstackptr
= boot_cpu_stack
;
815 pda
->irqstackptr
+= IRQSTACKSIZE
- 64;
817 if (!pda
->irqstackptr
) {
818 pda
->irqstackptr
= (char *)
819 __get_free_pages(GFP_ATOMIC
, IRQSTACK_ORDER
);
820 if (!pda
->irqstackptr
)
821 panic("cannot allocate irqstack for cpu %d",
823 pda
->irqstackptr
+= IRQSTACKSIZE
- 64;
826 if (pda
->nodenumber
== 0 && cpu_to_node(cpu
) != NUMA_NO_NODE
)
827 pda
->nodenumber
= cpu_to_node(cpu
);
831 char boot_exception_stacks
[(N_EXCEPTION_STACKS
- 1) * EXCEPTION_STKSZ
+
832 DEBUG_STKSZ
] __page_aligned_bss
;
834 extern asmlinkage
void ignore_sysret(void);
836 /* May not be marked __init: used by software suspend */
837 void syscall_init(void)
840 * LSTAR and STAR live in a bit strange symbiosis.
841 * They both write to the same internal register. STAR allows to
842 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
844 wrmsrl(MSR_STAR
, ((u64
)__USER32_CS
)<<48 | ((u64
)__KERNEL_CS
)<<32);
845 wrmsrl(MSR_LSTAR
, system_call
);
846 wrmsrl(MSR_CSTAR
, ignore_sysret
);
848 #ifdef CONFIG_IA32_EMULATION
849 syscall32_cpu_init();
852 /* Flags to clear on syscall */
853 wrmsrl(MSR_SYSCALL_MASK
,
854 X86_EFLAGS_TF
|X86_EFLAGS_DF
|X86_EFLAGS_IF
|X86_EFLAGS_IOPL
);
857 void __cpuinit
check_efer(void)
861 rdmsrl(MSR_EFER
, efer
);
862 if (!(efer
& EFER_NX
) || do_not_nx
)
863 __supported_pte_mask
&= ~_PAGE_NX
;
866 unsigned long kernel_eflags
;
869 * Copies of the original ist values from the tss are only accessed during
870 * debugging, no special alignment required.
872 DEFINE_PER_CPU(struct orig_ist
, orig_ist
);
876 /* Make sure %fs is initialized properly in idle threads */
877 struct pt_regs
* __cpuinit
idle_regs(struct pt_regs
*regs
)
879 memset(regs
, 0, sizeof(struct pt_regs
));
880 regs
->fs
= __KERNEL_PERCPU
;
886 * cpu_init() initializes state that is per-CPU. Some data is already
887 * initialized (naturally) in the bootstrap process, such as the GDT
888 * and IDT. We reload them nevertheless, this function acts as a
889 * 'CPU state barrier', nothing should get across.
890 * A lot of state is already set up in PDA init for 64 bit
893 void __cpuinit
cpu_init(void)
895 int cpu
= stack_smp_processor_id();
896 struct tss_struct
*t
= &per_cpu(init_tss
, cpu
);
897 struct orig_ist
*orig_ist
= &per_cpu(orig_ist
, cpu
);
899 char *estacks
= NULL
;
900 struct task_struct
*me
;
903 /* CPU 0 is initialised in head64.c */
907 estacks
= boot_exception_stacks
;
911 if (cpu_test_and_set(cpu
, cpu_initialized
))
912 panic("CPU#%d already initialized!\n", cpu
);
914 printk(KERN_INFO
"Initializing CPU#%d\n", cpu
);
916 clear_in_cr4(X86_CR4_VME
|X86_CR4_PVI
|X86_CR4_TSD
|X86_CR4_DE
);
919 * Initialize the per-CPU GDT with the boot GDT,
920 * and set up the GDT descriptor:
924 load_idt((const struct desc_ptr
*)&idt_descr
);
926 memset(me
->thread
.tls_array
, 0, GDT_ENTRY_TLS_ENTRIES
* 8);
929 wrmsrl(MSR_FS_BASE
, 0);
930 wrmsrl(MSR_KERNEL_GS_BASE
, 0);
934 if (cpu
!= 0 && x2apic
)
938 * set up and load the per-CPU TSS
940 if (!orig_ist
->ist
[0]) {
941 static const unsigned int order
[N_EXCEPTION_STACKS
] = {
942 [0 ... N_EXCEPTION_STACKS
- 1] = EXCEPTION_STACK_ORDER
,
943 [DEBUG_STACK
- 1] = DEBUG_STACK_ORDER
945 for (v
= 0; v
< N_EXCEPTION_STACKS
; v
++) {
947 estacks
= (char *)__get_free_pages(GFP_ATOMIC
, order
[v
]);
949 panic("Cannot allocate exception "
950 "stack %ld %d\n", v
, cpu
);
952 estacks
+= PAGE_SIZE
<< order
[v
];
953 orig_ist
->ist
[v
] = t
->x86_tss
.ist
[v
] =
954 (unsigned long)estacks
;
958 t
->x86_tss
.io_bitmap_base
= offsetof(struct tss_struct
, io_bitmap
);
960 * <= is required because the CPU will access up to
961 * 8 bits beyond the end of the IO permission bitmap.
963 for (i
= 0; i
<= IO_BITMAP_LONGS
; i
++)
964 t
->io_bitmap
[i
] = ~0UL;
966 atomic_inc(&init_mm
.mm_count
);
967 me
->active_mm
= &init_mm
;
970 enter_lazy_tlb(&init_mm
, me
);
972 load_sp0(t
, ¤t
->thread
);
973 set_tss_desc(cpu
, t
);
975 load_LDT(&init_mm
.context
);
979 * If the kgdb is connected no debug regs should be altered. This
980 * is only applicable when KGDB and a KGDB I/O module are built
981 * into the kernel and you are using early debugging with
982 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
984 if (kgdb_connected
&& arch_kgdb_ops
.correct_hw_break
)
985 arch_kgdb_ops
.correct_hw_break();
989 * Clear all 6 debug registers:
992 set_debugreg(0UL, 0);
993 set_debugreg(0UL, 1);
994 set_debugreg(0UL, 2);
995 set_debugreg(0UL, 3);
996 set_debugreg(0UL, 6);
997 set_debugreg(0UL, 7);
999 /* If the kgdb is connected no debug regs should be altered. */
1005 raw_local_save_flags(kernel_eflags
);
1013 void __cpuinit
cpu_init(void)
1015 int cpu
= smp_processor_id();
1016 struct task_struct
*curr
= current
;
1017 struct tss_struct
*t
= &per_cpu(init_tss
, cpu
);
1018 struct thread_struct
*thread
= &curr
->thread
;
1020 if (cpu_test_and_set(cpu
, cpu_initialized
)) {
1021 printk(KERN_WARNING
"CPU#%d already initialized!\n", cpu
);
1022 for (;;) local_irq_enable();
1025 printk(KERN_INFO
"Initializing CPU#%d\n", cpu
);
1027 if (cpu_has_vme
|| cpu_has_tsc
|| cpu_has_de
)
1028 clear_in_cr4(X86_CR4_VME
|X86_CR4_PVI
|X86_CR4_TSD
|X86_CR4_DE
);
1030 load_idt(&idt_descr
);
1031 switch_to_new_gdt();
1034 * Set up and load the per-CPU TSS and LDT
1036 atomic_inc(&init_mm
.mm_count
);
1037 curr
->active_mm
= &init_mm
;
1040 enter_lazy_tlb(&init_mm
, curr
);
1042 load_sp0(t
, thread
);
1043 set_tss_desc(cpu
, t
);
1045 load_LDT(&init_mm
.context
);
1047 #ifdef CONFIG_DOUBLEFAULT
1048 /* Set up doublefault TSS pointer in the GDT */
1049 __set_tss_desc(cpu
, GDT_ENTRY_DOUBLEFAULT_TSS
, &doublefault_tss
);
1053 asm volatile ("mov %0, %%gs" : : "r" (0));
1055 /* Clear all 6 debug registers: */
1064 * Force FPU initialization:
1067 current_thread_info()->status
= TS_XSAVE
;
1069 current_thread_info()->status
= 0;
1071 mxcsr_feature_mask_init();
1074 * Boot processor to setup the FP and extended state context info.
1076 if (!smp_processor_id())
1077 init_thread_xstate();
1082 #ifdef CONFIG_HOTPLUG_CPU
1083 void __cpuinit
cpu_uninit(void)
1085 int cpu
= raw_smp_processor_id();
1086 cpu_clear(cpu
, cpu_initialized
);
1088 /* lazy TLB state */
1089 per_cpu(cpu_tlbstate
, cpu
).state
= 0;
1090 per_cpu(cpu_tlbstate
, cpu
).active_mm
= &init_mm
;