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>
20 #include <asm/perf_counter.h>
26 #include <asm/cpumask.h>
29 #ifdef CONFIG_X86_LOCAL_APIC
30 #include <asm/uv/uv.h>
33 #include <asm/pgtable.h>
34 #include <asm/processor.h>
36 #include <asm/atomic.h>
37 #include <asm/proto.h>
38 #include <asm/sections.h>
39 #include <asm/setup.h>
40 #include <asm/hypervisor.h>
41 #include <asm/stackprotector.h>
47 /* all of these masks are initialized in setup_cpu_local_masks() */
48 cpumask_var_t cpu_callin_mask
;
49 cpumask_var_t cpu_callout_mask
;
50 cpumask_var_t cpu_initialized_mask
;
52 /* representing cpus for which sibling maps can be computed */
53 cpumask_var_t cpu_sibling_setup_mask
;
55 /* correctly size the local cpu masks */
56 void __init
setup_cpu_local_masks(void)
58 alloc_bootmem_cpumask_var(&cpu_initialized_mask
);
59 alloc_bootmem_cpumask_var(&cpu_callin_mask
);
60 alloc_bootmem_cpumask_var(&cpu_callout_mask
);
61 alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask
);
64 #else /* CONFIG_X86_32 */
66 cpumask_t cpu_callin_map
;
67 cpumask_t cpu_callout_map
;
68 cpumask_t cpu_initialized
;
69 cpumask_t cpu_sibling_setup_map
;
71 #endif /* CONFIG_X86_32 */
74 static struct cpu_dev
*this_cpu __cpuinitdata
;
76 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page
, gdt_page
) = { .gdt
= {
79 * We need valid kernel segments for data and code in long mode too
80 * IRET will check the segment types kkeil 2000/10/28
81 * Also sysret mandates a special GDT layout
83 * The TLS descriptors are currently at a different place compared to i386.
84 * Hopefully nobody expects them at a fixed place (Wine?)
86 [GDT_ENTRY_KERNEL32_CS
] = { { { 0x0000ffff, 0x00cf9b00 } } },
87 [GDT_ENTRY_KERNEL_CS
] = { { { 0x0000ffff, 0x00af9b00 } } },
88 [GDT_ENTRY_KERNEL_DS
] = { { { 0x0000ffff, 0x00cf9300 } } },
89 [GDT_ENTRY_DEFAULT_USER32_CS
] = { { { 0x0000ffff, 0x00cffb00 } } },
90 [GDT_ENTRY_DEFAULT_USER_DS
] = { { { 0x0000ffff, 0x00cff300 } } },
91 [GDT_ENTRY_DEFAULT_USER_CS
] = { { { 0x0000ffff, 0x00affb00 } } },
93 [GDT_ENTRY_KERNEL_CS
] = { { { 0x0000ffff, 0x00cf9a00 } } },
94 [GDT_ENTRY_KERNEL_DS
] = { { { 0x0000ffff, 0x00cf9200 } } },
95 [GDT_ENTRY_DEFAULT_USER_CS
] = { { { 0x0000ffff, 0x00cffa00 } } },
96 [GDT_ENTRY_DEFAULT_USER_DS
] = { { { 0x0000ffff, 0x00cff200 } } },
98 * Segments used for calling PnP BIOS have byte granularity.
99 * They code segments and data segments have fixed 64k limits,
100 * the transfer segment sizes are set at run time.
103 [GDT_ENTRY_PNPBIOS_CS32
] = { { { 0x0000ffff, 0x00409a00 } } },
105 [GDT_ENTRY_PNPBIOS_CS16
] = { { { 0x0000ffff, 0x00009a00 } } },
107 [GDT_ENTRY_PNPBIOS_DS
] = { { { 0x0000ffff, 0x00009200 } } },
109 [GDT_ENTRY_PNPBIOS_TS1
] = { { { 0x00000000, 0x00009200 } } },
111 [GDT_ENTRY_PNPBIOS_TS2
] = { { { 0x00000000, 0x00009200 } } },
113 * The APM segments have byte granularity and their bases
114 * are set at run time. All have 64k limits.
117 [GDT_ENTRY_APMBIOS_BASE
] = { { { 0x0000ffff, 0x00409a00 } } },
119 [GDT_ENTRY_APMBIOS_BASE
+1] = { { { 0x0000ffff, 0x00009a00 } } },
121 [GDT_ENTRY_APMBIOS_BASE
+2] = { { { 0x0000ffff, 0x00409200 } } },
123 [GDT_ENTRY_ESPFIX_SS
] = { { { 0x00000000, 0x00c09200 } } },
124 [GDT_ENTRY_PERCPU
] = { { { 0x0000ffff, 0x00cf9200 } } },
125 GDT_STACK_CANARY_INIT
128 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page
);
131 static int cachesize_override __cpuinitdata
= -1;
132 static int disable_x86_serial_nr __cpuinitdata
= 1;
134 static int __init
cachesize_setup(char *str
)
136 get_option(&str
, &cachesize_override
);
139 __setup("cachesize=", cachesize_setup
);
141 static int __init
x86_fxsr_setup(char *s
)
143 setup_clear_cpu_cap(X86_FEATURE_FXSR
);
144 setup_clear_cpu_cap(X86_FEATURE_XMM
);
147 __setup("nofxsr", x86_fxsr_setup
);
149 static int __init
x86_sep_setup(char *s
)
151 setup_clear_cpu_cap(X86_FEATURE_SEP
);
154 __setup("nosep", x86_sep_setup
);
156 /* Standard macro to see if a specific flag is changeable */
157 static inline int flag_is_changeable_p(u32 flag
)
162 * Cyrix and IDT cpus allow disabling of CPUID
163 * so the code below may return different results
164 * when it is executed before and after enabling
165 * the CPUID. Add "volatile" to not allow gcc to
166 * optimize the subsequent calls to this function.
168 asm volatile ("pushfl\n\t"
178 : "=&r" (f1
), "=&r" (f2
)
181 return ((f1
^f2
) & flag
) != 0;
184 /* Probe for the CPUID instruction */
185 static int __cpuinit
have_cpuid_p(void)
187 return flag_is_changeable_p(X86_EFLAGS_ID
);
190 static void __cpuinit
squash_the_stupid_serial_number(struct cpuinfo_x86
*c
)
192 if (cpu_has(c
, X86_FEATURE_PN
) && disable_x86_serial_nr
) {
193 /* Disable processor serial number */
194 unsigned long lo
, hi
;
195 rdmsr(MSR_IA32_BBL_CR_CTL
, lo
, hi
);
197 wrmsr(MSR_IA32_BBL_CR_CTL
, lo
, hi
);
198 printk(KERN_NOTICE
"CPU serial number disabled.\n");
199 clear_cpu_cap(c
, X86_FEATURE_PN
);
201 /* Disabling the serial number may affect the cpuid level */
202 c
->cpuid_level
= cpuid_eax(0);
206 static int __init
x86_serial_nr_setup(char *s
)
208 disable_x86_serial_nr
= 0;
211 __setup("serialnumber", x86_serial_nr_setup
);
213 static inline int flag_is_changeable_p(u32 flag
)
217 /* Probe for the CPUID instruction */
218 static inline int have_cpuid_p(void)
222 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86
*c
)
228 * Some CPU features depend on higher CPUID levels, which may not always
229 * be available due to CPUID level capping or broken virtualization
230 * software. Add those features to this table to auto-disable them.
232 struct cpuid_dependent_feature
{
236 static const struct cpuid_dependent_feature __cpuinitconst
237 cpuid_dependent_features
[] = {
238 { X86_FEATURE_MWAIT
, 0x00000005 },
239 { X86_FEATURE_DCA
, 0x00000009 },
240 { X86_FEATURE_XSAVE
, 0x0000000d },
244 static void __cpuinit
filter_cpuid_features(struct cpuinfo_x86
*c
, bool warn
)
246 const struct cpuid_dependent_feature
*df
;
247 for (df
= cpuid_dependent_features
; df
->feature
; df
++) {
249 * Note: cpuid_level is set to -1 if unavailable, but
250 * extended_extended_level is set to 0 if unavailable
251 * and the legitimate extended levels are all negative
252 * when signed; hence the weird messing around with
255 if (cpu_has(c
, df
->feature
) &&
256 ((s32
)df
->level
< 0 ?
257 (u32
)df
->level
> (u32
)c
->extended_cpuid_level
:
258 (s32
)df
->level
> (s32
)c
->cpuid_level
)) {
259 clear_cpu_cap(c
, df
->feature
);
262 "CPU: CPU feature %s disabled "
263 "due to lack of CPUID level 0x%x\n",
264 x86_cap_flags
[df
->feature
],
271 * Naming convention should be: <Name> [(<Codename>)]
272 * This table only is used unless init_<vendor>() below doesn't set it;
273 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
277 /* Look up CPU names by table lookup. */
278 static char __cpuinit
*table_lookup_model(struct cpuinfo_x86
*c
)
280 struct cpu_model_info
*info
;
282 if (c
->x86_model
>= 16)
283 return NULL
; /* Range check */
288 info
= this_cpu
->c_models
;
290 while (info
&& info
->family
) {
291 if (info
->family
== c
->x86
)
292 return info
->model_names
[c
->x86_model
];
295 return NULL
; /* Not found */
298 __u32 cleared_cpu_caps
[NCAPINTS
] __cpuinitdata
;
300 void load_percpu_segment(int cpu
)
303 loadsegment(fs
, __KERNEL_PERCPU
);
306 wrmsrl(MSR_GS_BASE
, (unsigned long)per_cpu(irq_stack_union
.gs_base
, cpu
));
308 load_stack_canary_segment();
311 /* Current gdt points %fs at the "master" per-cpu area: after this,
312 * it's on the real one. */
313 void switch_to_new_gdt(int cpu
)
315 struct desc_ptr gdt_descr
;
317 gdt_descr
.address
= (long)get_cpu_gdt_table(cpu
);
318 gdt_descr
.size
= GDT_SIZE
- 1;
319 load_gdt(&gdt_descr
);
320 /* Reload the per-cpu base */
322 load_percpu_segment(cpu
);
325 static struct cpu_dev
*cpu_devs
[X86_VENDOR_NUM
] = {};
327 static void __cpuinit
default_init(struct cpuinfo_x86
*c
)
330 display_cacheinfo(c
);
332 /* Not much we can do here... */
333 /* Check if at least it has cpuid */
334 if (c
->cpuid_level
== -1) {
335 /* No cpuid. It must be an ancient CPU */
337 strcpy(c
->x86_model_id
, "486");
338 else if (c
->x86
== 3)
339 strcpy(c
->x86_model_id
, "386");
344 static struct cpu_dev __cpuinitdata default_cpu
= {
345 .c_init
= default_init
,
346 .c_vendor
= "Unknown",
347 .c_x86_vendor
= X86_VENDOR_UNKNOWN
,
350 static void __cpuinit
get_model_name(struct cpuinfo_x86
*c
)
355 if (c
->extended_cpuid_level
< 0x80000004)
358 v
= (unsigned int *) c
->x86_model_id
;
359 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
360 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
361 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
362 c
->x86_model_id
[48] = 0;
364 /* Intel chips right-justify this string for some dumb reason;
365 undo that brain damage */
366 p
= q
= &c
->x86_model_id
[0];
372 while (q
<= &c
->x86_model_id
[48])
373 *q
++ = '\0'; /* Zero-pad the rest */
377 void __cpuinit
display_cacheinfo(struct cpuinfo_x86
*c
)
379 unsigned int n
, dummy
, ebx
, ecx
, edx
, l2size
;
381 n
= c
->extended_cpuid_level
;
383 if (n
>= 0x80000005) {
384 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
385 printk(KERN_INFO
"CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
386 edx
>>24, edx
&0xFF, ecx
>>24, ecx
&0xFF);
387 c
->x86_cache_size
= (ecx
>>24) + (edx
>>24);
389 /* On K8 L1 TLB is inclusive, so don't count it */
394 if (n
< 0x80000006) /* Some chips just has a large L1. */
397 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
401 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
403 /* do processor-specific cache resizing */
404 if (this_cpu
->c_size_cache
)
405 l2size
= this_cpu
->c_size_cache(c
, l2size
);
407 /* Allow user to override all this if necessary. */
408 if (cachesize_override
!= -1)
409 l2size
= cachesize_override
;
412 return; /* Again, no L2 cache is possible */
415 c
->x86_cache_size
= l2size
;
417 printk(KERN_INFO
"CPU: L2 Cache: %dK (%d bytes/line)\n",
421 void __cpuinit
detect_ht(struct cpuinfo_x86
*c
)
424 u32 eax
, ebx
, ecx
, edx
;
425 int index_msb
, core_bits
;
427 if (!cpu_has(c
, X86_FEATURE_HT
))
430 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
433 if (cpu_has(c
, X86_FEATURE_XTOPOLOGY
))
436 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
438 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
440 if (smp_num_siblings
== 1) {
441 printk(KERN_INFO
"CPU: Hyper-Threading is disabled\n");
442 } else if (smp_num_siblings
> 1) {
444 if (smp_num_siblings
> nr_cpu_ids
) {
445 printk(KERN_WARNING
"CPU: Unsupported number of siblings %d",
447 smp_num_siblings
= 1;
451 index_msb
= get_count_order(smp_num_siblings
);
452 c
->phys_proc_id
= apic
->phys_pkg_id(c
->initial_apicid
, index_msb
);
454 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
456 index_msb
= get_count_order(smp_num_siblings
);
458 core_bits
= get_count_order(c
->x86_max_cores
);
460 c
->cpu_core_id
= apic
->phys_pkg_id(c
->initial_apicid
, index_msb
) &
461 ((1 << core_bits
) - 1);
465 if ((c
->x86_max_cores
* smp_num_siblings
) > 1) {
466 printk(KERN_INFO
"CPU: Physical Processor ID: %d\n",
468 printk(KERN_INFO
"CPU: Processor Core ID: %d\n",
474 static void __cpuinit
get_cpu_vendor(struct cpuinfo_x86
*c
)
476 char *v
= c
->x86_vendor_id
;
480 for (i
= 0; i
< X86_VENDOR_NUM
; i
++) {
484 if (!strcmp(v
, cpu_devs
[i
]->c_ident
[0]) ||
485 (cpu_devs
[i
]->c_ident
[1] &&
486 !strcmp(v
, cpu_devs
[i
]->c_ident
[1]))) {
487 this_cpu
= cpu_devs
[i
];
488 c
->x86_vendor
= this_cpu
->c_x86_vendor
;
495 printk(KERN_ERR
"CPU: vendor_id '%s' unknown, using generic init.\n", v
);
496 printk(KERN_ERR
"CPU: Your system may be unstable.\n");
499 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
500 this_cpu
= &default_cpu
;
503 void __cpuinit
cpu_detect(struct cpuinfo_x86
*c
)
505 /* Get vendor name */
506 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
507 (unsigned int *)&c
->x86_vendor_id
[0],
508 (unsigned int *)&c
->x86_vendor_id
[8],
509 (unsigned int *)&c
->x86_vendor_id
[4]);
512 /* Intel-defined flags: level 0x00000001 */
513 if (c
->cpuid_level
>= 0x00000001) {
514 u32 junk
, tfms
, cap0
, misc
;
515 cpuid(0x00000001, &tfms
, &misc
, &junk
, &cap0
);
516 c
->x86
= (tfms
>> 8) & 0xf;
517 c
->x86_model
= (tfms
>> 4) & 0xf;
518 c
->x86_mask
= tfms
& 0xf;
520 c
->x86
+= (tfms
>> 20) & 0xff;
522 c
->x86_model
+= ((tfms
>> 16) & 0xf) << 4;
523 if (cap0
& (1<<19)) {
524 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
525 c
->x86_cache_alignment
= c
->x86_clflush_size
;
530 static void __cpuinit
get_cpu_cap(struct cpuinfo_x86
*c
)
535 /* Intel-defined flags: level 0x00000001 */
536 if (c
->cpuid_level
>= 0x00000001) {
537 u32 capability
, excap
;
538 cpuid(0x00000001, &tfms
, &ebx
, &excap
, &capability
);
539 c
->x86_capability
[0] = capability
;
540 c
->x86_capability
[4] = excap
;
543 /* AMD-defined flags: level 0x80000001 */
544 xlvl
= cpuid_eax(0x80000000);
545 c
->extended_cpuid_level
= xlvl
;
546 if ((xlvl
& 0xffff0000) == 0x80000000) {
547 if (xlvl
>= 0x80000001) {
548 c
->x86_capability
[1] = cpuid_edx(0x80000001);
549 c
->x86_capability
[6] = cpuid_ecx(0x80000001);
554 if (c
->extended_cpuid_level
>= 0x80000008) {
555 u32 eax
= cpuid_eax(0x80000008);
557 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
558 c
->x86_phys_bits
= eax
& 0xff;
562 if (c
->extended_cpuid_level
>= 0x80000007)
563 c
->x86_power
= cpuid_edx(0x80000007);
567 static void __cpuinit
identify_cpu_without_cpuid(struct cpuinfo_x86
*c
)
573 * First of all, decide if this is a 486 or higher
574 * It's a 486 if we can modify the AC flag
576 if (flag_is_changeable_p(X86_EFLAGS_AC
))
581 for (i
= 0; i
< X86_VENDOR_NUM
; i
++)
582 if (cpu_devs
[i
] && cpu_devs
[i
]->c_identify
) {
583 c
->x86_vendor_id
[0] = 0;
584 cpu_devs
[i
]->c_identify(c
);
585 if (c
->x86_vendor_id
[0]) {
594 * Do minimum CPU detection early.
595 * Fields really needed: vendor, cpuid_level, family, model, mask,
597 * The others are not touched to avoid unwanted side effects.
599 * WARNING: this function is only called on the BP. Don't add code here
600 * that is supposed to run on all CPUs.
602 static void __init
early_identify_cpu(struct cpuinfo_x86
*c
)
605 c
->x86_clflush_size
= 64;
607 c
->x86_clflush_size
= 32;
609 c
->x86_cache_alignment
= c
->x86_clflush_size
;
611 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
612 c
->extended_cpuid_level
= 0;
615 identify_cpu_without_cpuid(c
);
617 /* cyrix could have cpuid enabled via c_identify()*/
627 if (this_cpu
->c_early_init
)
628 this_cpu
->c_early_init(c
);
631 c
->cpu_index
= boot_cpu_id
;
633 filter_cpuid_features(c
, false);
636 void __init
early_cpu_init(void)
638 struct cpu_dev
**cdev
;
641 printk("KERNEL supported cpus:\n");
642 for (cdev
= __x86_cpu_dev_start
; cdev
< __x86_cpu_dev_end
; cdev
++) {
643 struct cpu_dev
*cpudev
= *cdev
;
646 if (count
>= X86_VENDOR_NUM
)
648 cpu_devs
[count
] = cpudev
;
651 for (j
= 0; j
< 2; j
++) {
652 if (!cpudev
->c_ident
[j
])
654 printk(" %s %s\n", cpudev
->c_vendor
,
659 early_identify_cpu(&boot_cpu_data
);
663 * The NOPL instruction is supposed to exist on all CPUs with
664 * family >= 6; unfortunately, that's not true in practice because
665 * of early VIA chips and (more importantly) broken virtualizers that
666 * are not easy to detect. In the latter case it doesn't even *fail*
667 * reliably, so probing for it doesn't even work. Disable it completely
668 * unless we can find a reliable way to detect all the broken cases.
670 static void __cpuinit
detect_nopl(struct cpuinfo_x86
*c
)
672 clear_cpu_cap(c
, X86_FEATURE_NOPL
);
675 static void __cpuinit
generic_identify(struct cpuinfo_x86
*c
)
677 c
->extended_cpuid_level
= 0;
680 identify_cpu_without_cpuid(c
);
682 /* cyrix could have cpuid enabled via c_identify()*/
692 if (c
->cpuid_level
>= 0x00000001) {
693 c
->initial_apicid
= (cpuid_ebx(1) >> 24) & 0xFF;
695 # ifdef CONFIG_X86_HT
696 c
->apicid
= apic
->phys_pkg_id(c
->initial_apicid
, 0);
698 c
->apicid
= c
->initial_apicid
;
703 c
->phys_proc_id
= c
->initial_apicid
;
707 get_model_name(c
); /* Default name */
709 init_scattered_cpuid_features(c
);
714 * This does the hard work of actually picking apart the CPU stuff...
716 static void __cpuinit
identify_cpu(struct cpuinfo_x86
*c
)
720 c
->loops_per_jiffy
= loops_per_jiffy
;
721 c
->x86_cache_size
= -1;
722 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
723 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
724 c
->x86_vendor_id
[0] = '\0'; /* Unset */
725 c
->x86_model_id
[0] = '\0'; /* Unset */
726 c
->x86_max_cores
= 1;
727 c
->x86_coreid_bits
= 0;
729 c
->x86_clflush_size
= 64;
731 c
->cpuid_level
= -1; /* CPUID not detected */
732 c
->x86_clflush_size
= 32;
734 c
->x86_cache_alignment
= c
->x86_clflush_size
;
735 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
739 if (this_cpu
->c_identify
)
740 this_cpu
->c_identify(c
);
743 c
->apicid
= apic
->phys_pkg_id(c
->initial_apicid
, 0);
747 * Vendor-specific initialization. In this section we
748 * canonicalize the feature flags, meaning if there are
749 * features a certain CPU supports which CPUID doesn't
750 * tell us, CPUID claiming incorrect flags, or other bugs,
751 * we handle them here.
753 * At the end of this section, c->x86_capability better
754 * indicate the features this CPU genuinely supports!
756 if (this_cpu
->c_init
)
759 /* Disable the PN if appropriate */
760 squash_the_stupid_serial_number(c
);
763 * The vendor-specific functions might have changed features. Now
764 * we do "generic changes."
767 /* Filter out anything that depends on CPUID levels we don't have */
768 filter_cpuid_features(c
, true);
770 /* If the model name is still unset, do table lookup. */
771 if (!c
->x86_model_id
[0]) {
773 p
= table_lookup_model(c
);
775 strcpy(c
->x86_model_id
, p
);
778 sprintf(c
->x86_model_id
, "%02x/%02x",
779 c
->x86
, c
->x86_model
);
788 * On SMP, boot_cpu_data holds the common feature set between
789 * all CPUs; so make sure that we indicate which features are
790 * common between the CPUs. The first time this routine gets
791 * executed, c == &boot_cpu_data.
793 if (c
!= &boot_cpu_data
) {
794 /* AND the already accumulated flags with these */
795 for (i
= 0; i
< NCAPINTS
; i
++)
796 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
799 /* Clear all flags overriden by options */
800 for (i
= 0; i
< NCAPINTS
; i
++)
801 c
->x86_capability
[i
] &= ~cleared_cpu_caps
[i
];
803 #ifdef CONFIG_X86_MCE
804 /* Init Machine Check Exception if available. */
808 select_idle_routine(c
);
810 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
811 numa_add_cpu(smp_processor_id());
816 static void vgetcpu_set_mode(void)
818 if (cpu_has(&boot_cpu_data
, X86_FEATURE_RDTSCP
))
819 vgetcpu_mode
= VGETCPU_RDTSCP
;
821 vgetcpu_mode
= VGETCPU_LSL
;
825 void __init
identify_boot_cpu(void)
827 identify_cpu(&boot_cpu_data
);
834 init_hw_perf_counters();
837 void __cpuinit
identify_secondary_cpu(struct cpuinfo_x86
*c
)
839 BUG_ON(c
== &boot_cpu_data
);
852 static struct msr_range msr_range_array
[] __cpuinitdata
= {
853 { 0x00000000, 0x00000418},
854 { 0xc0000000, 0xc000040b},
855 { 0xc0010000, 0xc0010142},
856 { 0xc0011000, 0xc001103b},
859 static void __cpuinit
print_cpu_msr(void)
864 unsigned index_min
, index_max
;
866 for (i
= 0; i
< ARRAY_SIZE(msr_range_array
); i
++) {
867 index_min
= msr_range_array
[i
].min
;
868 index_max
= msr_range_array
[i
].max
;
869 for (index
= index_min
; index
< index_max
; index
++) {
870 if (rdmsrl_amd_safe(index
, &val
))
872 printk(KERN_INFO
" MSR%08x: %016llx\n", index
, val
);
877 static int show_msr __cpuinitdata
;
878 static __init
int setup_show_msr(char *arg
)
882 get_option(&arg
, &num
);
888 __setup("show_msr=", setup_show_msr
);
890 static __init
int setup_noclflush(char *arg
)
892 setup_clear_cpu_cap(X86_FEATURE_CLFLSH
);
895 __setup("noclflush", setup_noclflush
);
897 void __cpuinit
print_cpu_info(struct cpuinfo_x86
*c
)
901 if (c
->x86_vendor
< X86_VENDOR_NUM
)
902 vendor
= this_cpu
->c_vendor
;
903 else if (c
->cpuid_level
>= 0)
904 vendor
= c
->x86_vendor_id
;
906 if (vendor
&& !strstr(c
->x86_model_id
, vendor
))
907 printk(KERN_CONT
"%s ", vendor
);
909 if (c
->x86_model_id
[0])
910 printk(KERN_CONT
"%s", c
->x86_model_id
);
912 printk(KERN_CONT
"%d86", c
->x86
);
914 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
915 printk(KERN_CONT
" stepping %02x\n", c
->x86_mask
);
917 printk(KERN_CONT
"\n");
920 if (c
->cpu_index
< show_msr
)
928 static __init
int setup_disablecpuid(char *arg
)
931 if (get_option(&arg
, &bit
) && bit
< NCAPINTS
*32)
932 setup_clear_cpu_cap(bit
);
937 __setup("clearcpuid=", setup_disablecpuid
);
940 struct desc_ptr idt_descr
= { 256 * 16 - 1, (unsigned long) idt_table
};
942 DEFINE_PER_CPU_FIRST(union irq_stack_union
,
943 irq_stack_union
) __aligned(PAGE_SIZE
);
944 DEFINE_PER_CPU(char *, irq_stack_ptr
) =
945 init_per_cpu_var(irq_stack_union
.irq_stack
) + IRQ_STACK_SIZE
- 64;
947 DEFINE_PER_CPU(unsigned long, kernel_stack
) =
948 (unsigned long)&init_thread_union
- KERNEL_STACK_OFFSET
+ THREAD_SIZE
;
949 EXPORT_PER_CPU_SYMBOL(kernel_stack
);
951 DEFINE_PER_CPU(unsigned int, irq_count
) = -1;
953 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
954 [(N_EXCEPTION_STACKS
- 1) * EXCEPTION_STKSZ
+ DEBUG_STKSZ
])
955 __aligned(PAGE_SIZE
);
957 extern asmlinkage
void ignore_sysret(void);
959 /* May not be marked __init: used by software suspend */
960 void syscall_init(void)
963 * LSTAR and STAR live in a bit strange symbiosis.
964 * They both write to the same internal register. STAR allows to
965 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
967 wrmsrl(MSR_STAR
, ((u64
)__USER32_CS
)<<48 | ((u64
)__KERNEL_CS
)<<32);
968 wrmsrl(MSR_LSTAR
, system_call
);
969 wrmsrl(MSR_CSTAR
, ignore_sysret
);
971 #ifdef CONFIG_IA32_EMULATION
972 syscall32_cpu_init();
975 /* Flags to clear on syscall */
976 wrmsrl(MSR_SYSCALL_MASK
,
977 X86_EFLAGS_TF
|X86_EFLAGS_DF
|X86_EFLAGS_IF
|X86_EFLAGS_IOPL
);
980 unsigned long kernel_eflags
;
983 * Copies of the original ist values from the tss are only accessed during
984 * debugging, no special alignment required.
986 DEFINE_PER_CPU(struct orig_ist
, orig_ist
);
990 #ifdef CONFIG_CC_STACKPROTECTOR
991 DEFINE_PER_CPU(unsigned long, stack_canary
);
994 /* Make sure %fs and %gs are initialized properly in idle threads */
995 struct pt_regs
* __cpuinit
idle_regs(struct pt_regs
*regs
)
997 memset(regs
, 0, sizeof(struct pt_regs
));
998 regs
->fs
= __KERNEL_PERCPU
;
999 regs
->gs
= __KERNEL_STACK_CANARY
;
1005 * cpu_init() initializes state that is per-CPU. Some data is already
1006 * initialized (naturally) in the bootstrap process, such as the GDT
1007 * and IDT. We reload them nevertheless, this function acts as a
1008 * 'CPU state barrier', nothing should get across.
1009 * A lot of state is already set up in PDA init for 64 bit
1011 #ifdef CONFIG_X86_64
1012 void __cpuinit
cpu_init(void)
1014 int cpu
= stack_smp_processor_id();
1015 struct tss_struct
*t
= &per_cpu(init_tss
, cpu
);
1016 struct orig_ist
*orig_ist
= &per_cpu(orig_ist
, cpu
);
1018 struct task_struct
*me
;
1022 if (cpu
!= 0 && percpu_read(node_number
) == 0 &&
1023 cpu_to_node(cpu
) != NUMA_NO_NODE
)
1024 percpu_write(node_number
, cpu_to_node(cpu
));
1029 if (cpumask_test_and_set_cpu(cpu
, cpu_initialized_mask
))
1030 panic("CPU#%d already initialized!\n", cpu
);
1032 printk(KERN_INFO
"Initializing CPU#%d\n", cpu
);
1034 clear_in_cr4(X86_CR4_VME
|X86_CR4_PVI
|X86_CR4_TSD
|X86_CR4_DE
);
1037 * Initialize the per-CPU GDT with the boot GDT,
1038 * and set up the GDT descriptor:
1041 switch_to_new_gdt(cpu
);
1044 load_idt((const struct desc_ptr
*)&idt_descr
);
1046 memset(me
->thread
.tls_array
, 0, GDT_ENTRY_TLS_ENTRIES
* 8);
1049 wrmsrl(MSR_FS_BASE
, 0);
1050 wrmsrl(MSR_KERNEL_GS_BASE
, 0);
1058 * set up and load the per-CPU TSS
1060 if (!orig_ist
->ist
[0]) {
1061 static const unsigned int sizes
[N_EXCEPTION_STACKS
] = {
1062 [0 ... N_EXCEPTION_STACKS
- 1] = EXCEPTION_STKSZ
,
1063 [DEBUG_STACK
- 1] = DEBUG_STKSZ
1065 char *estacks
= per_cpu(exception_stacks
, cpu
);
1066 for (v
= 0; v
< N_EXCEPTION_STACKS
; v
++) {
1067 estacks
+= sizes
[v
];
1068 orig_ist
->ist
[v
] = t
->x86_tss
.ist
[v
] =
1069 (unsigned long)estacks
;
1073 t
->x86_tss
.io_bitmap_base
= offsetof(struct tss_struct
, io_bitmap
);
1075 * <= is required because the CPU will access up to
1076 * 8 bits beyond the end of the IO permission bitmap.
1078 for (i
= 0; i
<= IO_BITMAP_LONGS
; i
++)
1079 t
->io_bitmap
[i
] = ~0UL;
1081 atomic_inc(&init_mm
.mm_count
);
1082 me
->active_mm
= &init_mm
;
1085 enter_lazy_tlb(&init_mm
, me
);
1087 load_sp0(t
, ¤t
->thread
);
1088 set_tss_desc(cpu
, t
);
1090 load_LDT(&init_mm
.context
);
1094 * If the kgdb is connected no debug regs should be altered. This
1095 * is only applicable when KGDB and a KGDB I/O module are built
1096 * into the kernel and you are using early debugging with
1097 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1099 if (kgdb_connected
&& arch_kgdb_ops
.correct_hw_break
)
1100 arch_kgdb_ops
.correct_hw_break();
1105 * Clear all 6 debug registers:
1107 set_debugreg(0UL, 0);
1108 set_debugreg(0UL, 1);
1109 set_debugreg(0UL, 2);
1110 set_debugreg(0UL, 3);
1111 set_debugreg(0UL, 6);
1112 set_debugreg(0UL, 7);
1117 raw_local_save_flags(kernel_eflags
);
1125 void __cpuinit
cpu_init(void)
1127 int cpu
= smp_processor_id();
1128 struct task_struct
*curr
= current
;
1129 struct tss_struct
*t
= &per_cpu(init_tss
, cpu
);
1130 struct thread_struct
*thread
= &curr
->thread
;
1132 if (cpumask_test_and_set_cpu(cpu
, cpu_initialized_mask
)) {
1133 printk(KERN_WARNING
"CPU#%d already initialized!\n", cpu
);
1134 for (;;) local_irq_enable();
1137 printk(KERN_INFO
"Initializing CPU#%d\n", cpu
);
1139 if (cpu_has_vme
|| cpu_has_tsc
|| cpu_has_de
)
1140 clear_in_cr4(X86_CR4_VME
|X86_CR4_PVI
|X86_CR4_TSD
|X86_CR4_DE
);
1142 load_idt(&idt_descr
);
1143 switch_to_new_gdt(cpu
);
1146 * Set up and load the per-CPU TSS and LDT
1148 atomic_inc(&init_mm
.mm_count
);
1149 curr
->active_mm
= &init_mm
;
1152 enter_lazy_tlb(&init_mm
, curr
);
1154 load_sp0(t
, thread
);
1155 set_tss_desc(cpu
, t
);
1157 load_LDT(&init_mm
.context
);
1159 #ifdef CONFIG_DOUBLEFAULT
1160 /* Set up doublefault TSS pointer in the GDT */
1161 __set_tss_desc(cpu
, GDT_ENTRY_DOUBLEFAULT_TSS
, &doublefault_tss
);
1164 /* Clear all 6 debug registers: */
1173 * Force FPU initialization:
1176 current_thread_info()->status
= TS_XSAVE
;
1178 current_thread_info()->status
= 0;
1180 mxcsr_feature_mask_init();
1183 * Boot processor to setup the FP and extended state context info.
1185 if (smp_processor_id() == boot_cpu_id
)
1186 init_thread_xstate();