1 #include <linux/bootmem.h>
2 #include <linux/linkage.h>
3 #include <linux/bitops.h>
4 #include <linux/kernel.h>
5 #include <linux/module.h>
6 #include <linux/percpu.h>
7 #include <linux/string.h>
8 #include <linux/delay.h>
9 #include <linux/sched.h>
10 #include <linux/init.h>
11 #include <linux/kprobes.h>
12 #include <linux/kgdb.h>
13 #include <linux/smp.h>
16 #include <asm/stackprotector.h>
17 #include <asm/perf_event.h>
18 #include <asm/mmu_context.h>
19 #include <asm/archrandom.h>
20 #include <asm/hypervisor.h>
21 #include <asm/processor.h>
22 #include <asm/tlbflush.h>
23 #include <asm/debugreg.h>
24 #include <asm/sections.h>
25 #include <asm/vsyscall.h>
26 #include <linux/topology.h>
27 #include <linux/cpumask.h>
28 #include <asm/pgtable.h>
29 #include <linux/atomic.h>
30 #include <asm/proto.h>
31 #include <asm/setup.h>
35 #include <asm/fpu-internal.h>
37 #include <linux/numa.h>
43 #include <asm/microcode.h>
44 #include <asm/microcode_intel.h>
46 #ifdef CONFIG_X86_LOCAL_APIC
47 #include <asm/uv/uv.h>
52 /* all of these masks are initialized in setup_cpu_local_masks() */
53 cpumask_var_t cpu_initialized_mask
;
54 cpumask_var_t cpu_callout_mask
;
55 cpumask_var_t cpu_callin_mask
;
57 /* representing cpus for which sibling maps can be computed */
58 cpumask_var_t cpu_sibling_setup_mask
;
60 /* correctly size the local cpu masks */
61 void __init
setup_cpu_local_masks(void)
63 alloc_bootmem_cpumask_var(&cpu_initialized_mask
);
64 alloc_bootmem_cpumask_var(&cpu_callin_mask
);
65 alloc_bootmem_cpumask_var(&cpu_callout_mask
);
66 alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask
);
69 static void default_init(struct cpuinfo_x86
*c
)
72 cpu_detect_cache_sizes(c
);
74 /* Not much we can do here... */
75 /* Check if at least it has cpuid */
76 if (c
->cpuid_level
== -1) {
77 /* No cpuid. It must be an ancient CPU */
79 strcpy(c
->x86_model_id
, "486");
81 strcpy(c
->x86_model_id
, "386");
86 static const struct cpu_dev default_cpu
= {
87 .c_init
= default_init
,
88 .c_vendor
= "Unknown",
89 .c_x86_vendor
= X86_VENDOR_UNKNOWN
,
92 static const struct cpu_dev
*this_cpu
= &default_cpu
;
94 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page
, gdt_page
) = { .gdt
= {
97 * We need valid kernel segments for data and code in long mode too
98 * IRET will check the segment types kkeil 2000/10/28
99 * Also sysret mandates a special GDT layout
101 * TLS descriptors are currently at a different place compared to i386.
102 * Hopefully nobody expects them at a fixed place (Wine?)
104 [GDT_ENTRY_KERNEL32_CS
] = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
105 [GDT_ENTRY_KERNEL_CS
] = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
106 [GDT_ENTRY_KERNEL_DS
] = GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
107 [GDT_ENTRY_DEFAULT_USER32_CS
] = GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
108 [GDT_ENTRY_DEFAULT_USER_DS
] = GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
109 [GDT_ENTRY_DEFAULT_USER_CS
] = GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
111 [GDT_ENTRY_KERNEL_CS
] = GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
112 [GDT_ENTRY_KERNEL_DS
] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
113 [GDT_ENTRY_DEFAULT_USER_CS
] = GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
114 [GDT_ENTRY_DEFAULT_USER_DS
] = GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
116 * Segments used for calling PnP BIOS have byte granularity.
117 * They code segments and data segments have fixed 64k limits,
118 * the transfer segment sizes are set at run time.
121 [GDT_ENTRY_PNPBIOS_CS32
] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
123 [GDT_ENTRY_PNPBIOS_CS16
] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
125 [GDT_ENTRY_PNPBIOS_DS
] = GDT_ENTRY_INIT(0x0092, 0, 0xffff),
127 [GDT_ENTRY_PNPBIOS_TS1
] = GDT_ENTRY_INIT(0x0092, 0, 0),
129 [GDT_ENTRY_PNPBIOS_TS2
] = GDT_ENTRY_INIT(0x0092, 0, 0),
131 * The APM segments have byte granularity and their bases
132 * are set at run time. All have 64k limits.
135 [GDT_ENTRY_APMBIOS_BASE
] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
137 [GDT_ENTRY_APMBIOS_BASE
+1] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
139 [GDT_ENTRY_APMBIOS_BASE
+2] = GDT_ENTRY_INIT(0x4092, 0, 0xffff),
141 [GDT_ENTRY_ESPFIX_SS
] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
142 [GDT_ENTRY_PERCPU
] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
143 GDT_STACK_CANARY_INIT
146 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page
);
148 static int __init
x86_xsave_setup(char *s
)
152 setup_clear_cpu_cap(X86_FEATURE_XSAVE
);
153 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT
);
154 setup_clear_cpu_cap(X86_FEATURE_XSAVES
);
155 setup_clear_cpu_cap(X86_FEATURE_AVX
);
156 setup_clear_cpu_cap(X86_FEATURE_AVX2
);
159 __setup("noxsave", x86_xsave_setup
);
161 static int __init
x86_xsaveopt_setup(char *s
)
163 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT
);
166 __setup("noxsaveopt", x86_xsaveopt_setup
);
168 static int __init
x86_xsaves_setup(char *s
)
170 setup_clear_cpu_cap(X86_FEATURE_XSAVES
);
173 __setup("noxsaves", x86_xsaves_setup
);
176 static int cachesize_override
= -1;
177 static int disable_x86_serial_nr
= 1;
179 static int __init
cachesize_setup(char *str
)
181 get_option(&str
, &cachesize_override
);
184 __setup("cachesize=", cachesize_setup
);
186 static int __init
x86_fxsr_setup(char *s
)
188 setup_clear_cpu_cap(X86_FEATURE_FXSR
);
189 setup_clear_cpu_cap(X86_FEATURE_XMM
);
192 __setup("nofxsr", x86_fxsr_setup
);
194 static int __init
x86_sep_setup(char *s
)
196 setup_clear_cpu_cap(X86_FEATURE_SEP
);
199 __setup("nosep", x86_sep_setup
);
201 /* Standard macro to see if a specific flag is changeable */
202 static inline int flag_is_changeable_p(u32 flag
)
207 * Cyrix and IDT cpus allow disabling of CPUID
208 * so the code below may return different results
209 * when it is executed before and after enabling
210 * the CPUID. Add "volatile" to not allow gcc to
211 * optimize the subsequent calls to this function.
213 asm volatile ("pushfl \n\t"
224 : "=&r" (f1
), "=&r" (f2
)
227 return ((f1
^f2
) & flag
) != 0;
230 /* Probe for the CPUID instruction */
231 int have_cpuid_p(void)
233 return flag_is_changeable_p(X86_EFLAGS_ID
);
236 static void squash_the_stupid_serial_number(struct cpuinfo_x86
*c
)
238 unsigned long lo
, hi
;
240 if (!cpu_has(c
, X86_FEATURE_PN
) || !disable_x86_serial_nr
)
243 /* Disable processor serial number: */
245 rdmsr(MSR_IA32_BBL_CR_CTL
, lo
, hi
);
247 wrmsr(MSR_IA32_BBL_CR_CTL
, lo
, hi
);
249 printk(KERN_NOTICE
"CPU serial number disabled.\n");
250 clear_cpu_cap(c
, X86_FEATURE_PN
);
252 /* Disabling the serial number may affect the cpuid level */
253 c
->cpuid_level
= cpuid_eax(0);
256 static int __init
x86_serial_nr_setup(char *s
)
258 disable_x86_serial_nr
= 0;
261 __setup("serialnumber", x86_serial_nr_setup
);
263 static inline int flag_is_changeable_p(u32 flag
)
267 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86
*c
)
272 static __init
int setup_disable_smep(char *arg
)
274 setup_clear_cpu_cap(X86_FEATURE_SMEP
);
277 __setup("nosmep", setup_disable_smep
);
279 static __always_inline
void setup_smep(struct cpuinfo_x86
*c
)
281 if (cpu_has(c
, X86_FEATURE_SMEP
))
282 cr4_set_bits(X86_CR4_SMEP
);
285 static __init
int setup_disable_smap(char *arg
)
287 setup_clear_cpu_cap(X86_FEATURE_SMAP
);
290 __setup("nosmap", setup_disable_smap
);
292 static __always_inline
void setup_smap(struct cpuinfo_x86
*c
)
294 unsigned long eflags
= native_save_fl();
296 /* This should have been cleared long ago */
297 BUG_ON(eflags
& X86_EFLAGS_AC
);
299 if (cpu_has(c
, X86_FEATURE_SMAP
)) {
300 #ifdef CONFIG_X86_SMAP
301 cr4_set_bits(X86_CR4_SMAP
);
303 cr4_clear_bits(X86_CR4_SMAP
);
309 * Some CPU features depend on higher CPUID levels, which may not always
310 * be available due to CPUID level capping or broken virtualization
311 * software. Add those features to this table to auto-disable them.
313 struct cpuid_dependent_feature
{
318 static const struct cpuid_dependent_feature
319 cpuid_dependent_features
[] = {
320 { X86_FEATURE_MWAIT
, 0x00000005 },
321 { X86_FEATURE_DCA
, 0x00000009 },
322 { X86_FEATURE_XSAVE
, 0x0000000d },
326 static void filter_cpuid_features(struct cpuinfo_x86
*c
, bool warn
)
328 const struct cpuid_dependent_feature
*df
;
330 for (df
= cpuid_dependent_features
; df
->feature
; df
++) {
332 if (!cpu_has(c
, df
->feature
))
335 * Note: cpuid_level is set to -1 if unavailable, but
336 * extended_extended_level is set to 0 if unavailable
337 * and the legitimate extended levels are all negative
338 * when signed; hence the weird messing around with
341 if (!((s32
)df
->level
< 0 ?
342 (u32
)df
->level
> (u32
)c
->extended_cpuid_level
:
343 (s32
)df
->level
> (s32
)c
->cpuid_level
))
346 clear_cpu_cap(c
, df
->feature
);
351 "CPU: CPU feature " X86_CAP_FMT
" disabled, no CPUID level 0x%x\n",
352 x86_cap_flag(df
->feature
), df
->level
);
357 * Naming convention should be: <Name> [(<Codename>)]
358 * This table only is used unless init_<vendor>() below doesn't set it;
359 * in particular, if CPUID levels 0x80000002..4 are supported, this
363 /* Look up CPU names by table lookup. */
364 static const char *table_lookup_model(struct cpuinfo_x86
*c
)
367 const struct legacy_cpu_model_info
*info
;
369 if (c
->x86_model
>= 16)
370 return NULL
; /* Range check */
375 info
= this_cpu
->legacy_models
;
377 while (info
->family
) {
378 if (info
->family
== c
->x86
)
379 return info
->model_names
[c
->x86_model
];
383 return NULL
; /* Not found */
386 __u32 cpu_caps_cleared
[NCAPINTS
];
387 __u32 cpu_caps_set
[NCAPINTS
];
389 void load_percpu_segment(int cpu
)
392 loadsegment(fs
, __KERNEL_PERCPU
);
395 wrmsrl(MSR_GS_BASE
, (unsigned long)per_cpu(irq_stack_union
.gs_base
, cpu
));
397 load_stack_canary_segment();
401 * Current gdt points %fs at the "master" per-cpu area: after this,
402 * it's on the real one.
404 void switch_to_new_gdt(int cpu
)
406 struct desc_ptr gdt_descr
;
408 gdt_descr
.address
= (long)get_cpu_gdt_table(cpu
);
409 gdt_descr
.size
= GDT_SIZE
- 1;
410 load_gdt(&gdt_descr
);
411 /* Reload the per-cpu base */
413 load_percpu_segment(cpu
);
416 static const struct cpu_dev
*cpu_devs
[X86_VENDOR_NUM
] = {};
418 static void get_model_name(struct cpuinfo_x86
*c
)
423 if (c
->extended_cpuid_level
< 0x80000004)
426 v
= (unsigned int *)c
->x86_model_id
;
427 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
428 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
429 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
430 c
->x86_model_id
[48] = 0;
433 * Intel chips right-justify this string for some dumb reason;
434 * undo that brain damage:
436 p
= q
= &c
->x86_model_id
[0];
442 while (q
<= &c
->x86_model_id
[48])
443 *q
++ = '\0'; /* Zero-pad the rest */
447 void cpu_detect_cache_sizes(struct cpuinfo_x86
*c
)
449 unsigned int n
, dummy
, ebx
, ecx
, edx
, l2size
;
451 n
= c
->extended_cpuid_level
;
453 if (n
>= 0x80000005) {
454 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
455 c
->x86_cache_size
= (ecx
>>24) + (edx
>>24);
457 /* On K8 L1 TLB is inclusive, so don't count it */
462 if (n
< 0x80000006) /* Some chips just has a large L1. */
465 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
469 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
471 /* do processor-specific cache resizing */
472 if (this_cpu
->legacy_cache_size
)
473 l2size
= this_cpu
->legacy_cache_size(c
, l2size
);
475 /* Allow user to override all this if necessary. */
476 if (cachesize_override
!= -1)
477 l2size
= cachesize_override
;
480 return; /* Again, no L2 cache is possible */
483 c
->x86_cache_size
= l2size
;
486 u16 __read_mostly tlb_lli_4k
[NR_INFO
];
487 u16 __read_mostly tlb_lli_2m
[NR_INFO
];
488 u16 __read_mostly tlb_lli_4m
[NR_INFO
];
489 u16 __read_mostly tlb_lld_4k
[NR_INFO
];
490 u16 __read_mostly tlb_lld_2m
[NR_INFO
];
491 u16 __read_mostly tlb_lld_4m
[NR_INFO
];
492 u16 __read_mostly tlb_lld_1g
[NR_INFO
];
494 static void cpu_detect_tlb(struct cpuinfo_x86
*c
)
496 if (this_cpu
->c_detect_tlb
)
497 this_cpu
->c_detect_tlb(c
);
499 pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
500 tlb_lli_4k
[ENTRIES
], tlb_lli_2m
[ENTRIES
],
501 tlb_lli_4m
[ENTRIES
]);
503 pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
504 tlb_lld_4k
[ENTRIES
], tlb_lld_2m
[ENTRIES
],
505 tlb_lld_4m
[ENTRIES
], tlb_lld_1g
[ENTRIES
]);
508 void detect_ht(struct cpuinfo_x86
*c
)
511 u32 eax
, ebx
, ecx
, edx
;
512 int index_msb
, core_bits
;
515 if (!cpu_has(c
, X86_FEATURE_HT
))
518 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
521 if (cpu_has(c
, X86_FEATURE_XTOPOLOGY
))
524 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
526 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
528 if (smp_num_siblings
== 1) {
529 printk_once(KERN_INFO
"CPU0: Hyper-Threading is disabled\n");
533 if (smp_num_siblings
<= 1)
536 index_msb
= get_count_order(smp_num_siblings
);
537 c
->phys_proc_id
= apic
->phys_pkg_id(c
->initial_apicid
, index_msb
);
539 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
541 index_msb
= get_count_order(smp_num_siblings
);
543 core_bits
= get_count_order(c
->x86_max_cores
);
545 c
->cpu_core_id
= apic
->phys_pkg_id(c
->initial_apicid
, index_msb
) &
546 ((1 << core_bits
) - 1);
549 if (!printed
&& (c
->x86_max_cores
* smp_num_siblings
) > 1) {
550 printk(KERN_INFO
"CPU: Physical Processor ID: %d\n",
552 printk(KERN_INFO
"CPU: Processor Core ID: %d\n",
559 static void get_cpu_vendor(struct cpuinfo_x86
*c
)
561 char *v
= c
->x86_vendor_id
;
564 for (i
= 0; i
< X86_VENDOR_NUM
; i
++) {
568 if (!strcmp(v
, cpu_devs
[i
]->c_ident
[0]) ||
569 (cpu_devs
[i
]->c_ident
[1] &&
570 !strcmp(v
, cpu_devs
[i
]->c_ident
[1]))) {
572 this_cpu
= cpu_devs
[i
];
573 c
->x86_vendor
= this_cpu
->c_x86_vendor
;
579 "CPU: vendor_id '%s' unknown, using generic init.\n" \
580 "CPU: Your system may be unstable.\n", v
);
582 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
583 this_cpu
= &default_cpu
;
586 void cpu_detect(struct cpuinfo_x86
*c
)
588 /* Get vendor name */
589 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
590 (unsigned int *)&c
->x86_vendor_id
[0],
591 (unsigned int *)&c
->x86_vendor_id
[8],
592 (unsigned int *)&c
->x86_vendor_id
[4]);
595 /* Intel-defined flags: level 0x00000001 */
596 if (c
->cpuid_level
>= 0x00000001) {
597 u32 junk
, tfms
, cap0
, misc
;
599 cpuid(0x00000001, &tfms
, &misc
, &junk
, &cap0
);
600 c
->x86
= (tfms
>> 8) & 0xf;
601 c
->x86_model
= (tfms
>> 4) & 0xf;
602 c
->x86_mask
= tfms
& 0xf;
605 c
->x86
+= (tfms
>> 20) & 0xff;
607 c
->x86_model
+= ((tfms
>> 16) & 0xf) << 4;
609 if (cap0
& (1<<19)) {
610 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
611 c
->x86_cache_alignment
= c
->x86_clflush_size
;
616 void get_cpu_cap(struct cpuinfo_x86
*c
)
621 /* Intel-defined flags: level 0x00000001 */
622 if (c
->cpuid_level
>= 0x00000001) {
623 u32 capability
, excap
;
625 cpuid(0x00000001, &tfms
, &ebx
, &excap
, &capability
);
626 c
->x86_capability
[0] = capability
;
627 c
->x86_capability
[4] = excap
;
630 /* Additional Intel-defined flags: level 0x00000007 */
631 if (c
->cpuid_level
>= 0x00000007) {
632 u32 eax
, ebx
, ecx
, edx
;
634 cpuid_count(0x00000007, 0, &eax
, &ebx
, &ecx
, &edx
);
636 c
->x86_capability
[9] = ebx
;
639 /* Extended state features: level 0x0000000d */
640 if (c
->cpuid_level
>= 0x0000000d) {
641 u32 eax
, ebx
, ecx
, edx
;
643 cpuid_count(0x0000000d, 1, &eax
, &ebx
, &ecx
, &edx
);
645 c
->x86_capability
[10] = eax
;
648 /* Additional Intel-defined flags: level 0x0000000F */
649 if (c
->cpuid_level
>= 0x0000000F) {
650 u32 eax
, ebx
, ecx
, edx
;
652 /* QoS sub-leaf, EAX=0Fh, ECX=0 */
653 cpuid_count(0x0000000F, 0, &eax
, &ebx
, &ecx
, &edx
);
654 c
->x86_capability
[11] = edx
;
655 if (cpu_has(c
, X86_FEATURE_CQM_LLC
)) {
656 /* will be overridden if occupancy monitoring exists */
657 c
->x86_cache_max_rmid
= ebx
;
659 /* QoS sub-leaf, EAX=0Fh, ECX=1 */
660 cpuid_count(0x0000000F, 1, &eax
, &ebx
, &ecx
, &edx
);
661 c
->x86_capability
[12] = edx
;
662 if (cpu_has(c
, X86_FEATURE_CQM_OCCUP_LLC
)) {
663 c
->x86_cache_max_rmid
= ecx
;
664 c
->x86_cache_occ_scale
= ebx
;
667 c
->x86_cache_max_rmid
= -1;
668 c
->x86_cache_occ_scale
= -1;
672 /* AMD-defined flags: level 0x80000001 */
673 xlvl
= cpuid_eax(0x80000000);
674 c
->extended_cpuid_level
= xlvl
;
676 if ((xlvl
& 0xffff0000) == 0x80000000) {
677 if (xlvl
>= 0x80000001) {
678 c
->x86_capability
[1] = cpuid_edx(0x80000001);
679 c
->x86_capability
[6] = cpuid_ecx(0x80000001);
683 if (c
->extended_cpuid_level
>= 0x80000008) {
684 u32 eax
= cpuid_eax(0x80000008);
686 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
687 c
->x86_phys_bits
= eax
& 0xff;
690 else if (cpu_has(c
, X86_FEATURE_PAE
) || cpu_has(c
, X86_FEATURE_PSE36
))
691 c
->x86_phys_bits
= 36;
694 if (c
->extended_cpuid_level
>= 0x80000007)
695 c
->x86_power
= cpuid_edx(0x80000007);
697 init_scattered_cpuid_features(c
);
700 static void identify_cpu_without_cpuid(struct cpuinfo_x86
*c
)
706 * First of all, decide if this is a 486 or higher
707 * It's a 486 if we can modify the AC flag
709 if (flag_is_changeable_p(X86_EFLAGS_AC
))
714 for (i
= 0; i
< X86_VENDOR_NUM
; i
++)
715 if (cpu_devs
[i
] && cpu_devs
[i
]->c_identify
) {
716 c
->x86_vendor_id
[0] = 0;
717 cpu_devs
[i
]->c_identify(c
);
718 if (c
->x86_vendor_id
[0]) {
727 * Do minimum CPU detection early.
728 * Fields really needed: vendor, cpuid_level, family, model, mask,
730 * The others are not touched to avoid unwanted side effects.
732 * WARNING: this function is only called on the BP. Don't add code here
733 * that is supposed to run on all CPUs.
735 static void __init
early_identify_cpu(struct cpuinfo_x86
*c
)
738 c
->x86_clflush_size
= 64;
739 c
->x86_phys_bits
= 36;
740 c
->x86_virt_bits
= 48;
742 c
->x86_clflush_size
= 32;
743 c
->x86_phys_bits
= 32;
744 c
->x86_virt_bits
= 32;
746 c
->x86_cache_alignment
= c
->x86_clflush_size
;
748 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
749 c
->extended_cpuid_level
= 0;
752 identify_cpu_without_cpuid(c
);
754 /* cyrix could have cpuid enabled via c_identify()*/
763 if (this_cpu
->c_early_init
)
764 this_cpu
->c_early_init(c
);
767 filter_cpuid_features(c
, false);
769 if (this_cpu
->c_bsp_init
)
770 this_cpu
->c_bsp_init(c
);
772 setup_force_cpu_cap(X86_FEATURE_ALWAYS
);
775 void __init
early_cpu_init(void)
777 const struct cpu_dev
*const *cdev
;
780 #ifdef CONFIG_PROCESSOR_SELECT
781 printk(KERN_INFO
"KERNEL supported cpus:\n");
784 for (cdev
= __x86_cpu_dev_start
; cdev
< __x86_cpu_dev_end
; cdev
++) {
785 const struct cpu_dev
*cpudev
= *cdev
;
787 if (count
>= X86_VENDOR_NUM
)
789 cpu_devs
[count
] = cpudev
;
792 #ifdef CONFIG_PROCESSOR_SELECT
796 for (j
= 0; j
< 2; j
++) {
797 if (!cpudev
->c_ident
[j
])
799 printk(KERN_INFO
" %s %s\n", cpudev
->c_vendor
,
805 early_identify_cpu(&boot_cpu_data
);
809 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
810 * unfortunately, that's not true in practice because of early VIA
811 * chips and (more importantly) broken virtualizers that are not easy
812 * to detect. In the latter case it doesn't even *fail* reliably, so
813 * probing for it doesn't even work. Disable it completely on 32-bit
814 * unless we can find a reliable way to detect all the broken cases.
815 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
817 static void detect_nopl(struct cpuinfo_x86
*c
)
820 clear_cpu_cap(c
, X86_FEATURE_NOPL
);
822 set_cpu_cap(c
, X86_FEATURE_NOPL
);
826 static void generic_identify(struct cpuinfo_x86
*c
)
828 c
->extended_cpuid_level
= 0;
831 identify_cpu_without_cpuid(c
);
833 /* cyrix could have cpuid enabled via c_identify()*/
843 if (c
->cpuid_level
>= 0x00000001) {
844 c
->initial_apicid
= (cpuid_ebx(1) >> 24) & 0xFF;
846 # ifdef CONFIG_X86_HT
847 c
->apicid
= apic
->phys_pkg_id(c
->initial_apicid
, 0);
849 c
->apicid
= c
->initial_apicid
;
852 c
->phys_proc_id
= c
->initial_apicid
;
855 get_model_name(c
); /* Default name */
860 static void x86_init_cache_qos(struct cpuinfo_x86
*c
)
863 * The heavy lifting of max_rmid and cache_occ_scale are handled
864 * in get_cpu_cap(). Here we just set the max_rmid for the boot_cpu
865 * in case CQM bits really aren't there in this CPU.
867 if (c
!= &boot_cpu_data
) {
868 boot_cpu_data
.x86_cache_max_rmid
=
869 min(boot_cpu_data
.x86_cache_max_rmid
,
870 c
->x86_cache_max_rmid
);
875 * This does the hard work of actually picking apart the CPU stuff...
877 static void identify_cpu(struct cpuinfo_x86
*c
)
881 c
->loops_per_jiffy
= loops_per_jiffy
;
882 c
->x86_cache_size
= -1;
883 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
884 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
885 c
->x86_vendor_id
[0] = '\0'; /* Unset */
886 c
->x86_model_id
[0] = '\0'; /* Unset */
887 c
->x86_max_cores
= 1;
888 c
->x86_coreid_bits
= 0;
890 c
->x86_clflush_size
= 64;
891 c
->x86_phys_bits
= 36;
892 c
->x86_virt_bits
= 48;
894 c
->cpuid_level
= -1; /* CPUID not detected */
895 c
->x86_clflush_size
= 32;
896 c
->x86_phys_bits
= 32;
897 c
->x86_virt_bits
= 32;
899 c
->x86_cache_alignment
= c
->x86_clflush_size
;
900 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
904 if (this_cpu
->c_identify
)
905 this_cpu
->c_identify(c
);
907 /* Clear/Set all flags overriden by options, after probe */
908 for (i
= 0; i
< NCAPINTS
; i
++) {
909 c
->x86_capability
[i
] &= ~cpu_caps_cleared
[i
];
910 c
->x86_capability
[i
] |= cpu_caps_set
[i
];
914 c
->apicid
= apic
->phys_pkg_id(c
->initial_apicid
, 0);
918 * Vendor-specific initialization. In this section we
919 * canonicalize the feature flags, meaning if there are
920 * features a certain CPU supports which CPUID doesn't
921 * tell us, CPUID claiming incorrect flags, or other bugs,
922 * we handle them here.
924 * At the end of this section, c->x86_capability better
925 * indicate the features this CPU genuinely supports!
927 if (this_cpu
->c_init
)
930 /* Disable the PN if appropriate */
931 squash_the_stupid_serial_number(c
);
933 /* Set up SMEP/SMAP */
938 * The vendor-specific functions might have changed features.
939 * Now we do "generic changes."
942 /* Filter out anything that depends on CPUID levels we don't have */
943 filter_cpuid_features(c
, true);
945 /* If the model name is still unset, do table lookup. */
946 if (!c
->x86_model_id
[0]) {
948 p
= table_lookup_model(c
);
950 strcpy(c
->x86_model_id
, p
);
953 sprintf(c
->x86_model_id
, "%02x/%02x",
954 c
->x86
, c
->x86_model
);
963 x86_init_cache_qos(c
);
966 * Clear/Set all flags overriden by options, need do it
967 * before following smp all cpus cap AND.
969 for (i
= 0; i
< NCAPINTS
; i
++) {
970 c
->x86_capability
[i
] &= ~cpu_caps_cleared
[i
];
971 c
->x86_capability
[i
] |= cpu_caps_set
[i
];
975 * On SMP, boot_cpu_data holds the common feature set between
976 * all CPUs; so make sure that we indicate which features are
977 * common between the CPUs. The first time this routine gets
978 * executed, c == &boot_cpu_data.
980 if (c
!= &boot_cpu_data
) {
981 /* AND the already accumulated flags with these */
982 for (i
= 0; i
< NCAPINTS
; i
++)
983 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
985 /* OR, i.e. replicate the bug flags */
986 for (i
= NCAPINTS
; i
< NCAPINTS
+ NBUGINTS
; i
++)
987 c
->x86_capability
[i
] |= boot_cpu_data
.x86_capability
[i
];
990 /* Init Machine Check Exception if available. */
993 select_idle_routine(c
);
996 numa_add_cpu(smp_processor_id());
1001 * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1002 * on 32-bit kernels:
1004 #ifdef CONFIG_X86_32
1005 void enable_sep_cpu(void)
1007 struct tss_struct
*tss
;
1011 tss
= &per_cpu(cpu_tss
, cpu
);
1013 if (!boot_cpu_has(X86_FEATURE_SEP
))
1017 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1018 * see the big comment in struct x86_hw_tss's definition.
1021 tss
->x86_tss
.ss1
= __KERNEL_CS
;
1022 wrmsr(MSR_IA32_SYSENTER_CS
, tss
->x86_tss
.ss1
, 0);
1024 wrmsr(MSR_IA32_SYSENTER_ESP
,
1025 (unsigned long)tss
+ offsetofend(struct tss_struct
, SYSENTER_stack
),
1028 wrmsr(MSR_IA32_SYSENTER_EIP
, (unsigned long)ia32_sysenter_target
, 0);
1035 void __init
identify_boot_cpu(void)
1037 identify_cpu(&boot_cpu_data
);
1038 init_amd_e400_c1e_mask();
1039 #ifdef CONFIG_X86_32
1043 cpu_detect_tlb(&boot_cpu_data
);
1046 void identify_secondary_cpu(struct cpuinfo_x86
*c
)
1048 BUG_ON(c
== &boot_cpu_data
);
1050 #ifdef CONFIG_X86_32
1061 static const struct msr_range msr_range_array
[] = {
1062 { 0x00000000, 0x00000418},
1063 { 0xc0000000, 0xc000040b},
1064 { 0xc0010000, 0xc0010142},
1065 { 0xc0011000, 0xc001103b},
1068 static void __print_cpu_msr(void)
1070 unsigned index_min
, index_max
;
1075 for (i
= 0; i
< ARRAY_SIZE(msr_range_array
); i
++) {
1076 index_min
= msr_range_array
[i
].min
;
1077 index_max
= msr_range_array
[i
].max
;
1079 for (index
= index_min
; index
< index_max
; index
++) {
1080 if (rdmsrl_safe(index
, &val
))
1082 printk(KERN_INFO
" MSR%08x: %016llx\n", index
, val
);
1087 static int show_msr
;
1089 static __init
int setup_show_msr(char *arg
)
1093 get_option(&arg
, &num
);
1099 __setup("show_msr=", setup_show_msr
);
1101 static __init
int setup_noclflush(char *arg
)
1103 setup_clear_cpu_cap(X86_FEATURE_CLFLUSH
);
1104 setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT
);
1107 __setup("noclflush", setup_noclflush
);
1109 void print_cpu_info(struct cpuinfo_x86
*c
)
1111 const char *vendor
= NULL
;
1113 if (c
->x86_vendor
< X86_VENDOR_NUM
) {
1114 vendor
= this_cpu
->c_vendor
;
1116 if (c
->cpuid_level
>= 0)
1117 vendor
= c
->x86_vendor_id
;
1120 if (vendor
&& !strstr(c
->x86_model_id
, vendor
))
1121 printk(KERN_CONT
"%s ", vendor
);
1123 if (c
->x86_model_id
[0])
1124 printk(KERN_CONT
"%s", strim(c
->x86_model_id
));
1126 printk(KERN_CONT
"%d86", c
->x86
);
1128 printk(KERN_CONT
" (fam: %02x, model: %02x", c
->x86
, c
->x86_model
);
1130 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1131 printk(KERN_CONT
", stepping: %02x)\n", c
->x86_mask
);
1133 printk(KERN_CONT
")\n");
1138 void print_cpu_msr(struct cpuinfo_x86
*c
)
1140 if (c
->cpu_index
< show_msr
)
1144 static __init
int setup_disablecpuid(char *arg
)
1148 if (get_option(&arg
, &bit
) && bit
< NCAPINTS
*32)
1149 setup_clear_cpu_cap(bit
);
1155 __setup("clearcpuid=", setup_disablecpuid
);
1157 DEFINE_PER_CPU(unsigned long, kernel_stack
) =
1158 (unsigned long)&init_thread_union
+ THREAD_SIZE
;
1159 EXPORT_PER_CPU_SYMBOL(kernel_stack
);
1161 #ifdef CONFIG_X86_64
1162 struct desc_ptr idt_descr
= { NR_VECTORS
* 16 - 1, (unsigned long) idt_table
};
1163 struct desc_ptr debug_idt_descr
= { NR_VECTORS
* 16 - 1,
1164 (unsigned long) debug_idt_table
};
1166 DEFINE_PER_CPU_FIRST(union irq_stack_union
,
1167 irq_stack_union
) __aligned(PAGE_SIZE
) __visible
;
1170 * The following percpu variables are hot. Align current_task to
1171 * cacheline size such that they fall in the same cacheline.
1173 DEFINE_PER_CPU(struct task_struct
*, current_task
) ____cacheline_aligned
=
1175 EXPORT_PER_CPU_SYMBOL(current_task
);
1177 DEFINE_PER_CPU(char *, irq_stack_ptr
) =
1178 init_per_cpu_var(irq_stack_union
.irq_stack
) + IRQ_STACK_SIZE
- 64;
1180 DEFINE_PER_CPU(unsigned int, irq_count
) __visible
= -1;
1182 DEFINE_PER_CPU(int, __preempt_count
) = INIT_PREEMPT_COUNT
;
1183 EXPORT_PER_CPU_SYMBOL(__preempt_count
);
1185 DEFINE_PER_CPU(struct task_struct
*, fpu_owner_task
);
1188 * Special IST stacks which the CPU switches to when it calls
1189 * an IST-marked descriptor entry. Up to 7 stacks (hardware
1190 * limit), all of them are 4K, except the debug stack which
1193 static const unsigned int exception_stack_sizes
[N_EXCEPTION_STACKS
] = {
1194 [0 ... N_EXCEPTION_STACKS
- 1] = EXCEPTION_STKSZ
,
1195 [DEBUG_STACK
- 1] = DEBUG_STKSZ
1198 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1199 [(N_EXCEPTION_STACKS
- 1) * EXCEPTION_STKSZ
+ DEBUG_STKSZ
]);
1201 /* May not be marked __init: used by software suspend */
1202 void syscall_init(void)
1205 * LSTAR and STAR live in a bit strange symbiosis.
1206 * They both write to the same internal register. STAR allows to
1207 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1209 wrmsrl(MSR_STAR
, ((u64
)__USER32_CS
)<<48 | ((u64
)__KERNEL_CS
)<<32);
1210 wrmsrl(MSR_LSTAR
, system_call
);
1212 #ifdef CONFIG_IA32_EMULATION
1213 wrmsrl(MSR_CSTAR
, ia32_cstar_target
);
1215 * This only works on Intel CPUs.
1216 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1217 * This does not cause SYSENTER to jump to the wrong location, because
1218 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1220 wrmsrl_safe(MSR_IA32_SYSENTER_CS
, (u64
)__KERNEL_CS
);
1221 wrmsrl_safe(MSR_IA32_SYSENTER_ESP
, 0ULL);
1222 wrmsrl_safe(MSR_IA32_SYSENTER_EIP
, (u64
)ia32_sysenter_target
);
1224 wrmsrl(MSR_CSTAR
, ignore_sysret
);
1225 wrmsrl_safe(MSR_IA32_SYSENTER_CS
, (u64
)GDT_ENTRY_INVALID_SEG
);
1226 wrmsrl_safe(MSR_IA32_SYSENTER_ESP
, 0ULL);
1227 wrmsrl_safe(MSR_IA32_SYSENTER_EIP
, 0ULL);
1230 /* Flags to clear on syscall */
1231 wrmsrl(MSR_SYSCALL_MASK
,
1232 X86_EFLAGS_TF
|X86_EFLAGS_DF
|X86_EFLAGS_IF
|
1233 X86_EFLAGS_IOPL
|X86_EFLAGS_AC
|X86_EFLAGS_NT
);
1237 * Copies of the original ist values from the tss are only accessed during
1238 * debugging, no special alignment required.
1240 DEFINE_PER_CPU(struct orig_ist
, orig_ist
);
1242 static DEFINE_PER_CPU(unsigned long, debug_stack_addr
);
1243 DEFINE_PER_CPU(int, debug_stack_usage
);
1245 int is_debug_stack(unsigned long addr
)
1247 return __this_cpu_read(debug_stack_usage
) ||
1248 (addr
<= __this_cpu_read(debug_stack_addr
) &&
1249 addr
> (__this_cpu_read(debug_stack_addr
) - DEBUG_STKSZ
));
1251 NOKPROBE_SYMBOL(is_debug_stack
);
1253 DEFINE_PER_CPU(u32
, debug_idt_ctr
);
1255 void debug_stack_set_zero(void)
1257 this_cpu_inc(debug_idt_ctr
);
1260 NOKPROBE_SYMBOL(debug_stack_set_zero
);
1262 void debug_stack_reset(void)
1264 if (WARN_ON(!this_cpu_read(debug_idt_ctr
)))
1266 if (this_cpu_dec_return(debug_idt_ctr
) == 0)
1269 NOKPROBE_SYMBOL(debug_stack_reset
);
1271 #else /* CONFIG_X86_64 */
1273 DEFINE_PER_CPU(struct task_struct
*, current_task
) = &init_task
;
1274 EXPORT_PER_CPU_SYMBOL(current_task
);
1275 DEFINE_PER_CPU(int, __preempt_count
) = INIT_PREEMPT_COUNT
;
1276 EXPORT_PER_CPU_SYMBOL(__preempt_count
);
1277 DEFINE_PER_CPU(struct task_struct
*, fpu_owner_task
);
1280 * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1281 * the top of the kernel stack. Use an extra percpu variable to track the
1282 * top of the kernel stack directly.
1284 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack
) =
1285 (unsigned long)&init_thread_union
+ THREAD_SIZE
;
1286 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack
);
1288 #ifdef CONFIG_CC_STACKPROTECTOR
1289 DEFINE_PER_CPU_ALIGNED(struct stack_canary
, stack_canary
);
1292 #endif /* CONFIG_X86_64 */
1295 * Clear all 6 debug registers:
1297 static void clear_all_debug_regs(void)
1301 for (i
= 0; i
< 8; i
++) {
1302 /* Ignore db4, db5 */
1303 if ((i
== 4) || (i
== 5))
1312 * Restore debug regs if using kgdbwait and you have a kernel debugger
1313 * connection established.
1315 static void dbg_restore_debug_regs(void)
1317 if (unlikely(kgdb_connected
&& arch_kgdb_ops
.correct_hw_break
))
1318 arch_kgdb_ops
.correct_hw_break();
1320 #else /* ! CONFIG_KGDB */
1321 #define dbg_restore_debug_regs()
1322 #endif /* ! CONFIG_KGDB */
1324 static void wait_for_master_cpu(int cpu
)
1328 * wait for ACK from master CPU before continuing
1329 * with AP initialization
1331 WARN_ON(cpumask_test_and_set_cpu(cpu
, cpu_initialized_mask
));
1332 while (!cpumask_test_cpu(cpu
, cpu_callout_mask
))
1338 * cpu_init() initializes state that is per-CPU. Some data is already
1339 * initialized (naturally) in the bootstrap process, such as the GDT
1340 * and IDT. We reload them nevertheless, this function acts as a
1341 * 'CPU state barrier', nothing should get across.
1342 * A lot of state is already set up in PDA init for 64 bit
1344 #ifdef CONFIG_X86_64
1348 struct orig_ist
*oist
;
1349 struct task_struct
*me
;
1350 struct tss_struct
*t
;
1352 int cpu
= stack_smp_processor_id();
1355 wait_for_master_cpu(cpu
);
1358 * Initialize the CR4 shadow before doing anything that could
1364 * Load microcode on this cpu if a valid microcode is available.
1365 * This is early microcode loading procedure.
1369 t
= &per_cpu(cpu_tss
, cpu
);
1370 oist
= &per_cpu(orig_ist
, cpu
);
1373 if (this_cpu_read(numa_node
) == 0 &&
1374 early_cpu_to_node(cpu
) != NUMA_NO_NODE
)
1375 set_numa_node(early_cpu_to_node(cpu
));
1380 pr_debug("Initializing CPU#%d\n", cpu
);
1382 cr4_clear_bits(X86_CR4_VME
|X86_CR4_PVI
|X86_CR4_TSD
|X86_CR4_DE
);
1385 * Initialize the per-CPU GDT with the boot GDT,
1386 * and set up the GDT descriptor:
1389 switch_to_new_gdt(cpu
);
1394 memset(me
->thread
.tls_array
, 0, GDT_ENTRY_TLS_ENTRIES
* 8);
1397 wrmsrl(MSR_FS_BASE
, 0);
1398 wrmsrl(MSR_KERNEL_GS_BASE
, 0);
1405 * set up and load the per-CPU TSS
1407 if (!oist
->ist
[0]) {
1408 char *estacks
= per_cpu(exception_stacks
, cpu
);
1410 for (v
= 0; v
< N_EXCEPTION_STACKS
; v
++) {
1411 estacks
+= exception_stack_sizes
[v
];
1412 oist
->ist
[v
] = t
->x86_tss
.ist
[v
] =
1413 (unsigned long)estacks
;
1414 if (v
== DEBUG_STACK
-1)
1415 per_cpu(debug_stack_addr
, cpu
) = (unsigned long)estacks
;
1419 t
->x86_tss
.io_bitmap_base
= offsetof(struct tss_struct
, io_bitmap
);
1422 * <= is required because the CPU will access up to
1423 * 8 bits beyond the end of the IO permission bitmap.
1425 for (i
= 0; i
<= IO_BITMAP_LONGS
; i
++)
1426 t
->io_bitmap
[i
] = ~0UL;
1428 atomic_inc(&init_mm
.mm_count
);
1429 me
->active_mm
= &init_mm
;
1431 enter_lazy_tlb(&init_mm
, me
);
1433 load_sp0(t
, ¤t
->thread
);
1434 set_tss_desc(cpu
, t
);
1436 load_mm_ldt(&init_mm
);
1438 clear_all_debug_regs();
1439 dbg_restore_debug_regs();
1451 int cpu
= smp_processor_id();
1452 struct task_struct
*curr
= current
;
1453 struct tss_struct
*t
= &per_cpu(cpu_tss
, cpu
);
1454 struct thread_struct
*thread
= &curr
->thread
;
1456 wait_for_master_cpu(cpu
);
1459 * Initialize the CR4 shadow before doing anything that could
1464 show_ucode_info_early();
1466 printk(KERN_INFO
"Initializing CPU#%d\n", cpu
);
1468 if (cpu_feature_enabled(X86_FEATURE_VME
) || cpu_has_tsc
|| cpu_has_de
)
1469 cr4_clear_bits(X86_CR4_VME
|X86_CR4_PVI
|X86_CR4_TSD
|X86_CR4_DE
);
1472 switch_to_new_gdt(cpu
);
1475 * Set up and load the per-CPU TSS and LDT
1477 atomic_inc(&init_mm
.mm_count
);
1478 curr
->active_mm
= &init_mm
;
1480 enter_lazy_tlb(&init_mm
, curr
);
1482 load_sp0(t
, thread
);
1483 set_tss_desc(cpu
, t
);
1485 load_mm_ldt(&init_mm
);
1487 t
->x86_tss
.io_bitmap_base
= offsetof(struct tss_struct
, io_bitmap
);
1489 #ifdef CONFIG_DOUBLEFAULT
1490 /* Set up doublefault TSS pointer in the GDT */
1491 __set_tss_desc(cpu
, GDT_ENTRY_DOUBLEFAULT_TSS
, &doublefault_tss
);
1494 clear_all_debug_regs();
1495 dbg_restore_debug_regs();
1501 #ifdef CONFIG_X86_DEBUG_STATIC_CPU_HAS
1502 void warn_pre_alternatives(void)
1504 WARN(1, "You're using static_cpu_has before alternatives have run!\n");
1506 EXPORT_SYMBOL_GPL(warn_pre_alternatives
);
1509 inline bool __static_cpu_has_safe(u16 bit
)
1511 return boot_cpu_has(bit
);
1513 EXPORT_SYMBOL_GPL(__static_cpu_has_safe
);