1 #include <linux/bootmem.h>
2 #include <linux/linkage.h>
3 #include <linux/bitops.h>
4 #include <linux/kernel.h>
5 #include <linux/export.h>
6 #include <linux/percpu.h>
7 #include <linux/string.h>
8 #include <linux/ctype.h>
9 #include <linux/delay.h>
10 #include <linux/sched/mm.h>
11 #include <linux/sched/clock.h>
12 #include <linux/sched/task.h>
13 #include <linux/init.h>
14 #include <linux/kprobes.h>
15 #include <linux/kgdb.h>
16 #include <linux/smp.h>
18 #include <linux/syscore_ops.h>
20 #include <asm/stackprotector.h>
21 #include <asm/perf_event.h>
22 #include <asm/mmu_context.h>
23 #include <asm/archrandom.h>
24 #include <asm/hypervisor.h>
25 #include <asm/processor.h>
26 #include <asm/tlbflush.h>
27 #include <asm/debugreg.h>
28 #include <asm/sections.h>
29 #include <asm/vsyscall.h>
30 #include <linux/topology.h>
31 #include <linux/cpumask.h>
32 #include <asm/pgtable.h>
33 #include <linux/atomic.h>
34 #include <asm/proto.h>
35 #include <asm/setup.h>
38 #include <asm/fpu/internal.h>
40 #include <asm/hwcap2.h>
41 #include <linux/numa.h>
48 #include <asm/microcode.h>
49 #include <asm/microcode_intel.h>
50 #include <asm/intel-family.h>
51 #include <asm/cpu_device_id.h>
53 #ifdef CONFIG_X86_LOCAL_APIC
54 #include <asm/uv/uv.h>
59 u32 elf_hwcap2 __read_mostly
;
61 /* all of these masks are initialized in setup_cpu_local_masks() */
62 cpumask_var_t cpu_initialized_mask
;
63 cpumask_var_t cpu_callout_mask
;
64 cpumask_var_t cpu_callin_mask
;
66 /* representing cpus for which sibling maps can be computed */
67 cpumask_var_t cpu_sibling_setup_mask
;
69 /* correctly size the local cpu masks */
70 void __init
setup_cpu_local_masks(void)
72 alloc_bootmem_cpumask_var(&cpu_initialized_mask
);
73 alloc_bootmem_cpumask_var(&cpu_callin_mask
);
74 alloc_bootmem_cpumask_var(&cpu_callout_mask
);
75 alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask
);
78 static void default_init(struct cpuinfo_x86
*c
)
81 cpu_detect_cache_sizes(c
);
83 /* Not much we can do here... */
84 /* Check if at least it has cpuid */
85 if (c
->cpuid_level
== -1) {
86 /* No cpuid. It must be an ancient CPU */
88 strcpy(c
->x86_model_id
, "486");
90 strcpy(c
->x86_model_id
, "386");
95 static const struct cpu_dev default_cpu
= {
96 .c_init
= default_init
,
97 .c_vendor
= "Unknown",
98 .c_x86_vendor
= X86_VENDOR_UNKNOWN
,
101 static const struct cpu_dev
*this_cpu
= &default_cpu
;
103 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page
, gdt_page
) = { .gdt
= {
106 * We need valid kernel segments for data and code in long mode too
107 * IRET will check the segment types kkeil 2000/10/28
108 * Also sysret mandates a special GDT layout
110 * TLS descriptors are currently at a different place compared to i386.
111 * Hopefully nobody expects them at a fixed place (Wine?)
113 [GDT_ENTRY_KERNEL32_CS
] = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
114 [GDT_ENTRY_KERNEL_CS
] = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
115 [GDT_ENTRY_KERNEL_DS
] = GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
116 [GDT_ENTRY_DEFAULT_USER32_CS
] = GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
117 [GDT_ENTRY_DEFAULT_USER_DS
] = GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
118 [GDT_ENTRY_DEFAULT_USER_CS
] = GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
120 [GDT_ENTRY_KERNEL_CS
] = GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
121 [GDT_ENTRY_KERNEL_DS
] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
122 [GDT_ENTRY_DEFAULT_USER_CS
] = GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
123 [GDT_ENTRY_DEFAULT_USER_DS
] = GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
125 * Segments used for calling PnP BIOS have byte granularity.
126 * They code segments and data segments have fixed 64k limits,
127 * the transfer segment sizes are set at run time.
130 [GDT_ENTRY_PNPBIOS_CS32
] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
132 [GDT_ENTRY_PNPBIOS_CS16
] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
134 [GDT_ENTRY_PNPBIOS_DS
] = GDT_ENTRY_INIT(0x0092, 0, 0xffff),
136 [GDT_ENTRY_PNPBIOS_TS1
] = GDT_ENTRY_INIT(0x0092, 0, 0),
138 [GDT_ENTRY_PNPBIOS_TS2
] = GDT_ENTRY_INIT(0x0092, 0, 0),
140 * The APM segments have byte granularity and their bases
141 * are set at run time. All have 64k limits.
144 [GDT_ENTRY_APMBIOS_BASE
] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
146 [GDT_ENTRY_APMBIOS_BASE
+1] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
148 [GDT_ENTRY_APMBIOS_BASE
+2] = GDT_ENTRY_INIT(0x4092, 0, 0xffff),
150 [GDT_ENTRY_ESPFIX_SS
] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
151 [GDT_ENTRY_PERCPU
] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
152 GDT_STACK_CANARY_INIT
155 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page
);
157 static int __init
x86_mpx_setup(char *s
)
159 /* require an exact match without trailing characters */
163 /* do not emit a message if the feature is not present */
164 if (!boot_cpu_has(X86_FEATURE_MPX
))
167 setup_clear_cpu_cap(X86_FEATURE_MPX
);
168 pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
171 __setup("nompx", x86_mpx_setup
);
174 static int __init
x86_nopcid_setup(char *s
)
176 /* nopcid doesn't accept parameters */
180 /* do not emit a message if the feature is not present */
181 if (!boot_cpu_has(X86_FEATURE_PCID
))
184 setup_clear_cpu_cap(X86_FEATURE_PCID
);
185 pr_info("nopcid: PCID feature disabled\n");
188 early_param("nopcid", x86_nopcid_setup
);
191 static int __init
x86_noinvpcid_setup(char *s
)
193 /* noinvpcid doesn't accept parameters */
197 /* do not emit a message if the feature is not present */
198 if (!boot_cpu_has(X86_FEATURE_INVPCID
))
201 setup_clear_cpu_cap(X86_FEATURE_INVPCID
);
202 pr_info("noinvpcid: INVPCID feature disabled\n");
205 early_param("noinvpcid", x86_noinvpcid_setup
);
208 static int cachesize_override
= -1;
209 static int disable_x86_serial_nr
= 1;
211 static int __init
cachesize_setup(char *str
)
213 get_option(&str
, &cachesize_override
);
216 __setup("cachesize=", cachesize_setup
);
218 static int __init
x86_sep_setup(char *s
)
220 setup_clear_cpu_cap(X86_FEATURE_SEP
);
223 __setup("nosep", x86_sep_setup
);
225 /* Standard macro to see if a specific flag is changeable */
226 static inline int flag_is_changeable_p(u32 flag
)
231 * Cyrix and IDT cpus allow disabling of CPUID
232 * so the code below may return different results
233 * when it is executed before and after enabling
234 * the CPUID. Add "volatile" to not allow gcc to
235 * optimize the subsequent calls to this function.
237 asm volatile ("pushfl \n\t"
248 : "=&r" (f1
), "=&r" (f2
)
251 return ((f1
^f2
) & flag
) != 0;
254 /* Probe for the CPUID instruction */
255 int have_cpuid_p(void)
257 return flag_is_changeable_p(X86_EFLAGS_ID
);
260 static void squash_the_stupid_serial_number(struct cpuinfo_x86
*c
)
262 unsigned long lo
, hi
;
264 if (!cpu_has(c
, X86_FEATURE_PN
) || !disable_x86_serial_nr
)
267 /* Disable processor serial number: */
269 rdmsr(MSR_IA32_BBL_CR_CTL
, lo
, hi
);
271 wrmsr(MSR_IA32_BBL_CR_CTL
, lo
, hi
);
273 pr_notice("CPU serial number disabled.\n");
274 clear_cpu_cap(c
, X86_FEATURE_PN
);
276 /* Disabling the serial number may affect the cpuid level */
277 c
->cpuid_level
= cpuid_eax(0);
280 static int __init
x86_serial_nr_setup(char *s
)
282 disable_x86_serial_nr
= 0;
285 __setup("serialnumber", x86_serial_nr_setup
);
287 static inline int flag_is_changeable_p(u32 flag
)
291 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86
*c
)
296 static __init
int setup_disable_smep(char *arg
)
298 setup_clear_cpu_cap(X86_FEATURE_SMEP
);
299 /* Check for things that depend on SMEP being enabled: */
300 check_mpx_erratum(&boot_cpu_data
);
303 __setup("nosmep", setup_disable_smep
);
305 static __always_inline
void setup_smep(struct cpuinfo_x86
*c
)
307 if (cpu_has(c
, X86_FEATURE_SMEP
))
308 cr4_set_bits(X86_CR4_SMEP
);
311 static __init
int setup_disable_smap(char *arg
)
313 setup_clear_cpu_cap(X86_FEATURE_SMAP
);
316 __setup("nosmap", setup_disable_smap
);
318 static __always_inline
void setup_smap(struct cpuinfo_x86
*c
)
320 unsigned long eflags
= native_save_fl();
322 /* This should have been cleared long ago */
323 BUG_ON(eflags
& X86_EFLAGS_AC
);
325 if (cpu_has(c
, X86_FEATURE_SMAP
)) {
326 #ifdef CONFIG_X86_SMAP
327 cr4_set_bits(X86_CR4_SMAP
);
329 cr4_clear_bits(X86_CR4_SMAP
);
335 * Protection Keys are not available in 32-bit mode.
337 static bool pku_disabled
;
339 static __always_inline
void setup_pku(struct cpuinfo_x86
*c
)
341 /* check the boot processor, plus compile options for PKU: */
342 if (!cpu_feature_enabled(X86_FEATURE_PKU
))
344 /* checks the actual processor's cpuid bits: */
345 if (!cpu_has(c
, X86_FEATURE_PKU
))
350 cr4_set_bits(X86_CR4_PKE
);
352 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
353 * cpuid bit to be set. We need to ensure that we
354 * update that bit in this CPU's "cpu_info".
359 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
360 static __init
int setup_disable_pku(char *arg
)
363 * Do not clear the X86_FEATURE_PKU bit. All of the
364 * runtime checks are against OSPKE so clearing the
367 * This way, we will see "pku" in cpuinfo, but not
368 * "ospke", which is exactly what we want. It shows
369 * that the CPU has PKU, but the OS has not enabled it.
370 * This happens to be exactly how a system would look
371 * if we disabled the config option.
373 pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
377 __setup("nopku", setup_disable_pku
);
378 #endif /* CONFIG_X86_64 */
381 * Some CPU features depend on higher CPUID levels, which may not always
382 * be available due to CPUID level capping or broken virtualization
383 * software. Add those features to this table to auto-disable them.
385 struct cpuid_dependent_feature
{
390 static const struct cpuid_dependent_feature
391 cpuid_dependent_features
[] = {
392 { X86_FEATURE_MWAIT
, 0x00000005 },
393 { X86_FEATURE_DCA
, 0x00000009 },
394 { X86_FEATURE_XSAVE
, 0x0000000d },
398 static void filter_cpuid_features(struct cpuinfo_x86
*c
, bool warn
)
400 const struct cpuid_dependent_feature
*df
;
402 for (df
= cpuid_dependent_features
; df
->feature
; df
++) {
404 if (!cpu_has(c
, df
->feature
))
407 * Note: cpuid_level is set to -1 if unavailable, but
408 * extended_extended_level is set to 0 if unavailable
409 * and the legitimate extended levels are all negative
410 * when signed; hence the weird messing around with
413 if (!((s32
)df
->level
< 0 ?
414 (u32
)df
->level
> (u32
)c
->extended_cpuid_level
:
415 (s32
)df
->level
> (s32
)c
->cpuid_level
))
418 clear_cpu_cap(c
, df
->feature
);
422 pr_warn("CPU: CPU feature " X86_CAP_FMT
" disabled, no CPUID level 0x%x\n",
423 x86_cap_flag(df
->feature
), df
->level
);
428 * Naming convention should be: <Name> [(<Codename>)]
429 * This table only is used unless init_<vendor>() below doesn't set it;
430 * in particular, if CPUID levels 0x80000002..4 are supported, this
434 /* Look up CPU names by table lookup. */
435 static const char *table_lookup_model(struct cpuinfo_x86
*c
)
438 const struct legacy_cpu_model_info
*info
;
440 if (c
->x86_model
>= 16)
441 return NULL
; /* Range check */
446 info
= this_cpu
->legacy_models
;
448 while (info
->family
) {
449 if (info
->family
== c
->x86
)
450 return info
->model_names
[c
->x86_model
];
454 return NULL
; /* Not found */
457 __u32 cpu_caps_cleared
[NCAPINTS
+ NBUGINTS
];
458 __u32 cpu_caps_set
[NCAPINTS
+ NBUGINTS
];
460 void load_percpu_segment(int cpu
)
463 loadsegment(fs
, __KERNEL_PERCPU
);
465 __loadsegment_simple(gs
, 0);
466 wrmsrl(MSR_GS_BASE
, (unsigned long)per_cpu(irq_stack_union
.gs_base
, cpu
));
468 load_stack_canary_segment();
472 /* The 32-bit entry code needs to find cpu_entry_area. */
473 DEFINE_PER_CPU(struct cpu_entry_area
*, cpu_entry_area
);
478 * Special IST stacks which the CPU switches to when it calls
479 * an IST-marked descriptor entry. Up to 7 stacks (hardware
480 * limit), all of them are 4K, except the debug stack which
483 static const unsigned int exception_stack_sizes
[N_EXCEPTION_STACKS
] = {
484 [0 ... N_EXCEPTION_STACKS
- 1] = EXCEPTION_STKSZ
,
485 [DEBUG_STACK
- 1] = DEBUG_STKSZ
489 /* Load the original GDT from the per-cpu structure */
490 void load_direct_gdt(int cpu
)
492 struct desc_ptr gdt_descr
;
494 gdt_descr
.address
= (long)get_cpu_gdt_rw(cpu
);
495 gdt_descr
.size
= GDT_SIZE
- 1;
496 load_gdt(&gdt_descr
);
498 EXPORT_SYMBOL_GPL(load_direct_gdt
);
500 /* Load a fixmap remapping of the per-cpu GDT */
501 void load_fixmap_gdt(int cpu
)
503 struct desc_ptr gdt_descr
;
505 gdt_descr
.address
= (long)get_cpu_gdt_ro(cpu
);
506 gdt_descr
.size
= GDT_SIZE
- 1;
507 load_gdt(&gdt_descr
);
509 EXPORT_SYMBOL_GPL(load_fixmap_gdt
);
512 * Current gdt points %fs at the "master" per-cpu area: after this,
513 * it's on the real one.
515 void switch_to_new_gdt(int cpu
)
517 /* Load the original GDT */
518 load_direct_gdt(cpu
);
519 /* Reload the per-cpu base */
520 load_percpu_segment(cpu
);
523 static const struct cpu_dev
*cpu_devs
[X86_VENDOR_NUM
] = {};
525 static void get_model_name(struct cpuinfo_x86
*c
)
530 if (c
->extended_cpuid_level
< 0x80000004)
533 v
= (unsigned int *)c
->x86_model_id
;
534 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
535 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
536 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
537 c
->x86_model_id
[48] = 0;
539 /* Trim whitespace */
540 p
= q
= s
= &c
->x86_model_id
[0];
546 /* Note the last non-whitespace index */
556 void cpu_detect_cache_sizes(struct cpuinfo_x86
*c
)
558 unsigned int n
, dummy
, ebx
, ecx
, edx
, l2size
;
560 n
= c
->extended_cpuid_level
;
562 if (n
>= 0x80000005) {
563 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
564 c
->x86_cache_size
= (ecx
>>24) + (edx
>>24);
566 /* On K8 L1 TLB is inclusive, so don't count it */
571 if (n
< 0x80000006) /* Some chips just has a large L1. */
574 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
578 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
580 /* do processor-specific cache resizing */
581 if (this_cpu
->legacy_cache_size
)
582 l2size
= this_cpu
->legacy_cache_size(c
, l2size
);
584 /* Allow user to override all this if necessary. */
585 if (cachesize_override
!= -1)
586 l2size
= cachesize_override
;
589 return; /* Again, no L2 cache is possible */
592 c
->x86_cache_size
= l2size
;
595 u16 __read_mostly tlb_lli_4k
[NR_INFO
];
596 u16 __read_mostly tlb_lli_2m
[NR_INFO
];
597 u16 __read_mostly tlb_lli_4m
[NR_INFO
];
598 u16 __read_mostly tlb_lld_4k
[NR_INFO
];
599 u16 __read_mostly tlb_lld_2m
[NR_INFO
];
600 u16 __read_mostly tlb_lld_4m
[NR_INFO
];
601 u16 __read_mostly tlb_lld_1g
[NR_INFO
];
603 static void cpu_detect_tlb(struct cpuinfo_x86
*c
)
605 if (this_cpu
->c_detect_tlb
)
606 this_cpu
->c_detect_tlb(c
);
608 pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
609 tlb_lli_4k
[ENTRIES
], tlb_lli_2m
[ENTRIES
],
610 tlb_lli_4m
[ENTRIES
]);
612 pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
613 tlb_lld_4k
[ENTRIES
], tlb_lld_2m
[ENTRIES
],
614 tlb_lld_4m
[ENTRIES
], tlb_lld_1g
[ENTRIES
]);
617 void detect_ht(struct cpuinfo_x86
*c
)
620 u32 eax
, ebx
, ecx
, edx
;
621 int index_msb
, core_bits
;
624 if (!cpu_has(c
, X86_FEATURE_HT
))
627 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
630 if (cpu_has(c
, X86_FEATURE_XTOPOLOGY
))
633 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
635 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
637 if (smp_num_siblings
== 1) {
638 pr_info_once("CPU0: Hyper-Threading is disabled\n");
642 if (smp_num_siblings
<= 1)
645 index_msb
= get_count_order(smp_num_siblings
);
646 c
->phys_proc_id
= apic
->phys_pkg_id(c
->initial_apicid
, index_msb
);
648 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
650 index_msb
= get_count_order(smp_num_siblings
);
652 core_bits
= get_count_order(c
->x86_max_cores
);
654 c
->cpu_core_id
= apic
->phys_pkg_id(c
->initial_apicid
, index_msb
) &
655 ((1 << core_bits
) - 1);
658 if (!printed
&& (c
->x86_max_cores
* smp_num_siblings
) > 1) {
659 pr_info("CPU: Physical Processor ID: %d\n",
661 pr_info("CPU: Processor Core ID: %d\n",
668 static void get_cpu_vendor(struct cpuinfo_x86
*c
)
670 char *v
= c
->x86_vendor_id
;
673 for (i
= 0; i
< X86_VENDOR_NUM
; i
++) {
677 if (!strcmp(v
, cpu_devs
[i
]->c_ident
[0]) ||
678 (cpu_devs
[i
]->c_ident
[1] &&
679 !strcmp(v
, cpu_devs
[i
]->c_ident
[1]))) {
681 this_cpu
= cpu_devs
[i
];
682 c
->x86_vendor
= this_cpu
->c_x86_vendor
;
687 pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
688 "CPU: Your system may be unstable.\n", v
);
690 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
691 this_cpu
= &default_cpu
;
694 void cpu_detect(struct cpuinfo_x86
*c
)
696 /* Get vendor name */
697 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
698 (unsigned int *)&c
->x86_vendor_id
[0],
699 (unsigned int *)&c
->x86_vendor_id
[8],
700 (unsigned int *)&c
->x86_vendor_id
[4]);
703 /* Intel-defined flags: level 0x00000001 */
704 if (c
->cpuid_level
>= 0x00000001) {
705 u32 junk
, tfms
, cap0
, misc
;
707 cpuid(0x00000001, &tfms
, &misc
, &junk
, &cap0
);
708 c
->x86
= x86_family(tfms
);
709 c
->x86_model
= x86_model(tfms
);
710 c
->x86_mask
= x86_stepping(tfms
);
712 if (cap0
& (1<<19)) {
713 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
714 c
->x86_cache_alignment
= c
->x86_clflush_size
;
719 static void apply_forced_caps(struct cpuinfo_x86
*c
)
723 for (i
= 0; i
< NCAPINTS
+ NBUGINTS
; i
++) {
724 c
->x86_capability
[i
] &= ~cpu_caps_cleared
[i
];
725 c
->x86_capability
[i
] |= cpu_caps_set
[i
];
729 void get_cpu_cap(struct cpuinfo_x86
*c
)
731 u32 eax
, ebx
, ecx
, edx
;
733 /* Intel-defined flags: level 0x00000001 */
734 if (c
->cpuid_level
>= 0x00000001) {
735 cpuid(0x00000001, &eax
, &ebx
, &ecx
, &edx
);
737 c
->x86_capability
[CPUID_1_ECX
] = ecx
;
738 c
->x86_capability
[CPUID_1_EDX
] = edx
;
741 /* Thermal and Power Management Leaf: level 0x00000006 (eax) */
742 if (c
->cpuid_level
>= 0x00000006)
743 c
->x86_capability
[CPUID_6_EAX
] = cpuid_eax(0x00000006);
745 /* Additional Intel-defined flags: level 0x00000007 */
746 if (c
->cpuid_level
>= 0x00000007) {
747 cpuid_count(0x00000007, 0, &eax
, &ebx
, &ecx
, &edx
);
748 c
->x86_capability
[CPUID_7_0_EBX
] = ebx
;
749 c
->x86_capability
[CPUID_7_ECX
] = ecx
;
752 /* Extended state features: level 0x0000000d */
753 if (c
->cpuid_level
>= 0x0000000d) {
754 cpuid_count(0x0000000d, 1, &eax
, &ebx
, &ecx
, &edx
);
756 c
->x86_capability
[CPUID_D_1_EAX
] = eax
;
759 /* Additional Intel-defined flags: level 0x0000000F */
760 if (c
->cpuid_level
>= 0x0000000F) {
762 /* QoS sub-leaf, EAX=0Fh, ECX=0 */
763 cpuid_count(0x0000000F, 0, &eax
, &ebx
, &ecx
, &edx
);
764 c
->x86_capability
[CPUID_F_0_EDX
] = edx
;
766 if (cpu_has(c
, X86_FEATURE_CQM_LLC
)) {
767 /* will be overridden if occupancy monitoring exists */
768 c
->x86_cache_max_rmid
= ebx
;
770 /* QoS sub-leaf, EAX=0Fh, ECX=1 */
771 cpuid_count(0x0000000F, 1, &eax
, &ebx
, &ecx
, &edx
);
772 c
->x86_capability
[CPUID_F_1_EDX
] = edx
;
774 if ((cpu_has(c
, X86_FEATURE_CQM_OCCUP_LLC
)) ||
775 ((cpu_has(c
, X86_FEATURE_CQM_MBM_TOTAL
)) ||
776 (cpu_has(c
, X86_FEATURE_CQM_MBM_LOCAL
)))) {
777 c
->x86_cache_max_rmid
= ecx
;
778 c
->x86_cache_occ_scale
= ebx
;
781 c
->x86_cache_max_rmid
= -1;
782 c
->x86_cache_occ_scale
= -1;
786 /* AMD-defined flags: level 0x80000001 */
787 eax
= cpuid_eax(0x80000000);
788 c
->extended_cpuid_level
= eax
;
790 if ((eax
& 0xffff0000) == 0x80000000) {
791 if (eax
>= 0x80000001) {
792 cpuid(0x80000001, &eax
, &ebx
, &ecx
, &edx
);
794 c
->x86_capability
[CPUID_8000_0001_ECX
] = ecx
;
795 c
->x86_capability
[CPUID_8000_0001_EDX
] = edx
;
799 if (c
->extended_cpuid_level
>= 0x80000007) {
800 cpuid(0x80000007, &eax
, &ebx
, &ecx
, &edx
);
802 c
->x86_capability
[CPUID_8000_0007_EBX
] = ebx
;
806 if (c
->extended_cpuid_level
>= 0x80000008) {
807 cpuid(0x80000008, &eax
, &ebx
, &ecx
, &edx
);
809 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
810 c
->x86_phys_bits
= eax
& 0xff;
811 c
->x86_capability
[CPUID_8000_0008_EBX
] = ebx
;
814 else if (cpu_has(c
, X86_FEATURE_PAE
) || cpu_has(c
, X86_FEATURE_PSE36
))
815 c
->x86_phys_bits
= 36;
818 if (c
->extended_cpuid_level
>= 0x8000000a)
819 c
->x86_capability
[CPUID_8000_000A_EDX
] = cpuid_edx(0x8000000a);
821 init_scattered_cpuid_features(c
);
824 * Clear/Set all flags overridden by options, after probe.
825 * This needs to happen each time we re-probe, which may happen
826 * several times during CPU initialization.
828 apply_forced_caps(c
);
831 static void identify_cpu_without_cpuid(struct cpuinfo_x86
*c
)
837 * First of all, decide if this is a 486 or higher
838 * It's a 486 if we can modify the AC flag
840 if (flag_is_changeable_p(X86_EFLAGS_AC
))
845 for (i
= 0; i
< X86_VENDOR_NUM
; i
++)
846 if (cpu_devs
[i
] && cpu_devs
[i
]->c_identify
) {
847 c
->x86_vendor_id
[0] = 0;
848 cpu_devs
[i
]->c_identify(c
);
849 if (c
->x86_vendor_id
[0]) {
857 static const __initdata
struct x86_cpu_id cpu_no_speculation
[] = {
858 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_CEDARVIEW
, X86_FEATURE_ANY
},
859 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_CLOVERVIEW
, X86_FEATURE_ANY
},
860 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_LINCROFT
, X86_FEATURE_ANY
},
861 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_PENWELL
, X86_FEATURE_ANY
},
862 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_PINEVIEW
, X86_FEATURE_ANY
},
863 { X86_VENDOR_CENTAUR
, 5 },
864 { X86_VENDOR_INTEL
, 5 },
865 { X86_VENDOR_NSC
, 5 },
866 { X86_VENDOR_ANY
, 4 },
870 static const __initdata
struct x86_cpu_id cpu_no_meltdown
[] = {
875 static const __initconst
struct x86_cpu_id cpu_no_spec_store_bypass
[] = {
876 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_PINEVIEW
},
877 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_LINCROFT
},
878 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_PENWELL
},
879 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_CLOVERVIEW
},
880 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_CEDARVIEW
},
881 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_SILVERMONT1
},
882 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_AIRMONT
},
883 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_SILVERMONT2
},
884 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_ATOM_MERRIFIELD
},
885 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_CORE_YONAH
},
886 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_XEON_PHI_KNL
},
887 { X86_VENDOR_INTEL
, 6, INTEL_FAM6_XEON_PHI_KNM
},
888 { X86_VENDOR_CENTAUR
, 5, },
889 { X86_VENDOR_INTEL
, 5, },
890 { X86_VENDOR_NSC
, 5, },
891 { X86_VENDOR_AMD
, 0x12, },
892 { X86_VENDOR_AMD
, 0x11, },
893 { X86_VENDOR_AMD
, 0x10, },
894 { X86_VENDOR_AMD
, 0xf, },
895 { X86_VENDOR_ANY
, 4, },
899 static void __init
cpu_set_bug_bits(struct cpuinfo_x86
*c
)
903 if (cpu_has(c
, X86_FEATURE_ARCH_CAPABILITIES
))
904 rdmsrl(MSR_IA32_ARCH_CAPABILITIES
, ia32_cap
);
906 if (!x86_match_cpu(cpu_no_spec_store_bypass
) &&
907 !(ia32_cap
& ARCH_CAP_SSBD_NO
))
908 setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS
);
910 if (x86_match_cpu(cpu_no_speculation
))
913 setup_force_cpu_bug(X86_BUG_SPECTRE_V1
);
914 setup_force_cpu_bug(X86_BUG_SPECTRE_V2
);
916 if (x86_match_cpu(cpu_no_meltdown
))
919 /* Rogue Data Cache Load? No! */
920 if (ia32_cap
& ARCH_CAP_RDCL_NO
)
923 setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN
);
927 * Do minimum CPU detection early.
928 * Fields really needed: vendor, cpuid_level, family, model, mask,
930 * The others are not touched to avoid unwanted side effects.
932 * WARNING: this function is only called on the BP. Don't add code here
933 * that is supposed to run on all CPUs.
935 static void __init
early_identify_cpu(struct cpuinfo_x86
*c
)
938 c
->x86_clflush_size
= 64;
939 c
->x86_phys_bits
= 36;
940 c
->x86_virt_bits
= 48;
942 c
->x86_clflush_size
= 32;
943 c
->x86_phys_bits
= 32;
944 c
->x86_virt_bits
= 32;
946 c
->x86_cache_alignment
= c
->x86_clflush_size
;
948 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
949 c
->extended_cpuid_level
= 0;
951 /* cyrix could have cpuid enabled via c_identify()*/
952 if (have_cpuid_p()) {
956 setup_force_cpu_cap(X86_FEATURE_CPUID
);
958 if (this_cpu
->c_early_init
)
959 this_cpu
->c_early_init(c
);
962 filter_cpuid_features(c
, false);
964 if (this_cpu
->c_bsp_init
)
965 this_cpu
->c_bsp_init(c
);
967 identify_cpu_without_cpuid(c
);
968 setup_clear_cpu_cap(X86_FEATURE_CPUID
);
971 setup_force_cpu_cap(X86_FEATURE_ALWAYS
);
978 void __init
early_cpu_init(void)
980 const struct cpu_dev
*const *cdev
;
983 #ifdef CONFIG_PROCESSOR_SELECT
984 pr_info("KERNEL supported cpus:\n");
987 for (cdev
= __x86_cpu_dev_start
; cdev
< __x86_cpu_dev_end
; cdev
++) {
988 const struct cpu_dev
*cpudev
= *cdev
;
990 if (count
>= X86_VENDOR_NUM
)
992 cpu_devs
[count
] = cpudev
;
995 #ifdef CONFIG_PROCESSOR_SELECT
999 for (j
= 0; j
< 2; j
++) {
1000 if (!cpudev
->c_ident
[j
])
1002 pr_info(" %s %s\n", cpudev
->c_vendor
,
1003 cpudev
->c_ident
[j
]);
1008 early_identify_cpu(&boot_cpu_data
);
1012 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1013 * unfortunately, that's not true in practice because of early VIA
1014 * chips and (more importantly) broken virtualizers that are not easy
1015 * to detect. In the latter case it doesn't even *fail* reliably, so
1016 * probing for it doesn't even work. Disable it completely on 32-bit
1017 * unless we can find a reliable way to detect all the broken cases.
1018 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1020 static void detect_nopl(struct cpuinfo_x86
*c
)
1022 #ifdef CONFIG_X86_32
1023 clear_cpu_cap(c
, X86_FEATURE_NOPL
);
1025 set_cpu_cap(c
, X86_FEATURE_NOPL
);
1029 static void detect_null_seg_behavior(struct cpuinfo_x86
*c
)
1031 #ifdef CONFIG_X86_64
1033 * Empirically, writing zero to a segment selector on AMD does
1034 * not clear the base, whereas writing zero to a segment
1035 * selector on Intel does clear the base. Intel's behavior
1036 * allows slightly faster context switches in the common case
1037 * where GS is unused by the prev and next threads.
1039 * Since neither vendor documents this anywhere that I can see,
1040 * detect it directly instead of hardcoding the choice by
1043 * I've designated AMD's behavior as the "bug" because it's
1044 * counterintuitive and less friendly.
1047 unsigned long old_base
, tmp
;
1048 rdmsrl(MSR_FS_BASE
, old_base
);
1049 wrmsrl(MSR_FS_BASE
, 1);
1051 rdmsrl(MSR_FS_BASE
, tmp
);
1053 set_cpu_bug(c
, X86_BUG_NULL_SEG
);
1054 wrmsrl(MSR_FS_BASE
, old_base
);
1058 static void generic_identify(struct cpuinfo_x86
*c
)
1060 c
->extended_cpuid_level
= 0;
1062 if (!have_cpuid_p())
1063 identify_cpu_without_cpuid(c
);
1065 /* cyrix could have cpuid enabled via c_identify()*/
1066 if (!have_cpuid_p())
1075 if (c
->cpuid_level
>= 0x00000001) {
1076 c
->initial_apicid
= (cpuid_ebx(1) >> 24) & 0xFF;
1077 #ifdef CONFIG_X86_32
1079 c
->apicid
= apic
->phys_pkg_id(c
->initial_apicid
, 0);
1081 c
->apicid
= c
->initial_apicid
;
1084 c
->phys_proc_id
= c
->initial_apicid
;
1087 get_model_name(c
); /* Default name */
1091 detect_null_seg_behavior(c
);
1094 * ESPFIX is a strange bug. All real CPUs have it. Paravirt
1095 * systems that run Linux at CPL > 0 may or may not have the
1096 * issue, but, even if they have the issue, there's absolutely
1097 * nothing we can do about it because we can't use the real IRET
1100 * NB: For the time being, only 32-bit kernels support
1101 * X86_BUG_ESPFIX as such. 64-bit kernels directly choose
1102 * whether to apply espfix using paravirt hooks. If any
1103 * non-paravirt system ever shows up that does *not* have the
1104 * ESPFIX issue, we can change this.
1106 #ifdef CONFIG_X86_32
1107 # ifdef CONFIG_PARAVIRT
1109 extern void native_iret(void);
1110 if (pv_cpu_ops
.iret
== native_iret
)
1111 set_cpu_bug(c
, X86_BUG_ESPFIX
);
1114 set_cpu_bug(c
, X86_BUG_ESPFIX
);
1119 static void x86_init_cache_qos(struct cpuinfo_x86
*c
)
1122 * The heavy lifting of max_rmid and cache_occ_scale are handled
1123 * in get_cpu_cap(). Here we just set the max_rmid for the boot_cpu
1124 * in case CQM bits really aren't there in this CPU.
1126 if (c
!= &boot_cpu_data
) {
1127 boot_cpu_data
.x86_cache_max_rmid
=
1128 min(boot_cpu_data
.x86_cache_max_rmid
,
1129 c
->x86_cache_max_rmid
);
1134 * Validate that ACPI/mptables have the same information about the
1135 * effective APIC id and update the package map.
1137 static void validate_apic_and_package_id(struct cpuinfo_x86
*c
)
1140 unsigned int apicid
, cpu
= smp_processor_id();
1142 apicid
= apic
->cpu_present_to_apicid(cpu
);
1144 if (apicid
!= c
->apicid
) {
1145 pr_err(FW_BUG
"CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1146 cpu
, apicid
, c
->initial_apicid
);
1148 BUG_ON(topology_update_package_map(c
->phys_proc_id
, cpu
));
1150 c
->logical_proc_id
= 0;
1155 * This does the hard work of actually picking apart the CPU stuff...
1157 static void identify_cpu(struct cpuinfo_x86
*c
)
1161 c
->loops_per_jiffy
= loops_per_jiffy
;
1162 c
->x86_cache_size
= -1;
1163 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
1164 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
1165 c
->x86_vendor_id
[0] = '\0'; /* Unset */
1166 c
->x86_model_id
[0] = '\0'; /* Unset */
1167 c
->x86_max_cores
= 1;
1168 c
->x86_coreid_bits
= 0;
1170 #ifdef CONFIG_X86_64
1171 c
->x86_clflush_size
= 64;
1172 c
->x86_phys_bits
= 36;
1173 c
->x86_virt_bits
= 48;
1175 c
->cpuid_level
= -1; /* CPUID not detected */
1176 c
->x86_clflush_size
= 32;
1177 c
->x86_phys_bits
= 32;
1178 c
->x86_virt_bits
= 32;
1180 c
->x86_cache_alignment
= c
->x86_clflush_size
;
1181 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
1183 generic_identify(c
);
1185 if (this_cpu
->c_identify
)
1186 this_cpu
->c_identify(c
);
1188 /* Clear/Set all flags overridden by options, after probe */
1189 apply_forced_caps(c
);
1191 #ifdef CONFIG_X86_64
1192 c
->apicid
= apic
->phys_pkg_id(c
->initial_apicid
, 0);
1196 * Vendor-specific initialization. In this section we
1197 * canonicalize the feature flags, meaning if there are
1198 * features a certain CPU supports which CPUID doesn't
1199 * tell us, CPUID claiming incorrect flags, or other bugs,
1200 * we handle them here.
1202 * At the end of this section, c->x86_capability better
1203 * indicate the features this CPU genuinely supports!
1205 if (this_cpu
->c_init
)
1206 this_cpu
->c_init(c
);
1208 /* Disable the PN if appropriate */
1209 squash_the_stupid_serial_number(c
);
1211 /* Set up SMEP/SMAP */
1216 * The vendor-specific functions might have changed features.
1217 * Now we do "generic changes."
1220 /* Filter out anything that depends on CPUID levels we don't have */
1221 filter_cpuid_features(c
, true);
1223 /* If the model name is still unset, do table lookup. */
1224 if (!c
->x86_model_id
[0]) {
1226 p
= table_lookup_model(c
);
1228 strcpy(c
->x86_model_id
, p
);
1230 /* Last resort... */
1231 sprintf(c
->x86_model_id
, "%02x/%02x",
1232 c
->x86
, c
->x86_model
);
1235 #ifdef CONFIG_X86_64
1240 x86_init_cache_qos(c
);
1244 * Clear/Set all flags overridden by options, need do it
1245 * before following smp all cpus cap AND.
1247 apply_forced_caps(c
);
1250 * On SMP, boot_cpu_data holds the common feature set between
1251 * all CPUs; so make sure that we indicate which features are
1252 * common between the CPUs. The first time this routine gets
1253 * executed, c == &boot_cpu_data.
1255 if (c
!= &boot_cpu_data
) {
1256 /* AND the already accumulated flags with these */
1257 for (i
= 0; i
< NCAPINTS
; i
++)
1258 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
1260 /* OR, i.e. replicate the bug flags */
1261 for (i
= NCAPINTS
; i
< NCAPINTS
+ NBUGINTS
; i
++)
1262 c
->x86_capability
[i
] |= boot_cpu_data
.x86_capability
[i
];
1265 /* Init Machine Check Exception if available. */
1268 select_idle_routine(c
);
1271 numa_add_cpu(smp_processor_id());
1276 * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1277 * on 32-bit kernels:
1279 #ifdef CONFIG_X86_32
1280 void enable_sep_cpu(void)
1282 struct tss_struct
*tss
;
1285 if (!boot_cpu_has(X86_FEATURE_SEP
))
1289 tss
= &per_cpu(cpu_tss_rw
, cpu
);
1292 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1293 * see the big comment in struct x86_hw_tss's definition.
1296 tss
->x86_tss
.ss1
= __KERNEL_CS
;
1297 wrmsr(MSR_IA32_SYSENTER_CS
, tss
->x86_tss
.ss1
, 0);
1298 wrmsr(MSR_IA32_SYSENTER_ESP
, (unsigned long)(cpu_entry_stack(cpu
) + 1), 0);
1299 wrmsr(MSR_IA32_SYSENTER_EIP
, (unsigned long)entry_SYSENTER_32
, 0);
1305 void __init
identify_boot_cpu(void)
1307 identify_cpu(&boot_cpu_data
);
1308 #ifdef CONFIG_X86_32
1312 cpu_detect_tlb(&boot_cpu_data
);
1315 void identify_secondary_cpu(struct cpuinfo_x86
*c
)
1317 BUG_ON(c
== &boot_cpu_data
);
1319 #ifdef CONFIG_X86_32
1323 validate_apic_and_package_id(c
);
1324 x86_spec_ctrl_setup_ap();
1327 static __init
int setup_noclflush(char *arg
)
1329 setup_clear_cpu_cap(X86_FEATURE_CLFLUSH
);
1330 setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT
);
1333 __setup("noclflush", setup_noclflush
);
1335 void print_cpu_info(struct cpuinfo_x86
*c
)
1337 const char *vendor
= NULL
;
1339 if (c
->x86_vendor
< X86_VENDOR_NUM
) {
1340 vendor
= this_cpu
->c_vendor
;
1342 if (c
->cpuid_level
>= 0)
1343 vendor
= c
->x86_vendor_id
;
1346 if (vendor
&& !strstr(c
->x86_model_id
, vendor
))
1347 pr_cont("%s ", vendor
);
1349 if (c
->x86_model_id
[0])
1350 pr_cont("%s", c
->x86_model_id
);
1352 pr_cont("%d86", c
->x86
);
1354 pr_cont(" (family: 0x%x, model: 0x%x", c
->x86
, c
->x86_model
);
1356 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1357 pr_cont(", stepping: 0x%x)\n", c
->x86_mask
);
1363 * clearcpuid= was already parsed in fpu__init_parse_early_param.
1364 * But we need to keep a dummy __setup around otherwise it would
1365 * show up as an environment variable for init.
1367 static __init
int setup_clearcpuid(char *arg
)
1371 __setup("clearcpuid=", setup_clearcpuid
);
1373 #ifdef CONFIG_X86_64
1374 struct desc_ptr idt_descr __ro_after_init
= {
1375 .size
= NR_VECTORS
* 16 - 1,
1376 .address
= (unsigned long) idt_table
,
1378 const struct desc_ptr debug_idt_descr
= {
1379 .size
= NR_VECTORS
* 16 - 1,
1380 .address
= (unsigned long) debug_idt_table
,
1383 DEFINE_PER_CPU_FIRST(union irq_stack_union
,
1384 irq_stack_union
) __aligned(PAGE_SIZE
) __visible
;
1387 * The following percpu variables are hot. Align current_task to
1388 * cacheline size such that they fall in the same cacheline.
1390 DEFINE_PER_CPU(struct task_struct
*, current_task
) ____cacheline_aligned
=
1392 EXPORT_PER_CPU_SYMBOL(current_task
);
1394 DEFINE_PER_CPU(char *, irq_stack_ptr
) =
1395 init_per_cpu_var(irq_stack_union
.irq_stack
) + IRQ_STACK_SIZE
;
1397 DEFINE_PER_CPU(unsigned int, irq_count
) __visible
= -1;
1399 DEFINE_PER_CPU(int, __preempt_count
) = INIT_PREEMPT_COUNT
;
1400 EXPORT_PER_CPU_SYMBOL(__preempt_count
);
1402 /* May not be marked __init: used by software suspend */
1403 void syscall_init(void)
1405 extern char _entry_trampoline
[];
1406 extern char entry_SYSCALL_64_trampoline
[];
1408 int cpu
= smp_processor_id();
1409 unsigned long SYSCALL64_entry_trampoline
=
1410 (unsigned long)get_cpu_entry_area(cpu
)->entry_trampoline
+
1411 (entry_SYSCALL_64_trampoline
- _entry_trampoline
);
1413 wrmsr(MSR_STAR
, 0, (__USER32_CS
<< 16) | __KERNEL_CS
);
1414 if (static_cpu_has(X86_FEATURE_PTI
))
1415 wrmsrl(MSR_LSTAR
, SYSCALL64_entry_trampoline
);
1417 wrmsrl(MSR_LSTAR
, (unsigned long)entry_SYSCALL_64
);
1419 #ifdef CONFIG_IA32_EMULATION
1420 wrmsrl(MSR_CSTAR
, (unsigned long)entry_SYSCALL_compat
);
1422 * This only works on Intel CPUs.
1423 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1424 * This does not cause SYSENTER to jump to the wrong location, because
1425 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1427 wrmsrl_safe(MSR_IA32_SYSENTER_CS
, (u64
)__KERNEL_CS
);
1428 wrmsrl_safe(MSR_IA32_SYSENTER_ESP
, (unsigned long)(cpu_entry_stack(cpu
) + 1));
1429 wrmsrl_safe(MSR_IA32_SYSENTER_EIP
, (u64
)entry_SYSENTER_compat
);
1431 wrmsrl(MSR_CSTAR
, (unsigned long)ignore_sysret
);
1432 wrmsrl_safe(MSR_IA32_SYSENTER_CS
, (u64
)GDT_ENTRY_INVALID_SEG
);
1433 wrmsrl_safe(MSR_IA32_SYSENTER_ESP
, 0ULL);
1434 wrmsrl_safe(MSR_IA32_SYSENTER_EIP
, 0ULL);
1437 /* Flags to clear on syscall */
1438 wrmsrl(MSR_SYSCALL_MASK
,
1439 X86_EFLAGS_TF
|X86_EFLAGS_DF
|X86_EFLAGS_IF
|
1440 X86_EFLAGS_IOPL
|X86_EFLAGS_AC
|X86_EFLAGS_NT
);
1444 * Copies of the original ist values from the tss are only accessed during
1445 * debugging, no special alignment required.
1447 DEFINE_PER_CPU(struct orig_ist
, orig_ist
);
1449 static DEFINE_PER_CPU(unsigned long, debug_stack_addr
);
1450 DEFINE_PER_CPU(int, debug_stack_usage
);
1452 int is_debug_stack(unsigned long addr
)
1454 return __this_cpu_read(debug_stack_usage
) ||
1455 (addr
<= __this_cpu_read(debug_stack_addr
) &&
1456 addr
> (__this_cpu_read(debug_stack_addr
) - DEBUG_STKSZ
));
1458 NOKPROBE_SYMBOL(is_debug_stack
);
1460 DEFINE_PER_CPU(u32
, debug_idt_ctr
);
1462 void debug_stack_set_zero(void)
1464 this_cpu_inc(debug_idt_ctr
);
1467 NOKPROBE_SYMBOL(debug_stack_set_zero
);
1469 void debug_stack_reset(void)
1471 if (WARN_ON(!this_cpu_read(debug_idt_ctr
)))
1473 if (this_cpu_dec_return(debug_idt_ctr
) == 0)
1476 NOKPROBE_SYMBOL(debug_stack_reset
);
1478 #else /* CONFIG_X86_64 */
1480 DEFINE_PER_CPU(struct task_struct
*, current_task
) = &init_task
;
1481 EXPORT_PER_CPU_SYMBOL(current_task
);
1482 DEFINE_PER_CPU(int, __preempt_count
) = INIT_PREEMPT_COUNT
;
1483 EXPORT_PER_CPU_SYMBOL(__preempt_count
);
1486 * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1487 * the top of the kernel stack. Use an extra percpu variable to track the
1488 * top of the kernel stack directly.
1490 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack
) =
1491 (unsigned long)&init_thread_union
+ THREAD_SIZE
;
1492 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack
);
1494 #ifdef CONFIG_CC_STACKPROTECTOR
1495 DEFINE_PER_CPU_ALIGNED(struct stack_canary
, stack_canary
);
1498 #endif /* CONFIG_X86_64 */
1501 * Clear all 6 debug registers:
1503 static void clear_all_debug_regs(void)
1507 for (i
= 0; i
< 8; i
++) {
1508 /* Ignore db4, db5 */
1509 if ((i
== 4) || (i
== 5))
1518 * Restore debug regs if using kgdbwait and you have a kernel debugger
1519 * connection established.
1521 static void dbg_restore_debug_regs(void)
1523 if (unlikely(kgdb_connected
&& arch_kgdb_ops
.correct_hw_break
))
1524 arch_kgdb_ops
.correct_hw_break();
1526 #else /* ! CONFIG_KGDB */
1527 #define dbg_restore_debug_regs()
1528 #endif /* ! CONFIG_KGDB */
1530 static void wait_for_master_cpu(int cpu
)
1534 * wait for ACK from master CPU before continuing
1535 * with AP initialization
1537 WARN_ON(cpumask_test_and_set_cpu(cpu
, cpu_initialized_mask
));
1538 while (!cpumask_test_cpu(cpu
, cpu_callout_mask
))
1544 * cpu_init() initializes state that is per-CPU. Some data is already
1545 * initialized (naturally) in the bootstrap process, such as the GDT
1546 * and IDT. We reload them nevertheless, this function acts as a
1547 * 'CPU state barrier', nothing should get across.
1548 * A lot of state is already set up in PDA init for 64 bit
1550 #ifdef CONFIG_X86_64
1554 struct orig_ist
*oist
;
1555 struct task_struct
*me
;
1556 struct tss_struct
*t
;
1558 int cpu
= raw_smp_processor_id();
1561 wait_for_master_cpu(cpu
);
1564 * Initialize the CR4 shadow before doing anything that could
1572 t
= &per_cpu(cpu_tss_rw
, cpu
);
1573 oist
= &per_cpu(orig_ist
, cpu
);
1576 if (this_cpu_read(numa_node
) == 0 &&
1577 early_cpu_to_node(cpu
) != NUMA_NO_NODE
)
1578 set_numa_node(early_cpu_to_node(cpu
));
1583 pr_debug("Initializing CPU#%d\n", cpu
);
1585 cr4_clear_bits(X86_CR4_VME
|X86_CR4_PVI
|X86_CR4_TSD
|X86_CR4_DE
);
1588 * Initialize the per-CPU GDT with the boot GDT,
1589 * and set up the GDT descriptor:
1592 switch_to_new_gdt(cpu
);
1597 memset(me
->thread
.tls_array
, 0, GDT_ENTRY_TLS_ENTRIES
* 8);
1600 wrmsrl(MSR_FS_BASE
, 0);
1601 wrmsrl(MSR_KERNEL_GS_BASE
, 0);
1608 * set up and load the per-CPU TSS
1610 if (!oist
->ist
[0]) {
1611 char *estacks
= get_cpu_entry_area(cpu
)->exception_stacks
;
1613 for (v
= 0; v
< N_EXCEPTION_STACKS
; v
++) {
1614 estacks
+= exception_stack_sizes
[v
];
1615 oist
->ist
[v
] = t
->x86_tss
.ist
[v
] =
1616 (unsigned long)estacks
;
1617 if (v
== DEBUG_STACK
-1)
1618 per_cpu(debug_stack_addr
, cpu
) = (unsigned long)estacks
;
1622 t
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
1625 * <= is required because the CPU will access up to
1626 * 8 bits beyond the end of the IO permission bitmap.
1628 for (i
= 0; i
<= IO_BITMAP_LONGS
; i
++)
1629 t
->io_bitmap
[i
] = ~0UL;
1632 me
->active_mm
= &init_mm
;
1634 initialize_tlbstate_and_flush();
1635 enter_lazy_tlb(&init_mm
, me
);
1638 * Initialize the TSS. sp0 points to the entry trampoline stack
1639 * regardless of what task is running.
1641 set_tss_desc(cpu
, &get_cpu_entry_area(cpu
)->tss
.x86_tss
);
1643 load_sp0((unsigned long)(cpu_entry_stack(cpu
) + 1));
1645 load_mm_ldt(&init_mm
);
1647 clear_all_debug_regs();
1648 dbg_restore_debug_regs();
1655 load_fixmap_gdt(cpu
);
1662 int cpu
= smp_processor_id();
1663 struct task_struct
*curr
= current
;
1664 struct tss_struct
*t
= &per_cpu(cpu_tss_rw
, cpu
);
1666 wait_for_master_cpu(cpu
);
1669 * Initialize the CR4 shadow before doing anything that could
1674 show_ucode_info_early();
1676 pr_info("Initializing CPU#%d\n", cpu
);
1678 if (cpu_feature_enabled(X86_FEATURE_VME
) ||
1679 boot_cpu_has(X86_FEATURE_TSC
) ||
1680 boot_cpu_has(X86_FEATURE_DE
))
1681 cr4_clear_bits(X86_CR4_VME
|X86_CR4_PVI
|X86_CR4_TSD
|X86_CR4_DE
);
1684 switch_to_new_gdt(cpu
);
1687 * Set up and load the per-CPU TSS and LDT
1690 curr
->active_mm
= &init_mm
;
1692 initialize_tlbstate_and_flush();
1693 enter_lazy_tlb(&init_mm
, curr
);
1696 * Initialize the TSS. Don't bother initializing sp0, as the initial
1697 * task never enters user mode.
1699 set_tss_desc(cpu
, &get_cpu_entry_area(cpu
)->tss
.x86_tss
);
1702 load_mm_ldt(&init_mm
);
1704 t
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
1706 #ifdef CONFIG_DOUBLEFAULT
1707 /* Set up doublefault TSS pointer in the GDT */
1708 __set_tss_desc(cpu
, GDT_ENTRY_DOUBLEFAULT_TSS
, &doublefault_tss
);
1711 clear_all_debug_regs();
1712 dbg_restore_debug_regs();
1716 load_fixmap_gdt(cpu
);
1720 static void bsp_resume(void)
1722 if (this_cpu
->c_bsp_resume
)
1723 this_cpu
->c_bsp_resume(&boot_cpu_data
);
1726 static struct syscore_ops cpu_syscore_ops
= {
1727 .resume
= bsp_resume
,
1730 static int __init
init_cpu_syscore(void)
1732 register_syscore_ops(&cpu_syscore_ops
);
1735 core_initcall(init_cpu_syscore
);