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>
51 #ifdef CONFIG_X86_LOCAL_APIC
52 #include <asm/uv/uv.h>
57 u32 elf_hwcap2 __read_mostly
;
59 /* all of these masks are initialized in setup_cpu_local_masks() */
60 cpumask_var_t cpu_initialized_mask
;
61 cpumask_var_t cpu_callout_mask
;
62 cpumask_var_t cpu_callin_mask
;
64 /* representing cpus for which sibling maps can be computed */
65 cpumask_var_t cpu_sibling_setup_mask
;
67 /* correctly size the local cpu masks */
68 void __init
setup_cpu_local_masks(void)
70 alloc_bootmem_cpumask_var(&cpu_initialized_mask
);
71 alloc_bootmem_cpumask_var(&cpu_callin_mask
);
72 alloc_bootmem_cpumask_var(&cpu_callout_mask
);
73 alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask
);
76 static void default_init(struct cpuinfo_x86
*c
)
79 cpu_detect_cache_sizes(c
);
81 /* Not much we can do here... */
82 /* Check if at least it has cpuid */
83 if (c
->cpuid_level
== -1) {
84 /* No cpuid. It must be an ancient CPU */
86 strcpy(c
->x86_model_id
, "486");
88 strcpy(c
->x86_model_id
, "386");
93 static const struct cpu_dev default_cpu
= {
94 .c_init
= default_init
,
95 .c_vendor
= "Unknown",
96 .c_x86_vendor
= X86_VENDOR_UNKNOWN
,
99 static const struct cpu_dev
*this_cpu
= &default_cpu
;
101 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page
, gdt_page
) = { .gdt
= {
104 * We need valid kernel segments for data and code in long mode too
105 * IRET will check the segment types kkeil 2000/10/28
106 * Also sysret mandates a special GDT layout
108 * TLS descriptors are currently at a different place compared to i386.
109 * Hopefully nobody expects them at a fixed place (Wine?)
111 [GDT_ENTRY_KERNEL32_CS
] = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
112 [GDT_ENTRY_KERNEL_CS
] = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
113 [GDT_ENTRY_KERNEL_DS
] = GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
114 [GDT_ENTRY_DEFAULT_USER32_CS
] = GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
115 [GDT_ENTRY_DEFAULT_USER_DS
] = GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
116 [GDT_ENTRY_DEFAULT_USER_CS
] = GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
118 [GDT_ENTRY_KERNEL_CS
] = GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
119 [GDT_ENTRY_KERNEL_DS
] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
120 [GDT_ENTRY_DEFAULT_USER_CS
] = GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
121 [GDT_ENTRY_DEFAULT_USER_DS
] = GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
123 * Segments used for calling PnP BIOS have byte granularity.
124 * They code segments and data segments have fixed 64k limits,
125 * the transfer segment sizes are set at run time.
128 [GDT_ENTRY_PNPBIOS_CS32
] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
130 [GDT_ENTRY_PNPBIOS_CS16
] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
132 [GDT_ENTRY_PNPBIOS_DS
] = GDT_ENTRY_INIT(0x0092, 0, 0xffff),
134 [GDT_ENTRY_PNPBIOS_TS1
] = GDT_ENTRY_INIT(0x0092, 0, 0),
136 [GDT_ENTRY_PNPBIOS_TS2
] = GDT_ENTRY_INIT(0x0092, 0, 0),
138 * The APM segments have byte granularity and their bases
139 * are set at run time. All have 64k limits.
142 [GDT_ENTRY_APMBIOS_BASE
] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
144 [GDT_ENTRY_APMBIOS_BASE
+1] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
146 [GDT_ENTRY_APMBIOS_BASE
+2] = GDT_ENTRY_INIT(0x4092, 0, 0xffff),
148 [GDT_ENTRY_ESPFIX_SS
] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
149 [GDT_ENTRY_PERCPU
] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
150 GDT_STACK_CANARY_INIT
153 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page
);
155 static int __init
x86_mpx_setup(char *s
)
157 /* require an exact match without trailing characters */
161 /* do not emit a message if the feature is not present */
162 if (!boot_cpu_has(X86_FEATURE_MPX
))
165 setup_clear_cpu_cap(X86_FEATURE_MPX
);
166 pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
169 __setup("nompx", x86_mpx_setup
);
172 static int __init
x86_nopcid_setup(char *s
)
174 /* nopcid doesn't accept parameters */
178 /* do not emit a message if the feature is not present */
179 if (!boot_cpu_has(X86_FEATURE_PCID
))
182 setup_clear_cpu_cap(X86_FEATURE_PCID
);
183 pr_info("nopcid: PCID feature disabled\n");
186 early_param("nopcid", x86_nopcid_setup
);
189 static int __init
x86_noinvpcid_setup(char *s
)
191 /* noinvpcid doesn't accept parameters */
195 /* do not emit a message if the feature is not present */
196 if (!boot_cpu_has(X86_FEATURE_INVPCID
))
199 setup_clear_cpu_cap(X86_FEATURE_INVPCID
);
200 pr_info("noinvpcid: INVPCID feature disabled\n");
203 early_param("noinvpcid", x86_noinvpcid_setup
);
206 static int cachesize_override
= -1;
207 static int disable_x86_serial_nr
= 1;
209 static int __init
cachesize_setup(char *str
)
211 get_option(&str
, &cachesize_override
);
214 __setup("cachesize=", cachesize_setup
);
216 static int __init
x86_sep_setup(char *s
)
218 setup_clear_cpu_cap(X86_FEATURE_SEP
);
221 __setup("nosep", x86_sep_setup
);
223 /* Standard macro to see if a specific flag is changeable */
224 static inline int flag_is_changeable_p(u32 flag
)
229 * Cyrix and IDT cpus allow disabling of CPUID
230 * so the code below may return different results
231 * when it is executed before and after enabling
232 * the CPUID. Add "volatile" to not allow gcc to
233 * optimize the subsequent calls to this function.
235 asm volatile ("pushfl \n\t"
246 : "=&r" (f1
), "=&r" (f2
)
249 return ((f1
^f2
) & flag
) != 0;
252 /* Probe for the CPUID instruction */
253 int have_cpuid_p(void)
255 return flag_is_changeable_p(X86_EFLAGS_ID
);
258 static void squash_the_stupid_serial_number(struct cpuinfo_x86
*c
)
260 unsigned long lo
, hi
;
262 if (!cpu_has(c
, X86_FEATURE_PN
) || !disable_x86_serial_nr
)
265 /* Disable processor serial number: */
267 rdmsr(MSR_IA32_BBL_CR_CTL
, lo
, hi
);
269 wrmsr(MSR_IA32_BBL_CR_CTL
, lo
, hi
);
271 pr_notice("CPU serial number disabled.\n");
272 clear_cpu_cap(c
, X86_FEATURE_PN
);
274 /* Disabling the serial number may affect the cpuid level */
275 c
->cpuid_level
= cpuid_eax(0);
278 static int __init
x86_serial_nr_setup(char *s
)
280 disable_x86_serial_nr
= 0;
283 __setup("serialnumber", x86_serial_nr_setup
);
285 static inline int flag_is_changeable_p(u32 flag
)
289 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86
*c
)
294 static __init
int setup_disable_smep(char *arg
)
296 setup_clear_cpu_cap(X86_FEATURE_SMEP
);
297 /* Check for things that depend on SMEP being enabled: */
298 check_mpx_erratum(&boot_cpu_data
);
301 __setup("nosmep", setup_disable_smep
);
303 static __always_inline
void setup_smep(struct cpuinfo_x86
*c
)
305 if (cpu_has(c
, X86_FEATURE_SMEP
))
306 cr4_set_bits(X86_CR4_SMEP
);
309 static __init
int setup_disable_smap(char *arg
)
311 setup_clear_cpu_cap(X86_FEATURE_SMAP
);
314 __setup("nosmap", setup_disable_smap
);
316 static __always_inline
void setup_smap(struct cpuinfo_x86
*c
)
318 unsigned long eflags
= native_save_fl();
320 /* This should have been cleared long ago */
321 BUG_ON(eflags
& X86_EFLAGS_AC
);
323 if (cpu_has(c
, X86_FEATURE_SMAP
)) {
324 #ifdef CONFIG_X86_SMAP
325 cr4_set_bits(X86_CR4_SMAP
);
327 cr4_clear_bits(X86_CR4_SMAP
);
333 * Protection Keys are not available in 32-bit mode.
335 static bool pku_disabled
;
337 static __always_inline
void setup_pku(struct cpuinfo_x86
*c
)
339 /* check the boot processor, plus compile options for PKU: */
340 if (!cpu_feature_enabled(X86_FEATURE_PKU
))
342 /* checks the actual processor's cpuid bits: */
343 if (!cpu_has(c
, X86_FEATURE_PKU
))
348 cr4_set_bits(X86_CR4_PKE
);
350 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
351 * cpuid bit to be set. We need to ensure that we
352 * update that bit in this CPU's "cpu_info".
357 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
358 static __init
int setup_disable_pku(char *arg
)
361 * Do not clear the X86_FEATURE_PKU bit. All of the
362 * runtime checks are against OSPKE so clearing the
365 * This way, we will see "pku" in cpuinfo, but not
366 * "ospke", which is exactly what we want. It shows
367 * that the CPU has PKU, but the OS has not enabled it.
368 * This happens to be exactly how a system would look
369 * if we disabled the config option.
371 pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
375 __setup("nopku", setup_disable_pku
);
376 #endif /* CONFIG_X86_64 */
379 * Some CPU features depend on higher CPUID levels, which may not always
380 * be available due to CPUID level capping or broken virtualization
381 * software. Add those features to this table to auto-disable them.
383 struct cpuid_dependent_feature
{
388 static const struct cpuid_dependent_feature
389 cpuid_dependent_features
[] = {
390 { X86_FEATURE_MWAIT
, 0x00000005 },
391 { X86_FEATURE_DCA
, 0x00000009 },
392 { X86_FEATURE_XSAVE
, 0x0000000d },
396 static void filter_cpuid_features(struct cpuinfo_x86
*c
, bool warn
)
398 const struct cpuid_dependent_feature
*df
;
400 for (df
= cpuid_dependent_features
; df
->feature
; df
++) {
402 if (!cpu_has(c
, df
->feature
))
405 * Note: cpuid_level is set to -1 if unavailable, but
406 * extended_extended_level is set to 0 if unavailable
407 * and the legitimate extended levels are all negative
408 * when signed; hence the weird messing around with
411 if (!((s32
)df
->level
< 0 ?
412 (u32
)df
->level
> (u32
)c
->extended_cpuid_level
:
413 (s32
)df
->level
> (s32
)c
->cpuid_level
))
416 clear_cpu_cap(c
, df
->feature
);
420 pr_warn("CPU: CPU feature " X86_CAP_FMT
" disabled, no CPUID level 0x%x\n",
421 x86_cap_flag(df
->feature
), df
->level
);
426 * Naming convention should be: <Name> [(<Codename>)]
427 * This table only is used unless init_<vendor>() below doesn't set it;
428 * in particular, if CPUID levels 0x80000002..4 are supported, this
432 /* Look up CPU names by table lookup. */
433 static const char *table_lookup_model(struct cpuinfo_x86
*c
)
436 const struct legacy_cpu_model_info
*info
;
438 if (c
->x86_model
>= 16)
439 return NULL
; /* Range check */
444 info
= this_cpu
->legacy_models
;
446 while (info
->family
) {
447 if (info
->family
== c
->x86
)
448 return info
->model_names
[c
->x86_model
];
452 return NULL
; /* Not found */
455 __u32 cpu_caps_cleared
[NCAPINTS
];
456 __u32 cpu_caps_set
[NCAPINTS
];
458 void load_percpu_segment(int cpu
)
461 loadsegment(fs
, __KERNEL_PERCPU
);
463 __loadsegment_simple(gs
, 0);
464 wrmsrl(MSR_GS_BASE
, (unsigned long)per_cpu(irq_stack_union
.gs_base
, cpu
));
466 load_stack_canary_segment();
470 /* The 32-bit entry code needs to find cpu_entry_area. */
471 DEFINE_PER_CPU(struct cpu_entry_area
*, cpu_entry_area
);
476 * Special IST stacks which the CPU switches to when it calls
477 * an IST-marked descriptor entry. Up to 7 stacks (hardware
478 * limit), all of them are 4K, except the debug stack which
481 static const unsigned int exception_stack_sizes
[N_EXCEPTION_STACKS
] = {
482 [0 ... N_EXCEPTION_STACKS
- 1] = EXCEPTION_STKSZ
,
483 [DEBUG_STACK
- 1] = DEBUG_STKSZ
486 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
487 [(N_EXCEPTION_STACKS
- 1) * EXCEPTION_STKSZ
+ DEBUG_STKSZ
]);
491 set_percpu_fixmap_pages(int idx
, void *ptr
, int pages
, pgprot_t prot
)
493 for ( ; pages
; pages
--, idx
--, ptr
+= PAGE_SIZE
)
494 __set_fixmap(idx
, per_cpu_ptr_to_phys(ptr
), prot
);
497 /* Setup the fixmap mappings only once per-processor */
498 static void __init
setup_cpu_entry_area(int cpu
)
501 extern char _entry_trampoline
[];
503 /* On 64-bit systems, we use a read-only fixmap GDT. */
504 pgprot_t gdt_prot
= PAGE_KERNEL_RO
;
507 * On native 32-bit systems, the GDT cannot be read-only because
508 * our double fault handler uses a task gate, and entering through
509 * a task gate needs to change an available TSS to busy. If the GDT
510 * is read-only, that will triple fault.
512 * On Xen PV, the GDT must be read-only because the hypervisor requires
515 pgprot_t gdt_prot
= boot_cpu_has(X86_FEATURE_XENPV
) ?
516 PAGE_KERNEL_RO
: PAGE_KERNEL
;
519 __set_fixmap(get_cpu_entry_area_index(cpu
, gdt
), get_cpu_gdt_paddr(cpu
), gdt_prot
);
522 * The Intel SDM says (Volume 3, 7.2.1):
524 * Avoid placing a page boundary in the part of the TSS that the
525 * processor reads during a task switch (the first 104 bytes). The
526 * processor may not correctly perform address translations if a
527 * boundary occurs in this area. During a task switch, the processor
528 * reads and writes into the first 104 bytes of each TSS (using
529 * contiguous physical addresses beginning with the physical address
530 * of the first byte of the TSS). So, after TSS access begins, if
531 * part of the 104 bytes is not physically contiguous, the processor
532 * will access incorrect information without generating a page-fault
535 * There are also a lot of errata involving the TSS spanning a page
536 * boundary. Assert that we're not doing that.
538 BUILD_BUG_ON((offsetof(struct tss_struct
, x86_tss
) ^
539 offsetofend(struct tss_struct
, x86_tss
)) & PAGE_MASK
);
540 BUILD_BUG_ON(sizeof(struct tss_struct
) % PAGE_SIZE
!= 0);
541 set_percpu_fixmap_pages(get_cpu_entry_area_index(cpu
, tss
),
542 &per_cpu(cpu_tss
, cpu
),
543 sizeof(struct tss_struct
) / PAGE_SIZE
,
547 per_cpu(cpu_entry_area
, cpu
) = get_cpu_entry_area(cpu
);
551 BUILD_BUG_ON(sizeof(exception_stacks
) % PAGE_SIZE
!= 0);
552 BUILD_BUG_ON(sizeof(exception_stacks
) !=
553 sizeof(((struct cpu_entry_area
*)0)->exception_stacks
));
554 set_percpu_fixmap_pages(get_cpu_entry_area_index(cpu
, exception_stacks
),
555 &per_cpu(exception_stacks
, cpu
),
556 sizeof(exception_stacks
) / PAGE_SIZE
,
559 __set_fixmap(get_cpu_entry_area_index(cpu
, entry_trampoline
),
560 __pa_symbol(_entry_trampoline
), PAGE_KERNEL_RX
);
564 void __init
setup_cpu_entry_areas(void)
568 for_each_possible_cpu(cpu
)
569 setup_cpu_entry_area(cpu
);
572 /* Load the original GDT from the per-cpu structure */
573 void load_direct_gdt(int cpu
)
575 struct desc_ptr gdt_descr
;
577 gdt_descr
.address
= (long)get_cpu_gdt_rw(cpu
);
578 gdt_descr
.size
= GDT_SIZE
- 1;
579 load_gdt(&gdt_descr
);
581 EXPORT_SYMBOL_GPL(load_direct_gdt
);
583 /* Load a fixmap remapping of the per-cpu GDT */
584 void load_fixmap_gdt(int cpu
)
586 struct desc_ptr gdt_descr
;
588 gdt_descr
.address
= (long)get_cpu_gdt_ro(cpu
);
589 gdt_descr
.size
= GDT_SIZE
- 1;
590 load_gdt(&gdt_descr
);
592 EXPORT_SYMBOL_GPL(load_fixmap_gdt
);
595 * Current gdt points %fs at the "master" per-cpu area: after this,
596 * it's on the real one.
598 void switch_to_new_gdt(int cpu
)
600 /* Load the original GDT */
601 load_direct_gdt(cpu
);
602 /* Reload the per-cpu base */
603 load_percpu_segment(cpu
);
606 static const struct cpu_dev
*cpu_devs
[X86_VENDOR_NUM
] = {};
608 static void get_model_name(struct cpuinfo_x86
*c
)
613 if (c
->extended_cpuid_level
< 0x80000004)
616 v
= (unsigned int *)c
->x86_model_id
;
617 cpuid(0x80000002, &v
[0], &v
[1], &v
[2], &v
[3]);
618 cpuid(0x80000003, &v
[4], &v
[5], &v
[6], &v
[7]);
619 cpuid(0x80000004, &v
[8], &v
[9], &v
[10], &v
[11]);
620 c
->x86_model_id
[48] = 0;
622 /* Trim whitespace */
623 p
= q
= s
= &c
->x86_model_id
[0];
629 /* Note the last non-whitespace index */
639 void cpu_detect_cache_sizes(struct cpuinfo_x86
*c
)
641 unsigned int n
, dummy
, ebx
, ecx
, edx
, l2size
;
643 n
= c
->extended_cpuid_level
;
645 if (n
>= 0x80000005) {
646 cpuid(0x80000005, &dummy
, &ebx
, &ecx
, &edx
);
647 c
->x86_cache_size
= (ecx
>>24) + (edx
>>24);
649 /* On K8 L1 TLB is inclusive, so don't count it */
654 if (n
< 0x80000006) /* Some chips just has a large L1. */
657 cpuid(0x80000006, &dummy
, &ebx
, &ecx
, &edx
);
661 c
->x86_tlbsize
+= ((ebx
>> 16) & 0xfff) + (ebx
& 0xfff);
663 /* do processor-specific cache resizing */
664 if (this_cpu
->legacy_cache_size
)
665 l2size
= this_cpu
->legacy_cache_size(c
, l2size
);
667 /* Allow user to override all this if necessary. */
668 if (cachesize_override
!= -1)
669 l2size
= cachesize_override
;
672 return; /* Again, no L2 cache is possible */
675 c
->x86_cache_size
= l2size
;
678 u16 __read_mostly tlb_lli_4k
[NR_INFO
];
679 u16 __read_mostly tlb_lli_2m
[NR_INFO
];
680 u16 __read_mostly tlb_lli_4m
[NR_INFO
];
681 u16 __read_mostly tlb_lld_4k
[NR_INFO
];
682 u16 __read_mostly tlb_lld_2m
[NR_INFO
];
683 u16 __read_mostly tlb_lld_4m
[NR_INFO
];
684 u16 __read_mostly tlb_lld_1g
[NR_INFO
];
686 static void cpu_detect_tlb(struct cpuinfo_x86
*c
)
688 if (this_cpu
->c_detect_tlb
)
689 this_cpu
->c_detect_tlb(c
);
691 pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
692 tlb_lli_4k
[ENTRIES
], tlb_lli_2m
[ENTRIES
],
693 tlb_lli_4m
[ENTRIES
]);
695 pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
696 tlb_lld_4k
[ENTRIES
], tlb_lld_2m
[ENTRIES
],
697 tlb_lld_4m
[ENTRIES
], tlb_lld_1g
[ENTRIES
]);
700 void detect_ht(struct cpuinfo_x86
*c
)
703 u32 eax
, ebx
, ecx
, edx
;
704 int index_msb
, core_bits
;
707 if (!cpu_has(c
, X86_FEATURE_HT
))
710 if (cpu_has(c
, X86_FEATURE_CMP_LEGACY
))
713 if (cpu_has(c
, X86_FEATURE_XTOPOLOGY
))
716 cpuid(1, &eax
, &ebx
, &ecx
, &edx
);
718 smp_num_siblings
= (ebx
& 0xff0000) >> 16;
720 if (smp_num_siblings
== 1) {
721 pr_info_once("CPU0: Hyper-Threading is disabled\n");
725 if (smp_num_siblings
<= 1)
728 index_msb
= get_count_order(smp_num_siblings
);
729 c
->phys_proc_id
= apic
->phys_pkg_id(c
->initial_apicid
, index_msb
);
731 smp_num_siblings
= smp_num_siblings
/ c
->x86_max_cores
;
733 index_msb
= get_count_order(smp_num_siblings
);
735 core_bits
= get_count_order(c
->x86_max_cores
);
737 c
->cpu_core_id
= apic
->phys_pkg_id(c
->initial_apicid
, index_msb
) &
738 ((1 << core_bits
) - 1);
741 if (!printed
&& (c
->x86_max_cores
* smp_num_siblings
) > 1) {
742 pr_info("CPU: Physical Processor ID: %d\n",
744 pr_info("CPU: Processor Core ID: %d\n",
751 static void get_cpu_vendor(struct cpuinfo_x86
*c
)
753 char *v
= c
->x86_vendor_id
;
756 for (i
= 0; i
< X86_VENDOR_NUM
; i
++) {
760 if (!strcmp(v
, cpu_devs
[i
]->c_ident
[0]) ||
761 (cpu_devs
[i
]->c_ident
[1] &&
762 !strcmp(v
, cpu_devs
[i
]->c_ident
[1]))) {
764 this_cpu
= cpu_devs
[i
];
765 c
->x86_vendor
= this_cpu
->c_x86_vendor
;
770 pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
771 "CPU: Your system may be unstable.\n", v
);
773 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
774 this_cpu
= &default_cpu
;
777 void cpu_detect(struct cpuinfo_x86
*c
)
779 /* Get vendor name */
780 cpuid(0x00000000, (unsigned int *)&c
->cpuid_level
,
781 (unsigned int *)&c
->x86_vendor_id
[0],
782 (unsigned int *)&c
->x86_vendor_id
[8],
783 (unsigned int *)&c
->x86_vendor_id
[4]);
786 /* Intel-defined flags: level 0x00000001 */
787 if (c
->cpuid_level
>= 0x00000001) {
788 u32 junk
, tfms
, cap0
, misc
;
790 cpuid(0x00000001, &tfms
, &misc
, &junk
, &cap0
);
791 c
->x86
= x86_family(tfms
);
792 c
->x86_model
= x86_model(tfms
);
793 c
->x86_mask
= x86_stepping(tfms
);
795 if (cap0
& (1<<19)) {
796 c
->x86_clflush_size
= ((misc
>> 8) & 0xff) * 8;
797 c
->x86_cache_alignment
= c
->x86_clflush_size
;
802 static void apply_forced_caps(struct cpuinfo_x86
*c
)
806 for (i
= 0; i
< NCAPINTS
; i
++) {
807 c
->x86_capability
[i
] &= ~cpu_caps_cleared
[i
];
808 c
->x86_capability
[i
] |= cpu_caps_set
[i
];
812 void get_cpu_cap(struct cpuinfo_x86
*c
)
814 u32 eax
, ebx
, ecx
, edx
;
816 /* Intel-defined flags: level 0x00000001 */
817 if (c
->cpuid_level
>= 0x00000001) {
818 cpuid(0x00000001, &eax
, &ebx
, &ecx
, &edx
);
820 c
->x86_capability
[CPUID_1_ECX
] = ecx
;
821 c
->x86_capability
[CPUID_1_EDX
] = edx
;
824 /* Thermal and Power Management Leaf: level 0x00000006 (eax) */
825 if (c
->cpuid_level
>= 0x00000006)
826 c
->x86_capability
[CPUID_6_EAX
] = cpuid_eax(0x00000006);
828 /* Additional Intel-defined flags: level 0x00000007 */
829 if (c
->cpuid_level
>= 0x00000007) {
830 cpuid_count(0x00000007, 0, &eax
, &ebx
, &ecx
, &edx
);
831 c
->x86_capability
[CPUID_7_0_EBX
] = ebx
;
832 c
->x86_capability
[CPUID_7_ECX
] = ecx
;
835 /* Extended state features: level 0x0000000d */
836 if (c
->cpuid_level
>= 0x0000000d) {
837 cpuid_count(0x0000000d, 1, &eax
, &ebx
, &ecx
, &edx
);
839 c
->x86_capability
[CPUID_D_1_EAX
] = eax
;
842 /* Additional Intel-defined flags: level 0x0000000F */
843 if (c
->cpuid_level
>= 0x0000000F) {
845 /* QoS sub-leaf, EAX=0Fh, ECX=0 */
846 cpuid_count(0x0000000F, 0, &eax
, &ebx
, &ecx
, &edx
);
847 c
->x86_capability
[CPUID_F_0_EDX
] = edx
;
849 if (cpu_has(c
, X86_FEATURE_CQM_LLC
)) {
850 /* will be overridden if occupancy monitoring exists */
851 c
->x86_cache_max_rmid
= ebx
;
853 /* QoS sub-leaf, EAX=0Fh, ECX=1 */
854 cpuid_count(0x0000000F, 1, &eax
, &ebx
, &ecx
, &edx
);
855 c
->x86_capability
[CPUID_F_1_EDX
] = edx
;
857 if ((cpu_has(c
, X86_FEATURE_CQM_OCCUP_LLC
)) ||
858 ((cpu_has(c
, X86_FEATURE_CQM_MBM_TOTAL
)) ||
859 (cpu_has(c
, X86_FEATURE_CQM_MBM_LOCAL
)))) {
860 c
->x86_cache_max_rmid
= ecx
;
861 c
->x86_cache_occ_scale
= ebx
;
864 c
->x86_cache_max_rmid
= -1;
865 c
->x86_cache_occ_scale
= -1;
869 /* AMD-defined flags: level 0x80000001 */
870 eax
= cpuid_eax(0x80000000);
871 c
->extended_cpuid_level
= eax
;
873 if ((eax
& 0xffff0000) == 0x80000000) {
874 if (eax
>= 0x80000001) {
875 cpuid(0x80000001, &eax
, &ebx
, &ecx
, &edx
);
877 c
->x86_capability
[CPUID_8000_0001_ECX
] = ecx
;
878 c
->x86_capability
[CPUID_8000_0001_EDX
] = edx
;
882 if (c
->extended_cpuid_level
>= 0x80000007) {
883 cpuid(0x80000007, &eax
, &ebx
, &ecx
, &edx
);
885 c
->x86_capability
[CPUID_8000_0007_EBX
] = ebx
;
889 if (c
->extended_cpuid_level
>= 0x80000008) {
890 cpuid(0x80000008, &eax
, &ebx
, &ecx
, &edx
);
892 c
->x86_virt_bits
= (eax
>> 8) & 0xff;
893 c
->x86_phys_bits
= eax
& 0xff;
894 c
->x86_capability
[CPUID_8000_0008_EBX
] = ebx
;
897 else if (cpu_has(c
, X86_FEATURE_PAE
) || cpu_has(c
, X86_FEATURE_PSE36
))
898 c
->x86_phys_bits
= 36;
901 if (c
->extended_cpuid_level
>= 0x8000000a)
902 c
->x86_capability
[CPUID_8000_000A_EDX
] = cpuid_edx(0x8000000a);
904 init_scattered_cpuid_features(c
);
907 * Clear/Set all flags overridden by options, after probe.
908 * This needs to happen each time we re-probe, which may happen
909 * several times during CPU initialization.
911 apply_forced_caps(c
);
914 static void identify_cpu_without_cpuid(struct cpuinfo_x86
*c
)
920 * First of all, decide if this is a 486 or higher
921 * It's a 486 if we can modify the AC flag
923 if (flag_is_changeable_p(X86_EFLAGS_AC
))
928 for (i
= 0; i
< X86_VENDOR_NUM
; i
++)
929 if (cpu_devs
[i
] && cpu_devs
[i
]->c_identify
) {
930 c
->x86_vendor_id
[0] = 0;
931 cpu_devs
[i
]->c_identify(c
);
932 if (c
->x86_vendor_id
[0]) {
941 * Do minimum CPU detection early.
942 * Fields really needed: vendor, cpuid_level, family, model, mask,
944 * The others are not touched to avoid unwanted side effects.
946 * WARNING: this function is only called on the BP. Don't add code here
947 * that is supposed to run on all CPUs.
949 static void __init
early_identify_cpu(struct cpuinfo_x86
*c
)
952 c
->x86_clflush_size
= 64;
953 c
->x86_phys_bits
= 36;
954 c
->x86_virt_bits
= 48;
956 c
->x86_clflush_size
= 32;
957 c
->x86_phys_bits
= 32;
958 c
->x86_virt_bits
= 32;
960 c
->x86_cache_alignment
= c
->x86_clflush_size
;
962 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
963 c
->extended_cpuid_level
= 0;
965 /* cyrix could have cpuid enabled via c_identify()*/
966 if (have_cpuid_p()) {
970 setup_force_cpu_cap(X86_FEATURE_CPUID
);
972 if (this_cpu
->c_early_init
)
973 this_cpu
->c_early_init(c
);
976 filter_cpuid_features(c
, false);
978 if (this_cpu
->c_bsp_init
)
979 this_cpu
->c_bsp_init(c
);
981 identify_cpu_without_cpuid(c
);
982 setup_clear_cpu_cap(X86_FEATURE_CPUID
);
985 setup_force_cpu_cap(X86_FEATURE_ALWAYS
);
989 void __init
early_cpu_init(void)
991 const struct cpu_dev
*const *cdev
;
994 #ifdef CONFIG_PROCESSOR_SELECT
995 pr_info("KERNEL supported cpus:\n");
998 for (cdev
= __x86_cpu_dev_start
; cdev
< __x86_cpu_dev_end
; cdev
++) {
999 const struct cpu_dev
*cpudev
= *cdev
;
1001 if (count
>= X86_VENDOR_NUM
)
1003 cpu_devs
[count
] = cpudev
;
1006 #ifdef CONFIG_PROCESSOR_SELECT
1010 for (j
= 0; j
< 2; j
++) {
1011 if (!cpudev
->c_ident
[j
])
1013 pr_info(" %s %s\n", cpudev
->c_vendor
,
1014 cpudev
->c_ident
[j
]);
1019 early_identify_cpu(&boot_cpu_data
);
1023 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1024 * unfortunately, that's not true in practice because of early VIA
1025 * chips and (more importantly) broken virtualizers that are not easy
1026 * to detect. In the latter case it doesn't even *fail* reliably, so
1027 * probing for it doesn't even work. Disable it completely on 32-bit
1028 * unless we can find a reliable way to detect all the broken cases.
1029 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1031 static void detect_nopl(struct cpuinfo_x86
*c
)
1033 #ifdef CONFIG_X86_32
1034 clear_cpu_cap(c
, X86_FEATURE_NOPL
);
1036 set_cpu_cap(c
, X86_FEATURE_NOPL
);
1040 static void detect_null_seg_behavior(struct cpuinfo_x86
*c
)
1042 #ifdef CONFIG_X86_64
1044 * Empirically, writing zero to a segment selector on AMD does
1045 * not clear the base, whereas writing zero to a segment
1046 * selector on Intel does clear the base. Intel's behavior
1047 * allows slightly faster context switches in the common case
1048 * where GS is unused by the prev and next threads.
1050 * Since neither vendor documents this anywhere that I can see,
1051 * detect it directly instead of hardcoding the choice by
1054 * I've designated AMD's behavior as the "bug" because it's
1055 * counterintuitive and less friendly.
1058 unsigned long old_base
, tmp
;
1059 rdmsrl(MSR_FS_BASE
, old_base
);
1060 wrmsrl(MSR_FS_BASE
, 1);
1062 rdmsrl(MSR_FS_BASE
, tmp
);
1064 set_cpu_bug(c
, X86_BUG_NULL_SEG
);
1065 wrmsrl(MSR_FS_BASE
, old_base
);
1069 static void generic_identify(struct cpuinfo_x86
*c
)
1071 c
->extended_cpuid_level
= 0;
1073 if (!have_cpuid_p())
1074 identify_cpu_without_cpuid(c
);
1076 /* cyrix could have cpuid enabled via c_identify()*/
1077 if (!have_cpuid_p())
1086 if (c
->cpuid_level
>= 0x00000001) {
1087 c
->initial_apicid
= (cpuid_ebx(1) >> 24) & 0xFF;
1088 #ifdef CONFIG_X86_32
1090 c
->apicid
= apic
->phys_pkg_id(c
->initial_apicid
, 0);
1092 c
->apicid
= c
->initial_apicid
;
1095 c
->phys_proc_id
= c
->initial_apicid
;
1098 get_model_name(c
); /* Default name */
1102 detect_null_seg_behavior(c
);
1105 * ESPFIX is a strange bug. All real CPUs have it. Paravirt
1106 * systems that run Linux at CPL > 0 may or may not have the
1107 * issue, but, even if they have the issue, there's absolutely
1108 * nothing we can do about it because we can't use the real IRET
1111 * NB: For the time being, only 32-bit kernels support
1112 * X86_BUG_ESPFIX as such. 64-bit kernels directly choose
1113 * whether to apply espfix using paravirt hooks. If any
1114 * non-paravirt system ever shows up that does *not* have the
1115 * ESPFIX issue, we can change this.
1117 #ifdef CONFIG_X86_32
1118 # ifdef CONFIG_PARAVIRT
1120 extern void native_iret(void);
1121 if (pv_cpu_ops
.iret
== native_iret
)
1122 set_cpu_bug(c
, X86_BUG_ESPFIX
);
1125 set_cpu_bug(c
, X86_BUG_ESPFIX
);
1130 static void x86_init_cache_qos(struct cpuinfo_x86
*c
)
1133 * The heavy lifting of max_rmid and cache_occ_scale are handled
1134 * in get_cpu_cap(). Here we just set the max_rmid for the boot_cpu
1135 * in case CQM bits really aren't there in this CPU.
1137 if (c
!= &boot_cpu_data
) {
1138 boot_cpu_data
.x86_cache_max_rmid
=
1139 min(boot_cpu_data
.x86_cache_max_rmid
,
1140 c
->x86_cache_max_rmid
);
1145 * Validate that ACPI/mptables have the same information about the
1146 * effective APIC id and update the package map.
1148 static void validate_apic_and_package_id(struct cpuinfo_x86
*c
)
1151 unsigned int apicid
, cpu
= smp_processor_id();
1153 apicid
= apic
->cpu_present_to_apicid(cpu
);
1155 if (apicid
!= c
->apicid
) {
1156 pr_err(FW_BUG
"CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1157 cpu
, apicid
, c
->initial_apicid
);
1159 BUG_ON(topology_update_package_map(c
->phys_proc_id
, cpu
));
1161 c
->logical_proc_id
= 0;
1166 * This does the hard work of actually picking apart the CPU stuff...
1168 static void identify_cpu(struct cpuinfo_x86
*c
)
1172 c
->loops_per_jiffy
= loops_per_jiffy
;
1173 c
->x86_cache_size
= -1;
1174 c
->x86_vendor
= X86_VENDOR_UNKNOWN
;
1175 c
->x86_model
= c
->x86_mask
= 0; /* So far unknown... */
1176 c
->x86_vendor_id
[0] = '\0'; /* Unset */
1177 c
->x86_model_id
[0] = '\0'; /* Unset */
1178 c
->x86_max_cores
= 1;
1179 c
->x86_coreid_bits
= 0;
1181 #ifdef CONFIG_X86_64
1182 c
->x86_clflush_size
= 64;
1183 c
->x86_phys_bits
= 36;
1184 c
->x86_virt_bits
= 48;
1186 c
->cpuid_level
= -1; /* CPUID not detected */
1187 c
->x86_clflush_size
= 32;
1188 c
->x86_phys_bits
= 32;
1189 c
->x86_virt_bits
= 32;
1191 c
->x86_cache_alignment
= c
->x86_clflush_size
;
1192 memset(&c
->x86_capability
, 0, sizeof c
->x86_capability
);
1194 generic_identify(c
);
1196 if (this_cpu
->c_identify
)
1197 this_cpu
->c_identify(c
);
1199 /* Clear/Set all flags overridden by options, after probe */
1200 apply_forced_caps(c
);
1202 #ifdef CONFIG_X86_64
1203 c
->apicid
= apic
->phys_pkg_id(c
->initial_apicid
, 0);
1207 * Vendor-specific initialization. In this section we
1208 * canonicalize the feature flags, meaning if there are
1209 * features a certain CPU supports which CPUID doesn't
1210 * tell us, CPUID claiming incorrect flags, or other bugs,
1211 * we handle them here.
1213 * At the end of this section, c->x86_capability better
1214 * indicate the features this CPU genuinely supports!
1216 if (this_cpu
->c_init
)
1217 this_cpu
->c_init(c
);
1219 /* Disable the PN if appropriate */
1220 squash_the_stupid_serial_number(c
);
1222 /* Set up SMEP/SMAP */
1227 * The vendor-specific functions might have changed features.
1228 * Now we do "generic changes."
1231 /* Filter out anything that depends on CPUID levels we don't have */
1232 filter_cpuid_features(c
, true);
1234 /* If the model name is still unset, do table lookup. */
1235 if (!c
->x86_model_id
[0]) {
1237 p
= table_lookup_model(c
);
1239 strcpy(c
->x86_model_id
, p
);
1241 /* Last resort... */
1242 sprintf(c
->x86_model_id
, "%02x/%02x",
1243 c
->x86
, c
->x86_model
);
1246 #ifdef CONFIG_X86_64
1251 x86_init_cache_qos(c
);
1255 * Clear/Set all flags overridden by options, need do it
1256 * before following smp all cpus cap AND.
1258 apply_forced_caps(c
);
1261 * On SMP, boot_cpu_data holds the common feature set between
1262 * all CPUs; so make sure that we indicate which features are
1263 * common between the CPUs. The first time this routine gets
1264 * executed, c == &boot_cpu_data.
1266 if (c
!= &boot_cpu_data
) {
1267 /* AND the already accumulated flags with these */
1268 for (i
= 0; i
< NCAPINTS
; i
++)
1269 boot_cpu_data
.x86_capability
[i
] &= c
->x86_capability
[i
];
1271 /* OR, i.e. replicate the bug flags */
1272 for (i
= NCAPINTS
; i
< NCAPINTS
+ NBUGINTS
; i
++)
1273 c
->x86_capability
[i
] |= boot_cpu_data
.x86_capability
[i
];
1276 /* Init Machine Check Exception if available. */
1279 select_idle_routine(c
);
1282 numa_add_cpu(smp_processor_id());
1287 * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1288 * on 32-bit kernels:
1290 #ifdef CONFIG_X86_32
1291 void enable_sep_cpu(void)
1293 struct tss_struct
*tss
;
1296 if (!boot_cpu_has(X86_FEATURE_SEP
))
1300 tss
= &per_cpu(cpu_tss
, cpu
);
1303 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1304 * see the big comment in struct x86_hw_tss's definition.
1307 tss
->x86_tss
.ss1
= __KERNEL_CS
;
1308 wrmsr(MSR_IA32_SYSENTER_CS
, tss
->x86_tss
.ss1
, 0);
1309 wrmsr(MSR_IA32_SYSENTER_ESP
, (unsigned long)(cpu_SYSENTER_stack(cpu
) + 1), 0);
1310 wrmsr(MSR_IA32_SYSENTER_EIP
, (unsigned long)entry_SYSENTER_32
, 0);
1316 void __init
identify_boot_cpu(void)
1318 identify_cpu(&boot_cpu_data
);
1319 #ifdef CONFIG_X86_32
1323 cpu_detect_tlb(&boot_cpu_data
);
1326 void identify_secondary_cpu(struct cpuinfo_x86
*c
)
1328 BUG_ON(c
== &boot_cpu_data
);
1330 #ifdef CONFIG_X86_32
1334 validate_apic_and_package_id(c
);
1337 static __init
int setup_noclflush(char *arg
)
1339 setup_clear_cpu_cap(X86_FEATURE_CLFLUSH
);
1340 setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT
);
1343 __setup("noclflush", setup_noclflush
);
1345 void print_cpu_info(struct cpuinfo_x86
*c
)
1347 const char *vendor
= NULL
;
1349 if (c
->x86_vendor
< X86_VENDOR_NUM
) {
1350 vendor
= this_cpu
->c_vendor
;
1352 if (c
->cpuid_level
>= 0)
1353 vendor
= c
->x86_vendor_id
;
1356 if (vendor
&& !strstr(c
->x86_model_id
, vendor
))
1357 pr_cont("%s ", vendor
);
1359 if (c
->x86_model_id
[0])
1360 pr_cont("%s", c
->x86_model_id
);
1362 pr_cont("%d86", c
->x86
);
1364 pr_cont(" (family: 0x%x, model: 0x%x", c
->x86
, c
->x86_model
);
1366 if (c
->x86_mask
|| c
->cpuid_level
>= 0)
1367 pr_cont(", stepping: 0x%x)\n", c
->x86_mask
);
1373 * clearcpuid= was already parsed in fpu__init_parse_early_param.
1374 * But we need to keep a dummy __setup around otherwise it would
1375 * show up as an environment variable for init.
1377 static __init
int setup_clearcpuid(char *arg
)
1381 __setup("clearcpuid=", setup_clearcpuid
);
1383 #ifdef CONFIG_X86_64
1384 struct desc_ptr idt_descr __ro_after_init
= {
1385 .size
= NR_VECTORS
* 16 - 1,
1386 .address
= (unsigned long) idt_table
,
1388 const struct desc_ptr debug_idt_descr
= {
1389 .size
= NR_VECTORS
* 16 - 1,
1390 .address
= (unsigned long) debug_idt_table
,
1393 DEFINE_PER_CPU_FIRST(union irq_stack_union
,
1394 irq_stack_union
) __aligned(PAGE_SIZE
) __visible
;
1397 * The following percpu variables are hot. Align current_task to
1398 * cacheline size such that they fall in the same cacheline.
1400 DEFINE_PER_CPU(struct task_struct
*, current_task
) ____cacheline_aligned
=
1402 EXPORT_PER_CPU_SYMBOL(current_task
);
1404 DEFINE_PER_CPU(char *, irq_stack_ptr
) =
1405 init_per_cpu_var(irq_stack_union
.irq_stack
) + IRQ_STACK_SIZE
;
1407 DEFINE_PER_CPU(unsigned int, irq_count
) __visible
= -1;
1409 DEFINE_PER_CPU(int, __preempt_count
) = INIT_PREEMPT_COUNT
;
1410 EXPORT_PER_CPU_SYMBOL(__preempt_count
);
1412 /* May not be marked __init: used by software suspend */
1413 void syscall_init(void)
1415 extern char _entry_trampoline
[];
1416 extern char entry_SYSCALL_64_trampoline
[];
1418 int cpu
= smp_processor_id();
1419 unsigned long SYSCALL64_entry_trampoline
=
1420 (unsigned long)get_cpu_entry_area(cpu
)->entry_trampoline
+
1421 (entry_SYSCALL_64_trampoline
- _entry_trampoline
);
1423 wrmsr(MSR_STAR
, 0, (__USER32_CS
<< 16) | __KERNEL_CS
);
1424 wrmsrl(MSR_LSTAR
, SYSCALL64_entry_trampoline
);
1426 #ifdef CONFIG_IA32_EMULATION
1427 wrmsrl(MSR_CSTAR
, (unsigned long)entry_SYSCALL_compat
);
1429 * This only works on Intel CPUs.
1430 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1431 * This does not cause SYSENTER to jump to the wrong location, because
1432 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1434 wrmsrl_safe(MSR_IA32_SYSENTER_CS
, (u64
)__KERNEL_CS
);
1435 wrmsrl_safe(MSR_IA32_SYSENTER_ESP
, (unsigned long)(cpu_SYSENTER_stack(cpu
) + 1));
1436 wrmsrl_safe(MSR_IA32_SYSENTER_EIP
, (u64
)entry_SYSENTER_compat
);
1438 wrmsrl(MSR_CSTAR
, (unsigned long)ignore_sysret
);
1439 wrmsrl_safe(MSR_IA32_SYSENTER_CS
, (u64
)GDT_ENTRY_INVALID_SEG
);
1440 wrmsrl_safe(MSR_IA32_SYSENTER_ESP
, 0ULL);
1441 wrmsrl_safe(MSR_IA32_SYSENTER_EIP
, 0ULL);
1444 /* Flags to clear on syscall */
1445 wrmsrl(MSR_SYSCALL_MASK
,
1446 X86_EFLAGS_TF
|X86_EFLAGS_DF
|X86_EFLAGS_IF
|
1447 X86_EFLAGS_IOPL
|X86_EFLAGS_AC
|X86_EFLAGS_NT
);
1451 * Copies of the original ist values from the tss are only accessed during
1452 * debugging, no special alignment required.
1454 DEFINE_PER_CPU(struct orig_ist
, orig_ist
);
1456 static DEFINE_PER_CPU(unsigned long, debug_stack_addr
);
1457 DEFINE_PER_CPU(int, debug_stack_usage
);
1459 int is_debug_stack(unsigned long addr
)
1461 return __this_cpu_read(debug_stack_usage
) ||
1462 (addr
<= __this_cpu_read(debug_stack_addr
) &&
1463 addr
> (__this_cpu_read(debug_stack_addr
) - DEBUG_STKSZ
));
1465 NOKPROBE_SYMBOL(is_debug_stack
);
1467 DEFINE_PER_CPU(u32
, debug_idt_ctr
);
1469 void debug_stack_set_zero(void)
1471 this_cpu_inc(debug_idt_ctr
);
1474 NOKPROBE_SYMBOL(debug_stack_set_zero
);
1476 void debug_stack_reset(void)
1478 if (WARN_ON(!this_cpu_read(debug_idt_ctr
)))
1480 if (this_cpu_dec_return(debug_idt_ctr
) == 0)
1483 NOKPROBE_SYMBOL(debug_stack_reset
);
1485 #else /* CONFIG_X86_64 */
1487 DEFINE_PER_CPU(struct task_struct
*, current_task
) = &init_task
;
1488 EXPORT_PER_CPU_SYMBOL(current_task
);
1489 DEFINE_PER_CPU(int, __preempt_count
) = INIT_PREEMPT_COUNT
;
1490 EXPORT_PER_CPU_SYMBOL(__preempt_count
);
1493 * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1494 * the top of the kernel stack. Use an extra percpu variable to track the
1495 * top of the kernel stack directly.
1497 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack
) =
1498 (unsigned long)&init_thread_union
+ THREAD_SIZE
;
1499 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack
);
1501 #ifdef CONFIG_CC_STACKPROTECTOR
1502 DEFINE_PER_CPU_ALIGNED(struct stack_canary
, stack_canary
);
1505 #endif /* CONFIG_X86_64 */
1508 * Clear all 6 debug registers:
1510 static void clear_all_debug_regs(void)
1514 for (i
= 0; i
< 8; i
++) {
1515 /* Ignore db4, db5 */
1516 if ((i
== 4) || (i
== 5))
1525 * Restore debug regs if using kgdbwait and you have a kernel debugger
1526 * connection established.
1528 static void dbg_restore_debug_regs(void)
1530 if (unlikely(kgdb_connected
&& arch_kgdb_ops
.correct_hw_break
))
1531 arch_kgdb_ops
.correct_hw_break();
1533 #else /* ! CONFIG_KGDB */
1534 #define dbg_restore_debug_regs()
1535 #endif /* ! CONFIG_KGDB */
1537 static void wait_for_master_cpu(int cpu
)
1541 * wait for ACK from master CPU before continuing
1542 * with AP initialization
1544 WARN_ON(cpumask_test_and_set_cpu(cpu
, cpu_initialized_mask
));
1545 while (!cpumask_test_cpu(cpu
, cpu_callout_mask
))
1551 * cpu_init() initializes state that is per-CPU. Some data is already
1552 * initialized (naturally) in the bootstrap process, such as the GDT
1553 * and IDT. We reload them nevertheless, this function acts as a
1554 * 'CPU state barrier', nothing should get across.
1555 * A lot of state is already set up in PDA init for 64 bit
1557 #ifdef CONFIG_X86_64
1561 struct orig_ist
*oist
;
1562 struct task_struct
*me
;
1563 struct tss_struct
*t
;
1565 int cpu
= raw_smp_processor_id();
1568 wait_for_master_cpu(cpu
);
1571 * Initialize the CR4 shadow before doing anything that could
1579 t
= &per_cpu(cpu_tss
, cpu
);
1580 oist
= &per_cpu(orig_ist
, cpu
);
1583 if (this_cpu_read(numa_node
) == 0 &&
1584 early_cpu_to_node(cpu
) != NUMA_NO_NODE
)
1585 set_numa_node(early_cpu_to_node(cpu
));
1590 pr_debug("Initializing CPU#%d\n", cpu
);
1592 cr4_clear_bits(X86_CR4_VME
|X86_CR4_PVI
|X86_CR4_TSD
|X86_CR4_DE
);
1595 * Initialize the per-CPU GDT with the boot GDT,
1596 * and set up the GDT descriptor:
1599 switch_to_new_gdt(cpu
);
1604 memset(me
->thread
.tls_array
, 0, GDT_ENTRY_TLS_ENTRIES
* 8);
1607 wrmsrl(MSR_FS_BASE
, 0);
1608 wrmsrl(MSR_KERNEL_GS_BASE
, 0);
1615 * set up and load the per-CPU TSS
1617 if (!oist
->ist
[0]) {
1618 char *estacks
= get_cpu_entry_area(cpu
)->exception_stacks
;
1620 for (v
= 0; v
< N_EXCEPTION_STACKS
; v
++) {
1621 estacks
+= exception_stack_sizes
[v
];
1622 oist
->ist
[v
] = t
->x86_tss
.ist
[v
] =
1623 (unsigned long)estacks
;
1624 if (v
== DEBUG_STACK
-1)
1625 per_cpu(debug_stack_addr
, cpu
) = (unsigned long)estacks
;
1629 t
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
1632 * <= is required because the CPU will access up to
1633 * 8 bits beyond the end of the IO permission bitmap.
1635 for (i
= 0; i
<= IO_BITMAP_LONGS
; i
++)
1636 t
->io_bitmap
[i
] = ~0UL;
1639 me
->active_mm
= &init_mm
;
1641 enter_lazy_tlb(&init_mm
, me
);
1644 * Initialize the TSS. sp0 points to the entry trampoline stack
1645 * regardless of what task is running.
1647 set_tss_desc(cpu
, &get_cpu_entry_area(cpu
)->tss
.x86_tss
);
1649 load_sp0((unsigned long)(cpu_SYSENTER_stack(cpu
) + 1));
1651 load_mm_ldt(&init_mm
);
1653 clear_all_debug_regs();
1654 dbg_restore_debug_regs();
1661 load_fixmap_gdt(cpu
);
1668 int cpu
= smp_processor_id();
1669 struct task_struct
*curr
= current
;
1670 struct tss_struct
*t
= &per_cpu(cpu_tss
, cpu
);
1672 wait_for_master_cpu(cpu
);
1675 * Initialize the CR4 shadow before doing anything that could
1680 show_ucode_info_early();
1682 pr_info("Initializing CPU#%d\n", cpu
);
1684 if (cpu_feature_enabled(X86_FEATURE_VME
) ||
1685 boot_cpu_has(X86_FEATURE_TSC
) ||
1686 boot_cpu_has(X86_FEATURE_DE
))
1687 cr4_clear_bits(X86_CR4_VME
|X86_CR4_PVI
|X86_CR4_TSD
|X86_CR4_DE
);
1690 switch_to_new_gdt(cpu
);
1693 * Set up and load the per-CPU TSS and LDT
1696 curr
->active_mm
= &init_mm
;
1698 enter_lazy_tlb(&init_mm
, curr
);
1701 * Initialize the TSS. Don't bother initializing sp0, as the initial
1702 * task never enters user mode.
1704 set_tss_desc(cpu
, &get_cpu_entry_area(cpu
)->tss
.x86_tss
);
1707 load_mm_ldt(&init_mm
);
1709 t
->x86_tss
.io_bitmap_base
= IO_BITMAP_OFFSET
;
1711 #ifdef CONFIG_DOUBLEFAULT
1712 /* Set up doublefault TSS pointer in the GDT */
1713 __set_tss_desc(cpu
, GDT_ENTRY_DOUBLEFAULT_TSS
, &doublefault_tss
);
1716 clear_all_debug_regs();
1717 dbg_restore_debug_regs();
1721 load_fixmap_gdt(cpu
);
1725 static void bsp_resume(void)
1727 if (this_cpu
->c_bsp_resume
)
1728 this_cpu
->c_bsp_resume(&boot_cpu_data
);
1731 static struct syscore_ops cpu_syscore_ops
= {
1732 .resume
= bsp_resume
,
1735 static int __init
init_cpu_syscore(void)
1737 register_syscore_ops(&cpu_syscore_ops
);
1740 core_initcall(init_cpu_syscore
);