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/debugreg.h>
23 #include <asm/sections.h>
24 #include <asm/vsyscall.h>
25 #include <linux/topology.h>
26 #include <linux/cpumask.h>
27 #include <asm/pgtable.h>
28 #include <linux/atomic.h>
29 #include <asm/proto.h>
30 #include <asm/setup.h>
34 #include <asm/fpu-internal.h>
36 #include <linux/numa.h>
42 #include <asm/microcode.h>
43 #include <asm/microcode_intel.h>
45 #ifdef CONFIG_X86_LOCAL_APIC
46 #include <asm/uv/uv.h>
51 /* all of these masks are initialized in setup_cpu_local_masks() */
52 cpumask_var_t cpu_initialized_mask;
53 cpumask_var_t cpu_callout_mask;
54 cpumask_var_t cpu_callin_mask;
56 /* representing cpus for which sibling maps can be computed */
57 cpumask_var_t cpu_sibling_setup_mask;
59 /* correctly size the local cpu masks */
60 void __init setup_cpu_local_masks(void)
62 alloc_bootmem_cpumask_var(&cpu_initialized_mask);
63 alloc_bootmem_cpumask_var(&cpu_callin_mask);
64 alloc_bootmem_cpumask_var(&cpu_callout_mask);
65 alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
68 static void default_init(struct cpuinfo_x86 *c)
71 cpu_detect_cache_sizes(c);
73 /* Not much we can do here... */
74 /* Check if at least it has cpuid */
75 if (c->cpuid_level == -1) {
76 /* No cpuid. It must be an ancient CPU */
78 strcpy(c->x86_model_id, "486");
80 strcpy(c->x86_model_id, "386");
85 static const struct cpu_dev default_cpu = {
86 .c_init = default_init,
87 .c_vendor = "Unknown",
88 .c_x86_vendor = X86_VENDOR_UNKNOWN,
91 static const struct cpu_dev *this_cpu = &default_cpu;
93 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
96 * We need valid kernel segments for data and code in long mode too
97 * IRET will check the segment types kkeil 2000/10/28
98 * Also sysret mandates a special GDT layout
100 * TLS descriptors are currently at a different place compared to i386.
101 * Hopefully nobody expects them at a fixed place (Wine?)
103 [GDT_ENTRY_KERNEL32_CS] = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
104 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
105 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
106 [GDT_ENTRY_DEFAULT_USER32_CS] = GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
107 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
108 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
110 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
111 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
112 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
113 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
115 * Segments used for calling PnP BIOS have byte granularity.
116 * They code segments and data segments have fixed 64k limits,
117 * the transfer segment sizes are set at run time.
120 [GDT_ENTRY_PNPBIOS_CS32] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
122 [GDT_ENTRY_PNPBIOS_CS16] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
124 [GDT_ENTRY_PNPBIOS_DS] = GDT_ENTRY_INIT(0x0092, 0, 0xffff),
126 [GDT_ENTRY_PNPBIOS_TS1] = GDT_ENTRY_INIT(0x0092, 0, 0),
128 [GDT_ENTRY_PNPBIOS_TS2] = GDT_ENTRY_INIT(0x0092, 0, 0),
130 * The APM segments have byte granularity and their bases
131 * are set at run time. All have 64k limits.
134 [GDT_ENTRY_APMBIOS_BASE] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
136 [GDT_ENTRY_APMBIOS_BASE+1] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
138 [GDT_ENTRY_APMBIOS_BASE+2] = GDT_ENTRY_INIT(0x4092, 0, 0xffff),
140 [GDT_ENTRY_ESPFIX_SS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
141 [GDT_ENTRY_PERCPU] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
142 GDT_STACK_CANARY_INIT
145 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
147 static int __init x86_xsave_setup(char *s)
149 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
150 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
151 setup_clear_cpu_cap(X86_FEATURE_XSAVES);
152 setup_clear_cpu_cap(X86_FEATURE_AVX);
153 setup_clear_cpu_cap(X86_FEATURE_AVX2);
156 __setup("noxsave", x86_xsave_setup);
158 static int __init x86_xsaveopt_setup(char *s)
160 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
163 __setup("noxsaveopt", x86_xsaveopt_setup);
165 static int __init x86_xsaves_setup(char *s)
167 setup_clear_cpu_cap(X86_FEATURE_XSAVES);
170 __setup("noxsaves", x86_xsaves_setup);
173 static int cachesize_override = -1;
174 static int disable_x86_serial_nr = 1;
176 static int __init cachesize_setup(char *str)
178 get_option(&str, &cachesize_override);
181 __setup("cachesize=", cachesize_setup);
183 static int __init x86_fxsr_setup(char *s)
185 setup_clear_cpu_cap(X86_FEATURE_FXSR);
186 setup_clear_cpu_cap(X86_FEATURE_XMM);
189 __setup("nofxsr", x86_fxsr_setup);
191 static int __init x86_sep_setup(char *s)
193 setup_clear_cpu_cap(X86_FEATURE_SEP);
196 __setup("nosep", x86_sep_setup);
198 /* Standard macro to see if a specific flag is changeable */
199 static inline int flag_is_changeable_p(u32 flag)
204 * Cyrix and IDT cpus allow disabling of CPUID
205 * so the code below may return different results
206 * when it is executed before and after enabling
207 * the CPUID. Add "volatile" to not allow gcc to
208 * optimize the subsequent calls to this function.
210 asm volatile ("pushfl \n\t"
221 : "=&r" (f1), "=&r" (f2)
224 return ((f1^f2) & flag) != 0;
227 /* Probe for the CPUID instruction */
228 int have_cpuid_p(void)
230 return flag_is_changeable_p(X86_EFLAGS_ID);
233 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
235 unsigned long lo, hi;
237 if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
240 /* Disable processor serial number: */
242 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
244 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
246 printk(KERN_NOTICE "CPU serial number disabled.\n");
247 clear_cpu_cap(c, X86_FEATURE_PN);
249 /* Disabling the serial number may affect the cpuid level */
250 c->cpuid_level = cpuid_eax(0);
253 static int __init x86_serial_nr_setup(char *s)
255 disable_x86_serial_nr = 0;
258 __setup("serialnumber", x86_serial_nr_setup);
260 static inline int flag_is_changeable_p(u32 flag)
264 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
269 static __init int setup_disable_smep(char *arg)
271 setup_clear_cpu_cap(X86_FEATURE_SMEP);
274 __setup("nosmep", setup_disable_smep);
276 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
278 if (cpu_has(c, X86_FEATURE_SMEP))
279 set_in_cr4(X86_CR4_SMEP);
282 static __init int setup_disable_smap(char *arg)
284 setup_clear_cpu_cap(X86_FEATURE_SMAP);
287 __setup("nosmap", setup_disable_smap);
289 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
291 unsigned long eflags;
293 /* This should have been cleared long ago */
294 raw_local_save_flags(eflags);
295 BUG_ON(eflags & X86_EFLAGS_AC);
297 if (cpu_has(c, X86_FEATURE_SMAP)) {
298 #ifdef CONFIG_X86_SMAP
299 set_in_cr4(X86_CR4_SMAP);
301 clear_in_cr4(X86_CR4_SMAP);
307 * Some CPU features depend on higher CPUID levels, which may not always
308 * be available due to CPUID level capping or broken virtualization
309 * software. Add those features to this table to auto-disable them.
311 struct cpuid_dependent_feature {
316 static const struct cpuid_dependent_feature
317 cpuid_dependent_features[] = {
318 { X86_FEATURE_MWAIT, 0x00000005 },
319 { X86_FEATURE_DCA, 0x00000009 },
320 { X86_FEATURE_XSAVE, 0x0000000d },
324 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
326 const struct cpuid_dependent_feature *df;
328 for (df = cpuid_dependent_features; df->feature; df++) {
330 if (!cpu_has(c, df->feature))
333 * Note: cpuid_level is set to -1 if unavailable, but
334 * extended_extended_level is set to 0 if unavailable
335 * and the legitimate extended levels are all negative
336 * when signed; hence the weird messing around with
339 if (!((s32)df->level < 0 ?
340 (u32)df->level > (u32)c->extended_cpuid_level :
341 (s32)df->level > (s32)c->cpuid_level))
344 clear_cpu_cap(c, df->feature);
349 "CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
350 x86_cap_flag(df->feature), df->level);
355 * Naming convention should be: <Name> [(<Codename>)]
356 * This table only is used unless init_<vendor>() below doesn't set it;
357 * in particular, if CPUID levels 0x80000002..4 are supported, this
361 /* Look up CPU names by table lookup. */
362 static const char *table_lookup_model(struct cpuinfo_x86 *c)
365 const struct legacy_cpu_model_info *info;
367 if (c->x86_model >= 16)
368 return NULL; /* Range check */
373 info = this_cpu->legacy_models;
375 while (info->family) {
376 if (info->family == c->x86)
377 return info->model_names[c->x86_model];
381 return NULL; /* Not found */
384 __u32 cpu_caps_cleared[NCAPINTS];
385 __u32 cpu_caps_set[NCAPINTS];
387 void load_percpu_segment(int cpu)
390 loadsegment(fs, __KERNEL_PERCPU);
393 wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
395 load_stack_canary_segment();
399 * Current gdt points %fs at the "master" per-cpu area: after this,
400 * it's on the real one.
402 void switch_to_new_gdt(int cpu)
404 struct desc_ptr gdt_descr;
406 gdt_descr.address = (long)get_cpu_gdt_table(cpu);
407 gdt_descr.size = GDT_SIZE - 1;
408 load_gdt(&gdt_descr);
409 /* Reload the per-cpu base */
411 load_percpu_segment(cpu);
414 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
416 static void get_model_name(struct cpuinfo_x86 *c)
421 if (c->extended_cpuid_level < 0x80000004)
424 v = (unsigned int *)c->x86_model_id;
425 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
426 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
427 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
428 c->x86_model_id[48] = 0;
431 * Intel chips right-justify this string for some dumb reason;
432 * undo that brain damage:
434 p = q = &c->x86_model_id[0];
440 while (q <= &c->x86_model_id[48])
441 *q++ = '\0'; /* Zero-pad the rest */
445 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
447 unsigned int n, dummy, ebx, ecx, edx, l2size;
449 n = c->extended_cpuid_level;
451 if (n >= 0x80000005) {
452 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
453 c->x86_cache_size = (ecx>>24) + (edx>>24);
455 /* On K8 L1 TLB is inclusive, so don't count it */
460 if (n < 0x80000006) /* Some chips just has a large L1. */
463 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
467 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
469 /* do processor-specific cache resizing */
470 if (this_cpu->legacy_cache_size)
471 l2size = this_cpu->legacy_cache_size(c, l2size);
473 /* Allow user to override all this if necessary. */
474 if (cachesize_override != -1)
475 l2size = cachesize_override;
478 return; /* Again, no L2 cache is possible */
481 c->x86_cache_size = l2size;
484 u16 __read_mostly tlb_lli_4k[NR_INFO];
485 u16 __read_mostly tlb_lli_2m[NR_INFO];
486 u16 __read_mostly tlb_lli_4m[NR_INFO];
487 u16 __read_mostly tlb_lld_4k[NR_INFO];
488 u16 __read_mostly tlb_lld_2m[NR_INFO];
489 u16 __read_mostly tlb_lld_4m[NR_INFO];
490 u16 __read_mostly tlb_lld_1g[NR_INFO];
492 void cpu_detect_tlb(struct cpuinfo_x86 *c)
494 if (this_cpu->c_detect_tlb)
495 this_cpu->c_detect_tlb(c);
497 printk(KERN_INFO "Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n"
498 "Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
499 tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
500 tlb_lli_4m[ENTRIES], tlb_lld_4k[ENTRIES],
501 tlb_lld_2m[ENTRIES], tlb_lld_4m[ENTRIES],
502 tlb_lld_1g[ENTRIES]);
505 void detect_ht(struct cpuinfo_x86 *c)
508 u32 eax, ebx, ecx, edx;
509 int index_msb, core_bits;
512 if (!cpu_has(c, X86_FEATURE_HT))
515 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
518 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
521 cpuid(1, &eax, &ebx, &ecx, &edx);
523 smp_num_siblings = (ebx & 0xff0000) >> 16;
525 if (smp_num_siblings == 1) {
526 printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
530 if (smp_num_siblings <= 1)
533 index_msb = get_count_order(smp_num_siblings);
534 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
536 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
538 index_msb = get_count_order(smp_num_siblings);
540 core_bits = get_count_order(c->x86_max_cores);
542 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
543 ((1 << core_bits) - 1);
546 if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
547 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
549 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
556 static void get_cpu_vendor(struct cpuinfo_x86 *c)
558 char *v = c->x86_vendor_id;
561 for (i = 0; i < X86_VENDOR_NUM; i++) {
565 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
566 (cpu_devs[i]->c_ident[1] &&
567 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
569 this_cpu = cpu_devs[i];
570 c->x86_vendor = this_cpu->c_x86_vendor;
576 "CPU: vendor_id '%s' unknown, using generic init.\n" \
577 "CPU: Your system may be unstable.\n", v);
579 c->x86_vendor = X86_VENDOR_UNKNOWN;
580 this_cpu = &default_cpu;
583 void cpu_detect(struct cpuinfo_x86 *c)
585 /* Get vendor name */
586 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
587 (unsigned int *)&c->x86_vendor_id[0],
588 (unsigned int *)&c->x86_vendor_id[8],
589 (unsigned int *)&c->x86_vendor_id[4]);
592 /* Intel-defined flags: level 0x00000001 */
593 if (c->cpuid_level >= 0x00000001) {
594 u32 junk, tfms, cap0, misc;
596 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
597 c->x86 = (tfms >> 8) & 0xf;
598 c->x86_model = (tfms >> 4) & 0xf;
599 c->x86_mask = tfms & 0xf;
602 c->x86 += (tfms >> 20) & 0xff;
604 c->x86_model += ((tfms >> 16) & 0xf) << 4;
606 if (cap0 & (1<<19)) {
607 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
608 c->x86_cache_alignment = c->x86_clflush_size;
613 void get_cpu_cap(struct cpuinfo_x86 *c)
618 /* Intel-defined flags: level 0x00000001 */
619 if (c->cpuid_level >= 0x00000001) {
620 u32 capability, excap;
622 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
623 c->x86_capability[0] = capability;
624 c->x86_capability[4] = excap;
627 /* Additional Intel-defined flags: level 0x00000007 */
628 if (c->cpuid_level >= 0x00000007) {
629 u32 eax, ebx, ecx, edx;
631 cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
633 c->x86_capability[9] = ebx;
636 /* Extended state features: level 0x0000000d */
637 if (c->cpuid_level >= 0x0000000d) {
638 u32 eax, ebx, ecx, edx;
640 cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
642 c->x86_capability[10] = eax;
645 /* AMD-defined flags: level 0x80000001 */
646 xlvl = cpuid_eax(0x80000000);
647 c->extended_cpuid_level = xlvl;
649 if ((xlvl & 0xffff0000) == 0x80000000) {
650 if (xlvl >= 0x80000001) {
651 c->x86_capability[1] = cpuid_edx(0x80000001);
652 c->x86_capability[6] = cpuid_ecx(0x80000001);
656 if (c->extended_cpuid_level >= 0x80000008) {
657 u32 eax = cpuid_eax(0x80000008);
659 c->x86_virt_bits = (eax >> 8) & 0xff;
660 c->x86_phys_bits = eax & 0xff;
663 else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
664 c->x86_phys_bits = 36;
667 if (c->extended_cpuid_level >= 0x80000007)
668 c->x86_power = cpuid_edx(0x80000007);
670 init_scattered_cpuid_features(c);
673 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
679 * First of all, decide if this is a 486 or higher
680 * It's a 486 if we can modify the AC flag
682 if (flag_is_changeable_p(X86_EFLAGS_AC))
687 for (i = 0; i < X86_VENDOR_NUM; i++)
688 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
689 c->x86_vendor_id[0] = 0;
690 cpu_devs[i]->c_identify(c);
691 if (c->x86_vendor_id[0]) {
700 * Do minimum CPU detection early.
701 * Fields really needed: vendor, cpuid_level, family, model, mask,
703 * The others are not touched to avoid unwanted side effects.
705 * WARNING: this function is only called on the BP. Don't add code here
706 * that is supposed to run on all CPUs.
708 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
711 c->x86_clflush_size = 64;
712 c->x86_phys_bits = 36;
713 c->x86_virt_bits = 48;
715 c->x86_clflush_size = 32;
716 c->x86_phys_bits = 32;
717 c->x86_virt_bits = 32;
719 c->x86_cache_alignment = c->x86_clflush_size;
721 memset(&c->x86_capability, 0, sizeof c->x86_capability);
722 c->extended_cpuid_level = 0;
725 identify_cpu_without_cpuid(c);
727 /* cyrix could have cpuid enabled via c_identify()*/
736 if (this_cpu->c_early_init)
737 this_cpu->c_early_init(c);
740 filter_cpuid_features(c, false);
742 if (this_cpu->c_bsp_init)
743 this_cpu->c_bsp_init(c);
745 setup_force_cpu_cap(X86_FEATURE_ALWAYS);
748 void __init early_cpu_init(void)
750 const struct cpu_dev *const *cdev;
753 #ifdef CONFIG_PROCESSOR_SELECT
754 printk(KERN_INFO "KERNEL supported cpus:\n");
757 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
758 const struct cpu_dev *cpudev = *cdev;
760 if (count >= X86_VENDOR_NUM)
762 cpu_devs[count] = cpudev;
765 #ifdef CONFIG_PROCESSOR_SELECT
769 for (j = 0; j < 2; j++) {
770 if (!cpudev->c_ident[j])
772 printk(KERN_INFO " %s %s\n", cpudev->c_vendor,
778 early_identify_cpu(&boot_cpu_data);
782 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
783 * unfortunately, that's not true in practice because of early VIA
784 * chips and (more importantly) broken virtualizers that are not easy
785 * to detect. In the latter case it doesn't even *fail* reliably, so
786 * probing for it doesn't even work. Disable it completely on 32-bit
787 * unless we can find a reliable way to detect all the broken cases.
788 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
790 static void detect_nopl(struct cpuinfo_x86 *c)
793 clear_cpu_cap(c, X86_FEATURE_NOPL);
795 set_cpu_cap(c, X86_FEATURE_NOPL);
799 static void generic_identify(struct cpuinfo_x86 *c)
801 c->extended_cpuid_level = 0;
804 identify_cpu_without_cpuid(c);
806 /* cyrix could have cpuid enabled via c_identify()*/
816 if (c->cpuid_level >= 0x00000001) {
817 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
819 # ifdef CONFIG_X86_HT
820 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
822 c->apicid = c->initial_apicid;
825 c->phys_proc_id = c->initial_apicid;
828 get_model_name(c); /* Default name */
834 * This does the hard work of actually picking apart the CPU stuff...
836 static void identify_cpu(struct cpuinfo_x86 *c)
840 c->loops_per_jiffy = loops_per_jiffy;
841 c->x86_cache_size = -1;
842 c->x86_vendor = X86_VENDOR_UNKNOWN;
843 c->x86_model = c->x86_mask = 0; /* So far unknown... */
844 c->x86_vendor_id[0] = '\0'; /* Unset */
845 c->x86_model_id[0] = '\0'; /* Unset */
846 c->x86_max_cores = 1;
847 c->x86_coreid_bits = 0;
849 c->x86_clflush_size = 64;
850 c->x86_phys_bits = 36;
851 c->x86_virt_bits = 48;
853 c->cpuid_level = -1; /* CPUID not detected */
854 c->x86_clflush_size = 32;
855 c->x86_phys_bits = 32;
856 c->x86_virt_bits = 32;
858 c->x86_cache_alignment = c->x86_clflush_size;
859 memset(&c->x86_capability, 0, sizeof c->x86_capability);
863 if (this_cpu->c_identify)
864 this_cpu->c_identify(c);
866 /* Clear/Set all flags overriden by options, after probe */
867 for (i = 0; i < NCAPINTS; i++) {
868 c->x86_capability[i] &= ~cpu_caps_cleared[i];
869 c->x86_capability[i] |= cpu_caps_set[i];
873 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
877 * Vendor-specific initialization. In this section we
878 * canonicalize the feature flags, meaning if there are
879 * features a certain CPU supports which CPUID doesn't
880 * tell us, CPUID claiming incorrect flags, or other bugs,
881 * we handle them here.
883 * At the end of this section, c->x86_capability better
884 * indicate the features this CPU genuinely supports!
886 if (this_cpu->c_init)
889 /* Disable the PN if appropriate */
890 squash_the_stupid_serial_number(c);
892 /* Set up SMEP/SMAP */
897 * The vendor-specific functions might have changed features.
898 * Now we do "generic changes."
901 /* Filter out anything that depends on CPUID levels we don't have */
902 filter_cpuid_features(c, true);
904 /* If the model name is still unset, do table lookup. */
905 if (!c->x86_model_id[0]) {
907 p = table_lookup_model(c);
909 strcpy(c->x86_model_id, p);
912 sprintf(c->x86_model_id, "%02x/%02x",
913 c->x86, c->x86_model);
924 * Clear/Set all flags overriden by options, need do it
925 * before following smp all cpus cap AND.
927 for (i = 0; i < NCAPINTS; i++) {
928 c->x86_capability[i] &= ~cpu_caps_cleared[i];
929 c->x86_capability[i] |= cpu_caps_set[i];
933 * On SMP, boot_cpu_data holds the common feature set between
934 * all CPUs; so make sure that we indicate which features are
935 * common between the CPUs. The first time this routine gets
936 * executed, c == &boot_cpu_data.
938 if (c != &boot_cpu_data) {
939 /* AND the already accumulated flags with these */
940 for (i = 0; i < NCAPINTS; i++)
941 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
943 /* OR, i.e. replicate the bug flags */
944 for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
945 c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
948 /* Init Machine Check Exception if available. */
951 select_idle_routine(c);
954 numa_add_cpu(smp_processor_id());
959 static void vgetcpu_set_mode(void)
961 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
962 vgetcpu_mode = VGETCPU_RDTSCP;
964 vgetcpu_mode = VGETCPU_LSL;
967 /* May not be __init: called during resume */
968 static void syscall32_cpu_init(void)
970 /* Load these always in case some future AMD CPU supports
971 SYSENTER from compat mode too. */
972 wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
973 wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
974 wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)ia32_sysenter_target);
976 wrmsrl(MSR_CSTAR, ia32_cstar_target);
981 void enable_sep_cpu(void)
984 struct tss_struct *tss = &per_cpu(init_tss, cpu);
986 if (!boot_cpu_has(X86_FEATURE_SEP)) {
991 tss->x86_tss.ss1 = __KERNEL_CS;
992 tss->x86_tss.sp1 = sizeof(struct tss_struct) + (unsigned long) tss;
993 wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
994 wrmsr(MSR_IA32_SYSENTER_ESP, tss->x86_tss.sp1, 0);
995 wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long) ia32_sysenter_target, 0);
1000 void __init identify_boot_cpu(void)
1002 identify_cpu(&boot_cpu_data);
1003 init_amd_e400_c1e_mask();
1004 #ifdef CONFIG_X86_32
1010 cpu_detect_tlb(&boot_cpu_data);
1013 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1015 BUG_ON(c == &boot_cpu_data);
1017 #ifdef CONFIG_X86_32
1028 static const struct msr_range msr_range_array[] = {
1029 { 0x00000000, 0x00000418},
1030 { 0xc0000000, 0xc000040b},
1031 { 0xc0010000, 0xc0010142},
1032 { 0xc0011000, 0xc001103b},
1035 static void __print_cpu_msr(void)
1037 unsigned index_min, index_max;
1042 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
1043 index_min = msr_range_array[i].min;
1044 index_max = msr_range_array[i].max;
1046 for (index = index_min; index < index_max; index++) {
1047 if (rdmsrl_safe(index, &val))
1049 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
1054 static int show_msr;
1056 static __init int setup_show_msr(char *arg)
1060 get_option(&arg, &num);
1066 __setup("show_msr=", setup_show_msr);
1068 static __init int setup_noclflush(char *arg)
1070 setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1071 setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1074 __setup("noclflush", setup_noclflush);
1076 void print_cpu_info(struct cpuinfo_x86 *c)
1078 const char *vendor = NULL;
1080 if (c->x86_vendor < X86_VENDOR_NUM) {
1081 vendor = this_cpu->c_vendor;
1083 if (c->cpuid_level >= 0)
1084 vendor = c->x86_vendor_id;
1087 if (vendor && !strstr(c->x86_model_id, vendor))
1088 printk(KERN_CONT "%s ", vendor);
1090 if (c->x86_model_id[0])
1091 printk(KERN_CONT "%s", strim(c->x86_model_id));
1093 printk(KERN_CONT "%d86", c->x86);
1095 printk(KERN_CONT " (fam: %02x, model: %02x", c->x86, c->x86_model);
1097 if (c->x86_mask || c->cpuid_level >= 0)
1098 printk(KERN_CONT ", stepping: %02x)\n", c->x86_mask);
1100 printk(KERN_CONT ")\n");
1105 void print_cpu_msr(struct cpuinfo_x86 *c)
1107 if (c->cpu_index < show_msr)
1111 static __init int setup_disablecpuid(char *arg)
1115 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1116 setup_clear_cpu_cap(bit);
1122 __setup("clearcpuid=", setup_disablecpuid);
1124 DEFINE_PER_CPU(unsigned long, kernel_stack) =
1125 (unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
1126 EXPORT_PER_CPU_SYMBOL(kernel_stack);
1128 #ifdef CONFIG_X86_64
1129 struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
1130 struct desc_ptr debug_idt_descr = { NR_VECTORS * 16 - 1,
1131 (unsigned long) debug_idt_table };
1133 DEFINE_PER_CPU_FIRST(union irq_stack_union,
1134 irq_stack_union) __aligned(PAGE_SIZE) __visible;
1137 * The following four percpu variables are hot. Align current_task to
1138 * cacheline size such that all four fall in the same cacheline.
1140 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1142 EXPORT_PER_CPU_SYMBOL(current_task);
1144 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1145 init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1147 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1149 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1150 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1152 DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1155 * Special IST stacks which the CPU switches to when it calls
1156 * an IST-marked descriptor entry. Up to 7 stacks (hardware
1157 * limit), all of them are 4K, except the debug stack which
1160 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1161 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1162 [DEBUG_STACK - 1] = DEBUG_STKSZ
1165 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1166 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1168 /* May not be marked __init: used by software suspend */
1169 void syscall_init(void)
1172 * LSTAR and STAR live in a bit strange symbiosis.
1173 * They both write to the same internal register. STAR allows to
1174 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1176 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
1177 wrmsrl(MSR_LSTAR, system_call);
1178 wrmsrl(MSR_CSTAR, ignore_sysret);
1180 #ifdef CONFIG_IA32_EMULATION
1181 syscall32_cpu_init();
1184 /* Flags to clear on syscall */
1185 wrmsrl(MSR_SYSCALL_MASK,
1186 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1187 X86_EFLAGS_IOPL|X86_EFLAGS_AC);
1191 * Copies of the original ist values from the tss are only accessed during
1192 * debugging, no special alignment required.
1194 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1196 static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1197 DEFINE_PER_CPU(int, debug_stack_usage);
1199 int is_debug_stack(unsigned long addr)
1201 return __get_cpu_var(debug_stack_usage) ||
1202 (addr <= __get_cpu_var(debug_stack_addr) &&
1203 addr > (__get_cpu_var(debug_stack_addr) - DEBUG_STKSZ));
1205 NOKPROBE_SYMBOL(is_debug_stack);
1207 DEFINE_PER_CPU(u32, debug_idt_ctr);
1209 void debug_stack_set_zero(void)
1211 this_cpu_inc(debug_idt_ctr);
1214 NOKPROBE_SYMBOL(debug_stack_set_zero);
1216 void debug_stack_reset(void)
1218 if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1220 if (this_cpu_dec_return(debug_idt_ctr) == 0)
1223 NOKPROBE_SYMBOL(debug_stack_reset);
1225 #else /* CONFIG_X86_64 */
1227 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1228 EXPORT_PER_CPU_SYMBOL(current_task);
1229 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1230 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1231 DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1233 #ifdef CONFIG_CC_STACKPROTECTOR
1234 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1237 #endif /* CONFIG_X86_64 */
1240 * Clear all 6 debug registers:
1242 static void clear_all_debug_regs(void)
1246 for (i = 0; i < 8; i++) {
1247 /* Ignore db4, db5 */
1248 if ((i == 4) || (i == 5))
1257 * Restore debug regs if using kgdbwait and you have a kernel debugger
1258 * connection established.
1260 static void dbg_restore_debug_regs(void)
1262 if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1263 arch_kgdb_ops.correct_hw_break();
1265 #else /* ! CONFIG_KGDB */
1266 #define dbg_restore_debug_regs()
1267 #endif /* ! CONFIG_KGDB */
1270 * cpu_init() initializes state that is per-CPU. Some data is already
1271 * initialized (naturally) in the bootstrap process, such as the GDT
1272 * and IDT. We reload them nevertheless, this function acts as a
1273 * 'CPU state barrier', nothing should get across.
1274 * A lot of state is already set up in PDA init for 64 bit
1276 #ifdef CONFIG_X86_64
1280 struct orig_ist *oist;
1281 struct task_struct *me;
1282 struct tss_struct *t;
1288 * Load microcode on this cpu if a valid microcode is available.
1289 * This is early microcode loading procedure.
1293 cpu = stack_smp_processor_id();
1294 t = &per_cpu(init_tss, cpu);
1295 oist = &per_cpu(orig_ist, cpu);
1298 if (this_cpu_read(numa_node) == 0 &&
1299 early_cpu_to_node(cpu) != NUMA_NO_NODE)
1300 set_numa_node(early_cpu_to_node(cpu));
1305 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1306 panic("CPU#%d already initialized!\n", cpu);
1308 pr_debug("Initializing CPU#%d\n", cpu);
1310 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1313 * Initialize the per-CPU GDT with the boot GDT,
1314 * and set up the GDT descriptor:
1317 switch_to_new_gdt(cpu);
1322 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1325 wrmsrl(MSR_FS_BASE, 0);
1326 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1333 * set up and load the per-CPU TSS
1335 if (!oist->ist[0]) {
1336 char *estacks = per_cpu(exception_stacks, cpu);
1338 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1339 estacks += exception_stack_sizes[v];
1340 oist->ist[v] = t->x86_tss.ist[v] =
1341 (unsigned long)estacks;
1342 if (v == DEBUG_STACK-1)
1343 per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1347 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1350 * <= is required because the CPU will access up to
1351 * 8 bits beyond the end of the IO permission bitmap.
1353 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1354 t->io_bitmap[i] = ~0UL;
1356 atomic_inc(&init_mm.mm_count);
1357 me->active_mm = &init_mm;
1359 enter_lazy_tlb(&init_mm, me);
1361 load_sp0(t, ¤t->thread);
1362 set_tss_desc(cpu, t);
1364 load_LDT(&init_mm.context);
1366 clear_all_debug_regs();
1367 dbg_restore_debug_regs();
1379 int cpu = smp_processor_id();
1380 struct task_struct *curr = current;
1381 struct tss_struct *t = &per_cpu(init_tss, cpu);
1382 struct thread_struct *thread = &curr->thread;
1384 show_ucode_info_early();
1386 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1387 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1392 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1394 if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1395 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1398 switch_to_new_gdt(cpu);
1401 * Set up and load the per-CPU TSS and LDT
1403 atomic_inc(&init_mm.mm_count);
1404 curr->active_mm = &init_mm;
1406 enter_lazy_tlb(&init_mm, curr);
1408 load_sp0(t, thread);
1409 set_tss_desc(cpu, t);
1411 load_LDT(&init_mm.context);
1413 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1415 #ifdef CONFIG_DOUBLEFAULT
1416 /* Set up doublefault TSS pointer in the GDT */
1417 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1420 clear_all_debug_regs();
1421 dbg_restore_debug_regs();
1427 #ifdef CONFIG_X86_DEBUG_STATIC_CPU_HAS
1428 void warn_pre_alternatives(void)
1430 WARN(1, "You're using static_cpu_has before alternatives have run!\n");
1432 EXPORT_SYMBOL_GPL(warn_pre_alternatives);
1435 inline bool __static_cpu_has_safe(u16 bit)
1437 return boot_cpu_has(bit);
1439 EXPORT_SYMBOL_GPL(__static_cpu_has_safe);