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/kgdb.h>
12 #include <linux/smp.h>
15 #include <asm/stackprotector.h>
16 #include <asm/perf_event.h>
17 #include <asm/mmu_context.h>
18 #include <asm/archrandom.h>
19 #include <asm/hypervisor.h>
20 #include <asm/processor.h>
21 #include <asm/debugreg.h>
22 #include <asm/sections.h>
23 #include <linux/topology.h>
24 #include <linux/cpumask.h>
25 #include <asm/pgtable.h>
26 #include <linux/atomic.h>
27 #include <asm/proto.h>
28 #include <asm/setup.h>
32 #include <asm/fpu-internal.h>
34 #include <linux/numa.h>
41 #ifdef CONFIG_X86_LOCAL_APIC
42 #include <asm/uv/uv.h>
47 /* all of these masks are initialized in setup_cpu_local_masks() */
48 cpumask_var_t cpu_initialized_mask;
49 cpumask_var_t cpu_callout_mask;
50 cpumask_var_t cpu_callin_mask;
52 /* representing cpus for which sibling maps can be computed */
53 cpumask_var_t cpu_sibling_setup_mask;
55 /* correctly size the local cpu masks */
56 void __init setup_cpu_local_masks(void)
58 alloc_bootmem_cpumask_var(&cpu_initialized_mask);
59 alloc_bootmem_cpumask_var(&cpu_callin_mask);
60 alloc_bootmem_cpumask_var(&cpu_callout_mask);
61 alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
64 static void __cpuinit default_init(struct cpuinfo_x86 *c)
67 cpu_detect_cache_sizes(c);
69 /* Not much we can do here... */
70 /* Check if at least it has cpuid */
71 if (c->cpuid_level == -1) {
72 /* No cpuid. It must be an ancient CPU */
74 strcpy(c->x86_model_id, "486");
76 strcpy(c->x86_model_id, "386");
81 static const struct cpu_dev __cpuinitconst default_cpu = {
82 .c_init = default_init,
83 .c_vendor = "Unknown",
84 .c_x86_vendor = X86_VENDOR_UNKNOWN,
87 static const struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
89 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
92 * We need valid kernel segments for data and code in long mode too
93 * IRET will check the segment types kkeil 2000/10/28
94 * Also sysret mandates a special GDT layout
96 * TLS descriptors are currently at a different place compared to i386.
97 * Hopefully nobody expects them at a fixed place (Wine?)
99 [GDT_ENTRY_KERNEL32_CS] = GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
100 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
101 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
102 [GDT_ENTRY_DEFAULT_USER32_CS] = GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
103 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
104 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
106 [GDT_ENTRY_KERNEL_CS] = GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
107 [GDT_ENTRY_KERNEL_DS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
108 [GDT_ENTRY_DEFAULT_USER_CS] = GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
109 [GDT_ENTRY_DEFAULT_USER_DS] = GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
111 * Segments used for calling PnP BIOS have byte granularity.
112 * They code segments and data segments have fixed 64k limits,
113 * the transfer segment sizes are set at run time.
116 [GDT_ENTRY_PNPBIOS_CS32] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
118 [GDT_ENTRY_PNPBIOS_CS16] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
120 [GDT_ENTRY_PNPBIOS_DS] = GDT_ENTRY_INIT(0x0092, 0, 0xffff),
122 [GDT_ENTRY_PNPBIOS_TS1] = GDT_ENTRY_INIT(0x0092, 0, 0),
124 [GDT_ENTRY_PNPBIOS_TS2] = GDT_ENTRY_INIT(0x0092, 0, 0),
126 * The APM segments have byte granularity and their bases
127 * are set at run time. All have 64k limits.
130 [GDT_ENTRY_APMBIOS_BASE] = GDT_ENTRY_INIT(0x409a, 0, 0xffff),
132 [GDT_ENTRY_APMBIOS_BASE+1] = GDT_ENTRY_INIT(0x009a, 0, 0xffff),
134 [GDT_ENTRY_APMBIOS_BASE+2] = GDT_ENTRY_INIT(0x4092, 0, 0xffff),
136 [GDT_ENTRY_ESPFIX_SS] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
137 [GDT_ENTRY_PERCPU] = GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
138 GDT_STACK_CANARY_INIT
141 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
143 static int __init x86_xsave_setup(char *s)
145 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
146 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
147 setup_clear_cpu_cap(X86_FEATURE_AVX);
148 setup_clear_cpu_cap(X86_FEATURE_AVX2);
151 __setup("noxsave", x86_xsave_setup);
153 static int __init x86_xsaveopt_setup(char *s)
155 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
158 __setup("noxsaveopt", x86_xsaveopt_setup);
161 static int cachesize_override __cpuinitdata = -1;
162 static int disable_x86_serial_nr __cpuinitdata = 1;
164 static int __init cachesize_setup(char *str)
166 get_option(&str, &cachesize_override);
169 __setup("cachesize=", cachesize_setup);
171 static int __init x86_fxsr_setup(char *s)
173 setup_clear_cpu_cap(X86_FEATURE_FXSR);
174 setup_clear_cpu_cap(X86_FEATURE_XMM);
177 __setup("nofxsr", x86_fxsr_setup);
179 static int __init x86_sep_setup(char *s)
181 setup_clear_cpu_cap(X86_FEATURE_SEP);
184 __setup("nosep", x86_sep_setup);
186 /* Standard macro to see if a specific flag is changeable */
187 static inline int flag_is_changeable_p(u32 flag)
192 * Cyrix and IDT cpus allow disabling of CPUID
193 * so the code below may return different results
194 * when it is executed before and after enabling
195 * the CPUID. Add "volatile" to not allow gcc to
196 * optimize the subsequent calls to this function.
198 asm volatile ("pushfl \n\t"
209 : "=&r" (f1), "=&r" (f2)
212 return ((f1^f2) & flag) != 0;
215 /* Probe for the CPUID instruction */
216 static int __cpuinit have_cpuid_p(void)
218 return flag_is_changeable_p(X86_EFLAGS_ID);
221 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
223 unsigned long lo, hi;
225 if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
228 /* Disable processor serial number: */
230 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
232 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
234 printk(KERN_NOTICE "CPU serial number disabled.\n");
235 clear_cpu_cap(c, X86_FEATURE_PN);
237 /* Disabling the serial number may affect the cpuid level */
238 c->cpuid_level = cpuid_eax(0);
241 static int __init x86_serial_nr_setup(char *s)
243 disable_x86_serial_nr = 0;
246 __setup("serialnumber", x86_serial_nr_setup);
248 static inline int flag_is_changeable_p(u32 flag)
252 /* Probe for the CPUID instruction */
253 static inline int have_cpuid_p(void)
257 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
262 static int disable_smep __cpuinitdata;
263 static __init int setup_disable_smep(char *arg)
268 __setup("nosmep", setup_disable_smep);
270 static __cpuinit void setup_smep(struct cpuinfo_x86 *c)
272 if (cpu_has(c, X86_FEATURE_SMEP)) {
273 if (unlikely(disable_smep)) {
274 setup_clear_cpu_cap(X86_FEATURE_SMEP);
275 clear_in_cr4(X86_CR4_SMEP);
277 set_in_cr4(X86_CR4_SMEP);
282 * Some CPU features depend on higher CPUID levels, which may not always
283 * be available due to CPUID level capping or broken virtualization
284 * software. Add those features to this table to auto-disable them.
286 struct cpuid_dependent_feature {
291 static const struct cpuid_dependent_feature __cpuinitconst
292 cpuid_dependent_features[] = {
293 { X86_FEATURE_MWAIT, 0x00000005 },
294 { X86_FEATURE_DCA, 0x00000009 },
295 { X86_FEATURE_XSAVE, 0x0000000d },
299 static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
301 const struct cpuid_dependent_feature *df;
303 for (df = cpuid_dependent_features; df->feature; df++) {
305 if (!cpu_has(c, df->feature))
308 * Note: cpuid_level is set to -1 if unavailable, but
309 * extended_extended_level is set to 0 if unavailable
310 * and the legitimate extended levels are all negative
311 * when signed; hence the weird messing around with
314 if (!((s32)df->level < 0 ?
315 (u32)df->level > (u32)c->extended_cpuid_level :
316 (s32)df->level > (s32)c->cpuid_level))
319 clear_cpu_cap(c, df->feature);
324 "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
325 x86_cap_flags[df->feature], df->level);
330 * Naming convention should be: <Name> [(<Codename>)]
331 * This table only is used unless init_<vendor>() below doesn't set it;
332 * in particular, if CPUID levels 0x80000002..4 are supported, this
336 /* Look up CPU names by table lookup. */
337 static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
339 const struct cpu_model_info *info;
341 if (c->x86_model >= 16)
342 return NULL; /* Range check */
347 info = this_cpu->c_models;
349 while (info && info->family) {
350 if (info->family == c->x86)
351 return info->model_names[c->x86_model];
354 return NULL; /* Not found */
357 __u32 cpu_caps_cleared[NCAPINTS] __cpuinitdata;
358 __u32 cpu_caps_set[NCAPINTS] __cpuinitdata;
360 void load_percpu_segment(int cpu)
363 loadsegment(fs, __KERNEL_PERCPU);
366 wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
368 load_stack_canary_segment();
372 * Current gdt points %fs at the "master" per-cpu area: after this,
373 * it's on the real one.
375 void switch_to_new_gdt(int cpu)
377 struct desc_ptr gdt_descr;
379 gdt_descr.address = (long)get_cpu_gdt_table(cpu);
380 gdt_descr.size = GDT_SIZE - 1;
381 load_gdt(&gdt_descr);
382 /* Reload the per-cpu base */
384 load_percpu_segment(cpu);
387 static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
389 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
394 if (c->extended_cpuid_level < 0x80000004)
397 v = (unsigned int *)c->x86_model_id;
398 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
399 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
400 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
401 c->x86_model_id[48] = 0;
404 * Intel chips right-justify this string for some dumb reason;
405 * undo that brain damage:
407 p = q = &c->x86_model_id[0];
413 while (q <= &c->x86_model_id[48])
414 *q++ = '\0'; /* Zero-pad the rest */
418 void __cpuinit cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
420 unsigned int n, dummy, ebx, ecx, edx, l2size;
422 n = c->extended_cpuid_level;
424 if (n >= 0x80000005) {
425 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
426 c->x86_cache_size = (ecx>>24) + (edx>>24);
428 /* On K8 L1 TLB is inclusive, so don't count it */
433 if (n < 0x80000006) /* Some chips just has a large L1. */
436 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
440 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
442 /* do processor-specific cache resizing */
443 if (this_cpu->c_size_cache)
444 l2size = this_cpu->c_size_cache(c, l2size);
446 /* Allow user to override all this if necessary. */
447 if (cachesize_override != -1)
448 l2size = cachesize_override;
451 return; /* Again, no L2 cache is possible */
454 c->x86_cache_size = l2size;
457 u16 __read_mostly tlb_lli_4k[NR_INFO];
458 u16 __read_mostly tlb_lli_2m[NR_INFO];
459 u16 __read_mostly tlb_lli_4m[NR_INFO];
460 u16 __read_mostly tlb_lld_4k[NR_INFO];
461 u16 __read_mostly tlb_lld_2m[NR_INFO];
462 u16 __read_mostly tlb_lld_4m[NR_INFO];
465 * tlb_flushall_shift shows the balance point in replacing cr3 write
466 * with multiple 'invlpg'. It will do this replacement when
467 * flush_tlb_lines <= active_lines/2^tlb_flushall_shift.
468 * If tlb_flushall_shift is -1, means the replacement will be disabled.
470 s8 __read_mostly tlb_flushall_shift = -1;
472 void __cpuinit cpu_detect_tlb(struct cpuinfo_x86 *c)
474 if (this_cpu->c_detect_tlb)
475 this_cpu->c_detect_tlb(c);
477 printk(KERN_INFO "Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n" \
478 "Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d\n" \
479 "tlb_flushall_shift is 0x%x\n",
480 tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
481 tlb_lli_4m[ENTRIES], tlb_lld_4k[ENTRIES],
482 tlb_lld_2m[ENTRIES], tlb_lld_4m[ENTRIES],
486 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
489 u32 eax, ebx, ecx, edx;
490 int index_msb, core_bits;
493 if (!cpu_has(c, X86_FEATURE_HT))
496 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
499 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
502 cpuid(1, &eax, &ebx, &ecx, &edx);
504 smp_num_siblings = (ebx & 0xff0000) >> 16;
506 if (smp_num_siblings == 1) {
507 printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
511 if (smp_num_siblings <= 1)
514 index_msb = get_count_order(smp_num_siblings);
515 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
517 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
519 index_msb = get_count_order(smp_num_siblings);
521 core_bits = get_count_order(c->x86_max_cores);
523 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
524 ((1 << core_bits) - 1);
527 if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
528 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
530 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
537 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
539 char *v = c->x86_vendor_id;
542 for (i = 0; i < X86_VENDOR_NUM; i++) {
546 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
547 (cpu_devs[i]->c_ident[1] &&
548 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
550 this_cpu = cpu_devs[i];
551 c->x86_vendor = this_cpu->c_x86_vendor;
557 "CPU: vendor_id '%s' unknown, using generic init.\n" \
558 "CPU: Your system may be unstable.\n", v);
560 c->x86_vendor = X86_VENDOR_UNKNOWN;
561 this_cpu = &default_cpu;
564 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
566 /* Get vendor name */
567 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
568 (unsigned int *)&c->x86_vendor_id[0],
569 (unsigned int *)&c->x86_vendor_id[8],
570 (unsigned int *)&c->x86_vendor_id[4]);
573 /* Intel-defined flags: level 0x00000001 */
574 if (c->cpuid_level >= 0x00000001) {
575 u32 junk, tfms, cap0, misc;
577 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
578 c->x86 = (tfms >> 8) & 0xf;
579 c->x86_model = (tfms >> 4) & 0xf;
580 c->x86_mask = tfms & 0xf;
583 c->x86 += (tfms >> 20) & 0xff;
585 c->x86_model += ((tfms >> 16) & 0xf) << 4;
587 if (cap0 & (1<<19)) {
588 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
589 c->x86_cache_alignment = c->x86_clflush_size;
594 void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
599 /* Intel-defined flags: level 0x00000001 */
600 if (c->cpuid_level >= 0x00000001) {
601 u32 capability, excap;
603 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
604 c->x86_capability[0] = capability;
605 c->x86_capability[4] = excap;
608 /* Additional Intel-defined flags: level 0x00000007 */
609 if (c->cpuid_level >= 0x00000007) {
610 u32 eax, ebx, ecx, edx;
612 cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
614 c->x86_capability[9] = ebx;
617 /* AMD-defined flags: level 0x80000001 */
618 xlvl = cpuid_eax(0x80000000);
619 c->extended_cpuid_level = xlvl;
621 if ((xlvl & 0xffff0000) == 0x80000000) {
622 if (xlvl >= 0x80000001) {
623 c->x86_capability[1] = cpuid_edx(0x80000001);
624 c->x86_capability[6] = cpuid_ecx(0x80000001);
628 if (c->extended_cpuid_level >= 0x80000008) {
629 u32 eax = cpuid_eax(0x80000008);
631 c->x86_virt_bits = (eax >> 8) & 0xff;
632 c->x86_phys_bits = eax & 0xff;
635 else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
636 c->x86_phys_bits = 36;
639 if (c->extended_cpuid_level >= 0x80000007)
640 c->x86_power = cpuid_edx(0x80000007);
642 init_scattered_cpuid_features(c);
645 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
651 * First of all, decide if this is a 486 or higher
652 * It's a 486 if we can modify the AC flag
654 if (flag_is_changeable_p(X86_EFLAGS_AC))
659 for (i = 0; i < X86_VENDOR_NUM; i++)
660 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
661 c->x86_vendor_id[0] = 0;
662 cpu_devs[i]->c_identify(c);
663 if (c->x86_vendor_id[0]) {
672 * Do minimum CPU detection early.
673 * Fields really needed: vendor, cpuid_level, family, model, mask,
675 * The others are not touched to avoid unwanted side effects.
677 * WARNING: this function is only called on the BP. Don't add code here
678 * that is supposed to run on all CPUs.
680 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
683 c->x86_clflush_size = 64;
684 c->x86_phys_bits = 36;
685 c->x86_virt_bits = 48;
687 c->x86_clflush_size = 32;
688 c->x86_phys_bits = 32;
689 c->x86_virt_bits = 32;
691 c->x86_cache_alignment = c->x86_clflush_size;
693 memset(&c->x86_capability, 0, sizeof c->x86_capability);
694 c->extended_cpuid_level = 0;
697 identify_cpu_without_cpuid(c);
699 /* cyrix could have cpuid enabled via c_identify()*/
709 if (this_cpu->c_early_init)
710 this_cpu->c_early_init(c);
713 filter_cpuid_features(c, false);
717 if (this_cpu->c_bsp_init)
718 this_cpu->c_bsp_init(c);
721 void __init early_cpu_init(void)
723 const struct cpu_dev *const *cdev;
726 #ifdef CONFIG_PROCESSOR_SELECT
727 printk(KERN_INFO "KERNEL supported cpus:\n");
730 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
731 const struct cpu_dev *cpudev = *cdev;
733 if (count >= X86_VENDOR_NUM)
735 cpu_devs[count] = cpudev;
738 #ifdef CONFIG_PROCESSOR_SELECT
742 for (j = 0; j < 2; j++) {
743 if (!cpudev->c_ident[j])
745 printk(KERN_INFO " %s %s\n", cpudev->c_vendor,
751 early_identify_cpu(&boot_cpu_data);
755 * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
756 * unfortunately, that's not true in practice because of early VIA
757 * chips and (more importantly) broken virtualizers that are not easy
758 * to detect. In the latter case it doesn't even *fail* reliably, so
759 * probing for it doesn't even work. Disable it completely on 32-bit
760 * unless we can find a reliable way to detect all the broken cases.
761 * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
763 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
766 clear_cpu_cap(c, X86_FEATURE_NOPL);
768 set_cpu_cap(c, X86_FEATURE_NOPL);
772 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
774 c->extended_cpuid_level = 0;
777 identify_cpu_without_cpuid(c);
779 /* cyrix could have cpuid enabled via c_identify()*/
789 if (c->cpuid_level >= 0x00000001) {
790 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
792 # ifdef CONFIG_X86_HT
793 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
795 c->apicid = c->initial_apicid;
798 c->phys_proc_id = c->initial_apicid;
803 get_model_name(c); /* Default name */
809 * This does the hard work of actually picking apart the CPU stuff...
811 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
815 c->loops_per_jiffy = loops_per_jiffy;
816 c->x86_cache_size = -1;
817 c->x86_vendor = X86_VENDOR_UNKNOWN;
818 c->x86_model = c->x86_mask = 0; /* So far unknown... */
819 c->x86_vendor_id[0] = '\0'; /* Unset */
820 c->x86_model_id[0] = '\0'; /* Unset */
821 c->x86_max_cores = 1;
822 c->x86_coreid_bits = 0;
824 c->x86_clflush_size = 64;
825 c->x86_phys_bits = 36;
826 c->x86_virt_bits = 48;
828 c->cpuid_level = -1; /* CPUID not detected */
829 c->x86_clflush_size = 32;
830 c->x86_phys_bits = 32;
831 c->x86_virt_bits = 32;
833 c->x86_cache_alignment = c->x86_clflush_size;
834 memset(&c->x86_capability, 0, sizeof c->x86_capability);
838 if (this_cpu->c_identify)
839 this_cpu->c_identify(c);
841 /* Clear/Set all flags overriden by options, after probe */
842 for (i = 0; i < NCAPINTS; i++) {
843 c->x86_capability[i] &= ~cpu_caps_cleared[i];
844 c->x86_capability[i] |= cpu_caps_set[i];
848 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
852 * Vendor-specific initialization. In this section we
853 * canonicalize the feature flags, meaning if there are
854 * features a certain CPU supports which CPUID doesn't
855 * tell us, CPUID claiming incorrect flags, or other bugs,
856 * we handle them here.
858 * At the end of this section, c->x86_capability better
859 * indicate the features this CPU genuinely supports!
861 if (this_cpu->c_init)
864 /* Disable the PN if appropriate */
865 squash_the_stupid_serial_number(c);
868 * The vendor-specific functions might have changed features.
869 * Now we do "generic changes."
872 /* Filter out anything that depends on CPUID levels we don't have */
873 filter_cpuid_features(c, true);
875 /* If the model name is still unset, do table lookup. */
876 if (!c->x86_model_id[0]) {
878 p = table_lookup_model(c);
880 strcpy(c->x86_model_id, p);
883 sprintf(c->x86_model_id, "%02x/%02x",
884 c->x86, c->x86_model);
895 * Clear/Set all flags overriden by options, need do it
896 * before following smp all cpus cap AND.
898 for (i = 0; i < NCAPINTS; i++) {
899 c->x86_capability[i] &= ~cpu_caps_cleared[i];
900 c->x86_capability[i] |= cpu_caps_set[i];
904 * On SMP, boot_cpu_data holds the common feature set between
905 * all CPUs; so make sure that we indicate which features are
906 * common between the CPUs. The first time this routine gets
907 * executed, c == &boot_cpu_data.
909 if (c != &boot_cpu_data) {
910 /* AND the already accumulated flags with these */
911 for (i = 0; i < NCAPINTS; i++)
912 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
915 /* Init Machine Check Exception if available. */
918 select_idle_routine(c);
921 numa_add_cpu(smp_processor_id());
926 static void vgetcpu_set_mode(void)
928 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
929 vgetcpu_mode = VGETCPU_RDTSCP;
931 vgetcpu_mode = VGETCPU_LSL;
935 void __init identify_boot_cpu(void)
937 identify_cpu(&boot_cpu_data);
938 init_amd_e400_c1e_mask();
945 if (boot_cpu_data.cpuid_level >= 2)
946 cpu_detect_tlb(&boot_cpu_data);
949 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
951 BUG_ON(c == &boot_cpu_data);
964 static const struct msr_range msr_range_array[] __cpuinitconst = {
965 { 0x00000000, 0x00000418},
966 { 0xc0000000, 0xc000040b},
967 { 0xc0010000, 0xc0010142},
968 { 0xc0011000, 0xc001103b},
971 static void __cpuinit __print_cpu_msr(void)
973 unsigned index_min, index_max;
978 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
979 index_min = msr_range_array[i].min;
980 index_max = msr_range_array[i].max;
982 for (index = index_min; index < index_max; index++) {
983 if (rdmsrl_safe(index, &val))
985 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
990 static int show_msr __cpuinitdata;
992 static __init int setup_show_msr(char *arg)
996 get_option(&arg, &num);
1002 __setup("show_msr=", setup_show_msr);
1004 static __init int setup_noclflush(char *arg)
1006 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
1009 __setup("noclflush", setup_noclflush);
1011 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1013 const char *vendor = NULL;
1015 if (c->x86_vendor < X86_VENDOR_NUM) {
1016 vendor = this_cpu->c_vendor;
1018 if (c->cpuid_level >= 0)
1019 vendor = c->x86_vendor_id;
1022 if (vendor && !strstr(c->x86_model_id, vendor))
1023 printk(KERN_CONT "%s ", vendor);
1025 if (c->x86_model_id[0])
1026 printk(KERN_CONT "%s", strim(c->x86_model_id));
1028 printk(KERN_CONT "%d86", c->x86);
1030 printk(KERN_CONT " (fam: %02x, model: %02x", c->x86, c->x86_model);
1032 if (c->x86_mask || c->cpuid_level >= 0)
1033 printk(KERN_CONT ", stepping: %02x)\n", c->x86_mask);
1035 printk(KERN_CONT ")\n");
1040 void __cpuinit print_cpu_msr(struct cpuinfo_x86 *c)
1042 if (c->cpu_index < show_msr)
1046 static __init int setup_disablecpuid(char *arg)
1050 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1051 setup_clear_cpu_cap(bit);
1057 __setup("clearcpuid=", setup_disablecpuid);
1059 #ifdef CONFIG_X86_64
1060 struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
1061 struct desc_ptr nmi_idt_descr = { NR_VECTORS * 16 - 1,
1062 (unsigned long) nmi_idt_table };
1064 DEFINE_PER_CPU_FIRST(union irq_stack_union,
1065 irq_stack_union) __aligned(PAGE_SIZE);
1068 * The following four percpu variables are hot. Align current_task to
1069 * cacheline size such that all four fall in the same cacheline.
1071 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1073 EXPORT_PER_CPU_SYMBOL(current_task);
1075 DEFINE_PER_CPU(unsigned long, kernel_stack) =
1076 (unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
1077 EXPORT_PER_CPU_SYMBOL(kernel_stack);
1079 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1080 init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1082 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
1084 DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1087 * Special IST stacks which the CPU switches to when it calls
1088 * an IST-marked descriptor entry. Up to 7 stacks (hardware
1089 * limit), all of them are 4K, except the debug stack which
1092 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1093 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1094 [DEBUG_STACK - 1] = DEBUG_STKSZ
1097 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1098 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1100 /* May not be marked __init: used by software suspend */
1101 void syscall_init(void)
1104 * LSTAR and STAR live in a bit strange symbiosis.
1105 * They both write to the same internal register. STAR allows to
1106 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1108 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
1109 wrmsrl(MSR_LSTAR, system_call);
1110 wrmsrl(MSR_CSTAR, ignore_sysret);
1112 #ifdef CONFIG_IA32_EMULATION
1113 syscall32_cpu_init();
1116 /* Flags to clear on syscall */
1117 wrmsrl(MSR_SYSCALL_MASK,
1118 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
1122 * Copies of the original ist values from the tss are only accessed during
1123 * debugging, no special alignment required.
1125 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1127 static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1128 DEFINE_PER_CPU(int, debug_stack_usage);
1130 int is_debug_stack(unsigned long addr)
1132 return __get_cpu_var(debug_stack_usage) ||
1133 (addr <= __get_cpu_var(debug_stack_addr) &&
1134 addr > (__get_cpu_var(debug_stack_addr) - DEBUG_STKSZ));
1137 static DEFINE_PER_CPU(u32, debug_stack_use_ctr);
1139 void debug_stack_set_zero(void)
1141 this_cpu_inc(debug_stack_use_ctr);
1142 load_idt((const struct desc_ptr *)&nmi_idt_descr);
1145 void debug_stack_reset(void)
1147 if (WARN_ON(!this_cpu_read(debug_stack_use_ctr)))
1149 if (this_cpu_dec_return(debug_stack_use_ctr) == 0)
1150 load_idt((const struct desc_ptr *)&idt_descr);
1153 #else /* CONFIG_X86_64 */
1155 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1156 EXPORT_PER_CPU_SYMBOL(current_task);
1157 DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1159 #ifdef CONFIG_CC_STACKPROTECTOR
1160 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1163 /* Make sure %fs and %gs are initialized properly in idle threads */
1164 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
1166 memset(regs, 0, sizeof(struct pt_regs));
1167 regs->fs = __KERNEL_PERCPU;
1168 regs->gs = __KERNEL_STACK_CANARY;
1172 #endif /* CONFIG_X86_64 */
1175 * Clear all 6 debug registers:
1177 static void clear_all_debug_regs(void)
1181 for (i = 0; i < 8; i++) {
1182 /* Ignore db4, db5 */
1183 if ((i == 4) || (i == 5))
1192 * Restore debug regs if using kgdbwait and you have a kernel debugger
1193 * connection established.
1195 static void dbg_restore_debug_regs(void)
1197 if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1198 arch_kgdb_ops.correct_hw_break();
1200 #else /* ! CONFIG_KGDB */
1201 #define dbg_restore_debug_regs()
1202 #endif /* ! CONFIG_KGDB */
1205 * cpu_init() initializes state that is per-CPU. Some data is already
1206 * initialized (naturally) in the bootstrap process, such as the GDT
1207 * and IDT. We reload them nevertheless, this function acts as a
1208 * 'CPU state barrier', nothing should get across.
1209 * A lot of state is already set up in PDA init for 64 bit
1211 #ifdef CONFIG_X86_64
1213 void __cpuinit cpu_init(void)
1215 struct orig_ist *oist;
1216 struct task_struct *me;
1217 struct tss_struct *t;
1222 cpu = stack_smp_processor_id();
1223 t = &per_cpu(init_tss, cpu);
1224 oist = &per_cpu(orig_ist, cpu);
1227 if (cpu != 0 && this_cpu_read(numa_node) == 0 &&
1228 early_cpu_to_node(cpu) != NUMA_NO_NODE)
1229 set_numa_node(early_cpu_to_node(cpu));
1234 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1235 panic("CPU#%d already initialized!\n", cpu);
1237 pr_debug("Initializing CPU#%d\n", cpu);
1239 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1242 * Initialize the per-CPU GDT with the boot GDT,
1243 * and set up the GDT descriptor:
1246 switch_to_new_gdt(cpu);
1249 load_idt((const struct desc_ptr *)&idt_descr);
1251 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1254 wrmsrl(MSR_FS_BASE, 0);
1255 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1263 * set up and load the per-CPU TSS
1265 if (!oist->ist[0]) {
1266 char *estacks = per_cpu(exception_stacks, cpu);
1268 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1269 estacks += exception_stack_sizes[v];
1270 oist->ist[v] = t->x86_tss.ist[v] =
1271 (unsigned long)estacks;
1272 if (v == DEBUG_STACK-1)
1273 per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1277 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1280 * <= is required because the CPU will access up to
1281 * 8 bits beyond the end of the IO permission bitmap.
1283 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1284 t->io_bitmap[i] = ~0UL;
1286 atomic_inc(&init_mm.mm_count);
1287 me->active_mm = &init_mm;
1289 enter_lazy_tlb(&init_mm, me);
1291 load_sp0(t, ¤t->thread);
1292 set_tss_desc(cpu, t);
1294 load_LDT(&init_mm.context);
1296 clear_all_debug_regs();
1297 dbg_restore_debug_regs();
1307 void __cpuinit cpu_init(void)
1309 int cpu = smp_processor_id();
1310 struct task_struct *curr = current;
1311 struct tss_struct *t = &per_cpu(init_tss, cpu);
1312 struct thread_struct *thread = &curr->thread;
1314 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1315 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1320 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1322 if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1323 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1325 load_idt(&idt_descr);
1326 switch_to_new_gdt(cpu);
1329 * Set up and load the per-CPU TSS and LDT
1331 atomic_inc(&init_mm.mm_count);
1332 curr->active_mm = &init_mm;
1334 enter_lazy_tlb(&init_mm, curr);
1336 load_sp0(t, thread);
1337 set_tss_desc(cpu, t);
1339 load_LDT(&init_mm.context);
1341 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1343 #ifdef CONFIG_DOUBLEFAULT
1344 /* Set up doublefault TSS pointer in the GDT */
1345 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1348 clear_all_debug_regs();
1349 dbg_restore_debug_regs();