1 #include <linux/init.h>
2 #include <linux/kernel.h>
3 #include <linux/sched.h>
4 #include <linux/string.h>
5 #include <linux/bootmem.h>
6 #include <linux/bitops.h>
7 #include <linux/module.h>
8 #include <linux/kgdb.h>
9 #include <linux/topology.h>
10 #include <linux/delay.h>
11 #include <linux/smp.h>
12 #include <linux/percpu.h>
16 #include <asm/linkage.h>
17 #include <asm/mmu_context.h>
25 #include <asm/cpumask.h>
26 #ifdef CONFIG_X86_LOCAL_APIC
27 #include <asm/mpspec.h>
29 #include <asm/genapic.h>
30 #include <asm/genapic.h>
31 #include <asm/uv/uv.h>
34 #include <asm/pgtable.h>
35 #include <asm/processor.h>
37 #include <asm/atomic.h>
38 #include <asm/proto.h>
39 #include <asm/sections.h>
40 #include <asm/setup.h>
41 #include <asm/hypervisor.h>
47 /* all of these masks are initialized in setup_cpu_local_masks() */
48 cpumask_var_t cpu_callin_mask;
49 cpumask_var_t cpu_callout_mask;
50 cpumask_var_t cpu_initialized_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 #else /* CONFIG_X86_32 */
66 cpumask_t cpu_callin_map;
67 cpumask_t cpu_callout_map;
68 cpumask_t cpu_initialized;
69 cpumask_t cpu_sibling_setup_map;
71 #endif /* CONFIG_X86_32 */
74 static struct cpu_dev *this_cpu __cpuinitdata;
76 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
79 * We need valid kernel segments for data and code in long mode too
80 * IRET will check the segment types kkeil 2000/10/28
81 * Also sysret mandates a special GDT layout
83 * The TLS descriptors are currently at a different place compared to i386.
84 * Hopefully nobody expects them at a fixed place (Wine?)
86 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
87 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
88 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
89 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
90 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
91 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
93 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } },
94 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } },
95 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } },
96 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff200 } } },
98 * Segments used for calling PnP BIOS have byte granularity.
99 * They code segments and data segments have fixed 64k limits,
100 * the transfer segment sizes are set at run time.
103 [GDT_ENTRY_PNPBIOS_CS32] = { { { 0x0000ffff, 0x00409a00 } } },
105 [GDT_ENTRY_PNPBIOS_CS16] = { { { 0x0000ffff, 0x00009a00 } } },
107 [GDT_ENTRY_PNPBIOS_DS] = { { { 0x0000ffff, 0x00009200 } } },
109 [GDT_ENTRY_PNPBIOS_TS1] = { { { 0x00000000, 0x00009200 } } },
111 [GDT_ENTRY_PNPBIOS_TS2] = { { { 0x00000000, 0x00009200 } } },
113 * The APM segments have byte granularity and their bases
114 * are set at run time. All have 64k limits.
117 [GDT_ENTRY_APMBIOS_BASE] = { { { 0x0000ffff, 0x00409a00 } } },
119 [GDT_ENTRY_APMBIOS_BASE+1] = { { { 0x0000ffff, 0x00009a00 } } },
121 [GDT_ENTRY_APMBIOS_BASE+2] = { { { 0x0000ffff, 0x00409200 } } },
123 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } },
124 [GDT_ENTRY_PERCPU] = { { { 0x0000ffff, 0x00cf9200 } } },
127 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
130 static int cachesize_override __cpuinitdata = -1;
131 static int disable_x86_serial_nr __cpuinitdata = 1;
133 static int __init cachesize_setup(char *str)
135 get_option(&str, &cachesize_override);
138 __setup("cachesize=", cachesize_setup);
140 static int __init x86_fxsr_setup(char *s)
142 setup_clear_cpu_cap(X86_FEATURE_FXSR);
143 setup_clear_cpu_cap(X86_FEATURE_XMM);
146 __setup("nofxsr", x86_fxsr_setup);
148 static int __init x86_sep_setup(char *s)
150 setup_clear_cpu_cap(X86_FEATURE_SEP);
153 __setup("nosep", x86_sep_setup);
155 /* Standard macro to see if a specific flag is changeable */
156 static inline int flag_is_changeable_p(u32 flag)
161 * Cyrix and IDT cpus allow disabling of CPUID
162 * so the code below may return different results
163 * when it is executed before and after enabling
164 * the CPUID. Add "volatile" to not allow gcc to
165 * optimize the subsequent calls to this function.
167 asm volatile ("pushfl\n\t"
177 : "=&r" (f1), "=&r" (f2)
180 return ((f1^f2) & flag) != 0;
183 /* Probe for the CPUID instruction */
184 static int __cpuinit have_cpuid_p(void)
186 return flag_is_changeable_p(X86_EFLAGS_ID);
189 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
191 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
192 /* Disable processor serial number */
193 unsigned long lo, hi;
194 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
196 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
197 printk(KERN_NOTICE "CPU serial number disabled.\n");
198 clear_cpu_cap(c, X86_FEATURE_PN);
200 /* Disabling the serial number may affect the cpuid level */
201 c->cpuid_level = cpuid_eax(0);
205 static int __init x86_serial_nr_setup(char *s)
207 disable_x86_serial_nr = 0;
210 __setup("serialnumber", x86_serial_nr_setup);
212 static inline int flag_is_changeable_p(u32 flag)
216 /* Probe for the CPUID instruction */
217 static inline int have_cpuid_p(void)
221 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
227 * Some CPU features depend on higher CPUID levels, which may not always
228 * be available due to CPUID level capping or broken virtualization
229 * software. Add those features to this table to auto-disable them.
231 struct cpuid_dependent_feature {
235 static const struct cpuid_dependent_feature __cpuinitconst
236 cpuid_dependent_features[] = {
237 { X86_FEATURE_MWAIT, 0x00000005 },
238 { X86_FEATURE_DCA, 0x00000009 },
239 { X86_FEATURE_XSAVE, 0x0000000d },
243 static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
245 const struct cpuid_dependent_feature *df;
246 for (df = cpuid_dependent_features; df->feature; df++) {
248 * Note: cpuid_level is set to -1 if unavailable, but
249 * extended_extended_level is set to 0 if unavailable
250 * and the legitimate extended levels are all negative
251 * when signed; hence the weird messing around with
254 if (cpu_has(c, df->feature) &&
255 ((s32)df->feature < 0 ?
256 (u32)df->feature > (u32)c->extended_cpuid_level :
257 (s32)df->feature > (s32)c->cpuid_level)) {
258 clear_cpu_cap(c, df->feature);
261 "CPU: CPU feature %s disabled "
262 "due to lack of CPUID level 0x%x\n",
263 x86_cap_flags[df->feature],
270 * Naming convention should be: <Name> [(<Codename>)]
271 * This table only is used unless init_<vendor>() below doesn't set it;
272 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
276 /* Look up CPU names by table lookup. */
277 static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
279 struct cpu_model_info *info;
281 if (c->x86_model >= 16)
282 return NULL; /* Range check */
287 info = this_cpu->c_models;
289 while (info && info->family) {
290 if (info->family == c->x86)
291 return info->model_names[c->x86_model];
294 return NULL; /* Not found */
297 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
299 /* Current gdt points %fs at the "master" per-cpu area: after this,
300 * it's on the real one. */
301 void switch_to_new_gdt(void)
303 struct desc_ptr gdt_descr;
304 int cpu = smp_processor_id();
306 gdt_descr.address = (long)get_cpu_gdt_table(cpu);
307 gdt_descr.size = GDT_SIZE - 1;
308 load_gdt(&gdt_descr);
309 /* Reload the per-cpu base */
311 loadsegment(fs, __KERNEL_PERCPU);
314 wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
318 static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
320 static void __cpuinit default_init(struct cpuinfo_x86 *c)
323 display_cacheinfo(c);
325 /* Not much we can do here... */
326 /* Check if at least it has cpuid */
327 if (c->cpuid_level == -1) {
328 /* No cpuid. It must be an ancient CPU */
330 strcpy(c->x86_model_id, "486");
331 else if (c->x86 == 3)
332 strcpy(c->x86_model_id, "386");
337 static struct cpu_dev __cpuinitdata default_cpu = {
338 .c_init = default_init,
339 .c_vendor = "Unknown",
340 .c_x86_vendor = X86_VENDOR_UNKNOWN,
343 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
348 if (c->extended_cpuid_level < 0x80000004)
351 v = (unsigned int *) c->x86_model_id;
352 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
353 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
354 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
355 c->x86_model_id[48] = 0;
357 /* Intel chips right-justify this string for some dumb reason;
358 undo that brain damage */
359 p = q = &c->x86_model_id[0];
365 while (q <= &c->x86_model_id[48])
366 *q++ = '\0'; /* Zero-pad the rest */
370 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
372 unsigned int n, dummy, ebx, ecx, edx, l2size;
374 n = c->extended_cpuid_level;
376 if (n >= 0x80000005) {
377 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
378 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
379 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
380 c->x86_cache_size = (ecx>>24) + (edx>>24);
382 /* On K8 L1 TLB is inclusive, so don't count it */
387 if (n < 0x80000006) /* Some chips just has a large L1. */
390 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
394 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
396 /* do processor-specific cache resizing */
397 if (this_cpu->c_size_cache)
398 l2size = this_cpu->c_size_cache(c, l2size);
400 /* Allow user to override all this if necessary. */
401 if (cachesize_override != -1)
402 l2size = cachesize_override;
405 return; /* Again, no L2 cache is possible */
408 c->x86_cache_size = l2size;
410 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
414 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
417 u32 eax, ebx, ecx, edx;
418 int index_msb, core_bits;
420 if (!cpu_has(c, X86_FEATURE_HT))
423 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
426 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
429 cpuid(1, &eax, &ebx, &ecx, &edx);
431 smp_num_siblings = (ebx & 0xff0000) >> 16;
433 if (smp_num_siblings == 1) {
434 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
435 } else if (smp_num_siblings > 1) {
437 if (smp_num_siblings > nr_cpu_ids) {
438 printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
440 smp_num_siblings = 1;
444 index_msb = get_count_order(smp_num_siblings);
445 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
447 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
449 index_msb = get_count_order(smp_num_siblings);
451 core_bits = get_count_order(c->x86_max_cores);
453 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
454 ((1 << core_bits) - 1);
458 if ((c->x86_max_cores * smp_num_siblings) > 1) {
459 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
461 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
467 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
469 char *v = c->x86_vendor_id;
473 for (i = 0; i < X86_VENDOR_NUM; i++) {
477 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
478 (cpu_devs[i]->c_ident[1] &&
479 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
480 this_cpu = cpu_devs[i];
481 c->x86_vendor = this_cpu->c_x86_vendor;
488 printk(KERN_ERR "CPU: vendor_id '%s' unknown, using generic init.\n", v);
489 printk(KERN_ERR "CPU: Your system may be unstable.\n");
492 c->x86_vendor = X86_VENDOR_UNKNOWN;
493 this_cpu = &default_cpu;
496 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
498 /* Get vendor name */
499 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
500 (unsigned int *)&c->x86_vendor_id[0],
501 (unsigned int *)&c->x86_vendor_id[8],
502 (unsigned int *)&c->x86_vendor_id[4]);
505 /* Intel-defined flags: level 0x00000001 */
506 if (c->cpuid_level >= 0x00000001) {
507 u32 junk, tfms, cap0, misc;
508 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
509 c->x86 = (tfms >> 8) & 0xf;
510 c->x86_model = (tfms >> 4) & 0xf;
511 c->x86_mask = tfms & 0xf;
513 c->x86 += (tfms >> 20) & 0xff;
515 c->x86_model += ((tfms >> 16) & 0xf) << 4;
516 if (cap0 & (1<<19)) {
517 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
518 c->x86_cache_alignment = c->x86_clflush_size;
523 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
528 /* Intel-defined flags: level 0x00000001 */
529 if (c->cpuid_level >= 0x00000001) {
530 u32 capability, excap;
531 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
532 c->x86_capability[0] = capability;
533 c->x86_capability[4] = excap;
536 /* AMD-defined flags: level 0x80000001 */
537 xlvl = cpuid_eax(0x80000000);
538 c->extended_cpuid_level = xlvl;
539 if ((xlvl & 0xffff0000) == 0x80000000) {
540 if (xlvl >= 0x80000001) {
541 c->x86_capability[1] = cpuid_edx(0x80000001);
542 c->x86_capability[6] = cpuid_ecx(0x80000001);
547 if (c->extended_cpuid_level >= 0x80000008) {
548 u32 eax = cpuid_eax(0x80000008);
550 c->x86_virt_bits = (eax >> 8) & 0xff;
551 c->x86_phys_bits = eax & 0xff;
555 if (c->extended_cpuid_level >= 0x80000007)
556 c->x86_power = cpuid_edx(0x80000007);
560 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
566 * First of all, decide if this is a 486 or higher
567 * It's a 486 if we can modify the AC flag
569 if (flag_is_changeable_p(X86_EFLAGS_AC))
574 for (i = 0; i < X86_VENDOR_NUM; i++)
575 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
576 c->x86_vendor_id[0] = 0;
577 cpu_devs[i]->c_identify(c);
578 if (c->x86_vendor_id[0]) {
587 * Do minimum CPU detection early.
588 * Fields really needed: vendor, cpuid_level, family, model, mask,
590 * The others are not touched to avoid unwanted side effects.
592 * WARNING: this function is only called on the BP. Don't add code here
593 * that is supposed to run on all CPUs.
595 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
598 c->x86_clflush_size = 64;
600 c->x86_clflush_size = 32;
602 c->x86_cache_alignment = c->x86_clflush_size;
604 memset(&c->x86_capability, 0, sizeof c->x86_capability);
605 c->extended_cpuid_level = 0;
608 identify_cpu_without_cpuid(c);
610 /* cyrix could have cpuid enabled via c_identify()*/
620 if (this_cpu->c_early_init)
621 this_cpu->c_early_init(c);
624 c->cpu_index = boot_cpu_id;
626 filter_cpuid_features(c, false);
629 void __init early_cpu_init(void)
631 struct cpu_dev **cdev;
634 printk("KERNEL supported cpus:\n");
635 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
636 struct cpu_dev *cpudev = *cdev;
639 if (count >= X86_VENDOR_NUM)
641 cpu_devs[count] = cpudev;
644 for (j = 0; j < 2; j++) {
645 if (!cpudev->c_ident[j])
647 printk(" %s %s\n", cpudev->c_vendor,
652 early_identify_cpu(&boot_cpu_data);
656 * The NOPL instruction is supposed to exist on all CPUs with
657 * family >= 6; unfortunately, that's not true in practice because
658 * of early VIA chips and (more importantly) broken virtualizers that
659 * are not easy to detect. In the latter case it doesn't even *fail*
660 * reliably, so probing for it doesn't even work. Disable it completely
661 * unless we can find a reliable way to detect all the broken cases.
663 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
665 clear_cpu_cap(c, X86_FEATURE_NOPL);
668 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
670 c->extended_cpuid_level = 0;
673 identify_cpu_without_cpuid(c);
675 /* cyrix could have cpuid enabled via c_identify()*/
685 if (c->cpuid_level >= 0x00000001) {
686 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
688 # ifdef CONFIG_X86_HT
689 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
691 c->apicid = c->initial_apicid;
696 c->phys_proc_id = c->initial_apicid;
700 get_model_name(c); /* Default name */
702 init_scattered_cpuid_features(c);
707 * This does the hard work of actually picking apart the CPU stuff...
709 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
713 c->loops_per_jiffy = loops_per_jiffy;
714 c->x86_cache_size = -1;
715 c->x86_vendor = X86_VENDOR_UNKNOWN;
716 c->x86_model = c->x86_mask = 0; /* So far unknown... */
717 c->x86_vendor_id[0] = '\0'; /* Unset */
718 c->x86_model_id[0] = '\0'; /* Unset */
719 c->x86_max_cores = 1;
720 c->x86_coreid_bits = 0;
722 c->x86_clflush_size = 64;
724 c->cpuid_level = -1; /* CPUID not detected */
725 c->x86_clflush_size = 32;
727 c->x86_cache_alignment = c->x86_clflush_size;
728 memset(&c->x86_capability, 0, sizeof c->x86_capability);
732 if (this_cpu->c_identify)
733 this_cpu->c_identify(c);
736 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
740 * Vendor-specific initialization. In this section we
741 * canonicalize the feature flags, meaning if there are
742 * features a certain CPU supports which CPUID doesn't
743 * tell us, CPUID claiming incorrect flags, or other bugs,
744 * we handle them here.
746 * At the end of this section, c->x86_capability better
747 * indicate the features this CPU genuinely supports!
749 if (this_cpu->c_init)
752 /* Disable the PN if appropriate */
753 squash_the_stupid_serial_number(c);
756 * The vendor-specific functions might have changed features. Now
757 * we do "generic changes."
760 /* Filter out anything that depends on CPUID levels we don't have */
761 filter_cpuid_features(c, true);
763 /* If the model name is still unset, do table lookup. */
764 if (!c->x86_model_id[0]) {
766 p = table_lookup_model(c);
768 strcpy(c->x86_model_id, p);
771 sprintf(c->x86_model_id, "%02x/%02x",
772 c->x86, c->x86_model);
781 * On SMP, boot_cpu_data holds the common feature set between
782 * all CPUs; so make sure that we indicate which features are
783 * common between the CPUs. The first time this routine gets
784 * executed, c == &boot_cpu_data.
786 if (c != &boot_cpu_data) {
787 /* AND the already accumulated flags with these */
788 for (i = 0; i < NCAPINTS; i++)
789 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
792 /* Clear all flags overriden by options */
793 for (i = 0; i < NCAPINTS; i++)
794 c->x86_capability[i] &= ~cleared_cpu_caps[i];
796 #ifdef CONFIG_X86_MCE
797 /* Init Machine Check Exception if available. */
801 select_idle_routine(c);
803 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
804 numa_add_cpu(smp_processor_id());
809 static void vgetcpu_set_mode(void)
811 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
812 vgetcpu_mode = VGETCPU_RDTSCP;
814 vgetcpu_mode = VGETCPU_LSL;
818 void __init identify_boot_cpu(void)
820 identify_cpu(&boot_cpu_data);
829 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
831 BUG_ON(c == &boot_cpu_data);
844 static struct msr_range msr_range_array[] __cpuinitdata = {
845 { 0x00000000, 0x00000418},
846 { 0xc0000000, 0xc000040b},
847 { 0xc0010000, 0xc0010142},
848 { 0xc0011000, 0xc001103b},
851 static void __cpuinit print_cpu_msr(void)
856 unsigned index_min, index_max;
858 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
859 index_min = msr_range_array[i].min;
860 index_max = msr_range_array[i].max;
861 for (index = index_min; index < index_max; index++) {
862 if (rdmsrl_amd_safe(index, &val))
864 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
869 static int show_msr __cpuinitdata;
870 static __init int setup_show_msr(char *arg)
874 get_option(&arg, &num);
880 __setup("show_msr=", setup_show_msr);
882 static __init int setup_noclflush(char *arg)
884 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
887 __setup("noclflush", setup_noclflush);
889 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
893 if (c->x86_vendor < X86_VENDOR_NUM)
894 vendor = this_cpu->c_vendor;
895 else if (c->cpuid_level >= 0)
896 vendor = c->x86_vendor_id;
898 if (vendor && !strstr(c->x86_model_id, vendor))
899 printk(KERN_CONT "%s ", vendor);
901 if (c->x86_model_id[0])
902 printk(KERN_CONT "%s", c->x86_model_id);
904 printk(KERN_CONT "%d86", c->x86);
906 if (c->x86_mask || c->cpuid_level >= 0)
907 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
909 printk(KERN_CONT "\n");
912 if (c->cpu_index < show_msr)
920 static __init int setup_disablecpuid(char *arg)
923 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
924 setup_clear_cpu_cap(bit);
929 __setup("clearcpuid=", setup_disablecpuid);
932 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
934 DEFINE_PER_CPU_FIRST(union irq_stack_union,
935 irq_stack_union) __aligned(PAGE_SIZE);
937 DEFINE_PER_CPU(char *, irq_stack_ptr); /* will be set during per cpu init */
939 DEFINE_PER_CPU(char *, irq_stack_ptr) =
940 per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
943 DEFINE_PER_CPU(unsigned long, kernel_stack) =
944 (unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
945 EXPORT_PER_CPU_SYMBOL(kernel_stack);
947 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
949 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
950 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ])
951 __aligned(PAGE_SIZE);
953 extern asmlinkage void ignore_sysret(void);
955 /* May not be marked __init: used by software suspend */
956 void syscall_init(void)
959 * LSTAR and STAR live in a bit strange symbiosis.
960 * They both write to the same internal register. STAR allows to
961 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
963 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
964 wrmsrl(MSR_LSTAR, system_call);
965 wrmsrl(MSR_CSTAR, ignore_sysret);
967 #ifdef CONFIG_IA32_EMULATION
968 syscall32_cpu_init();
971 /* Flags to clear on syscall */
972 wrmsrl(MSR_SYSCALL_MASK,
973 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
976 unsigned long kernel_eflags;
979 * Copies of the original ist values from the tss are only accessed during
980 * debugging, no special alignment required.
982 DEFINE_PER_CPU(struct orig_ist, orig_ist);
986 /* Make sure %fs is initialized properly in idle threads */
987 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
989 memset(regs, 0, sizeof(struct pt_regs));
990 regs->fs = __KERNEL_PERCPU;
996 * cpu_init() initializes state that is per-CPU. Some data is already
997 * initialized (naturally) in the bootstrap process, such as the GDT
998 * and IDT. We reload them nevertheless, this function acts as a
999 * 'CPU state barrier', nothing should get across.
1000 * A lot of state is already set up in PDA init for 64 bit
1002 #ifdef CONFIG_X86_64
1003 void __cpuinit cpu_init(void)
1005 int cpu = stack_smp_processor_id();
1006 struct tss_struct *t = &per_cpu(init_tss, cpu);
1007 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
1009 struct task_struct *me;
1013 if (cpu != 0 && percpu_read(node_number) == 0 &&
1014 cpu_to_node(cpu) != NUMA_NO_NODE)
1015 percpu_write(node_number, cpu_to_node(cpu));
1020 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1021 panic("CPU#%d already initialized!\n", cpu);
1023 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1025 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1028 * Initialize the per-CPU GDT with the boot GDT,
1029 * and set up the GDT descriptor:
1032 switch_to_new_gdt();
1035 load_idt((const struct desc_ptr *)&idt_descr);
1037 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1040 wrmsrl(MSR_FS_BASE, 0);
1041 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1045 if (cpu != 0 && x2apic)
1049 * set up and load the per-CPU TSS
1051 if (!orig_ist->ist[0]) {
1052 static const unsigned int sizes[N_EXCEPTION_STACKS] = {
1053 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1054 [DEBUG_STACK - 1] = DEBUG_STKSZ
1056 char *estacks = per_cpu(exception_stacks, cpu);
1057 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1058 estacks += sizes[v];
1059 orig_ist->ist[v] = t->x86_tss.ist[v] =
1060 (unsigned long)estacks;
1064 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1066 * <= is required because the CPU will access up to
1067 * 8 bits beyond the end of the IO permission bitmap.
1069 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1070 t->io_bitmap[i] = ~0UL;
1072 atomic_inc(&init_mm.mm_count);
1073 me->active_mm = &init_mm;
1076 enter_lazy_tlb(&init_mm, me);
1078 load_sp0(t, ¤t->thread);
1079 set_tss_desc(cpu, t);
1081 load_LDT(&init_mm.context);
1085 * If the kgdb is connected no debug regs should be altered. This
1086 * is only applicable when KGDB and a KGDB I/O module are built
1087 * into the kernel and you are using early debugging with
1088 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1090 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1091 arch_kgdb_ops.correct_hw_break();
1096 * Clear all 6 debug registers:
1098 set_debugreg(0UL, 0);
1099 set_debugreg(0UL, 1);
1100 set_debugreg(0UL, 2);
1101 set_debugreg(0UL, 3);
1102 set_debugreg(0UL, 6);
1103 set_debugreg(0UL, 7);
1108 raw_local_save_flags(kernel_eflags);
1116 void __cpuinit cpu_init(void)
1118 int cpu = smp_processor_id();
1119 struct task_struct *curr = current;
1120 struct tss_struct *t = &per_cpu(init_tss, cpu);
1121 struct thread_struct *thread = &curr->thread;
1123 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1124 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1125 for (;;) local_irq_enable();
1128 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1130 if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1131 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1133 load_idt(&idt_descr);
1134 switch_to_new_gdt();
1137 * Set up and load the per-CPU TSS and LDT
1139 atomic_inc(&init_mm.mm_count);
1140 curr->active_mm = &init_mm;
1143 enter_lazy_tlb(&init_mm, curr);
1145 load_sp0(t, thread);
1146 set_tss_desc(cpu, t);
1148 load_LDT(&init_mm.context);
1150 #ifdef CONFIG_DOUBLEFAULT
1151 /* Set up doublefault TSS pointer in the GDT */
1152 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1156 asm volatile ("mov %0, %%gs" : : "r" (0));
1158 /* Clear all 6 debug registers: */
1167 * Force FPU initialization:
1170 current_thread_info()->status = TS_XSAVE;
1172 current_thread_info()->status = 0;
1174 mxcsr_feature_mask_init();
1177 * Boot processor to setup the FP and extended state context info.
1179 if (smp_processor_id() == boot_cpu_id)
1180 init_thread_xstate();