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>
42 #include <asm/stackprotector.h>
48 /* all of these masks are initialized in setup_cpu_local_masks() */
49 cpumask_var_t cpu_callin_mask;
50 cpumask_var_t cpu_callout_mask;
51 cpumask_var_t cpu_initialized_mask;
53 /* representing cpus for which sibling maps can be computed */
54 cpumask_var_t cpu_sibling_setup_mask;
56 /* correctly size the local cpu masks */
57 void __init setup_cpu_local_masks(void)
59 alloc_bootmem_cpumask_var(&cpu_initialized_mask);
60 alloc_bootmem_cpumask_var(&cpu_callin_mask);
61 alloc_bootmem_cpumask_var(&cpu_callout_mask);
62 alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
65 #else /* CONFIG_X86_32 */
67 cpumask_t cpu_callin_map;
68 cpumask_t cpu_callout_map;
69 cpumask_t cpu_initialized;
70 cpumask_t cpu_sibling_setup_map;
72 #endif /* CONFIG_X86_32 */
75 static struct cpu_dev *this_cpu __cpuinitdata;
77 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
80 * We need valid kernel segments for data and code in long mode too
81 * IRET will check the segment types kkeil 2000/10/28
82 * Also sysret mandates a special GDT layout
84 * The TLS descriptors are currently at a different place compared to i386.
85 * Hopefully nobody expects them at a fixed place (Wine?)
87 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
88 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
89 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
90 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
91 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
92 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
94 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } },
95 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } },
96 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } },
97 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff200 } } },
99 * Segments used for calling PnP BIOS have byte granularity.
100 * They code segments and data segments have fixed 64k limits,
101 * the transfer segment sizes are set at run time.
104 [GDT_ENTRY_PNPBIOS_CS32] = { { { 0x0000ffff, 0x00409a00 } } },
106 [GDT_ENTRY_PNPBIOS_CS16] = { { { 0x0000ffff, 0x00009a00 } } },
108 [GDT_ENTRY_PNPBIOS_DS] = { { { 0x0000ffff, 0x00009200 } } },
110 [GDT_ENTRY_PNPBIOS_TS1] = { { { 0x00000000, 0x00009200 } } },
112 [GDT_ENTRY_PNPBIOS_TS2] = { { { 0x00000000, 0x00009200 } } },
114 * The APM segments have byte granularity and their bases
115 * are set at run time. All have 64k limits.
118 [GDT_ENTRY_APMBIOS_BASE] = { { { 0x0000ffff, 0x00409a00 } } },
120 [GDT_ENTRY_APMBIOS_BASE+1] = { { { 0x0000ffff, 0x00009a00 } } },
122 [GDT_ENTRY_APMBIOS_BASE+2] = { { { 0x0000ffff, 0x00409200 } } },
124 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } },
125 [GDT_ENTRY_PERCPU] = { { { 0x0000ffff, 0x00cf9200 } } },
126 GDT_STACK_CANARY_INIT
129 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
132 static int cachesize_override __cpuinitdata = -1;
133 static int disable_x86_serial_nr __cpuinitdata = 1;
135 static int __init cachesize_setup(char *str)
137 get_option(&str, &cachesize_override);
140 __setup("cachesize=", cachesize_setup);
142 static int __init x86_fxsr_setup(char *s)
144 setup_clear_cpu_cap(X86_FEATURE_FXSR);
145 setup_clear_cpu_cap(X86_FEATURE_XMM);
148 __setup("nofxsr", x86_fxsr_setup);
150 static int __init x86_sep_setup(char *s)
152 setup_clear_cpu_cap(X86_FEATURE_SEP);
155 __setup("nosep", x86_sep_setup);
157 /* Standard macro to see if a specific flag is changeable */
158 static inline int flag_is_changeable_p(u32 flag)
163 * Cyrix and IDT cpus allow disabling of CPUID
164 * so the code below may return different results
165 * when it is executed before and after enabling
166 * the CPUID. Add "volatile" to not allow gcc to
167 * optimize the subsequent calls to this function.
169 asm volatile ("pushfl\n\t"
179 : "=&r" (f1), "=&r" (f2)
182 return ((f1^f2) & flag) != 0;
185 /* Probe for the CPUID instruction */
186 static int __cpuinit have_cpuid_p(void)
188 return flag_is_changeable_p(X86_EFLAGS_ID);
191 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
193 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
194 /* Disable processor serial number */
195 unsigned long lo, hi;
196 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
198 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
199 printk(KERN_NOTICE "CPU serial number disabled.\n");
200 clear_cpu_cap(c, X86_FEATURE_PN);
202 /* Disabling the serial number may affect the cpuid level */
203 c->cpuid_level = cpuid_eax(0);
207 static int __init x86_serial_nr_setup(char *s)
209 disable_x86_serial_nr = 0;
212 __setup("serialnumber", x86_serial_nr_setup);
214 static inline int flag_is_changeable_p(u32 flag)
218 /* Probe for the CPUID instruction */
219 static inline int have_cpuid_p(void)
223 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
229 * Some CPU features depend on higher CPUID levels, which may not always
230 * be available due to CPUID level capping or broken virtualization
231 * software. Add those features to this table to auto-disable them.
233 struct cpuid_dependent_feature {
237 static const struct cpuid_dependent_feature __cpuinitconst
238 cpuid_dependent_features[] = {
239 { X86_FEATURE_MWAIT, 0x00000005 },
240 { X86_FEATURE_DCA, 0x00000009 },
241 { X86_FEATURE_XSAVE, 0x0000000d },
245 static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
247 const struct cpuid_dependent_feature *df;
248 for (df = cpuid_dependent_features; df->feature; df++) {
250 * Note: cpuid_level is set to -1 if unavailable, but
251 * extended_extended_level is set to 0 if unavailable
252 * and the legitimate extended levels are all negative
253 * when signed; hence the weird messing around with
256 if (cpu_has(c, df->feature) &&
257 ((s32)df->feature < 0 ?
258 (u32)df->feature > (u32)c->extended_cpuid_level :
259 (s32)df->feature > (s32)c->cpuid_level)) {
260 clear_cpu_cap(c, df->feature);
263 "CPU: CPU feature %s disabled "
264 "due to lack of CPUID level 0x%x\n",
265 x86_cap_flags[df->feature],
272 * Naming convention should be: <Name> [(<Codename>)]
273 * This table only is used unless init_<vendor>() below doesn't set it;
274 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
278 /* Look up CPU names by table lookup. */
279 static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
281 struct cpu_model_info *info;
283 if (c->x86_model >= 16)
284 return NULL; /* Range check */
289 info = this_cpu->c_models;
291 while (info && info->family) {
292 if (info->family == c->x86)
293 return info->model_names[c->x86_model];
296 return NULL; /* Not found */
299 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
301 void load_percpu_segment(int cpu)
304 loadsegment(fs, __KERNEL_PERCPU);
307 wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
309 load_stack_canary_segment();
312 /* Current gdt points %fs at the "master" per-cpu area: after this,
313 * it's on the real one. */
314 void switch_to_new_gdt(int cpu)
316 struct desc_ptr gdt_descr;
318 gdt_descr.address = (long)get_cpu_gdt_table(cpu);
319 gdt_descr.size = GDT_SIZE - 1;
320 load_gdt(&gdt_descr);
321 /* Reload the per-cpu base */
323 load_percpu_segment(cpu);
326 static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
328 static void __cpuinit default_init(struct cpuinfo_x86 *c)
331 display_cacheinfo(c);
333 /* Not much we can do here... */
334 /* Check if at least it has cpuid */
335 if (c->cpuid_level == -1) {
336 /* No cpuid. It must be an ancient CPU */
338 strcpy(c->x86_model_id, "486");
339 else if (c->x86 == 3)
340 strcpy(c->x86_model_id, "386");
345 static struct cpu_dev __cpuinitdata default_cpu = {
346 .c_init = default_init,
347 .c_vendor = "Unknown",
348 .c_x86_vendor = X86_VENDOR_UNKNOWN,
351 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
356 if (c->extended_cpuid_level < 0x80000004)
359 v = (unsigned int *) c->x86_model_id;
360 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
361 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
362 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
363 c->x86_model_id[48] = 0;
365 /* Intel chips right-justify this string for some dumb reason;
366 undo that brain damage */
367 p = q = &c->x86_model_id[0];
373 while (q <= &c->x86_model_id[48])
374 *q++ = '\0'; /* Zero-pad the rest */
378 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
380 unsigned int n, dummy, ebx, ecx, edx, l2size;
382 n = c->extended_cpuid_level;
384 if (n >= 0x80000005) {
385 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
386 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
387 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
388 c->x86_cache_size = (ecx>>24) + (edx>>24);
390 /* On K8 L1 TLB is inclusive, so don't count it */
395 if (n < 0x80000006) /* Some chips just has a large L1. */
398 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
402 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
404 /* do processor-specific cache resizing */
405 if (this_cpu->c_size_cache)
406 l2size = this_cpu->c_size_cache(c, l2size);
408 /* Allow user to override all this if necessary. */
409 if (cachesize_override != -1)
410 l2size = cachesize_override;
413 return; /* Again, no L2 cache is possible */
416 c->x86_cache_size = l2size;
418 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
422 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
425 u32 eax, ebx, ecx, edx;
426 int index_msb, core_bits;
428 if (!cpu_has(c, X86_FEATURE_HT))
431 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
434 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
437 cpuid(1, &eax, &ebx, &ecx, &edx);
439 smp_num_siblings = (ebx & 0xff0000) >> 16;
441 if (smp_num_siblings == 1) {
442 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
443 } else if (smp_num_siblings > 1) {
445 if (smp_num_siblings > nr_cpu_ids) {
446 printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
448 smp_num_siblings = 1;
452 index_msb = get_count_order(smp_num_siblings);
453 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
455 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
457 index_msb = get_count_order(smp_num_siblings);
459 core_bits = get_count_order(c->x86_max_cores);
461 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
462 ((1 << core_bits) - 1);
466 if ((c->x86_max_cores * smp_num_siblings) > 1) {
467 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
469 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
475 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
477 char *v = c->x86_vendor_id;
481 for (i = 0; i < X86_VENDOR_NUM; i++) {
485 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
486 (cpu_devs[i]->c_ident[1] &&
487 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
488 this_cpu = cpu_devs[i];
489 c->x86_vendor = this_cpu->c_x86_vendor;
496 printk(KERN_ERR "CPU: vendor_id '%s' unknown, using generic init.\n", v);
497 printk(KERN_ERR "CPU: Your system may be unstable.\n");
500 c->x86_vendor = X86_VENDOR_UNKNOWN;
501 this_cpu = &default_cpu;
504 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
506 /* Get vendor name */
507 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
508 (unsigned int *)&c->x86_vendor_id[0],
509 (unsigned int *)&c->x86_vendor_id[8],
510 (unsigned int *)&c->x86_vendor_id[4]);
513 /* Intel-defined flags: level 0x00000001 */
514 if (c->cpuid_level >= 0x00000001) {
515 u32 junk, tfms, cap0, misc;
516 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
517 c->x86 = (tfms >> 8) & 0xf;
518 c->x86_model = (tfms >> 4) & 0xf;
519 c->x86_mask = tfms & 0xf;
521 c->x86 += (tfms >> 20) & 0xff;
523 c->x86_model += ((tfms >> 16) & 0xf) << 4;
524 if (cap0 & (1<<19)) {
525 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
526 c->x86_cache_alignment = c->x86_clflush_size;
531 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
536 /* Intel-defined flags: level 0x00000001 */
537 if (c->cpuid_level >= 0x00000001) {
538 u32 capability, excap;
539 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
540 c->x86_capability[0] = capability;
541 c->x86_capability[4] = excap;
544 /* AMD-defined flags: level 0x80000001 */
545 xlvl = cpuid_eax(0x80000000);
546 c->extended_cpuid_level = xlvl;
547 if ((xlvl & 0xffff0000) == 0x80000000) {
548 if (xlvl >= 0x80000001) {
549 c->x86_capability[1] = cpuid_edx(0x80000001);
550 c->x86_capability[6] = cpuid_ecx(0x80000001);
555 if (c->extended_cpuid_level >= 0x80000008) {
556 u32 eax = cpuid_eax(0x80000008);
558 c->x86_virt_bits = (eax >> 8) & 0xff;
559 c->x86_phys_bits = eax & 0xff;
563 if (c->extended_cpuid_level >= 0x80000007)
564 c->x86_power = cpuid_edx(0x80000007);
568 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
574 * First of all, decide if this is a 486 or higher
575 * It's a 486 if we can modify the AC flag
577 if (flag_is_changeable_p(X86_EFLAGS_AC))
582 for (i = 0; i < X86_VENDOR_NUM; i++)
583 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
584 c->x86_vendor_id[0] = 0;
585 cpu_devs[i]->c_identify(c);
586 if (c->x86_vendor_id[0]) {
595 * Do minimum CPU detection early.
596 * Fields really needed: vendor, cpuid_level, family, model, mask,
598 * The others are not touched to avoid unwanted side effects.
600 * WARNING: this function is only called on the BP. Don't add code here
601 * that is supposed to run on all CPUs.
603 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
606 c->x86_clflush_size = 64;
608 c->x86_clflush_size = 32;
610 c->x86_cache_alignment = c->x86_clflush_size;
612 memset(&c->x86_capability, 0, sizeof c->x86_capability);
613 c->extended_cpuid_level = 0;
616 identify_cpu_without_cpuid(c);
618 /* cyrix could have cpuid enabled via c_identify()*/
628 if (this_cpu->c_early_init)
629 this_cpu->c_early_init(c);
632 c->cpu_index = boot_cpu_id;
634 filter_cpuid_features(c, false);
637 void __init early_cpu_init(void)
639 struct cpu_dev **cdev;
642 printk("KERNEL supported cpus:\n");
643 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
644 struct cpu_dev *cpudev = *cdev;
647 if (count >= X86_VENDOR_NUM)
649 cpu_devs[count] = cpudev;
652 for (j = 0; j < 2; j++) {
653 if (!cpudev->c_ident[j])
655 printk(" %s %s\n", cpudev->c_vendor,
660 early_identify_cpu(&boot_cpu_data);
664 * The NOPL instruction is supposed to exist on all CPUs with
665 * family >= 6; unfortunately, that's not true in practice because
666 * of early VIA chips and (more importantly) broken virtualizers that
667 * are not easy to detect. In the latter case it doesn't even *fail*
668 * reliably, so probing for it doesn't even work. Disable it completely
669 * unless we can find a reliable way to detect all the broken cases.
671 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
673 clear_cpu_cap(c, X86_FEATURE_NOPL);
676 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
678 c->extended_cpuid_level = 0;
681 identify_cpu_without_cpuid(c);
683 /* cyrix could have cpuid enabled via c_identify()*/
693 if (c->cpuid_level >= 0x00000001) {
694 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
696 # ifdef CONFIG_X86_HT
697 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
699 c->apicid = c->initial_apicid;
704 c->phys_proc_id = c->initial_apicid;
708 get_model_name(c); /* Default name */
710 init_scattered_cpuid_features(c);
715 * This does the hard work of actually picking apart the CPU stuff...
717 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
721 c->loops_per_jiffy = loops_per_jiffy;
722 c->x86_cache_size = -1;
723 c->x86_vendor = X86_VENDOR_UNKNOWN;
724 c->x86_model = c->x86_mask = 0; /* So far unknown... */
725 c->x86_vendor_id[0] = '\0'; /* Unset */
726 c->x86_model_id[0] = '\0'; /* Unset */
727 c->x86_max_cores = 1;
728 c->x86_coreid_bits = 0;
730 c->x86_clflush_size = 64;
732 c->cpuid_level = -1; /* CPUID not detected */
733 c->x86_clflush_size = 32;
735 c->x86_cache_alignment = c->x86_clflush_size;
736 memset(&c->x86_capability, 0, sizeof c->x86_capability);
740 if (this_cpu->c_identify)
741 this_cpu->c_identify(c);
744 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
748 * Vendor-specific initialization. In this section we
749 * canonicalize the feature flags, meaning if there are
750 * features a certain CPU supports which CPUID doesn't
751 * tell us, CPUID claiming incorrect flags, or other bugs,
752 * we handle them here.
754 * At the end of this section, c->x86_capability better
755 * indicate the features this CPU genuinely supports!
757 if (this_cpu->c_init)
760 /* Disable the PN if appropriate */
761 squash_the_stupid_serial_number(c);
764 * The vendor-specific functions might have changed features. Now
765 * we do "generic changes."
768 /* Filter out anything that depends on CPUID levels we don't have */
769 filter_cpuid_features(c, true);
771 /* If the model name is still unset, do table lookup. */
772 if (!c->x86_model_id[0]) {
774 p = table_lookup_model(c);
776 strcpy(c->x86_model_id, p);
779 sprintf(c->x86_model_id, "%02x/%02x",
780 c->x86, c->x86_model);
789 * On SMP, boot_cpu_data holds the common feature set between
790 * all CPUs; so make sure that we indicate which features are
791 * common between the CPUs. The first time this routine gets
792 * executed, c == &boot_cpu_data.
794 if (c != &boot_cpu_data) {
795 /* AND the already accumulated flags with these */
796 for (i = 0; i < NCAPINTS; i++)
797 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
800 /* Clear all flags overriden by options */
801 for (i = 0; i < NCAPINTS; i++)
802 c->x86_capability[i] &= ~cleared_cpu_caps[i];
804 #ifdef CONFIG_X86_MCE
805 /* Init Machine Check Exception if available. */
809 select_idle_routine(c);
811 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
812 numa_add_cpu(smp_processor_id());
817 static void vgetcpu_set_mode(void)
819 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
820 vgetcpu_mode = VGETCPU_RDTSCP;
822 vgetcpu_mode = VGETCPU_LSL;
826 void __init identify_boot_cpu(void)
828 identify_cpu(&boot_cpu_data);
837 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
839 BUG_ON(c == &boot_cpu_data);
852 static struct msr_range msr_range_array[] __cpuinitdata = {
853 { 0x00000000, 0x00000418},
854 { 0xc0000000, 0xc000040b},
855 { 0xc0010000, 0xc0010142},
856 { 0xc0011000, 0xc001103b},
859 static void __cpuinit print_cpu_msr(void)
864 unsigned index_min, index_max;
866 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
867 index_min = msr_range_array[i].min;
868 index_max = msr_range_array[i].max;
869 for (index = index_min; index < index_max; index++) {
870 if (rdmsrl_amd_safe(index, &val))
872 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
877 static int show_msr __cpuinitdata;
878 static __init int setup_show_msr(char *arg)
882 get_option(&arg, &num);
888 __setup("show_msr=", setup_show_msr);
890 static __init int setup_noclflush(char *arg)
892 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
895 __setup("noclflush", setup_noclflush);
897 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
901 if (c->x86_vendor < X86_VENDOR_NUM)
902 vendor = this_cpu->c_vendor;
903 else if (c->cpuid_level >= 0)
904 vendor = c->x86_vendor_id;
906 if (vendor && !strstr(c->x86_model_id, vendor))
907 printk(KERN_CONT "%s ", vendor);
909 if (c->x86_model_id[0])
910 printk(KERN_CONT "%s", c->x86_model_id);
912 printk(KERN_CONT "%d86", c->x86);
914 if (c->x86_mask || c->cpuid_level >= 0)
915 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
917 printk(KERN_CONT "\n");
920 if (c->cpu_index < show_msr)
928 static __init int setup_disablecpuid(char *arg)
931 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
932 setup_clear_cpu_cap(bit);
937 __setup("clearcpuid=", setup_disablecpuid);
940 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
942 DEFINE_PER_CPU_FIRST(union irq_stack_union,
943 irq_stack_union) __aligned(PAGE_SIZE);
944 DEFINE_PER_CPU(char *, irq_stack_ptr) =
945 init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
947 DEFINE_PER_CPU(unsigned long, kernel_stack) =
948 (unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
949 EXPORT_PER_CPU_SYMBOL(kernel_stack);
951 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
953 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
954 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ])
955 __aligned(PAGE_SIZE);
957 extern asmlinkage void ignore_sysret(void);
959 /* May not be marked __init: used by software suspend */
960 void syscall_init(void)
963 * LSTAR and STAR live in a bit strange symbiosis.
964 * They both write to the same internal register. STAR allows to
965 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
967 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
968 wrmsrl(MSR_LSTAR, system_call);
969 wrmsrl(MSR_CSTAR, ignore_sysret);
971 #ifdef CONFIG_IA32_EMULATION
972 syscall32_cpu_init();
975 /* Flags to clear on syscall */
976 wrmsrl(MSR_SYSCALL_MASK,
977 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
980 unsigned long kernel_eflags;
983 * Copies of the original ist values from the tss are only accessed during
984 * debugging, no special alignment required.
986 DEFINE_PER_CPU(struct orig_ist, orig_ist);
990 #ifdef CONFIG_CC_STACKPROTECTOR
991 DEFINE_PER_CPU(unsigned long, stack_canary);
994 /* Make sure %fs and %gs are initialized properly in idle threads */
995 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
997 memset(regs, 0, sizeof(struct pt_regs));
998 regs->fs = __KERNEL_PERCPU;
999 regs->gs = __KERNEL_STACK_CANARY;
1005 * cpu_init() initializes state that is per-CPU. Some data is already
1006 * initialized (naturally) in the bootstrap process, such as the GDT
1007 * and IDT. We reload them nevertheless, this function acts as a
1008 * 'CPU state barrier', nothing should get across.
1009 * A lot of state is already set up in PDA init for 64 bit
1011 #ifdef CONFIG_X86_64
1012 void __cpuinit cpu_init(void)
1014 int cpu = stack_smp_processor_id();
1015 struct tss_struct *t = &per_cpu(init_tss, cpu);
1016 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
1018 struct task_struct *me;
1022 if (cpu != 0 && percpu_read(node_number) == 0 &&
1023 cpu_to_node(cpu) != NUMA_NO_NODE)
1024 percpu_write(node_number, cpu_to_node(cpu));
1029 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1030 panic("CPU#%d already initialized!\n", cpu);
1032 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1034 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1037 * Initialize the per-CPU GDT with the boot GDT,
1038 * and set up the GDT descriptor:
1041 switch_to_new_gdt(cpu);
1044 load_idt((const struct desc_ptr *)&idt_descr);
1046 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1049 wrmsrl(MSR_FS_BASE, 0);
1050 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1054 if (cpu != 0 && x2apic)
1058 * set up and load the per-CPU TSS
1060 if (!orig_ist->ist[0]) {
1061 static const unsigned int sizes[N_EXCEPTION_STACKS] = {
1062 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1063 [DEBUG_STACK - 1] = DEBUG_STKSZ
1065 char *estacks = per_cpu(exception_stacks, cpu);
1066 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1067 estacks += sizes[v];
1068 orig_ist->ist[v] = t->x86_tss.ist[v] =
1069 (unsigned long)estacks;
1073 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1075 * <= is required because the CPU will access up to
1076 * 8 bits beyond the end of the IO permission bitmap.
1078 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1079 t->io_bitmap[i] = ~0UL;
1081 atomic_inc(&init_mm.mm_count);
1082 me->active_mm = &init_mm;
1085 enter_lazy_tlb(&init_mm, me);
1087 load_sp0(t, ¤t->thread);
1088 set_tss_desc(cpu, t);
1090 load_LDT(&init_mm.context);
1094 * If the kgdb is connected no debug regs should be altered. This
1095 * is only applicable when KGDB and a KGDB I/O module are built
1096 * into the kernel and you are using early debugging with
1097 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1099 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1100 arch_kgdb_ops.correct_hw_break();
1105 * Clear all 6 debug registers:
1107 set_debugreg(0UL, 0);
1108 set_debugreg(0UL, 1);
1109 set_debugreg(0UL, 2);
1110 set_debugreg(0UL, 3);
1111 set_debugreg(0UL, 6);
1112 set_debugreg(0UL, 7);
1117 raw_local_save_flags(kernel_eflags);
1125 void __cpuinit cpu_init(void)
1127 int cpu = smp_processor_id();
1128 struct task_struct *curr = current;
1129 struct tss_struct *t = &per_cpu(init_tss, cpu);
1130 struct thread_struct *thread = &curr->thread;
1132 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1133 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1134 for (;;) local_irq_enable();
1137 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1139 if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1140 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1142 load_idt(&idt_descr);
1143 switch_to_new_gdt(cpu);
1146 * Set up and load the per-CPU TSS and LDT
1148 atomic_inc(&init_mm.mm_count);
1149 curr->active_mm = &init_mm;
1152 enter_lazy_tlb(&init_mm, curr);
1154 load_sp0(t, thread);
1155 set_tss_desc(cpu, t);
1157 load_LDT(&init_mm.context);
1159 #ifdef CONFIG_DOUBLEFAULT
1160 /* Set up doublefault TSS pointer in the GDT */
1161 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1164 /* Clear all 6 debug registers: */
1173 * Force FPU initialization:
1176 current_thread_info()->status = TS_XSAVE;
1178 current_thread_info()->status = 0;
1180 mxcsr_feature_mask_init();
1183 * Boot processor to setup the FP and extended state context info.
1185 if (smp_processor_id() == boot_cpu_id)
1186 init_thread_xstate();