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/mmu_context.h>
17 #include <asm/hypervisor.h>
18 #include <asm/processor.h>
19 #include <asm/sections.h>
20 #include <asm/topology.h>
21 #include <asm/cpumask.h>
22 #include <asm/pgtable.h>
23 #include <asm/atomic.h>
24 #include <asm/proto.h>
25 #include <asm/setup.h>
38 #ifdef CONFIG_X86_LOCAL_APIC
39 #include <asm/uv/uv.h>
46 /* all of these masks are initialized in setup_cpu_local_masks() */
47 cpumask_var_t cpu_initialized_mask;
48 cpumask_var_t cpu_callout_mask;
49 cpumask_var_t cpu_callin_mask;
51 /* representing cpus for which sibling maps can be computed */
52 cpumask_var_t cpu_sibling_setup_mask;
54 /* correctly size the local cpu masks */
55 void __init setup_cpu_local_masks(void)
57 alloc_bootmem_cpumask_var(&cpu_initialized_mask);
58 alloc_bootmem_cpumask_var(&cpu_callin_mask);
59 alloc_bootmem_cpumask_var(&cpu_callout_mask);
60 alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
63 #else /* CONFIG_X86_32 */
65 cpumask_t cpu_sibling_setup_map;
66 cpumask_t cpu_callout_map;
67 cpumask_t cpu_initialized;
68 cpumask_t cpu_callin_map;
70 #endif /* CONFIG_X86_32 */
73 static const struct cpu_dev *this_cpu __cpuinitdata;
75 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
78 * We need valid kernel segments for data and code in long mode too
79 * IRET will check the segment types kkeil 2000/10/28
80 * Also sysret mandates a special GDT layout
82 * TLS descriptors are currently at a different place compared to i386.
83 * Hopefully nobody expects them at a fixed place (Wine?)
85 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
86 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
87 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
88 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
89 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
90 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
92 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } },
93 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } },
94 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } },
95 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff200 } } },
97 * Segments used for calling PnP BIOS have byte granularity.
98 * They code segments and data segments have fixed 64k limits,
99 * the transfer segment sizes are set at run time.
102 [GDT_ENTRY_PNPBIOS_CS32] = { { { 0x0000ffff, 0x00409a00 } } },
104 [GDT_ENTRY_PNPBIOS_CS16] = { { { 0x0000ffff, 0x00009a00 } } },
106 [GDT_ENTRY_PNPBIOS_DS] = { { { 0x0000ffff, 0x00009200 } } },
108 [GDT_ENTRY_PNPBIOS_TS1] = { { { 0x00000000, 0x00009200 } } },
110 [GDT_ENTRY_PNPBIOS_TS2] = { { { 0x00000000, 0x00009200 } } },
112 * The APM segments have byte granularity and their bases
113 * are set at run time. All have 64k limits.
116 [GDT_ENTRY_APMBIOS_BASE] = { { { 0x0000ffff, 0x00409a00 } } },
118 [GDT_ENTRY_APMBIOS_BASE+1] = { { { 0x0000ffff, 0x00009a00 } } },
120 [GDT_ENTRY_APMBIOS_BASE+2] = { { { 0x0000ffff, 0x00409200 } } },
122 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } },
123 [GDT_ENTRY_PERCPU] = { { { 0x0000ffff, 0x00cf9200 } } },
124 GDT_STACK_CANARY_INIT
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"
178 : "=&r" (f1), "=&r" (f2)
181 return ((f1^f2) & flag) != 0;
184 /* Probe for the CPUID instruction */
185 static int __cpuinit have_cpuid_p(void)
187 return flag_is_changeable_p(X86_EFLAGS_ID);
190 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
192 unsigned long lo, hi;
194 if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
197 /* Disable processor serial number: */
199 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
201 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
203 printk(KERN_NOTICE "CPU serial number disabled.\n");
204 clear_cpu_cap(c, X86_FEATURE_PN);
206 /* Disabling the serial number may affect the cpuid level */
207 c->cpuid_level = cpuid_eax(0);
210 static int __init x86_serial_nr_setup(char *s)
212 disable_x86_serial_nr = 0;
215 __setup("serialnumber", x86_serial_nr_setup);
217 static inline int flag_is_changeable_p(u32 flag)
221 /* Probe for the CPUID instruction */
222 static inline int have_cpuid_p(void)
226 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
232 * Some CPU features depend on higher CPUID levels, which may not always
233 * be available due to CPUID level capping or broken virtualization
234 * software. Add those features to this table to auto-disable them.
236 struct cpuid_dependent_feature {
241 static const struct cpuid_dependent_feature __cpuinitconst
242 cpuid_dependent_features[] = {
243 { X86_FEATURE_MWAIT, 0x00000005 },
244 { X86_FEATURE_DCA, 0x00000009 },
245 { X86_FEATURE_XSAVE, 0x0000000d },
249 static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
251 const struct cpuid_dependent_feature *df;
253 for (df = cpuid_dependent_features; df->feature; df++) {
255 if (!cpu_has(c, df->feature))
258 * Note: cpuid_level is set to -1 if unavailable, but
259 * extended_extended_level is set to 0 if unavailable
260 * and the legitimate extended levels are all negative
261 * when signed; hence the weird messing around with
264 if (!((s32)df->level < 0 ?
265 (u32)df->level > (u32)c->extended_cpuid_level :
266 (s32)df->level > (s32)c->cpuid_level))
269 clear_cpu_cap(c, df->feature);
274 "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
275 x86_cap_flags[df->feature], df->level);
280 * Naming convention should be: <Name> [(<Codename>)]
281 * This table only is used unless init_<vendor>() below doesn't set it;
282 * in particular, if CPUID levels 0x80000002..4 are supported, this
286 /* Look up CPU names by table lookup. */
287 static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
289 const struct cpu_model_info *info;
291 if (c->x86_model >= 16)
292 return NULL; /* Range check */
297 info = this_cpu->c_models;
299 while (info && info->family) {
300 if (info->family == c->x86)
301 return info->model_names[c->x86_model];
304 return NULL; /* Not found */
307 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
309 void load_percpu_segment(int cpu)
312 loadsegment(fs, __KERNEL_PERCPU);
315 wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
317 load_stack_canary_segment();
321 * Current gdt points %fs at the "master" per-cpu area: after this,
322 * it's on the real one.
324 void switch_to_new_gdt(int cpu)
326 struct desc_ptr gdt_descr;
328 gdt_descr.address = (long)get_cpu_gdt_table(cpu);
329 gdt_descr.size = GDT_SIZE - 1;
330 load_gdt(&gdt_descr);
331 /* Reload the per-cpu base */
333 load_percpu_segment(cpu);
336 static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
338 static void __cpuinit default_init(struct cpuinfo_x86 *c)
341 display_cacheinfo(c);
343 /* Not much we can do here... */
344 /* Check if at least it has cpuid */
345 if (c->cpuid_level == -1) {
346 /* No cpuid. It must be an ancient CPU */
348 strcpy(c->x86_model_id, "486");
349 else if (c->x86 == 3)
350 strcpy(c->x86_model_id, "386");
355 static const struct cpu_dev __cpuinitconst default_cpu = {
356 .c_init = default_init,
357 .c_vendor = "Unknown",
358 .c_x86_vendor = X86_VENDOR_UNKNOWN,
361 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
366 if (c->extended_cpuid_level < 0x80000004)
369 v = (unsigned int *)c->x86_model_id;
370 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
371 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
372 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
373 c->x86_model_id[48] = 0;
376 * Intel chips right-justify this string for some dumb reason;
377 * undo that brain damage:
379 p = q = &c->x86_model_id[0];
385 while (q <= &c->x86_model_id[48])
386 *q++ = '\0'; /* Zero-pad the rest */
390 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
392 unsigned int n, dummy, ebx, ecx, edx, l2size;
394 n = c->extended_cpuid_level;
396 if (n >= 0x80000005) {
397 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
398 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
399 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
400 c->x86_cache_size = (ecx>>24) + (edx>>24);
402 /* On K8 L1 TLB is inclusive, so don't count it */
407 if (n < 0x80000006) /* Some chips just has a large L1. */
410 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
414 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
416 /* do processor-specific cache resizing */
417 if (this_cpu->c_size_cache)
418 l2size = this_cpu->c_size_cache(c, l2size);
420 /* Allow user to override all this if necessary. */
421 if (cachesize_override != -1)
422 l2size = cachesize_override;
425 return; /* Again, no L2 cache is possible */
428 c->x86_cache_size = l2size;
430 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
434 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
437 u32 eax, ebx, ecx, edx;
438 int index_msb, core_bits;
440 if (!cpu_has(c, X86_FEATURE_HT))
443 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
446 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
449 cpuid(1, &eax, &ebx, &ecx, &edx);
451 smp_num_siblings = (ebx & 0xff0000) >> 16;
453 if (smp_num_siblings == 1) {
454 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
458 if (smp_num_siblings <= 1)
461 if (smp_num_siblings > nr_cpu_ids) {
462 pr_warning("CPU: Unsupported number of siblings %d",
464 smp_num_siblings = 1;
468 index_msb = get_count_order(smp_num_siblings);
469 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
471 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
473 index_msb = get_count_order(smp_num_siblings);
475 core_bits = get_count_order(c->x86_max_cores);
477 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
478 ((1 << core_bits) - 1);
481 if ((c->x86_max_cores * smp_num_siblings) > 1) {
482 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
484 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
490 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
492 char *v = c->x86_vendor_id;
496 for (i = 0; i < X86_VENDOR_NUM; i++) {
500 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
501 (cpu_devs[i]->c_ident[1] &&
502 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
504 this_cpu = cpu_devs[i];
505 c->x86_vendor = this_cpu->c_x86_vendor;
513 "CPU: vendor_id '%s' unknown, using generic init.\n", v);
515 printk(KERN_ERR "CPU: Your system may be unstable.\n");
518 c->x86_vendor = X86_VENDOR_UNKNOWN;
519 this_cpu = &default_cpu;
522 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
524 /* Get vendor name */
525 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
526 (unsigned int *)&c->x86_vendor_id[0],
527 (unsigned int *)&c->x86_vendor_id[8],
528 (unsigned int *)&c->x86_vendor_id[4]);
531 /* Intel-defined flags: level 0x00000001 */
532 if (c->cpuid_level >= 0x00000001) {
533 u32 junk, tfms, cap0, misc;
535 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
536 c->x86 = (tfms >> 8) & 0xf;
537 c->x86_model = (tfms >> 4) & 0xf;
538 c->x86_mask = tfms & 0xf;
541 c->x86 += (tfms >> 20) & 0xff;
543 c->x86_model += ((tfms >> 16) & 0xf) << 4;
545 if (cap0 & (1<<19)) {
546 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
547 c->x86_cache_alignment = c->x86_clflush_size;
552 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
557 /* Intel-defined flags: level 0x00000001 */
558 if (c->cpuid_level >= 0x00000001) {
559 u32 capability, excap;
561 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
562 c->x86_capability[0] = capability;
563 c->x86_capability[4] = excap;
566 /* AMD-defined flags: level 0x80000001 */
567 xlvl = cpuid_eax(0x80000000);
568 c->extended_cpuid_level = xlvl;
570 if ((xlvl & 0xffff0000) == 0x80000000) {
571 if (xlvl >= 0x80000001) {
572 c->x86_capability[1] = cpuid_edx(0x80000001);
573 c->x86_capability[6] = cpuid_ecx(0x80000001);
577 if (c->extended_cpuid_level >= 0x80000008) {
578 u32 eax = cpuid_eax(0x80000008);
580 c->x86_virt_bits = (eax >> 8) & 0xff;
581 c->x86_phys_bits = eax & 0xff;
584 else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
585 c->x86_phys_bits = 36;
588 if (c->extended_cpuid_level >= 0x80000007)
589 c->x86_power = cpuid_edx(0x80000007);
593 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
599 * First of all, decide if this is a 486 or higher
600 * It's a 486 if we can modify the AC flag
602 if (flag_is_changeable_p(X86_EFLAGS_AC))
607 for (i = 0; i < X86_VENDOR_NUM; i++)
608 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
609 c->x86_vendor_id[0] = 0;
610 cpu_devs[i]->c_identify(c);
611 if (c->x86_vendor_id[0]) {
620 * Do minimum CPU detection early.
621 * Fields really needed: vendor, cpuid_level, family, model, mask,
623 * The others are not touched to avoid unwanted side effects.
625 * WARNING: this function is only called on the BP. Don't add code here
626 * that is supposed to run on all CPUs.
628 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
631 c->x86_clflush_size = 64;
632 c->x86_phys_bits = 36;
633 c->x86_virt_bits = 48;
635 c->x86_clflush_size = 32;
636 c->x86_phys_bits = 32;
637 c->x86_virt_bits = 32;
639 c->x86_cache_alignment = c->x86_clflush_size;
641 memset(&c->x86_capability, 0, sizeof c->x86_capability);
642 c->extended_cpuid_level = 0;
645 identify_cpu_without_cpuid(c);
647 /* cyrix could have cpuid enabled via c_identify()*/
657 if (this_cpu->c_early_init)
658 this_cpu->c_early_init(c);
661 c->cpu_index = boot_cpu_id;
663 filter_cpuid_features(c, false);
666 void __init early_cpu_init(void)
668 const struct cpu_dev *const *cdev;
671 printk(KERN_INFO "KERNEL supported cpus:\n");
672 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
673 const struct cpu_dev *cpudev = *cdev;
676 if (count >= X86_VENDOR_NUM)
678 cpu_devs[count] = cpudev;
681 for (j = 0; j < 2; j++) {
682 if (!cpudev->c_ident[j])
684 printk(KERN_INFO " %s %s\n", cpudev->c_vendor,
689 early_identify_cpu(&boot_cpu_data);
693 * The NOPL instruction is supposed to exist on all CPUs with
694 * family >= 6; unfortunately, that's not true in practice because
695 * of early VIA chips and (more importantly) broken virtualizers that
696 * are not easy to detect. In the latter case it doesn't even *fail*
697 * reliably, so probing for it doesn't even work. Disable it completely
698 * unless we can find a reliable way to detect all the broken cases.
700 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
702 clear_cpu_cap(c, X86_FEATURE_NOPL);
705 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
707 c->extended_cpuid_level = 0;
710 identify_cpu_without_cpuid(c);
712 /* cyrix could have cpuid enabled via c_identify()*/
722 if (c->cpuid_level >= 0x00000001) {
723 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
725 # ifdef CONFIG_X86_HT
726 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
728 c->apicid = c->initial_apicid;
733 c->phys_proc_id = c->initial_apicid;
737 get_model_name(c); /* Default name */
739 init_scattered_cpuid_features(c);
744 * This does the hard work of actually picking apart the CPU stuff...
746 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
750 c->loops_per_jiffy = loops_per_jiffy;
751 c->x86_cache_size = -1;
752 c->x86_vendor = X86_VENDOR_UNKNOWN;
753 c->x86_model = c->x86_mask = 0; /* So far unknown... */
754 c->x86_vendor_id[0] = '\0'; /* Unset */
755 c->x86_model_id[0] = '\0'; /* Unset */
756 c->x86_max_cores = 1;
757 c->x86_coreid_bits = 0;
759 c->x86_clflush_size = 64;
760 c->x86_phys_bits = 36;
761 c->x86_virt_bits = 48;
763 c->cpuid_level = -1; /* CPUID not detected */
764 c->x86_clflush_size = 32;
765 c->x86_phys_bits = 32;
766 c->x86_virt_bits = 32;
768 c->x86_cache_alignment = c->x86_clflush_size;
769 memset(&c->x86_capability, 0, sizeof c->x86_capability);
773 if (this_cpu->c_identify)
774 this_cpu->c_identify(c);
777 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
781 * Vendor-specific initialization. In this section we
782 * canonicalize the feature flags, meaning if there are
783 * features a certain CPU supports which CPUID doesn't
784 * tell us, CPUID claiming incorrect flags, or other bugs,
785 * we handle them here.
787 * At the end of this section, c->x86_capability better
788 * indicate the features this CPU genuinely supports!
790 if (this_cpu->c_init)
793 /* Disable the PN if appropriate */
794 squash_the_stupid_serial_number(c);
797 * The vendor-specific functions might have changed features.
798 * Now we do "generic changes."
801 /* Filter out anything that depends on CPUID levels we don't have */
802 filter_cpuid_features(c, true);
804 /* If the model name is still unset, do table lookup. */
805 if (!c->x86_model_id[0]) {
807 p = table_lookup_model(c);
809 strcpy(c->x86_model_id, p);
812 sprintf(c->x86_model_id, "%02x/%02x",
813 c->x86, c->x86_model);
822 * On SMP, boot_cpu_data holds the common feature set between
823 * all CPUs; so make sure that we indicate which features are
824 * common between the CPUs. The first time this routine gets
825 * executed, c == &boot_cpu_data.
827 if (c != &boot_cpu_data) {
828 /* AND the already accumulated flags with these */
829 for (i = 0; i < NCAPINTS; i++)
830 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
833 /* Clear all flags overriden by options */
834 for (i = 0; i < NCAPINTS; i++)
835 c->x86_capability[i] &= ~cleared_cpu_caps[i];
837 #ifdef CONFIG_X86_MCE
838 /* Init Machine Check Exception if available. */
842 select_idle_routine(c);
844 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
845 numa_add_cpu(smp_processor_id());
850 static void vgetcpu_set_mode(void)
852 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
853 vgetcpu_mode = VGETCPU_RDTSCP;
855 vgetcpu_mode = VGETCPU_LSL;
859 void __init identify_boot_cpu(void)
861 identify_cpu(&boot_cpu_data);
870 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
872 BUG_ON(c == &boot_cpu_data);
885 static const struct msr_range msr_range_array[] __cpuinitconst = {
886 { 0x00000000, 0x00000418},
887 { 0xc0000000, 0xc000040b},
888 { 0xc0010000, 0xc0010142},
889 { 0xc0011000, 0xc001103b},
892 static void __cpuinit print_cpu_msr(void)
894 unsigned index_min, index_max;
899 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
900 index_min = msr_range_array[i].min;
901 index_max = msr_range_array[i].max;
903 for (index = index_min; index < index_max; index++) {
904 if (rdmsrl_amd_safe(index, &val))
906 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
911 static int show_msr __cpuinitdata;
913 static __init int setup_show_msr(char *arg)
917 get_option(&arg, &num);
923 __setup("show_msr=", setup_show_msr);
925 static __init int setup_noclflush(char *arg)
927 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
930 __setup("noclflush", setup_noclflush);
932 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
934 const char *vendor = NULL;
936 if (c->x86_vendor < X86_VENDOR_NUM) {
937 vendor = this_cpu->c_vendor;
939 if (c->cpuid_level >= 0)
940 vendor = c->x86_vendor_id;
943 if (vendor && !strstr(c->x86_model_id, vendor))
944 printk(KERN_CONT "%s ", vendor);
946 if (c->x86_model_id[0])
947 printk(KERN_CONT "%s", c->x86_model_id);
949 printk(KERN_CONT "%d86", c->x86);
951 if (c->x86_mask || c->cpuid_level >= 0)
952 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
954 printk(KERN_CONT "\n");
957 if (c->cpu_index < show_msr)
965 static __init int setup_disablecpuid(char *arg)
969 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
970 setup_clear_cpu_cap(bit);
976 __setup("clearcpuid=", setup_disablecpuid);
979 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
981 DEFINE_PER_CPU_FIRST(union irq_stack_union,
982 irq_stack_union) __aligned(PAGE_SIZE);
984 DEFINE_PER_CPU(char *, irq_stack_ptr) =
985 init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
987 DEFINE_PER_CPU(unsigned long, kernel_stack) =
988 (unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
989 EXPORT_PER_CPU_SYMBOL(kernel_stack);
991 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
994 * Special IST stacks which the CPU switches to when it calls
995 * an IST-marked descriptor entry. Up to 7 stacks (hardware
996 * limit), all of them are 4K, except the debug stack which
999 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1000 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1001 [DEBUG_STACK - 1] = DEBUG_STKSZ
1004 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1005 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ])
1006 __aligned(PAGE_SIZE);
1008 /* May not be marked __init: used by software suspend */
1009 void syscall_init(void)
1012 * LSTAR and STAR live in a bit strange symbiosis.
1013 * They both write to the same internal register. STAR allows to
1014 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1016 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
1017 wrmsrl(MSR_LSTAR, system_call);
1018 wrmsrl(MSR_CSTAR, ignore_sysret);
1020 #ifdef CONFIG_IA32_EMULATION
1021 syscall32_cpu_init();
1024 /* Flags to clear on syscall */
1025 wrmsrl(MSR_SYSCALL_MASK,
1026 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
1029 unsigned long kernel_eflags;
1032 * Copies of the original ist values from the tss are only accessed during
1033 * debugging, no special alignment required.
1035 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1037 #else /* CONFIG_X86_64 */
1039 #ifdef CONFIG_CC_STACKPROTECTOR
1040 DEFINE_PER_CPU(unsigned long, stack_canary);
1043 /* Make sure %fs and %gs are initialized properly in idle threads */
1044 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
1046 memset(regs, 0, sizeof(struct pt_regs));
1047 regs->fs = __KERNEL_PERCPU;
1048 regs->gs = __KERNEL_STACK_CANARY;
1052 #endif /* CONFIG_X86_64 */
1055 * Clear all 6 debug registers:
1057 static void clear_all_debug_regs(void)
1061 for (i = 0; i < 8; i++) {
1062 /* Ignore db4, db5 */
1063 if ((i == 4) || (i == 5))
1071 * cpu_init() initializes state that is per-CPU. Some data is already
1072 * initialized (naturally) in the bootstrap process, such as the GDT
1073 * and IDT. We reload them nevertheless, this function acts as a
1074 * 'CPU state barrier', nothing should get across.
1075 * A lot of state is already set up in PDA init for 64 bit
1077 #ifdef CONFIG_X86_64
1079 void __cpuinit cpu_init(void)
1081 struct orig_ist *orig_ist;
1082 struct task_struct *me;
1083 struct tss_struct *t;
1088 cpu = stack_smp_processor_id();
1089 t = &per_cpu(init_tss, cpu);
1090 orig_ist = &per_cpu(orig_ist, cpu);
1093 if (cpu != 0 && percpu_read(node_number) == 0 &&
1094 cpu_to_node(cpu) != NUMA_NO_NODE)
1095 percpu_write(node_number, cpu_to_node(cpu));
1100 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1101 panic("CPU#%d already initialized!\n", cpu);
1103 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1105 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1108 * Initialize the per-CPU GDT with the boot GDT,
1109 * and set up the GDT descriptor:
1112 switch_to_new_gdt(cpu);
1115 load_idt((const struct desc_ptr *)&idt_descr);
1117 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1120 wrmsrl(MSR_FS_BASE, 0);
1121 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1129 * set up and load the per-CPU TSS
1131 if (!orig_ist->ist[0]) {
1132 char *estacks = per_cpu(exception_stacks, cpu);
1134 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1135 estacks += exception_stack_sizes[v];
1136 orig_ist->ist[v] = t->x86_tss.ist[v] =
1137 (unsigned long)estacks;
1141 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1144 * <= is required because the CPU will access up to
1145 * 8 bits beyond the end of the IO permission bitmap.
1147 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1148 t->io_bitmap[i] = ~0UL;
1150 atomic_inc(&init_mm.mm_count);
1151 me->active_mm = &init_mm;
1153 enter_lazy_tlb(&init_mm, me);
1155 load_sp0(t, ¤t->thread);
1156 set_tss_desc(cpu, t);
1158 load_LDT(&init_mm.context);
1162 * If the kgdb is connected no debug regs should be altered. This
1163 * is only applicable when KGDB and a KGDB I/O module are built
1164 * into the kernel and you are using early debugging with
1165 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1167 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1168 arch_kgdb_ops.correct_hw_break();
1171 clear_all_debug_regs();
1175 raw_local_save_flags(kernel_eflags);
1183 void __cpuinit cpu_init(void)
1185 int cpu = smp_processor_id();
1186 struct task_struct *curr = current;
1187 struct tss_struct *t = &per_cpu(init_tss, cpu);
1188 struct thread_struct *thread = &curr->thread;
1190 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1191 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1196 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1198 if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1199 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1201 load_idt(&idt_descr);
1202 switch_to_new_gdt(cpu);
1205 * Set up and load the per-CPU TSS and LDT
1207 atomic_inc(&init_mm.mm_count);
1208 curr->active_mm = &init_mm;
1210 enter_lazy_tlb(&init_mm, curr);
1212 load_sp0(t, thread);
1213 set_tss_desc(cpu, t);
1215 load_LDT(&init_mm.context);
1217 #ifdef CONFIG_DOUBLEFAULT
1218 /* Set up doublefault TSS pointer in the GDT */
1219 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1222 clear_all_debug_regs();
1225 * Force FPU initialization:
1228 current_thread_info()->status = TS_XSAVE;
1230 current_thread_info()->status = 0;
1232 mxcsr_feature_mask_init();
1235 * Boot processor to setup the FP and extended state context info.
1237 if (smp_processor_id() == boot_cpu_id)
1238 init_thread_xstate();