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_counter.h>
17 #include <asm/mmu_context.h>
18 #include <asm/hypervisor.h>
19 #include <asm/processor.h>
20 #include <asm/sections.h>
21 #include <asm/topology.h>
22 #include <asm/cpumask.h>
23 #include <asm/pgtable.h>
24 #include <asm/atomic.h>
25 #include <asm/proto.h>
26 #include <asm/setup.h>
39 #ifdef CONFIG_X86_LOCAL_APIC
40 #include <asm/uv/uv.h>
45 /* all of these masks are initialized in setup_cpu_local_masks() */
46 cpumask_var_t cpu_initialized_mask;
47 cpumask_var_t cpu_callout_mask;
48 cpumask_var_t cpu_callin_mask;
50 /* representing cpus for which sibling maps can be computed */
51 cpumask_var_t cpu_sibling_setup_mask;
53 /* correctly size the local cpu masks */
54 void __init setup_cpu_local_masks(void)
56 alloc_bootmem_cpumask_var(&cpu_initialized_mask);
57 alloc_bootmem_cpumask_var(&cpu_callin_mask);
58 alloc_bootmem_cpumask_var(&cpu_callout_mask);
59 alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
62 static const struct cpu_dev *this_cpu __cpuinitdata;
64 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
67 * We need valid kernel segments for data and code in long mode too
68 * IRET will check the segment types kkeil 2000/10/28
69 * Also sysret mandates a special GDT layout
71 * TLS descriptors are currently at a different place compared to i386.
72 * Hopefully nobody expects them at a fixed place (Wine?)
74 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
75 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
76 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
77 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
78 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
79 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
81 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } },
82 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } },
83 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } },
84 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff200 } } },
86 * Segments used for calling PnP BIOS have byte granularity.
87 * They code segments and data segments have fixed 64k limits,
88 * the transfer segment sizes are set at run time.
91 [GDT_ENTRY_PNPBIOS_CS32] = { { { 0x0000ffff, 0x00409a00 } } },
93 [GDT_ENTRY_PNPBIOS_CS16] = { { { 0x0000ffff, 0x00009a00 } } },
95 [GDT_ENTRY_PNPBIOS_DS] = { { { 0x0000ffff, 0x00009200 } } },
97 [GDT_ENTRY_PNPBIOS_TS1] = { { { 0x00000000, 0x00009200 } } },
99 [GDT_ENTRY_PNPBIOS_TS2] = { { { 0x00000000, 0x00009200 } } },
101 * The APM segments have byte granularity and their bases
102 * are set at run time. All have 64k limits.
105 [GDT_ENTRY_APMBIOS_BASE] = { { { 0x0000ffff, 0x00409a00 } } },
107 [GDT_ENTRY_APMBIOS_BASE+1] = { { { 0x0000ffff, 0x00009a00 } } },
109 [GDT_ENTRY_APMBIOS_BASE+2] = { { { 0x0000ffff, 0x00409200 } } },
111 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } },
112 [GDT_ENTRY_PERCPU] = { { { 0x0000ffff, 0x00cf9200 } } },
113 GDT_STACK_CANARY_INIT
116 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
118 static int __init x86_xsave_setup(char *s)
120 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
123 __setup("noxsave", x86_xsave_setup);
126 static int cachesize_override __cpuinitdata = -1;
127 static int disable_x86_serial_nr __cpuinitdata = 1;
129 static int __init cachesize_setup(char *str)
131 get_option(&str, &cachesize_override);
134 __setup("cachesize=", cachesize_setup);
136 static int __init x86_fxsr_setup(char *s)
138 setup_clear_cpu_cap(X86_FEATURE_FXSR);
139 setup_clear_cpu_cap(X86_FEATURE_XMM);
142 __setup("nofxsr", x86_fxsr_setup);
144 static int __init x86_sep_setup(char *s)
146 setup_clear_cpu_cap(X86_FEATURE_SEP);
149 __setup("nosep", x86_sep_setup);
151 /* Standard macro to see if a specific flag is changeable */
152 static inline int flag_is_changeable_p(u32 flag)
157 * Cyrix and IDT cpus allow disabling of CPUID
158 * so the code below may return different results
159 * when it is executed before and after enabling
160 * the CPUID. Add "volatile" to not allow gcc to
161 * optimize the subsequent calls to this function.
163 asm volatile ("pushfl \n\t"
174 : "=&r" (f1), "=&r" (f2)
177 return ((f1^f2) & flag) != 0;
180 /* Probe for the CPUID instruction */
181 static int __cpuinit have_cpuid_p(void)
183 return flag_is_changeable_p(X86_EFLAGS_ID);
186 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
188 unsigned long lo, hi;
190 if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
193 /* Disable processor serial number: */
195 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
197 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);
206 static int __init x86_serial_nr_setup(char *s)
208 disable_x86_serial_nr = 0;
211 __setup("serialnumber", x86_serial_nr_setup);
213 static inline int flag_is_changeable_p(u32 flag)
217 /* Probe for the CPUID instruction */
218 static inline int have_cpuid_p(void)
222 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
228 * Some CPU features depend on higher CPUID levels, which may not always
229 * be available due to CPUID level capping or broken virtualization
230 * software. Add those features to this table to auto-disable them.
232 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;
249 for (df = cpuid_dependent_features; df->feature; df++) {
251 if (!cpu_has(c, df->feature))
254 * Note: cpuid_level is set to -1 if unavailable, but
255 * extended_extended_level is set to 0 if unavailable
256 * and the legitimate extended levels are all negative
257 * when signed; hence the weird messing around with
260 if (!((s32)df->level < 0 ?
261 (u32)df->level > (u32)c->extended_cpuid_level :
262 (s32)df->level > (s32)c->cpuid_level))
265 clear_cpu_cap(c, df->feature);
270 "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
271 x86_cap_flags[df->feature], df->level);
276 * Naming convention should be: <Name> [(<Codename>)]
277 * This table only is used unless init_<vendor>() below doesn't set it;
278 * in particular, if CPUID levels 0x80000002..4 are supported, this
282 /* Look up CPU names by table lookup. */
283 static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
285 const struct cpu_model_info *info;
287 if (c->x86_model >= 16)
288 return NULL; /* Range check */
293 info = this_cpu->c_models;
295 while (info && info->family) {
296 if (info->family == c->x86)
297 return info->model_names[c->x86_model];
300 return NULL; /* Not found */
303 __u32 cpu_caps_cleared[NCAPINTS] __cpuinitdata;
304 __u32 cpu_caps_set[NCAPINTS] __cpuinitdata;
306 void load_percpu_segment(int cpu)
309 loadsegment(fs, __KERNEL_PERCPU);
312 wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
314 load_stack_canary_segment();
318 * Current gdt points %fs at the "master" per-cpu area: after this,
319 * it's on the real one.
321 void switch_to_new_gdt(int cpu)
323 struct desc_ptr gdt_descr;
325 gdt_descr.address = (long)get_cpu_gdt_table(cpu);
326 gdt_descr.size = GDT_SIZE - 1;
327 load_gdt(&gdt_descr);
328 /* Reload the per-cpu base */
330 load_percpu_segment(cpu);
333 static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
335 static void __cpuinit default_init(struct cpuinfo_x86 *c)
338 display_cacheinfo(c);
340 /* Not much we can do here... */
341 /* Check if at least it has cpuid */
342 if (c->cpuid_level == -1) {
343 /* No cpuid. It must be an ancient CPU */
345 strcpy(c->x86_model_id, "486");
346 else if (c->x86 == 3)
347 strcpy(c->x86_model_id, "386");
352 static const struct cpu_dev __cpuinitconst default_cpu = {
353 .c_init = default_init,
354 .c_vendor = "Unknown",
355 .c_x86_vendor = X86_VENDOR_UNKNOWN,
358 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
363 if (c->extended_cpuid_level < 0x80000004)
366 v = (unsigned int *)c->x86_model_id;
367 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
368 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
369 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
370 c->x86_model_id[48] = 0;
373 * Intel chips right-justify this string for some dumb reason;
374 * undo that brain damage:
376 p = q = &c->x86_model_id[0];
382 while (q <= &c->x86_model_id[48])
383 *q++ = '\0'; /* Zero-pad the rest */
387 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
389 unsigned int n, dummy, ebx, ecx, edx, l2size;
391 n = c->extended_cpuid_level;
393 if (n >= 0x80000005) {
394 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
395 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
396 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
397 c->x86_cache_size = (ecx>>24) + (edx>>24);
399 /* On K8 L1 TLB is inclusive, so don't count it */
404 if (n < 0x80000006) /* Some chips just has a large L1. */
407 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
411 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
413 /* do processor-specific cache resizing */
414 if (this_cpu->c_size_cache)
415 l2size = this_cpu->c_size_cache(c, l2size);
417 /* Allow user to override all this if necessary. */
418 if (cachesize_override != -1)
419 l2size = cachesize_override;
422 return; /* Again, no L2 cache is possible */
425 c->x86_cache_size = l2size;
427 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
431 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
434 u32 eax, ebx, ecx, edx;
435 int index_msb, core_bits;
437 if (!cpu_has(c, X86_FEATURE_HT))
440 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
443 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
446 cpuid(1, &eax, &ebx, &ecx, &edx);
448 smp_num_siblings = (ebx & 0xff0000) >> 16;
450 if (smp_num_siblings == 1) {
451 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
455 if (smp_num_siblings <= 1)
458 if (smp_num_siblings > nr_cpu_ids) {
459 pr_warning("CPU: Unsupported number of siblings %d",
461 smp_num_siblings = 1;
465 index_msb = get_count_order(smp_num_siblings);
466 c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
468 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
470 index_msb = get_count_order(smp_num_siblings);
472 core_bits = get_count_order(c->x86_max_cores);
474 c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
475 ((1 << core_bits) - 1);
478 if ((c->x86_max_cores * smp_num_siblings) > 1) {
479 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
481 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
487 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
489 char *v = c->x86_vendor_id;
493 for (i = 0; i < X86_VENDOR_NUM; i++) {
497 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
498 (cpu_devs[i]->c_ident[1] &&
499 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
501 this_cpu = cpu_devs[i];
502 c->x86_vendor = this_cpu->c_x86_vendor;
510 "CPU: vendor_id '%s' unknown, using generic init.\n", v);
512 printk(KERN_ERR "CPU: Your system may be unstable.\n");
515 c->x86_vendor = X86_VENDOR_UNKNOWN;
516 this_cpu = &default_cpu;
519 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
521 /* Get vendor name */
522 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
523 (unsigned int *)&c->x86_vendor_id[0],
524 (unsigned int *)&c->x86_vendor_id[8],
525 (unsigned int *)&c->x86_vendor_id[4]);
528 /* Intel-defined flags: level 0x00000001 */
529 if (c->cpuid_level >= 0x00000001) {
530 u32 junk, tfms, cap0, misc;
532 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
533 c->x86 = (tfms >> 8) & 0xf;
534 c->x86_model = (tfms >> 4) & 0xf;
535 c->x86_mask = tfms & 0xf;
538 c->x86 += (tfms >> 20) & 0xff;
540 c->x86_model += ((tfms >> 16) & 0xf) << 4;
542 if (cap0 & (1<<19)) {
543 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
544 c->x86_cache_alignment = c->x86_clflush_size;
549 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
554 /* Intel-defined flags: level 0x00000001 */
555 if (c->cpuid_level >= 0x00000001) {
556 u32 capability, excap;
558 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
559 c->x86_capability[0] = capability;
560 c->x86_capability[4] = excap;
563 /* AMD-defined flags: level 0x80000001 */
564 xlvl = cpuid_eax(0x80000000);
565 c->extended_cpuid_level = xlvl;
567 if ((xlvl & 0xffff0000) == 0x80000000) {
568 if (xlvl >= 0x80000001) {
569 c->x86_capability[1] = cpuid_edx(0x80000001);
570 c->x86_capability[6] = cpuid_ecx(0x80000001);
574 if (c->extended_cpuid_level >= 0x80000008) {
575 u32 eax = cpuid_eax(0x80000008);
577 c->x86_virt_bits = (eax >> 8) & 0xff;
578 c->x86_phys_bits = eax & 0xff;
581 else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
582 c->x86_phys_bits = 36;
585 if (c->extended_cpuid_level >= 0x80000007)
586 c->x86_power = cpuid_edx(0x80000007);
590 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
596 * First of all, decide if this is a 486 or higher
597 * It's a 486 if we can modify the AC flag
599 if (flag_is_changeable_p(X86_EFLAGS_AC))
604 for (i = 0; i < X86_VENDOR_NUM; i++)
605 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
606 c->x86_vendor_id[0] = 0;
607 cpu_devs[i]->c_identify(c);
608 if (c->x86_vendor_id[0]) {
617 * Do minimum CPU detection early.
618 * Fields really needed: vendor, cpuid_level, family, model, mask,
620 * The others are not touched to avoid unwanted side effects.
622 * WARNING: this function is only called on the BP. Don't add code here
623 * that is supposed to run on all CPUs.
625 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
628 c->x86_clflush_size = 64;
629 c->x86_phys_bits = 36;
630 c->x86_virt_bits = 48;
632 c->x86_clflush_size = 32;
633 c->x86_phys_bits = 32;
634 c->x86_virt_bits = 32;
636 c->x86_cache_alignment = c->x86_clflush_size;
638 memset(&c->x86_capability, 0, sizeof c->x86_capability);
639 c->extended_cpuid_level = 0;
642 identify_cpu_without_cpuid(c);
644 /* cyrix could have cpuid enabled via c_identify()*/
654 if (this_cpu->c_early_init)
655 this_cpu->c_early_init(c);
658 c->cpu_index = boot_cpu_id;
660 filter_cpuid_features(c, false);
663 void __init early_cpu_init(void)
665 const struct cpu_dev *const *cdev;
668 printk(KERN_INFO "KERNEL supported cpus:\n");
669 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
670 const struct cpu_dev *cpudev = *cdev;
673 if (count >= X86_VENDOR_NUM)
675 cpu_devs[count] = cpudev;
678 for (j = 0; j < 2; j++) {
679 if (!cpudev->c_ident[j])
681 printk(KERN_INFO " %s %s\n", cpudev->c_vendor,
686 early_identify_cpu(&boot_cpu_data);
690 * The NOPL instruction is supposed to exist on all CPUs with
691 * family >= 6; unfortunately, that's not true in practice because
692 * of early VIA chips and (more importantly) broken virtualizers that
693 * are not easy to detect. In the latter case it doesn't even *fail*
694 * reliably, so probing for it doesn't even work. Disable it completely
695 * unless we can find a reliable way to detect all the broken cases.
697 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
699 clear_cpu_cap(c, X86_FEATURE_NOPL);
702 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
704 c->extended_cpuid_level = 0;
707 identify_cpu_without_cpuid(c);
709 /* cyrix could have cpuid enabled via c_identify()*/
719 if (c->cpuid_level >= 0x00000001) {
720 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
722 # ifdef CONFIG_X86_HT
723 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
725 c->apicid = c->initial_apicid;
730 c->phys_proc_id = c->initial_apicid;
734 get_model_name(c); /* Default name */
736 init_scattered_cpuid_features(c);
741 * This does the hard work of actually picking apart the CPU stuff...
743 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
747 c->loops_per_jiffy = loops_per_jiffy;
748 c->x86_cache_size = -1;
749 c->x86_vendor = X86_VENDOR_UNKNOWN;
750 c->x86_model = c->x86_mask = 0; /* So far unknown... */
751 c->x86_vendor_id[0] = '\0'; /* Unset */
752 c->x86_model_id[0] = '\0'; /* Unset */
753 c->x86_max_cores = 1;
754 c->x86_coreid_bits = 0;
756 c->x86_clflush_size = 64;
757 c->x86_phys_bits = 36;
758 c->x86_virt_bits = 48;
760 c->cpuid_level = -1; /* CPUID not detected */
761 c->x86_clflush_size = 32;
762 c->x86_phys_bits = 32;
763 c->x86_virt_bits = 32;
765 c->x86_cache_alignment = c->x86_clflush_size;
766 memset(&c->x86_capability, 0, sizeof c->x86_capability);
770 if (this_cpu->c_identify)
771 this_cpu->c_identify(c);
773 /* Clear/Set all flags overriden by options, after probe */
774 for (i = 0; i < NCAPINTS; i++) {
775 c->x86_capability[i] &= ~cpu_caps_cleared[i];
776 c->x86_capability[i] |= cpu_caps_set[i];
780 c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
784 * Vendor-specific initialization. In this section we
785 * canonicalize the feature flags, meaning if there are
786 * features a certain CPU supports which CPUID doesn't
787 * tell us, CPUID claiming incorrect flags, or other bugs,
788 * we handle them here.
790 * At the end of this section, c->x86_capability better
791 * indicate the features this CPU genuinely supports!
793 if (this_cpu->c_init)
796 /* Disable the PN if appropriate */
797 squash_the_stupid_serial_number(c);
800 * The vendor-specific functions might have changed features.
801 * Now we do "generic changes."
804 /* Filter out anything that depends on CPUID levels we don't have */
805 filter_cpuid_features(c, true);
807 /* If the model name is still unset, do table lookup. */
808 if (!c->x86_model_id[0]) {
810 p = table_lookup_model(c);
812 strcpy(c->x86_model_id, p);
815 sprintf(c->x86_model_id, "%02x/%02x",
816 c->x86, c->x86_model);
826 * Clear/Set all flags overriden by options, need do it
827 * before following smp all cpus cap AND.
829 for (i = 0; i < NCAPINTS; i++) {
830 c->x86_capability[i] &= ~cpu_caps_cleared[i];
831 c->x86_capability[i] |= cpu_caps_set[i];
835 * On SMP, boot_cpu_data holds the common feature set between
836 * all CPUs; so make sure that we indicate which features are
837 * common between the CPUs. The first time this routine gets
838 * executed, c == &boot_cpu_data.
840 if (c != &boot_cpu_data) {
841 /* AND the already accumulated flags with these */
842 for (i = 0; i < NCAPINTS; i++)
843 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
846 #ifdef CONFIG_X86_MCE
847 /* Init Machine Check Exception if available. */
851 select_idle_routine(c);
853 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
854 numa_add_cpu(smp_processor_id());
857 /* Cap the iomem address space to what is addressable on all CPUs */
858 iomem_resource.end &= (1ULL << c->x86_phys_bits) - 1;
862 static void vgetcpu_set_mode(void)
864 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
865 vgetcpu_mode = VGETCPU_RDTSCP;
867 vgetcpu_mode = VGETCPU_LSL;
871 void __init identify_boot_cpu(void)
873 identify_cpu(&boot_cpu_data);
881 init_hw_perf_counters();
884 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
886 BUG_ON(c == &boot_cpu_data);
899 static const struct msr_range msr_range_array[] __cpuinitconst = {
900 { 0x00000000, 0x00000418},
901 { 0xc0000000, 0xc000040b},
902 { 0xc0010000, 0xc0010142},
903 { 0xc0011000, 0xc001103b},
906 static void __cpuinit print_cpu_msr(void)
908 unsigned index_min, index_max;
913 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
914 index_min = msr_range_array[i].min;
915 index_max = msr_range_array[i].max;
917 for (index = index_min; index < index_max; index++) {
918 if (rdmsrl_amd_safe(index, &val))
920 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
925 static int show_msr __cpuinitdata;
927 static __init int setup_show_msr(char *arg)
931 get_option(&arg, &num);
937 __setup("show_msr=", setup_show_msr);
939 static __init int setup_noclflush(char *arg)
941 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
944 __setup("noclflush", setup_noclflush);
946 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
948 const char *vendor = NULL;
950 if (c->x86_vendor < X86_VENDOR_NUM) {
951 vendor = this_cpu->c_vendor;
953 if (c->cpuid_level >= 0)
954 vendor = c->x86_vendor_id;
957 if (vendor && !strstr(c->x86_model_id, vendor))
958 printk(KERN_CONT "%s ", vendor);
960 if (c->x86_model_id[0])
961 printk(KERN_CONT "%s", c->x86_model_id);
963 printk(KERN_CONT "%d86", c->x86);
965 if (c->x86_mask || c->cpuid_level >= 0)
966 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
968 printk(KERN_CONT "\n");
971 if (c->cpu_index < show_msr)
979 static __init int setup_disablecpuid(char *arg)
983 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
984 setup_clear_cpu_cap(bit);
990 __setup("clearcpuid=", setup_disablecpuid);
993 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
995 DEFINE_PER_CPU_FIRST(union irq_stack_union,
996 irq_stack_union) __aligned(PAGE_SIZE);
998 DEFINE_PER_CPU(char *, irq_stack_ptr) =
999 init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1001 DEFINE_PER_CPU(unsigned long, kernel_stack) =
1002 (unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
1003 EXPORT_PER_CPU_SYMBOL(kernel_stack);
1005 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
1008 * Special IST stacks which the CPU switches to when it calls
1009 * an IST-marked descriptor entry. Up to 7 stacks (hardware
1010 * limit), all of them are 4K, except the debug stack which
1013 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1014 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1015 [DEBUG_STACK - 1] = DEBUG_STKSZ
1018 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1019 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ])
1020 __aligned(PAGE_SIZE);
1022 /* May not be marked __init: used by software suspend */
1023 void syscall_init(void)
1026 * LSTAR and STAR live in a bit strange symbiosis.
1027 * They both write to the same internal register. STAR allows to
1028 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1030 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
1031 wrmsrl(MSR_LSTAR, system_call);
1032 wrmsrl(MSR_CSTAR, ignore_sysret);
1034 #ifdef CONFIG_IA32_EMULATION
1035 syscall32_cpu_init();
1038 /* Flags to clear on syscall */
1039 wrmsrl(MSR_SYSCALL_MASK,
1040 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
1043 unsigned long kernel_eflags;
1046 * Copies of the original ist values from the tss are only accessed during
1047 * debugging, no special alignment required.
1049 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1051 #else /* CONFIG_X86_64 */
1053 #ifdef CONFIG_CC_STACKPROTECTOR
1054 DEFINE_PER_CPU(unsigned long, stack_canary);
1057 /* Make sure %fs and %gs are initialized properly in idle threads */
1058 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
1060 memset(regs, 0, sizeof(struct pt_regs));
1061 regs->fs = __KERNEL_PERCPU;
1062 regs->gs = __KERNEL_STACK_CANARY;
1066 #endif /* CONFIG_X86_64 */
1069 * Clear all 6 debug registers:
1071 static void clear_all_debug_regs(void)
1075 for (i = 0; i < 8; i++) {
1076 /* Ignore db4, db5 */
1077 if ((i == 4) || (i == 5))
1085 * cpu_init() initializes state that is per-CPU. Some data is already
1086 * initialized (naturally) in the bootstrap process, such as the GDT
1087 * and IDT. We reload them nevertheless, this function acts as a
1088 * 'CPU state barrier', nothing should get across.
1089 * A lot of state is already set up in PDA init for 64 bit
1091 #ifdef CONFIG_X86_64
1093 void __cpuinit cpu_init(void)
1095 struct orig_ist *orig_ist;
1096 struct task_struct *me;
1097 struct tss_struct *t;
1102 cpu = stack_smp_processor_id();
1103 t = &per_cpu(init_tss, cpu);
1104 orig_ist = &per_cpu(orig_ist, cpu);
1107 if (cpu != 0 && percpu_read(node_number) == 0 &&
1108 cpu_to_node(cpu) != NUMA_NO_NODE)
1109 percpu_write(node_number, cpu_to_node(cpu));
1114 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1115 panic("CPU#%d already initialized!\n", cpu);
1117 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1119 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1122 * Initialize the per-CPU GDT with the boot GDT,
1123 * and set up the GDT descriptor:
1126 switch_to_new_gdt(cpu);
1129 load_idt((const struct desc_ptr *)&idt_descr);
1131 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1134 wrmsrl(MSR_FS_BASE, 0);
1135 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1143 * set up and load the per-CPU TSS
1145 if (!orig_ist->ist[0]) {
1146 char *estacks = per_cpu(exception_stacks, cpu);
1148 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1149 estacks += exception_stack_sizes[v];
1150 orig_ist->ist[v] = t->x86_tss.ist[v] =
1151 (unsigned long)estacks;
1155 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1158 * <= is required because the CPU will access up to
1159 * 8 bits beyond the end of the IO permission bitmap.
1161 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1162 t->io_bitmap[i] = ~0UL;
1164 atomic_inc(&init_mm.mm_count);
1165 me->active_mm = &init_mm;
1167 enter_lazy_tlb(&init_mm, me);
1169 load_sp0(t, ¤t->thread);
1170 set_tss_desc(cpu, t);
1172 load_LDT(&init_mm.context);
1176 * If the kgdb is connected no debug regs should be altered. This
1177 * is only applicable when KGDB and a KGDB I/O module are built
1178 * into the kernel and you are using early debugging with
1179 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1181 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1182 arch_kgdb_ops.correct_hw_break();
1185 clear_all_debug_regs();
1189 raw_local_save_flags(kernel_eflags);
1197 void __cpuinit cpu_init(void)
1199 int cpu = smp_processor_id();
1200 struct task_struct *curr = current;
1201 struct tss_struct *t = &per_cpu(init_tss, cpu);
1202 struct thread_struct *thread = &curr->thread;
1204 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1205 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1210 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1212 if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1213 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1215 load_idt(&idt_descr);
1216 switch_to_new_gdt(cpu);
1219 * Set up and load the per-CPU TSS and LDT
1221 atomic_inc(&init_mm.mm_count);
1222 curr->active_mm = &init_mm;
1224 enter_lazy_tlb(&init_mm, curr);
1226 load_sp0(t, thread);
1227 set_tss_desc(cpu, t);
1229 load_LDT(&init_mm.context);
1231 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1233 #ifdef CONFIG_DOUBLEFAULT
1234 /* Set up doublefault TSS pointer in the GDT */
1235 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1238 clear_all_debug_regs();
1241 * Force FPU initialization:
1244 current_thread_info()->status = TS_XSAVE;
1246 current_thread_info()->status = 0;
1248 mxcsr_feature_mask_init();
1251 * Boot processor to setup the FP and extended state context info.
1253 if (smp_processor_id() == boot_cpu_id)
1254 init_thread_xstate();