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 <mach_apic.h>
30 #include <asm/genapic.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 #else /* CONFIG_X86_32 */
57 cpumask_t cpu_callin_map;
58 cpumask_t cpu_callout_map;
59 cpumask_t cpu_initialized;
60 cpumask_t cpu_sibling_setup_map;
62 #endif /* CONFIG_X86_32 */
65 static struct cpu_dev *this_cpu __cpuinitdata;
68 /* We need valid kernel segments for data and code in long mode too
69 * IRET will check the segment types kkeil 2000/10/28
70 * Also sysret mandates a special GDT layout
72 /* The TLS descriptors are currently at a different place compared to i386.
73 Hopefully nobody expects them at a fixed place (Wine?) */
74 DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
75 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
76 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
77 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
78 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
79 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
80 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
83 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
84 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } },
85 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } },
86 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } },
87 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff200 } } },
89 * Segments used for calling PnP BIOS have byte granularity.
90 * They code segments and data segments have fixed 64k limits,
91 * the transfer segment sizes are set at run time.
94 [GDT_ENTRY_PNPBIOS_CS32] = { { { 0x0000ffff, 0x00409a00 } } },
96 [GDT_ENTRY_PNPBIOS_CS16] = { { { 0x0000ffff, 0x00009a00 } } },
98 [GDT_ENTRY_PNPBIOS_DS] = { { { 0x0000ffff, 0x00009200 } } },
100 [GDT_ENTRY_PNPBIOS_TS1] = { { { 0x00000000, 0x00009200 } } },
102 [GDT_ENTRY_PNPBIOS_TS2] = { { { 0x00000000, 0x00009200 } } },
104 * The APM segments have byte granularity and their bases
105 * are set at run time. All have 64k limits.
108 [GDT_ENTRY_APMBIOS_BASE] = { { { 0x0000ffff, 0x00409a00 } } },
110 [GDT_ENTRY_APMBIOS_BASE+1] = { { { 0x0000ffff, 0x00009a00 } } },
112 [GDT_ENTRY_APMBIOS_BASE+2] = { { { 0x0000ffff, 0x00409200 } } },
114 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } },
115 [GDT_ENTRY_PERCPU] = { { { 0x00000000, 0x00000000 } } },
118 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
121 static int cachesize_override __cpuinitdata = -1;
122 static int disable_x86_serial_nr __cpuinitdata = 1;
124 static int __init cachesize_setup(char *str)
126 get_option(&str, &cachesize_override);
129 __setup("cachesize=", cachesize_setup);
131 static int __init x86_fxsr_setup(char *s)
133 setup_clear_cpu_cap(X86_FEATURE_FXSR);
134 setup_clear_cpu_cap(X86_FEATURE_XMM);
137 __setup("nofxsr", x86_fxsr_setup);
139 static int __init x86_sep_setup(char *s)
141 setup_clear_cpu_cap(X86_FEATURE_SEP);
144 __setup("nosep", x86_sep_setup);
146 /* Standard macro to see if a specific flag is changeable */
147 static inline int flag_is_changeable_p(u32 flag)
152 * Cyrix and IDT cpus allow disabling of CPUID
153 * so the code below may return different results
154 * when it is executed before and after enabling
155 * the CPUID. Add "volatile" to not allow gcc to
156 * optimize the subsequent calls to this function.
158 asm volatile ("pushfl\n\t"
168 : "=&r" (f1), "=&r" (f2)
171 return ((f1^f2) & flag) != 0;
174 /* Probe for the CPUID instruction */
175 static int __cpuinit have_cpuid_p(void)
177 return flag_is_changeable_p(X86_EFLAGS_ID);
180 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
182 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
183 /* Disable processor serial number */
184 unsigned long lo, hi;
185 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
187 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
188 printk(KERN_NOTICE "CPU serial number disabled.\n");
189 clear_cpu_cap(c, X86_FEATURE_PN);
191 /* Disabling the serial number may affect the cpuid level */
192 c->cpuid_level = cpuid_eax(0);
196 static int __init x86_serial_nr_setup(char *s)
198 disable_x86_serial_nr = 0;
201 __setup("serialnumber", x86_serial_nr_setup);
203 static inline int flag_is_changeable_p(u32 flag)
207 /* Probe for the CPUID instruction */
208 static inline int have_cpuid_p(void)
212 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
218 * Naming convention should be: <Name> [(<Codename>)]
219 * This table only is used unless init_<vendor>() below doesn't set it;
220 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
224 /* Look up CPU names by table lookup. */
225 static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
227 struct cpu_model_info *info;
229 if (c->x86_model >= 16)
230 return NULL; /* Range check */
235 info = this_cpu->c_models;
237 while (info && info->family) {
238 if (info->family == c->x86)
239 return info->model_names[c->x86_model];
242 return NULL; /* Not found */
245 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
247 /* Current gdt points %fs at the "master" per-cpu area: after this,
248 * it's on the real one. */
249 void switch_to_new_gdt(void)
251 struct desc_ptr gdt_descr;
253 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
254 gdt_descr.size = GDT_SIZE - 1;
255 load_gdt(&gdt_descr);
257 asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
261 static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
263 static void __cpuinit default_init(struct cpuinfo_x86 *c)
266 display_cacheinfo(c);
268 /* Not much we can do here... */
269 /* Check if at least it has cpuid */
270 if (c->cpuid_level == -1) {
271 /* No cpuid. It must be an ancient CPU */
273 strcpy(c->x86_model_id, "486");
274 else if (c->x86 == 3)
275 strcpy(c->x86_model_id, "386");
280 static struct cpu_dev __cpuinitdata default_cpu = {
281 .c_init = default_init,
282 .c_vendor = "Unknown",
283 .c_x86_vendor = X86_VENDOR_UNKNOWN,
286 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
291 if (c->extended_cpuid_level < 0x80000004)
294 v = (unsigned int *) c->x86_model_id;
295 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
296 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
297 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
298 c->x86_model_id[48] = 0;
300 /* Intel chips right-justify this string for some dumb reason;
301 undo that brain damage */
302 p = q = &c->x86_model_id[0];
308 while (q <= &c->x86_model_id[48])
309 *q++ = '\0'; /* Zero-pad the rest */
313 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
315 unsigned int n, dummy, ebx, ecx, edx, l2size;
317 n = c->extended_cpuid_level;
319 if (n >= 0x80000005) {
320 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
321 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
322 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
323 c->x86_cache_size = (ecx>>24) + (edx>>24);
325 /* On K8 L1 TLB is inclusive, so don't count it */
330 if (n < 0x80000006) /* Some chips just has a large L1. */
333 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
337 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
339 /* do processor-specific cache resizing */
340 if (this_cpu->c_size_cache)
341 l2size = this_cpu->c_size_cache(c, l2size);
343 /* Allow user to override all this if necessary. */
344 if (cachesize_override != -1)
345 l2size = cachesize_override;
348 return; /* Again, no L2 cache is possible */
351 c->x86_cache_size = l2size;
353 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
357 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
360 u32 eax, ebx, ecx, edx;
361 int index_msb, core_bits;
363 if (!cpu_has(c, X86_FEATURE_HT))
366 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
369 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
372 cpuid(1, &eax, &ebx, &ecx, &edx);
374 smp_num_siblings = (ebx & 0xff0000) >> 16;
376 if (smp_num_siblings == 1) {
377 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
378 } else if (smp_num_siblings > 1) {
380 if (smp_num_siblings > nr_cpu_ids) {
381 printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
383 smp_num_siblings = 1;
387 index_msb = get_count_order(smp_num_siblings);
389 c->phys_proc_id = phys_pkg_id(index_msb);
391 c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb);
394 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
396 index_msb = get_count_order(smp_num_siblings);
398 core_bits = get_count_order(c->x86_max_cores);
401 c->cpu_core_id = phys_pkg_id(index_msb) &
402 ((1 << core_bits) - 1);
404 c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) &
405 ((1 << core_bits) - 1);
410 if ((c->x86_max_cores * smp_num_siblings) > 1) {
411 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
413 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
419 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
421 char *v = c->x86_vendor_id;
425 for (i = 0; i < X86_VENDOR_NUM; i++) {
429 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
430 (cpu_devs[i]->c_ident[1] &&
431 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
432 this_cpu = cpu_devs[i];
433 c->x86_vendor = this_cpu->c_x86_vendor;
440 printk(KERN_ERR "CPU: vendor_id '%s' unknown, using generic init.\n", v);
441 printk(KERN_ERR "CPU: Your system may be unstable.\n");
444 c->x86_vendor = X86_VENDOR_UNKNOWN;
445 this_cpu = &default_cpu;
448 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
450 /* Get vendor name */
451 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
452 (unsigned int *)&c->x86_vendor_id[0],
453 (unsigned int *)&c->x86_vendor_id[8],
454 (unsigned int *)&c->x86_vendor_id[4]);
457 /* Intel-defined flags: level 0x00000001 */
458 if (c->cpuid_level >= 0x00000001) {
459 u32 junk, tfms, cap0, misc;
460 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
461 c->x86 = (tfms >> 8) & 0xf;
462 c->x86_model = (tfms >> 4) & 0xf;
463 c->x86_mask = tfms & 0xf;
465 c->x86 += (tfms >> 20) & 0xff;
467 c->x86_model += ((tfms >> 16) & 0xf) << 4;
468 if (cap0 & (1<<19)) {
469 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
470 c->x86_cache_alignment = c->x86_clflush_size;
475 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
480 /* Intel-defined flags: level 0x00000001 */
481 if (c->cpuid_level >= 0x00000001) {
482 u32 capability, excap;
483 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
484 c->x86_capability[0] = capability;
485 c->x86_capability[4] = excap;
488 /* AMD-defined flags: level 0x80000001 */
489 xlvl = cpuid_eax(0x80000000);
490 c->extended_cpuid_level = xlvl;
491 if ((xlvl & 0xffff0000) == 0x80000000) {
492 if (xlvl >= 0x80000001) {
493 c->x86_capability[1] = cpuid_edx(0x80000001);
494 c->x86_capability[6] = cpuid_ecx(0x80000001);
499 if (c->extended_cpuid_level >= 0x80000008) {
500 u32 eax = cpuid_eax(0x80000008);
502 c->x86_virt_bits = (eax >> 8) & 0xff;
503 c->x86_phys_bits = eax & 0xff;
507 if (c->extended_cpuid_level >= 0x80000007)
508 c->x86_power = cpuid_edx(0x80000007);
512 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
518 * First of all, decide if this is a 486 or higher
519 * It's a 486 if we can modify the AC flag
521 if (flag_is_changeable_p(X86_EFLAGS_AC))
526 for (i = 0; i < X86_VENDOR_NUM; i++)
527 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
528 c->x86_vendor_id[0] = 0;
529 cpu_devs[i]->c_identify(c);
530 if (c->x86_vendor_id[0]) {
539 * Do minimum CPU detection early.
540 * Fields really needed: vendor, cpuid_level, family, model, mask,
542 * The others are not touched to avoid unwanted side effects.
544 * WARNING: this function is only called on the BP. Don't add code here
545 * that is supposed to run on all CPUs.
547 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
550 c->x86_clflush_size = 64;
552 c->x86_clflush_size = 32;
554 c->x86_cache_alignment = c->x86_clflush_size;
556 memset(&c->x86_capability, 0, sizeof c->x86_capability);
557 c->extended_cpuid_level = 0;
560 identify_cpu_without_cpuid(c);
562 /* cyrix could have cpuid enabled via c_identify()*/
572 if (this_cpu->c_early_init)
573 this_cpu->c_early_init(c);
575 validate_pat_support(c);
578 c->cpu_index = boot_cpu_id;
582 void __init early_cpu_init(void)
584 struct cpu_dev **cdev;
587 printk("KERNEL supported cpus:\n");
588 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
589 struct cpu_dev *cpudev = *cdev;
592 if (count >= X86_VENDOR_NUM)
594 cpu_devs[count] = cpudev;
597 for (j = 0; j < 2; j++) {
598 if (!cpudev->c_ident[j])
600 printk(" %s %s\n", cpudev->c_vendor,
605 early_identify_cpu(&boot_cpu_data);
609 * The NOPL instruction is supposed to exist on all CPUs with
610 * family >= 6; unfortunately, that's not true in practice because
611 * of early VIA chips and (more importantly) broken virtualizers that
612 * are not easy to detect. In the latter case it doesn't even *fail*
613 * reliably, so probing for it doesn't even work. Disable it completely
614 * unless we can find a reliable way to detect all the broken cases.
616 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
618 clear_cpu_cap(c, X86_FEATURE_NOPL);
621 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
623 c->extended_cpuid_level = 0;
626 identify_cpu_without_cpuid(c);
628 /* cyrix could have cpuid enabled via c_identify()*/
638 if (c->cpuid_level >= 0x00000001) {
639 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
641 # ifdef CONFIG_X86_HT
642 c->apicid = phys_pkg_id(c->initial_apicid, 0);
644 c->apicid = c->initial_apicid;
649 c->phys_proc_id = c->initial_apicid;
653 get_model_name(c); /* Default name */
655 init_scattered_cpuid_features(c);
660 * This does the hard work of actually picking apart the CPU stuff...
662 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
666 c->loops_per_jiffy = loops_per_jiffy;
667 c->x86_cache_size = -1;
668 c->x86_vendor = X86_VENDOR_UNKNOWN;
669 c->x86_model = c->x86_mask = 0; /* So far unknown... */
670 c->x86_vendor_id[0] = '\0'; /* Unset */
671 c->x86_model_id[0] = '\0'; /* Unset */
672 c->x86_max_cores = 1;
673 c->x86_coreid_bits = 0;
675 c->x86_clflush_size = 64;
677 c->cpuid_level = -1; /* CPUID not detected */
678 c->x86_clflush_size = 32;
680 c->x86_cache_alignment = c->x86_clflush_size;
681 memset(&c->x86_capability, 0, sizeof c->x86_capability);
685 if (this_cpu->c_identify)
686 this_cpu->c_identify(c);
689 c->apicid = phys_pkg_id(0);
693 * Vendor-specific initialization. In this section we
694 * canonicalize the feature flags, meaning if there are
695 * features a certain CPU supports which CPUID doesn't
696 * tell us, CPUID claiming incorrect flags, or other bugs,
697 * we handle them here.
699 * At the end of this section, c->x86_capability better
700 * indicate the features this CPU genuinely supports!
702 if (this_cpu->c_init)
705 /* Disable the PN if appropriate */
706 squash_the_stupid_serial_number(c);
709 * The vendor-specific functions might have changed features. Now
710 * we do "generic changes."
713 /* If the model name is still unset, do table lookup. */
714 if (!c->x86_model_id[0]) {
716 p = table_lookup_model(c);
718 strcpy(c->x86_model_id, p);
721 sprintf(c->x86_model_id, "%02x/%02x",
722 c->x86, c->x86_model);
731 * On SMP, boot_cpu_data holds the common feature set between
732 * all CPUs; so make sure that we indicate which features are
733 * common between the CPUs. The first time this routine gets
734 * executed, c == &boot_cpu_data.
736 if (c != &boot_cpu_data) {
737 /* AND the already accumulated flags with these */
738 for (i = 0; i < NCAPINTS; i++)
739 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
742 /* Clear all flags overriden by options */
743 for (i = 0; i < NCAPINTS; i++)
744 c->x86_capability[i] &= ~cleared_cpu_caps[i];
746 #ifdef CONFIG_X86_MCE
747 /* Init Machine Check Exception if available. */
751 select_idle_routine(c);
753 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
754 numa_add_cpu(smp_processor_id());
759 static void vgetcpu_set_mode(void)
761 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
762 vgetcpu_mode = VGETCPU_RDTSCP;
764 vgetcpu_mode = VGETCPU_LSL;
768 void __init identify_boot_cpu(void)
770 identify_cpu(&boot_cpu_data);
779 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
781 BUG_ON(c == &boot_cpu_data);
794 static struct msr_range msr_range_array[] __cpuinitdata = {
795 { 0x00000000, 0x00000418},
796 { 0xc0000000, 0xc000040b},
797 { 0xc0010000, 0xc0010142},
798 { 0xc0011000, 0xc001103b},
801 static void __cpuinit print_cpu_msr(void)
806 unsigned index_min, index_max;
808 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
809 index_min = msr_range_array[i].min;
810 index_max = msr_range_array[i].max;
811 for (index = index_min; index < index_max; index++) {
812 if (rdmsrl_amd_safe(index, &val))
814 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
819 static int show_msr __cpuinitdata;
820 static __init int setup_show_msr(char *arg)
824 get_option(&arg, &num);
830 __setup("show_msr=", setup_show_msr);
832 static __init int setup_noclflush(char *arg)
834 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
837 __setup("noclflush", setup_noclflush);
839 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
843 if (c->x86_vendor < X86_VENDOR_NUM)
844 vendor = this_cpu->c_vendor;
845 else if (c->cpuid_level >= 0)
846 vendor = c->x86_vendor_id;
848 if (vendor && !strstr(c->x86_model_id, vendor))
849 printk(KERN_CONT "%s ", vendor);
851 if (c->x86_model_id[0])
852 printk(KERN_CONT "%s", c->x86_model_id);
854 printk(KERN_CONT "%d86", c->x86);
856 if (c->x86_mask || c->cpuid_level >= 0)
857 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
859 printk(KERN_CONT "\n");
862 if (c->cpu_index < show_msr)
870 static __init int setup_disablecpuid(char *arg)
873 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
874 setup_clear_cpu_cap(bit);
879 __setup("clearcpuid=", setup_disablecpuid);
882 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
884 DEFINE_PER_CPU_PAGE_ALIGNED(char[IRQ_STACK_SIZE], irq_stack);
886 DEFINE_PER_CPU(char *, irq_stack_ptr); /* will be set during per cpu init */
888 DEFINE_PER_CPU(char *, irq_stack_ptr) =
889 per_cpu_var(irq_stack) + IRQ_STACK_SIZE - 64;
892 void __cpuinit pda_init(int cpu)
894 struct x8664_pda *pda = cpu_pda(cpu);
896 /* Setup up data that may be needed in __get_free_pages early */
900 load_pda_offset(cpu);
902 pda->cpunumber = cpu;
904 pda->kernelstack = (unsigned long)stack_thread_info() -
905 PDA_STACKOFFSET + THREAD_SIZE;
908 /* others are initialized in smpboot.c */
909 pda->pcurrent = &init_task;
911 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
912 pda->nodenumber = cpu_to_node(cpu);
916 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
917 [(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ])
918 __aligned(PAGE_SIZE);
920 extern asmlinkage void ignore_sysret(void);
922 /* May not be marked __init: used by software suspend */
923 void syscall_init(void)
926 * LSTAR and STAR live in a bit strange symbiosis.
927 * They both write to the same internal register. STAR allows to
928 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
930 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
931 wrmsrl(MSR_LSTAR, system_call);
932 wrmsrl(MSR_CSTAR, ignore_sysret);
934 #ifdef CONFIG_IA32_EMULATION
935 syscall32_cpu_init();
938 /* Flags to clear on syscall */
939 wrmsrl(MSR_SYSCALL_MASK,
940 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
943 unsigned long kernel_eflags;
946 * Copies of the original ist values from the tss are only accessed during
947 * debugging, no special alignment required.
949 DEFINE_PER_CPU(struct orig_ist, orig_ist);
953 /* Make sure %fs is initialized properly in idle threads */
954 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
956 memset(regs, 0, sizeof(struct pt_regs));
957 regs->fs = __KERNEL_PERCPU;
963 * cpu_init() initializes state that is per-CPU. Some data is already
964 * initialized (naturally) in the bootstrap process, such as the GDT
965 * and IDT. We reload them nevertheless, this function acts as a
966 * 'CPU state barrier', nothing should get across.
967 * A lot of state is already set up in PDA init for 64 bit
970 void __cpuinit cpu_init(void)
972 int cpu = stack_smp_processor_id();
973 struct tss_struct *t = &per_cpu(init_tss, cpu);
974 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
976 struct task_struct *me;
979 /* CPU 0 is initialised in head64.c */
985 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
986 panic("CPU#%d already initialized!\n", cpu);
988 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
990 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
993 * Initialize the per-CPU GDT with the boot GDT,
994 * and set up the GDT descriptor:
998 load_idt((const struct desc_ptr *)&idt_descr);
1000 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1003 wrmsrl(MSR_FS_BASE, 0);
1004 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1008 if (cpu != 0 && x2apic)
1012 * set up and load the per-CPU TSS
1014 if (!orig_ist->ist[0]) {
1015 static const unsigned int sizes[N_EXCEPTION_STACKS] = {
1016 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STKSZ,
1017 [DEBUG_STACK - 1] = DEBUG_STKSZ
1019 char *estacks = per_cpu(exception_stacks, cpu);
1020 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1021 estacks += sizes[v];
1022 orig_ist->ist[v] = t->x86_tss.ist[v] =
1023 (unsigned long)estacks;
1027 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1029 * <= is required because the CPU will access up to
1030 * 8 bits beyond the end of the IO permission bitmap.
1032 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1033 t->io_bitmap[i] = ~0UL;
1035 atomic_inc(&init_mm.mm_count);
1036 me->active_mm = &init_mm;
1039 enter_lazy_tlb(&init_mm, me);
1041 load_sp0(t, ¤t->thread);
1042 set_tss_desc(cpu, t);
1044 load_LDT(&init_mm.context);
1048 * If the kgdb is connected no debug regs should be altered. This
1049 * is only applicable when KGDB and a KGDB I/O module are built
1050 * into the kernel and you are using early debugging with
1051 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1053 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1054 arch_kgdb_ops.correct_hw_break();
1058 * Clear all 6 debug registers:
1061 set_debugreg(0UL, 0);
1062 set_debugreg(0UL, 1);
1063 set_debugreg(0UL, 2);
1064 set_debugreg(0UL, 3);
1065 set_debugreg(0UL, 6);
1066 set_debugreg(0UL, 7);
1068 /* If the kgdb is connected no debug regs should be altered. */
1074 raw_local_save_flags(kernel_eflags);
1082 void __cpuinit cpu_init(void)
1084 int cpu = smp_processor_id();
1085 struct task_struct *curr = current;
1086 struct tss_struct *t = &per_cpu(init_tss, cpu);
1087 struct thread_struct *thread = &curr->thread;
1089 if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1090 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1091 for (;;) local_irq_enable();
1094 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1096 if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1097 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1099 load_idt(&idt_descr);
1100 switch_to_new_gdt();
1103 * Set up and load the per-CPU TSS and LDT
1105 atomic_inc(&init_mm.mm_count);
1106 curr->active_mm = &init_mm;
1109 enter_lazy_tlb(&init_mm, curr);
1111 load_sp0(t, thread);
1112 set_tss_desc(cpu, t);
1114 load_LDT(&init_mm.context);
1116 #ifdef CONFIG_DOUBLEFAULT
1117 /* Set up doublefault TSS pointer in the GDT */
1118 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1122 asm volatile ("mov %0, %%gs" : : "r" (0));
1124 /* Clear all 6 debug registers: */
1133 * Force FPU initialization:
1136 current_thread_info()->status = TS_XSAVE;
1138 current_thread_info()->status = 0;
1140 mxcsr_feature_mask_init();
1143 * Boot processor to setup the FP and extended state context info.
1145 if (smp_processor_id() == boot_cpu_id)
1146 init_thread_xstate();