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>
23 #ifdef CONFIG_X86_LOCAL_APIC
24 #include <asm/mpspec.h>
26 #include <mach_apic.h>
27 #include <asm/genapic.h>
31 #include <asm/pgtable.h>
32 #include <asm/processor.h>
34 #include <asm/atomic.h>
35 #include <asm/proto.h>
36 #include <asm/sections.h>
37 #include <asm/setup.h>
41 static struct cpu_dev *this_cpu __cpuinitdata;
44 /* We need valid kernel segments for data and code in long mode too
45 * IRET will check the segment types kkeil 2000/10/28
46 * Also sysret mandates a special GDT layout
48 /* The TLS descriptors are currently at a different place compared to i386.
49 Hopefully nobody expects them at a fixed place (Wine?) */
50 DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
51 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
52 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
53 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
54 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
55 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
56 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
59 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
60 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } },
61 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } },
62 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } },
63 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff200 } } },
65 * Segments used for calling PnP BIOS have byte granularity.
66 * They code segments and data segments have fixed 64k limits,
67 * the transfer segment sizes are set at run time.
70 [GDT_ENTRY_PNPBIOS_CS32] = { { { 0x0000ffff, 0x00409a00 } } },
72 [GDT_ENTRY_PNPBIOS_CS16] = { { { 0x0000ffff, 0x00009a00 } } },
74 [GDT_ENTRY_PNPBIOS_DS] = { { { 0x0000ffff, 0x00009200 } } },
76 [GDT_ENTRY_PNPBIOS_TS1] = { { { 0x00000000, 0x00009200 } } },
78 [GDT_ENTRY_PNPBIOS_TS2] = { { { 0x00000000, 0x00009200 } } },
80 * The APM segments have byte granularity and their bases
81 * are set at run time. All have 64k limits.
84 [GDT_ENTRY_APMBIOS_BASE] = { { { 0x0000ffff, 0x00409a00 } } },
86 [GDT_ENTRY_APMBIOS_BASE+1] = { { { 0x0000ffff, 0x00009a00 } } },
88 [GDT_ENTRY_APMBIOS_BASE+2] = { { { 0x0000ffff, 0x00409200 } } },
90 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } },
91 [GDT_ENTRY_PERCPU] = { { { 0x00000000, 0x00000000 } } },
94 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
97 static int cachesize_override __cpuinitdata = -1;
98 static int disable_x86_serial_nr __cpuinitdata = 1;
100 static int __init cachesize_setup(char *str)
102 get_option(&str, &cachesize_override);
105 __setup("cachesize=", cachesize_setup);
107 static int __init x86_fxsr_setup(char *s)
109 setup_clear_cpu_cap(X86_FEATURE_FXSR);
110 setup_clear_cpu_cap(X86_FEATURE_XMM);
113 __setup("nofxsr", x86_fxsr_setup);
115 static int __init x86_sep_setup(char *s)
117 setup_clear_cpu_cap(X86_FEATURE_SEP);
120 __setup("nosep", x86_sep_setup);
122 /* Standard macro to see if a specific flag is changeable */
123 static inline int flag_is_changeable_p(u32 flag)
137 : "=&r" (f1), "=&r" (f2)
140 return ((f1^f2) & flag) != 0;
143 /* Probe for the CPUID instruction */
144 static int __cpuinit have_cpuid_p(void)
146 return flag_is_changeable_p(X86_EFLAGS_ID);
149 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
151 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
152 /* Disable processor serial number */
153 unsigned long lo, hi;
154 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
156 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
157 printk(KERN_NOTICE "CPU serial number disabled.\n");
158 clear_cpu_cap(c, X86_FEATURE_PN);
160 /* Disabling the serial number may affect the cpuid level */
161 c->cpuid_level = cpuid_eax(0);
165 static int __init x86_serial_nr_setup(char *s)
167 disable_x86_serial_nr = 0;
170 __setup("serialnumber", x86_serial_nr_setup);
172 static inline int flag_is_changeable_p(u32 flag)
176 /* Probe for the CPUID instruction */
177 static inline int have_cpuid_p(void)
181 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
187 * Naming convention should be: <Name> [(<Codename>)]
188 * This table only is used unless init_<vendor>() below doesn't set it;
189 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
193 /* Look up CPU names by table lookup. */
194 static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
196 struct cpu_model_info *info;
198 if (c->x86_model >= 16)
199 return NULL; /* Range check */
204 info = this_cpu->c_models;
206 while (info && info->family) {
207 if (info->family == c->x86)
208 return info->model_names[c->x86_model];
211 return NULL; /* Not found */
214 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
216 /* Current gdt points %fs at the "master" per-cpu area: after this,
217 * it's on the real one. */
218 void switch_to_new_gdt(void)
220 struct desc_ptr gdt_descr;
222 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
223 gdt_descr.size = GDT_SIZE - 1;
224 load_gdt(&gdt_descr);
226 asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
230 static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
232 static void __cpuinit default_init(struct cpuinfo_x86 *c)
235 display_cacheinfo(c);
237 /* Not much we can do here... */
238 /* Check if at least it has cpuid */
239 if (c->cpuid_level == -1) {
240 /* No cpuid. It must be an ancient CPU */
242 strcpy(c->x86_model_id, "486");
243 else if (c->x86 == 3)
244 strcpy(c->x86_model_id, "386");
249 static struct cpu_dev __cpuinitdata default_cpu = {
250 .c_init = default_init,
251 .c_vendor = "Unknown",
252 .c_x86_vendor = X86_VENDOR_UNKNOWN,
255 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
260 if (c->extended_cpuid_level < 0x80000004)
263 v = (unsigned int *) c->x86_model_id;
264 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
265 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
266 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
267 c->x86_model_id[48] = 0;
269 /* Intel chips right-justify this string for some dumb reason;
270 undo that brain damage */
271 p = q = &c->x86_model_id[0];
277 while (q <= &c->x86_model_id[48])
278 *q++ = '\0'; /* Zero-pad the rest */
282 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
284 unsigned int n, dummy, ebx, ecx, edx, l2size;
286 n = c->extended_cpuid_level;
288 if (n >= 0x80000005) {
289 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
290 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
291 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
292 c->x86_cache_size = (ecx>>24) + (edx>>24);
294 /* On K8 L1 TLB is inclusive, so don't count it */
299 if (n < 0x80000006) /* Some chips just has a large L1. */
302 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
306 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
308 /* do processor-specific cache resizing */
309 if (this_cpu->c_size_cache)
310 l2size = this_cpu->c_size_cache(c, l2size);
312 /* Allow user to override all this if necessary. */
313 if (cachesize_override != -1)
314 l2size = cachesize_override;
317 return; /* Again, no L2 cache is possible */
320 c->x86_cache_size = l2size;
322 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
326 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
329 u32 eax, ebx, ecx, edx;
330 int index_msb, core_bits;
332 if (!cpu_has(c, X86_FEATURE_HT))
335 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
338 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
341 cpuid(1, &eax, &ebx, &ecx, &edx);
343 smp_num_siblings = (ebx & 0xff0000) >> 16;
345 if (smp_num_siblings == 1) {
346 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
347 } else if (smp_num_siblings > 1) {
349 if (smp_num_siblings > NR_CPUS) {
350 printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
352 smp_num_siblings = 1;
356 index_msb = get_count_order(smp_num_siblings);
358 c->phys_proc_id = phys_pkg_id(index_msb);
360 c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb);
363 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
365 index_msb = get_count_order(smp_num_siblings);
367 core_bits = get_count_order(c->x86_max_cores);
370 c->cpu_core_id = phys_pkg_id(index_msb) &
371 ((1 << core_bits) - 1);
373 c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) &
374 ((1 << core_bits) - 1);
379 if ((c->x86_max_cores * smp_num_siblings) > 1) {
380 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
382 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
388 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
390 char *v = c->x86_vendor_id;
394 for (i = 0; i < X86_VENDOR_NUM; i++) {
398 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
399 (cpu_devs[i]->c_ident[1] &&
400 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
401 this_cpu = cpu_devs[i];
402 c->x86_vendor = this_cpu->c_x86_vendor;
409 printk(KERN_ERR "CPU: vendor_id '%s' unknown, using generic init.\n", v);
410 printk(KERN_ERR "CPU: Your system may be unstable.\n");
413 c->x86_vendor = X86_VENDOR_UNKNOWN;
414 this_cpu = &default_cpu;
417 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
419 /* Get vendor name */
420 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
421 (unsigned int *)&c->x86_vendor_id[0],
422 (unsigned int *)&c->x86_vendor_id[8],
423 (unsigned int *)&c->x86_vendor_id[4]);
426 /* Intel-defined flags: level 0x00000001 */
427 if (c->cpuid_level >= 0x00000001) {
428 u32 junk, tfms, cap0, misc;
429 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
430 c->x86 = (tfms >> 8) & 0xf;
431 c->x86_model = (tfms >> 4) & 0xf;
432 c->x86_mask = tfms & 0xf;
434 c->x86 += (tfms >> 20) & 0xff;
436 c->x86_model += ((tfms >> 16) & 0xf) << 4;
437 if (cap0 & (1<<19)) {
438 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
439 c->x86_cache_alignment = c->x86_clflush_size;
444 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
449 /* Intel-defined flags: level 0x00000001 */
450 if (c->cpuid_level >= 0x00000001) {
451 u32 capability, excap;
452 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
453 c->x86_capability[0] = capability;
454 c->x86_capability[4] = excap;
457 /* AMD-defined flags: level 0x80000001 */
458 xlvl = cpuid_eax(0x80000000);
459 c->extended_cpuid_level = xlvl;
460 if ((xlvl & 0xffff0000) == 0x80000000) {
461 if (xlvl >= 0x80000001) {
462 c->x86_capability[1] = cpuid_edx(0x80000001);
463 c->x86_capability[6] = cpuid_ecx(0x80000001);
468 if (c->extended_cpuid_level >= 0x80000008) {
469 u32 eax = cpuid_eax(0x80000008);
471 c->x86_virt_bits = (eax >> 8) & 0xff;
472 c->x86_phys_bits = eax & 0xff;
476 if (c->extended_cpuid_level >= 0x80000007)
477 c->x86_power = cpuid_edx(0x80000007);
481 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
487 * First of all, decide if this is a 486 or higher
488 * It's a 486 if we can modify the AC flag
490 if (flag_is_changeable_p(X86_EFLAGS_AC))
495 for (i = 0; i < X86_VENDOR_NUM; i++)
496 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
497 c->x86_vendor_id[0] = 0;
498 cpu_devs[i]->c_identify(c);
499 if (c->x86_vendor_id[0]) {
508 * Do minimum CPU detection early.
509 * Fields really needed: vendor, cpuid_level, family, model, mask,
511 * The others are not touched to avoid unwanted side effects.
513 * WARNING: this function is only called on the BP. Don't add code here
514 * that is supposed to run on all CPUs.
516 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
519 c->x86_clflush_size = 64;
521 c->x86_clflush_size = 32;
523 c->x86_cache_alignment = c->x86_clflush_size;
525 memset(&c->x86_capability, 0, sizeof c->x86_capability);
526 c->extended_cpuid_level = 0;
529 identify_cpu_without_cpuid(c);
531 /* cyrix could have cpuid enabled via c_identify()*/
541 if (this_cpu->c_early_init)
542 this_cpu->c_early_init(c);
544 validate_pat_support(c);
547 void __init early_cpu_init(void)
549 struct cpu_dev **cdev;
552 printk("KERNEL supported cpus:\n");
553 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
554 struct cpu_dev *cpudev = *cdev;
557 if (count >= X86_VENDOR_NUM)
559 cpu_devs[count] = cpudev;
562 for (j = 0; j < 2; j++) {
563 if (!cpudev->c_ident[j])
565 printk(" %s %s\n", cpudev->c_vendor,
570 early_identify_cpu(&boot_cpu_data);
574 * The NOPL instruction is supposed to exist on all CPUs with
575 * family >= 6; unfortunately, that's not true in practice because
576 * of early VIA chips and (more importantly) broken virtualizers that
577 * are not easy to detect. In the latter case it doesn't even *fail*
578 * reliably, so probing for it doesn't even work. Disable it completely
579 * unless we can find a reliable way to detect all the broken cases.
581 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
583 clear_cpu_cap(c, X86_FEATURE_NOPL);
586 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
588 c->extended_cpuid_level = 0;
591 identify_cpu_without_cpuid(c);
593 /* cyrix could have cpuid enabled via c_identify()*/
603 if (c->cpuid_level >= 0x00000001) {
604 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
606 # ifdef CONFIG_X86_HT
607 c->apicid = phys_pkg_id(c->initial_apicid, 0);
609 c->apicid = c->initial_apicid;
614 c->phys_proc_id = c->initial_apicid;
618 get_model_name(c); /* Default name */
620 init_scattered_cpuid_features(c);
625 * This does the hard work of actually picking apart the CPU stuff...
627 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
631 c->loops_per_jiffy = loops_per_jiffy;
632 c->x86_cache_size = -1;
633 c->x86_vendor = X86_VENDOR_UNKNOWN;
634 c->x86_model = c->x86_mask = 0; /* So far unknown... */
635 c->x86_vendor_id[0] = '\0'; /* Unset */
636 c->x86_model_id[0] = '\0'; /* Unset */
637 c->x86_max_cores = 1;
638 c->x86_coreid_bits = 0;
640 c->x86_clflush_size = 64;
642 c->cpuid_level = -1; /* CPUID not detected */
643 c->x86_clflush_size = 32;
645 c->x86_cache_alignment = c->x86_clflush_size;
646 memset(&c->x86_capability, 0, sizeof c->x86_capability);
650 if (this_cpu->c_identify)
651 this_cpu->c_identify(c);
654 c->apicid = phys_pkg_id(0);
658 * Vendor-specific initialization. In this section we
659 * canonicalize the feature flags, meaning if there are
660 * features a certain CPU supports which CPUID doesn't
661 * tell us, CPUID claiming incorrect flags, or other bugs,
662 * we handle them here.
664 * At the end of this section, c->x86_capability better
665 * indicate the features this CPU genuinely supports!
667 if (this_cpu->c_init)
670 /* Disable the PN if appropriate */
671 squash_the_stupid_serial_number(c);
674 * The vendor-specific functions might have changed features. Now
675 * we do "generic changes."
678 /* If the model name is still unset, do table lookup. */
679 if (!c->x86_model_id[0]) {
681 p = table_lookup_model(c);
683 strcpy(c->x86_model_id, p);
686 sprintf(c->x86_model_id, "%02x/%02x",
687 c->x86, c->x86_model);
695 * On SMP, boot_cpu_data holds the common feature set between
696 * all CPUs; so make sure that we indicate which features are
697 * common between the CPUs. The first time this routine gets
698 * executed, c == &boot_cpu_data.
700 if (c != &boot_cpu_data) {
701 /* AND the already accumulated flags with these */
702 for (i = 0; i < NCAPINTS; i++)
703 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
706 /* Clear all flags overriden by options */
707 for (i = 0; i < NCAPINTS; i++)
708 c->x86_capability[i] &= ~cleared_cpu_caps[i];
710 #ifdef CONFIG_X86_MCE
711 /* Init Machine Check Exception if available. */
715 select_idle_routine(c);
717 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
718 numa_add_cpu(smp_processor_id());
722 void __init identify_boot_cpu(void)
724 identify_cpu(&boot_cpu_data);
731 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
733 BUG_ON(c == &boot_cpu_data);
746 static struct msr_range msr_range_array[] __cpuinitdata = {
747 { 0x00000000, 0x00000418},
748 { 0xc0000000, 0xc000040b},
749 { 0xc0010000, 0xc0010142},
750 { 0xc0011000, 0xc001103b},
753 static void __cpuinit print_cpu_msr(void)
758 unsigned index_min, index_max;
760 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
761 index_min = msr_range_array[i].min;
762 index_max = msr_range_array[i].max;
763 for (index = index_min; index < index_max; index++) {
764 if (rdmsrl_amd_safe(index, &val))
766 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
771 static int show_msr __cpuinitdata;
772 static __init int setup_show_msr(char *arg)
776 get_option(&arg, &num);
782 __setup("show_msr=", setup_show_msr);
784 static __init int setup_noclflush(char *arg)
786 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
789 __setup("noclflush", setup_noclflush);
791 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
795 if (c->x86_vendor < X86_VENDOR_NUM)
796 vendor = this_cpu->c_vendor;
797 else if (c->cpuid_level >= 0)
798 vendor = c->x86_vendor_id;
800 if (vendor && !strstr(c->x86_model_id, vendor))
801 printk(KERN_CONT "%s ", vendor);
803 if (c->x86_model_id[0])
804 printk(KERN_CONT "%s", c->x86_model_id);
806 printk(KERN_CONT "%d86", c->x86);
808 if (c->x86_mask || c->cpuid_level >= 0)
809 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
811 printk(KERN_CONT "\n");
814 if (c->cpu_index < show_msr)
822 static __init int setup_disablecpuid(char *arg)
825 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
826 setup_clear_cpu_cap(bit);
831 __setup("clearcpuid=", setup_disablecpuid);
833 cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
836 struct x8664_pda **_cpu_pda __read_mostly;
837 EXPORT_SYMBOL(_cpu_pda);
839 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
841 char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
843 void __cpuinit pda_init(int cpu)
845 struct x8664_pda *pda = cpu_pda(cpu);
847 /* Setup up data that may be needed in __get_free_pages early */
850 /* Memory clobbers used to order PDA accessed */
852 wrmsrl(MSR_GS_BASE, pda);
855 pda->cpunumber = cpu;
857 pda->kernelstack = (unsigned long)stack_thread_info() -
858 PDA_STACKOFFSET + THREAD_SIZE;
859 pda->active_mm = &init_mm;
863 /* others are initialized in smpboot.c */
864 pda->pcurrent = &init_task;
865 pda->irqstackptr = boot_cpu_stack;
866 pda->irqstackptr += IRQSTACKSIZE - 64;
868 if (!pda->irqstackptr) {
869 pda->irqstackptr = (char *)
870 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
871 if (!pda->irqstackptr)
872 panic("cannot allocate irqstack for cpu %d",
874 pda->irqstackptr += IRQSTACKSIZE - 64;
877 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
878 pda->nodenumber = cpu_to_node(cpu);
882 char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
883 DEBUG_STKSZ] __page_aligned_bss;
885 extern asmlinkage void ignore_sysret(void);
887 /* May not be marked __init: used by software suspend */
888 void syscall_init(void)
891 * LSTAR and STAR live in a bit strange symbiosis.
892 * They both write to the same internal register. STAR allows to
893 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
895 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
896 wrmsrl(MSR_LSTAR, system_call);
897 wrmsrl(MSR_CSTAR, ignore_sysret);
899 #ifdef CONFIG_IA32_EMULATION
900 syscall32_cpu_init();
903 /* Flags to clear on syscall */
904 wrmsrl(MSR_SYSCALL_MASK,
905 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
908 unsigned long kernel_eflags;
911 * Copies of the original ist values from the tss are only accessed during
912 * debugging, no special alignment required.
914 DEFINE_PER_CPU(struct orig_ist, orig_ist);
918 /* Make sure %fs is initialized properly in idle threads */
919 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
921 memset(regs, 0, sizeof(struct pt_regs));
922 regs->fs = __KERNEL_PERCPU;
928 * cpu_init() initializes state that is per-CPU. Some data is already
929 * initialized (naturally) in the bootstrap process, such as the GDT
930 * and IDT. We reload them nevertheless, this function acts as a
931 * 'CPU state barrier', nothing should get across.
932 * A lot of state is already set up in PDA init for 64 bit
935 void __cpuinit cpu_init(void)
937 int cpu = stack_smp_processor_id();
938 struct tss_struct *t = &per_cpu(init_tss, cpu);
939 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
941 char *estacks = NULL;
942 struct task_struct *me;
945 /* CPU 0 is initialised in head64.c */
949 estacks = boot_exception_stacks;
953 if (cpu_test_and_set(cpu, cpu_initialized))
954 panic("CPU#%d already initialized!\n", cpu);
956 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
958 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
961 * Initialize the per-CPU GDT with the boot GDT,
962 * and set up the GDT descriptor:
966 load_idt((const struct desc_ptr *)&idt_descr);
968 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
971 wrmsrl(MSR_FS_BASE, 0);
972 wrmsrl(MSR_KERNEL_GS_BASE, 0);
976 if (cpu != 0 && x2apic)
980 * set up and load the per-CPU TSS
982 if (!orig_ist->ist[0]) {
983 static const unsigned int order[N_EXCEPTION_STACKS] = {
984 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
985 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
987 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
989 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
991 panic("Cannot allocate exception "
992 "stack %ld %d\n", v, cpu);
994 estacks += PAGE_SIZE << order[v];
995 orig_ist->ist[v] = t->x86_tss.ist[v] =
996 (unsigned long)estacks;
1000 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1002 * <= is required because the CPU will access up to
1003 * 8 bits beyond the end of the IO permission bitmap.
1005 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1006 t->io_bitmap[i] = ~0UL;
1008 atomic_inc(&init_mm.mm_count);
1009 me->active_mm = &init_mm;
1012 enter_lazy_tlb(&init_mm, me);
1014 load_sp0(t, ¤t->thread);
1015 set_tss_desc(cpu, t);
1017 load_LDT(&init_mm.context);
1021 * If the kgdb is connected no debug regs should be altered. This
1022 * is only applicable when KGDB and a KGDB I/O module are built
1023 * into the kernel and you are using early debugging with
1024 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1026 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1027 arch_kgdb_ops.correct_hw_break();
1031 * Clear all 6 debug registers:
1034 set_debugreg(0UL, 0);
1035 set_debugreg(0UL, 1);
1036 set_debugreg(0UL, 2);
1037 set_debugreg(0UL, 3);
1038 set_debugreg(0UL, 6);
1039 set_debugreg(0UL, 7);
1041 /* If the kgdb is connected no debug regs should be altered. */
1047 raw_local_save_flags(kernel_eflags);
1055 void __cpuinit cpu_init(void)
1057 int cpu = smp_processor_id();
1058 struct task_struct *curr = current;
1059 struct tss_struct *t = &per_cpu(init_tss, cpu);
1060 struct thread_struct *thread = &curr->thread;
1062 if (cpu_test_and_set(cpu, cpu_initialized)) {
1063 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1064 for (;;) local_irq_enable();
1067 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1069 if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1070 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1072 load_idt(&idt_descr);
1073 switch_to_new_gdt();
1076 * Set up and load the per-CPU TSS and LDT
1078 atomic_inc(&init_mm.mm_count);
1079 curr->active_mm = &init_mm;
1082 enter_lazy_tlb(&init_mm, curr);
1084 load_sp0(t, thread);
1085 set_tss_desc(cpu, t);
1087 load_LDT(&init_mm.context);
1089 #ifdef CONFIG_DOUBLEFAULT
1090 /* Set up doublefault TSS pointer in the GDT */
1091 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1095 asm volatile ("mov %0, %%gs" : : "r" (0));
1097 /* Clear all 6 debug registers: */
1106 * Force FPU initialization:
1109 current_thread_info()->status = TS_XSAVE;
1111 current_thread_info()->status = 0;
1113 mxcsr_feature_mask_init();
1116 * Boot processor to setup the FP and extended state context info.
1118 if (!smp_processor_id())
1119 init_thread_xstate();