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>
24 #ifdef CONFIG_X86_LOCAL_APIC
25 #include <asm/mpspec.h>
27 #include <mach_apic.h>
28 #include <asm/genapic.h>
32 #include <asm/pgtable.h>
33 #include <asm/processor.h>
35 #include <asm/atomic.h>
36 #include <asm/proto.h>
37 #include <asm/sections.h>
38 #include <asm/setup.h>
39 #include <asm/hypervisor.h>
43 static struct cpu_dev *this_cpu __cpuinitdata;
46 /* We need valid kernel segments for data and code in long mode too
47 * IRET will check the segment types kkeil 2000/10/28
48 * Also sysret mandates a special GDT layout
50 /* The TLS descriptors are currently at a different place compared to i386.
51 Hopefully nobody expects them at a fixed place (Wine?) */
52 DEFINE_PER_CPU(struct gdt_page, gdt_page) = { .gdt = {
53 [GDT_ENTRY_KERNEL32_CS] = { { { 0x0000ffff, 0x00cf9b00 } } },
54 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00af9b00 } } },
55 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9300 } } },
56 [GDT_ENTRY_DEFAULT_USER32_CS] = { { { 0x0000ffff, 0x00cffb00 } } },
57 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff300 } } },
58 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00affb00 } } },
61 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
62 [GDT_ENTRY_KERNEL_CS] = { { { 0x0000ffff, 0x00cf9a00 } } },
63 [GDT_ENTRY_KERNEL_DS] = { { { 0x0000ffff, 0x00cf9200 } } },
64 [GDT_ENTRY_DEFAULT_USER_CS] = { { { 0x0000ffff, 0x00cffa00 } } },
65 [GDT_ENTRY_DEFAULT_USER_DS] = { { { 0x0000ffff, 0x00cff200 } } },
67 * Segments used for calling PnP BIOS have byte granularity.
68 * They code segments and data segments have fixed 64k limits,
69 * the transfer segment sizes are set at run time.
72 [GDT_ENTRY_PNPBIOS_CS32] = { { { 0x0000ffff, 0x00409a00 } } },
74 [GDT_ENTRY_PNPBIOS_CS16] = { { { 0x0000ffff, 0x00009a00 } } },
76 [GDT_ENTRY_PNPBIOS_DS] = { { { 0x0000ffff, 0x00009200 } } },
78 [GDT_ENTRY_PNPBIOS_TS1] = { { { 0x00000000, 0x00009200 } } },
80 [GDT_ENTRY_PNPBIOS_TS2] = { { { 0x00000000, 0x00009200 } } },
82 * The APM segments have byte granularity and their bases
83 * are set at run time. All have 64k limits.
86 [GDT_ENTRY_APMBIOS_BASE] = { { { 0x0000ffff, 0x00409a00 } } },
88 [GDT_ENTRY_APMBIOS_BASE+1] = { { { 0x0000ffff, 0x00009a00 } } },
90 [GDT_ENTRY_APMBIOS_BASE+2] = { { { 0x0000ffff, 0x00409200 } } },
92 [GDT_ENTRY_ESPFIX_SS] = { { { 0x00000000, 0x00c09200 } } },
93 [GDT_ENTRY_PERCPU] = { { { 0x00000000, 0x00000000 } } },
96 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
99 static int cachesize_override __cpuinitdata = -1;
100 static int disable_x86_serial_nr __cpuinitdata = 1;
102 static int __init cachesize_setup(char *str)
104 get_option(&str, &cachesize_override);
107 __setup("cachesize=", cachesize_setup);
109 static int __init x86_fxsr_setup(char *s)
111 setup_clear_cpu_cap(X86_FEATURE_FXSR);
112 setup_clear_cpu_cap(X86_FEATURE_XMM);
115 __setup("nofxsr", x86_fxsr_setup);
117 static int __init x86_sep_setup(char *s)
119 setup_clear_cpu_cap(X86_FEATURE_SEP);
122 __setup("nosep", x86_sep_setup);
124 /* Standard macro to see if a specific flag is changeable */
125 static inline int flag_is_changeable_p(u32 flag)
130 * Cyrix and IDT cpus allow disabling of CPUID
131 * so the code below may return different results
132 * when it is executed before and after enabling
133 * the CPUID. Add "volatile" to not allow gcc to
134 * optimize the subsequent calls to this function.
136 asm volatile ("pushfl\n\t"
146 : "=&r" (f1), "=&r" (f2)
149 return ((f1^f2) & flag) != 0;
152 /* Probe for the CPUID instruction */
153 static int __cpuinit have_cpuid_p(void)
155 return flag_is_changeable_p(X86_EFLAGS_ID);
158 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
160 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
161 /* Disable processor serial number */
162 unsigned long lo, hi;
163 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
165 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
166 printk(KERN_NOTICE "CPU serial number disabled.\n");
167 clear_cpu_cap(c, X86_FEATURE_PN);
169 /* Disabling the serial number may affect the cpuid level */
170 c->cpuid_level = cpuid_eax(0);
174 static int __init x86_serial_nr_setup(char *s)
176 disable_x86_serial_nr = 0;
179 __setup("serialnumber", x86_serial_nr_setup);
181 static inline int flag_is_changeable_p(u32 flag)
185 /* Probe for the CPUID instruction */
186 static inline int have_cpuid_p(void)
190 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
196 * Naming convention should be: <Name> [(<Codename>)]
197 * This table only is used unless init_<vendor>() below doesn't set it;
198 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
202 /* Look up CPU names by table lookup. */
203 static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
205 struct cpu_model_info *info;
207 if (c->x86_model >= 16)
208 return NULL; /* Range check */
213 info = this_cpu->c_models;
215 while (info && info->family) {
216 if (info->family == c->x86)
217 return info->model_names[c->x86_model];
220 return NULL; /* Not found */
223 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
225 /* Current gdt points %fs at the "master" per-cpu area: after this,
226 * it's on the real one. */
227 void switch_to_new_gdt(void)
229 struct desc_ptr gdt_descr;
231 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
232 gdt_descr.size = GDT_SIZE - 1;
233 load_gdt(&gdt_descr);
235 asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
239 static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
241 static void __cpuinit default_init(struct cpuinfo_x86 *c)
244 display_cacheinfo(c);
246 /* Not much we can do here... */
247 /* Check if at least it has cpuid */
248 if (c->cpuid_level == -1) {
249 /* No cpuid. It must be an ancient CPU */
251 strcpy(c->x86_model_id, "486");
252 else if (c->x86 == 3)
253 strcpy(c->x86_model_id, "386");
258 static struct cpu_dev __cpuinitdata default_cpu = {
259 .c_init = default_init,
260 .c_vendor = "Unknown",
261 .c_x86_vendor = X86_VENDOR_UNKNOWN,
264 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
269 if (c->extended_cpuid_level < 0x80000004)
272 v = (unsigned int *) c->x86_model_id;
273 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
274 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
275 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
276 c->x86_model_id[48] = 0;
278 /* Intel chips right-justify this string for some dumb reason;
279 undo that brain damage */
280 p = q = &c->x86_model_id[0];
286 while (q <= &c->x86_model_id[48])
287 *q++ = '\0'; /* Zero-pad the rest */
291 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
293 unsigned int n, dummy, ebx, ecx, edx, l2size;
295 n = c->extended_cpuid_level;
297 if (n >= 0x80000005) {
298 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
299 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
300 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
301 c->x86_cache_size = (ecx>>24) + (edx>>24);
303 /* On K8 L1 TLB is inclusive, so don't count it */
308 if (n < 0x80000006) /* Some chips just has a large L1. */
311 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
315 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
317 /* do processor-specific cache resizing */
318 if (this_cpu->c_size_cache)
319 l2size = this_cpu->c_size_cache(c, l2size);
321 /* Allow user to override all this if necessary. */
322 if (cachesize_override != -1)
323 l2size = cachesize_override;
326 return; /* Again, no L2 cache is possible */
329 c->x86_cache_size = l2size;
331 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
335 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
338 u32 eax, ebx, ecx, edx;
339 int index_msb, core_bits;
341 if (!cpu_has(c, X86_FEATURE_HT))
344 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
347 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
350 cpuid(1, &eax, &ebx, &ecx, &edx);
352 smp_num_siblings = (ebx & 0xff0000) >> 16;
354 if (smp_num_siblings == 1) {
355 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
356 } else if (smp_num_siblings > 1) {
358 if (smp_num_siblings > nr_cpu_ids) {
359 printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
361 smp_num_siblings = 1;
365 index_msb = get_count_order(smp_num_siblings);
367 c->phys_proc_id = phys_pkg_id(index_msb);
369 c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb);
372 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
374 index_msb = get_count_order(smp_num_siblings);
376 core_bits = get_count_order(c->x86_max_cores);
379 c->cpu_core_id = phys_pkg_id(index_msb) &
380 ((1 << core_bits) - 1);
382 c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) &
383 ((1 << core_bits) - 1);
388 if ((c->x86_max_cores * smp_num_siblings) > 1) {
389 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
391 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
397 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
399 char *v = c->x86_vendor_id;
403 for (i = 0; i < X86_VENDOR_NUM; i++) {
407 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
408 (cpu_devs[i]->c_ident[1] &&
409 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
410 this_cpu = cpu_devs[i];
411 c->x86_vendor = this_cpu->c_x86_vendor;
418 printk(KERN_ERR "CPU: vendor_id '%s' unknown, using generic init.\n", v);
419 printk(KERN_ERR "CPU: Your system may be unstable.\n");
422 c->x86_vendor = X86_VENDOR_UNKNOWN;
423 this_cpu = &default_cpu;
426 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
428 /* Get vendor name */
429 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
430 (unsigned int *)&c->x86_vendor_id[0],
431 (unsigned int *)&c->x86_vendor_id[8],
432 (unsigned int *)&c->x86_vendor_id[4]);
435 /* Intel-defined flags: level 0x00000001 */
436 if (c->cpuid_level >= 0x00000001) {
437 u32 junk, tfms, cap0, misc;
438 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
439 c->x86 = (tfms >> 8) & 0xf;
440 c->x86_model = (tfms >> 4) & 0xf;
441 c->x86_mask = tfms & 0xf;
443 c->x86 += (tfms >> 20) & 0xff;
445 c->x86_model += ((tfms >> 16) & 0xf) << 4;
446 if (cap0 & (1<<19)) {
447 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
448 c->x86_cache_alignment = c->x86_clflush_size;
453 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
458 /* Intel-defined flags: level 0x00000001 */
459 if (c->cpuid_level >= 0x00000001) {
460 u32 capability, excap;
461 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
462 c->x86_capability[0] = capability;
463 c->x86_capability[4] = excap;
466 /* AMD-defined flags: level 0x80000001 */
467 xlvl = cpuid_eax(0x80000000);
468 c->extended_cpuid_level = xlvl;
469 if ((xlvl & 0xffff0000) == 0x80000000) {
470 if (xlvl >= 0x80000001) {
471 c->x86_capability[1] = cpuid_edx(0x80000001);
472 c->x86_capability[6] = cpuid_ecx(0x80000001);
477 if (c->extended_cpuid_level >= 0x80000008) {
478 u32 eax = cpuid_eax(0x80000008);
480 c->x86_virt_bits = (eax >> 8) & 0xff;
481 c->x86_phys_bits = eax & 0xff;
485 if (c->extended_cpuid_level >= 0x80000007)
486 c->x86_power = cpuid_edx(0x80000007);
490 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
496 * First of all, decide if this is a 486 or higher
497 * It's a 486 if we can modify the AC flag
499 if (flag_is_changeable_p(X86_EFLAGS_AC))
504 for (i = 0; i < X86_VENDOR_NUM; i++)
505 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
506 c->x86_vendor_id[0] = 0;
507 cpu_devs[i]->c_identify(c);
508 if (c->x86_vendor_id[0]) {
517 * Do minimum CPU detection early.
518 * Fields really needed: vendor, cpuid_level, family, model, mask,
520 * The others are not touched to avoid unwanted side effects.
522 * WARNING: this function is only called on the BP. Don't add code here
523 * that is supposed to run on all CPUs.
525 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
528 c->x86_clflush_size = 64;
530 c->x86_clflush_size = 32;
532 c->x86_cache_alignment = c->x86_clflush_size;
534 memset(&c->x86_capability, 0, sizeof c->x86_capability);
535 c->extended_cpuid_level = 0;
538 identify_cpu_without_cpuid(c);
540 /* cyrix could have cpuid enabled via c_identify()*/
550 if (this_cpu->c_early_init)
551 this_cpu->c_early_init(c);
553 validate_pat_support(c);
556 c->cpu_index = boot_cpu_id;
560 void __init early_cpu_init(void)
562 struct cpu_dev **cdev;
565 printk("KERNEL supported cpus:\n");
566 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
567 struct cpu_dev *cpudev = *cdev;
570 if (count >= X86_VENDOR_NUM)
572 cpu_devs[count] = cpudev;
575 for (j = 0; j < 2; j++) {
576 if (!cpudev->c_ident[j])
578 printk(" %s %s\n", cpudev->c_vendor,
583 early_identify_cpu(&boot_cpu_data);
587 * The NOPL instruction is supposed to exist on all CPUs with
588 * family >= 6; unfortunately, that's not true in practice because
589 * of early VIA chips and (more importantly) broken virtualizers that
590 * are not easy to detect. In the latter case it doesn't even *fail*
591 * reliably, so probing for it doesn't even work. Disable it completely
592 * unless we can find a reliable way to detect all the broken cases.
594 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
596 clear_cpu_cap(c, X86_FEATURE_NOPL);
599 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
601 c->extended_cpuid_level = 0;
604 identify_cpu_without_cpuid(c);
606 /* cyrix could have cpuid enabled via c_identify()*/
616 if (c->cpuid_level >= 0x00000001) {
617 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
619 # ifdef CONFIG_X86_HT
620 c->apicid = phys_pkg_id(c->initial_apicid, 0);
622 c->apicid = c->initial_apicid;
627 c->phys_proc_id = c->initial_apicid;
631 get_model_name(c); /* Default name */
633 init_scattered_cpuid_features(c);
638 * This does the hard work of actually picking apart the CPU stuff...
640 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
644 c->loops_per_jiffy = loops_per_jiffy;
645 c->x86_cache_size = -1;
646 c->x86_vendor = X86_VENDOR_UNKNOWN;
647 c->x86_model = c->x86_mask = 0; /* So far unknown... */
648 c->x86_vendor_id[0] = '\0'; /* Unset */
649 c->x86_model_id[0] = '\0'; /* Unset */
650 c->x86_max_cores = 1;
651 c->x86_coreid_bits = 0;
653 c->x86_clflush_size = 64;
655 c->cpuid_level = -1; /* CPUID not detected */
656 c->x86_clflush_size = 32;
658 c->x86_cache_alignment = c->x86_clflush_size;
659 memset(&c->x86_capability, 0, sizeof c->x86_capability);
663 if (this_cpu->c_identify)
664 this_cpu->c_identify(c);
667 c->apicid = phys_pkg_id(0);
671 * Vendor-specific initialization. In this section we
672 * canonicalize the feature flags, meaning if there are
673 * features a certain CPU supports which CPUID doesn't
674 * tell us, CPUID claiming incorrect flags, or other bugs,
675 * we handle them here.
677 * At the end of this section, c->x86_capability better
678 * indicate the features this CPU genuinely supports!
680 if (this_cpu->c_init)
683 /* Disable the PN if appropriate */
684 squash_the_stupid_serial_number(c);
687 * The vendor-specific functions might have changed features. Now
688 * we do "generic changes."
691 /* If the model name is still unset, do table lookup. */
692 if (!c->x86_model_id[0]) {
694 p = table_lookup_model(c);
696 strcpy(c->x86_model_id, p);
699 sprintf(c->x86_model_id, "%02x/%02x",
700 c->x86, c->x86_model);
709 * On SMP, boot_cpu_data holds the common feature set between
710 * all CPUs; so make sure that we indicate which features are
711 * common between the CPUs. The first time this routine gets
712 * executed, c == &boot_cpu_data.
714 if (c != &boot_cpu_data) {
715 /* AND the already accumulated flags with these */
716 for (i = 0; i < NCAPINTS; i++)
717 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
720 /* Clear all flags overriden by options */
721 for (i = 0; i < NCAPINTS; i++)
722 c->x86_capability[i] &= ~cleared_cpu_caps[i];
724 #ifdef CONFIG_X86_MCE
725 /* Init Machine Check Exception if available. */
729 select_idle_routine(c);
731 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
732 numa_add_cpu(smp_processor_id());
737 static void vgetcpu_set_mode(void)
739 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
740 vgetcpu_mode = VGETCPU_RDTSCP;
742 vgetcpu_mode = VGETCPU_LSL;
746 void __init identify_boot_cpu(void)
748 identify_cpu(&boot_cpu_data);
757 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
759 BUG_ON(c == &boot_cpu_data);
772 static struct msr_range msr_range_array[] __cpuinitdata = {
773 { 0x00000000, 0x00000418},
774 { 0xc0000000, 0xc000040b},
775 { 0xc0010000, 0xc0010142},
776 { 0xc0011000, 0xc001103b},
779 static void __cpuinit print_cpu_msr(void)
784 unsigned index_min, index_max;
786 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
787 index_min = msr_range_array[i].min;
788 index_max = msr_range_array[i].max;
789 for (index = index_min; index < index_max; index++) {
790 if (rdmsrl_amd_safe(index, &val))
792 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
797 static int show_msr __cpuinitdata;
798 static __init int setup_show_msr(char *arg)
802 get_option(&arg, &num);
808 __setup("show_msr=", setup_show_msr);
810 static __init int setup_noclflush(char *arg)
812 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
815 __setup("noclflush", setup_noclflush);
817 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
821 if (c->x86_vendor < X86_VENDOR_NUM)
822 vendor = this_cpu->c_vendor;
823 else if (c->cpuid_level >= 0)
824 vendor = c->x86_vendor_id;
826 if (vendor && !strstr(c->x86_model_id, vendor))
827 printk(KERN_CONT "%s ", vendor);
829 if (c->x86_model_id[0])
830 printk(KERN_CONT "%s", c->x86_model_id);
832 printk(KERN_CONT "%d86", c->x86);
834 if (c->x86_mask || c->cpuid_level >= 0)
835 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
837 printk(KERN_CONT "\n");
840 if (c->cpu_index < show_msr)
848 static __init int setup_disablecpuid(char *arg)
851 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
852 setup_clear_cpu_cap(bit);
857 __setup("clearcpuid=", setup_disablecpuid);
859 cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
862 struct x8664_pda **_cpu_pda __read_mostly;
863 EXPORT_SYMBOL(_cpu_pda);
865 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
867 static char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
869 void __cpuinit pda_init(int cpu)
871 struct x8664_pda *pda = cpu_pda(cpu);
873 /* Setup up data that may be needed in __get_free_pages early */
876 /* Memory clobbers used to order PDA accessed */
878 wrmsrl(MSR_GS_BASE, pda);
881 pda->cpunumber = cpu;
883 pda->kernelstack = (unsigned long)stack_thread_info() -
884 PDA_STACKOFFSET + THREAD_SIZE;
885 pda->active_mm = &init_mm;
889 /* others are initialized in smpboot.c */
890 pda->pcurrent = &init_task;
891 pda->irqstackptr = boot_cpu_stack;
892 pda->irqstackptr += IRQSTACKSIZE - 64;
894 if (!pda->irqstackptr) {
895 pda->irqstackptr = (char *)
896 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
897 if (!pda->irqstackptr)
898 panic("cannot allocate irqstack for cpu %d",
900 pda->irqstackptr += IRQSTACKSIZE - 64;
903 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
904 pda->nodenumber = cpu_to_node(cpu);
908 static char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
909 DEBUG_STKSZ] __page_aligned_bss;
911 extern asmlinkage void ignore_sysret(void);
913 /* May not be marked __init: used by software suspend */
914 void syscall_init(void)
917 * LSTAR and STAR live in a bit strange symbiosis.
918 * They both write to the same internal register. STAR allows to
919 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
921 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
922 wrmsrl(MSR_LSTAR, system_call);
923 wrmsrl(MSR_CSTAR, ignore_sysret);
925 #ifdef CONFIG_IA32_EMULATION
926 syscall32_cpu_init();
929 /* Flags to clear on syscall */
930 wrmsrl(MSR_SYSCALL_MASK,
931 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
934 unsigned long kernel_eflags;
937 * Copies of the original ist values from the tss are only accessed during
938 * debugging, no special alignment required.
940 DEFINE_PER_CPU(struct orig_ist, orig_ist);
944 /* Make sure %fs is initialized properly in idle threads */
945 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
947 memset(regs, 0, sizeof(struct pt_regs));
948 regs->fs = __KERNEL_PERCPU;
954 * cpu_init() initializes state that is per-CPU. Some data is already
955 * initialized (naturally) in the bootstrap process, such as the GDT
956 * and IDT. We reload them nevertheless, this function acts as a
957 * 'CPU state barrier', nothing should get across.
958 * A lot of state is already set up in PDA init for 64 bit
961 void __cpuinit cpu_init(void)
963 int cpu = stack_smp_processor_id();
964 struct tss_struct *t = &per_cpu(init_tss, cpu);
965 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
967 char *estacks = NULL;
968 struct task_struct *me;
971 /* CPU 0 is initialised in head64.c */
975 estacks = boot_exception_stacks;
979 if (cpu_test_and_set(cpu, cpu_initialized))
980 panic("CPU#%d already initialized!\n", cpu);
982 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
984 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
987 * Initialize the per-CPU GDT with the boot GDT,
988 * and set up the GDT descriptor:
992 load_idt((const struct desc_ptr *)&idt_descr);
994 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
997 wrmsrl(MSR_FS_BASE, 0);
998 wrmsrl(MSR_KERNEL_GS_BASE, 0);
1002 if (cpu != 0 && x2apic)
1006 * set up and load the per-CPU TSS
1008 if (!orig_ist->ist[0]) {
1009 static const unsigned int order[N_EXCEPTION_STACKS] = {
1010 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
1011 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
1013 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1015 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
1017 panic("Cannot allocate exception "
1018 "stack %ld %d\n", v, cpu);
1020 estacks += PAGE_SIZE << order[v];
1021 orig_ist->ist[v] = t->x86_tss.ist[v] =
1022 (unsigned long)estacks;
1026 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1028 * <= is required because the CPU will access up to
1029 * 8 bits beyond the end of the IO permission bitmap.
1031 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1032 t->io_bitmap[i] = ~0UL;
1034 atomic_inc(&init_mm.mm_count);
1035 me->active_mm = &init_mm;
1038 enter_lazy_tlb(&init_mm, me);
1040 load_sp0(t, ¤t->thread);
1041 set_tss_desc(cpu, t);
1043 load_LDT(&init_mm.context);
1047 * If the kgdb is connected no debug regs should be altered. This
1048 * is only applicable when KGDB and a KGDB I/O module are built
1049 * into the kernel and you are using early debugging with
1050 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1052 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1053 arch_kgdb_ops.correct_hw_break();
1057 * Clear all 6 debug registers:
1060 set_debugreg(0UL, 0);
1061 set_debugreg(0UL, 1);
1062 set_debugreg(0UL, 2);
1063 set_debugreg(0UL, 3);
1064 set_debugreg(0UL, 6);
1065 set_debugreg(0UL, 7);
1067 /* If the kgdb is connected no debug regs should be altered. */
1073 raw_local_save_flags(kernel_eflags);
1081 void __cpuinit cpu_init(void)
1083 int cpu = smp_processor_id();
1084 struct task_struct *curr = current;
1085 struct tss_struct *t = &per_cpu(init_tss, cpu);
1086 struct thread_struct *thread = &curr->thread;
1088 if (cpu_test_and_set(cpu, cpu_initialized)) {
1089 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1090 for (;;) local_irq_enable();
1093 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1095 if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1096 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1098 load_idt(&idt_descr);
1099 switch_to_new_gdt();
1102 * Set up and load the per-CPU TSS and LDT
1104 atomic_inc(&init_mm.mm_count);
1105 curr->active_mm = &init_mm;
1108 enter_lazy_tlb(&init_mm, curr);
1110 load_sp0(t, thread);
1111 set_tss_desc(cpu, t);
1113 load_LDT(&init_mm.context);
1115 #ifdef CONFIG_DOUBLEFAULT
1116 /* Set up doublefault TSS pointer in the GDT */
1117 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1121 asm volatile ("mov %0, %%gs" : : "r" (0));
1123 /* Clear all 6 debug registers: */
1132 * Force FPU initialization:
1135 current_thread_info()->status = TS_XSAVE;
1137 current_thread_info()->status = 0;
1139 mxcsr_feature_mask_init();
1142 * Boot processor to setup the FP and extended state context info.
1144 if (smp_processor_id() == boot_cpu_id)
1145 init_thread_xstate();