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);
108 * Naming convention should be: <Name> [(<Codename>)]
109 * This table only is used unless init_<vendor>() below doesn't set it;
110 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
114 /* Look up CPU names by table lookup. */
115 static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
117 struct cpu_model_info *info;
119 if (c->x86_model >= 16)
120 return NULL; /* Range check */
125 info = this_cpu->c_models;
127 while (info && info->family) {
128 if (info->family == c->x86)
129 return info->model_names[c->x86_model];
132 return NULL; /* Not found */
135 static int __init x86_fxsr_setup(char *s)
137 setup_clear_cpu_cap(X86_FEATURE_FXSR);
138 setup_clear_cpu_cap(X86_FEATURE_XMM);
141 __setup("nofxsr", x86_fxsr_setup);
143 static int __init x86_sep_setup(char *s)
145 setup_clear_cpu_cap(X86_FEATURE_SEP);
148 __setup("nosep", x86_sep_setup);
150 /* Standard macro to see if a specific flag is changeable */
151 static inline int flag_is_changeable_p(u32 flag)
165 : "=&r" (f1), "=&r" (f2)
168 return ((f1^f2) & flag) != 0;
171 /* Probe for the CPUID instruction */
172 static int __cpuinit have_cpuid_p(void)
174 return flag_is_changeable_p(X86_EFLAGS_ID);
177 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
179 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
180 /* Disable processor serial number */
181 unsigned long lo, hi;
182 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
184 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
185 printk(KERN_NOTICE "CPU serial number disabled.\n");
186 clear_cpu_cap(c, X86_FEATURE_PN);
188 /* Disabling the serial number may affect the cpuid level */
189 c->cpuid_level = cpuid_eax(0);
193 static int __init x86_serial_nr_setup(char *s)
195 disable_x86_serial_nr = 0;
198 __setup("serialnumber", x86_serial_nr_setup);
200 /* Probe for the CPUID instruction */
201 static inline int have_cpuid_p(void)
207 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
209 /* Current gdt points %fs at the "master" per-cpu area: after this,
210 * it's on the real one. */
211 void switch_to_new_gdt(void)
213 struct desc_ptr gdt_descr;
215 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
216 gdt_descr.size = GDT_SIZE - 1;
217 load_gdt(&gdt_descr);
219 asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
223 static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
225 static void __cpuinit default_init(struct cpuinfo_x86 *c)
228 display_cacheinfo(c);
230 /* Not much we can do here... */
231 /* Check if at least it has cpuid */
232 if (c->cpuid_level == -1) {
233 /* No cpuid. It must be an ancient CPU */
235 strcpy(c->x86_model_id, "486");
236 else if (c->x86 == 3)
237 strcpy(c->x86_model_id, "386");
242 static struct cpu_dev __cpuinitdata default_cpu = {
243 .c_init = default_init,
244 .c_vendor = "Unknown",
245 .c_x86_vendor = X86_VENDOR_UNKNOWN,
248 int __cpuinit get_model_name(struct cpuinfo_x86 *c)
253 if (c->extended_cpuid_level < 0x80000004)
256 v = (unsigned int *) c->x86_model_id;
257 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
258 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
259 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
260 c->x86_model_id[48] = 0;
262 /* Intel chips right-justify this string for some dumb reason;
263 undo that brain damage */
264 p = q = &c->x86_model_id[0];
270 while (q <= &c->x86_model_id[48])
271 *q++ = '\0'; /* Zero-pad the rest */
277 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
279 unsigned int n, dummy, ebx, ecx, edx, l2size;
281 n = c->extended_cpuid_level;
283 if (n >= 0x80000005) {
284 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
285 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
286 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
287 c->x86_cache_size = (ecx>>24) + (edx>>24);
289 /* On K8 L1 TLB is inclusive, so don't count it */
294 if (n < 0x80000006) /* Some chips just has a large L1. */
297 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
301 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
303 /* do processor-specific cache resizing */
304 if (this_cpu->c_size_cache)
305 l2size = this_cpu->c_size_cache(c, l2size);
307 /* Allow user to override all this if necessary. */
308 if (cachesize_override != -1)
309 l2size = cachesize_override;
312 return; /* Again, no L2 cache is possible */
315 c->x86_cache_size = l2size;
317 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
321 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
324 u32 eax, ebx, ecx, edx;
325 int index_msb, core_bits;
327 if (!cpu_has(c, X86_FEATURE_HT))
330 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
333 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
336 cpuid(1, &eax, &ebx, &ecx, &edx);
338 smp_num_siblings = (ebx & 0xff0000) >> 16;
340 if (smp_num_siblings == 1) {
341 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
342 } else if (smp_num_siblings > 1) {
344 if (smp_num_siblings > NR_CPUS) {
345 printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
347 smp_num_siblings = 1;
351 index_msb = get_count_order(smp_num_siblings);
353 c->phys_proc_id = phys_pkg_id(index_msb);
355 c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb);
358 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
360 index_msb = get_count_order(smp_num_siblings);
362 core_bits = get_count_order(c->x86_max_cores);
365 c->cpu_core_id = phys_pkg_id(index_msb) &
366 ((1 << core_bits) - 1);
368 c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) &
369 ((1 << core_bits) - 1);
374 if ((c->x86_max_cores * smp_num_siblings) > 1) {
375 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
377 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
383 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
385 char *v = c->x86_vendor_id;
389 for (i = 0; i < X86_VENDOR_NUM; i++) {
393 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
394 (cpu_devs[i]->c_ident[1] &&
395 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
396 this_cpu = cpu_devs[i];
397 c->x86_vendor = this_cpu->c_x86_vendor;
404 printk(KERN_ERR "CPU: Vendor unknown, using generic init.\n");
405 printk(KERN_ERR "CPU: Your system may be unstable.\n");
408 c->x86_vendor = X86_VENDOR_UNKNOWN;
409 this_cpu = &default_cpu;
412 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
414 /* Get vendor name */
415 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
416 (unsigned int *)&c->x86_vendor_id[0],
417 (unsigned int *)&c->x86_vendor_id[8],
418 (unsigned int *)&c->x86_vendor_id[4]);
421 /* Intel-defined flags: level 0x00000001 */
422 if (c->cpuid_level >= 0x00000001) {
423 u32 junk, tfms, cap0, misc;
424 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
425 c->x86 = (tfms >> 8) & 0xf;
426 c->x86_model = (tfms >> 4) & 0xf;
427 c->x86_mask = tfms & 0xf;
429 c->x86 += (tfms >> 20) & 0xff;
431 c->x86_model += ((tfms >> 16) & 0xf) << 4;
432 if (cap0 & (1<<19)) {
433 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
434 c->x86_cache_alignment = c->x86_clflush_size;
439 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
444 /* Intel-defined flags: level 0x00000001 */
445 if (c->cpuid_level >= 0x00000001) {
446 u32 capability, excap;
447 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
448 c->x86_capability[0] = capability;
449 c->x86_capability[4] = excap;
452 /* AMD-defined flags: level 0x80000001 */
453 xlvl = cpuid_eax(0x80000000);
454 c->extended_cpuid_level = xlvl;
455 if ((xlvl & 0xffff0000) == 0x80000000) {
456 if (xlvl >= 0x80000001) {
457 c->x86_capability[1] = cpuid_edx(0x80000001);
458 c->x86_capability[6] = cpuid_ecx(0x80000001);
463 * Do minimum CPU detection early.
464 * Fields really needed: vendor, cpuid_level, family, model, mask,
466 * The others are not touched to avoid unwanted side effects.
468 * WARNING: this function is only called on the BP. Don't add code here
469 * that is supposed to run on all CPUs.
471 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
473 c->x86_clflush_size = 32;
474 c->x86_cache_alignment = c->x86_clflush_size;
479 memset(&c->x86_capability, 0, sizeof c->x86_capability);
481 c->extended_cpuid_level = 0;
489 if (this_cpu->c_early_init)
490 this_cpu->c_early_init(c);
492 validate_pat_support(c);
495 void __init early_cpu_init(void)
497 struct cpu_dev **cdev;
500 printk("KERNEL supported cpus:\n");
501 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
502 struct cpu_dev *cpudev = *cdev;
505 if (count >= X86_VENDOR_NUM)
507 cpu_devs[count] = cpudev;
510 for (j = 0; j < 2; j++) {
511 if (!cpudev->c_ident[j])
513 printk(" %s %s\n", cpudev->c_vendor,
518 early_identify_cpu(&boot_cpu_data);
522 * The NOPL instruction is supposed to exist on all CPUs with
523 * family >= 6, unfortunately, that's not true in practice because
524 * of early VIA chips and (more importantly) broken virtualizers that
525 * are not easy to detect. Hence, probe for it based on first
528 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
530 const u32 nopl_signature = 0x888c53b1; /* Random number */
531 u32 has_nopl = nopl_signature;
533 clear_cpu_cap(c, X86_FEATURE_NOPL);
536 "1: .byte 0x0f,0x1f,0xc0\n" /* nopl %eax */
538 " .section .fixup,\"ax\"\n"
545 if (has_nopl == nopl_signature)
546 set_cpu_cap(c, X86_FEATURE_NOPL);
550 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
555 c->extended_cpuid_level = 0;
563 if (c->cpuid_level >= 0x00000001) {
564 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
566 c->apicid = phys_pkg_id(c->initial_apicid, 0);
567 c->phys_proc_id = c->initial_apicid;
569 c->apicid = c->initial_apicid;
573 if (c->extended_cpuid_level >= 0x80000004)
574 get_model_name(c); /* Default name */
576 init_scattered_cpuid_features(c);
581 * This does the hard work of actually picking apart the CPU stuff...
583 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
587 c->loops_per_jiffy = loops_per_jiffy;
588 c->x86_cache_size = -1;
589 c->x86_vendor = X86_VENDOR_UNKNOWN;
590 c->cpuid_level = -1; /* CPUID not detected */
591 c->x86_model = c->x86_mask = 0; /* So far unknown... */
592 c->x86_vendor_id[0] = '\0'; /* Unset */
593 c->x86_model_id[0] = '\0'; /* Unset */
594 c->x86_max_cores = 1;
595 c->x86_clflush_size = 32;
596 memset(&c->x86_capability, 0, sizeof c->x86_capability);
598 if (!have_cpuid_p()) {
600 * First of all, decide if this is a 486 or higher
601 * It's a 486 if we can modify the AC flag
603 if (flag_is_changeable_p(X86_EFLAGS_AC))
611 if (this_cpu->c_identify)
612 this_cpu->c_identify(c);
615 * Vendor-specific initialization. In this section we
616 * canonicalize the feature flags, meaning if there are
617 * features a certain CPU supports which CPUID doesn't
618 * tell us, CPUID claiming incorrect flags, or other bugs,
619 * we handle them here.
621 * At the end of this section, c->x86_capability better
622 * indicate the features this CPU genuinely supports!
624 if (this_cpu->c_init)
627 /* Disable the PN if appropriate */
628 squash_the_stupid_serial_number(c);
631 * The vendor-specific functions might have changed features. Now
632 * we do "generic changes."
635 /* If the model name is still unset, do table lookup. */
636 if (!c->x86_model_id[0]) {
638 p = table_lookup_model(c);
640 strcpy(c->x86_model_id, p);
643 sprintf(c->x86_model_id, "%02x/%02x",
644 c->x86, c->x86_model);
648 * On SMP, boot_cpu_data holds the common feature set between
649 * all CPUs; so make sure that we indicate which features are
650 * common between the CPUs. The first time this routine gets
651 * executed, c == &boot_cpu_data.
653 if (c != &boot_cpu_data) {
654 /* AND the already accumulated flags with these */
655 for (i = 0; i < NCAPINTS; i++)
656 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
659 /* Clear all flags overriden by options */
660 for (i = 0; i < NCAPINTS; i++)
661 c->x86_capability[i] &= ~cleared_cpu_caps[i];
663 /* Init Machine Check Exception if available. */
666 select_idle_routine(c);
669 void __init identify_boot_cpu(void)
671 identify_cpu(&boot_cpu_data);
676 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
678 BUG_ON(c == &boot_cpu_data);
689 static struct msr_range msr_range_array[] __cpuinitdata = {
690 { 0x00000000, 0x00000418},
691 { 0xc0000000, 0xc000040b},
692 { 0xc0010000, 0xc0010142},
693 { 0xc0011000, 0xc001103b},
696 static void __cpuinit print_cpu_msr(void)
701 unsigned index_min, index_max;
703 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
704 index_min = msr_range_array[i].min;
705 index_max = msr_range_array[i].max;
706 for (index = index_min; index < index_max; index++) {
707 if (rdmsrl_amd_safe(index, &val))
709 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
714 static int show_msr __cpuinitdata;
715 static __init int setup_show_msr(char *arg)
719 get_option(&arg, &num);
725 __setup("show_msr=", setup_show_msr);
727 static __init int setup_noclflush(char *arg)
729 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
732 __setup("noclflush", setup_noclflush);
734 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
738 if (c->x86_vendor < X86_VENDOR_NUM)
739 vendor = this_cpu->c_vendor;
740 else if (c->cpuid_level >= 0)
741 vendor = c->x86_vendor_id;
743 if (vendor && strncmp(c->x86_model_id, vendor, strlen(vendor)))
744 printk(KERN_CONT "%s ", vendor);
746 if (c->x86_model_id[0])
747 printk(KERN_CONT "%s", c->x86_model_id);
749 printk(KERN_CONT "%d86", c->x86);
751 if (c->x86_mask || c->cpuid_level >= 0)
752 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
754 printk(KERN_CONT "\n");
757 if (c->cpu_index < show_msr)
765 static __init int setup_disablecpuid(char *arg)
768 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
769 setup_clear_cpu_cap(bit);
774 __setup("clearcpuid=", setup_disablecpuid);
776 cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
779 struct x8664_pda **_cpu_pda __read_mostly;
780 EXPORT_SYMBOL(_cpu_pda);
782 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
784 char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
786 unsigned long __supported_pte_mask __read_mostly = ~0UL;
787 EXPORT_SYMBOL_GPL(__supported_pte_mask);
789 static int do_not_nx __cpuinitdata;
792 Control non executable mappings for 64bit processes.
797 static int __init nonx_setup(char *str)
801 if (!strncmp(str, "on", 2)) {
802 __supported_pte_mask |= _PAGE_NX;
804 } else if (!strncmp(str, "off", 3)) {
806 __supported_pte_mask &= ~_PAGE_NX;
810 early_param("noexec", nonx_setup);
812 int force_personality32;
815 Control non executable heap for 32bit processes.
816 To control the stack too use noexec=off
818 on PROT_READ does not imply PROT_EXEC for 32bit processes (default)
819 off PROT_READ implies PROT_EXEC
821 static int __init nonx32_setup(char *str)
823 if (!strcmp(str, "on"))
824 force_personality32 &= ~READ_IMPLIES_EXEC;
825 else if (!strcmp(str, "off"))
826 force_personality32 |= READ_IMPLIES_EXEC;
829 __setup("noexec32=", nonx32_setup);
831 void pda_init(int cpu)
833 struct x8664_pda *pda = cpu_pda(cpu);
835 /* Setup up data that may be needed in __get_free_pages early */
838 /* Memory clobbers used to order PDA accessed */
840 wrmsrl(MSR_GS_BASE, pda);
843 pda->cpunumber = cpu;
845 pda->kernelstack = (unsigned long)stack_thread_info() -
846 PDA_STACKOFFSET + THREAD_SIZE;
847 pda->active_mm = &init_mm;
851 /* others are initialized in smpboot.c */
852 pda->pcurrent = &init_task;
853 pda->irqstackptr = boot_cpu_stack;
854 pda->irqstackptr += IRQSTACKSIZE - 64;
856 if (!pda->irqstackptr) {
857 pda->irqstackptr = (char *)
858 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
859 if (!pda->irqstackptr)
860 panic("cannot allocate irqstack for cpu %d",
862 pda->irqstackptr += IRQSTACKSIZE - 64;
865 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
866 pda->nodenumber = cpu_to_node(cpu);
870 char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
871 DEBUG_STKSZ] __page_aligned_bss;
873 extern asmlinkage void ignore_sysret(void);
875 /* May not be marked __init: used by software suspend */
876 void syscall_init(void)
879 * LSTAR and STAR live in a bit strange symbiosis.
880 * They both write to the same internal register. STAR allows to
881 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
883 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
884 wrmsrl(MSR_LSTAR, system_call);
885 wrmsrl(MSR_CSTAR, ignore_sysret);
887 #ifdef CONFIG_IA32_EMULATION
888 syscall32_cpu_init();
891 /* Flags to clear on syscall */
892 wrmsrl(MSR_SYSCALL_MASK,
893 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
896 void __cpuinit check_efer(void)
900 rdmsrl(MSR_EFER, efer);
901 if (!(efer & EFER_NX) || do_not_nx)
902 __supported_pte_mask &= ~_PAGE_NX;
905 unsigned long kernel_eflags;
908 * Copies of the original ist values from the tss are only accessed during
909 * debugging, no special alignment required.
911 DEFINE_PER_CPU(struct orig_ist, orig_ist);
915 /* Make sure %fs is initialized properly in idle threads */
916 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
918 memset(regs, 0, sizeof(struct pt_regs));
919 regs->fs = __KERNEL_PERCPU;
925 * cpu_init() initializes state that is per-CPU. Some data is already
926 * initialized (naturally) in the bootstrap process, such as the GDT
927 * and IDT. We reload them nevertheless, this function acts as a
928 * 'CPU state barrier', nothing should get across.
929 * A lot of state is already set up in PDA init for 64 bit
932 void __cpuinit cpu_init(void)
934 int cpu = stack_smp_processor_id();
935 struct tss_struct *t = &per_cpu(init_tss, cpu);
936 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
938 char *estacks = NULL;
939 struct task_struct *me;
942 /* CPU 0 is initialised in head64.c */
946 estacks = boot_exception_stacks;
950 if (cpu_test_and_set(cpu, cpu_initialized))
951 panic("CPU#%d already initialized!\n", cpu);
953 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
955 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
958 * Initialize the per-CPU GDT with the boot GDT,
959 * and set up the GDT descriptor:
963 load_idt((const struct desc_ptr *)&idt_descr);
965 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
968 wrmsrl(MSR_FS_BASE, 0);
969 wrmsrl(MSR_KERNEL_GS_BASE, 0);
973 if (cpu != 0 && x2apic)
977 * set up and load the per-CPU TSS
979 if (!orig_ist->ist[0]) {
980 static const unsigned int order[N_EXCEPTION_STACKS] = {
981 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
982 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
984 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
986 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
988 panic("Cannot allocate exception "
989 "stack %ld %d\n", v, cpu);
991 estacks += PAGE_SIZE << order[v];
992 orig_ist->ist[v] = t->x86_tss.ist[v] =
993 (unsigned long)estacks;
997 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
999 * <= is required because the CPU will access up to
1000 * 8 bits beyond the end of the IO permission bitmap.
1002 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1003 t->io_bitmap[i] = ~0UL;
1005 atomic_inc(&init_mm.mm_count);
1006 me->active_mm = &init_mm;
1009 enter_lazy_tlb(&init_mm, me);
1011 load_sp0(t, ¤t->thread);
1012 set_tss_desc(cpu, t);
1014 load_LDT(&init_mm.context);
1018 * If the kgdb is connected no debug regs should be altered. This
1019 * is only applicable when KGDB and a KGDB I/O module are built
1020 * into the kernel and you are using early debugging with
1021 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1023 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1024 arch_kgdb_ops.correct_hw_break();
1028 * Clear all 6 debug registers:
1031 set_debugreg(0UL, 0);
1032 set_debugreg(0UL, 1);
1033 set_debugreg(0UL, 2);
1034 set_debugreg(0UL, 3);
1035 set_debugreg(0UL, 6);
1036 set_debugreg(0UL, 7);
1038 /* If the kgdb is connected no debug regs should be altered. */
1044 raw_local_save_flags(kernel_eflags);
1052 void __cpuinit cpu_init(void)
1054 int cpu = smp_processor_id();
1055 struct task_struct *curr = current;
1056 struct tss_struct *t = &per_cpu(init_tss, cpu);
1057 struct thread_struct *thread = &curr->thread;
1059 if (cpu_test_and_set(cpu, cpu_initialized)) {
1060 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1061 for (;;) local_irq_enable();
1064 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1066 if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1067 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1069 load_idt(&idt_descr);
1070 switch_to_new_gdt();
1073 * Set up and load the per-CPU TSS and LDT
1075 atomic_inc(&init_mm.mm_count);
1076 curr->active_mm = &init_mm;
1079 enter_lazy_tlb(&init_mm, curr);
1081 load_sp0(t, thread);
1082 set_tss_desc(cpu, t);
1084 load_LDT(&init_mm.context);
1086 #ifdef CONFIG_DOUBLEFAULT
1087 /* Set up doublefault TSS pointer in the GDT */
1088 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1092 asm volatile ("mov %0, %%gs" : : "r" (0));
1094 /* Clear all 6 debug registers: */
1103 * Force FPU initialization:
1106 current_thread_info()->status = TS_XSAVE;
1108 current_thread_info()->status = 0;
1110 mxcsr_feature_mask_init();
1113 * Boot processor to setup the FP and extended state context info.
1115 if (!smp_processor_id())
1116 init_thread_xstate();
1121 #ifdef CONFIG_HOTPLUG_CPU
1122 void __cpuinit cpu_uninit(void)
1124 int cpu = raw_smp_processor_id();
1125 cpu_clear(cpu, cpu_initialized);
1127 /* lazy TLB state */
1128 per_cpu(cpu_tlbstate, cpu).state = 0;
1129 per_cpu(cpu_tlbstate, cpu).active_mm = &init_mm;