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)
128 * Cyrix and IDT cpus allow disabling of CPUID
129 * so the code below may return different results
130 * when it is executed before and after enabling
131 * the CPUID. Add "volatile" to not allow gcc to
132 * optimize the subsequent calls to this function.
134 asm volatile ("pushfl\n\t"
144 : "=&r" (f1), "=&r" (f2)
147 return ((f1^f2) & flag) != 0;
150 /* Probe for the CPUID instruction */
151 static int __cpuinit have_cpuid_p(void)
153 return flag_is_changeable_p(X86_EFLAGS_ID);
156 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
158 if (cpu_has(c, X86_FEATURE_PN) && disable_x86_serial_nr) {
159 /* Disable processor serial number */
160 unsigned long lo, hi;
161 rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
163 wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
164 printk(KERN_NOTICE "CPU serial number disabled.\n");
165 clear_cpu_cap(c, X86_FEATURE_PN);
167 /* Disabling the serial number may affect the cpuid level */
168 c->cpuid_level = cpuid_eax(0);
172 static int __init x86_serial_nr_setup(char *s)
174 disable_x86_serial_nr = 0;
177 __setup("serialnumber", x86_serial_nr_setup);
179 static inline int flag_is_changeable_p(u32 flag)
183 /* Probe for the CPUID instruction */
184 static inline int have_cpuid_p(void)
188 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
194 * Naming convention should be: <Name> [(<Codename>)]
195 * This table only is used unless init_<vendor>() below doesn't set it;
196 * in particular, if CPUID levels 0x80000002..4 are supported, this isn't used
200 /* Look up CPU names by table lookup. */
201 static char __cpuinit *table_lookup_model(struct cpuinfo_x86 *c)
203 struct cpu_model_info *info;
205 if (c->x86_model >= 16)
206 return NULL; /* Range check */
211 info = this_cpu->c_models;
213 while (info && info->family) {
214 if (info->family == c->x86)
215 return info->model_names[c->x86_model];
218 return NULL; /* Not found */
221 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
223 /* Current gdt points %fs at the "master" per-cpu area: after this,
224 * it's on the real one. */
225 void switch_to_new_gdt(void)
227 struct desc_ptr gdt_descr;
229 gdt_descr.address = (long)get_cpu_gdt_table(smp_processor_id());
230 gdt_descr.size = GDT_SIZE - 1;
231 load_gdt(&gdt_descr);
233 asm("mov %0, %%fs" : : "r" (__KERNEL_PERCPU) : "memory");
237 static struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
239 static void __cpuinit default_init(struct cpuinfo_x86 *c)
242 display_cacheinfo(c);
244 /* Not much we can do here... */
245 /* Check if at least it has cpuid */
246 if (c->cpuid_level == -1) {
247 /* No cpuid. It must be an ancient CPU */
249 strcpy(c->x86_model_id, "486");
250 else if (c->x86 == 3)
251 strcpy(c->x86_model_id, "386");
256 static struct cpu_dev __cpuinitdata default_cpu = {
257 .c_init = default_init,
258 .c_vendor = "Unknown",
259 .c_x86_vendor = X86_VENDOR_UNKNOWN,
262 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
267 if (c->extended_cpuid_level < 0x80000004)
270 v = (unsigned int *) c->x86_model_id;
271 cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
272 cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
273 cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
274 c->x86_model_id[48] = 0;
276 /* Intel chips right-justify this string for some dumb reason;
277 undo that brain damage */
278 p = q = &c->x86_model_id[0];
284 while (q <= &c->x86_model_id[48])
285 *q++ = '\0'; /* Zero-pad the rest */
289 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
291 unsigned int n, dummy, ebx, ecx, edx, l2size;
293 n = c->extended_cpuid_level;
295 if (n >= 0x80000005) {
296 cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
297 printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
298 edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
299 c->x86_cache_size = (ecx>>24) + (edx>>24);
301 /* On K8 L1 TLB is inclusive, so don't count it */
306 if (n < 0x80000006) /* Some chips just has a large L1. */
309 cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
313 c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
315 /* do processor-specific cache resizing */
316 if (this_cpu->c_size_cache)
317 l2size = this_cpu->c_size_cache(c, l2size);
319 /* Allow user to override all this if necessary. */
320 if (cachesize_override != -1)
321 l2size = cachesize_override;
324 return; /* Again, no L2 cache is possible */
327 c->x86_cache_size = l2size;
329 printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
333 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
336 u32 eax, ebx, ecx, edx;
337 int index_msb, core_bits;
339 if (!cpu_has(c, X86_FEATURE_HT))
342 if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
345 if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
348 cpuid(1, &eax, &ebx, &ecx, &edx);
350 smp_num_siblings = (ebx & 0xff0000) >> 16;
352 if (smp_num_siblings == 1) {
353 printk(KERN_INFO "CPU: Hyper-Threading is disabled\n");
354 } else if (smp_num_siblings > 1) {
356 if (smp_num_siblings > NR_CPUS) {
357 printk(KERN_WARNING "CPU: Unsupported number of siblings %d",
359 smp_num_siblings = 1;
363 index_msb = get_count_order(smp_num_siblings);
365 c->phys_proc_id = phys_pkg_id(index_msb);
367 c->phys_proc_id = phys_pkg_id(c->initial_apicid, index_msb);
370 smp_num_siblings = smp_num_siblings / c->x86_max_cores;
372 index_msb = get_count_order(smp_num_siblings);
374 core_bits = get_count_order(c->x86_max_cores);
377 c->cpu_core_id = phys_pkg_id(index_msb) &
378 ((1 << core_bits) - 1);
380 c->cpu_core_id = phys_pkg_id(c->initial_apicid, index_msb) &
381 ((1 << core_bits) - 1);
386 if ((c->x86_max_cores * smp_num_siblings) > 1) {
387 printk(KERN_INFO "CPU: Physical Processor ID: %d\n",
389 printk(KERN_INFO "CPU: Processor Core ID: %d\n",
395 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
397 char *v = c->x86_vendor_id;
401 for (i = 0; i < X86_VENDOR_NUM; i++) {
405 if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
406 (cpu_devs[i]->c_ident[1] &&
407 !strcmp(v, cpu_devs[i]->c_ident[1]))) {
408 this_cpu = cpu_devs[i];
409 c->x86_vendor = this_cpu->c_x86_vendor;
416 printk(KERN_ERR "CPU: vendor_id '%s' unknown, using generic init.\n", v);
417 printk(KERN_ERR "CPU: Your system may be unstable.\n");
420 c->x86_vendor = X86_VENDOR_UNKNOWN;
421 this_cpu = &default_cpu;
424 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
426 /* Get vendor name */
427 cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
428 (unsigned int *)&c->x86_vendor_id[0],
429 (unsigned int *)&c->x86_vendor_id[8],
430 (unsigned int *)&c->x86_vendor_id[4]);
433 /* Intel-defined flags: level 0x00000001 */
434 if (c->cpuid_level >= 0x00000001) {
435 u32 junk, tfms, cap0, misc;
436 cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
437 c->x86 = (tfms >> 8) & 0xf;
438 c->x86_model = (tfms >> 4) & 0xf;
439 c->x86_mask = tfms & 0xf;
441 c->x86 += (tfms >> 20) & 0xff;
443 c->x86_model += ((tfms >> 16) & 0xf) << 4;
444 if (cap0 & (1<<19)) {
445 c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
446 c->x86_cache_alignment = c->x86_clflush_size;
451 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
456 /* Intel-defined flags: level 0x00000001 */
457 if (c->cpuid_level >= 0x00000001) {
458 u32 capability, excap;
459 cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
460 c->x86_capability[0] = capability;
461 c->x86_capability[4] = excap;
464 /* AMD-defined flags: level 0x80000001 */
465 xlvl = cpuid_eax(0x80000000);
466 c->extended_cpuid_level = xlvl;
467 if ((xlvl & 0xffff0000) == 0x80000000) {
468 if (xlvl >= 0x80000001) {
469 c->x86_capability[1] = cpuid_edx(0x80000001);
470 c->x86_capability[6] = cpuid_ecx(0x80000001);
475 if (c->extended_cpuid_level >= 0x80000008) {
476 u32 eax = cpuid_eax(0x80000008);
478 c->x86_virt_bits = (eax >> 8) & 0xff;
479 c->x86_phys_bits = eax & 0xff;
483 if (c->extended_cpuid_level >= 0x80000007)
484 c->x86_power = cpuid_edx(0x80000007);
488 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
494 * First of all, decide if this is a 486 or higher
495 * It's a 486 if we can modify the AC flag
497 if (flag_is_changeable_p(X86_EFLAGS_AC))
502 for (i = 0; i < X86_VENDOR_NUM; i++)
503 if (cpu_devs[i] && cpu_devs[i]->c_identify) {
504 c->x86_vendor_id[0] = 0;
505 cpu_devs[i]->c_identify(c);
506 if (c->x86_vendor_id[0]) {
515 * Do minimum CPU detection early.
516 * Fields really needed: vendor, cpuid_level, family, model, mask,
518 * The others are not touched to avoid unwanted side effects.
520 * WARNING: this function is only called on the BP. Don't add code here
521 * that is supposed to run on all CPUs.
523 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
526 c->x86_clflush_size = 64;
528 c->x86_clflush_size = 32;
530 c->x86_cache_alignment = c->x86_clflush_size;
532 memset(&c->x86_capability, 0, sizeof c->x86_capability);
533 c->extended_cpuid_level = 0;
536 identify_cpu_without_cpuid(c);
538 /* cyrix could have cpuid enabled via c_identify()*/
548 if (this_cpu->c_early_init)
549 this_cpu->c_early_init(c);
551 validate_pat_support(c);
553 c->cpu_index = boot_cpu_id;
556 void __init early_cpu_init(void)
558 struct cpu_dev **cdev;
561 printk("KERNEL supported cpus:\n");
562 for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
563 struct cpu_dev *cpudev = *cdev;
566 if (count >= X86_VENDOR_NUM)
568 cpu_devs[count] = cpudev;
571 for (j = 0; j < 2; j++) {
572 if (!cpudev->c_ident[j])
574 printk(" %s %s\n", cpudev->c_vendor,
579 early_identify_cpu(&boot_cpu_data);
583 * The NOPL instruction is supposed to exist on all CPUs with
584 * family >= 6; unfortunately, that's not true in practice because
585 * of early VIA chips and (more importantly) broken virtualizers that
586 * are not easy to detect. In the latter case it doesn't even *fail*
587 * reliably, so probing for it doesn't even work. Disable it completely
588 * unless we can find a reliable way to detect all the broken cases.
590 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
592 clear_cpu_cap(c, X86_FEATURE_NOPL);
595 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
597 c->extended_cpuid_level = 0;
600 identify_cpu_without_cpuid(c);
602 /* cyrix could have cpuid enabled via c_identify()*/
612 if (c->cpuid_level >= 0x00000001) {
613 c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
615 # ifdef CONFIG_X86_HT
616 c->apicid = phys_pkg_id(c->initial_apicid, 0);
618 c->apicid = c->initial_apicid;
623 c->phys_proc_id = c->initial_apicid;
627 get_model_name(c); /* Default name */
629 init_scattered_cpuid_features(c);
634 * This does the hard work of actually picking apart the CPU stuff...
636 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
640 c->loops_per_jiffy = loops_per_jiffy;
641 c->x86_cache_size = -1;
642 c->x86_vendor = X86_VENDOR_UNKNOWN;
643 c->x86_model = c->x86_mask = 0; /* So far unknown... */
644 c->x86_vendor_id[0] = '\0'; /* Unset */
645 c->x86_model_id[0] = '\0'; /* Unset */
646 c->x86_max_cores = 1;
647 c->x86_coreid_bits = 0;
649 c->x86_clflush_size = 64;
651 c->cpuid_level = -1; /* CPUID not detected */
652 c->x86_clflush_size = 32;
654 c->x86_cache_alignment = c->x86_clflush_size;
655 memset(&c->x86_capability, 0, sizeof c->x86_capability);
659 if (this_cpu->c_identify)
660 this_cpu->c_identify(c);
663 c->apicid = phys_pkg_id(0);
667 * Vendor-specific initialization. In this section we
668 * canonicalize the feature flags, meaning if there are
669 * features a certain CPU supports which CPUID doesn't
670 * tell us, CPUID claiming incorrect flags, or other bugs,
671 * we handle them here.
673 * At the end of this section, c->x86_capability better
674 * indicate the features this CPU genuinely supports!
676 if (this_cpu->c_init)
679 /* Disable the PN if appropriate */
680 squash_the_stupid_serial_number(c);
683 * The vendor-specific functions might have changed features. Now
684 * we do "generic changes."
687 /* If the model name is still unset, do table lookup. */
688 if (!c->x86_model_id[0]) {
690 p = table_lookup_model(c);
692 strcpy(c->x86_model_id, p);
695 sprintf(c->x86_model_id, "%02x/%02x",
696 c->x86, c->x86_model);
704 * On SMP, boot_cpu_data holds the common feature set between
705 * all CPUs; so make sure that we indicate which features are
706 * common between the CPUs. The first time this routine gets
707 * executed, c == &boot_cpu_data.
709 if (c != &boot_cpu_data) {
710 /* AND the already accumulated flags with these */
711 for (i = 0; i < NCAPINTS; i++)
712 boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
715 /* Clear all flags overriden by options */
716 for (i = 0; i < NCAPINTS; i++)
717 c->x86_capability[i] &= ~cleared_cpu_caps[i];
719 #ifdef CONFIG_X86_MCE
720 /* Init Machine Check Exception if available. */
724 select_idle_routine(c);
726 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
727 numa_add_cpu(smp_processor_id());
732 static void vgetcpu_set_mode(void)
734 if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
735 vgetcpu_mode = VGETCPU_RDTSCP;
737 vgetcpu_mode = VGETCPU_LSL;
741 void __init identify_boot_cpu(void)
743 identify_cpu(&boot_cpu_data);
752 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
754 BUG_ON(c == &boot_cpu_data);
767 static struct msr_range msr_range_array[] __cpuinitdata = {
768 { 0x00000000, 0x00000418},
769 { 0xc0000000, 0xc000040b},
770 { 0xc0010000, 0xc0010142},
771 { 0xc0011000, 0xc001103b},
774 static void __cpuinit print_cpu_msr(void)
779 unsigned index_min, index_max;
781 for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
782 index_min = msr_range_array[i].min;
783 index_max = msr_range_array[i].max;
784 for (index = index_min; index < index_max; index++) {
785 if (rdmsrl_amd_safe(index, &val))
787 printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
792 static int show_msr __cpuinitdata;
793 static __init int setup_show_msr(char *arg)
797 get_option(&arg, &num);
803 __setup("show_msr=", setup_show_msr);
805 static __init int setup_noclflush(char *arg)
807 setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
810 __setup("noclflush", setup_noclflush);
812 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
816 if (c->x86_vendor < X86_VENDOR_NUM)
817 vendor = this_cpu->c_vendor;
818 else if (c->cpuid_level >= 0)
819 vendor = c->x86_vendor_id;
821 if (vendor && !strstr(c->x86_model_id, vendor))
822 printk(KERN_CONT "%s ", vendor);
824 if (c->x86_model_id[0])
825 printk(KERN_CONT "%s", c->x86_model_id);
827 printk(KERN_CONT "%d86", c->x86);
829 if (c->x86_mask || c->cpuid_level >= 0)
830 printk(KERN_CONT " stepping %02x\n", c->x86_mask);
832 printk(KERN_CONT "\n");
835 if (c->cpu_index < show_msr)
843 static __init int setup_disablecpuid(char *arg)
846 if (get_option(&arg, &bit) && bit < NCAPINTS*32)
847 setup_clear_cpu_cap(bit);
852 __setup("clearcpuid=", setup_disablecpuid);
854 cpumask_t cpu_initialized __cpuinitdata = CPU_MASK_NONE;
857 struct x8664_pda **_cpu_pda __read_mostly;
858 EXPORT_SYMBOL(_cpu_pda);
860 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
862 char boot_cpu_stack[IRQSTACKSIZE] __page_aligned_bss;
864 void __cpuinit pda_init(int cpu)
866 struct x8664_pda *pda = cpu_pda(cpu);
868 /* Setup up data that may be needed in __get_free_pages early */
871 /* Memory clobbers used to order PDA accessed */
873 wrmsrl(MSR_GS_BASE, pda);
876 pda->cpunumber = cpu;
878 pda->kernelstack = (unsigned long)stack_thread_info() -
879 PDA_STACKOFFSET + THREAD_SIZE;
880 pda->active_mm = &init_mm;
884 /* others are initialized in smpboot.c */
885 pda->pcurrent = &init_task;
886 pda->irqstackptr = boot_cpu_stack;
887 pda->irqstackptr += IRQSTACKSIZE - 64;
889 if (!pda->irqstackptr) {
890 pda->irqstackptr = (char *)
891 __get_free_pages(GFP_ATOMIC, IRQSTACK_ORDER);
892 if (!pda->irqstackptr)
893 panic("cannot allocate irqstack for cpu %d",
895 pda->irqstackptr += IRQSTACKSIZE - 64;
898 if (pda->nodenumber == 0 && cpu_to_node(cpu) != NUMA_NO_NODE)
899 pda->nodenumber = cpu_to_node(cpu);
903 char boot_exception_stacks[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ +
904 DEBUG_STKSZ] __page_aligned_bss;
906 extern asmlinkage void ignore_sysret(void);
908 /* May not be marked __init: used by software suspend */
909 void syscall_init(void)
912 * LSTAR and STAR live in a bit strange symbiosis.
913 * They both write to the same internal register. STAR allows to
914 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
916 wrmsrl(MSR_STAR, ((u64)__USER32_CS)<<48 | ((u64)__KERNEL_CS)<<32);
917 wrmsrl(MSR_LSTAR, system_call);
918 wrmsrl(MSR_CSTAR, ignore_sysret);
920 #ifdef CONFIG_IA32_EMULATION
921 syscall32_cpu_init();
924 /* Flags to clear on syscall */
925 wrmsrl(MSR_SYSCALL_MASK,
926 X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
929 unsigned long kernel_eflags;
932 * Copies of the original ist values from the tss are only accessed during
933 * debugging, no special alignment required.
935 DEFINE_PER_CPU(struct orig_ist, orig_ist);
939 /* Make sure %fs is initialized properly in idle threads */
940 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
942 memset(regs, 0, sizeof(struct pt_regs));
943 regs->fs = __KERNEL_PERCPU;
949 * cpu_init() initializes state that is per-CPU. Some data is already
950 * initialized (naturally) in the bootstrap process, such as the GDT
951 * and IDT. We reload them nevertheless, this function acts as a
952 * 'CPU state barrier', nothing should get across.
953 * A lot of state is already set up in PDA init for 64 bit
956 void __cpuinit cpu_init(void)
958 int cpu = stack_smp_processor_id();
959 struct tss_struct *t = &per_cpu(init_tss, cpu);
960 struct orig_ist *orig_ist = &per_cpu(orig_ist, cpu);
962 char *estacks = NULL;
963 struct task_struct *me;
966 /* CPU 0 is initialised in head64.c */
970 estacks = boot_exception_stacks;
974 if (cpu_test_and_set(cpu, cpu_initialized))
975 panic("CPU#%d already initialized!\n", cpu);
977 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
979 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
982 * Initialize the per-CPU GDT with the boot GDT,
983 * and set up the GDT descriptor:
987 load_idt((const struct desc_ptr *)&idt_descr);
989 memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
992 wrmsrl(MSR_FS_BASE, 0);
993 wrmsrl(MSR_KERNEL_GS_BASE, 0);
997 if (cpu != 0 && x2apic)
1001 * set up and load the per-CPU TSS
1003 if (!orig_ist->ist[0]) {
1004 static const unsigned int order[N_EXCEPTION_STACKS] = {
1005 [0 ... N_EXCEPTION_STACKS - 1] = EXCEPTION_STACK_ORDER,
1006 [DEBUG_STACK - 1] = DEBUG_STACK_ORDER
1008 for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1010 estacks = (char *)__get_free_pages(GFP_ATOMIC, order[v]);
1012 panic("Cannot allocate exception "
1013 "stack %ld %d\n", v, cpu);
1015 estacks += PAGE_SIZE << order[v];
1016 orig_ist->ist[v] = t->x86_tss.ist[v] =
1017 (unsigned long)estacks;
1021 t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1023 * <= is required because the CPU will access up to
1024 * 8 bits beyond the end of the IO permission bitmap.
1026 for (i = 0; i <= IO_BITMAP_LONGS; i++)
1027 t->io_bitmap[i] = ~0UL;
1029 atomic_inc(&init_mm.mm_count);
1030 me->active_mm = &init_mm;
1033 enter_lazy_tlb(&init_mm, me);
1035 load_sp0(t, ¤t->thread);
1036 set_tss_desc(cpu, t);
1038 load_LDT(&init_mm.context);
1042 * If the kgdb is connected no debug regs should be altered. This
1043 * is only applicable when KGDB and a KGDB I/O module are built
1044 * into the kernel and you are using early debugging with
1045 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1047 if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1048 arch_kgdb_ops.correct_hw_break();
1052 * Clear all 6 debug registers:
1055 set_debugreg(0UL, 0);
1056 set_debugreg(0UL, 1);
1057 set_debugreg(0UL, 2);
1058 set_debugreg(0UL, 3);
1059 set_debugreg(0UL, 6);
1060 set_debugreg(0UL, 7);
1062 /* If the kgdb is connected no debug regs should be altered. */
1068 raw_local_save_flags(kernel_eflags);
1076 void __cpuinit cpu_init(void)
1078 int cpu = smp_processor_id();
1079 struct task_struct *curr = current;
1080 struct tss_struct *t = &per_cpu(init_tss, cpu);
1081 struct thread_struct *thread = &curr->thread;
1083 if (cpu_test_and_set(cpu, cpu_initialized)) {
1084 printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1085 for (;;) local_irq_enable();
1088 printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1090 if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1091 clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1093 load_idt(&idt_descr);
1094 switch_to_new_gdt();
1097 * Set up and load the per-CPU TSS and LDT
1099 atomic_inc(&init_mm.mm_count);
1100 curr->active_mm = &init_mm;
1103 enter_lazy_tlb(&init_mm, curr);
1105 load_sp0(t, thread);
1106 set_tss_desc(cpu, t);
1108 load_LDT(&init_mm.context);
1110 #ifdef CONFIG_DOUBLEFAULT
1111 /* Set up doublefault TSS pointer in the GDT */
1112 __set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1116 asm volatile ("mov %0, %%gs" : : "r" (0));
1118 /* Clear all 6 debug registers: */
1127 * Force FPU initialization:
1130 current_thread_info()->status = TS_XSAVE;
1132 current_thread_info()->status = 0;
1134 mxcsr_feature_mask_init();
1137 * Boot processor to setup the FP and extended state context info.
1139 if (smp_processor_id() == boot_cpu_id)
1140 init_thread_xstate();