xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 31b90347)
1 #include <linux/bootmem.h>
2 #include <linux/linkage.h>
3 #include <linux/bitops.h>
4 #include <linux/kernel.h>
5 #include <linux/module.h>
6 #include <linux/percpu.h>
7 #include <linux/string.h>
8 #include <linux/delay.h>
9 #include <linux/sched.h>
10 #include <linux/init.h>
11 #include <linux/kgdb.h>
12 #include <linux/smp.h>
13 #include <linux/io.h>
14 
15 #include <asm/stackprotector.h>
16 #include <asm/perf_event.h>
17 #include <asm/mmu_context.h>
18 #include <asm/archrandom.h>
19 #include <asm/hypervisor.h>
20 #include <asm/processor.h>
21 #include <asm/debugreg.h>
22 #include <asm/sections.h>
23 #include <linux/topology.h>
24 #include <linux/cpumask.h>
25 #include <asm/pgtable.h>
26 #include <linux/atomic.h>
27 #include <asm/proto.h>
28 #include <asm/setup.h>
29 #include <asm/apic.h>
30 #include <asm/desc.h>
31 #include <asm/i387.h>
32 #include <asm/fpu-internal.h>
33 #include <asm/mtrr.h>
34 #include <linux/numa.h>
35 #include <asm/asm.h>
36 #include <asm/cpu.h>
37 #include <asm/mce.h>
38 #include <asm/msr.h>
39 #include <asm/pat.h>
40 #include <asm/microcode.h>
41 #include <asm/microcode_intel.h>
42 
43 #ifdef CONFIG_X86_LOCAL_APIC
44 #include <asm/uv/uv.h>
45 #endif
46 
47 #include "cpu.h"
48 
49 /* all of these masks are initialized in setup_cpu_local_masks() */
50 cpumask_var_t cpu_initialized_mask;
51 cpumask_var_t cpu_callout_mask;
52 cpumask_var_t cpu_callin_mask;
53 
54 /* representing cpus for which sibling maps can be computed */
55 cpumask_var_t cpu_sibling_setup_mask;
56 
57 /* correctly size the local cpu masks */
58 void __init setup_cpu_local_masks(void)
59 {
60 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
61 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
62 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
63 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
64 }
65 
66 static void default_init(struct cpuinfo_x86 *c)
67 {
68 #ifdef CONFIG_X86_64
69 	cpu_detect_cache_sizes(c);
70 #else
71 	/* Not much we can do here... */
72 	/* Check if at least it has cpuid */
73 	if (c->cpuid_level == -1) {
74 		/* No cpuid. It must be an ancient CPU */
75 		if (c->x86 == 4)
76 			strcpy(c->x86_model_id, "486");
77 		else if (c->x86 == 3)
78 			strcpy(c->x86_model_id, "386");
79 	}
80 #endif
81 }
82 
83 static const struct cpu_dev default_cpu = {
84 	.c_init		= default_init,
85 	.c_vendor	= "Unknown",
86 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
87 };
88 
89 static const struct cpu_dev *this_cpu = &default_cpu;
90 
91 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
92 #ifdef CONFIG_X86_64
93 	/*
94 	 * We need valid kernel segments for data and code in long mode too
95 	 * IRET will check the segment types  kkeil 2000/10/28
96 	 * Also sysret mandates a special GDT layout
97 	 *
98 	 * TLS descriptors are currently at a different place compared to i386.
99 	 * Hopefully nobody expects them at a fixed place (Wine?)
100 	 */
101 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
102 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
103 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
104 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
105 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
106 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
107 #else
108 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
109 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
110 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
111 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
112 	/*
113 	 * Segments used for calling PnP BIOS have byte granularity.
114 	 * They code segments and data segments have fixed 64k limits,
115 	 * the transfer segment sizes are set at run time.
116 	 */
117 	/* 32-bit code */
118 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
119 	/* 16-bit code */
120 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
121 	/* 16-bit data */
122 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
123 	/* 16-bit data */
124 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
125 	/* 16-bit data */
126 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
127 	/*
128 	 * The APM segments have byte granularity and their bases
129 	 * are set at run time.  All have 64k limits.
130 	 */
131 	/* 32-bit code */
132 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
133 	/* 16-bit code */
134 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
135 	/* data */
136 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
137 
138 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
139 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
140 	GDT_STACK_CANARY_INIT
141 #endif
142 } };
143 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
144 
145 static int __init x86_xsave_setup(char *s)
146 {
147 	setup_clear_cpu_cap(X86_FEATURE_XSAVE);
148 	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
149 	setup_clear_cpu_cap(X86_FEATURE_AVX);
150 	setup_clear_cpu_cap(X86_FEATURE_AVX2);
151 	return 1;
152 }
153 __setup("noxsave", x86_xsave_setup);
154 
155 static int __init x86_xsaveopt_setup(char *s)
156 {
157 	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
158 	return 1;
159 }
160 __setup("noxsaveopt", x86_xsaveopt_setup);
161 
162 #ifdef CONFIG_X86_32
163 static int cachesize_override = -1;
164 static int disable_x86_serial_nr = 1;
165 
166 static int __init cachesize_setup(char *str)
167 {
168 	get_option(&str, &cachesize_override);
169 	return 1;
170 }
171 __setup("cachesize=", cachesize_setup);
172 
173 static int __init x86_fxsr_setup(char *s)
174 {
175 	setup_clear_cpu_cap(X86_FEATURE_FXSR);
176 	setup_clear_cpu_cap(X86_FEATURE_XMM);
177 	return 1;
178 }
179 __setup("nofxsr", x86_fxsr_setup);
180 
181 static int __init x86_sep_setup(char *s)
182 {
183 	setup_clear_cpu_cap(X86_FEATURE_SEP);
184 	return 1;
185 }
186 __setup("nosep", x86_sep_setup);
187 
188 /* Standard macro to see if a specific flag is changeable */
189 static inline int flag_is_changeable_p(u32 flag)
190 {
191 	u32 f1, f2;
192 
193 	/*
194 	 * Cyrix and IDT cpus allow disabling of CPUID
195 	 * so the code below may return different results
196 	 * when it is executed before and after enabling
197 	 * the CPUID. Add "volatile" to not allow gcc to
198 	 * optimize the subsequent calls to this function.
199 	 */
200 	asm volatile ("pushfl		\n\t"
201 		      "pushfl		\n\t"
202 		      "popl %0		\n\t"
203 		      "movl %0, %1	\n\t"
204 		      "xorl %2, %0	\n\t"
205 		      "pushl %0		\n\t"
206 		      "popfl		\n\t"
207 		      "pushfl		\n\t"
208 		      "popl %0		\n\t"
209 		      "popfl		\n\t"
210 
211 		      : "=&r" (f1), "=&r" (f2)
212 		      : "ir" (flag));
213 
214 	return ((f1^f2) & flag) != 0;
215 }
216 
217 /* Probe for the CPUID instruction */
218 int have_cpuid_p(void)
219 {
220 	return flag_is_changeable_p(X86_EFLAGS_ID);
221 }
222 
223 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
224 {
225 	unsigned long lo, hi;
226 
227 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
228 		return;
229 
230 	/* Disable processor serial number: */
231 
232 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
233 	lo |= 0x200000;
234 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
235 
236 	printk(KERN_NOTICE "CPU serial number disabled.\n");
237 	clear_cpu_cap(c, X86_FEATURE_PN);
238 
239 	/* Disabling the serial number may affect the cpuid level */
240 	c->cpuid_level = cpuid_eax(0);
241 }
242 
243 static int __init x86_serial_nr_setup(char *s)
244 {
245 	disable_x86_serial_nr = 0;
246 	return 1;
247 }
248 __setup("serialnumber", x86_serial_nr_setup);
249 #else
250 static inline int flag_is_changeable_p(u32 flag)
251 {
252 	return 1;
253 }
254 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
255 {
256 }
257 #endif
258 
259 static __init int setup_disable_smep(char *arg)
260 {
261 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
262 	return 1;
263 }
264 __setup("nosmep", setup_disable_smep);
265 
266 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
267 {
268 	if (cpu_has(c, X86_FEATURE_SMEP))
269 		set_in_cr4(X86_CR4_SMEP);
270 }
271 
272 static __init int setup_disable_smap(char *arg)
273 {
274 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
275 	return 1;
276 }
277 __setup("nosmap", setup_disable_smap);
278 
279 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
280 {
281 	unsigned long eflags;
282 
283 	/* This should have been cleared long ago */
284 	raw_local_save_flags(eflags);
285 	BUG_ON(eflags & X86_EFLAGS_AC);
286 
287 	if (cpu_has(c, X86_FEATURE_SMAP))
288 		set_in_cr4(X86_CR4_SMAP);
289 }
290 
291 /*
292  * Some CPU features depend on higher CPUID levels, which may not always
293  * be available due to CPUID level capping or broken virtualization
294  * software.  Add those features to this table to auto-disable them.
295  */
296 struct cpuid_dependent_feature {
297 	u32 feature;
298 	u32 level;
299 };
300 
301 static const struct cpuid_dependent_feature
302 cpuid_dependent_features[] = {
303 	{ X86_FEATURE_MWAIT,		0x00000005 },
304 	{ X86_FEATURE_DCA,		0x00000009 },
305 	{ X86_FEATURE_XSAVE,		0x0000000d },
306 	{ 0, 0 }
307 };
308 
309 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
310 {
311 	const struct cpuid_dependent_feature *df;
312 
313 	for (df = cpuid_dependent_features; df->feature; df++) {
314 
315 		if (!cpu_has(c, df->feature))
316 			continue;
317 		/*
318 		 * Note: cpuid_level is set to -1 if unavailable, but
319 		 * extended_extended_level is set to 0 if unavailable
320 		 * and the legitimate extended levels are all negative
321 		 * when signed; hence the weird messing around with
322 		 * signs here...
323 		 */
324 		if (!((s32)df->level < 0 ?
325 		     (u32)df->level > (u32)c->extended_cpuid_level :
326 		     (s32)df->level > (s32)c->cpuid_level))
327 			continue;
328 
329 		clear_cpu_cap(c, df->feature);
330 		if (!warn)
331 			continue;
332 
333 		printk(KERN_WARNING
334 		       "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
335 				x86_cap_flags[df->feature], df->level);
336 	}
337 }
338 
339 /*
340  * Naming convention should be: <Name> [(<Codename>)]
341  * This table only is used unless init_<vendor>() below doesn't set it;
342  * in particular, if CPUID levels 0x80000002..4 are supported, this
343  * isn't used
344  */
345 
346 /* Look up CPU names by table lookup. */
347 static const char *table_lookup_model(struct cpuinfo_x86 *c)
348 {
349 #ifdef CONFIG_X86_32
350 	const struct legacy_cpu_model_info *info;
351 
352 	if (c->x86_model >= 16)
353 		return NULL;	/* Range check */
354 
355 	if (!this_cpu)
356 		return NULL;
357 
358 	info = this_cpu->legacy_models;
359 
360 	while (info->family) {
361 		if (info->family == c->x86)
362 			return info->model_names[c->x86_model];
363 		info++;
364 	}
365 #endif
366 	return NULL;		/* Not found */
367 }
368 
369 __u32 cpu_caps_cleared[NCAPINTS];
370 __u32 cpu_caps_set[NCAPINTS];
371 
372 void load_percpu_segment(int cpu)
373 {
374 #ifdef CONFIG_X86_32
375 	loadsegment(fs, __KERNEL_PERCPU);
376 #else
377 	loadsegment(gs, 0);
378 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
379 #endif
380 	load_stack_canary_segment();
381 }
382 
383 /*
384  * Current gdt points %fs at the "master" per-cpu area: after this,
385  * it's on the real one.
386  */
387 void switch_to_new_gdt(int cpu)
388 {
389 	struct desc_ptr gdt_descr;
390 
391 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
392 	gdt_descr.size = GDT_SIZE - 1;
393 	load_gdt(&gdt_descr);
394 	/* Reload the per-cpu base */
395 
396 	load_percpu_segment(cpu);
397 }
398 
399 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
400 
401 static void get_model_name(struct cpuinfo_x86 *c)
402 {
403 	unsigned int *v;
404 	char *p, *q;
405 
406 	if (c->extended_cpuid_level < 0x80000004)
407 		return;
408 
409 	v = (unsigned int *)c->x86_model_id;
410 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
411 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
412 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
413 	c->x86_model_id[48] = 0;
414 
415 	/*
416 	 * Intel chips right-justify this string for some dumb reason;
417 	 * undo that brain damage:
418 	 */
419 	p = q = &c->x86_model_id[0];
420 	while (*p == ' ')
421 		p++;
422 	if (p != q) {
423 		while (*p)
424 			*q++ = *p++;
425 		while (q <= &c->x86_model_id[48])
426 			*q++ = '\0';	/* Zero-pad the rest */
427 	}
428 }
429 
430 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
431 {
432 	unsigned int n, dummy, ebx, ecx, edx, l2size;
433 
434 	n = c->extended_cpuid_level;
435 
436 	if (n >= 0x80000005) {
437 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
438 		c->x86_cache_size = (ecx>>24) + (edx>>24);
439 #ifdef CONFIG_X86_64
440 		/* On K8 L1 TLB is inclusive, so don't count it */
441 		c->x86_tlbsize = 0;
442 #endif
443 	}
444 
445 	if (n < 0x80000006)	/* Some chips just has a large L1. */
446 		return;
447 
448 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
449 	l2size = ecx >> 16;
450 
451 #ifdef CONFIG_X86_64
452 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
453 #else
454 	/* do processor-specific cache resizing */
455 	if (this_cpu->legacy_cache_size)
456 		l2size = this_cpu->legacy_cache_size(c, l2size);
457 
458 	/* Allow user to override all this if necessary. */
459 	if (cachesize_override != -1)
460 		l2size = cachesize_override;
461 
462 	if (l2size == 0)
463 		return;		/* Again, no L2 cache is possible */
464 #endif
465 
466 	c->x86_cache_size = l2size;
467 }
468 
469 u16 __read_mostly tlb_lli_4k[NR_INFO];
470 u16 __read_mostly tlb_lli_2m[NR_INFO];
471 u16 __read_mostly tlb_lli_4m[NR_INFO];
472 u16 __read_mostly tlb_lld_4k[NR_INFO];
473 u16 __read_mostly tlb_lld_2m[NR_INFO];
474 u16 __read_mostly tlb_lld_4m[NR_INFO];
475 
476 /*
477  * tlb_flushall_shift shows the balance point in replacing cr3 write
478  * with multiple 'invlpg'. It will do this replacement when
479  *   flush_tlb_lines <= active_lines/2^tlb_flushall_shift.
480  * If tlb_flushall_shift is -1, means the replacement will be disabled.
481  */
482 s8  __read_mostly tlb_flushall_shift = -1;
483 
484 void cpu_detect_tlb(struct cpuinfo_x86 *c)
485 {
486 	if (this_cpu->c_detect_tlb)
487 		this_cpu->c_detect_tlb(c);
488 
489 	printk(KERN_INFO "Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n" \
490 		"Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d\n"	     \
491 		"tlb_flushall_shift: %d\n",
492 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
493 		tlb_lli_4m[ENTRIES], tlb_lld_4k[ENTRIES],
494 		tlb_lld_2m[ENTRIES], tlb_lld_4m[ENTRIES],
495 		tlb_flushall_shift);
496 }
497 
498 void detect_ht(struct cpuinfo_x86 *c)
499 {
500 #ifdef CONFIG_X86_HT
501 	u32 eax, ebx, ecx, edx;
502 	int index_msb, core_bits;
503 	static bool printed;
504 
505 	if (!cpu_has(c, X86_FEATURE_HT))
506 		return;
507 
508 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
509 		goto out;
510 
511 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
512 		return;
513 
514 	cpuid(1, &eax, &ebx, &ecx, &edx);
515 
516 	smp_num_siblings = (ebx & 0xff0000) >> 16;
517 
518 	if (smp_num_siblings == 1) {
519 		printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
520 		goto out;
521 	}
522 
523 	if (smp_num_siblings <= 1)
524 		goto out;
525 
526 	index_msb = get_count_order(smp_num_siblings);
527 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
528 
529 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
530 
531 	index_msb = get_count_order(smp_num_siblings);
532 
533 	core_bits = get_count_order(c->x86_max_cores);
534 
535 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
536 				       ((1 << core_bits) - 1);
537 
538 out:
539 	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
540 		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
541 		       c->phys_proc_id);
542 		printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
543 		       c->cpu_core_id);
544 		printed = 1;
545 	}
546 #endif
547 }
548 
549 static void get_cpu_vendor(struct cpuinfo_x86 *c)
550 {
551 	char *v = c->x86_vendor_id;
552 	int i;
553 
554 	for (i = 0; i < X86_VENDOR_NUM; i++) {
555 		if (!cpu_devs[i])
556 			break;
557 
558 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
559 		    (cpu_devs[i]->c_ident[1] &&
560 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
561 
562 			this_cpu = cpu_devs[i];
563 			c->x86_vendor = this_cpu->c_x86_vendor;
564 			return;
565 		}
566 	}
567 
568 	printk_once(KERN_ERR
569 			"CPU: vendor_id '%s' unknown, using generic init.\n" \
570 			"CPU: Your system may be unstable.\n", v);
571 
572 	c->x86_vendor = X86_VENDOR_UNKNOWN;
573 	this_cpu = &default_cpu;
574 }
575 
576 void cpu_detect(struct cpuinfo_x86 *c)
577 {
578 	/* Get vendor name */
579 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
580 	      (unsigned int *)&c->x86_vendor_id[0],
581 	      (unsigned int *)&c->x86_vendor_id[8],
582 	      (unsigned int *)&c->x86_vendor_id[4]);
583 
584 	c->x86 = 4;
585 	/* Intel-defined flags: level 0x00000001 */
586 	if (c->cpuid_level >= 0x00000001) {
587 		u32 junk, tfms, cap0, misc;
588 
589 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
590 		c->x86 = (tfms >> 8) & 0xf;
591 		c->x86_model = (tfms >> 4) & 0xf;
592 		c->x86_mask = tfms & 0xf;
593 
594 		if (c->x86 == 0xf)
595 			c->x86 += (tfms >> 20) & 0xff;
596 		if (c->x86 >= 0x6)
597 			c->x86_model += ((tfms >> 16) & 0xf) << 4;
598 
599 		if (cap0 & (1<<19)) {
600 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
601 			c->x86_cache_alignment = c->x86_clflush_size;
602 		}
603 	}
604 }
605 
606 void get_cpu_cap(struct cpuinfo_x86 *c)
607 {
608 	u32 tfms, xlvl;
609 	u32 ebx;
610 
611 	/* Intel-defined flags: level 0x00000001 */
612 	if (c->cpuid_level >= 0x00000001) {
613 		u32 capability, excap;
614 
615 		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
616 		c->x86_capability[0] = capability;
617 		c->x86_capability[4] = excap;
618 	}
619 
620 	/* Additional Intel-defined flags: level 0x00000007 */
621 	if (c->cpuid_level >= 0x00000007) {
622 		u32 eax, ebx, ecx, edx;
623 
624 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
625 
626 		c->x86_capability[9] = ebx;
627 	}
628 
629 	/* AMD-defined flags: level 0x80000001 */
630 	xlvl = cpuid_eax(0x80000000);
631 	c->extended_cpuid_level = xlvl;
632 
633 	if ((xlvl & 0xffff0000) == 0x80000000) {
634 		if (xlvl >= 0x80000001) {
635 			c->x86_capability[1] = cpuid_edx(0x80000001);
636 			c->x86_capability[6] = cpuid_ecx(0x80000001);
637 		}
638 	}
639 
640 	if (c->extended_cpuid_level >= 0x80000008) {
641 		u32 eax = cpuid_eax(0x80000008);
642 
643 		c->x86_virt_bits = (eax >> 8) & 0xff;
644 		c->x86_phys_bits = eax & 0xff;
645 	}
646 #ifdef CONFIG_X86_32
647 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
648 		c->x86_phys_bits = 36;
649 #endif
650 
651 	if (c->extended_cpuid_level >= 0x80000007)
652 		c->x86_power = cpuid_edx(0x80000007);
653 
654 	init_scattered_cpuid_features(c);
655 }
656 
657 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
658 {
659 #ifdef CONFIG_X86_32
660 	int i;
661 
662 	/*
663 	 * First of all, decide if this is a 486 or higher
664 	 * It's a 486 if we can modify the AC flag
665 	 */
666 	if (flag_is_changeable_p(X86_EFLAGS_AC))
667 		c->x86 = 4;
668 	else
669 		c->x86 = 3;
670 
671 	for (i = 0; i < X86_VENDOR_NUM; i++)
672 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
673 			c->x86_vendor_id[0] = 0;
674 			cpu_devs[i]->c_identify(c);
675 			if (c->x86_vendor_id[0]) {
676 				get_cpu_vendor(c);
677 				break;
678 			}
679 		}
680 #endif
681 }
682 
683 /*
684  * Do minimum CPU detection early.
685  * Fields really needed: vendor, cpuid_level, family, model, mask,
686  * cache alignment.
687  * The others are not touched to avoid unwanted side effects.
688  *
689  * WARNING: this function is only called on the BP.  Don't add code here
690  * that is supposed to run on all CPUs.
691  */
692 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
693 {
694 #ifdef CONFIG_X86_64
695 	c->x86_clflush_size = 64;
696 	c->x86_phys_bits = 36;
697 	c->x86_virt_bits = 48;
698 #else
699 	c->x86_clflush_size = 32;
700 	c->x86_phys_bits = 32;
701 	c->x86_virt_bits = 32;
702 #endif
703 	c->x86_cache_alignment = c->x86_clflush_size;
704 
705 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
706 	c->extended_cpuid_level = 0;
707 
708 	if (!have_cpuid_p())
709 		identify_cpu_without_cpuid(c);
710 
711 	/* cyrix could have cpuid enabled via c_identify()*/
712 	if (!have_cpuid_p())
713 		return;
714 
715 	cpu_detect(c);
716 	get_cpu_vendor(c);
717 	get_cpu_cap(c);
718 	fpu_detect(c);
719 
720 	if (this_cpu->c_early_init)
721 		this_cpu->c_early_init(c);
722 
723 	c->cpu_index = 0;
724 	filter_cpuid_features(c, false);
725 
726 	if (this_cpu->c_bsp_init)
727 		this_cpu->c_bsp_init(c);
728 
729 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
730 }
731 
732 void __init early_cpu_init(void)
733 {
734 	const struct cpu_dev *const *cdev;
735 	int count = 0;
736 
737 #ifdef CONFIG_PROCESSOR_SELECT
738 	printk(KERN_INFO "KERNEL supported cpus:\n");
739 #endif
740 
741 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
742 		const struct cpu_dev *cpudev = *cdev;
743 
744 		if (count >= X86_VENDOR_NUM)
745 			break;
746 		cpu_devs[count] = cpudev;
747 		count++;
748 
749 #ifdef CONFIG_PROCESSOR_SELECT
750 		{
751 			unsigned int j;
752 
753 			for (j = 0; j < 2; j++) {
754 				if (!cpudev->c_ident[j])
755 					continue;
756 				printk(KERN_INFO "  %s %s\n", cpudev->c_vendor,
757 					cpudev->c_ident[j]);
758 			}
759 		}
760 #endif
761 	}
762 	early_identify_cpu(&boot_cpu_data);
763 }
764 
765 /*
766  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
767  * unfortunately, that's not true in practice because of early VIA
768  * chips and (more importantly) broken virtualizers that are not easy
769  * to detect. In the latter case it doesn't even *fail* reliably, so
770  * probing for it doesn't even work. Disable it completely on 32-bit
771  * unless we can find a reliable way to detect all the broken cases.
772  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
773  */
774 static void detect_nopl(struct cpuinfo_x86 *c)
775 {
776 #ifdef CONFIG_X86_32
777 	clear_cpu_cap(c, X86_FEATURE_NOPL);
778 #else
779 	set_cpu_cap(c, X86_FEATURE_NOPL);
780 #endif
781 }
782 
783 static void generic_identify(struct cpuinfo_x86 *c)
784 {
785 	c->extended_cpuid_level = 0;
786 
787 	if (!have_cpuid_p())
788 		identify_cpu_without_cpuid(c);
789 
790 	/* cyrix could have cpuid enabled via c_identify()*/
791 	if (!have_cpuid_p())
792 		return;
793 
794 	cpu_detect(c);
795 
796 	get_cpu_vendor(c);
797 
798 	get_cpu_cap(c);
799 
800 	if (c->cpuid_level >= 0x00000001) {
801 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
802 #ifdef CONFIG_X86_32
803 # ifdef CONFIG_X86_HT
804 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
805 # else
806 		c->apicid = c->initial_apicid;
807 # endif
808 #endif
809 		c->phys_proc_id = c->initial_apicid;
810 	}
811 
812 	get_model_name(c); /* Default name */
813 
814 	detect_nopl(c);
815 }
816 
817 /*
818  * This does the hard work of actually picking apart the CPU stuff...
819  */
820 static void identify_cpu(struct cpuinfo_x86 *c)
821 {
822 	int i;
823 
824 	c->loops_per_jiffy = loops_per_jiffy;
825 	c->x86_cache_size = -1;
826 	c->x86_vendor = X86_VENDOR_UNKNOWN;
827 	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
828 	c->x86_vendor_id[0] = '\0'; /* Unset */
829 	c->x86_model_id[0] = '\0';  /* Unset */
830 	c->x86_max_cores = 1;
831 	c->x86_coreid_bits = 0;
832 #ifdef CONFIG_X86_64
833 	c->x86_clflush_size = 64;
834 	c->x86_phys_bits = 36;
835 	c->x86_virt_bits = 48;
836 #else
837 	c->cpuid_level = -1;	/* CPUID not detected */
838 	c->x86_clflush_size = 32;
839 	c->x86_phys_bits = 32;
840 	c->x86_virt_bits = 32;
841 #endif
842 	c->x86_cache_alignment = c->x86_clflush_size;
843 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
844 
845 	generic_identify(c);
846 
847 	if (this_cpu->c_identify)
848 		this_cpu->c_identify(c);
849 
850 	/* Clear/Set all flags overriden by options, after probe */
851 	for (i = 0; i < NCAPINTS; i++) {
852 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
853 		c->x86_capability[i] |= cpu_caps_set[i];
854 	}
855 
856 #ifdef CONFIG_X86_64
857 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
858 #endif
859 
860 	/*
861 	 * Vendor-specific initialization.  In this section we
862 	 * canonicalize the feature flags, meaning if there are
863 	 * features a certain CPU supports which CPUID doesn't
864 	 * tell us, CPUID claiming incorrect flags, or other bugs,
865 	 * we handle them here.
866 	 *
867 	 * At the end of this section, c->x86_capability better
868 	 * indicate the features this CPU genuinely supports!
869 	 */
870 	if (this_cpu->c_init)
871 		this_cpu->c_init(c);
872 
873 	/* Disable the PN if appropriate */
874 	squash_the_stupid_serial_number(c);
875 
876 	/* Set up SMEP/SMAP */
877 	setup_smep(c);
878 	setup_smap(c);
879 
880 	/*
881 	 * The vendor-specific functions might have changed features.
882 	 * Now we do "generic changes."
883 	 */
884 
885 	/* Filter out anything that depends on CPUID levels we don't have */
886 	filter_cpuid_features(c, true);
887 
888 	/* If the model name is still unset, do table lookup. */
889 	if (!c->x86_model_id[0]) {
890 		const char *p;
891 		p = table_lookup_model(c);
892 		if (p)
893 			strcpy(c->x86_model_id, p);
894 		else
895 			/* Last resort... */
896 			sprintf(c->x86_model_id, "%02x/%02x",
897 				c->x86, c->x86_model);
898 	}
899 
900 #ifdef CONFIG_X86_64
901 	detect_ht(c);
902 #endif
903 
904 	init_hypervisor(c);
905 	x86_init_rdrand(c);
906 
907 	/*
908 	 * Clear/Set all flags overriden by options, need do it
909 	 * before following smp all cpus cap AND.
910 	 */
911 	for (i = 0; i < NCAPINTS; i++) {
912 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
913 		c->x86_capability[i] |= cpu_caps_set[i];
914 	}
915 
916 	/*
917 	 * On SMP, boot_cpu_data holds the common feature set between
918 	 * all CPUs; so make sure that we indicate which features are
919 	 * common between the CPUs.  The first time this routine gets
920 	 * executed, c == &boot_cpu_data.
921 	 */
922 	if (c != &boot_cpu_data) {
923 		/* AND the already accumulated flags with these */
924 		for (i = 0; i < NCAPINTS; i++)
925 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
926 
927 		/* OR, i.e. replicate the bug flags */
928 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
929 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
930 	}
931 
932 	/* Init Machine Check Exception if available. */
933 	mcheck_cpu_init(c);
934 
935 	select_idle_routine(c);
936 
937 #ifdef CONFIG_NUMA
938 	numa_add_cpu(smp_processor_id());
939 #endif
940 }
941 
942 #ifdef CONFIG_X86_64
943 static void vgetcpu_set_mode(void)
944 {
945 	if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
946 		vgetcpu_mode = VGETCPU_RDTSCP;
947 	else
948 		vgetcpu_mode = VGETCPU_LSL;
949 }
950 #endif
951 
952 void __init identify_boot_cpu(void)
953 {
954 	identify_cpu(&boot_cpu_data);
955 	init_amd_e400_c1e_mask();
956 #ifdef CONFIG_X86_32
957 	sysenter_setup();
958 	enable_sep_cpu();
959 #else
960 	vgetcpu_set_mode();
961 #endif
962 	cpu_detect_tlb(&boot_cpu_data);
963 }
964 
965 void identify_secondary_cpu(struct cpuinfo_x86 *c)
966 {
967 	BUG_ON(c == &boot_cpu_data);
968 	identify_cpu(c);
969 #ifdef CONFIG_X86_32
970 	enable_sep_cpu();
971 #endif
972 	mtrr_ap_init();
973 }
974 
975 struct msr_range {
976 	unsigned	min;
977 	unsigned	max;
978 };
979 
980 static const struct msr_range msr_range_array[] = {
981 	{ 0x00000000, 0x00000418},
982 	{ 0xc0000000, 0xc000040b},
983 	{ 0xc0010000, 0xc0010142},
984 	{ 0xc0011000, 0xc001103b},
985 };
986 
987 static void __print_cpu_msr(void)
988 {
989 	unsigned index_min, index_max;
990 	unsigned index;
991 	u64 val;
992 	int i;
993 
994 	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
995 		index_min = msr_range_array[i].min;
996 		index_max = msr_range_array[i].max;
997 
998 		for (index = index_min; index < index_max; index++) {
999 			if (rdmsrl_safe(index, &val))
1000 				continue;
1001 			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
1002 		}
1003 	}
1004 }
1005 
1006 static int show_msr;
1007 
1008 static __init int setup_show_msr(char *arg)
1009 {
1010 	int num;
1011 
1012 	get_option(&arg, &num);
1013 
1014 	if (num > 0)
1015 		show_msr = num;
1016 	return 1;
1017 }
1018 __setup("show_msr=", setup_show_msr);
1019 
1020 static __init int setup_noclflush(char *arg)
1021 {
1022 	setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
1023 	return 1;
1024 }
1025 __setup("noclflush", setup_noclflush);
1026 
1027 void print_cpu_info(struct cpuinfo_x86 *c)
1028 {
1029 	const char *vendor = NULL;
1030 
1031 	if (c->x86_vendor < X86_VENDOR_NUM) {
1032 		vendor = this_cpu->c_vendor;
1033 	} else {
1034 		if (c->cpuid_level >= 0)
1035 			vendor = c->x86_vendor_id;
1036 	}
1037 
1038 	if (vendor && !strstr(c->x86_model_id, vendor))
1039 		printk(KERN_CONT "%s ", vendor);
1040 
1041 	if (c->x86_model_id[0])
1042 		printk(KERN_CONT "%s", strim(c->x86_model_id));
1043 	else
1044 		printk(KERN_CONT "%d86", c->x86);
1045 
1046 	printk(KERN_CONT " (fam: %02x, model: %02x", c->x86, c->x86_model);
1047 
1048 	if (c->x86_mask || c->cpuid_level >= 0)
1049 		printk(KERN_CONT ", stepping: %02x)\n", c->x86_mask);
1050 	else
1051 		printk(KERN_CONT ")\n");
1052 
1053 	print_cpu_msr(c);
1054 }
1055 
1056 void print_cpu_msr(struct cpuinfo_x86 *c)
1057 {
1058 	if (c->cpu_index < show_msr)
1059 		__print_cpu_msr();
1060 }
1061 
1062 static __init int setup_disablecpuid(char *arg)
1063 {
1064 	int bit;
1065 
1066 	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1067 		setup_clear_cpu_cap(bit);
1068 	else
1069 		return 0;
1070 
1071 	return 1;
1072 }
1073 __setup("clearcpuid=", setup_disablecpuid);
1074 
1075 #ifdef CONFIG_X86_64
1076 struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
1077 struct desc_ptr debug_idt_descr = { NR_VECTORS * 16 - 1,
1078 				    (unsigned long) debug_idt_table };
1079 
1080 DEFINE_PER_CPU_FIRST(union irq_stack_union,
1081 		     irq_stack_union) __aligned(PAGE_SIZE) __visible;
1082 
1083 /*
1084  * The following four percpu variables are hot.  Align current_task to
1085  * cacheline size such that all four fall in the same cacheline.
1086  */
1087 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1088 	&init_task;
1089 EXPORT_PER_CPU_SYMBOL(current_task);
1090 
1091 DEFINE_PER_CPU(unsigned long, kernel_stack) =
1092 	(unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
1093 EXPORT_PER_CPU_SYMBOL(kernel_stack);
1094 
1095 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1096 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1097 
1098 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1099 
1100 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1101 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1102 
1103 DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1104 
1105 /*
1106  * Special IST stacks which the CPU switches to when it calls
1107  * an IST-marked descriptor entry. Up to 7 stacks (hardware
1108  * limit), all of them are 4K, except the debug stack which
1109  * is 8K.
1110  */
1111 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1112 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1113 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1114 };
1115 
1116 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1117 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1118 
1119 /* May not be marked __init: used by software suspend */
1120 void syscall_init(void)
1121 {
1122 	/*
1123 	 * LSTAR and STAR live in a bit strange symbiosis.
1124 	 * They both write to the same internal register. STAR allows to
1125 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1126 	 */
1127 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1128 	wrmsrl(MSR_LSTAR, system_call);
1129 	wrmsrl(MSR_CSTAR, ignore_sysret);
1130 
1131 #ifdef CONFIG_IA32_EMULATION
1132 	syscall32_cpu_init();
1133 #endif
1134 
1135 	/* Flags to clear on syscall */
1136 	wrmsrl(MSR_SYSCALL_MASK,
1137 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1138 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC);
1139 }
1140 
1141 /*
1142  * Copies of the original ist values from the tss are only accessed during
1143  * debugging, no special alignment required.
1144  */
1145 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1146 
1147 static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1148 DEFINE_PER_CPU(int, debug_stack_usage);
1149 
1150 int is_debug_stack(unsigned long addr)
1151 {
1152 	return __get_cpu_var(debug_stack_usage) ||
1153 		(addr <= __get_cpu_var(debug_stack_addr) &&
1154 		 addr > (__get_cpu_var(debug_stack_addr) - DEBUG_STKSZ));
1155 }
1156 
1157 DEFINE_PER_CPU(u32, debug_idt_ctr);
1158 
1159 void debug_stack_set_zero(void)
1160 {
1161 	this_cpu_inc(debug_idt_ctr);
1162 	load_current_idt();
1163 }
1164 
1165 void debug_stack_reset(void)
1166 {
1167 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1168 		return;
1169 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1170 		load_current_idt();
1171 }
1172 
1173 #else	/* CONFIG_X86_64 */
1174 
1175 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1176 EXPORT_PER_CPU_SYMBOL(current_task);
1177 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1178 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1179 DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1180 
1181 #ifdef CONFIG_CC_STACKPROTECTOR
1182 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1183 #endif
1184 
1185 #endif	/* CONFIG_X86_64 */
1186 
1187 /*
1188  * Clear all 6 debug registers:
1189  */
1190 static void clear_all_debug_regs(void)
1191 {
1192 	int i;
1193 
1194 	for (i = 0; i < 8; i++) {
1195 		/* Ignore db4, db5 */
1196 		if ((i == 4) || (i == 5))
1197 			continue;
1198 
1199 		set_debugreg(0, i);
1200 	}
1201 }
1202 
1203 #ifdef CONFIG_KGDB
1204 /*
1205  * Restore debug regs if using kgdbwait and you have a kernel debugger
1206  * connection established.
1207  */
1208 static void dbg_restore_debug_regs(void)
1209 {
1210 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1211 		arch_kgdb_ops.correct_hw_break();
1212 }
1213 #else /* ! CONFIG_KGDB */
1214 #define dbg_restore_debug_regs()
1215 #endif /* ! CONFIG_KGDB */
1216 
1217 /*
1218  * cpu_init() initializes state that is per-CPU. Some data is already
1219  * initialized (naturally) in the bootstrap process, such as the GDT
1220  * and IDT. We reload them nevertheless, this function acts as a
1221  * 'CPU state barrier', nothing should get across.
1222  * A lot of state is already set up in PDA init for 64 bit
1223  */
1224 #ifdef CONFIG_X86_64
1225 
1226 void cpu_init(void)
1227 {
1228 	struct orig_ist *oist;
1229 	struct task_struct *me;
1230 	struct tss_struct *t;
1231 	unsigned long v;
1232 	int cpu;
1233 	int i;
1234 
1235 	/*
1236 	 * Load microcode on this cpu if a valid microcode is available.
1237 	 * This is early microcode loading procedure.
1238 	 */
1239 	load_ucode_ap();
1240 
1241 	cpu = stack_smp_processor_id();
1242 	t = &per_cpu(init_tss, cpu);
1243 	oist = &per_cpu(orig_ist, cpu);
1244 
1245 #ifdef CONFIG_NUMA
1246 	if (this_cpu_read(numa_node) == 0 &&
1247 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1248 		set_numa_node(early_cpu_to_node(cpu));
1249 #endif
1250 
1251 	me = current;
1252 
1253 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1254 		panic("CPU#%d already initialized!\n", cpu);
1255 
1256 	pr_debug("Initializing CPU#%d\n", cpu);
1257 
1258 	clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1259 
1260 	/*
1261 	 * Initialize the per-CPU GDT with the boot GDT,
1262 	 * and set up the GDT descriptor:
1263 	 */
1264 
1265 	switch_to_new_gdt(cpu);
1266 	loadsegment(fs, 0);
1267 
1268 	load_current_idt();
1269 
1270 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1271 	syscall_init();
1272 
1273 	wrmsrl(MSR_FS_BASE, 0);
1274 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1275 	barrier();
1276 
1277 	x86_configure_nx();
1278 	enable_x2apic();
1279 
1280 	/*
1281 	 * set up and load the per-CPU TSS
1282 	 */
1283 	if (!oist->ist[0]) {
1284 		char *estacks = per_cpu(exception_stacks, cpu);
1285 
1286 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1287 			estacks += exception_stack_sizes[v];
1288 			oist->ist[v] = t->x86_tss.ist[v] =
1289 					(unsigned long)estacks;
1290 			if (v == DEBUG_STACK-1)
1291 				per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1292 		}
1293 	}
1294 
1295 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1296 
1297 	/*
1298 	 * <= is required because the CPU will access up to
1299 	 * 8 bits beyond the end of the IO permission bitmap.
1300 	 */
1301 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1302 		t->io_bitmap[i] = ~0UL;
1303 
1304 	atomic_inc(&init_mm.mm_count);
1305 	me->active_mm = &init_mm;
1306 	BUG_ON(me->mm);
1307 	enter_lazy_tlb(&init_mm, me);
1308 
1309 	load_sp0(t, &current->thread);
1310 	set_tss_desc(cpu, t);
1311 	load_TR_desc();
1312 	load_LDT(&init_mm.context);
1313 
1314 	clear_all_debug_regs();
1315 	dbg_restore_debug_regs();
1316 
1317 	fpu_init();
1318 
1319 	if (is_uv_system())
1320 		uv_cpu_init();
1321 }
1322 
1323 #else
1324 
1325 void cpu_init(void)
1326 {
1327 	int cpu = smp_processor_id();
1328 	struct task_struct *curr = current;
1329 	struct tss_struct *t = &per_cpu(init_tss, cpu);
1330 	struct thread_struct *thread = &curr->thread;
1331 
1332 	show_ucode_info_early();
1333 
1334 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1335 		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1336 		for (;;)
1337 			local_irq_enable();
1338 	}
1339 
1340 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1341 
1342 	if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1343 		clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1344 
1345 	load_current_idt();
1346 	switch_to_new_gdt(cpu);
1347 
1348 	/*
1349 	 * Set up and load the per-CPU TSS and LDT
1350 	 */
1351 	atomic_inc(&init_mm.mm_count);
1352 	curr->active_mm = &init_mm;
1353 	BUG_ON(curr->mm);
1354 	enter_lazy_tlb(&init_mm, curr);
1355 
1356 	load_sp0(t, thread);
1357 	set_tss_desc(cpu, t);
1358 	load_TR_desc();
1359 	load_LDT(&init_mm.context);
1360 
1361 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1362 
1363 #ifdef CONFIG_DOUBLEFAULT
1364 	/* Set up doublefault TSS pointer in the GDT */
1365 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1366 #endif
1367 
1368 	clear_all_debug_regs();
1369 	dbg_restore_debug_regs();
1370 
1371 	fpu_init();
1372 }
1373 #endif
1374 
1375 #ifdef CONFIG_X86_DEBUG_STATIC_CPU_HAS
1376 void warn_pre_alternatives(void)
1377 {
1378 	WARN(1, "You're using static_cpu_has before alternatives have run!\n");
1379 }
1380 EXPORT_SYMBOL_GPL(warn_pre_alternatives);
1381 #endif
1382 
1383 inline bool __static_cpu_has_safe(u16 bit)
1384 {
1385 	return boot_cpu_has(bit);
1386 }
1387 EXPORT_SYMBOL_GPL(__static_cpu_has_safe);
1388