xref: /linux/arch/x86/kernel/cpu/common.c (revision b889fcf63cb62e7fdb7816565e28f44dbe4a76a5)
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 
41 #ifdef CONFIG_X86_LOCAL_APIC
42 #include <asm/uv/uv.h>
43 #endif
44 
45 #include "cpu.h"
46 
47 /* all of these masks are initialized in setup_cpu_local_masks() */
48 cpumask_var_t cpu_initialized_mask;
49 cpumask_var_t cpu_callout_mask;
50 cpumask_var_t cpu_callin_mask;
51 
52 /* representing cpus for which sibling maps can be computed */
53 cpumask_var_t cpu_sibling_setup_mask;
54 
55 /* correctly size the local cpu masks */
56 void __init setup_cpu_local_masks(void)
57 {
58 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
59 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
60 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
61 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
62 }
63 
64 static void __cpuinit default_init(struct cpuinfo_x86 *c)
65 {
66 #ifdef CONFIG_X86_64
67 	cpu_detect_cache_sizes(c);
68 #else
69 	/* Not much we can do here... */
70 	/* Check if at least it has cpuid */
71 	if (c->cpuid_level == -1) {
72 		/* No cpuid. It must be an ancient CPU */
73 		if (c->x86 == 4)
74 			strcpy(c->x86_model_id, "486");
75 		else if (c->x86 == 3)
76 			strcpy(c->x86_model_id, "386");
77 	}
78 #endif
79 }
80 
81 static const struct cpu_dev __cpuinitconst default_cpu = {
82 	.c_init		= default_init,
83 	.c_vendor	= "Unknown",
84 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
85 };
86 
87 static const struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
88 
89 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
90 #ifdef CONFIG_X86_64
91 	/*
92 	 * We need valid kernel segments for data and code in long mode too
93 	 * IRET will check the segment types  kkeil 2000/10/28
94 	 * Also sysret mandates a special GDT layout
95 	 *
96 	 * TLS descriptors are currently at a different place compared to i386.
97 	 * Hopefully nobody expects them at a fixed place (Wine?)
98 	 */
99 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
100 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
101 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
102 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
103 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
104 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
105 #else
106 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
107 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
108 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
109 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
110 	/*
111 	 * Segments used for calling PnP BIOS have byte granularity.
112 	 * They code segments and data segments have fixed 64k limits,
113 	 * the transfer segment sizes are set at run time.
114 	 */
115 	/* 32-bit code */
116 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
117 	/* 16-bit code */
118 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
119 	/* 16-bit data */
120 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
121 	/* 16-bit data */
122 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
123 	/* 16-bit data */
124 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
125 	/*
126 	 * The APM segments have byte granularity and their bases
127 	 * are set at run time.  All have 64k limits.
128 	 */
129 	/* 32-bit code */
130 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
131 	/* 16-bit code */
132 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
133 	/* data */
134 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
135 
136 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
137 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
138 	GDT_STACK_CANARY_INIT
139 #endif
140 } };
141 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
142 
143 static int __init x86_xsave_setup(char *s)
144 {
145 	setup_clear_cpu_cap(X86_FEATURE_XSAVE);
146 	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
147 	setup_clear_cpu_cap(X86_FEATURE_AVX);
148 	setup_clear_cpu_cap(X86_FEATURE_AVX2);
149 	return 1;
150 }
151 __setup("noxsave", x86_xsave_setup);
152 
153 static int __init x86_xsaveopt_setup(char *s)
154 {
155 	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
156 	return 1;
157 }
158 __setup("noxsaveopt", x86_xsaveopt_setup);
159 
160 #ifdef CONFIG_X86_32
161 static int cachesize_override __cpuinitdata = -1;
162 static int disable_x86_serial_nr __cpuinitdata = 1;
163 
164 static int __init cachesize_setup(char *str)
165 {
166 	get_option(&str, &cachesize_override);
167 	return 1;
168 }
169 __setup("cachesize=", cachesize_setup);
170 
171 static int __init x86_fxsr_setup(char *s)
172 {
173 	setup_clear_cpu_cap(X86_FEATURE_FXSR);
174 	setup_clear_cpu_cap(X86_FEATURE_XMM);
175 	return 1;
176 }
177 __setup("nofxsr", x86_fxsr_setup);
178 
179 static int __init x86_sep_setup(char *s)
180 {
181 	setup_clear_cpu_cap(X86_FEATURE_SEP);
182 	return 1;
183 }
184 __setup("nosep", x86_sep_setup);
185 
186 /* Standard macro to see if a specific flag is changeable */
187 static inline int flag_is_changeable_p(u32 flag)
188 {
189 	u32 f1, f2;
190 
191 	/*
192 	 * Cyrix and IDT cpus allow disabling of CPUID
193 	 * so the code below may return different results
194 	 * when it is executed before and after enabling
195 	 * the CPUID. Add "volatile" to not allow gcc to
196 	 * optimize the subsequent calls to this function.
197 	 */
198 	asm volatile ("pushfl		\n\t"
199 		      "pushfl		\n\t"
200 		      "popl %0		\n\t"
201 		      "movl %0, %1	\n\t"
202 		      "xorl %2, %0	\n\t"
203 		      "pushl %0		\n\t"
204 		      "popfl		\n\t"
205 		      "pushfl		\n\t"
206 		      "popl %0		\n\t"
207 		      "popfl		\n\t"
208 
209 		      : "=&r" (f1), "=&r" (f2)
210 		      : "ir" (flag));
211 
212 	return ((f1^f2) & flag) != 0;
213 }
214 
215 /* Probe for the CPUID instruction */
216 static int __cpuinit have_cpuid_p(void)
217 {
218 	return flag_is_changeable_p(X86_EFLAGS_ID);
219 }
220 
221 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
222 {
223 	unsigned long lo, hi;
224 
225 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
226 		return;
227 
228 	/* Disable processor serial number: */
229 
230 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
231 	lo |= 0x200000;
232 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
233 
234 	printk(KERN_NOTICE "CPU serial number disabled.\n");
235 	clear_cpu_cap(c, X86_FEATURE_PN);
236 
237 	/* Disabling the serial number may affect the cpuid level */
238 	c->cpuid_level = cpuid_eax(0);
239 }
240 
241 static int __init x86_serial_nr_setup(char *s)
242 {
243 	disable_x86_serial_nr = 0;
244 	return 1;
245 }
246 __setup("serialnumber", x86_serial_nr_setup);
247 #else
248 static inline int flag_is_changeable_p(u32 flag)
249 {
250 	return 1;
251 }
252 /* Probe for the CPUID instruction */
253 static inline int have_cpuid_p(void)
254 {
255 	return 1;
256 }
257 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
258 {
259 }
260 #endif
261 
262 static __init int setup_disable_smep(char *arg)
263 {
264 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
265 	return 1;
266 }
267 __setup("nosmep", setup_disable_smep);
268 
269 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
270 {
271 	if (cpu_has(c, X86_FEATURE_SMEP))
272 		set_in_cr4(X86_CR4_SMEP);
273 }
274 
275 static __init int setup_disable_smap(char *arg)
276 {
277 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
278 	return 1;
279 }
280 __setup("nosmap", setup_disable_smap);
281 
282 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
283 {
284 	unsigned long eflags;
285 
286 	/* This should have been cleared long ago */
287 	raw_local_save_flags(eflags);
288 	BUG_ON(eflags & X86_EFLAGS_AC);
289 
290 	if (cpu_has(c, X86_FEATURE_SMAP))
291 		set_in_cr4(X86_CR4_SMAP);
292 }
293 
294 /*
295  * Some CPU features depend on higher CPUID levels, which may not always
296  * be available due to CPUID level capping or broken virtualization
297  * software.  Add those features to this table to auto-disable them.
298  */
299 struct cpuid_dependent_feature {
300 	u32 feature;
301 	u32 level;
302 };
303 
304 static const struct cpuid_dependent_feature __cpuinitconst
305 cpuid_dependent_features[] = {
306 	{ X86_FEATURE_MWAIT,		0x00000005 },
307 	{ X86_FEATURE_DCA,		0x00000009 },
308 	{ X86_FEATURE_XSAVE,		0x0000000d },
309 	{ 0, 0 }
310 };
311 
312 static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
313 {
314 	const struct cpuid_dependent_feature *df;
315 
316 	for (df = cpuid_dependent_features; df->feature; df++) {
317 
318 		if (!cpu_has(c, df->feature))
319 			continue;
320 		/*
321 		 * Note: cpuid_level is set to -1 if unavailable, but
322 		 * extended_extended_level is set to 0 if unavailable
323 		 * and the legitimate extended levels are all negative
324 		 * when signed; hence the weird messing around with
325 		 * signs here...
326 		 */
327 		if (!((s32)df->level < 0 ?
328 		     (u32)df->level > (u32)c->extended_cpuid_level :
329 		     (s32)df->level > (s32)c->cpuid_level))
330 			continue;
331 
332 		clear_cpu_cap(c, df->feature);
333 		if (!warn)
334 			continue;
335 
336 		printk(KERN_WARNING
337 		       "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
338 				x86_cap_flags[df->feature], df->level);
339 	}
340 }
341 
342 /*
343  * Naming convention should be: <Name> [(<Codename>)]
344  * This table only is used unless init_<vendor>() below doesn't set it;
345  * in particular, if CPUID levels 0x80000002..4 are supported, this
346  * isn't used
347  */
348 
349 /* Look up CPU names by table lookup. */
350 static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
351 {
352 	const struct cpu_model_info *info;
353 
354 	if (c->x86_model >= 16)
355 		return NULL;	/* Range check */
356 
357 	if (!this_cpu)
358 		return NULL;
359 
360 	info = this_cpu->c_models;
361 
362 	while (info && info->family) {
363 		if (info->family == c->x86)
364 			return info->model_names[c->x86_model];
365 		info++;
366 	}
367 	return NULL;		/* Not found */
368 }
369 
370 __u32 cpu_caps_cleared[NCAPINTS] __cpuinitdata;
371 __u32 cpu_caps_set[NCAPINTS] __cpuinitdata;
372 
373 void load_percpu_segment(int cpu)
374 {
375 #ifdef CONFIG_X86_32
376 	loadsegment(fs, __KERNEL_PERCPU);
377 #else
378 	loadsegment(gs, 0);
379 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
380 #endif
381 	load_stack_canary_segment();
382 }
383 
384 /*
385  * Current gdt points %fs at the "master" per-cpu area: after this,
386  * it's on the real one.
387  */
388 void switch_to_new_gdt(int cpu)
389 {
390 	struct desc_ptr gdt_descr;
391 
392 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
393 	gdt_descr.size = GDT_SIZE - 1;
394 	load_gdt(&gdt_descr);
395 	/* Reload the per-cpu base */
396 
397 	load_percpu_segment(cpu);
398 }
399 
400 static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
401 
402 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
403 {
404 	unsigned int *v;
405 	char *p, *q;
406 
407 	if (c->extended_cpuid_level < 0x80000004)
408 		return;
409 
410 	v = (unsigned int *)c->x86_model_id;
411 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
412 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
413 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
414 	c->x86_model_id[48] = 0;
415 
416 	/*
417 	 * Intel chips right-justify this string for some dumb reason;
418 	 * undo that brain damage:
419 	 */
420 	p = q = &c->x86_model_id[0];
421 	while (*p == ' ')
422 		p++;
423 	if (p != q) {
424 		while (*p)
425 			*q++ = *p++;
426 		while (q <= &c->x86_model_id[48])
427 			*q++ = '\0';	/* Zero-pad the rest */
428 	}
429 }
430 
431 void __cpuinit cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
432 {
433 	unsigned int n, dummy, ebx, ecx, edx, l2size;
434 
435 	n = c->extended_cpuid_level;
436 
437 	if (n >= 0x80000005) {
438 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
439 		c->x86_cache_size = (ecx>>24) + (edx>>24);
440 #ifdef CONFIG_X86_64
441 		/* On K8 L1 TLB is inclusive, so don't count it */
442 		c->x86_tlbsize = 0;
443 #endif
444 	}
445 
446 	if (n < 0x80000006)	/* Some chips just has a large L1. */
447 		return;
448 
449 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
450 	l2size = ecx >> 16;
451 
452 #ifdef CONFIG_X86_64
453 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
454 #else
455 	/* do processor-specific cache resizing */
456 	if (this_cpu->c_size_cache)
457 		l2size = this_cpu->c_size_cache(c, l2size);
458 
459 	/* Allow user to override all this if necessary. */
460 	if (cachesize_override != -1)
461 		l2size = cachesize_override;
462 
463 	if (l2size == 0)
464 		return;		/* Again, no L2 cache is possible */
465 #endif
466 
467 	c->x86_cache_size = l2size;
468 }
469 
470 u16 __read_mostly tlb_lli_4k[NR_INFO];
471 u16 __read_mostly tlb_lli_2m[NR_INFO];
472 u16 __read_mostly tlb_lli_4m[NR_INFO];
473 u16 __read_mostly tlb_lld_4k[NR_INFO];
474 u16 __read_mostly tlb_lld_2m[NR_INFO];
475 u16 __read_mostly tlb_lld_4m[NR_INFO];
476 
477 /*
478  * tlb_flushall_shift shows the balance point in replacing cr3 write
479  * with multiple 'invlpg'. It will do this replacement when
480  *   flush_tlb_lines <= active_lines/2^tlb_flushall_shift.
481  * If tlb_flushall_shift is -1, means the replacement will be disabled.
482  */
483 s8  __read_mostly tlb_flushall_shift = -1;
484 
485 void __cpuinit cpu_detect_tlb(struct cpuinfo_x86 *c)
486 {
487 	if (this_cpu->c_detect_tlb)
488 		this_cpu->c_detect_tlb(c);
489 
490 	printk(KERN_INFO "Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n" \
491 		"Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d\n"	     \
492 		"tlb_flushall_shift: %d\n",
493 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
494 		tlb_lli_4m[ENTRIES], tlb_lld_4k[ENTRIES],
495 		tlb_lld_2m[ENTRIES], tlb_lld_4m[ENTRIES],
496 		tlb_flushall_shift);
497 }
498 
499 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
500 {
501 #ifdef CONFIG_X86_HT
502 	u32 eax, ebx, ecx, edx;
503 	int index_msb, core_bits;
504 	static bool printed;
505 
506 	if (!cpu_has(c, X86_FEATURE_HT))
507 		return;
508 
509 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
510 		goto out;
511 
512 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
513 		return;
514 
515 	cpuid(1, &eax, &ebx, &ecx, &edx);
516 
517 	smp_num_siblings = (ebx & 0xff0000) >> 16;
518 
519 	if (smp_num_siblings == 1) {
520 		printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
521 		goto out;
522 	}
523 
524 	if (smp_num_siblings <= 1)
525 		goto out;
526 
527 	index_msb = get_count_order(smp_num_siblings);
528 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
529 
530 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
531 
532 	index_msb = get_count_order(smp_num_siblings);
533 
534 	core_bits = get_count_order(c->x86_max_cores);
535 
536 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
537 				       ((1 << core_bits) - 1);
538 
539 out:
540 	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
541 		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
542 		       c->phys_proc_id);
543 		printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
544 		       c->cpu_core_id);
545 		printed = 1;
546 	}
547 #endif
548 }
549 
550 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
551 {
552 	char *v = c->x86_vendor_id;
553 	int i;
554 
555 	for (i = 0; i < X86_VENDOR_NUM; i++) {
556 		if (!cpu_devs[i])
557 			break;
558 
559 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
560 		    (cpu_devs[i]->c_ident[1] &&
561 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
562 
563 			this_cpu = cpu_devs[i];
564 			c->x86_vendor = this_cpu->c_x86_vendor;
565 			return;
566 		}
567 	}
568 
569 	printk_once(KERN_ERR
570 			"CPU: vendor_id '%s' unknown, using generic init.\n" \
571 			"CPU: Your system may be unstable.\n", v);
572 
573 	c->x86_vendor = X86_VENDOR_UNKNOWN;
574 	this_cpu = &default_cpu;
575 }
576 
577 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
578 {
579 	/* Get vendor name */
580 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
581 	      (unsigned int *)&c->x86_vendor_id[0],
582 	      (unsigned int *)&c->x86_vendor_id[8],
583 	      (unsigned int *)&c->x86_vendor_id[4]);
584 
585 	c->x86 = 4;
586 	/* Intel-defined flags: level 0x00000001 */
587 	if (c->cpuid_level >= 0x00000001) {
588 		u32 junk, tfms, cap0, misc;
589 
590 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
591 		c->x86 = (tfms >> 8) & 0xf;
592 		c->x86_model = (tfms >> 4) & 0xf;
593 		c->x86_mask = tfms & 0xf;
594 
595 		if (c->x86 == 0xf)
596 			c->x86 += (tfms >> 20) & 0xff;
597 		if (c->x86 >= 0x6)
598 			c->x86_model += ((tfms >> 16) & 0xf) << 4;
599 
600 		if (cap0 & (1<<19)) {
601 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
602 			c->x86_cache_alignment = c->x86_clflush_size;
603 		}
604 	}
605 }
606 
607 void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
608 {
609 	u32 tfms, xlvl;
610 	u32 ebx;
611 
612 	/* Intel-defined flags: level 0x00000001 */
613 	if (c->cpuid_level >= 0x00000001) {
614 		u32 capability, excap;
615 
616 		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
617 		c->x86_capability[0] = capability;
618 		c->x86_capability[4] = excap;
619 	}
620 
621 	/* Additional Intel-defined flags: level 0x00000007 */
622 	if (c->cpuid_level >= 0x00000007) {
623 		u32 eax, ebx, ecx, edx;
624 
625 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
626 
627 		c->x86_capability[9] = ebx;
628 	}
629 
630 	/* AMD-defined flags: level 0x80000001 */
631 	xlvl = cpuid_eax(0x80000000);
632 	c->extended_cpuid_level = xlvl;
633 
634 	if ((xlvl & 0xffff0000) == 0x80000000) {
635 		if (xlvl >= 0x80000001) {
636 			c->x86_capability[1] = cpuid_edx(0x80000001);
637 			c->x86_capability[6] = cpuid_ecx(0x80000001);
638 		}
639 	}
640 
641 	if (c->extended_cpuid_level >= 0x80000008) {
642 		u32 eax = cpuid_eax(0x80000008);
643 
644 		c->x86_virt_bits = (eax >> 8) & 0xff;
645 		c->x86_phys_bits = eax & 0xff;
646 	}
647 #ifdef CONFIG_X86_32
648 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
649 		c->x86_phys_bits = 36;
650 #endif
651 
652 	if (c->extended_cpuid_level >= 0x80000007)
653 		c->x86_power = cpuid_edx(0x80000007);
654 
655 	init_scattered_cpuid_features(c);
656 }
657 
658 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
659 {
660 #ifdef CONFIG_X86_32
661 	int i;
662 
663 	/*
664 	 * First of all, decide if this is a 486 or higher
665 	 * It's a 486 if we can modify the AC flag
666 	 */
667 	if (flag_is_changeable_p(X86_EFLAGS_AC))
668 		c->x86 = 4;
669 	else
670 		c->x86 = 3;
671 
672 	for (i = 0; i < X86_VENDOR_NUM; i++)
673 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
674 			c->x86_vendor_id[0] = 0;
675 			cpu_devs[i]->c_identify(c);
676 			if (c->x86_vendor_id[0]) {
677 				get_cpu_vendor(c);
678 				break;
679 			}
680 		}
681 #endif
682 }
683 
684 /*
685  * Do minimum CPU detection early.
686  * Fields really needed: vendor, cpuid_level, family, model, mask,
687  * cache alignment.
688  * The others are not touched to avoid unwanted side effects.
689  *
690  * WARNING: this function is only called on the BP.  Don't add code here
691  * that is supposed to run on all CPUs.
692  */
693 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
694 {
695 #ifdef CONFIG_X86_64
696 	c->x86_clflush_size = 64;
697 	c->x86_phys_bits = 36;
698 	c->x86_virt_bits = 48;
699 #else
700 	c->x86_clflush_size = 32;
701 	c->x86_phys_bits = 32;
702 	c->x86_virt_bits = 32;
703 #endif
704 	c->x86_cache_alignment = c->x86_clflush_size;
705 
706 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
707 	c->extended_cpuid_level = 0;
708 
709 	if (!have_cpuid_p())
710 		identify_cpu_without_cpuid(c);
711 
712 	/* cyrix could have cpuid enabled via c_identify()*/
713 	if (!have_cpuid_p())
714 		return;
715 
716 	cpu_detect(c);
717 
718 	get_cpu_vendor(c);
719 
720 	get_cpu_cap(c);
721 
722 	if (this_cpu->c_early_init)
723 		this_cpu->c_early_init(c);
724 
725 	c->cpu_index = 0;
726 	filter_cpuid_features(c, false);
727 
728 	if (this_cpu->c_bsp_init)
729 		this_cpu->c_bsp_init(c);
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 __cpuinit 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 __cpuinit 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 __cpuinit 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 
928 	/* Init Machine Check Exception if available. */
929 	mcheck_cpu_init(c);
930 
931 	select_idle_routine(c);
932 
933 #ifdef CONFIG_NUMA
934 	numa_add_cpu(smp_processor_id());
935 #endif
936 }
937 
938 #ifdef CONFIG_X86_64
939 static void vgetcpu_set_mode(void)
940 {
941 	if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
942 		vgetcpu_mode = VGETCPU_RDTSCP;
943 	else
944 		vgetcpu_mode = VGETCPU_LSL;
945 }
946 #endif
947 
948 void __init identify_boot_cpu(void)
949 {
950 	identify_cpu(&boot_cpu_data);
951 	init_amd_e400_c1e_mask();
952 #ifdef CONFIG_X86_32
953 	sysenter_setup();
954 	enable_sep_cpu();
955 #else
956 	vgetcpu_set_mode();
957 #endif
958 	cpu_detect_tlb(&boot_cpu_data);
959 }
960 
961 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
962 {
963 	BUG_ON(c == &boot_cpu_data);
964 	identify_cpu(c);
965 #ifdef CONFIG_X86_32
966 	enable_sep_cpu();
967 #endif
968 	mtrr_ap_init();
969 }
970 
971 struct msr_range {
972 	unsigned	min;
973 	unsigned	max;
974 };
975 
976 static const struct msr_range msr_range_array[] __cpuinitconst = {
977 	{ 0x00000000, 0x00000418},
978 	{ 0xc0000000, 0xc000040b},
979 	{ 0xc0010000, 0xc0010142},
980 	{ 0xc0011000, 0xc001103b},
981 };
982 
983 static void __cpuinit __print_cpu_msr(void)
984 {
985 	unsigned index_min, index_max;
986 	unsigned index;
987 	u64 val;
988 	int i;
989 
990 	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
991 		index_min = msr_range_array[i].min;
992 		index_max = msr_range_array[i].max;
993 
994 		for (index = index_min; index < index_max; index++) {
995 			if (rdmsrl_safe(index, &val))
996 				continue;
997 			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
998 		}
999 	}
1000 }
1001 
1002 static int show_msr __cpuinitdata;
1003 
1004 static __init int setup_show_msr(char *arg)
1005 {
1006 	int num;
1007 
1008 	get_option(&arg, &num);
1009 
1010 	if (num > 0)
1011 		show_msr = num;
1012 	return 1;
1013 }
1014 __setup("show_msr=", setup_show_msr);
1015 
1016 static __init int setup_noclflush(char *arg)
1017 {
1018 	setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
1019 	return 1;
1020 }
1021 __setup("noclflush", setup_noclflush);
1022 
1023 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
1024 {
1025 	const char *vendor = NULL;
1026 
1027 	if (c->x86_vendor < X86_VENDOR_NUM) {
1028 		vendor = this_cpu->c_vendor;
1029 	} else {
1030 		if (c->cpuid_level >= 0)
1031 			vendor = c->x86_vendor_id;
1032 	}
1033 
1034 	if (vendor && !strstr(c->x86_model_id, vendor))
1035 		printk(KERN_CONT "%s ", vendor);
1036 
1037 	if (c->x86_model_id[0])
1038 		printk(KERN_CONT "%s", strim(c->x86_model_id));
1039 	else
1040 		printk(KERN_CONT "%d86", c->x86);
1041 
1042 	printk(KERN_CONT " (fam: %02x, model: %02x", c->x86, c->x86_model);
1043 
1044 	if (c->x86_mask || c->cpuid_level >= 0)
1045 		printk(KERN_CONT ", stepping: %02x)\n", c->x86_mask);
1046 	else
1047 		printk(KERN_CONT ")\n");
1048 
1049 	print_cpu_msr(c);
1050 }
1051 
1052 void __cpuinit print_cpu_msr(struct cpuinfo_x86 *c)
1053 {
1054 	if (c->cpu_index < show_msr)
1055 		__print_cpu_msr();
1056 }
1057 
1058 static __init int setup_disablecpuid(char *arg)
1059 {
1060 	int bit;
1061 
1062 	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1063 		setup_clear_cpu_cap(bit);
1064 	else
1065 		return 0;
1066 
1067 	return 1;
1068 }
1069 __setup("clearcpuid=", setup_disablecpuid);
1070 
1071 #ifdef CONFIG_X86_64
1072 struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
1073 struct desc_ptr nmi_idt_descr = { NR_VECTORS * 16 - 1,
1074 				    (unsigned long) nmi_idt_table };
1075 
1076 DEFINE_PER_CPU_FIRST(union irq_stack_union,
1077 		     irq_stack_union) __aligned(PAGE_SIZE);
1078 
1079 /*
1080  * The following four percpu variables are hot.  Align current_task to
1081  * cacheline size such that all four fall in the same cacheline.
1082  */
1083 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1084 	&init_task;
1085 EXPORT_PER_CPU_SYMBOL(current_task);
1086 
1087 DEFINE_PER_CPU(unsigned long, kernel_stack) =
1088 	(unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
1089 EXPORT_PER_CPU_SYMBOL(kernel_stack);
1090 
1091 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1092 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1093 
1094 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
1095 
1096 DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1097 
1098 /*
1099  * Special IST stacks which the CPU switches to when it calls
1100  * an IST-marked descriptor entry. Up to 7 stacks (hardware
1101  * limit), all of them are 4K, except the debug stack which
1102  * is 8K.
1103  */
1104 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1105 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1106 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1107 };
1108 
1109 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1110 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1111 
1112 /* May not be marked __init: used by software suspend */
1113 void syscall_init(void)
1114 {
1115 	/*
1116 	 * LSTAR and STAR live in a bit strange symbiosis.
1117 	 * They both write to the same internal register. STAR allows to
1118 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1119 	 */
1120 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1121 	wrmsrl(MSR_LSTAR, system_call);
1122 	wrmsrl(MSR_CSTAR, ignore_sysret);
1123 
1124 #ifdef CONFIG_IA32_EMULATION
1125 	syscall32_cpu_init();
1126 #endif
1127 
1128 	/* Flags to clear on syscall */
1129 	wrmsrl(MSR_SYSCALL_MASK,
1130 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1131 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC);
1132 }
1133 
1134 /*
1135  * Copies of the original ist values from the tss are only accessed during
1136  * debugging, no special alignment required.
1137  */
1138 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1139 
1140 static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1141 DEFINE_PER_CPU(int, debug_stack_usage);
1142 
1143 int is_debug_stack(unsigned long addr)
1144 {
1145 	return __get_cpu_var(debug_stack_usage) ||
1146 		(addr <= __get_cpu_var(debug_stack_addr) &&
1147 		 addr > (__get_cpu_var(debug_stack_addr) - DEBUG_STKSZ));
1148 }
1149 
1150 static DEFINE_PER_CPU(u32, debug_stack_use_ctr);
1151 
1152 void debug_stack_set_zero(void)
1153 {
1154 	this_cpu_inc(debug_stack_use_ctr);
1155 	load_idt((const struct desc_ptr *)&nmi_idt_descr);
1156 }
1157 
1158 void debug_stack_reset(void)
1159 {
1160 	if (WARN_ON(!this_cpu_read(debug_stack_use_ctr)))
1161 		return;
1162 	if (this_cpu_dec_return(debug_stack_use_ctr) == 0)
1163 		load_idt((const struct desc_ptr *)&idt_descr);
1164 }
1165 
1166 #else	/* CONFIG_X86_64 */
1167 
1168 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1169 EXPORT_PER_CPU_SYMBOL(current_task);
1170 DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1171 
1172 #ifdef CONFIG_CC_STACKPROTECTOR
1173 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1174 #endif
1175 
1176 #endif	/* CONFIG_X86_64 */
1177 
1178 /*
1179  * Clear all 6 debug registers:
1180  */
1181 static void clear_all_debug_regs(void)
1182 {
1183 	int i;
1184 
1185 	for (i = 0; i < 8; i++) {
1186 		/* Ignore db4, db5 */
1187 		if ((i == 4) || (i == 5))
1188 			continue;
1189 
1190 		set_debugreg(0, i);
1191 	}
1192 }
1193 
1194 #ifdef CONFIG_KGDB
1195 /*
1196  * Restore debug regs if using kgdbwait and you have a kernel debugger
1197  * connection established.
1198  */
1199 static void dbg_restore_debug_regs(void)
1200 {
1201 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1202 		arch_kgdb_ops.correct_hw_break();
1203 }
1204 #else /* ! CONFIG_KGDB */
1205 #define dbg_restore_debug_regs()
1206 #endif /* ! CONFIG_KGDB */
1207 
1208 /*
1209  * cpu_init() initializes state that is per-CPU. Some data is already
1210  * initialized (naturally) in the bootstrap process, such as the GDT
1211  * and IDT. We reload them nevertheless, this function acts as a
1212  * 'CPU state barrier', nothing should get across.
1213  * A lot of state is already set up in PDA init for 64 bit
1214  */
1215 #ifdef CONFIG_X86_64
1216 
1217 void __cpuinit cpu_init(void)
1218 {
1219 	struct orig_ist *oist;
1220 	struct task_struct *me;
1221 	struct tss_struct *t;
1222 	unsigned long v;
1223 	int cpu;
1224 	int i;
1225 
1226 	cpu = stack_smp_processor_id();
1227 	t = &per_cpu(init_tss, cpu);
1228 	oist = &per_cpu(orig_ist, cpu);
1229 
1230 #ifdef CONFIG_NUMA
1231 	if (this_cpu_read(numa_node) == 0 &&
1232 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1233 		set_numa_node(early_cpu_to_node(cpu));
1234 #endif
1235 
1236 	me = current;
1237 
1238 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1239 		panic("CPU#%d already initialized!\n", cpu);
1240 
1241 	pr_debug("Initializing CPU#%d\n", cpu);
1242 
1243 	clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1244 
1245 	/*
1246 	 * Initialize the per-CPU GDT with the boot GDT,
1247 	 * and set up the GDT descriptor:
1248 	 */
1249 
1250 	switch_to_new_gdt(cpu);
1251 	loadsegment(fs, 0);
1252 
1253 	load_idt((const struct desc_ptr *)&idt_descr);
1254 
1255 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1256 	syscall_init();
1257 
1258 	wrmsrl(MSR_FS_BASE, 0);
1259 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1260 	barrier();
1261 
1262 	x86_configure_nx();
1263 	enable_x2apic();
1264 
1265 	/*
1266 	 * set up and load the per-CPU TSS
1267 	 */
1268 	if (!oist->ist[0]) {
1269 		char *estacks = per_cpu(exception_stacks, cpu);
1270 
1271 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1272 			estacks += exception_stack_sizes[v];
1273 			oist->ist[v] = t->x86_tss.ist[v] =
1274 					(unsigned long)estacks;
1275 			if (v == DEBUG_STACK-1)
1276 				per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1277 		}
1278 	}
1279 
1280 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1281 
1282 	/*
1283 	 * <= is required because the CPU will access up to
1284 	 * 8 bits beyond the end of the IO permission bitmap.
1285 	 */
1286 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1287 		t->io_bitmap[i] = ~0UL;
1288 
1289 	atomic_inc(&init_mm.mm_count);
1290 	me->active_mm = &init_mm;
1291 	BUG_ON(me->mm);
1292 	enter_lazy_tlb(&init_mm, me);
1293 
1294 	load_sp0(t, &current->thread);
1295 	set_tss_desc(cpu, t);
1296 	load_TR_desc();
1297 	load_LDT(&init_mm.context);
1298 
1299 	clear_all_debug_regs();
1300 	dbg_restore_debug_regs();
1301 
1302 	fpu_init();
1303 
1304 	if (is_uv_system())
1305 		uv_cpu_init();
1306 }
1307 
1308 #else
1309 
1310 void __cpuinit cpu_init(void)
1311 {
1312 	int cpu = smp_processor_id();
1313 	struct task_struct *curr = current;
1314 	struct tss_struct *t = &per_cpu(init_tss, cpu);
1315 	struct thread_struct *thread = &curr->thread;
1316 
1317 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1318 		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1319 		for (;;)
1320 			local_irq_enable();
1321 	}
1322 
1323 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1324 
1325 	if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1326 		clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1327 
1328 	load_idt(&idt_descr);
1329 	switch_to_new_gdt(cpu);
1330 
1331 	/*
1332 	 * Set up and load the per-CPU TSS and LDT
1333 	 */
1334 	atomic_inc(&init_mm.mm_count);
1335 	curr->active_mm = &init_mm;
1336 	BUG_ON(curr->mm);
1337 	enter_lazy_tlb(&init_mm, curr);
1338 
1339 	load_sp0(t, thread);
1340 	set_tss_desc(cpu, t);
1341 	load_TR_desc();
1342 	load_LDT(&init_mm.context);
1343 
1344 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1345 
1346 #ifdef CONFIG_DOUBLEFAULT
1347 	/* Set up doublefault TSS pointer in the GDT */
1348 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1349 #endif
1350 
1351 	clear_all_debug_regs();
1352 	dbg_restore_debug_regs();
1353 
1354 	fpu_init();
1355 }
1356 #endif
1357