xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision db181ce0)
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/kprobes.h>
12 #include <linux/kgdb.h>
13 #include <linux/smp.h>
14 #include <linux/io.h>
15 
16 #include <asm/stackprotector.h>
17 #include <asm/perf_event.h>
18 #include <asm/mmu_context.h>
19 #include <asm/archrandom.h>
20 #include <asm/hypervisor.h>
21 #include <asm/processor.h>
22 #include <asm/debugreg.h>
23 #include <asm/sections.h>
24 #include <asm/vsyscall.h>
25 #include <linux/topology.h>
26 #include <linux/cpumask.h>
27 #include <asm/pgtable.h>
28 #include <linux/atomic.h>
29 #include <asm/proto.h>
30 #include <asm/setup.h>
31 #include <asm/apic.h>
32 #include <asm/desc.h>
33 #include <asm/i387.h>
34 #include <asm/fpu-internal.h>
35 #include <asm/mtrr.h>
36 #include <linux/numa.h>
37 #include <asm/asm.h>
38 #include <asm/cpu.h>
39 #include <asm/mce.h>
40 #include <asm/msr.h>
41 #include <asm/pat.h>
42 #include <asm/microcode.h>
43 #include <asm/microcode_intel.h>
44 
45 #ifdef CONFIG_X86_LOCAL_APIC
46 #include <asm/uv/uv.h>
47 #endif
48 
49 #include "cpu.h"
50 
51 /* all of these masks are initialized in setup_cpu_local_masks() */
52 cpumask_var_t cpu_initialized_mask;
53 cpumask_var_t cpu_callout_mask;
54 cpumask_var_t cpu_callin_mask;
55 
56 /* representing cpus for which sibling maps can be computed */
57 cpumask_var_t cpu_sibling_setup_mask;
58 
59 /* correctly size the local cpu masks */
60 void __init setup_cpu_local_masks(void)
61 {
62 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
63 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
64 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
65 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
66 }
67 
68 static void default_init(struct cpuinfo_x86 *c)
69 {
70 #ifdef CONFIG_X86_64
71 	cpu_detect_cache_sizes(c);
72 #else
73 	/* Not much we can do here... */
74 	/* Check if at least it has cpuid */
75 	if (c->cpuid_level == -1) {
76 		/* No cpuid. It must be an ancient CPU */
77 		if (c->x86 == 4)
78 			strcpy(c->x86_model_id, "486");
79 		else if (c->x86 == 3)
80 			strcpy(c->x86_model_id, "386");
81 	}
82 #endif
83 }
84 
85 static const struct cpu_dev default_cpu = {
86 	.c_init		= default_init,
87 	.c_vendor	= "Unknown",
88 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
89 };
90 
91 static const struct cpu_dev *this_cpu = &default_cpu;
92 
93 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
94 #ifdef CONFIG_X86_64
95 	/*
96 	 * We need valid kernel segments for data and code in long mode too
97 	 * IRET will check the segment types  kkeil 2000/10/28
98 	 * Also sysret mandates a special GDT layout
99 	 *
100 	 * TLS descriptors are currently at a different place compared to i386.
101 	 * Hopefully nobody expects them at a fixed place (Wine?)
102 	 */
103 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
104 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
105 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
106 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
107 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
108 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
109 #else
110 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
111 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
112 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
113 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
114 	/*
115 	 * Segments used for calling PnP BIOS have byte granularity.
116 	 * They code segments and data segments have fixed 64k limits,
117 	 * the transfer segment sizes are set at run time.
118 	 */
119 	/* 32-bit code */
120 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
121 	/* 16-bit code */
122 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
123 	/* 16-bit data */
124 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
125 	/* 16-bit data */
126 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
127 	/* 16-bit data */
128 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
129 	/*
130 	 * The APM segments have byte granularity and their bases
131 	 * are set at run time.  All have 64k limits.
132 	 */
133 	/* 32-bit code */
134 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
135 	/* 16-bit code */
136 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
137 	/* data */
138 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
139 
140 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
141 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
142 	GDT_STACK_CANARY_INIT
143 #endif
144 } };
145 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
146 
147 static int __init x86_xsave_setup(char *s)
148 {
149 	setup_clear_cpu_cap(X86_FEATURE_XSAVE);
150 	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
151 	setup_clear_cpu_cap(X86_FEATURE_AVX);
152 	setup_clear_cpu_cap(X86_FEATURE_AVX2);
153 	return 1;
154 }
155 __setup("noxsave", x86_xsave_setup);
156 
157 static int __init x86_xsaveopt_setup(char *s)
158 {
159 	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
160 	return 1;
161 }
162 __setup("noxsaveopt", x86_xsaveopt_setup);
163 
164 #ifdef CONFIG_X86_32
165 static int cachesize_override = -1;
166 static int disable_x86_serial_nr = 1;
167 
168 static int __init cachesize_setup(char *str)
169 {
170 	get_option(&str, &cachesize_override);
171 	return 1;
172 }
173 __setup("cachesize=", cachesize_setup);
174 
175 static int __init x86_fxsr_setup(char *s)
176 {
177 	setup_clear_cpu_cap(X86_FEATURE_FXSR);
178 	setup_clear_cpu_cap(X86_FEATURE_XMM);
179 	return 1;
180 }
181 __setup("nofxsr", x86_fxsr_setup);
182 
183 static int __init x86_sep_setup(char *s)
184 {
185 	setup_clear_cpu_cap(X86_FEATURE_SEP);
186 	return 1;
187 }
188 __setup("nosep", x86_sep_setup);
189 
190 /* Standard macro to see if a specific flag is changeable */
191 static inline int flag_is_changeable_p(u32 flag)
192 {
193 	u32 f1, f2;
194 
195 	/*
196 	 * Cyrix and IDT cpus allow disabling of CPUID
197 	 * so the code below may return different results
198 	 * when it is executed before and after enabling
199 	 * the CPUID. Add "volatile" to not allow gcc to
200 	 * optimize the subsequent calls to this function.
201 	 */
202 	asm volatile ("pushfl		\n\t"
203 		      "pushfl		\n\t"
204 		      "popl %0		\n\t"
205 		      "movl %0, %1	\n\t"
206 		      "xorl %2, %0	\n\t"
207 		      "pushl %0		\n\t"
208 		      "popfl		\n\t"
209 		      "pushfl		\n\t"
210 		      "popl %0		\n\t"
211 		      "popfl		\n\t"
212 
213 		      : "=&r" (f1), "=&r" (f2)
214 		      : "ir" (flag));
215 
216 	return ((f1^f2) & flag) != 0;
217 }
218 
219 /* Probe for the CPUID instruction */
220 int have_cpuid_p(void)
221 {
222 	return flag_is_changeable_p(X86_EFLAGS_ID);
223 }
224 
225 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
226 {
227 	unsigned long lo, hi;
228 
229 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
230 		return;
231 
232 	/* Disable processor serial number: */
233 
234 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
235 	lo |= 0x200000;
236 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
237 
238 	printk(KERN_NOTICE "CPU serial number disabled.\n");
239 	clear_cpu_cap(c, X86_FEATURE_PN);
240 
241 	/* Disabling the serial number may affect the cpuid level */
242 	c->cpuid_level = cpuid_eax(0);
243 }
244 
245 static int __init x86_serial_nr_setup(char *s)
246 {
247 	disable_x86_serial_nr = 0;
248 	return 1;
249 }
250 __setup("serialnumber", x86_serial_nr_setup);
251 #else
252 static inline int flag_is_changeable_p(u32 flag)
253 {
254 	return 1;
255 }
256 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
257 {
258 }
259 #endif
260 
261 static __init int setup_disable_smep(char *arg)
262 {
263 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
264 	return 1;
265 }
266 __setup("nosmep", setup_disable_smep);
267 
268 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
269 {
270 	if (cpu_has(c, X86_FEATURE_SMEP))
271 		set_in_cr4(X86_CR4_SMEP);
272 }
273 
274 static __init int setup_disable_smap(char *arg)
275 {
276 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
277 	return 1;
278 }
279 __setup("nosmap", setup_disable_smap);
280 
281 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
282 {
283 	unsigned long eflags;
284 
285 	/* This should have been cleared long ago */
286 	raw_local_save_flags(eflags);
287 	BUG_ON(eflags & X86_EFLAGS_AC);
288 
289 	if (cpu_has(c, X86_FEATURE_SMAP)) {
290 #ifdef CONFIG_X86_SMAP
291 		set_in_cr4(X86_CR4_SMAP);
292 #else
293 		clear_in_cr4(X86_CR4_SMAP);
294 #endif
295 	}
296 }
297 
298 /*
299  * Some CPU features depend on higher CPUID levels, which may not always
300  * be available due to CPUID level capping or broken virtualization
301  * software.  Add those features to this table to auto-disable them.
302  */
303 struct cpuid_dependent_feature {
304 	u32 feature;
305 	u32 level;
306 };
307 
308 static const struct cpuid_dependent_feature
309 cpuid_dependent_features[] = {
310 	{ X86_FEATURE_MWAIT,		0x00000005 },
311 	{ X86_FEATURE_DCA,		0x00000009 },
312 	{ X86_FEATURE_XSAVE,		0x0000000d },
313 	{ 0, 0 }
314 };
315 
316 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
317 {
318 	const struct cpuid_dependent_feature *df;
319 
320 	for (df = cpuid_dependent_features; df->feature; df++) {
321 
322 		if (!cpu_has(c, df->feature))
323 			continue;
324 		/*
325 		 * Note: cpuid_level is set to -1 if unavailable, but
326 		 * extended_extended_level is set to 0 if unavailable
327 		 * and the legitimate extended levels are all negative
328 		 * when signed; hence the weird messing around with
329 		 * signs here...
330 		 */
331 		if (!((s32)df->level < 0 ?
332 		     (u32)df->level > (u32)c->extended_cpuid_level :
333 		     (s32)df->level > (s32)c->cpuid_level))
334 			continue;
335 
336 		clear_cpu_cap(c, df->feature);
337 		if (!warn)
338 			continue;
339 
340 		printk(KERN_WARNING
341 		       "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
342 				x86_cap_flags[df->feature], df->level);
343 	}
344 }
345 
346 /*
347  * Naming convention should be: <Name> [(<Codename>)]
348  * This table only is used unless init_<vendor>() below doesn't set it;
349  * in particular, if CPUID levels 0x80000002..4 are supported, this
350  * isn't used
351  */
352 
353 /* Look up CPU names by table lookup. */
354 static const char *table_lookup_model(struct cpuinfo_x86 *c)
355 {
356 #ifdef CONFIG_X86_32
357 	const struct legacy_cpu_model_info *info;
358 
359 	if (c->x86_model >= 16)
360 		return NULL;	/* Range check */
361 
362 	if (!this_cpu)
363 		return NULL;
364 
365 	info = this_cpu->legacy_models;
366 
367 	while (info->family) {
368 		if (info->family == c->x86)
369 			return info->model_names[c->x86_model];
370 		info++;
371 	}
372 #endif
373 	return NULL;		/* Not found */
374 }
375 
376 __u32 cpu_caps_cleared[NCAPINTS];
377 __u32 cpu_caps_set[NCAPINTS];
378 
379 void load_percpu_segment(int cpu)
380 {
381 #ifdef CONFIG_X86_32
382 	loadsegment(fs, __KERNEL_PERCPU);
383 #else
384 	loadsegment(gs, 0);
385 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
386 #endif
387 	load_stack_canary_segment();
388 }
389 
390 /*
391  * Current gdt points %fs at the "master" per-cpu area: after this,
392  * it's on the real one.
393  */
394 void switch_to_new_gdt(int cpu)
395 {
396 	struct desc_ptr gdt_descr;
397 
398 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
399 	gdt_descr.size = GDT_SIZE - 1;
400 	load_gdt(&gdt_descr);
401 	/* Reload the per-cpu base */
402 
403 	load_percpu_segment(cpu);
404 }
405 
406 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
407 
408 static void get_model_name(struct cpuinfo_x86 *c)
409 {
410 	unsigned int *v;
411 	char *p, *q;
412 
413 	if (c->extended_cpuid_level < 0x80000004)
414 		return;
415 
416 	v = (unsigned int *)c->x86_model_id;
417 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
418 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
419 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
420 	c->x86_model_id[48] = 0;
421 
422 	/*
423 	 * Intel chips right-justify this string for some dumb reason;
424 	 * undo that brain damage:
425 	 */
426 	p = q = &c->x86_model_id[0];
427 	while (*p == ' ')
428 		p++;
429 	if (p != q) {
430 		while (*p)
431 			*q++ = *p++;
432 		while (q <= &c->x86_model_id[48])
433 			*q++ = '\0';	/* Zero-pad the rest */
434 	}
435 }
436 
437 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
438 {
439 	unsigned int n, dummy, ebx, ecx, edx, l2size;
440 
441 	n = c->extended_cpuid_level;
442 
443 	if (n >= 0x80000005) {
444 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
445 		c->x86_cache_size = (ecx>>24) + (edx>>24);
446 #ifdef CONFIG_X86_64
447 		/* On K8 L1 TLB is inclusive, so don't count it */
448 		c->x86_tlbsize = 0;
449 #endif
450 	}
451 
452 	if (n < 0x80000006)	/* Some chips just has a large L1. */
453 		return;
454 
455 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
456 	l2size = ecx >> 16;
457 
458 #ifdef CONFIG_X86_64
459 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
460 #else
461 	/* do processor-specific cache resizing */
462 	if (this_cpu->legacy_cache_size)
463 		l2size = this_cpu->legacy_cache_size(c, l2size);
464 
465 	/* Allow user to override all this if necessary. */
466 	if (cachesize_override != -1)
467 		l2size = cachesize_override;
468 
469 	if (l2size == 0)
470 		return;		/* Again, no L2 cache is possible */
471 #endif
472 
473 	c->x86_cache_size = l2size;
474 }
475 
476 u16 __read_mostly tlb_lli_4k[NR_INFO];
477 u16 __read_mostly tlb_lli_2m[NR_INFO];
478 u16 __read_mostly tlb_lli_4m[NR_INFO];
479 u16 __read_mostly tlb_lld_4k[NR_INFO];
480 u16 __read_mostly tlb_lld_2m[NR_INFO];
481 u16 __read_mostly tlb_lld_4m[NR_INFO];
482 u16 __read_mostly tlb_lld_1g[NR_INFO];
483 
484 /*
485  * tlb_flushall_shift shows the balance point in replacing cr3 write
486  * with multiple 'invlpg'. It will do this replacement when
487  *   flush_tlb_lines <= active_lines/2^tlb_flushall_shift.
488  * If tlb_flushall_shift is -1, means the replacement will be disabled.
489  */
490 s8  __read_mostly tlb_flushall_shift = -1;
491 
492 void cpu_detect_tlb(struct cpuinfo_x86 *c)
493 {
494 	if (this_cpu->c_detect_tlb)
495 		this_cpu->c_detect_tlb(c);
496 
497 	printk(KERN_INFO "Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n"
498 		"Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n"
499 		"tlb_flushall_shift: %d\n",
500 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
501 		tlb_lli_4m[ENTRIES], tlb_lld_4k[ENTRIES],
502 		tlb_lld_2m[ENTRIES], tlb_lld_4m[ENTRIES],
503 		tlb_lld_1g[ENTRIES], tlb_flushall_shift);
504 }
505 
506 void detect_ht(struct cpuinfo_x86 *c)
507 {
508 #ifdef CONFIG_X86_HT
509 	u32 eax, ebx, ecx, edx;
510 	int index_msb, core_bits;
511 	static bool printed;
512 
513 	if (!cpu_has(c, X86_FEATURE_HT))
514 		return;
515 
516 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
517 		goto out;
518 
519 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
520 		return;
521 
522 	cpuid(1, &eax, &ebx, &ecx, &edx);
523 
524 	smp_num_siblings = (ebx & 0xff0000) >> 16;
525 
526 	if (smp_num_siblings == 1) {
527 		printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
528 		goto out;
529 	}
530 
531 	if (smp_num_siblings <= 1)
532 		goto out;
533 
534 	index_msb = get_count_order(smp_num_siblings);
535 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
536 
537 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
538 
539 	index_msb = get_count_order(smp_num_siblings);
540 
541 	core_bits = get_count_order(c->x86_max_cores);
542 
543 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
544 				       ((1 << core_bits) - 1);
545 
546 out:
547 	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
548 		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
549 		       c->phys_proc_id);
550 		printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
551 		       c->cpu_core_id);
552 		printed = 1;
553 	}
554 #endif
555 }
556 
557 static void get_cpu_vendor(struct cpuinfo_x86 *c)
558 {
559 	char *v = c->x86_vendor_id;
560 	int i;
561 
562 	for (i = 0; i < X86_VENDOR_NUM; i++) {
563 		if (!cpu_devs[i])
564 			break;
565 
566 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
567 		    (cpu_devs[i]->c_ident[1] &&
568 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
569 
570 			this_cpu = cpu_devs[i];
571 			c->x86_vendor = this_cpu->c_x86_vendor;
572 			return;
573 		}
574 	}
575 
576 	printk_once(KERN_ERR
577 			"CPU: vendor_id '%s' unknown, using generic init.\n" \
578 			"CPU: Your system may be unstable.\n", v);
579 
580 	c->x86_vendor = X86_VENDOR_UNKNOWN;
581 	this_cpu = &default_cpu;
582 }
583 
584 void cpu_detect(struct cpuinfo_x86 *c)
585 {
586 	/* Get vendor name */
587 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
588 	      (unsigned int *)&c->x86_vendor_id[0],
589 	      (unsigned int *)&c->x86_vendor_id[8],
590 	      (unsigned int *)&c->x86_vendor_id[4]);
591 
592 	c->x86 = 4;
593 	/* Intel-defined flags: level 0x00000001 */
594 	if (c->cpuid_level >= 0x00000001) {
595 		u32 junk, tfms, cap0, misc;
596 
597 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
598 		c->x86 = (tfms >> 8) & 0xf;
599 		c->x86_model = (tfms >> 4) & 0xf;
600 		c->x86_mask = tfms & 0xf;
601 
602 		if (c->x86 == 0xf)
603 			c->x86 += (tfms >> 20) & 0xff;
604 		if (c->x86 >= 0x6)
605 			c->x86_model += ((tfms >> 16) & 0xf) << 4;
606 
607 		if (cap0 & (1<<19)) {
608 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
609 			c->x86_cache_alignment = c->x86_clflush_size;
610 		}
611 	}
612 }
613 
614 void get_cpu_cap(struct cpuinfo_x86 *c)
615 {
616 	u32 tfms, xlvl;
617 	u32 ebx;
618 
619 	/* Intel-defined flags: level 0x00000001 */
620 	if (c->cpuid_level >= 0x00000001) {
621 		u32 capability, excap;
622 
623 		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
624 		c->x86_capability[0] = capability;
625 		c->x86_capability[4] = excap;
626 	}
627 
628 	/* Additional Intel-defined flags: level 0x00000007 */
629 	if (c->cpuid_level >= 0x00000007) {
630 		u32 eax, ebx, ecx, edx;
631 
632 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
633 
634 		c->x86_capability[9] = ebx;
635 	}
636 
637 	/* AMD-defined flags: level 0x80000001 */
638 	xlvl = cpuid_eax(0x80000000);
639 	c->extended_cpuid_level = xlvl;
640 
641 	if ((xlvl & 0xffff0000) == 0x80000000) {
642 		if (xlvl >= 0x80000001) {
643 			c->x86_capability[1] = cpuid_edx(0x80000001);
644 			c->x86_capability[6] = cpuid_ecx(0x80000001);
645 		}
646 	}
647 
648 	if (c->extended_cpuid_level >= 0x80000008) {
649 		u32 eax = cpuid_eax(0x80000008);
650 
651 		c->x86_virt_bits = (eax >> 8) & 0xff;
652 		c->x86_phys_bits = eax & 0xff;
653 	}
654 #ifdef CONFIG_X86_32
655 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
656 		c->x86_phys_bits = 36;
657 #endif
658 
659 	if (c->extended_cpuid_level >= 0x80000007)
660 		c->x86_power = cpuid_edx(0x80000007);
661 
662 	init_scattered_cpuid_features(c);
663 }
664 
665 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
666 {
667 #ifdef CONFIG_X86_32
668 	int i;
669 
670 	/*
671 	 * First of all, decide if this is a 486 or higher
672 	 * It's a 486 if we can modify the AC flag
673 	 */
674 	if (flag_is_changeable_p(X86_EFLAGS_AC))
675 		c->x86 = 4;
676 	else
677 		c->x86 = 3;
678 
679 	for (i = 0; i < X86_VENDOR_NUM; i++)
680 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
681 			c->x86_vendor_id[0] = 0;
682 			cpu_devs[i]->c_identify(c);
683 			if (c->x86_vendor_id[0]) {
684 				get_cpu_vendor(c);
685 				break;
686 			}
687 		}
688 #endif
689 }
690 
691 /*
692  * Do minimum CPU detection early.
693  * Fields really needed: vendor, cpuid_level, family, model, mask,
694  * cache alignment.
695  * The others are not touched to avoid unwanted side effects.
696  *
697  * WARNING: this function is only called on the BP.  Don't add code here
698  * that is supposed to run on all CPUs.
699  */
700 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
701 {
702 #ifdef CONFIG_X86_64
703 	c->x86_clflush_size = 64;
704 	c->x86_phys_bits = 36;
705 	c->x86_virt_bits = 48;
706 #else
707 	c->x86_clflush_size = 32;
708 	c->x86_phys_bits = 32;
709 	c->x86_virt_bits = 32;
710 #endif
711 	c->x86_cache_alignment = c->x86_clflush_size;
712 
713 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
714 	c->extended_cpuid_level = 0;
715 
716 	if (!have_cpuid_p())
717 		identify_cpu_without_cpuid(c);
718 
719 	/* cyrix could have cpuid enabled via c_identify()*/
720 	if (!have_cpuid_p())
721 		return;
722 
723 	cpu_detect(c);
724 	get_cpu_vendor(c);
725 	get_cpu_cap(c);
726 	fpu_detect(c);
727 
728 	if (this_cpu->c_early_init)
729 		this_cpu->c_early_init(c);
730 
731 	c->cpu_index = 0;
732 	filter_cpuid_features(c, false);
733 
734 	if (this_cpu->c_bsp_init)
735 		this_cpu->c_bsp_init(c);
736 
737 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
738 }
739 
740 void __init early_cpu_init(void)
741 {
742 	const struct cpu_dev *const *cdev;
743 	int count = 0;
744 
745 #ifdef CONFIG_PROCESSOR_SELECT
746 	printk(KERN_INFO "KERNEL supported cpus:\n");
747 #endif
748 
749 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
750 		const struct cpu_dev *cpudev = *cdev;
751 
752 		if (count >= X86_VENDOR_NUM)
753 			break;
754 		cpu_devs[count] = cpudev;
755 		count++;
756 
757 #ifdef CONFIG_PROCESSOR_SELECT
758 		{
759 			unsigned int j;
760 
761 			for (j = 0; j < 2; j++) {
762 				if (!cpudev->c_ident[j])
763 					continue;
764 				printk(KERN_INFO "  %s %s\n", cpudev->c_vendor,
765 					cpudev->c_ident[j]);
766 			}
767 		}
768 #endif
769 	}
770 	early_identify_cpu(&boot_cpu_data);
771 }
772 
773 /*
774  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
775  * unfortunately, that's not true in practice because of early VIA
776  * chips and (more importantly) broken virtualizers that are not easy
777  * to detect. In the latter case it doesn't even *fail* reliably, so
778  * probing for it doesn't even work. Disable it completely on 32-bit
779  * unless we can find a reliable way to detect all the broken cases.
780  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
781  */
782 static void detect_nopl(struct cpuinfo_x86 *c)
783 {
784 #ifdef CONFIG_X86_32
785 	clear_cpu_cap(c, X86_FEATURE_NOPL);
786 #else
787 	set_cpu_cap(c, X86_FEATURE_NOPL);
788 #endif
789 }
790 
791 static void generic_identify(struct cpuinfo_x86 *c)
792 {
793 	c->extended_cpuid_level = 0;
794 
795 	if (!have_cpuid_p())
796 		identify_cpu_without_cpuid(c);
797 
798 	/* cyrix could have cpuid enabled via c_identify()*/
799 	if (!have_cpuid_p())
800 		return;
801 
802 	cpu_detect(c);
803 
804 	get_cpu_vendor(c);
805 
806 	get_cpu_cap(c);
807 
808 	if (c->cpuid_level >= 0x00000001) {
809 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
810 #ifdef CONFIG_X86_32
811 # ifdef CONFIG_X86_HT
812 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
813 # else
814 		c->apicid = c->initial_apicid;
815 # endif
816 #endif
817 		c->phys_proc_id = c->initial_apicid;
818 	}
819 
820 	get_model_name(c); /* Default name */
821 
822 	detect_nopl(c);
823 }
824 
825 /*
826  * This does the hard work of actually picking apart the CPU stuff...
827  */
828 static void identify_cpu(struct cpuinfo_x86 *c)
829 {
830 	int i;
831 
832 	c->loops_per_jiffy = loops_per_jiffy;
833 	c->x86_cache_size = -1;
834 	c->x86_vendor = X86_VENDOR_UNKNOWN;
835 	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
836 	c->x86_vendor_id[0] = '\0'; /* Unset */
837 	c->x86_model_id[0] = '\0';  /* Unset */
838 	c->x86_max_cores = 1;
839 	c->x86_coreid_bits = 0;
840 #ifdef CONFIG_X86_64
841 	c->x86_clflush_size = 64;
842 	c->x86_phys_bits = 36;
843 	c->x86_virt_bits = 48;
844 #else
845 	c->cpuid_level = -1;	/* CPUID not detected */
846 	c->x86_clflush_size = 32;
847 	c->x86_phys_bits = 32;
848 	c->x86_virt_bits = 32;
849 #endif
850 	c->x86_cache_alignment = c->x86_clflush_size;
851 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
852 
853 	generic_identify(c);
854 
855 	if (this_cpu->c_identify)
856 		this_cpu->c_identify(c);
857 
858 	/* Clear/Set all flags overriden by options, after probe */
859 	for (i = 0; i < NCAPINTS; i++) {
860 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
861 		c->x86_capability[i] |= cpu_caps_set[i];
862 	}
863 
864 #ifdef CONFIG_X86_64
865 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
866 #endif
867 
868 	/*
869 	 * Vendor-specific initialization.  In this section we
870 	 * canonicalize the feature flags, meaning if there are
871 	 * features a certain CPU supports which CPUID doesn't
872 	 * tell us, CPUID claiming incorrect flags, or other bugs,
873 	 * we handle them here.
874 	 *
875 	 * At the end of this section, c->x86_capability better
876 	 * indicate the features this CPU genuinely supports!
877 	 */
878 	if (this_cpu->c_init)
879 		this_cpu->c_init(c);
880 
881 	/* Disable the PN if appropriate */
882 	squash_the_stupid_serial_number(c);
883 
884 	/* Set up SMEP/SMAP */
885 	setup_smep(c);
886 	setup_smap(c);
887 
888 	/*
889 	 * The vendor-specific functions might have changed features.
890 	 * Now we do "generic changes."
891 	 */
892 
893 	/* Filter out anything that depends on CPUID levels we don't have */
894 	filter_cpuid_features(c, true);
895 
896 	/* If the model name is still unset, do table lookup. */
897 	if (!c->x86_model_id[0]) {
898 		const char *p;
899 		p = table_lookup_model(c);
900 		if (p)
901 			strcpy(c->x86_model_id, p);
902 		else
903 			/* Last resort... */
904 			sprintf(c->x86_model_id, "%02x/%02x",
905 				c->x86, c->x86_model);
906 	}
907 
908 #ifdef CONFIG_X86_64
909 	detect_ht(c);
910 #endif
911 
912 	init_hypervisor(c);
913 	x86_init_rdrand(c);
914 
915 	/*
916 	 * Clear/Set all flags overriden by options, need do it
917 	 * before following smp all cpus cap AND.
918 	 */
919 	for (i = 0; i < NCAPINTS; i++) {
920 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
921 		c->x86_capability[i] |= cpu_caps_set[i];
922 	}
923 
924 	/*
925 	 * On SMP, boot_cpu_data holds the common feature set between
926 	 * all CPUs; so make sure that we indicate which features are
927 	 * common between the CPUs.  The first time this routine gets
928 	 * executed, c == &boot_cpu_data.
929 	 */
930 	if (c != &boot_cpu_data) {
931 		/* AND the already accumulated flags with these */
932 		for (i = 0; i < NCAPINTS; i++)
933 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
934 
935 		/* OR, i.e. replicate the bug flags */
936 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
937 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
938 	}
939 
940 	/* Init Machine Check Exception if available. */
941 	mcheck_cpu_init(c);
942 
943 	select_idle_routine(c);
944 
945 #ifdef CONFIG_NUMA
946 	numa_add_cpu(smp_processor_id());
947 #endif
948 }
949 
950 #ifdef CONFIG_X86_64
951 static void vgetcpu_set_mode(void)
952 {
953 	if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
954 		vgetcpu_mode = VGETCPU_RDTSCP;
955 	else
956 		vgetcpu_mode = VGETCPU_LSL;
957 }
958 
959 /* May not be __init: called during resume */
960 static void syscall32_cpu_init(void)
961 {
962 	/* Load these always in case some future AMD CPU supports
963 	   SYSENTER from compat mode too. */
964 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
965 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
966 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)ia32_sysenter_target);
967 
968 	wrmsrl(MSR_CSTAR, ia32_cstar_target);
969 }
970 #endif
971 
972 #ifdef CONFIG_X86_32
973 void enable_sep_cpu(void)
974 {
975 	int cpu = get_cpu();
976 	struct tss_struct *tss = &per_cpu(init_tss, cpu);
977 
978 	if (!boot_cpu_has(X86_FEATURE_SEP)) {
979 		put_cpu();
980 		return;
981 	}
982 
983 	tss->x86_tss.ss1 = __KERNEL_CS;
984 	tss->x86_tss.sp1 = sizeof(struct tss_struct) + (unsigned long) tss;
985 	wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
986 	wrmsr(MSR_IA32_SYSENTER_ESP, tss->x86_tss.sp1, 0);
987 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long) ia32_sysenter_target, 0);
988 	put_cpu();
989 }
990 #endif
991 
992 void __init identify_boot_cpu(void)
993 {
994 	identify_cpu(&boot_cpu_data);
995 	init_amd_e400_c1e_mask();
996 #ifdef CONFIG_X86_32
997 	sysenter_setup();
998 	enable_sep_cpu();
999 #else
1000 	vgetcpu_set_mode();
1001 #endif
1002 	cpu_detect_tlb(&boot_cpu_data);
1003 }
1004 
1005 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1006 {
1007 	BUG_ON(c == &boot_cpu_data);
1008 	identify_cpu(c);
1009 #ifdef CONFIG_X86_32
1010 	enable_sep_cpu();
1011 #endif
1012 	mtrr_ap_init();
1013 }
1014 
1015 struct msr_range {
1016 	unsigned	min;
1017 	unsigned	max;
1018 };
1019 
1020 static const struct msr_range msr_range_array[] = {
1021 	{ 0x00000000, 0x00000418},
1022 	{ 0xc0000000, 0xc000040b},
1023 	{ 0xc0010000, 0xc0010142},
1024 	{ 0xc0011000, 0xc001103b},
1025 };
1026 
1027 static void __print_cpu_msr(void)
1028 {
1029 	unsigned index_min, index_max;
1030 	unsigned index;
1031 	u64 val;
1032 	int i;
1033 
1034 	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
1035 		index_min = msr_range_array[i].min;
1036 		index_max = msr_range_array[i].max;
1037 
1038 		for (index = index_min; index < index_max; index++) {
1039 			if (rdmsrl_safe(index, &val))
1040 				continue;
1041 			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
1042 		}
1043 	}
1044 }
1045 
1046 static int show_msr;
1047 
1048 static __init int setup_show_msr(char *arg)
1049 {
1050 	int num;
1051 
1052 	get_option(&arg, &num);
1053 
1054 	if (num > 0)
1055 		show_msr = num;
1056 	return 1;
1057 }
1058 __setup("show_msr=", setup_show_msr);
1059 
1060 static __init int setup_noclflush(char *arg)
1061 {
1062 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1063 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1064 	return 1;
1065 }
1066 __setup("noclflush", setup_noclflush);
1067 
1068 void print_cpu_info(struct cpuinfo_x86 *c)
1069 {
1070 	const char *vendor = NULL;
1071 
1072 	if (c->x86_vendor < X86_VENDOR_NUM) {
1073 		vendor = this_cpu->c_vendor;
1074 	} else {
1075 		if (c->cpuid_level >= 0)
1076 			vendor = c->x86_vendor_id;
1077 	}
1078 
1079 	if (vendor && !strstr(c->x86_model_id, vendor))
1080 		printk(KERN_CONT "%s ", vendor);
1081 
1082 	if (c->x86_model_id[0])
1083 		printk(KERN_CONT "%s", strim(c->x86_model_id));
1084 	else
1085 		printk(KERN_CONT "%d86", c->x86);
1086 
1087 	printk(KERN_CONT " (fam: %02x, model: %02x", c->x86, c->x86_model);
1088 
1089 	if (c->x86_mask || c->cpuid_level >= 0)
1090 		printk(KERN_CONT ", stepping: %02x)\n", c->x86_mask);
1091 	else
1092 		printk(KERN_CONT ")\n");
1093 
1094 	print_cpu_msr(c);
1095 }
1096 
1097 void print_cpu_msr(struct cpuinfo_x86 *c)
1098 {
1099 	if (c->cpu_index < show_msr)
1100 		__print_cpu_msr();
1101 }
1102 
1103 static __init int setup_disablecpuid(char *arg)
1104 {
1105 	int bit;
1106 
1107 	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1108 		setup_clear_cpu_cap(bit);
1109 	else
1110 		return 0;
1111 
1112 	return 1;
1113 }
1114 __setup("clearcpuid=", setup_disablecpuid);
1115 
1116 DEFINE_PER_CPU(unsigned long, kernel_stack) =
1117 	(unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
1118 EXPORT_PER_CPU_SYMBOL(kernel_stack);
1119 
1120 #ifdef CONFIG_X86_64
1121 struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
1122 struct desc_ptr debug_idt_descr = { NR_VECTORS * 16 - 1,
1123 				    (unsigned long) debug_idt_table };
1124 
1125 DEFINE_PER_CPU_FIRST(union irq_stack_union,
1126 		     irq_stack_union) __aligned(PAGE_SIZE) __visible;
1127 
1128 /*
1129  * The following four percpu variables are hot.  Align current_task to
1130  * cacheline size such that all four fall in the same cacheline.
1131  */
1132 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1133 	&init_task;
1134 EXPORT_PER_CPU_SYMBOL(current_task);
1135 
1136 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1137 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1138 
1139 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1140 
1141 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1142 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1143 
1144 DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1145 
1146 /*
1147  * Special IST stacks which the CPU switches to when it calls
1148  * an IST-marked descriptor entry. Up to 7 stacks (hardware
1149  * limit), all of them are 4K, except the debug stack which
1150  * is 8K.
1151  */
1152 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1153 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1154 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1155 };
1156 
1157 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1158 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1159 
1160 /* May not be marked __init: used by software suspend */
1161 void syscall_init(void)
1162 {
1163 	/*
1164 	 * LSTAR and STAR live in a bit strange symbiosis.
1165 	 * They both write to the same internal register. STAR allows to
1166 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1167 	 */
1168 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1169 	wrmsrl(MSR_LSTAR, system_call);
1170 	wrmsrl(MSR_CSTAR, ignore_sysret);
1171 
1172 #ifdef CONFIG_IA32_EMULATION
1173 	syscall32_cpu_init();
1174 #endif
1175 
1176 	/* Flags to clear on syscall */
1177 	wrmsrl(MSR_SYSCALL_MASK,
1178 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1179 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC);
1180 }
1181 
1182 /*
1183  * Copies of the original ist values from the tss are only accessed during
1184  * debugging, no special alignment required.
1185  */
1186 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1187 
1188 static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1189 DEFINE_PER_CPU(int, debug_stack_usage);
1190 
1191 int is_debug_stack(unsigned long addr)
1192 {
1193 	return __get_cpu_var(debug_stack_usage) ||
1194 		(addr <= __get_cpu_var(debug_stack_addr) &&
1195 		 addr > (__get_cpu_var(debug_stack_addr) - DEBUG_STKSZ));
1196 }
1197 NOKPROBE_SYMBOL(is_debug_stack);
1198 
1199 DEFINE_PER_CPU(u32, debug_idt_ctr);
1200 
1201 void debug_stack_set_zero(void)
1202 {
1203 	this_cpu_inc(debug_idt_ctr);
1204 	load_current_idt();
1205 }
1206 NOKPROBE_SYMBOL(debug_stack_set_zero);
1207 
1208 void debug_stack_reset(void)
1209 {
1210 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1211 		return;
1212 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1213 		load_current_idt();
1214 }
1215 NOKPROBE_SYMBOL(debug_stack_reset);
1216 
1217 #else	/* CONFIG_X86_64 */
1218 
1219 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1220 EXPORT_PER_CPU_SYMBOL(current_task);
1221 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1222 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1223 DEFINE_PER_CPU(struct task_struct *, fpu_owner_task);
1224 
1225 #ifdef CONFIG_CC_STACKPROTECTOR
1226 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1227 #endif
1228 
1229 #endif	/* CONFIG_X86_64 */
1230 
1231 /*
1232  * Clear all 6 debug registers:
1233  */
1234 static void clear_all_debug_regs(void)
1235 {
1236 	int i;
1237 
1238 	for (i = 0; i < 8; i++) {
1239 		/* Ignore db4, db5 */
1240 		if ((i == 4) || (i == 5))
1241 			continue;
1242 
1243 		set_debugreg(0, i);
1244 	}
1245 }
1246 
1247 #ifdef CONFIG_KGDB
1248 /*
1249  * Restore debug regs if using kgdbwait and you have a kernel debugger
1250  * connection established.
1251  */
1252 static void dbg_restore_debug_regs(void)
1253 {
1254 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1255 		arch_kgdb_ops.correct_hw_break();
1256 }
1257 #else /* ! CONFIG_KGDB */
1258 #define dbg_restore_debug_regs()
1259 #endif /* ! CONFIG_KGDB */
1260 
1261 /*
1262  * cpu_init() initializes state that is per-CPU. Some data is already
1263  * initialized (naturally) in the bootstrap process, such as the GDT
1264  * and IDT. We reload them nevertheless, this function acts as a
1265  * 'CPU state barrier', nothing should get across.
1266  * A lot of state is already set up in PDA init for 64 bit
1267  */
1268 #ifdef CONFIG_X86_64
1269 
1270 void cpu_init(void)
1271 {
1272 	struct orig_ist *oist;
1273 	struct task_struct *me;
1274 	struct tss_struct *t;
1275 	unsigned long v;
1276 	int cpu;
1277 	int i;
1278 
1279 	/*
1280 	 * Load microcode on this cpu if a valid microcode is available.
1281 	 * This is early microcode loading procedure.
1282 	 */
1283 	load_ucode_ap();
1284 
1285 	cpu = stack_smp_processor_id();
1286 	t = &per_cpu(init_tss, cpu);
1287 	oist = &per_cpu(orig_ist, cpu);
1288 
1289 #ifdef CONFIG_NUMA
1290 	if (this_cpu_read(numa_node) == 0 &&
1291 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1292 		set_numa_node(early_cpu_to_node(cpu));
1293 #endif
1294 
1295 	me = current;
1296 
1297 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1298 		panic("CPU#%d already initialized!\n", cpu);
1299 
1300 	pr_debug("Initializing CPU#%d\n", cpu);
1301 
1302 	clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1303 
1304 	/*
1305 	 * Initialize the per-CPU GDT with the boot GDT,
1306 	 * and set up the GDT descriptor:
1307 	 */
1308 
1309 	switch_to_new_gdt(cpu);
1310 	loadsegment(fs, 0);
1311 
1312 	load_current_idt();
1313 
1314 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1315 	syscall_init();
1316 
1317 	wrmsrl(MSR_FS_BASE, 0);
1318 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1319 	barrier();
1320 
1321 	x86_configure_nx();
1322 	enable_x2apic();
1323 
1324 	/*
1325 	 * set up and load the per-CPU TSS
1326 	 */
1327 	if (!oist->ist[0]) {
1328 		char *estacks = per_cpu(exception_stacks, cpu);
1329 
1330 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1331 			estacks += exception_stack_sizes[v];
1332 			oist->ist[v] = t->x86_tss.ist[v] =
1333 					(unsigned long)estacks;
1334 			if (v == DEBUG_STACK-1)
1335 				per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1336 		}
1337 	}
1338 
1339 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1340 
1341 	/*
1342 	 * <= is required because the CPU will access up to
1343 	 * 8 bits beyond the end of the IO permission bitmap.
1344 	 */
1345 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1346 		t->io_bitmap[i] = ~0UL;
1347 
1348 	atomic_inc(&init_mm.mm_count);
1349 	me->active_mm = &init_mm;
1350 	BUG_ON(me->mm);
1351 	enter_lazy_tlb(&init_mm, me);
1352 
1353 	load_sp0(t, &current->thread);
1354 	set_tss_desc(cpu, t);
1355 	load_TR_desc();
1356 	load_LDT(&init_mm.context);
1357 
1358 	clear_all_debug_regs();
1359 	dbg_restore_debug_regs();
1360 
1361 	fpu_init();
1362 
1363 	if (is_uv_system())
1364 		uv_cpu_init();
1365 }
1366 
1367 #else
1368 
1369 void cpu_init(void)
1370 {
1371 	int cpu = smp_processor_id();
1372 	struct task_struct *curr = current;
1373 	struct tss_struct *t = &per_cpu(init_tss, cpu);
1374 	struct thread_struct *thread = &curr->thread;
1375 
1376 	show_ucode_info_early();
1377 
1378 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1379 		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1380 		for (;;)
1381 			local_irq_enable();
1382 	}
1383 
1384 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1385 
1386 	if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1387 		clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1388 
1389 	load_current_idt();
1390 	switch_to_new_gdt(cpu);
1391 
1392 	/*
1393 	 * Set up and load the per-CPU TSS and LDT
1394 	 */
1395 	atomic_inc(&init_mm.mm_count);
1396 	curr->active_mm = &init_mm;
1397 	BUG_ON(curr->mm);
1398 	enter_lazy_tlb(&init_mm, curr);
1399 
1400 	load_sp0(t, thread);
1401 	set_tss_desc(cpu, t);
1402 	load_TR_desc();
1403 	load_LDT(&init_mm.context);
1404 
1405 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1406 
1407 #ifdef CONFIG_DOUBLEFAULT
1408 	/* Set up doublefault TSS pointer in the GDT */
1409 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1410 #endif
1411 
1412 	clear_all_debug_regs();
1413 	dbg_restore_debug_regs();
1414 
1415 	fpu_init();
1416 }
1417 #endif
1418 
1419 #ifdef CONFIG_X86_DEBUG_STATIC_CPU_HAS
1420 void warn_pre_alternatives(void)
1421 {
1422 	WARN(1, "You're using static_cpu_has before alternatives have run!\n");
1423 }
1424 EXPORT_SYMBOL_GPL(warn_pre_alternatives);
1425 #endif
1426 
1427 inline bool __static_cpu_has_safe(u16 bit)
1428 {
1429 	return boot_cpu_has(bit);
1430 }
1431 EXPORT_SYMBOL_GPL(__static_cpu_has_safe);
1432