xref: /linux/arch/x86/kernel/cpu/common.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
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/sections.h>
22 #include <linux/topology.h>
23 #include <linux/cpumask.h>
24 #include <asm/pgtable.h>
25 #include <linux/atomic.h>
26 #include <asm/proto.h>
27 #include <asm/setup.h>
28 #include <asm/apic.h>
29 #include <asm/desc.h>
30 #include <asm/i387.h>
31 #include <asm/mtrr.h>
32 #include <linux/numa.h>
33 #include <asm/asm.h>
34 #include <asm/cpu.h>
35 #include <asm/mce.h>
36 #include <asm/msr.h>
37 #include <asm/pat.h>
38 
39 #ifdef CONFIG_X86_LOCAL_APIC
40 #include <asm/uv/uv.h>
41 #endif
42 
43 #include "cpu.h"
44 
45 /* all of these masks are initialized in setup_cpu_local_masks() */
46 cpumask_var_t cpu_initialized_mask;
47 cpumask_var_t cpu_callout_mask;
48 cpumask_var_t cpu_callin_mask;
49 
50 /* representing cpus for which sibling maps can be computed */
51 cpumask_var_t cpu_sibling_setup_mask;
52 
53 /* correctly size the local cpu masks */
54 void __init setup_cpu_local_masks(void)
55 {
56 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
57 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
58 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
59 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
60 }
61 
62 static void __cpuinit default_init(struct cpuinfo_x86 *c)
63 {
64 #ifdef CONFIG_X86_64
65 	cpu_detect_cache_sizes(c);
66 #else
67 	/* Not much we can do here... */
68 	/* Check if at least it has cpuid */
69 	if (c->cpuid_level == -1) {
70 		/* No cpuid. It must be an ancient CPU */
71 		if (c->x86 == 4)
72 			strcpy(c->x86_model_id, "486");
73 		else if (c->x86 == 3)
74 			strcpy(c->x86_model_id, "386");
75 	}
76 #endif
77 }
78 
79 static const struct cpu_dev __cpuinitconst default_cpu = {
80 	.c_init		= default_init,
81 	.c_vendor	= "Unknown",
82 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
83 };
84 
85 static const struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
86 
87 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
88 #ifdef CONFIG_X86_64
89 	/*
90 	 * We need valid kernel segments for data and code in long mode too
91 	 * IRET will check the segment types  kkeil 2000/10/28
92 	 * Also sysret mandates a special GDT layout
93 	 *
94 	 * TLS descriptors are currently at a different place compared to i386.
95 	 * Hopefully nobody expects them at a fixed place (Wine?)
96 	 */
97 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
98 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
99 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
100 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
101 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
102 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
103 #else
104 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
105 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
106 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
107 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
108 	/*
109 	 * Segments used for calling PnP BIOS have byte granularity.
110 	 * They code segments and data segments have fixed 64k limits,
111 	 * the transfer segment sizes are set at run time.
112 	 */
113 	/* 32-bit code */
114 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
115 	/* 16-bit code */
116 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
117 	/* 16-bit data */
118 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
119 	/* 16-bit data */
120 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
121 	/* 16-bit data */
122 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
123 	/*
124 	 * The APM segments have byte granularity and their bases
125 	 * are set at run time.  All have 64k limits.
126 	 */
127 	/* 32-bit code */
128 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
129 	/* 16-bit code */
130 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
131 	/* data */
132 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
133 
134 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
135 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
136 	GDT_STACK_CANARY_INIT
137 #endif
138 } };
139 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
140 
141 static int __init x86_xsave_setup(char *s)
142 {
143 	setup_clear_cpu_cap(X86_FEATURE_XSAVE);
144 	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
145 	return 1;
146 }
147 __setup("noxsave", x86_xsave_setup);
148 
149 static int __init x86_xsaveopt_setup(char *s)
150 {
151 	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
152 	return 1;
153 }
154 __setup("noxsaveopt", x86_xsaveopt_setup);
155 
156 #ifdef CONFIG_X86_32
157 static int cachesize_override __cpuinitdata = -1;
158 static int disable_x86_serial_nr __cpuinitdata = 1;
159 
160 static int __init cachesize_setup(char *str)
161 {
162 	get_option(&str, &cachesize_override);
163 	return 1;
164 }
165 __setup("cachesize=", cachesize_setup);
166 
167 static int __init x86_fxsr_setup(char *s)
168 {
169 	setup_clear_cpu_cap(X86_FEATURE_FXSR);
170 	setup_clear_cpu_cap(X86_FEATURE_XMM);
171 	return 1;
172 }
173 __setup("nofxsr", x86_fxsr_setup);
174 
175 static int __init x86_sep_setup(char *s)
176 {
177 	setup_clear_cpu_cap(X86_FEATURE_SEP);
178 	return 1;
179 }
180 __setup("nosep", x86_sep_setup);
181 
182 /* Standard macro to see if a specific flag is changeable */
183 static inline int flag_is_changeable_p(u32 flag)
184 {
185 	u32 f1, f2;
186 
187 	/*
188 	 * Cyrix and IDT cpus allow disabling of CPUID
189 	 * so the code below may return different results
190 	 * when it is executed before and after enabling
191 	 * the CPUID. Add "volatile" to not allow gcc to
192 	 * optimize the subsequent calls to this function.
193 	 */
194 	asm volatile ("pushfl		\n\t"
195 		      "pushfl		\n\t"
196 		      "popl %0		\n\t"
197 		      "movl %0, %1	\n\t"
198 		      "xorl %2, %0	\n\t"
199 		      "pushl %0		\n\t"
200 		      "popfl		\n\t"
201 		      "pushfl		\n\t"
202 		      "popl %0		\n\t"
203 		      "popfl		\n\t"
204 
205 		      : "=&r" (f1), "=&r" (f2)
206 		      : "ir" (flag));
207 
208 	return ((f1^f2) & flag) != 0;
209 }
210 
211 /* Probe for the CPUID instruction */
212 static int __cpuinit have_cpuid_p(void)
213 {
214 	return flag_is_changeable_p(X86_EFLAGS_ID);
215 }
216 
217 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
218 {
219 	unsigned long lo, hi;
220 
221 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
222 		return;
223 
224 	/* Disable processor serial number: */
225 
226 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
227 	lo |= 0x200000;
228 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
229 
230 	printk(KERN_NOTICE "CPU serial number disabled.\n");
231 	clear_cpu_cap(c, X86_FEATURE_PN);
232 
233 	/* Disabling the serial number may affect the cpuid level */
234 	c->cpuid_level = cpuid_eax(0);
235 }
236 
237 static int __init x86_serial_nr_setup(char *s)
238 {
239 	disable_x86_serial_nr = 0;
240 	return 1;
241 }
242 __setup("serialnumber", x86_serial_nr_setup);
243 #else
244 static inline int flag_is_changeable_p(u32 flag)
245 {
246 	return 1;
247 }
248 /* Probe for the CPUID instruction */
249 static inline int have_cpuid_p(void)
250 {
251 	return 1;
252 }
253 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
254 {
255 }
256 #endif
257 
258 static int disable_smep __cpuinitdata;
259 static __init int setup_disable_smep(char *arg)
260 {
261 	disable_smep = 1;
262 	return 1;
263 }
264 __setup("nosmep", setup_disable_smep);
265 
266 static __cpuinit void setup_smep(struct cpuinfo_x86 *c)
267 {
268 	if (cpu_has(c, X86_FEATURE_SMEP)) {
269 		if (unlikely(disable_smep)) {
270 			setup_clear_cpu_cap(X86_FEATURE_SMEP);
271 			clear_in_cr4(X86_CR4_SMEP);
272 		} else
273 			set_in_cr4(X86_CR4_SMEP);
274 	}
275 }
276 
277 /*
278  * Some CPU features depend on higher CPUID levels, which may not always
279  * be available due to CPUID level capping or broken virtualization
280  * software.  Add those features to this table to auto-disable them.
281  */
282 struct cpuid_dependent_feature {
283 	u32 feature;
284 	u32 level;
285 };
286 
287 static const struct cpuid_dependent_feature __cpuinitconst
288 cpuid_dependent_features[] = {
289 	{ X86_FEATURE_MWAIT,		0x00000005 },
290 	{ X86_FEATURE_DCA,		0x00000009 },
291 	{ X86_FEATURE_XSAVE,		0x0000000d },
292 	{ 0, 0 }
293 };
294 
295 static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
296 {
297 	const struct cpuid_dependent_feature *df;
298 
299 	for (df = cpuid_dependent_features; df->feature; df++) {
300 
301 		if (!cpu_has(c, df->feature))
302 			continue;
303 		/*
304 		 * Note: cpuid_level is set to -1 if unavailable, but
305 		 * extended_extended_level is set to 0 if unavailable
306 		 * and the legitimate extended levels are all negative
307 		 * when signed; hence the weird messing around with
308 		 * signs here...
309 		 */
310 		if (!((s32)df->level < 0 ?
311 		     (u32)df->level > (u32)c->extended_cpuid_level :
312 		     (s32)df->level > (s32)c->cpuid_level))
313 			continue;
314 
315 		clear_cpu_cap(c, df->feature);
316 		if (!warn)
317 			continue;
318 
319 		printk(KERN_WARNING
320 		       "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
321 				x86_cap_flags[df->feature], df->level);
322 	}
323 }
324 
325 /*
326  * Naming convention should be: <Name> [(<Codename>)]
327  * This table only is used unless init_<vendor>() below doesn't set it;
328  * in particular, if CPUID levels 0x80000002..4 are supported, this
329  * isn't used
330  */
331 
332 /* Look up CPU names by table lookup. */
333 static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
334 {
335 	const struct cpu_model_info *info;
336 
337 	if (c->x86_model >= 16)
338 		return NULL;	/* Range check */
339 
340 	if (!this_cpu)
341 		return NULL;
342 
343 	info = this_cpu->c_models;
344 
345 	while (info && info->family) {
346 		if (info->family == c->x86)
347 			return info->model_names[c->x86_model];
348 		info++;
349 	}
350 	return NULL;		/* Not found */
351 }
352 
353 __u32 cpu_caps_cleared[NCAPINTS] __cpuinitdata;
354 __u32 cpu_caps_set[NCAPINTS] __cpuinitdata;
355 
356 void load_percpu_segment(int cpu)
357 {
358 #ifdef CONFIG_X86_32
359 	loadsegment(fs, __KERNEL_PERCPU);
360 #else
361 	loadsegment(gs, 0);
362 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
363 #endif
364 	load_stack_canary_segment();
365 }
366 
367 /*
368  * Current gdt points %fs at the "master" per-cpu area: after this,
369  * it's on the real one.
370  */
371 void switch_to_new_gdt(int cpu)
372 {
373 	struct desc_ptr gdt_descr;
374 
375 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
376 	gdt_descr.size = GDT_SIZE - 1;
377 	load_gdt(&gdt_descr);
378 	/* Reload the per-cpu base */
379 
380 	load_percpu_segment(cpu);
381 }
382 
383 static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
384 
385 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
386 {
387 	unsigned int *v;
388 	char *p, *q;
389 
390 	if (c->extended_cpuid_level < 0x80000004)
391 		return;
392 
393 	v = (unsigned int *)c->x86_model_id;
394 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
395 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
396 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
397 	c->x86_model_id[48] = 0;
398 
399 	/*
400 	 * Intel chips right-justify this string for some dumb reason;
401 	 * undo that brain damage:
402 	 */
403 	p = q = &c->x86_model_id[0];
404 	while (*p == ' ')
405 		p++;
406 	if (p != q) {
407 		while (*p)
408 			*q++ = *p++;
409 		while (q <= &c->x86_model_id[48])
410 			*q++ = '\0';	/* Zero-pad the rest */
411 	}
412 }
413 
414 void __cpuinit cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
415 {
416 	unsigned int n, dummy, ebx, ecx, edx, l2size;
417 
418 	n = c->extended_cpuid_level;
419 
420 	if (n >= 0x80000005) {
421 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
422 		c->x86_cache_size = (ecx>>24) + (edx>>24);
423 #ifdef CONFIG_X86_64
424 		/* On K8 L1 TLB is inclusive, so don't count it */
425 		c->x86_tlbsize = 0;
426 #endif
427 	}
428 
429 	if (n < 0x80000006)	/* Some chips just has a large L1. */
430 		return;
431 
432 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
433 	l2size = ecx >> 16;
434 
435 #ifdef CONFIG_X86_64
436 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
437 #else
438 	/* do processor-specific cache resizing */
439 	if (this_cpu->c_size_cache)
440 		l2size = this_cpu->c_size_cache(c, l2size);
441 
442 	/* Allow user to override all this if necessary. */
443 	if (cachesize_override != -1)
444 		l2size = cachesize_override;
445 
446 	if (l2size == 0)
447 		return;		/* Again, no L2 cache is possible */
448 #endif
449 
450 	c->x86_cache_size = l2size;
451 }
452 
453 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
454 {
455 #ifdef CONFIG_X86_HT
456 	u32 eax, ebx, ecx, edx;
457 	int index_msb, core_bits;
458 	static bool printed;
459 
460 	if (!cpu_has(c, X86_FEATURE_HT))
461 		return;
462 
463 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
464 		goto out;
465 
466 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
467 		return;
468 
469 	cpuid(1, &eax, &ebx, &ecx, &edx);
470 
471 	smp_num_siblings = (ebx & 0xff0000) >> 16;
472 
473 	if (smp_num_siblings == 1) {
474 		printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
475 		goto out;
476 	}
477 
478 	if (smp_num_siblings <= 1)
479 		goto out;
480 
481 	index_msb = get_count_order(smp_num_siblings);
482 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
483 
484 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
485 
486 	index_msb = get_count_order(smp_num_siblings);
487 
488 	core_bits = get_count_order(c->x86_max_cores);
489 
490 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
491 				       ((1 << core_bits) - 1);
492 
493 out:
494 	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
495 		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
496 		       c->phys_proc_id);
497 		printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
498 		       c->cpu_core_id);
499 		printed = 1;
500 	}
501 #endif
502 }
503 
504 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
505 {
506 	char *v = c->x86_vendor_id;
507 	int i;
508 
509 	for (i = 0; i < X86_VENDOR_NUM; i++) {
510 		if (!cpu_devs[i])
511 			break;
512 
513 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
514 		    (cpu_devs[i]->c_ident[1] &&
515 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
516 
517 			this_cpu = cpu_devs[i];
518 			c->x86_vendor = this_cpu->c_x86_vendor;
519 			return;
520 		}
521 	}
522 
523 	printk_once(KERN_ERR
524 			"CPU: vendor_id '%s' unknown, using generic init.\n" \
525 			"CPU: Your system may be unstable.\n", v);
526 
527 	c->x86_vendor = X86_VENDOR_UNKNOWN;
528 	this_cpu = &default_cpu;
529 }
530 
531 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
532 {
533 	/* Get vendor name */
534 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
535 	      (unsigned int *)&c->x86_vendor_id[0],
536 	      (unsigned int *)&c->x86_vendor_id[8],
537 	      (unsigned int *)&c->x86_vendor_id[4]);
538 
539 	c->x86 = 4;
540 	/* Intel-defined flags: level 0x00000001 */
541 	if (c->cpuid_level >= 0x00000001) {
542 		u32 junk, tfms, cap0, misc;
543 
544 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
545 		c->x86 = (tfms >> 8) & 0xf;
546 		c->x86_model = (tfms >> 4) & 0xf;
547 		c->x86_mask = tfms & 0xf;
548 
549 		if (c->x86 == 0xf)
550 			c->x86 += (tfms >> 20) & 0xff;
551 		if (c->x86 >= 0x6)
552 			c->x86_model += ((tfms >> 16) & 0xf) << 4;
553 
554 		if (cap0 & (1<<19)) {
555 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
556 			c->x86_cache_alignment = c->x86_clflush_size;
557 		}
558 	}
559 }
560 
561 void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
562 {
563 	u32 tfms, xlvl;
564 	u32 ebx;
565 
566 	/* Intel-defined flags: level 0x00000001 */
567 	if (c->cpuid_level >= 0x00000001) {
568 		u32 capability, excap;
569 
570 		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
571 		c->x86_capability[0] = capability;
572 		c->x86_capability[4] = excap;
573 	}
574 
575 	/* Additional Intel-defined flags: level 0x00000007 */
576 	if (c->cpuid_level >= 0x00000007) {
577 		u32 eax, ebx, ecx, edx;
578 
579 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
580 
581 		c->x86_capability[9] = ebx;
582 	}
583 
584 	/* AMD-defined flags: level 0x80000001 */
585 	xlvl = cpuid_eax(0x80000000);
586 	c->extended_cpuid_level = xlvl;
587 
588 	if ((xlvl & 0xffff0000) == 0x80000000) {
589 		if (xlvl >= 0x80000001) {
590 			c->x86_capability[1] = cpuid_edx(0x80000001);
591 			c->x86_capability[6] = cpuid_ecx(0x80000001);
592 		}
593 	}
594 
595 	if (c->extended_cpuid_level >= 0x80000008) {
596 		u32 eax = cpuid_eax(0x80000008);
597 
598 		c->x86_virt_bits = (eax >> 8) & 0xff;
599 		c->x86_phys_bits = eax & 0xff;
600 	}
601 #ifdef CONFIG_X86_32
602 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
603 		c->x86_phys_bits = 36;
604 #endif
605 
606 	if (c->extended_cpuid_level >= 0x80000007)
607 		c->x86_power = cpuid_edx(0x80000007);
608 
609 	init_scattered_cpuid_features(c);
610 }
611 
612 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
613 {
614 #ifdef CONFIG_X86_32
615 	int i;
616 
617 	/*
618 	 * First of all, decide if this is a 486 or higher
619 	 * It's a 486 if we can modify the AC flag
620 	 */
621 	if (flag_is_changeable_p(X86_EFLAGS_AC))
622 		c->x86 = 4;
623 	else
624 		c->x86 = 3;
625 
626 	for (i = 0; i < X86_VENDOR_NUM; i++)
627 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
628 			c->x86_vendor_id[0] = 0;
629 			cpu_devs[i]->c_identify(c);
630 			if (c->x86_vendor_id[0]) {
631 				get_cpu_vendor(c);
632 				break;
633 			}
634 		}
635 #endif
636 }
637 
638 /*
639  * Do minimum CPU detection early.
640  * Fields really needed: vendor, cpuid_level, family, model, mask,
641  * cache alignment.
642  * The others are not touched to avoid unwanted side effects.
643  *
644  * WARNING: this function is only called on the BP.  Don't add code here
645  * that is supposed to run on all CPUs.
646  */
647 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
648 {
649 #ifdef CONFIG_X86_64
650 	c->x86_clflush_size = 64;
651 	c->x86_phys_bits = 36;
652 	c->x86_virt_bits = 48;
653 #else
654 	c->x86_clflush_size = 32;
655 	c->x86_phys_bits = 32;
656 	c->x86_virt_bits = 32;
657 #endif
658 	c->x86_cache_alignment = c->x86_clflush_size;
659 
660 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
661 	c->extended_cpuid_level = 0;
662 
663 	if (!have_cpuid_p())
664 		identify_cpu_without_cpuid(c);
665 
666 	/* cyrix could have cpuid enabled via c_identify()*/
667 	if (!have_cpuid_p())
668 		return;
669 
670 	cpu_detect(c);
671 
672 	get_cpu_vendor(c);
673 
674 	get_cpu_cap(c);
675 
676 	if (this_cpu->c_early_init)
677 		this_cpu->c_early_init(c);
678 
679 #ifdef CONFIG_SMP
680 	c->cpu_index = 0;
681 #endif
682 	filter_cpuid_features(c, false);
683 
684 	setup_smep(c);
685 
686 	if (this_cpu->c_bsp_init)
687 		this_cpu->c_bsp_init(c);
688 }
689 
690 void __init early_cpu_init(void)
691 {
692 	const struct cpu_dev *const *cdev;
693 	int count = 0;
694 
695 #ifdef CONFIG_PROCESSOR_SELECT
696 	printk(KERN_INFO "KERNEL supported cpus:\n");
697 #endif
698 
699 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
700 		const struct cpu_dev *cpudev = *cdev;
701 
702 		if (count >= X86_VENDOR_NUM)
703 			break;
704 		cpu_devs[count] = cpudev;
705 		count++;
706 
707 #ifdef CONFIG_PROCESSOR_SELECT
708 		{
709 			unsigned int j;
710 
711 			for (j = 0; j < 2; j++) {
712 				if (!cpudev->c_ident[j])
713 					continue;
714 				printk(KERN_INFO "  %s %s\n", cpudev->c_vendor,
715 					cpudev->c_ident[j]);
716 			}
717 		}
718 #endif
719 	}
720 	early_identify_cpu(&boot_cpu_data);
721 }
722 
723 /*
724  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
725  * unfortunately, that's not true in practice because of early VIA
726  * chips and (more importantly) broken virtualizers that are not easy
727  * to detect. In the latter case it doesn't even *fail* reliably, so
728  * probing for it doesn't even work. Disable it completely on 32-bit
729  * unless we can find a reliable way to detect all the broken cases.
730  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
731  */
732 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
733 {
734 #ifdef CONFIG_X86_32
735 	clear_cpu_cap(c, X86_FEATURE_NOPL);
736 #else
737 	set_cpu_cap(c, X86_FEATURE_NOPL);
738 #endif
739 }
740 
741 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
742 {
743 	c->extended_cpuid_level = 0;
744 
745 	if (!have_cpuid_p())
746 		identify_cpu_without_cpuid(c);
747 
748 	/* cyrix could have cpuid enabled via c_identify()*/
749 	if (!have_cpuid_p())
750 		return;
751 
752 	cpu_detect(c);
753 
754 	get_cpu_vendor(c);
755 
756 	get_cpu_cap(c);
757 
758 	if (c->cpuid_level >= 0x00000001) {
759 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
760 #ifdef CONFIG_X86_32
761 # ifdef CONFIG_X86_HT
762 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
763 # else
764 		c->apicid = c->initial_apicid;
765 # endif
766 #endif
767 
768 #ifdef CONFIG_X86_HT
769 		c->phys_proc_id = c->initial_apicid;
770 #endif
771 	}
772 
773 	setup_smep(c);
774 
775 	get_model_name(c); /* Default name */
776 
777 	detect_nopl(c);
778 }
779 
780 /*
781  * This does the hard work of actually picking apart the CPU stuff...
782  */
783 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
784 {
785 	int i;
786 
787 	c->loops_per_jiffy = loops_per_jiffy;
788 	c->x86_cache_size = -1;
789 	c->x86_vendor = X86_VENDOR_UNKNOWN;
790 	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
791 	c->x86_vendor_id[0] = '\0'; /* Unset */
792 	c->x86_model_id[0] = '\0';  /* Unset */
793 	c->x86_max_cores = 1;
794 	c->x86_coreid_bits = 0;
795 #ifdef CONFIG_X86_64
796 	c->x86_clflush_size = 64;
797 	c->x86_phys_bits = 36;
798 	c->x86_virt_bits = 48;
799 #else
800 	c->cpuid_level = -1;	/* CPUID not detected */
801 	c->x86_clflush_size = 32;
802 	c->x86_phys_bits = 32;
803 	c->x86_virt_bits = 32;
804 #endif
805 	c->x86_cache_alignment = c->x86_clflush_size;
806 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
807 
808 	generic_identify(c);
809 
810 	if (this_cpu->c_identify)
811 		this_cpu->c_identify(c);
812 
813 	/* Clear/Set all flags overriden by options, after probe */
814 	for (i = 0; i < NCAPINTS; i++) {
815 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
816 		c->x86_capability[i] |= cpu_caps_set[i];
817 	}
818 
819 #ifdef CONFIG_X86_64
820 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
821 #endif
822 
823 	/*
824 	 * Vendor-specific initialization.  In this section we
825 	 * canonicalize the feature flags, meaning if there are
826 	 * features a certain CPU supports which CPUID doesn't
827 	 * tell us, CPUID claiming incorrect flags, or other bugs,
828 	 * we handle them here.
829 	 *
830 	 * At the end of this section, c->x86_capability better
831 	 * indicate the features this CPU genuinely supports!
832 	 */
833 	if (this_cpu->c_init)
834 		this_cpu->c_init(c);
835 
836 	/* Disable the PN if appropriate */
837 	squash_the_stupid_serial_number(c);
838 
839 	/*
840 	 * The vendor-specific functions might have changed features.
841 	 * Now we do "generic changes."
842 	 */
843 
844 	/* Filter out anything that depends on CPUID levels we don't have */
845 	filter_cpuid_features(c, true);
846 
847 	/* If the model name is still unset, do table lookup. */
848 	if (!c->x86_model_id[0]) {
849 		const char *p;
850 		p = table_lookup_model(c);
851 		if (p)
852 			strcpy(c->x86_model_id, p);
853 		else
854 			/* Last resort... */
855 			sprintf(c->x86_model_id, "%02x/%02x",
856 				c->x86, c->x86_model);
857 	}
858 
859 #ifdef CONFIG_X86_64
860 	detect_ht(c);
861 #endif
862 
863 	init_hypervisor(c);
864 	x86_init_rdrand(c);
865 
866 	/*
867 	 * Clear/Set all flags overriden by options, need do it
868 	 * before following smp all cpus cap AND.
869 	 */
870 	for (i = 0; i < NCAPINTS; i++) {
871 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
872 		c->x86_capability[i] |= cpu_caps_set[i];
873 	}
874 
875 	/*
876 	 * On SMP, boot_cpu_data holds the common feature set between
877 	 * all CPUs; so make sure that we indicate which features are
878 	 * common between the CPUs.  The first time this routine gets
879 	 * executed, c == &boot_cpu_data.
880 	 */
881 	if (c != &boot_cpu_data) {
882 		/* AND the already accumulated flags with these */
883 		for (i = 0; i < NCAPINTS; i++)
884 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
885 	}
886 
887 	/* Init Machine Check Exception if available. */
888 	mcheck_cpu_init(c);
889 
890 	select_idle_routine(c);
891 
892 #ifdef CONFIG_NUMA
893 	numa_add_cpu(smp_processor_id());
894 #endif
895 }
896 
897 #ifdef CONFIG_X86_64
898 static void vgetcpu_set_mode(void)
899 {
900 	if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
901 		vgetcpu_mode = VGETCPU_RDTSCP;
902 	else
903 		vgetcpu_mode = VGETCPU_LSL;
904 }
905 #endif
906 
907 void __init identify_boot_cpu(void)
908 {
909 	identify_cpu(&boot_cpu_data);
910 	init_amd_e400_c1e_mask();
911 #ifdef CONFIG_X86_32
912 	sysenter_setup();
913 	enable_sep_cpu();
914 #else
915 	vgetcpu_set_mode();
916 #endif
917 }
918 
919 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
920 {
921 	BUG_ON(c == &boot_cpu_data);
922 	identify_cpu(c);
923 #ifdef CONFIG_X86_32
924 	enable_sep_cpu();
925 #endif
926 	mtrr_ap_init();
927 }
928 
929 struct msr_range {
930 	unsigned	min;
931 	unsigned	max;
932 };
933 
934 static const struct msr_range msr_range_array[] __cpuinitconst = {
935 	{ 0x00000000, 0x00000418},
936 	{ 0xc0000000, 0xc000040b},
937 	{ 0xc0010000, 0xc0010142},
938 	{ 0xc0011000, 0xc001103b},
939 };
940 
941 static void __cpuinit print_cpu_msr(void)
942 {
943 	unsigned index_min, index_max;
944 	unsigned index;
945 	u64 val;
946 	int i;
947 
948 	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
949 		index_min = msr_range_array[i].min;
950 		index_max = msr_range_array[i].max;
951 
952 		for (index = index_min; index < index_max; index++) {
953 			if (rdmsrl_amd_safe(index, &val))
954 				continue;
955 			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
956 		}
957 	}
958 }
959 
960 static int show_msr __cpuinitdata;
961 
962 static __init int setup_show_msr(char *arg)
963 {
964 	int num;
965 
966 	get_option(&arg, &num);
967 
968 	if (num > 0)
969 		show_msr = num;
970 	return 1;
971 }
972 __setup("show_msr=", setup_show_msr);
973 
974 static __init int setup_noclflush(char *arg)
975 {
976 	setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
977 	return 1;
978 }
979 __setup("noclflush", setup_noclflush);
980 
981 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
982 {
983 	const char *vendor = NULL;
984 
985 	if (c->x86_vendor < X86_VENDOR_NUM) {
986 		vendor = this_cpu->c_vendor;
987 	} else {
988 		if (c->cpuid_level >= 0)
989 			vendor = c->x86_vendor_id;
990 	}
991 
992 	if (vendor && !strstr(c->x86_model_id, vendor))
993 		printk(KERN_CONT "%s ", vendor);
994 
995 	if (c->x86_model_id[0])
996 		printk(KERN_CONT "%s", c->x86_model_id);
997 	else
998 		printk(KERN_CONT "%d86", c->x86);
999 
1000 	if (c->x86_mask || c->cpuid_level >= 0)
1001 		printk(KERN_CONT " stepping %02x\n", c->x86_mask);
1002 	else
1003 		printk(KERN_CONT "\n");
1004 
1005 #ifdef CONFIG_SMP
1006 	if (c->cpu_index < show_msr)
1007 		print_cpu_msr();
1008 #else
1009 	if (show_msr)
1010 		print_cpu_msr();
1011 #endif
1012 }
1013 
1014 static __init int setup_disablecpuid(char *arg)
1015 {
1016 	int bit;
1017 
1018 	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1019 		setup_clear_cpu_cap(bit);
1020 	else
1021 		return 0;
1022 
1023 	return 1;
1024 }
1025 __setup("clearcpuid=", setup_disablecpuid);
1026 
1027 #ifdef CONFIG_X86_64
1028 struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
1029 
1030 DEFINE_PER_CPU_FIRST(union irq_stack_union,
1031 		     irq_stack_union) __aligned(PAGE_SIZE);
1032 
1033 /*
1034  * The following four percpu variables are hot.  Align current_task to
1035  * cacheline size such that all four fall in the same cacheline.
1036  */
1037 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1038 	&init_task;
1039 EXPORT_PER_CPU_SYMBOL(current_task);
1040 
1041 DEFINE_PER_CPU(unsigned long, kernel_stack) =
1042 	(unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
1043 EXPORT_PER_CPU_SYMBOL(kernel_stack);
1044 
1045 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1046 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1047 
1048 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
1049 
1050 /*
1051  * Special IST stacks which the CPU switches to when it calls
1052  * an IST-marked descriptor entry. Up to 7 stacks (hardware
1053  * limit), all of them are 4K, except the debug stack which
1054  * is 8K.
1055  */
1056 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1057 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1058 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1059 };
1060 
1061 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1062 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1063 
1064 /* May not be marked __init: used by software suspend */
1065 void syscall_init(void)
1066 {
1067 	/*
1068 	 * LSTAR and STAR live in a bit strange symbiosis.
1069 	 * They both write to the same internal register. STAR allows to
1070 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1071 	 */
1072 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1073 	wrmsrl(MSR_LSTAR, system_call);
1074 	wrmsrl(MSR_CSTAR, ignore_sysret);
1075 
1076 #ifdef CONFIG_IA32_EMULATION
1077 	syscall32_cpu_init();
1078 #endif
1079 
1080 	/* Flags to clear on syscall */
1081 	wrmsrl(MSR_SYSCALL_MASK,
1082 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
1083 }
1084 
1085 unsigned long kernel_eflags;
1086 
1087 /*
1088  * Copies of the original ist values from the tss are only accessed during
1089  * debugging, no special alignment required.
1090  */
1091 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1092 
1093 #else	/* CONFIG_X86_64 */
1094 
1095 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1096 EXPORT_PER_CPU_SYMBOL(current_task);
1097 
1098 #ifdef CONFIG_CC_STACKPROTECTOR
1099 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1100 #endif
1101 
1102 /* Make sure %fs and %gs are initialized properly in idle threads */
1103 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
1104 {
1105 	memset(regs, 0, sizeof(struct pt_regs));
1106 	regs->fs = __KERNEL_PERCPU;
1107 	regs->gs = __KERNEL_STACK_CANARY;
1108 
1109 	return regs;
1110 }
1111 #endif	/* CONFIG_X86_64 */
1112 
1113 /*
1114  * Clear all 6 debug registers:
1115  */
1116 static void clear_all_debug_regs(void)
1117 {
1118 	int i;
1119 
1120 	for (i = 0; i < 8; i++) {
1121 		/* Ignore db4, db5 */
1122 		if ((i == 4) || (i == 5))
1123 			continue;
1124 
1125 		set_debugreg(0, i);
1126 	}
1127 }
1128 
1129 #ifdef CONFIG_KGDB
1130 /*
1131  * Restore debug regs if using kgdbwait and you have a kernel debugger
1132  * connection established.
1133  */
1134 static void dbg_restore_debug_regs(void)
1135 {
1136 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1137 		arch_kgdb_ops.correct_hw_break();
1138 }
1139 #else /* ! CONFIG_KGDB */
1140 #define dbg_restore_debug_regs()
1141 #endif /* ! CONFIG_KGDB */
1142 
1143 /*
1144  * cpu_init() initializes state that is per-CPU. Some data is already
1145  * initialized (naturally) in the bootstrap process, such as the GDT
1146  * and IDT. We reload them nevertheless, this function acts as a
1147  * 'CPU state barrier', nothing should get across.
1148  * A lot of state is already set up in PDA init for 64 bit
1149  */
1150 #ifdef CONFIG_X86_64
1151 
1152 void __cpuinit cpu_init(void)
1153 {
1154 	struct orig_ist *oist;
1155 	struct task_struct *me;
1156 	struct tss_struct *t;
1157 	unsigned long v;
1158 	int cpu;
1159 	int i;
1160 
1161 	cpu = stack_smp_processor_id();
1162 	t = &per_cpu(init_tss, cpu);
1163 	oist = &per_cpu(orig_ist, cpu);
1164 
1165 #ifdef CONFIG_NUMA
1166 	if (cpu != 0 && percpu_read(numa_node) == 0 &&
1167 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1168 		set_numa_node(early_cpu_to_node(cpu));
1169 #endif
1170 
1171 	me = current;
1172 
1173 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1174 		panic("CPU#%d already initialized!\n", cpu);
1175 
1176 	pr_debug("Initializing CPU#%d\n", cpu);
1177 
1178 	clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1179 
1180 	/*
1181 	 * Initialize the per-CPU GDT with the boot GDT,
1182 	 * and set up the GDT descriptor:
1183 	 */
1184 
1185 	switch_to_new_gdt(cpu);
1186 	loadsegment(fs, 0);
1187 
1188 	load_idt((const struct desc_ptr *)&idt_descr);
1189 
1190 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1191 	syscall_init();
1192 
1193 	wrmsrl(MSR_FS_BASE, 0);
1194 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1195 	barrier();
1196 
1197 	x86_configure_nx();
1198 	if (cpu != 0)
1199 		enable_x2apic();
1200 
1201 	/*
1202 	 * set up and load the per-CPU TSS
1203 	 */
1204 	if (!oist->ist[0]) {
1205 		char *estacks = per_cpu(exception_stacks, cpu);
1206 
1207 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1208 			estacks += exception_stack_sizes[v];
1209 			oist->ist[v] = t->x86_tss.ist[v] =
1210 					(unsigned long)estacks;
1211 		}
1212 	}
1213 
1214 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1215 
1216 	/*
1217 	 * <= is required because the CPU will access up to
1218 	 * 8 bits beyond the end of the IO permission bitmap.
1219 	 */
1220 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1221 		t->io_bitmap[i] = ~0UL;
1222 
1223 	atomic_inc(&init_mm.mm_count);
1224 	me->active_mm = &init_mm;
1225 	BUG_ON(me->mm);
1226 	enter_lazy_tlb(&init_mm, me);
1227 
1228 	load_sp0(t, &current->thread);
1229 	set_tss_desc(cpu, t);
1230 	load_TR_desc();
1231 	load_LDT(&init_mm.context);
1232 
1233 	clear_all_debug_regs();
1234 	dbg_restore_debug_regs();
1235 
1236 	fpu_init();
1237 	xsave_init();
1238 
1239 	raw_local_save_flags(kernel_eflags);
1240 
1241 	if (is_uv_system())
1242 		uv_cpu_init();
1243 }
1244 
1245 #else
1246 
1247 void __cpuinit cpu_init(void)
1248 {
1249 	int cpu = smp_processor_id();
1250 	struct task_struct *curr = current;
1251 	struct tss_struct *t = &per_cpu(init_tss, cpu);
1252 	struct thread_struct *thread = &curr->thread;
1253 
1254 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1255 		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1256 		for (;;)
1257 			local_irq_enable();
1258 	}
1259 
1260 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1261 
1262 	if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1263 		clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1264 
1265 	load_idt(&idt_descr);
1266 	switch_to_new_gdt(cpu);
1267 
1268 	/*
1269 	 * Set up and load the per-CPU TSS and LDT
1270 	 */
1271 	atomic_inc(&init_mm.mm_count);
1272 	curr->active_mm = &init_mm;
1273 	BUG_ON(curr->mm);
1274 	enter_lazy_tlb(&init_mm, curr);
1275 
1276 	load_sp0(t, thread);
1277 	set_tss_desc(cpu, t);
1278 	load_TR_desc();
1279 	load_LDT(&init_mm.context);
1280 
1281 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1282 
1283 #ifdef CONFIG_DOUBLEFAULT
1284 	/* Set up doublefault TSS pointer in the GDT */
1285 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1286 #endif
1287 
1288 	clear_all_debug_regs();
1289 	dbg_restore_debug_regs();
1290 
1291 	fpu_init();
1292 	xsave_init();
1293 }
1294 #endif
1295