xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 261a9af6)
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/hypervisor.h>
19 #include <asm/processor.h>
20 #include <asm/sections.h>
21 #include <linux/topology.h>
22 #include <linux/cpumask.h>
23 #include <asm/pgtable.h>
24 #include <asm/atomic.h>
25 #include <asm/proto.h>
26 #include <asm/setup.h>
27 #include <asm/apic.h>
28 #include <asm/desc.h>
29 #include <asm/i387.h>
30 #include <asm/mtrr.h>
31 #include <linux/numa.h>
32 #include <asm/asm.h>
33 #include <asm/cpu.h>
34 #include <asm/mce.h>
35 #include <asm/msr.h>
36 #include <asm/pat.h>
37 
38 #ifdef CONFIG_X86_LOCAL_APIC
39 #include <asm/uv/uv.h>
40 #endif
41 
42 #include "cpu.h"
43 
44 /* all of these masks are initialized in setup_cpu_local_masks() */
45 cpumask_var_t cpu_initialized_mask;
46 cpumask_var_t cpu_callout_mask;
47 cpumask_var_t cpu_callin_mask;
48 
49 /* representing cpus for which sibling maps can be computed */
50 cpumask_var_t cpu_sibling_setup_mask;
51 
52 /* correctly size the local cpu masks */
53 void __init setup_cpu_local_masks(void)
54 {
55 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
56 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
57 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
58 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
59 }
60 
61 static void __cpuinit default_init(struct cpuinfo_x86 *c)
62 {
63 #ifdef CONFIG_X86_64
64 	cpu_detect_cache_sizes(c);
65 #else
66 	/* Not much we can do here... */
67 	/* Check if at least it has cpuid */
68 	if (c->cpuid_level == -1) {
69 		/* No cpuid. It must be an ancient CPU */
70 		if (c->x86 == 4)
71 			strcpy(c->x86_model_id, "486");
72 		else if (c->x86 == 3)
73 			strcpy(c->x86_model_id, "386");
74 	}
75 #endif
76 }
77 
78 static const struct cpu_dev __cpuinitconst default_cpu = {
79 	.c_init		= default_init,
80 	.c_vendor	= "Unknown",
81 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
82 };
83 
84 static const struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
85 
86 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
87 #ifdef CONFIG_X86_64
88 	/*
89 	 * We need valid kernel segments for data and code in long mode too
90 	 * IRET will check the segment types  kkeil 2000/10/28
91 	 * Also sysret mandates a special GDT layout
92 	 *
93 	 * TLS descriptors are currently at a different place compared to i386.
94 	 * Hopefully nobody expects them at a fixed place (Wine?)
95 	 */
96 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
97 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
98 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
99 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
100 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
101 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
102 #else
103 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
104 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
105 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
106 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
107 	/*
108 	 * Segments used for calling PnP BIOS have byte granularity.
109 	 * They code segments and data segments have fixed 64k limits,
110 	 * the transfer segment sizes are set at run time.
111 	 */
112 	/* 32-bit code */
113 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
114 	/* 16-bit code */
115 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
116 	/* 16-bit data */
117 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
118 	/* 16-bit data */
119 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
120 	/* 16-bit data */
121 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
122 	/*
123 	 * The APM segments have byte granularity and their bases
124 	 * are set at run time.  All have 64k limits.
125 	 */
126 	/* 32-bit code */
127 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
128 	/* 16-bit code */
129 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
130 	/* data */
131 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
132 
133 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
134 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
135 	GDT_STACK_CANARY_INIT
136 #endif
137 } };
138 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
139 
140 static int __init x86_xsave_setup(char *s)
141 {
142 	setup_clear_cpu_cap(X86_FEATURE_XSAVE);
143 	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
144 	return 1;
145 }
146 __setup("noxsave", x86_xsave_setup);
147 
148 static int __init x86_xsaveopt_setup(char *s)
149 {
150 	setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
151 	return 1;
152 }
153 __setup("noxsaveopt", x86_xsaveopt_setup);
154 
155 #ifdef CONFIG_X86_32
156 static int cachesize_override __cpuinitdata = -1;
157 static int disable_x86_serial_nr __cpuinitdata = 1;
158 
159 static int __init cachesize_setup(char *str)
160 {
161 	get_option(&str, &cachesize_override);
162 	return 1;
163 }
164 __setup("cachesize=", cachesize_setup);
165 
166 static int __init x86_fxsr_setup(char *s)
167 {
168 	setup_clear_cpu_cap(X86_FEATURE_FXSR);
169 	setup_clear_cpu_cap(X86_FEATURE_XMM);
170 	return 1;
171 }
172 __setup("nofxsr", x86_fxsr_setup);
173 
174 static int __init x86_sep_setup(char *s)
175 {
176 	setup_clear_cpu_cap(X86_FEATURE_SEP);
177 	return 1;
178 }
179 __setup("nosep", x86_sep_setup);
180 
181 /* Standard macro to see if a specific flag is changeable */
182 static inline int flag_is_changeable_p(u32 flag)
183 {
184 	u32 f1, f2;
185 
186 	/*
187 	 * Cyrix and IDT cpus allow disabling of CPUID
188 	 * so the code below may return different results
189 	 * when it is executed before and after enabling
190 	 * the CPUID. Add "volatile" to not allow gcc to
191 	 * optimize the subsequent calls to this function.
192 	 */
193 	asm volatile ("pushfl		\n\t"
194 		      "pushfl		\n\t"
195 		      "popl %0		\n\t"
196 		      "movl %0, %1	\n\t"
197 		      "xorl %2, %0	\n\t"
198 		      "pushl %0		\n\t"
199 		      "popfl		\n\t"
200 		      "pushfl		\n\t"
201 		      "popl %0		\n\t"
202 		      "popfl		\n\t"
203 
204 		      : "=&r" (f1), "=&r" (f2)
205 		      : "ir" (flag));
206 
207 	return ((f1^f2) & flag) != 0;
208 }
209 
210 /* Probe for the CPUID instruction */
211 static int __cpuinit have_cpuid_p(void)
212 {
213 	return flag_is_changeable_p(X86_EFLAGS_ID);
214 }
215 
216 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
217 {
218 	unsigned long lo, hi;
219 
220 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
221 		return;
222 
223 	/* Disable processor serial number: */
224 
225 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
226 	lo |= 0x200000;
227 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
228 
229 	printk(KERN_NOTICE "CPU serial number disabled.\n");
230 	clear_cpu_cap(c, X86_FEATURE_PN);
231 
232 	/* Disabling the serial number may affect the cpuid level */
233 	c->cpuid_level = cpuid_eax(0);
234 }
235 
236 static int __init x86_serial_nr_setup(char *s)
237 {
238 	disable_x86_serial_nr = 0;
239 	return 1;
240 }
241 __setup("serialnumber", x86_serial_nr_setup);
242 #else
243 static inline int flag_is_changeable_p(u32 flag)
244 {
245 	return 1;
246 }
247 /* Probe for the CPUID instruction */
248 static inline int have_cpuid_p(void)
249 {
250 	return 1;
251 }
252 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
253 {
254 }
255 #endif
256 
257 static int disable_smep __cpuinitdata;
258 static __init int setup_disable_smep(char *arg)
259 {
260 	disable_smep = 1;
261 	return 1;
262 }
263 __setup("nosmep", setup_disable_smep);
264 
265 static __cpuinit void setup_smep(struct cpuinfo_x86 *c)
266 {
267 	if (cpu_has(c, X86_FEATURE_SMEP)) {
268 		if (unlikely(disable_smep)) {
269 			setup_clear_cpu_cap(X86_FEATURE_SMEP);
270 			clear_in_cr4(X86_CR4_SMEP);
271 		} else
272 			set_in_cr4(X86_CR4_SMEP);
273 	}
274 }
275 
276 /*
277  * Some CPU features depend on higher CPUID levels, which may not always
278  * be available due to CPUID level capping or broken virtualization
279  * software.  Add those features to this table to auto-disable them.
280  */
281 struct cpuid_dependent_feature {
282 	u32 feature;
283 	u32 level;
284 };
285 
286 static const struct cpuid_dependent_feature __cpuinitconst
287 cpuid_dependent_features[] = {
288 	{ X86_FEATURE_MWAIT,		0x00000005 },
289 	{ X86_FEATURE_DCA,		0x00000009 },
290 	{ X86_FEATURE_XSAVE,		0x0000000d },
291 	{ 0, 0 }
292 };
293 
294 static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
295 {
296 	const struct cpuid_dependent_feature *df;
297 
298 	for (df = cpuid_dependent_features; df->feature; df++) {
299 
300 		if (!cpu_has(c, df->feature))
301 			continue;
302 		/*
303 		 * Note: cpuid_level is set to -1 if unavailable, but
304 		 * extended_extended_level is set to 0 if unavailable
305 		 * and the legitimate extended levels are all negative
306 		 * when signed; hence the weird messing around with
307 		 * signs here...
308 		 */
309 		if (!((s32)df->level < 0 ?
310 		     (u32)df->level > (u32)c->extended_cpuid_level :
311 		     (s32)df->level > (s32)c->cpuid_level))
312 			continue;
313 
314 		clear_cpu_cap(c, df->feature);
315 		if (!warn)
316 			continue;
317 
318 		printk(KERN_WARNING
319 		       "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
320 				x86_cap_flags[df->feature], df->level);
321 	}
322 }
323 
324 /*
325  * Naming convention should be: <Name> [(<Codename>)]
326  * This table only is used unless init_<vendor>() below doesn't set it;
327  * in particular, if CPUID levels 0x80000002..4 are supported, this
328  * isn't used
329  */
330 
331 /* Look up CPU names by table lookup. */
332 static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
333 {
334 	const struct cpu_model_info *info;
335 
336 	if (c->x86_model >= 16)
337 		return NULL;	/* Range check */
338 
339 	if (!this_cpu)
340 		return NULL;
341 
342 	info = this_cpu->c_models;
343 
344 	while (info && info->family) {
345 		if (info->family == c->x86)
346 			return info->model_names[c->x86_model];
347 		info++;
348 	}
349 	return NULL;		/* Not found */
350 }
351 
352 __u32 cpu_caps_cleared[NCAPINTS] __cpuinitdata;
353 __u32 cpu_caps_set[NCAPINTS] __cpuinitdata;
354 
355 void load_percpu_segment(int cpu)
356 {
357 #ifdef CONFIG_X86_32
358 	loadsegment(fs, __KERNEL_PERCPU);
359 #else
360 	loadsegment(gs, 0);
361 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
362 #endif
363 	load_stack_canary_segment();
364 }
365 
366 /*
367  * Current gdt points %fs at the "master" per-cpu area: after this,
368  * it's on the real one.
369  */
370 void switch_to_new_gdt(int cpu)
371 {
372 	struct desc_ptr gdt_descr;
373 
374 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
375 	gdt_descr.size = GDT_SIZE - 1;
376 	load_gdt(&gdt_descr);
377 	/* Reload the per-cpu base */
378 
379 	load_percpu_segment(cpu);
380 }
381 
382 static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
383 
384 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
385 {
386 	unsigned int *v;
387 	char *p, *q;
388 
389 	if (c->extended_cpuid_level < 0x80000004)
390 		return;
391 
392 	v = (unsigned int *)c->x86_model_id;
393 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
394 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
395 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
396 	c->x86_model_id[48] = 0;
397 
398 	/*
399 	 * Intel chips right-justify this string for some dumb reason;
400 	 * undo that brain damage:
401 	 */
402 	p = q = &c->x86_model_id[0];
403 	while (*p == ' ')
404 		p++;
405 	if (p != q) {
406 		while (*p)
407 			*q++ = *p++;
408 		while (q <= &c->x86_model_id[48])
409 			*q++ = '\0';	/* Zero-pad the rest */
410 	}
411 }
412 
413 void __cpuinit cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
414 {
415 	unsigned int n, dummy, ebx, ecx, edx, l2size;
416 
417 	n = c->extended_cpuid_level;
418 
419 	if (n >= 0x80000005) {
420 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
421 		c->x86_cache_size = (ecx>>24) + (edx>>24);
422 #ifdef CONFIG_X86_64
423 		/* On K8 L1 TLB is inclusive, so don't count it */
424 		c->x86_tlbsize = 0;
425 #endif
426 	}
427 
428 	if (n < 0x80000006)	/* Some chips just has a large L1. */
429 		return;
430 
431 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
432 	l2size = ecx >> 16;
433 
434 #ifdef CONFIG_X86_64
435 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
436 #else
437 	/* do processor-specific cache resizing */
438 	if (this_cpu->c_size_cache)
439 		l2size = this_cpu->c_size_cache(c, l2size);
440 
441 	/* Allow user to override all this if necessary. */
442 	if (cachesize_override != -1)
443 		l2size = cachesize_override;
444 
445 	if (l2size == 0)
446 		return;		/* Again, no L2 cache is possible */
447 #endif
448 
449 	c->x86_cache_size = l2size;
450 }
451 
452 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
453 {
454 #ifdef CONFIG_X86_HT
455 	u32 eax, ebx, ecx, edx;
456 	int index_msb, core_bits;
457 	static bool printed;
458 
459 	if (!cpu_has(c, X86_FEATURE_HT))
460 		return;
461 
462 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
463 		goto out;
464 
465 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
466 		return;
467 
468 	cpuid(1, &eax, &ebx, &ecx, &edx);
469 
470 	smp_num_siblings = (ebx & 0xff0000) >> 16;
471 
472 	if (smp_num_siblings == 1) {
473 		printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
474 		goto out;
475 	}
476 
477 	if (smp_num_siblings <= 1)
478 		goto out;
479 
480 	index_msb = get_count_order(smp_num_siblings);
481 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
482 
483 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
484 
485 	index_msb = get_count_order(smp_num_siblings);
486 
487 	core_bits = get_count_order(c->x86_max_cores);
488 
489 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
490 				       ((1 << core_bits) - 1);
491 
492 out:
493 	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
494 		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
495 		       c->phys_proc_id);
496 		printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
497 		       c->cpu_core_id);
498 		printed = 1;
499 	}
500 #endif
501 }
502 
503 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
504 {
505 	char *v = c->x86_vendor_id;
506 	int i;
507 
508 	for (i = 0; i < X86_VENDOR_NUM; i++) {
509 		if (!cpu_devs[i])
510 			break;
511 
512 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
513 		    (cpu_devs[i]->c_ident[1] &&
514 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
515 
516 			this_cpu = cpu_devs[i];
517 			c->x86_vendor = this_cpu->c_x86_vendor;
518 			return;
519 		}
520 	}
521 
522 	printk_once(KERN_ERR
523 			"CPU: vendor_id '%s' unknown, using generic init.\n" \
524 			"CPU: Your system may be unstable.\n", v);
525 
526 	c->x86_vendor = X86_VENDOR_UNKNOWN;
527 	this_cpu = &default_cpu;
528 }
529 
530 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
531 {
532 	/* Get vendor name */
533 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
534 	      (unsigned int *)&c->x86_vendor_id[0],
535 	      (unsigned int *)&c->x86_vendor_id[8],
536 	      (unsigned int *)&c->x86_vendor_id[4]);
537 
538 	c->x86 = 4;
539 	/* Intel-defined flags: level 0x00000001 */
540 	if (c->cpuid_level >= 0x00000001) {
541 		u32 junk, tfms, cap0, misc;
542 
543 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
544 		c->x86 = (tfms >> 8) & 0xf;
545 		c->x86_model = (tfms >> 4) & 0xf;
546 		c->x86_mask = tfms & 0xf;
547 
548 		if (c->x86 == 0xf)
549 			c->x86 += (tfms >> 20) & 0xff;
550 		if (c->x86 >= 0x6)
551 			c->x86_model += ((tfms >> 16) & 0xf) << 4;
552 
553 		if (cap0 & (1<<19)) {
554 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
555 			c->x86_cache_alignment = c->x86_clflush_size;
556 		}
557 	}
558 }
559 
560 void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
561 {
562 	u32 tfms, xlvl;
563 	u32 ebx;
564 
565 	/* Intel-defined flags: level 0x00000001 */
566 	if (c->cpuid_level >= 0x00000001) {
567 		u32 capability, excap;
568 
569 		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
570 		c->x86_capability[0] = capability;
571 		c->x86_capability[4] = excap;
572 	}
573 
574 	/* Additional Intel-defined flags: level 0x00000007 */
575 	if (c->cpuid_level >= 0x00000007) {
576 		u32 eax, ebx, ecx, edx;
577 
578 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
579 
580 		c->x86_capability[9] = ebx;
581 	}
582 
583 	/* AMD-defined flags: level 0x80000001 */
584 	xlvl = cpuid_eax(0x80000000);
585 	c->extended_cpuid_level = xlvl;
586 
587 	if ((xlvl & 0xffff0000) == 0x80000000) {
588 		if (xlvl >= 0x80000001) {
589 			c->x86_capability[1] = cpuid_edx(0x80000001);
590 			c->x86_capability[6] = cpuid_ecx(0x80000001);
591 		}
592 	}
593 
594 	if (c->extended_cpuid_level >= 0x80000008) {
595 		u32 eax = cpuid_eax(0x80000008);
596 
597 		c->x86_virt_bits = (eax >> 8) & 0xff;
598 		c->x86_phys_bits = eax & 0xff;
599 	}
600 #ifdef CONFIG_X86_32
601 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
602 		c->x86_phys_bits = 36;
603 #endif
604 
605 	if (c->extended_cpuid_level >= 0x80000007)
606 		c->x86_power = cpuid_edx(0x80000007);
607 
608 	init_scattered_cpuid_features(c);
609 }
610 
611 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
612 {
613 #ifdef CONFIG_X86_32
614 	int i;
615 
616 	/*
617 	 * First of all, decide if this is a 486 or higher
618 	 * It's a 486 if we can modify the AC flag
619 	 */
620 	if (flag_is_changeable_p(X86_EFLAGS_AC))
621 		c->x86 = 4;
622 	else
623 		c->x86 = 3;
624 
625 	for (i = 0; i < X86_VENDOR_NUM; i++)
626 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
627 			c->x86_vendor_id[0] = 0;
628 			cpu_devs[i]->c_identify(c);
629 			if (c->x86_vendor_id[0]) {
630 				get_cpu_vendor(c);
631 				break;
632 			}
633 		}
634 #endif
635 }
636 
637 /*
638  * Do minimum CPU detection early.
639  * Fields really needed: vendor, cpuid_level, family, model, mask,
640  * cache alignment.
641  * The others are not touched to avoid unwanted side effects.
642  *
643  * WARNING: this function is only called on the BP.  Don't add code here
644  * that is supposed to run on all CPUs.
645  */
646 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
647 {
648 #ifdef CONFIG_X86_64
649 	c->x86_clflush_size = 64;
650 	c->x86_phys_bits = 36;
651 	c->x86_virt_bits = 48;
652 #else
653 	c->x86_clflush_size = 32;
654 	c->x86_phys_bits = 32;
655 	c->x86_virt_bits = 32;
656 #endif
657 	c->x86_cache_alignment = c->x86_clflush_size;
658 
659 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
660 	c->extended_cpuid_level = 0;
661 
662 	if (!have_cpuid_p())
663 		identify_cpu_without_cpuid(c);
664 
665 	/* cyrix could have cpuid enabled via c_identify()*/
666 	if (!have_cpuid_p())
667 		return;
668 
669 	cpu_detect(c);
670 
671 	get_cpu_vendor(c);
672 
673 	get_cpu_cap(c);
674 
675 	if (this_cpu->c_early_init)
676 		this_cpu->c_early_init(c);
677 
678 #ifdef CONFIG_SMP
679 	c->cpu_index = 0;
680 #endif
681 	filter_cpuid_features(c, false);
682 
683 	setup_smep(c);
684 }
685 
686 void __init early_cpu_init(void)
687 {
688 	const struct cpu_dev *const *cdev;
689 	int count = 0;
690 
691 #ifdef CONFIG_PROCESSOR_SELECT
692 	printk(KERN_INFO "KERNEL supported cpus:\n");
693 #endif
694 
695 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
696 		const struct cpu_dev *cpudev = *cdev;
697 
698 		if (count >= X86_VENDOR_NUM)
699 			break;
700 		cpu_devs[count] = cpudev;
701 		count++;
702 
703 #ifdef CONFIG_PROCESSOR_SELECT
704 		{
705 			unsigned int j;
706 
707 			for (j = 0; j < 2; j++) {
708 				if (!cpudev->c_ident[j])
709 					continue;
710 				printk(KERN_INFO "  %s %s\n", cpudev->c_vendor,
711 					cpudev->c_ident[j]);
712 			}
713 		}
714 #endif
715 	}
716 	early_identify_cpu(&boot_cpu_data);
717 }
718 
719 /*
720  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
721  * unfortunately, that's not true in practice because of early VIA
722  * chips and (more importantly) broken virtualizers that are not easy
723  * to detect. In the latter case it doesn't even *fail* reliably, so
724  * probing for it doesn't even work. Disable it completely on 32-bit
725  * unless we can find a reliable way to detect all the broken cases.
726  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
727  */
728 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
729 {
730 #ifdef CONFIG_X86_32
731 	clear_cpu_cap(c, X86_FEATURE_NOPL);
732 #else
733 	set_cpu_cap(c, X86_FEATURE_NOPL);
734 #endif
735 }
736 
737 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
738 {
739 	c->extended_cpuid_level = 0;
740 
741 	if (!have_cpuid_p())
742 		identify_cpu_without_cpuid(c);
743 
744 	/* cyrix could have cpuid enabled via c_identify()*/
745 	if (!have_cpuid_p())
746 		return;
747 
748 	cpu_detect(c);
749 
750 	get_cpu_vendor(c);
751 
752 	get_cpu_cap(c);
753 
754 	if (c->cpuid_level >= 0x00000001) {
755 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
756 #ifdef CONFIG_X86_32
757 # ifdef CONFIG_X86_HT
758 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
759 # else
760 		c->apicid = c->initial_apicid;
761 # endif
762 #endif
763 
764 #ifdef CONFIG_X86_HT
765 		c->phys_proc_id = c->initial_apicid;
766 #endif
767 	}
768 
769 	setup_smep(c);
770 
771 	get_model_name(c); /* Default name */
772 
773 	detect_nopl(c);
774 }
775 
776 /*
777  * This does the hard work of actually picking apart the CPU stuff...
778  */
779 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
780 {
781 	int i;
782 
783 	c->loops_per_jiffy = loops_per_jiffy;
784 	c->x86_cache_size = -1;
785 	c->x86_vendor = X86_VENDOR_UNKNOWN;
786 	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
787 	c->x86_vendor_id[0] = '\0'; /* Unset */
788 	c->x86_model_id[0] = '\0';  /* Unset */
789 	c->x86_max_cores = 1;
790 	c->x86_coreid_bits = 0;
791 #ifdef CONFIG_X86_64
792 	c->x86_clflush_size = 64;
793 	c->x86_phys_bits = 36;
794 	c->x86_virt_bits = 48;
795 #else
796 	c->cpuid_level = -1;	/* CPUID not detected */
797 	c->x86_clflush_size = 32;
798 	c->x86_phys_bits = 32;
799 	c->x86_virt_bits = 32;
800 #endif
801 	c->x86_cache_alignment = c->x86_clflush_size;
802 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
803 
804 	generic_identify(c);
805 
806 	if (this_cpu->c_identify)
807 		this_cpu->c_identify(c);
808 
809 	/* Clear/Set all flags overriden by options, after probe */
810 	for (i = 0; i < NCAPINTS; i++) {
811 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
812 		c->x86_capability[i] |= cpu_caps_set[i];
813 	}
814 
815 #ifdef CONFIG_X86_64
816 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
817 #endif
818 
819 	/*
820 	 * Vendor-specific initialization.  In this section we
821 	 * canonicalize the feature flags, meaning if there are
822 	 * features a certain CPU supports which CPUID doesn't
823 	 * tell us, CPUID claiming incorrect flags, or other bugs,
824 	 * we handle them here.
825 	 *
826 	 * At the end of this section, c->x86_capability better
827 	 * indicate the features this CPU genuinely supports!
828 	 */
829 	if (this_cpu->c_init)
830 		this_cpu->c_init(c);
831 
832 	/* Disable the PN if appropriate */
833 	squash_the_stupid_serial_number(c);
834 
835 	/*
836 	 * The vendor-specific functions might have changed features.
837 	 * Now we do "generic changes."
838 	 */
839 
840 	/* Filter out anything that depends on CPUID levels we don't have */
841 	filter_cpuid_features(c, true);
842 
843 	/* If the model name is still unset, do table lookup. */
844 	if (!c->x86_model_id[0]) {
845 		const char *p;
846 		p = table_lookup_model(c);
847 		if (p)
848 			strcpy(c->x86_model_id, p);
849 		else
850 			/* Last resort... */
851 			sprintf(c->x86_model_id, "%02x/%02x",
852 				c->x86, c->x86_model);
853 	}
854 
855 #ifdef CONFIG_X86_64
856 	detect_ht(c);
857 #endif
858 
859 	init_hypervisor(c);
860 
861 	/*
862 	 * Clear/Set all flags overriden by options, need do it
863 	 * before following smp all cpus cap AND.
864 	 */
865 	for (i = 0; i < NCAPINTS; i++) {
866 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
867 		c->x86_capability[i] |= cpu_caps_set[i];
868 	}
869 
870 	/*
871 	 * On SMP, boot_cpu_data holds the common feature set between
872 	 * all CPUs; so make sure that we indicate which features are
873 	 * common between the CPUs.  The first time this routine gets
874 	 * executed, c == &boot_cpu_data.
875 	 */
876 	if (c != &boot_cpu_data) {
877 		/* AND the already accumulated flags with these */
878 		for (i = 0; i < NCAPINTS; i++)
879 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
880 	}
881 
882 	/* Init Machine Check Exception if available. */
883 	mcheck_cpu_init(c);
884 
885 	select_idle_routine(c);
886 
887 #ifdef CONFIG_NUMA
888 	numa_add_cpu(smp_processor_id());
889 #endif
890 }
891 
892 #ifdef CONFIG_X86_64
893 static void vgetcpu_set_mode(void)
894 {
895 	if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
896 		vgetcpu_mode = VGETCPU_RDTSCP;
897 	else
898 		vgetcpu_mode = VGETCPU_LSL;
899 }
900 #endif
901 
902 void __init identify_boot_cpu(void)
903 {
904 	identify_cpu(&boot_cpu_data);
905 	init_amd_e400_c1e_mask();
906 #ifdef CONFIG_X86_32
907 	sysenter_setup();
908 	enable_sep_cpu();
909 #else
910 	vgetcpu_set_mode();
911 #endif
912 }
913 
914 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
915 {
916 	BUG_ON(c == &boot_cpu_data);
917 	identify_cpu(c);
918 #ifdef CONFIG_X86_32
919 	enable_sep_cpu();
920 #endif
921 	mtrr_ap_init();
922 }
923 
924 struct msr_range {
925 	unsigned	min;
926 	unsigned	max;
927 };
928 
929 static const struct msr_range msr_range_array[] __cpuinitconst = {
930 	{ 0x00000000, 0x00000418},
931 	{ 0xc0000000, 0xc000040b},
932 	{ 0xc0010000, 0xc0010142},
933 	{ 0xc0011000, 0xc001103b},
934 };
935 
936 static void __cpuinit print_cpu_msr(void)
937 {
938 	unsigned index_min, index_max;
939 	unsigned index;
940 	u64 val;
941 	int i;
942 
943 	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
944 		index_min = msr_range_array[i].min;
945 		index_max = msr_range_array[i].max;
946 
947 		for (index = index_min; index < index_max; index++) {
948 			if (rdmsrl_amd_safe(index, &val))
949 				continue;
950 			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
951 		}
952 	}
953 }
954 
955 static int show_msr __cpuinitdata;
956 
957 static __init int setup_show_msr(char *arg)
958 {
959 	int num;
960 
961 	get_option(&arg, &num);
962 
963 	if (num > 0)
964 		show_msr = num;
965 	return 1;
966 }
967 __setup("show_msr=", setup_show_msr);
968 
969 static __init int setup_noclflush(char *arg)
970 {
971 	setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
972 	return 1;
973 }
974 __setup("noclflush", setup_noclflush);
975 
976 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
977 {
978 	const char *vendor = NULL;
979 
980 	if (c->x86_vendor < X86_VENDOR_NUM) {
981 		vendor = this_cpu->c_vendor;
982 	} else {
983 		if (c->cpuid_level >= 0)
984 			vendor = c->x86_vendor_id;
985 	}
986 
987 	if (vendor && !strstr(c->x86_model_id, vendor))
988 		printk(KERN_CONT "%s ", vendor);
989 
990 	if (c->x86_model_id[0])
991 		printk(KERN_CONT "%s", c->x86_model_id);
992 	else
993 		printk(KERN_CONT "%d86", c->x86);
994 
995 	if (c->x86_mask || c->cpuid_level >= 0)
996 		printk(KERN_CONT " stepping %02x\n", c->x86_mask);
997 	else
998 		printk(KERN_CONT "\n");
999 
1000 #ifdef CONFIG_SMP
1001 	if (c->cpu_index < show_msr)
1002 		print_cpu_msr();
1003 #else
1004 	if (show_msr)
1005 		print_cpu_msr();
1006 #endif
1007 }
1008 
1009 static __init int setup_disablecpuid(char *arg)
1010 {
1011 	int bit;
1012 
1013 	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1014 		setup_clear_cpu_cap(bit);
1015 	else
1016 		return 0;
1017 
1018 	return 1;
1019 }
1020 __setup("clearcpuid=", setup_disablecpuid);
1021 
1022 #ifdef CONFIG_X86_64
1023 struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
1024 
1025 DEFINE_PER_CPU_FIRST(union irq_stack_union,
1026 		     irq_stack_union) __aligned(PAGE_SIZE);
1027 
1028 /*
1029  * The following four percpu variables are hot.  Align current_task to
1030  * cacheline size such that all four fall in the same cacheline.
1031  */
1032 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1033 	&init_task;
1034 EXPORT_PER_CPU_SYMBOL(current_task);
1035 
1036 DEFINE_PER_CPU(unsigned long, kernel_stack) =
1037 	(unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
1038 EXPORT_PER_CPU_SYMBOL(kernel_stack);
1039 
1040 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1041 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1042 
1043 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
1044 
1045 /*
1046  * Special IST stacks which the CPU switches to when it calls
1047  * an IST-marked descriptor entry. Up to 7 stacks (hardware
1048  * limit), all of them are 4K, except the debug stack which
1049  * is 8K.
1050  */
1051 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1052 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1053 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1054 };
1055 
1056 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1057 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1058 
1059 /* May not be marked __init: used by software suspend */
1060 void syscall_init(void)
1061 {
1062 	/*
1063 	 * LSTAR and STAR live in a bit strange symbiosis.
1064 	 * They both write to the same internal register. STAR allows to
1065 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1066 	 */
1067 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1068 	wrmsrl(MSR_LSTAR, system_call);
1069 	wrmsrl(MSR_CSTAR, ignore_sysret);
1070 
1071 #ifdef CONFIG_IA32_EMULATION
1072 	syscall32_cpu_init();
1073 #endif
1074 
1075 	/* Flags to clear on syscall */
1076 	wrmsrl(MSR_SYSCALL_MASK,
1077 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
1078 }
1079 
1080 unsigned long kernel_eflags;
1081 
1082 /*
1083  * Copies of the original ist values from the tss are only accessed during
1084  * debugging, no special alignment required.
1085  */
1086 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1087 
1088 #else	/* CONFIG_X86_64 */
1089 
1090 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1091 EXPORT_PER_CPU_SYMBOL(current_task);
1092 
1093 #ifdef CONFIG_CC_STACKPROTECTOR
1094 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1095 #endif
1096 
1097 /* Make sure %fs and %gs are initialized properly in idle threads */
1098 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
1099 {
1100 	memset(regs, 0, sizeof(struct pt_regs));
1101 	regs->fs = __KERNEL_PERCPU;
1102 	regs->gs = __KERNEL_STACK_CANARY;
1103 
1104 	return regs;
1105 }
1106 #endif	/* CONFIG_X86_64 */
1107 
1108 /*
1109  * Clear all 6 debug registers:
1110  */
1111 static void clear_all_debug_regs(void)
1112 {
1113 	int i;
1114 
1115 	for (i = 0; i < 8; i++) {
1116 		/* Ignore db4, db5 */
1117 		if ((i == 4) || (i == 5))
1118 			continue;
1119 
1120 		set_debugreg(0, i);
1121 	}
1122 }
1123 
1124 #ifdef CONFIG_KGDB
1125 /*
1126  * Restore debug regs if using kgdbwait and you have a kernel debugger
1127  * connection established.
1128  */
1129 static void dbg_restore_debug_regs(void)
1130 {
1131 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1132 		arch_kgdb_ops.correct_hw_break();
1133 }
1134 #else /* ! CONFIG_KGDB */
1135 #define dbg_restore_debug_regs()
1136 #endif /* ! CONFIG_KGDB */
1137 
1138 /*
1139  * cpu_init() initializes state that is per-CPU. Some data is already
1140  * initialized (naturally) in the bootstrap process, such as the GDT
1141  * and IDT. We reload them nevertheless, this function acts as a
1142  * 'CPU state barrier', nothing should get across.
1143  * A lot of state is already set up in PDA init for 64 bit
1144  */
1145 #ifdef CONFIG_X86_64
1146 
1147 void __cpuinit cpu_init(void)
1148 {
1149 	struct orig_ist *oist;
1150 	struct task_struct *me;
1151 	struct tss_struct *t;
1152 	unsigned long v;
1153 	int cpu;
1154 	int i;
1155 
1156 	cpu = stack_smp_processor_id();
1157 	t = &per_cpu(init_tss, cpu);
1158 	oist = &per_cpu(orig_ist, cpu);
1159 
1160 #ifdef CONFIG_NUMA
1161 	if (cpu != 0 && percpu_read(numa_node) == 0 &&
1162 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1163 		set_numa_node(early_cpu_to_node(cpu));
1164 #endif
1165 
1166 	me = current;
1167 
1168 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1169 		panic("CPU#%d already initialized!\n", cpu);
1170 
1171 	pr_debug("Initializing CPU#%d\n", cpu);
1172 
1173 	clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1174 
1175 	/*
1176 	 * Initialize the per-CPU GDT with the boot GDT,
1177 	 * and set up the GDT descriptor:
1178 	 */
1179 
1180 	switch_to_new_gdt(cpu);
1181 	loadsegment(fs, 0);
1182 
1183 	load_idt((const struct desc_ptr *)&idt_descr);
1184 
1185 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1186 	syscall_init();
1187 
1188 	wrmsrl(MSR_FS_BASE, 0);
1189 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1190 	barrier();
1191 
1192 	x86_configure_nx();
1193 	if (cpu != 0)
1194 		enable_x2apic();
1195 
1196 	/*
1197 	 * set up and load the per-CPU TSS
1198 	 */
1199 	if (!oist->ist[0]) {
1200 		char *estacks = per_cpu(exception_stacks, cpu);
1201 
1202 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1203 			estacks += exception_stack_sizes[v];
1204 			oist->ist[v] = t->x86_tss.ist[v] =
1205 					(unsigned long)estacks;
1206 		}
1207 	}
1208 
1209 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1210 
1211 	/*
1212 	 * <= is required because the CPU will access up to
1213 	 * 8 bits beyond the end of the IO permission bitmap.
1214 	 */
1215 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1216 		t->io_bitmap[i] = ~0UL;
1217 
1218 	atomic_inc(&init_mm.mm_count);
1219 	me->active_mm = &init_mm;
1220 	BUG_ON(me->mm);
1221 	enter_lazy_tlb(&init_mm, me);
1222 
1223 	load_sp0(t, &current->thread);
1224 	set_tss_desc(cpu, t);
1225 	load_TR_desc();
1226 	load_LDT(&init_mm.context);
1227 
1228 	clear_all_debug_regs();
1229 	dbg_restore_debug_regs();
1230 
1231 	fpu_init();
1232 	xsave_init();
1233 
1234 	raw_local_save_flags(kernel_eflags);
1235 
1236 	if (is_uv_system())
1237 		uv_cpu_init();
1238 }
1239 
1240 #else
1241 
1242 void __cpuinit cpu_init(void)
1243 {
1244 	int cpu = smp_processor_id();
1245 	struct task_struct *curr = current;
1246 	struct tss_struct *t = &per_cpu(init_tss, cpu);
1247 	struct thread_struct *thread = &curr->thread;
1248 
1249 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1250 		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1251 		for (;;)
1252 			local_irq_enable();
1253 	}
1254 
1255 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1256 
1257 	if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1258 		clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1259 
1260 	load_idt(&idt_descr);
1261 	switch_to_new_gdt(cpu);
1262 
1263 	/*
1264 	 * Set up and load the per-CPU TSS and LDT
1265 	 */
1266 	atomic_inc(&init_mm.mm_count);
1267 	curr->active_mm = &init_mm;
1268 	BUG_ON(curr->mm);
1269 	enter_lazy_tlb(&init_mm, curr);
1270 
1271 	load_sp0(t, thread);
1272 	set_tss_desc(cpu, t);
1273 	load_TR_desc();
1274 	load_LDT(&init_mm.context);
1275 
1276 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1277 
1278 #ifdef CONFIG_DOUBLEFAULT
1279 	/* Set up doublefault TSS pointer in the GDT */
1280 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1281 #endif
1282 
1283 	clear_all_debug_regs();
1284 	dbg_restore_debug_regs();
1285 
1286 	fpu_init();
1287 	xsave_init();
1288 }
1289 #endif
1290