xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision b04b4f78)
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/mmu_context.h>
17 #include <asm/hypervisor.h>
18 #include <asm/processor.h>
19 #include <asm/sections.h>
20 #include <asm/topology.h>
21 #include <asm/cpumask.h>
22 #include <asm/pgtable.h>
23 #include <asm/atomic.h>
24 #include <asm/proto.h>
25 #include <asm/setup.h>
26 #include <asm/apic.h>
27 #include <asm/desc.h>
28 #include <asm/i387.h>
29 #include <asm/mtrr.h>
30 #include <asm/numa.h>
31 #include <asm/asm.h>
32 #include <asm/cpu.h>
33 #include <asm/mce.h>
34 #include <asm/msr.h>
35 #include <asm/pat.h>
36 #include <asm/smp.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 const struct cpu_dev *this_cpu __cpuinitdata;
62 
63 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
64 #ifdef CONFIG_X86_64
65 	/*
66 	 * We need valid kernel segments for data and code in long mode too
67 	 * IRET will check the segment types  kkeil 2000/10/28
68 	 * Also sysret mandates a special GDT layout
69 	 *
70 	 * TLS descriptors are currently at a different place compared to i386.
71 	 * Hopefully nobody expects them at a fixed place (Wine?)
72 	 */
73 	[GDT_ENTRY_KERNEL32_CS]		= { { { 0x0000ffff, 0x00cf9b00 } } },
74 	[GDT_ENTRY_KERNEL_CS]		= { { { 0x0000ffff, 0x00af9b00 } } },
75 	[GDT_ENTRY_KERNEL_DS]		= { { { 0x0000ffff, 0x00cf9300 } } },
76 	[GDT_ENTRY_DEFAULT_USER32_CS]	= { { { 0x0000ffff, 0x00cffb00 } } },
77 	[GDT_ENTRY_DEFAULT_USER_DS]	= { { { 0x0000ffff, 0x00cff300 } } },
78 	[GDT_ENTRY_DEFAULT_USER_CS]	= { { { 0x0000ffff, 0x00affb00 } } },
79 #else
80 	[GDT_ENTRY_KERNEL_CS]		= { { { 0x0000ffff, 0x00cf9a00 } } },
81 	[GDT_ENTRY_KERNEL_DS]		= { { { 0x0000ffff, 0x00cf9200 } } },
82 	[GDT_ENTRY_DEFAULT_USER_CS]	= { { { 0x0000ffff, 0x00cffa00 } } },
83 	[GDT_ENTRY_DEFAULT_USER_DS]	= { { { 0x0000ffff, 0x00cff200 } } },
84 	/*
85 	 * Segments used for calling PnP BIOS have byte granularity.
86 	 * They code segments and data segments have fixed 64k limits,
87 	 * the transfer segment sizes are set at run time.
88 	 */
89 	/* 32-bit code */
90 	[GDT_ENTRY_PNPBIOS_CS32]	= { { { 0x0000ffff, 0x00409a00 } } },
91 	/* 16-bit code */
92 	[GDT_ENTRY_PNPBIOS_CS16]	= { { { 0x0000ffff, 0x00009a00 } } },
93 	/* 16-bit data */
94 	[GDT_ENTRY_PNPBIOS_DS]		= { { { 0x0000ffff, 0x00009200 } } },
95 	/* 16-bit data */
96 	[GDT_ENTRY_PNPBIOS_TS1]		= { { { 0x00000000, 0x00009200 } } },
97 	/* 16-bit data */
98 	[GDT_ENTRY_PNPBIOS_TS2]		= { { { 0x00000000, 0x00009200 } } },
99 	/*
100 	 * The APM segments have byte granularity and their bases
101 	 * are set at run time.  All have 64k limits.
102 	 */
103 	/* 32-bit code */
104 	[GDT_ENTRY_APMBIOS_BASE]	= { { { 0x0000ffff, 0x00409a00 } } },
105 	/* 16-bit code */
106 	[GDT_ENTRY_APMBIOS_BASE+1]	= { { { 0x0000ffff, 0x00009a00 } } },
107 	/* data */
108 	[GDT_ENTRY_APMBIOS_BASE+2]	= { { { 0x0000ffff, 0x00409200 } } },
109 
110 	[GDT_ENTRY_ESPFIX_SS]		= { { { 0x00000000, 0x00c09200 } } },
111 	[GDT_ENTRY_PERCPU]		= { { { 0x0000ffff, 0x00cf9200 } } },
112 	GDT_STACK_CANARY_INIT
113 #endif
114 } };
115 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
116 
117 #ifdef CONFIG_X86_32
118 static int cachesize_override __cpuinitdata = -1;
119 static int disable_x86_serial_nr __cpuinitdata = 1;
120 
121 static int __init cachesize_setup(char *str)
122 {
123 	get_option(&str, &cachesize_override);
124 	return 1;
125 }
126 __setup("cachesize=", cachesize_setup);
127 
128 static int __init x86_fxsr_setup(char *s)
129 {
130 	setup_clear_cpu_cap(X86_FEATURE_FXSR);
131 	setup_clear_cpu_cap(X86_FEATURE_XMM);
132 	return 1;
133 }
134 __setup("nofxsr", x86_fxsr_setup);
135 
136 static int __init x86_sep_setup(char *s)
137 {
138 	setup_clear_cpu_cap(X86_FEATURE_SEP);
139 	return 1;
140 }
141 __setup("nosep", x86_sep_setup);
142 
143 /* Standard macro to see if a specific flag is changeable */
144 static inline int flag_is_changeable_p(u32 flag)
145 {
146 	u32 f1, f2;
147 
148 	/*
149 	 * Cyrix and IDT cpus allow disabling of CPUID
150 	 * so the code below may return different results
151 	 * when it is executed before and after enabling
152 	 * the CPUID. Add "volatile" to not allow gcc to
153 	 * optimize the subsequent calls to this function.
154 	 */
155 	asm volatile ("pushfl		\n\t"
156 		      "pushfl		\n\t"
157 		      "popl %0		\n\t"
158 		      "movl %0, %1	\n\t"
159 		      "xorl %2, %0	\n\t"
160 		      "pushl %0		\n\t"
161 		      "popfl		\n\t"
162 		      "pushfl		\n\t"
163 		      "popl %0		\n\t"
164 		      "popfl		\n\t"
165 
166 		      : "=&r" (f1), "=&r" (f2)
167 		      : "ir" (flag));
168 
169 	return ((f1^f2) & flag) != 0;
170 }
171 
172 /* Probe for the CPUID instruction */
173 static int __cpuinit have_cpuid_p(void)
174 {
175 	return flag_is_changeable_p(X86_EFLAGS_ID);
176 }
177 
178 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
179 {
180 	unsigned long lo, hi;
181 
182 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
183 		return;
184 
185 	/* Disable processor serial number: */
186 
187 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
188 	lo |= 0x200000;
189 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
190 
191 	printk(KERN_NOTICE "CPU serial number disabled.\n");
192 	clear_cpu_cap(c, X86_FEATURE_PN);
193 
194 	/* Disabling the serial number may affect the cpuid level */
195 	c->cpuid_level = cpuid_eax(0);
196 }
197 
198 static int __init x86_serial_nr_setup(char *s)
199 {
200 	disable_x86_serial_nr = 0;
201 	return 1;
202 }
203 __setup("serialnumber", x86_serial_nr_setup);
204 #else
205 static inline int flag_is_changeable_p(u32 flag)
206 {
207 	return 1;
208 }
209 /* Probe for the CPUID instruction */
210 static inline int have_cpuid_p(void)
211 {
212 	return 1;
213 }
214 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
215 {
216 }
217 #endif
218 
219 /*
220  * Some CPU features depend on higher CPUID levels, which may not always
221  * be available due to CPUID level capping or broken virtualization
222  * software.  Add those features to this table to auto-disable them.
223  */
224 struct cpuid_dependent_feature {
225 	u32 feature;
226 	u32 level;
227 };
228 
229 static const struct cpuid_dependent_feature __cpuinitconst
230 cpuid_dependent_features[] = {
231 	{ X86_FEATURE_MWAIT,		0x00000005 },
232 	{ X86_FEATURE_DCA,		0x00000009 },
233 	{ X86_FEATURE_XSAVE,		0x0000000d },
234 	{ 0, 0 }
235 };
236 
237 static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
238 {
239 	const struct cpuid_dependent_feature *df;
240 
241 	for (df = cpuid_dependent_features; df->feature; df++) {
242 
243 		if (!cpu_has(c, df->feature))
244 			continue;
245 		/*
246 		 * Note: cpuid_level is set to -1 if unavailable, but
247 		 * extended_extended_level is set to 0 if unavailable
248 		 * and the legitimate extended levels are all negative
249 		 * when signed; hence the weird messing around with
250 		 * signs here...
251 		 */
252 		if (!((s32)df->level < 0 ?
253 		     (u32)df->level > (u32)c->extended_cpuid_level :
254 		     (s32)df->level > (s32)c->cpuid_level))
255 			continue;
256 
257 		clear_cpu_cap(c, df->feature);
258 		if (!warn)
259 			continue;
260 
261 		printk(KERN_WARNING
262 		       "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
263 				x86_cap_flags[df->feature], df->level);
264 	}
265 }
266 
267 /*
268  * Naming convention should be: <Name> [(<Codename>)]
269  * This table only is used unless init_<vendor>() below doesn't set it;
270  * in particular, if CPUID levels 0x80000002..4 are supported, this
271  * isn't used
272  */
273 
274 /* Look up CPU names by table lookup. */
275 static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
276 {
277 	const struct cpu_model_info *info;
278 
279 	if (c->x86_model >= 16)
280 		return NULL;	/* Range check */
281 
282 	if (!this_cpu)
283 		return NULL;
284 
285 	info = this_cpu->c_models;
286 
287 	while (info && info->family) {
288 		if (info->family == c->x86)
289 			return info->model_names[c->x86_model];
290 		info++;
291 	}
292 	return NULL;		/* Not found */
293 }
294 
295 __u32 cleared_cpu_caps[NCAPINTS] __cpuinitdata;
296 
297 void load_percpu_segment(int cpu)
298 {
299 #ifdef CONFIG_X86_32
300 	loadsegment(fs, __KERNEL_PERCPU);
301 #else
302 	loadsegment(gs, 0);
303 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
304 #endif
305 	load_stack_canary_segment();
306 }
307 
308 /*
309  * Current gdt points %fs at the "master" per-cpu area: after this,
310  * it's on the real one.
311  */
312 void switch_to_new_gdt(int cpu)
313 {
314 	struct desc_ptr gdt_descr;
315 
316 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
317 	gdt_descr.size = GDT_SIZE - 1;
318 	load_gdt(&gdt_descr);
319 	/* Reload the per-cpu base */
320 
321 	load_percpu_segment(cpu);
322 }
323 
324 static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
325 
326 static void __cpuinit default_init(struct cpuinfo_x86 *c)
327 {
328 #ifdef CONFIG_X86_64
329 	display_cacheinfo(c);
330 #else
331 	/* Not much we can do here... */
332 	/* Check if at least it has cpuid */
333 	if (c->cpuid_level == -1) {
334 		/* No cpuid. It must be an ancient CPU */
335 		if (c->x86 == 4)
336 			strcpy(c->x86_model_id, "486");
337 		else if (c->x86 == 3)
338 			strcpy(c->x86_model_id, "386");
339 	}
340 #endif
341 }
342 
343 static const struct cpu_dev __cpuinitconst default_cpu = {
344 	.c_init	= default_init,
345 	.c_vendor = "Unknown",
346 	.c_x86_vendor = X86_VENDOR_UNKNOWN,
347 };
348 
349 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
350 {
351 	unsigned int *v;
352 	char *p, *q;
353 
354 	if (c->extended_cpuid_level < 0x80000004)
355 		return;
356 
357 	v = (unsigned int *)c->x86_model_id;
358 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
359 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
360 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
361 	c->x86_model_id[48] = 0;
362 
363 	/*
364 	 * Intel chips right-justify this string for some dumb reason;
365 	 * undo that brain damage:
366 	 */
367 	p = q = &c->x86_model_id[0];
368 	while (*p == ' ')
369 		p++;
370 	if (p != q) {
371 		while (*p)
372 			*q++ = *p++;
373 		while (q <= &c->x86_model_id[48])
374 			*q++ = '\0';	/* Zero-pad the rest */
375 	}
376 }
377 
378 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
379 {
380 	unsigned int n, dummy, ebx, ecx, edx, l2size;
381 
382 	n = c->extended_cpuid_level;
383 
384 	if (n >= 0x80000005) {
385 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
386 		printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
387 				edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
388 		c->x86_cache_size = (ecx>>24) + (edx>>24);
389 #ifdef CONFIG_X86_64
390 		/* On K8 L1 TLB is inclusive, so don't count it */
391 		c->x86_tlbsize = 0;
392 #endif
393 	}
394 
395 	if (n < 0x80000006)	/* Some chips just has a large L1. */
396 		return;
397 
398 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
399 	l2size = ecx >> 16;
400 
401 #ifdef CONFIG_X86_64
402 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
403 #else
404 	/* do processor-specific cache resizing */
405 	if (this_cpu->c_size_cache)
406 		l2size = this_cpu->c_size_cache(c, l2size);
407 
408 	/* Allow user to override all this if necessary. */
409 	if (cachesize_override != -1)
410 		l2size = cachesize_override;
411 
412 	if (l2size == 0)
413 		return;		/* Again, no L2 cache is possible */
414 #endif
415 
416 	c->x86_cache_size = l2size;
417 
418 	printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
419 			l2size, ecx & 0xFF);
420 }
421 
422 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
423 {
424 #ifdef CONFIG_X86_HT
425 	u32 eax, ebx, ecx, edx;
426 	int index_msb, core_bits;
427 
428 	if (!cpu_has(c, X86_FEATURE_HT))
429 		return;
430 
431 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
432 		goto out;
433 
434 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
435 		return;
436 
437 	cpuid(1, &eax, &ebx, &ecx, &edx);
438 
439 	smp_num_siblings = (ebx & 0xff0000) >> 16;
440 
441 	if (smp_num_siblings == 1) {
442 		printk(KERN_INFO  "CPU: Hyper-Threading is disabled\n");
443 		goto out;
444 	}
445 
446 	if (smp_num_siblings <= 1)
447 		goto out;
448 
449 	if (smp_num_siblings > nr_cpu_ids) {
450 		pr_warning("CPU: Unsupported number of siblings %d",
451 			   smp_num_siblings);
452 		smp_num_siblings = 1;
453 		return;
454 	}
455 
456 	index_msb = get_count_order(smp_num_siblings);
457 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
458 
459 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
460 
461 	index_msb = get_count_order(smp_num_siblings);
462 
463 	core_bits = get_count_order(c->x86_max_cores);
464 
465 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
466 				       ((1 << core_bits) - 1);
467 
468 out:
469 	if ((c->x86_max_cores * smp_num_siblings) > 1) {
470 		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
471 		       c->phys_proc_id);
472 		printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
473 		       c->cpu_core_id);
474 	}
475 #endif
476 }
477 
478 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
479 {
480 	char *v = c->x86_vendor_id;
481 	static int printed;
482 	int i;
483 
484 	for (i = 0; i < X86_VENDOR_NUM; i++) {
485 		if (!cpu_devs[i])
486 			break;
487 
488 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
489 		    (cpu_devs[i]->c_ident[1] &&
490 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
491 
492 			this_cpu = cpu_devs[i];
493 			c->x86_vendor = this_cpu->c_x86_vendor;
494 			return;
495 		}
496 	}
497 
498 	if (!printed) {
499 		printed++;
500 		printk(KERN_ERR
501 		    "CPU: vendor_id '%s' unknown, using generic init.\n", v);
502 
503 		printk(KERN_ERR "CPU: Your system may be unstable.\n");
504 	}
505 
506 	c->x86_vendor = X86_VENDOR_UNKNOWN;
507 	this_cpu = &default_cpu;
508 }
509 
510 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
511 {
512 	/* Get vendor name */
513 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
514 	      (unsigned int *)&c->x86_vendor_id[0],
515 	      (unsigned int *)&c->x86_vendor_id[8],
516 	      (unsigned int *)&c->x86_vendor_id[4]);
517 
518 	c->x86 = 4;
519 	/* Intel-defined flags: level 0x00000001 */
520 	if (c->cpuid_level >= 0x00000001) {
521 		u32 junk, tfms, cap0, misc;
522 
523 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
524 		c->x86 = (tfms >> 8) & 0xf;
525 		c->x86_model = (tfms >> 4) & 0xf;
526 		c->x86_mask = tfms & 0xf;
527 
528 		if (c->x86 == 0xf)
529 			c->x86 += (tfms >> 20) & 0xff;
530 		if (c->x86 >= 0x6)
531 			c->x86_model += ((tfms >> 16) & 0xf) << 4;
532 
533 		if (cap0 & (1<<19)) {
534 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
535 			c->x86_cache_alignment = c->x86_clflush_size;
536 		}
537 	}
538 }
539 
540 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
541 {
542 	u32 tfms, xlvl;
543 	u32 ebx;
544 
545 	/* Intel-defined flags: level 0x00000001 */
546 	if (c->cpuid_level >= 0x00000001) {
547 		u32 capability, excap;
548 
549 		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
550 		c->x86_capability[0] = capability;
551 		c->x86_capability[4] = excap;
552 	}
553 
554 	/* AMD-defined flags: level 0x80000001 */
555 	xlvl = cpuid_eax(0x80000000);
556 	c->extended_cpuid_level = xlvl;
557 
558 	if ((xlvl & 0xffff0000) == 0x80000000) {
559 		if (xlvl >= 0x80000001) {
560 			c->x86_capability[1] = cpuid_edx(0x80000001);
561 			c->x86_capability[6] = cpuid_ecx(0x80000001);
562 		}
563 	}
564 
565 	if (c->extended_cpuid_level >= 0x80000008) {
566 		u32 eax = cpuid_eax(0x80000008);
567 
568 		c->x86_virt_bits = (eax >> 8) & 0xff;
569 		c->x86_phys_bits = eax & 0xff;
570 	}
571 #ifdef CONFIG_X86_32
572 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
573 		c->x86_phys_bits = 36;
574 #endif
575 
576 	if (c->extended_cpuid_level >= 0x80000007)
577 		c->x86_power = cpuid_edx(0x80000007);
578 
579 }
580 
581 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
582 {
583 #ifdef CONFIG_X86_32
584 	int i;
585 
586 	/*
587 	 * First of all, decide if this is a 486 or higher
588 	 * It's a 486 if we can modify the AC flag
589 	 */
590 	if (flag_is_changeable_p(X86_EFLAGS_AC))
591 		c->x86 = 4;
592 	else
593 		c->x86 = 3;
594 
595 	for (i = 0; i < X86_VENDOR_NUM; i++)
596 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
597 			c->x86_vendor_id[0] = 0;
598 			cpu_devs[i]->c_identify(c);
599 			if (c->x86_vendor_id[0]) {
600 				get_cpu_vendor(c);
601 				break;
602 			}
603 		}
604 #endif
605 }
606 
607 /*
608  * Do minimum CPU detection early.
609  * Fields really needed: vendor, cpuid_level, family, model, mask,
610  * cache alignment.
611  * The others are not touched to avoid unwanted side effects.
612  *
613  * WARNING: this function is only called on the BP.  Don't add code here
614  * that is supposed to run on all CPUs.
615  */
616 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
617 {
618 #ifdef CONFIG_X86_64
619 	c->x86_clflush_size = 64;
620 	c->x86_phys_bits = 36;
621 	c->x86_virt_bits = 48;
622 #else
623 	c->x86_clflush_size = 32;
624 	c->x86_phys_bits = 32;
625 	c->x86_virt_bits = 32;
626 #endif
627 	c->x86_cache_alignment = c->x86_clflush_size;
628 
629 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
630 	c->extended_cpuid_level = 0;
631 
632 	if (!have_cpuid_p())
633 		identify_cpu_without_cpuid(c);
634 
635 	/* cyrix could have cpuid enabled via c_identify()*/
636 	if (!have_cpuid_p())
637 		return;
638 
639 	cpu_detect(c);
640 
641 	get_cpu_vendor(c);
642 
643 	get_cpu_cap(c);
644 
645 	if (this_cpu->c_early_init)
646 		this_cpu->c_early_init(c);
647 
648 #ifdef CONFIG_SMP
649 	c->cpu_index = boot_cpu_id;
650 #endif
651 	filter_cpuid_features(c, false);
652 }
653 
654 void __init early_cpu_init(void)
655 {
656 	const struct cpu_dev *const *cdev;
657 	int count = 0;
658 
659 	printk(KERN_INFO "KERNEL supported cpus:\n");
660 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
661 		const struct cpu_dev *cpudev = *cdev;
662 		unsigned int j;
663 
664 		if (count >= X86_VENDOR_NUM)
665 			break;
666 		cpu_devs[count] = cpudev;
667 		count++;
668 
669 		for (j = 0; j < 2; j++) {
670 			if (!cpudev->c_ident[j])
671 				continue;
672 			printk(KERN_INFO "  %s %s\n", cpudev->c_vendor,
673 				cpudev->c_ident[j]);
674 		}
675 	}
676 
677 	early_identify_cpu(&boot_cpu_data);
678 }
679 
680 /*
681  * The NOPL instruction is supposed to exist on all CPUs with
682  * family >= 6; unfortunately, that's not true in practice because
683  * of early VIA chips and (more importantly) broken virtualizers that
684  * are not easy to detect.  In the latter case it doesn't even *fail*
685  * reliably, so probing for it doesn't even work.  Disable it completely
686  * unless we can find a reliable way to detect all the broken cases.
687  */
688 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
689 {
690 	clear_cpu_cap(c, X86_FEATURE_NOPL);
691 }
692 
693 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
694 {
695 	c->extended_cpuid_level = 0;
696 
697 	if (!have_cpuid_p())
698 		identify_cpu_without_cpuid(c);
699 
700 	/* cyrix could have cpuid enabled via c_identify()*/
701 	if (!have_cpuid_p())
702 		return;
703 
704 	cpu_detect(c);
705 
706 	get_cpu_vendor(c);
707 
708 	get_cpu_cap(c);
709 
710 	if (c->cpuid_level >= 0x00000001) {
711 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
712 #ifdef CONFIG_X86_32
713 # ifdef CONFIG_X86_HT
714 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
715 # else
716 		c->apicid = c->initial_apicid;
717 # endif
718 #endif
719 
720 #ifdef CONFIG_X86_HT
721 		c->phys_proc_id = c->initial_apicid;
722 #endif
723 	}
724 
725 	get_model_name(c); /* Default name */
726 
727 	init_scattered_cpuid_features(c);
728 	detect_nopl(c);
729 }
730 
731 /*
732  * This does the hard work of actually picking apart the CPU stuff...
733  */
734 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
735 {
736 	int i;
737 
738 	c->loops_per_jiffy = loops_per_jiffy;
739 	c->x86_cache_size = -1;
740 	c->x86_vendor = X86_VENDOR_UNKNOWN;
741 	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
742 	c->x86_vendor_id[0] = '\0'; /* Unset */
743 	c->x86_model_id[0] = '\0';  /* Unset */
744 	c->x86_max_cores = 1;
745 	c->x86_coreid_bits = 0;
746 #ifdef CONFIG_X86_64
747 	c->x86_clflush_size = 64;
748 	c->x86_phys_bits = 36;
749 	c->x86_virt_bits = 48;
750 #else
751 	c->cpuid_level = -1;	/* CPUID not detected */
752 	c->x86_clflush_size = 32;
753 	c->x86_phys_bits = 32;
754 	c->x86_virt_bits = 32;
755 #endif
756 	c->x86_cache_alignment = c->x86_clflush_size;
757 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
758 
759 	generic_identify(c);
760 
761 	if (this_cpu->c_identify)
762 		this_cpu->c_identify(c);
763 
764 #ifdef CONFIG_X86_64
765 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
766 #endif
767 
768 	/*
769 	 * Vendor-specific initialization.  In this section we
770 	 * canonicalize the feature flags, meaning if there are
771 	 * features a certain CPU supports which CPUID doesn't
772 	 * tell us, CPUID claiming incorrect flags, or other bugs,
773 	 * we handle them here.
774 	 *
775 	 * At the end of this section, c->x86_capability better
776 	 * indicate the features this CPU genuinely supports!
777 	 */
778 	if (this_cpu->c_init)
779 		this_cpu->c_init(c);
780 
781 	/* Disable the PN if appropriate */
782 	squash_the_stupid_serial_number(c);
783 
784 	/*
785 	 * The vendor-specific functions might have changed features.
786 	 * Now we do "generic changes."
787 	 */
788 
789 	/* Filter out anything that depends on CPUID levels we don't have */
790 	filter_cpuid_features(c, true);
791 
792 	/* If the model name is still unset, do table lookup. */
793 	if (!c->x86_model_id[0]) {
794 		const char *p;
795 		p = table_lookup_model(c);
796 		if (p)
797 			strcpy(c->x86_model_id, p);
798 		else
799 			/* Last resort... */
800 			sprintf(c->x86_model_id, "%02x/%02x",
801 				c->x86, c->x86_model);
802 	}
803 
804 #ifdef CONFIG_X86_64
805 	detect_ht(c);
806 #endif
807 
808 	init_hypervisor(c);
809 	/*
810 	 * On SMP, boot_cpu_data holds the common feature set between
811 	 * all CPUs; so make sure that we indicate which features are
812 	 * common between the CPUs.  The first time this routine gets
813 	 * executed, c == &boot_cpu_data.
814 	 */
815 	if (c != &boot_cpu_data) {
816 		/* AND the already accumulated flags with these */
817 		for (i = 0; i < NCAPINTS; i++)
818 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
819 	}
820 
821 	/* Clear all flags overriden by options */
822 	for (i = 0; i < NCAPINTS; i++)
823 		c->x86_capability[i] &= ~cleared_cpu_caps[i];
824 
825 #ifdef CONFIG_X86_MCE
826 	/* Init Machine Check Exception if available. */
827 	mcheck_init(c);
828 #endif
829 
830 	select_idle_routine(c);
831 
832 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
833 	numa_add_cpu(smp_processor_id());
834 #endif
835 }
836 
837 #ifdef CONFIG_X86_64
838 static void vgetcpu_set_mode(void)
839 {
840 	if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
841 		vgetcpu_mode = VGETCPU_RDTSCP;
842 	else
843 		vgetcpu_mode = VGETCPU_LSL;
844 }
845 #endif
846 
847 void __init identify_boot_cpu(void)
848 {
849 	identify_cpu(&boot_cpu_data);
850 	init_c1e_mask();
851 #ifdef CONFIG_X86_32
852 	sysenter_setup();
853 	enable_sep_cpu();
854 #else
855 	vgetcpu_set_mode();
856 #endif
857 }
858 
859 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
860 {
861 	BUG_ON(c == &boot_cpu_data);
862 	identify_cpu(c);
863 #ifdef CONFIG_X86_32
864 	enable_sep_cpu();
865 #endif
866 	mtrr_ap_init();
867 }
868 
869 struct msr_range {
870 	unsigned	min;
871 	unsigned	max;
872 };
873 
874 static const struct msr_range msr_range_array[] __cpuinitconst = {
875 	{ 0x00000000, 0x00000418},
876 	{ 0xc0000000, 0xc000040b},
877 	{ 0xc0010000, 0xc0010142},
878 	{ 0xc0011000, 0xc001103b},
879 };
880 
881 static void __cpuinit print_cpu_msr(void)
882 {
883 	unsigned index_min, index_max;
884 	unsigned index;
885 	u64 val;
886 	int i;
887 
888 	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
889 		index_min = msr_range_array[i].min;
890 		index_max = msr_range_array[i].max;
891 
892 		for (index = index_min; index < index_max; index++) {
893 			if (rdmsrl_amd_safe(index, &val))
894 				continue;
895 			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
896 		}
897 	}
898 }
899 
900 static int show_msr __cpuinitdata;
901 
902 static __init int setup_show_msr(char *arg)
903 {
904 	int num;
905 
906 	get_option(&arg, &num);
907 
908 	if (num > 0)
909 		show_msr = num;
910 	return 1;
911 }
912 __setup("show_msr=", setup_show_msr);
913 
914 static __init int setup_noclflush(char *arg)
915 {
916 	setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
917 	return 1;
918 }
919 __setup("noclflush", setup_noclflush);
920 
921 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
922 {
923 	const char *vendor = NULL;
924 
925 	if (c->x86_vendor < X86_VENDOR_NUM) {
926 		vendor = this_cpu->c_vendor;
927 	} else {
928 		if (c->cpuid_level >= 0)
929 			vendor = c->x86_vendor_id;
930 	}
931 
932 	if (vendor && !strstr(c->x86_model_id, vendor))
933 		printk(KERN_CONT "%s ", vendor);
934 
935 	if (c->x86_model_id[0])
936 		printk(KERN_CONT "%s", c->x86_model_id);
937 	else
938 		printk(KERN_CONT "%d86", c->x86);
939 
940 	if (c->x86_mask || c->cpuid_level >= 0)
941 		printk(KERN_CONT " stepping %02x\n", c->x86_mask);
942 	else
943 		printk(KERN_CONT "\n");
944 
945 #ifdef CONFIG_SMP
946 	if (c->cpu_index < show_msr)
947 		print_cpu_msr();
948 #else
949 	if (show_msr)
950 		print_cpu_msr();
951 #endif
952 }
953 
954 static __init int setup_disablecpuid(char *arg)
955 {
956 	int bit;
957 
958 	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
959 		setup_clear_cpu_cap(bit);
960 	else
961 		return 0;
962 
963 	return 1;
964 }
965 __setup("clearcpuid=", setup_disablecpuid);
966 
967 #ifdef CONFIG_X86_64
968 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
969 
970 DEFINE_PER_CPU_FIRST(union irq_stack_union,
971 		     irq_stack_union) __aligned(PAGE_SIZE);
972 
973 DEFINE_PER_CPU(char *, irq_stack_ptr) =
974 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
975 
976 DEFINE_PER_CPU(unsigned long, kernel_stack) =
977 	(unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
978 EXPORT_PER_CPU_SYMBOL(kernel_stack);
979 
980 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
981 
982 /*
983  * Special IST stacks which the CPU switches to when it calls
984  * an IST-marked descriptor entry. Up to 7 stacks (hardware
985  * limit), all of them are 4K, except the debug stack which
986  * is 8K.
987  */
988 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
989 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
990 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
991 };
992 
993 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
994 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ])
995 	__aligned(PAGE_SIZE);
996 
997 /* May not be marked __init: used by software suspend */
998 void syscall_init(void)
999 {
1000 	/*
1001 	 * LSTAR and STAR live in a bit strange symbiosis.
1002 	 * They both write to the same internal register. STAR allows to
1003 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1004 	 */
1005 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1006 	wrmsrl(MSR_LSTAR, system_call);
1007 	wrmsrl(MSR_CSTAR, ignore_sysret);
1008 
1009 #ifdef CONFIG_IA32_EMULATION
1010 	syscall32_cpu_init();
1011 #endif
1012 
1013 	/* Flags to clear on syscall */
1014 	wrmsrl(MSR_SYSCALL_MASK,
1015 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
1016 }
1017 
1018 unsigned long kernel_eflags;
1019 
1020 /*
1021  * Copies of the original ist values from the tss are only accessed during
1022  * debugging, no special alignment required.
1023  */
1024 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1025 
1026 #else	/* CONFIG_X86_64 */
1027 
1028 #ifdef CONFIG_CC_STACKPROTECTOR
1029 DEFINE_PER_CPU(unsigned long, stack_canary);
1030 #endif
1031 
1032 /* Make sure %fs and %gs are initialized properly in idle threads */
1033 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
1034 {
1035 	memset(regs, 0, sizeof(struct pt_regs));
1036 	regs->fs = __KERNEL_PERCPU;
1037 	regs->gs = __KERNEL_STACK_CANARY;
1038 
1039 	return regs;
1040 }
1041 #endif	/* CONFIG_X86_64 */
1042 
1043 /*
1044  * Clear all 6 debug registers:
1045  */
1046 static void clear_all_debug_regs(void)
1047 {
1048 	int i;
1049 
1050 	for (i = 0; i < 8; i++) {
1051 		/* Ignore db4, db5 */
1052 		if ((i == 4) || (i == 5))
1053 			continue;
1054 
1055 		set_debugreg(0, i);
1056 	}
1057 }
1058 
1059 /*
1060  * cpu_init() initializes state that is per-CPU. Some data is already
1061  * initialized (naturally) in the bootstrap process, such as the GDT
1062  * and IDT. We reload them nevertheless, this function acts as a
1063  * 'CPU state barrier', nothing should get across.
1064  * A lot of state is already set up in PDA init for 64 bit
1065  */
1066 #ifdef CONFIG_X86_64
1067 
1068 void __cpuinit cpu_init(void)
1069 {
1070 	struct orig_ist *orig_ist;
1071 	struct task_struct *me;
1072 	struct tss_struct *t;
1073 	unsigned long v;
1074 	int cpu;
1075 	int i;
1076 
1077 	cpu = stack_smp_processor_id();
1078 	t = &per_cpu(init_tss, cpu);
1079 	orig_ist = &per_cpu(orig_ist, cpu);
1080 
1081 #ifdef CONFIG_NUMA
1082 	if (cpu != 0 && percpu_read(node_number) == 0 &&
1083 	    cpu_to_node(cpu) != NUMA_NO_NODE)
1084 		percpu_write(node_number, cpu_to_node(cpu));
1085 #endif
1086 
1087 	me = current;
1088 
1089 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1090 		panic("CPU#%d already initialized!\n", cpu);
1091 
1092 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1093 
1094 	clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1095 
1096 	/*
1097 	 * Initialize the per-CPU GDT with the boot GDT,
1098 	 * and set up the GDT descriptor:
1099 	 */
1100 
1101 	switch_to_new_gdt(cpu);
1102 	loadsegment(fs, 0);
1103 
1104 	load_idt((const struct desc_ptr *)&idt_descr);
1105 
1106 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1107 	syscall_init();
1108 
1109 	wrmsrl(MSR_FS_BASE, 0);
1110 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1111 	barrier();
1112 
1113 	check_efer();
1114 	if (cpu != 0)
1115 		enable_x2apic();
1116 
1117 	/*
1118 	 * set up and load the per-CPU TSS
1119 	 */
1120 	if (!orig_ist->ist[0]) {
1121 		char *estacks = per_cpu(exception_stacks, cpu);
1122 
1123 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1124 			estacks += exception_stack_sizes[v];
1125 			orig_ist->ist[v] = t->x86_tss.ist[v] =
1126 					(unsigned long)estacks;
1127 		}
1128 	}
1129 
1130 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1131 
1132 	/*
1133 	 * <= is required because the CPU will access up to
1134 	 * 8 bits beyond the end of the IO permission bitmap.
1135 	 */
1136 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1137 		t->io_bitmap[i] = ~0UL;
1138 
1139 	atomic_inc(&init_mm.mm_count);
1140 	me->active_mm = &init_mm;
1141 	BUG_ON(me->mm);
1142 	enter_lazy_tlb(&init_mm, me);
1143 
1144 	load_sp0(t, &current->thread);
1145 	set_tss_desc(cpu, t);
1146 	load_TR_desc();
1147 	load_LDT(&init_mm.context);
1148 
1149 #ifdef CONFIG_KGDB
1150 	/*
1151 	 * If the kgdb is connected no debug regs should be altered.  This
1152 	 * is only applicable when KGDB and a KGDB I/O module are built
1153 	 * into the kernel and you are using early debugging with
1154 	 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1155 	 */
1156 	if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1157 		arch_kgdb_ops.correct_hw_break();
1158 	else
1159 #endif
1160 		clear_all_debug_regs();
1161 
1162 	fpu_init();
1163 
1164 	raw_local_save_flags(kernel_eflags);
1165 
1166 	if (is_uv_system())
1167 		uv_cpu_init();
1168 }
1169 
1170 #else
1171 
1172 void __cpuinit cpu_init(void)
1173 {
1174 	int cpu = smp_processor_id();
1175 	struct task_struct *curr = current;
1176 	struct tss_struct *t = &per_cpu(init_tss, cpu);
1177 	struct thread_struct *thread = &curr->thread;
1178 
1179 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1180 		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1181 		for (;;)
1182 			local_irq_enable();
1183 	}
1184 
1185 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1186 
1187 	if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1188 		clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1189 
1190 	load_idt(&idt_descr);
1191 	switch_to_new_gdt(cpu);
1192 
1193 	/*
1194 	 * Set up and load the per-CPU TSS and LDT
1195 	 */
1196 	atomic_inc(&init_mm.mm_count);
1197 	curr->active_mm = &init_mm;
1198 	BUG_ON(curr->mm);
1199 	enter_lazy_tlb(&init_mm, curr);
1200 
1201 	load_sp0(t, thread);
1202 	set_tss_desc(cpu, t);
1203 	load_TR_desc();
1204 	load_LDT(&init_mm.context);
1205 
1206 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1207 
1208 #ifdef CONFIG_DOUBLEFAULT
1209 	/* Set up doublefault TSS pointer in the GDT */
1210 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1211 #endif
1212 
1213 	clear_all_debug_regs();
1214 
1215 	/*
1216 	 * Force FPU initialization:
1217 	 */
1218 	if (cpu_has_xsave)
1219 		current_thread_info()->status = TS_XSAVE;
1220 	else
1221 		current_thread_info()->status = 0;
1222 	clear_used_math();
1223 	mxcsr_feature_mask_init();
1224 
1225 	/*
1226 	 * Boot processor to setup the FP and extended state context info.
1227 	 */
1228 	if (smp_processor_id() == boot_cpu_id)
1229 		init_thread_xstate();
1230 
1231 	xsave_init();
1232 }
1233 #endif
1234