xref: /linux/arch/x86/kernel/cpu/common.c (revision fea966f7564205fcf5919af9bde031e753419c96)
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_counter.h>
17 #include <asm/mmu_context.h>
18 #include <asm/hypervisor.h>
19 #include <asm/processor.h>
20 #include <asm/sections.h>
21 #include <asm/topology.h>
22 #include <asm/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 <asm/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 #include <asm/smp.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 	display_cacheinfo(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]		= { { { 0x0000ffff, 0x00cf9b00 } } },
98 	[GDT_ENTRY_KERNEL_CS]		= { { { 0x0000ffff, 0x00af9b00 } } },
99 	[GDT_ENTRY_KERNEL_DS]		= { { { 0x0000ffff, 0x00cf9300 } } },
100 	[GDT_ENTRY_DEFAULT_USER32_CS]	= { { { 0x0000ffff, 0x00cffb00 } } },
101 	[GDT_ENTRY_DEFAULT_USER_DS]	= { { { 0x0000ffff, 0x00cff300 } } },
102 	[GDT_ENTRY_DEFAULT_USER_CS]	= { { { 0x0000ffff, 0x00affb00 } } },
103 #else
104 	[GDT_ENTRY_KERNEL_CS]		= { { { 0x0000ffff, 0x00cf9a00 } } },
105 	[GDT_ENTRY_KERNEL_DS]		= { { { 0x0000ffff, 0x00cf9200 } } },
106 	[GDT_ENTRY_DEFAULT_USER_CS]	= { { { 0x0000ffff, 0x00cffa00 } } },
107 	[GDT_ENTRY_DEFAULT_USER_DS]	= { { { 0x0000ffff, 0x00cff200 } } },
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]	= { { { 0x0000ffff, 0x00409a00 } } },
115 	/* 16-bit code */
116 	[GDT_ENTRY_PNPBIOS_CS16]	= { { { 0x0000ffff, 0x00009a00 } } },
117 	/* 16-bit data */
118 	[GDT_ENTRY_PNPBIOS_DS]		= { { { 0x0000ffff, 0x00009200 } } },
119 	/* 16-bit data */
120 	[GDT_ENTRY_PNPBIOS_TS1]		= { { { 0x00000000, 0x00009200 } } },
121 	/* 16-bit data */
122 	[GDT_ENTRY_PNPBIOS_TS2]		= { { { 0x00000000, 0x00009200 } } },
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]	= { { { 0x0000ffff, 0x00409a00 } } },
129 	/* 16-bit code */
130 	[GDT_ENTRY_APMBIOS_BASE+1]	= { { { 0x0000ffff, 0x00009a00 } } },
131 	/* data */
132 	[GDT_ENTRY_APMBIOS_BASE+2]	= { { { 0x0000ffff, 0x00409200 } } },
133 
134 	[GDT_ENTRY_ESPFIX_SS]		= { { { 0x0000ffff, 0x00cf9200 } } },
135 	[GDT_ENTRY_PERCPU]		= { { { 0x0000ffff, 0x00cf9200 } } },
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 	return 1;
145 }
146 __setup("noxsave", x86_xsave_setup);
147 
148 #ifdef CONFIG_X86_32
149 static int cachesize_override __cpuinitdata = -1;
150 static int disable_x86_serial_nr __cpuinitdata = 1;
151 
152 static int __init cachesize_setup(char *str)
153 {
154 	get_option(&str, &cachesize_override);
155 	return 1;
156 }
157 __setup("cachesize=", cachesize_setup);
158 
159 static int __init x86_fxsr_setup(char *s)
160 {
161 	setup_clear_cpu_cap(X86_FEATURE_FXSR);
162 	setup_clear_cpu_cap(X86_FEATURE_XMM);
163 	return 1;
164 }
165 __setup("nofxsr", x86_fxsr_setup);
166 
167 static int __init x86_sep_setup(char *s)
168 {
169 	setup_clear_cpu_cap(X86_FEATURE_SEP);
170 	return 1;
171 }
172 __setup("nosep", x86_sep_setup);
173 
174 /* Standard macro to see if a specific flag is changeable */
175 static inline int flag_is_changeable_p(u32 flag)
176 {
177 	u32 f1, f2;
178 
179 	/*
180 	 * Cyrix and IDT cpus allow disabling of CPUID
181 	 * so the code below may return different results
182 	 * when it is executed before and after enabling
183 	 * the CPUID. Add "volatile" to not allow gcc to
184 	 * optimize the subsequent calls to this function.
185 	 */
186 	asm volatile ("pushfl		\n\t"
187 		      "pushfl		\n\t"
188 		      "popl %0		\n\t"
189 		      "movl %0, %1	\n\t"
190 		      "xorl %2, %0	\n\t"
191 		      "pushl %0		\n\t"
192 		      "popfl		\n\t"
193 		      "pushfl		\n\t"
194 		      "popl %0		\n\t"
195 		      "popfl		\n\t"
196 
197 		      : "=&r" (f1), "=&r" (f2)
198 		      : "ir" (flag));
199 
200 	return ((f1^f2) & flag) != 0;
201 }
202 
203 /* Probe for the CPUID instruction */
204 static int __cpuinit have_cpuid_p(void)
205 {
206 	return flag_is_changeable_p(X86_EFLAGS_ID);
207 }
208 
209 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
210 {
211 	unsigned long lo, hi;
212 
213 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
214 		return;
215 
216 	/* Disable processor serial number: */
217 
218 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
219 	lo |= 0x200000;
220 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
221 
222 	printk(KERN_NOTICE "CPU serial number disabled.\n");
223 	clear_cpu_cap(c, X86_FEATURE_PN);
224 
225 	/* Disabling the serial number may affect the cpuid level */
226 	c->cpuid_level = cpuid_eax(0);
227 }
228 
229 static int __init x86_serial_nr_setup(char *s)
230 {
231 	disable_x86_serial_nr = 0;
232 	return 1;
233 }
234 __setup("serialnumber", x86_serial_nr_setup);
235 #else
236 static inline int flag_is_changeable_p(u32 flag)
237 {
238 	return 1;
239 }
240 /* Probe for the CPUID instruction */
241 static inline int have_cpuid_p(void)
242 {
243 	return 1;
244 }
245 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
246 {
247 }
248 #endif
249 
250 /*
251  * Some CPU features depend on higher CPUID levels, which may not always
252  * be available due to CPUID level capping or broken virtualization
253  * software.  Add those features to this table to auto-disable them.
254  */
255 struct cpuid_dependent_feature {
256 	u32 feature;
257 	u32 level;
258 };
259 
260 static const struct cpuid_dependent_feature __cpuinitconst
261 cpuid_dependent_features[] = {
262 	{ X86_FEATURE_MWAIT,		0x00000005 },
263 	{ X86_FEATURE_DCA,		0x00000009 },
264 	{ X86_FEATURE_XSAVE,		0x0000000d },
265 	{ 0, 0 }
266 };
267 
268 static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
269 {
270 	const struct cpuid_dependent_feature *df;
271 
272 	for (df = cpuid_dependent_features; df->feature; df++) {
273 
274 		if (!cpu_has(c, df->feature))
275 			continue;
276 		/*
277 		 * Note: cpuid_level is set to -1 if unavailable, but
278 		 * extended_extended_level is set to 0 if unavailable
279 		 * and the legitimate extended levels are all negative
280 		 * when signed; hence the weird messing around with
281 		 * signs here...
282 		 */
283 		if (!((s32)df->level < 0 ?
284 		     (u32)df->level > (u32)c->extended_cpuid_level :
285 		     (s32)df->level > (s32)c->cpuid_level))
286 			continue;
287 
288 		clear_cpu_cap(c, df->feature);
289 		if (!warn)
290 			continue;
291 
292 		printk(KERN_WARNING
293 		       "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
294 				x86_cap_flags[df->feature], df->level);
295 	}
296 }
297 
298 /*
299  * Naming convention should be: <Name> [(<Codename>)]
300  * This table only is used unless init_<vendor>() below doesn't set it;
301  * in particular, if CPUID levels 0x80000002..4 are supported, this
302  * isn't used
303  */
304 
305 /* Look up CPU names by table lookup. */
306 static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
307 {
308 	const struct cpu_model_info *info;
309 
310 	if (c->x86_model >= 16)
311 		return NULL;	/* Range check */
312 
313 	if (!this_cpu)
314 		return NULL;
315 
316 	info = this_cpu->c_models;
317 
318 	while (info && info->family) {
319 		if (info->family == c->x86)
320 			return info->model_names[c->x86_model];
321 		info++;
322 	}
323 	return NULL;		/* Not found */
324 }
325 
326 __u32 cpu_caps_cleared[NCAPINTS] __cpuinitdata;
327 __u32 cpu_caps_set[NCAPINTS] __cpuinitdata;
328 
329 void load_percpu_segment(int cpu)
330 {
331 #ifdef CONFIG_X86_32
332 	loadsegment(fs, __KERNEL_PERCPU);
333 #else
334 	loadsegment(gs, 0);
335 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
336 #endif
337 	load_stack_canary_segment();
338 }
339 
340 /*
341  * Current gdt points %fs at the "master" per-cpu area: after this,
342  * it's on the real one.
343  */
344 void switch_to_new_gdt(int cpu)
345 {
346 	struct desc_ptr gdt_descr;
347 
348 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
349 	gdt_descr.size = GDT_SIZE - 1;
350 	load_gdt(&gdt_descr);
351 	/* Reload the per-cpu base */
352 
353 	load_percpu_segment(cpu);
354 }
355 
356 static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
357 
358 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
359 {
360 	unsigned int *v;
361 	char *p, *q;
362 
363 	if (c->extended_cpuid_level < 0x80000004)
364 		return;
365 
366 	v = (unsigned int *)c->x86_model_id;
367 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
368 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
369 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
370 	c->x86_model_id[48] = 0;
371 
372 	/*
373 	 * Intel chips right-justify this string for some dumb reason;
374 	 * undo that brain damage:
375 	 */
376 	p = q = &c->x86_model_id[0];
377 	while (*p == ' ')
378 		p++;
379 	if (p != q) {
380 		while (*p)
381 			*q++ = *p++;
382 		while (q <= &c->x86_model_id[48])
383 			*q++ = '\0';	/* Zero-pad the rest */
384 	}
385 }
386 
387 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
388 {
389 	unsigned int n, dummy, ebx, ecx, edx, l2size;
390 
391 	n = c->extended_cpuid_level;
392 
393 	if (n >= 0x80000005) {
394 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
395 		printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
396 				edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
397 		c->x86_cache_size = (ecx>>24) + (edx>>24);
398 #ifdef CONFIG_X86_64
399 		/* On K8 L1 TLB is inclusive, so don't count it */
400 		c->x86_tlbsize = 0;
401 #endif
402 	}
403 
404 	if (n < 0x80000006)	/* Some chips just has a large L1. */
405 		return;
406 
407 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
408 	l2size = ecx >> 16;
409 
410 #ifdef CONFIG_X86_64
411 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
412 #else
413 	/* do processor-specific cache resizing */
414 	if (this_cpu->c_size_cache)
415 		l2size = this_cpu->c_size_cache(c, l2size);
416 
417 	/* Allow user to override all this if necessary. */
418 	if (cachesize_override != -1)
419 		l2size = cachesize_override;
420 
421 	if (l2size == 0)
422 		return;		/* Again, no L2 cache is possible */
423 #endif
424 
425 	c->x86_cache_size = l2size;
426 
427 	printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
428 			l2size, ecx & 0xFF);
429 }
430 
431 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
432 {
433 #ifdef CONFIG_X86_HT
434 	u32 eax, ebx, ecx, edx;
435 	int index_msb, core_bits;
436 
437 	if (!cpu_has(c, X86_FEATURE_HT))
438 		return;
439 
440 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
441 		goto out;
442 
443 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
444 		return;
445 
446 	cpuid(1, &eax, &ebx, &ecx, &edx);
447 
448 	smp_num_siblings = (ebx & 0xff0000) >> 16;
449 
450 	if (smp_num_siblings == 1) {
451 		printk(KERN_INFO  "CPU: Hyper-Threading is disabled\n");
452 		goto out;
453 	}
454 
455 	if (smp_num_siblings <= 1)
456 		goto out;
457 
458 	if (smp_num_siblings > nr_cpu_ids) {
459 		pr_warning("CPU: Unsupported number of siblings %d",
460 			   smp_num_siblings);
461 		smp_num_siblings = 1;
462 		return;
463 	}
464 
465 	index_msb = get_count_order(smp_num_siblings);
466 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
467 
468 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
469 
470 	index_msb = get_count_order(smp_num_siblings);
471 
472 	core_bits = get_count_order(c->x86_max_cores);
473 
474 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
475 				       ((1 << core_bits) - 1);
476 
477 out:
478 	if ((c->x86_max_cores * smp_num_siblings) > 1) {
479 		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
480 		       c->phys_proc_id);
481 		printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
482 		       c->cpu_core_id);
483 	}
484 #endif
485 }
486 
487 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
488 {
489 	char *v = c->x86_vendor_id;
490 	int i;
491 
492 	for (i = 0; i < X86_VENDOR_NUM; i++) {
493 		if (!cpu_devs[i])
494 			break;
495 
496 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
497 		    (cpu_devs[i]->c_ident[1] &&
498 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
499 
500 			this_cpu = cpu_devs[i];
501 			c->x86_vendor = this_cpu->c_x86_vendor;
502 			return;
503 		}
504 	}
505 
506 	printk_once(KERN_ERR
507 			"CPU: vendor_id '%s' unknown, using generic init.\n" \
508 			"CPU: Your system may be unstable.\n", v);
509 
510 	c->x86_vendor = X86_VENDOR_UNKNOWN;
511 	this_cpu = &default_cpu;
512 }
513 
514 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
515 {
516 	/* Get vendor name */
517 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
518 	      (unsigned int *)&c->x86_vendor_id[0],
519 	      (unsigned int *)&c->x86_vendor_id[8],
520 	      (unsigned int *)&c->x86_vendor_id[4]);
521 
522 	c->x86 = 4;
523 	/* Intel-defined flags: level 0x00000001 */
524 	if (c->cpuid_level >= 0x00000001) {
525 		u32 junk, tfms, cap0, misc;
526 
527 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
528 		c->x86 = (tfms >> 8) & 0xf;
529 		c->x86_model = (tfms >> 4) & 0xf;
530 		c->x86_mask = tfms & 0xf;
531 
532 		if (c->x86 == 0xf)
533 			c->x86 += (tfms >> 20) & 0xff;
534 		if (c->x86 >= 0x6)
535 			c->x86_model += ((tfms >> 16) & 0xf) << 4;
536 
537 		if (cap0 & (1<<19)) {
538 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
539 			c->x86_cache_alignment = c->x86_clflush_size;
540 		}
541 	}
542 }
543 
544 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
545 {
546 	u32 tfms, xlvl;
547 	u32 ebx;
548 
549 	/* Intel-defined flags: level 0x00000001 */
550 	if (c->cpuid_level >= 0x00000001) {
551 		u32 capability, excap;
552 
553 		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
554 		c->x86_capability[0] = capability;
555 		c->x86_capability[4] = excap;
556 	}
557 
558 	/* AMD-defined flags: level 0x80000001 */
559 	xlvl = cpuid_eax(0x80000000);
560 	c->extended_cpuid_level = xlvl;
561 
562 	if ((xlvl & 0xffff0000) == 0x80000000) {
563 		if (xlvl >= 0x80000001) {
564 			c->x86_capability[1] = cpuid_edx(0x80000001);
565 			c->x86_capability[6] = cpuid_ecx(0x80000001);
566 		}
567 	}
568 
569 	if (c->extended_cpuid_level >= 0x80000008) {
570 		u32 eax = cpuid_eax(0x80000008);
571 
572 		c->x86_virt_bits = (eax >> 8) & 0xff;
573 		c->x86_phys_bits = eax & 0xff;
574 	}
575 #ifdef CONFIG_X86_32
576 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
577 		c->x86_phys_bits = 36;
578 #endif
579 
580 	if (c->extended_cpuid_level >= 0x80000007)
581 		c->x86_power = cpuid_edx(0x80000007);
582 
583 }
584 
585 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
586 {
587 #ifdef CONFIG_X86_32
588 	int i;
589 
590 	/*
591 	 * First of all, decide if this is a 486 or higher
592 	 * It's a 486 if we can modify the AC flag
593 	 */
594 	if (flag_is_changeable_p(X86_EFLAGS_AC))
595 		c->x86 = 4;
596 	else
597 		c->x86 = 3;
598 
599 	for (i = 0; i < X86_VENDOR_NUM; i++)
600 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
601 			c->x86_vendor_id[0] = 0;
602 			cpu_devs[i]->c_identify(c);
603 			if (c->x86_vendor_id[0]) {
604 				get_cpu_vendor(c);
605 				break;
606 			}
607 		}
608 #endif
609 }
610 
611 /*
612  * Do minimum CPU detection early.
613  * Fields really needed: vendor, cpuid_level, family, model, mask,
614  * cache alignment.
615  * The others are not touched to avoid unwanted side effects.
616  *
617  * WARNING: this function is only called on the BP.  Don't add code here
618  * that is supposed to run on all CPUs.
619  */
620 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
621 {
622 #ifdef CONFIG_X86_64
623 	c->x86_clflush_size = 64;
624 	c->x86_phys_bits = 36;
625 	c->x86_virt_bits = 48;
626 #else
627 	c->x86_clflush_size = 32;
628 	c->x86_phys_bits = 32;
629 	c->x86_virt_bits = 32;
630 #endif
631 	c->x86_cache_alignment = c->x86_clflush_size;
632 
633 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
634 	c->extended_cpuid_level = 0;
635 
636 	if (!have_cpuid_p())
637 		identify_cpu_without_cpuid(c);
638 
639 	/* cyrix could have cpuid enabled via c_identify()*/
640 	if (!have_cpuid_p())
641 		return;
642 
643 	cpu_detect(c);
644 
645 	get_cpu_vendor(c);
646 
647 	get_cpu_cap(c);
648 
649 	if (this_cpu->c_early_init)
650 		this_cpu->c_early_init(c);
651 
652 #ifdef CONFIG_SMP
653 	c->cpu_index = boot_cpu_id;
654 #endif
655 	filter_cpuid_features(c, false);
656 }
657 
658 void __init early_cpu_init(void)
659 {
660 	const struct cpu_dev *const *cdev;
661 	int count = 0;
662 
663 	printk(KERN_INFO "KERNEL supported cpus:\n");
664 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
665 		const struct cpu_dev *cpudev = *cdev;
666 		unsigned int j;
667 
668 		if (count >= X86_VENDOR_NUM)
669 			break;
670 		cpu_devs[count] = cpudev;
671 		count++;
672 
673 		for (j = 0; j < 2; j++) {
674 			if (!cpudev->c_ident[j])
675 				continue;
676 			printk(KERN_INFO "  %s %s\n", cpudev->c_vendor,
677 				cpudev->c_ident[j]);
678 		}
679 	}
680 
681 	early_identify_cpu(&boot_cpu_data);
682 }
683 
684 /*
685  * The NOPL instruction is supposed to exist on all CPUs with
686  * family >= 6; unfortunately, that's not true in practice because
687  * of early VIA chips and (more importantly) broken virtualizers that
688  * are not easy to detect.  In the latter case it doesn't even *fail*
689  * reliably, so probing for it doesn't even work.  Disable it completely
690  * unless we can find a reliable way to detect all the broken cases.
691  */
692 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
693 {
694 	clear_cpu_cap(c, X86_FEATURE_NOPL);
695 }
696 
697 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
698 {
699 	c->extended_cpuid_level = 0;
700 
701 	if (!have_cpuid_p())
702 		identify_cpu_without_cpuid(c);
703 
704 	/* cyrix could have cpuid enabled via c_identify()*/
705 	if (!have_cpuid_p())
706 		return;
707 
708 	cpu_detect(c);
709 
710 	get_cpu_vendor(c);
711 
712 	get_cpu_cap(c);
713 
714 	if (c->cpuid_level >= 0x00000001) {
715 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
716 #ifdef CONFIG_X86_32
717 # ifdef CONFIG_X86_HT
718 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
719 # else
720 		c->apicid = c->initial_apicid;
721 # endif
722 #endif
723 
724 #ifdef CONFIG_X86_HT
725 		c->phys_proc_id = c->initial_apicid;
726 #endif
727 	}
728 
729 	get_model_name(c); /* Default name */
730 
731 	init_scattered_cpuid_features(c);
732 	detect_nopl(c);
733 }
734 
735 /*
736  * This does the hard work of actually picking apart the CPU stuff...
737  */
738 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
739 {
740 	int i;
741 
742 	c->loops_per_jiffy = loops_per_jiffy;
743 	c->x86_cache_size = -1;
744 	c->x86_vendor = X86_VENDOR_UNKNOWN;
745 	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
746 	c->x86_vendor_id[0] = '\0'; /* Unset */
747 	c->x86_model_id[0] = '\0';  /* Unset */
748 	c->x86_max_cores = 1;
749 	c->x86_coreid_bits = 0;
750 #ifdef CONFIG_X86_64
751 	c->x86_clflush_size = 64;
752 	c->x86_phys_bits = 36;
753 	c->x86_virt_bits = 48;
754 #else
755 	c->cpuid_level = -1;	/* CPUID not detected */
756 	c->x86_clflush_size = 32;
757 	c->x86_phys_bits = 32;
758 	c->x86_virt_bits = 32;
759 #endif
760 	c->x86_cache_alignment = c->x86_clflush_size;
761 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
762 
763 	generic_identify(c);
764 
765 	if (this_cpu->c_identify)
766 		this_cpu->c_identify(c);
767 
768 	/* Clear/Set all flags overriden by options, after probe */
769 	for (i = 0; i < NCAPINTS; i++) {
770 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
771 		c->x86_capability[i] |= cpu_caps_set[i];
772 	}
773 
774 #ifdef CONFIG_X86_64
775 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
776 #endif
777 
778 	/*
779 	 * Vendor-specific initialization.  In this section we
780 	 * canonicalize the feature flags, meaning if there are
781 	 * features a certain CPU supports which CPUID doesn't
782 	 * tell us, CPUID claiming incorrect flags, or other bugs,
783 	 * we handle them here.
784 	 *
785 	 * At the end of this section, c->x86_capability better
786 	 * indicate the features this CPU genuinely supports!
787 	 */
788 	if (this_cpu->c_init)
789 		this_cpu->c_init(c);
790 
791 	/* Disable the PN if appropriate */
792 	squash_the_stupid_serial_number(c);
793 
794 	/*
795 	 * The vendor-specific functions might have changed features.
796 	 * Now we do "generic changes."
797 	 */
798 
799 	/* Filter out anything that depends on CPUID levels we don't have */
800 	filter_cpuid_features(c, true);
801 
802 	/* If the model name is still unset, do table lookup. */
803 	if (!c->x86_model_id[0]) {
804 		const char *p;
805 		p = table_lookup_model(c);
806 		if (p)
807 			strcpy(c->x86_model_id, p);
808 		else
809 			/* Last resort... */
810 			sprintf(c->x86_model_id, "%02x/%02x",
811 				c->x86, c->x86_model);
812 	}
813 
814 #ifdef CONFIG_X86_64
815 	detect_ht(c);
816 #endif
817 
818 	init_hypervisor(c);
819 
820 	/*
821 	 * Clear/Set all flags overriden by options, need do it
822 	 * before following smp all cpus cap AND.
823 	 */
824 	for (i = 0; i < NCAPINTS; i++) {
825 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
826 		c->x86_capability[i] |= cpu_caps_set[i];
827 	}
828 
829 	/*
830 	 * On SMP, boot_cpu_data holds the common feature set between
831 	 * all CPUs; so make sure that we indicate which features are
832 	 * common between the CPUs.  The first time this routine gets
833 	 * executed, c == &boot_cpu_data.
834 	 */
835 	if (c != &boot_cpu_data) {
836 		/* AND the already accumulated flags with these */
837 		for (i = 0; i < NCAPINTS; i++)
838 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
839 	}
840 
841 #ifdef CONFIG_X86_MCE
842 	/* Init Machine Check Exception if available. */
843 	mcheck_init(c);
844 #endif
845 
846 	select_idle_routine(c);
847 
848 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
849 	numa_add_cpu(smp_processor_id());
850 #endif
851 }
852 
853 #ifdef CONFIG_X86_64
854 static void vgetcpu_set_mode(void)
855 {
856 	if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
857 		vgetcpu_mode = VGETCPU_RDTSCP;
858 	else
859 		vgetcpu_mode = VGETCPU_LSL;
860 }
861 #endif
862 
863 void __init identify_boot_cpu(void)
864 {
865 	identify_cpu(&boot_cpu_data);
866 	init_c1e_mask();
867 #ifdef CONFIG_X86_32
868 	sysenter_setup();
869 	enable_sep_cpu();
870 #else
871 	vgetcpu_set_mode();
872 #endif
873 	init_hw_perf_counters();
874 }
875 
876 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
877 {
878 	BUG_ON(c == &boot_cpu_data);
879 	identify_cpu(c);
880 #ifdef CONFIG_X86_32
881 	enable_sep_cpu();
882 #endif
883 	mtrr_ap_init();
884 }
885 
886 struct msr_range {
887 	unsigned	min;
888 	unsigned	max;
889 };
890 
891 static const struct msr_range msr_range_array[] __cpuinitconst = {
892 	{ 0x00000000, 0x00000418},
893 	{ 0xc0000000, 0xc000040b},
894 	{ 0xc0010000, 0xc0010142},
895 	{ 0xc0011000, 0xc001103b},
896 };
897 
898 static void __cpuinit print_cpu_msr(void)
899 {
900 	unsigned index_min, index_max;
901 	unsigned index;
902 	u64 val;
903 	int i;
904 
905 	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
906 		index_min = msr_range_array[i].min;
907 		index_max = msr_range_array[i].max;
908 
909 		for (index = index_min; index < index_max; index++) {
910 			if (rdmsrl_amd_safe(index, &val))
911 				continue;
912 			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
913 		}
914 	}
915 }
916 
917 static int show_msr __cpuinitdata;
918 
919 static __init int setup_show_msr(char *arg)
920 {
921 	int num;
922 
923 	get_option(&arg, &num);
924 
925 	if (num > 0)
926 		show_msr = num;
927 	return 1;
928 }
929 __setup("show_msr=", setup_show_msr);
930 
931 static __init int setup_noclflush(char *arg)
932 {
933 	setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
934 	return 1;
935 }
936 __setup("noclflush", setup_noclflush);
937 
938 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
939 {
940 	const char *vendor = NULL;
941 
942 	if (c->x86_vendor < X86_VENDOR_NUM) {
943 		vendor = this_cpu->c_vendor;
944 	} else {
945 		if (c->cpuid_level >= 0)
946 			vendor = c->x86_vendor_id;
947 	}
948 
949 	if (vendor && !strstr(c->x86_model_id, vendor))
950 		printk(KERN_CONT "%s ", vendor);
951 
952 	if (c->x86_model_id[0])
953 		printk(KERN_CONT "%s", c->x86_model_id);
954 	else
955 		printk(KERN_CONT "%d86", c->x86);
956 
957 	if (c->x86_mask || c->cpuid_level >= 0)
958 		printk(KERN_CONT " stepping %02x\n", c->x86_mask);
959 	else
960 		printk(KERN_CONT "\n");
961 
962 #ifdef CONFIG_SMP
963 	if (c->cpu_index < show_msr)
964 		print_cpu_msr();
965 #else
966 	if (show_msr)
967 		print_cpu_msr();
968 #endif
969 }
970 
971 static __init int setup_disablecpuid(char *arg)
972 {
973 	int bit;
974 
975 	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
976 		setup_clear_cpu_cap(bit);
977 	else
978 		return 0;
979 
980 	return 1;
981 }
982 __setup("clearcpuid=", setup_disablecpuid);
983 
984 #ifdef CONFIG_X86_64
985 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
986 
987 DEFINE_PER_CPU_FIRST(union irq_stack_union,
988 		     irq_stack_union) __aligned(PAGE_SIZE);
989 
990 DEFINE_PER_CPU(char *, irq_stack_ptr) =
991 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
992 
993 DEFINE_PER_CPU(unsigned long, kernel_stack) =
994 	(unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
995 EXPORT_PER_CPU_SYMBOL(kernel_stack);
996 
997 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
998 
999 /*
1000  * Special IST stacks which the CPU switches to when it calls
1001  * an IST-marked descriptor entry. Up to 7 stacks (hardware
1002  * limit), all of them are 4K, except the debug stack which
1003  * is 8K.
1004  */
1005 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1006 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1007 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1008 };
1009 
1010 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1011 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ])
1012 	__aligned(PAGE_SIZE);
1013 
1014 /* May not be marked __init: used by software suspend */
1015 void syscall_init(void)
1016 {
1017 	/*
1018 	 * LSTAR and STAR live in a bit strange symbiosis.
1019 	 * They both write to the same internal register. STAR allows to
1020 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1021 	 */
1022 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1023 	wrmsrl(MSR_LSTAR, system_call);
1024 	wrmsrl(MSR_CSTAR, ignore_sysret);
1025 
1026 #ifdef CONFIG_IA32_EMULATION
1027 	syscall32_cpu_init();
1028 #endif
1029 
1030 	/* Flags to clear on syscall */
1031 	wrmsrl(MSR_SYSCALL_MASK,
1032 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
1033 }
1034 
1035 unsigned long kernel_eflags;
1036 
1037 /*
1038  * Copies of the original ist values from the tss are only accessed during
1039  * debugging, no special alignment required.
1040  */
1041 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1042 
1043 #else	/* CONFIG_X86_64 */
1044 
1045 #ifdef CONFIG_CC_STACKPROTECTOR
1046 DEFINE_PER_CPU(unsigned long, stack_canary);
1047 #endif
1048 
1049 /* Make sure %fs and %gs are initialized properly in idle threads */
1050 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
1051 {
1052 	memset(regs, 0, sizeof(struct pt_regs));
1053 	regs->fs = __KERNEL_PERCPU;
1054 	regs->gs = __KERNEL_STACK_CANARY;
1055 
1056 	return regs;
1057 }
1058 #endif	/* CONFIG_X86_64 */
1059 
1060 /*
1061  * Clear all 6 debug registers:
1062  */
1063 static void clear_all_debug_regs(void)
1064 {
1065 	int i;
1066 
1067 	for (i = 0; i < 8; i++) {
1068 		/* Ignore db4, db5 */
1069 		if ((i == 4) || (i == 5))
1070 			continue;
1071 
1072 		set_debugreg(0, i);
1073 	}
1074 }
1075 
1076 /*
1077  * cpu_init() initializes state that is per-CPU. Some data is already
1078  * initialized (naturally) in the bootstrap process, such as the GDT
1079  * and IDT. We reload them nevertheless, this function acts as a
1080  * 'CPU state barrier', nothing should get across.
1081  * A lot of state is already set up in PDA init for 64 bit
1082  */
1083 #ifdef CONFIG_X86_64
1084 
1085 void __cpuinit cpu_init(void)
1086 {
1087 	struct orig_ist *orig_ist;
1088 	struct task_struct *me;
1089 	struct tss_struct *t;
1090 	unsigned long v;
1091 	int cpu;
1092 	int i;
1093 
1094 	cpu = stack_smp_processor_id();
1095 	t = &per_cpu(init_tss, cpu);
1096 	orig_ist = &per_cpu(orig_ist, cpu);
1097 
1098 #ifdef CONFIG_NUMA
1099 	if (cpu != 0 && percpu_read(node_number) == 0 &&
1100 	    cpu_to_node(cpu) != NUMA_NO_NODE)
1101 		percpu_write(node_number, cpu_to_node(cpu));
1102 #endif
1103 
1104 	me = current;
1105 
1106 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1107 		panic("CPU#%d already initialized!\n", cpu);
1108 
1109 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1110 
1111 	clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1112 
1113 	/*
1114 	 * Initialize the per-CPU GDT with the boot GDT,
1115 	 * and set up the GDT descriptor:
1116 	 */
1117 
1118 	switch_to_new_gdt(cpu);
1119 	loadsegment(fs, 0);
1120 
1121 	load_idt((const struct desc_ptr *)&idt_descr);
1122 
1123 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1124 	syscall_init();
1125 
1126 	wrmsrl(MSR_FS_BASE, 0);
1127 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1128 	barrier();
1129 
1130 	check_efer();
1131 	if (cpu != 0)
1132 		enable_x2apic();
1133 
1134 	/*
1135 	 * set up and load the per-CPU TSS
1136 	 */
1137 	if (!orig_ist->ist[0]) {
1138 		char *estacks = per_cpu(exception_stacks, cpu);
1139 
1140 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1141 			estacks += exception_stack_sizes[v];
1142 			orig_ist->ist[v] = t->x86_tss.ist[v] =
1143 					(unsigned long)estacks;
1144 		}
1145 	}
1146 
1147 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1148 
1149 	/*
1150 	 * <= is required because the CPU will access up to
1151 	 * 8 bits beyond the end of the IO permission bitmap.
1152 	 */
1153 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1154 		t->io_bitmap[i] = ~0UL;
1155 
1156 	atomic_inc(&init_mm.mm_count);
1157 	me->active_mm = &init_mm;
1158 	BUG_ON(me->mm);
1159 	enter_lazy_tlb(&init_mm, me);
1160 
1161 	load_sp0(t, &current->thread);
1162 	set_tss_desc(cpu, t);
1163 	load_TR_desc();
1164 	load_LDT(&init_mm.context);
1165 
1166 #ifdef CONFIG_KGDB
1167 	/*
1168 	 * If the kgdb is connected no debug regs should be altered.  This
1169 	 * is only applicable when KGDB and a KGDB I/O module are built
1170 	 * into the kernel and you are using early debugging with
1171 	 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1172 	 */
1173 	if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1174 		arch_kgdb_ops.correct_hw_break();
1175 	else
1176 #endif
1177 		clear_all_debug_regs();
1178 
1179 	fpu_init();
1180 
1181 	raw_local_save_flags(kernel_eflags);
1182 
1183 	if (is_uv_system())
1184 		uv_cpu_init();
1185 }
1186 
1187 #else
1188 
1189 void __cpuinit cpu_init(void)
1190 {
1191 	int cpu = smp_processor_id();
1192 	struct task_struct *curr = current;
1193 	struct tss_struct *t = &per_cpu(init_tss, cpu);
1194 	struct thread_struct *thread = &curr->thread;
1195 
1196 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1197 		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1198 		for (;;)
1199 			local_irq_enable();
1200 	}
1201 
1202 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1203 
1204 	if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1205 		clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1206 
1207 	load_idt(&idt_descr);
1208 	switch_to_new_gdt(cpu);
1209 
1210 	/*
1211 	 * Set up and load the per-CPU TSS and LDT
1212 	 */
1213 	atomic_inc(&init_mm.mm_count);
1214 	curr->active_mm = &init_mm;
1215 	BUG_ON(curr->mm);
1216 	enter_lazy_tlb(&init_mm, curr);
1217 
1218 	load_sp0(t, thread);
1219 	set_tss_desc(cpu, t);
1220 	load_TR_desc();
1221 	load_LDT(&init_mm.context);
1222 
1223 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1224 
1225 #ifdef CONFIG_DOUBLEFAULT
1226 	/* Set up doublefault TSS pointer in the GDT */
1227 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1228 #endif
1229 
1230 	clear_all_debug_regs();
1231 
1232 	/*
1233 	 * Force FPU initialization:
1234 	 */
1235 	if (cpu_has_xsave)
1236 		current_thread_info()->status = TS_XSAVE;
1237 	else
1238 		current_thread_info()->status = 0;
1239 	clear_used_math();
1240 	mxcsr_feature_mask_init();
1241 
1242 	/*
1243 	 * Boot processor to setup the FP and extended state context info.
1244 	 */
1245 	if (smp_processor_id() == boot_cpu_id)
1246 		init_thread_xstate();
1247 
1248 	xsave_init();
1249 }
1250 #endif
1251