xref: /linux/arch/x86/kernel/cpu/common.c (revision e27ecdd94d81e5bc3d1f68591701db5adb342f0d)
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 const struct cpu_dev *this_cpu __cpuinitdata;
63 
64 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
65 #ifdef CONFIG_X86_64
66 	/*
67 	 * We need valid kernel segments for data and code in long mode too
68 	 * IRET will check the segment types  kkeil 2000/10/28
69 	 * Also sysret mandates a special GDT layout
70 	 *
71 	 * TLS descriptors are currently at a different place compared to i386.
72 	 * Hopefully nobody expects them at a fixed place (Wine?)
73 	 */
74 	[GDT_ENTRY_KERNEL32_CS]		= { { { 0x0000ffff, 0x00cf9b00 } } },
75 	[GDT_ENTRY_KERNEL_CS]		= { { { 0x0000ffff, 0x00af9b00 } } },
76 	[GDT_ENTRY_KERNEL_DS]		= { { { 0x0000ffff, 0x00cf9300 } } },
77 	[GDT_ENTRY_DEFAULT_USER32_CS]	= { { { 0x0000ffff, 0x00cffb00 } } },
78 	[GDT_ENTRY_DEFAULT_USER_DS]	= { { { 0x0000ffff, 0x00cff300 } } },
79 	[GDT_ENTRY_DEFAULT_USER_CS]	= { { { 0x0000ffff, 0x00affb00 } } },
80 #else
81 	[GDT_ENTRY_KERNEL_CS]		= { { { 0x0000ffff, 0x00cf9a00 } } },
82 	[GDT_ENTRY_KERNEL_DS]		= { { { 0x0000ffff, 0x00cf9200 } } },
83 	[GDT_ENTRY_DEFAULT_USER_CS]	= { { { 0x0000ffff, 0x00cffa00 } } },
84 	[GDT_ENTRY_DEFAULT_USER_DS]	= { { { 0x0000ffff, 0x00cff200 } } },
85 	/*
86 	 * Segments used for calling PnP BIOS have byte granularity.
87 	 * They code segments and data segments have fixed 64k limits,
88 	 * the transfer segment sizes are set at run time.
89 	 */
90 	/* 32-bit code */
91 	[GDT_ENTRY_PNPBIOS_CS32]	= { { { 0x0000ffff, 0x00409a00 } } },
92 	/* 16-bit code */
93 	[GDT_ENTRY_PNPBIOS_CS16]	= { { { 0x0000ffff, 0x00009a00 } } },
94 	/* 16-bit data */
95 	[GDT_ENTRY_PNPBIOS_DS]		= { { { 0x0000ffff, 0x00009200 } } },
96 	/* 16-bit data */
97 	[GDT_ENTRY_PNPBIOS_TS1]		= { { { 0x00000000, 0x00009200 } } },
98 	/* 16-bit data */
99 	[GDT_ENTRY_PNPBIOS_TS2]		= { { { 0x00000000, 0x00009200 } } },
100 	/*
101 	 * The APM segments have byte granularity and their bases
102 	 * are set at run time.  All have 64k limits.
103 	 */
104 	/* 32-bit code */
105 	[GDT_ENTRY_APMBIOS_BASE]	= { { { 0x0000ffff, 0x00409a00 } } },
106 	/* 16-bit code */
107 	[GDT_ENTRY_APMBIOS_BASE+1]	= { { { 0x0000ffff, 0x00009a00 } } },
108 	/* data */
109 	[GDT_ENTRY_APMBIOS_BASE+2]	= { { { 0x0000ffff, 0x00409200 } } },
110 
111 	[GDT_ENTRY_ESPFIX_SS]		= { { { 0x00000000, 0x00c09200 } } },
112 	[GDT_ENTRY_PERCPU]		= { { { 0x0000ffff, 0x00cf9200 } } },
113 	GDT_STACK_CANARY_INIT
114 #endif
115 } };
116 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
117 
118 static int __init x86_xsave_setup(char *s)
119 {
120 	setup_clear_cpu_cap(X86_FEATURE_XSAVE);
121 	return 1;
122 }
123 __setup("noxsave", x86_xsave_setup);
124 
125 #ifdef CONFIG_X86_32
126 static int cachesize_override __cpuinitdata = -1;
127 static int disable_x86_serial_nr __cpuinitdata = 1;
128 
129 static int __init cachesize_setup(char *str)
130 {
131 	get_option(&str, &cachesize_override);
132 	return 1;
133 }
134 __setup("cachesize=", cachesize_setup);
135 
136 static int __init x86_fxsr_setup(char *s)
137 {
138 	setup_clear_cpu_cap(X86_FEATURE_FXSR);
139 	setup_clear_cpu_cap(X86_FEATURE_XMM);
140 	return 1;
141 }
142 __setup("nofxsr", x86_fxsr_setup);
143 
144 static int __init x86_sep_setup(char *s)
145 {
146 	setup_clear_cpu_cap(X86_FEATURE_SEP);
147 	return 1;
148 }
149 __setup("nosep", x86_sep_setup);
150 
151 /* Standard macro to see if a specific flag is changeable */
152 static inline int flag_is_changeable_p(u32 flag)
153 {
154 	u32 f1, f2;
155 
156 	/*
157 	 * Cyrix and IDT cpus allow disabling of CPUID
158 	 * so the code below may return different results
159 	 * when it is executed before and after enabling
160 	 * the CPUID. Add "volatile" to not allow gcc to
161 	 * optimize the subsequent calls to this function.
162 	 */
163 	asm volatile ("pushfl		\n\t"
164 		      "pushfl		\n\t"
165 		      "popl %0		\n\t"
166 		      "movl %0, %1	\n\t"
167 		      "xorl %2, %0	\n\t"
168 		      "pushl %0		\n\t"
169 		      "popfl		\n\t"
170 		      "pushfl		\n\t"
171 		      "popl %0		\n\t"
172 		      "popfl		\n\t"
173 
174 		      : "=&r" (f1), "=&r" (f2)
175 		      : "ir" (flag));
176 
177 	return ((f1^f2) & flag) != 0;
178 }
179 
180 /* Probe for the CPUID instruction */
181 static int __cpuinit have_cpuid_p(void)
182 {
183 	return flag_is_changeable_p(X86_EFLAGS_ID);
184 }
185 
186 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
187 {
188 	unsigned long lo, hi;
189 
190 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
191 		return;
192 
193 	/* Disable processor serial number: */
194 
195 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
196 	lo |= 0x200000;
197 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
198 
199 	printk(KERN_NOTICE "CPU serial number disabled.\n");
200 	clear_cpu_cap(c, X86_FEATURE_PN);
201 
202 	/* Disabling the serial number may affect the cpuid level */
203 	c->cpuid_level = cpuid_eax(0);
204 }
205 
206 static int __init x86_serial_nr_setup(char *s)
207 {
208 	disable_x86_serial_nr = 0;
209 	return 1;
210 }
211 __setup("serialnumber", x86_serial_nr_setup);
212 #else
213 static inline int flag_is_changeable_p(u32 flag)
214 {
215 	return 1;
216 }
217 /* Probe for the CPUID instruction */
218 static inline int have_cpuid_p(void)
219 {
220 	return 1;
221 }
222 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
223 {
224 }
225 #endif
226 
227 /*
228  * Some CPU features depend on higher CPUID levels, which may not always
229  * be available due to CPUID level capping or broken virtualization
230  * software.  Add those features to this table to auto-disable them.
231  */
232 struct cpuid_dependent_feature {
233 	u32 feature;
234 	u32 level;
235 };
236 
237 static const struct cpuid_dependent_feature __cpuinitconst
238 cpuid_dependent_features[] = {
239 	{ X86_FEATURE_MWAIT,		0x00000005 },
240 	{ X86_FEATURE_DCA,		0x00000009 },
241 	{ X86_FEATURE_XSAVE,		0x0000000d },
242 	{ 0, 0 }
243 };
244 
245 static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
246 {
247 	const struct cpuid_dependent_feature *df;
248 
249 	for (df = cpuid_dependent_features; df->feature; df++) {
250 
251 		if (!cpu_has(c, df->feature))
252 			continue;
253 		/*
254 		 * Note: cpuid_level is set to -1 if unavailable, but
255 		 * extended_extended_level is set to 0 if unavailable
256 		 * and the legitimate extended levels are all negative
257 		 * when signed; hence the weird messing around with
258 		 * signs here...
259 		 */
260 		if (!((s32)df->level < 0 ?
261 		     (u32)df->level > (u32)c->extended_cpuid_level :
262 		     (s32)df->level > (s32)c->cpuid_level))
263 			continue;
264 
265 		clear_cpu_cap(c, df->feature);
266 		if (!warn)
267 			continue;
268 
269 		printk(KERN_WARNING
270 		       "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
271 				x86_cap_flags[df->feature], df->level);
272 	}
273 }
274 
275 /*
276  * Naming convention should be: <Name> [(<Codename>)]
277  * This table only is used unless init_<vendor>() below doesn't set it;
278  * in particular, if CPUID levels 0x80000002..4 are supported, this
279  * isn't used
280  */
281 
282 /* Look up CPU names by table lookup. */
283 static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
284 {
285 	const struct cpu_model_info *info;
286 
287 	if (c->x86_model >= 16)
288 		return NULL;	/* Range check */
289 
290 	if (!this_cpu)
291 		return NULL;
292 
293 	info = this_cpu->c_models;
294 
295 	while (info && info->family) {
296 		if (info->family == c->x86)
297 			return info->model_names[c->x86_model];
298 		info++;
299 	}
300 	return NULL;		/* Not found */
301 }
302 
303 __u32 cpu_caps_cleared[NCAPINTS] __cpuinitdata;
304 __u32 cpu_caps_set[NCAPINTS] __cpuinitdata;
305 
306 void load_percpu_segment(int cpu)
307 {
308 #ifdef CONFIG_X86_32
309 	loadsegment(fs, __KERNEL_PERCPU);
310 #else
311 	loadsegment(gs, 0);
312 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
313 #endif
314 	load_stack_canary_segment();
315 }
316 
317 /*
318  * Current gdt points %fs at the "master" per-cpu area: after this,
319  * it's on the real one.
320  */
321 void switch_to_new_gdt(int cpu)
322 {
323 	struct desc_ptr gdt_descr;
324 
325 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
326 	gdt_descr.size = GDT_SIZE - 1;
327 	load_gdt(&gdt_descr);
328 	/* Reload the per-cpu base */
329 
330 	load_percpu_segment(cpu);
331 }
332 
333 static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
334 
335 static void __cpuinit default_init(struct cpuinfo_x86 *c)
336 {
337 #ifdef CONFIG_X86_64
338 	display_cacheinfo(c);
339 #else
340 	/* Not much we can do here... */
341 	/* Check if at least it has cpuid */
342 	if (c->cpuid_level == -1) {
343 		/* No cpuid. It must be an ancient CPU */
344 		if (c->x86 == 4)
345 			strcpy(c->x86_model_id, "486");
346 		else if (c->x86 == 3)
347 			strcpy(c->x86_model_id, "386");
348 	}
349 #endif
350 }
351 
352 static const struct cpu_dev __cpuinitconst default_cpu = {
353 	.c_init	= default_init,
354 	.c_vendor = "Unknown",
355 	.c_x86_vendor = X86_VENDOR_UNKNOWN,
356 };
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 	static int printed;
491 	int i;
492 
493 	for (i = 0; i < X86_VENDOR_NUM; i++) {
494 		if (!cpu_devs[i])
495 			break;
496 
497 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
498 		    (cpu_devs[i]->c_ident[1] &&
499 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
500 
501 			this_cpu = cpu_devs[i];
502 			c->x86_vendor = this_cpu->c_x86_vendor;
503 			return;
504 		}
505 	}
506 
507 	if (!printed) {
508 		printed++;
509 		printk(KERN_ERR
510 		    "CPU: vendor_id '%s' unknown, using generic init.\n", v);
511 
512 		printk(KERN_ERR "CPU: Your system may be unstable.\n");
513 	}
514 
515 	c->x86_vendor = X86_VENDOR_UNKNOWN;
516 	this_cpu = &default_cpu;
517 }
518 
519 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
520 {
521 	/* Get vendor name */
522 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
523 	      (unsigned int *)&c->x86_vendor_id[0],
524 	      (unsigned int *)&c->x86_vendor_id[8],
525 	      (unsigned int *)&c->x86_vendor_id[4]);
526 
527 	c->x86 = 4;
528 	/* Intel-defined flags: level 0x00000001 */
529 	if (c->cpuid_level >= 0x00000001) {
530 		u32 junk, tfms, cap0, misc;
531 
532 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
533 		c->x86 = (tfms >> 8) & 0xf;
534 		c->x86_model = (tfms >> 4) & 0xf;
535 		c->x86_mask = tfms & 0xf;
536 
537 		if (c->x86 == 0xf)
538 			c->x86 += (tfms >> 20) & 0xff;
539 		if (c->x86 >= 0x6)
540 			c->x86_model += ((tfms >> 16) & 0xf) << 4;
541 
542 		if (cap0 & (1<<19)) {
543 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
544 			c->x86_cache_alignment = c->x86_clflush_size;
545 		}
546 	}
547 }
548 
549 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
550 {
551 	u32 tfms, xlvl;
552 	u32 ebx;
553 
554 	/* Intel-defined flags: level 0x00000001 */
555 	if (c->cpuid_level >= 0x00000001) {
556 		u32 capability, excap;
557 
558 		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
559 		c->x86_capability[0] = capability;
560 		c->x86_capability[4] = excap;
561 	}
562 
563 	/* AMD-defined flags: level 0x80000001 */
564 	xlvl = cpuid_eax(0x80000000);
565 	c->extended_cpuid_level = xlvl;
566 
567 	if ((xlvl & 0xffff0000) == 0x80000000) {
568 		if (xlvl >= 0x80000001) {
569 			c->x86_capability[1] = cpuid_edx(0x80000001);
570 			c->x86_capability[6] = cpuid_ecx(0x80000001);
571 		}
572 	}
573 
574 	if (c->extended_cpuid_level >= 0x80000008) {
575 		u32 eax = cpuid_eax(0x80000008);
576 
577 		c->x86_virt_bits = (eax >> 8) & 0xff;
578 		c->x86_phys_bits = eax & 0xff;
579 	}
580 #ifdef CONFIG_X86_32
581 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
582 		c->x86_phys_bits = 36;
583 #endif
584 
585 	if (c->extended_cpuid_level >= 0x80000007)
586 		c->x86_power = cpuid_edx(0x80000007);
587 
588 }
589 
590 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
591 {
592 #ifdef CONFIG_X86_32
593 	int i;
594 
595 	/*
596 	 * First of all, decide if this is a 486 or higher
597 	 * It's a 486 if we can modify the AC flag
598 	 */
599 	if (flag_is_changeable_p(X86_EFLAGS_AC))
600 		c->x86 = 4;
601 	else
602 		c->x86 = 3;
603 
604 	for (i = 0; i < X86_VENDOR_NUM; i++)
605 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
606 			c->x86_vendor_id[0] = 0;
607 			cpu_devs[i]->c_identify(c);
608 			if (c->x86_vendor_id[0]) {
609 				get_cpu_vendor(c);
610 				break;
611 			}
612 		}
613 #endif
614 }
615 
616 /*
617  * Do minimum CPU detection early.
618  * Fields really needed: vendor, cpuid_level, family, model, mask,
619  * cache alignment.
620  * The others are not touched to avoid unwanted side effects.
621  *
622  * WARNING: this function is only called on the BP.  Don't add code here
623  * that is supposed to run on all CPUs.
624  */
625 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
626 {
627 #ifdef CONFIG_X86_64
628 	c->x86_clflush_size = 64;
629 	c->x86_phys_bits = 36;
630 	c->x86_virt_bits = 48;
631 #else
632 	c->x86_clflush_size = 32;
633 	c->x86_phys_bits = 32;
634 	c->x86_virt_bits = 32;
635 #endif
636 	c->x86_cache_alignment = c->x86_clflush_size;
637 
638 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
639 	c->extended_cpuid_level = 0;
640 
641 	if (!have_cpuid_p())
642 		identify_cpu_without_cpuid(c);
643 
644 	/* cyrix could have cpuid enabled via c_identify()*/
645 	if (!have_cpuid_p())
646 		return;
647 
648 	cpu_detect(c);
649 
650 	get_cpu_vendor(c);
651 
652 	get_cpu_cap(c);
653 
654 	if (this_cpu->c_early_init)
655 		this_cpu->c_early_init(c);
656 
657 #ifdef CONFIG_SMP
658 	c->cpu_index = boot_cpu_id;
659 #endif
660 	filter_cpuid_features(c, false);
661 }
662 
663 void __init early_cpu_init(void)
664 {
665 	const struct cpu_dev *const *cdev;
666 	int count = 0;
667 
668 	printk(KERN_INFO "KERNEL supported cpus:\n");
669 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
670 		const struct cpu_dev *cpudev = *cdev;
671 		unsigned int j;
672 
673 		if (count >= X86_VENDOR_NUM)
674 			break;
675 		cpu_devs[count] = cpudev;
676 		count++;
677 
678 		for (j = 0; j < 2; j++) {
679 			if (!cpudev->c_ident[j])
680 				continue;
681 			printk(KERN_INFO "  %s %s\n", cpudev->c_vendor,
682 				cpudev->c_ident[j]);
683 		}
684 	}
685 
686 	early_identify_cpu(&boot_cpu_data);
687 }
688 
689 /*
690  * The NOPL instruction is supposed to exist on all CPUs with
691  * family >= 6; unfortunately, that's not true in practice because
692  * of early VIA chips and (more importantly) broken virtualizers that
693  * are not easy to detect.  In the latter case it doesn't even *fail*
694  * reliably, so probing for it doesn't even work.  Disable it completely
695  * unless we can find a reliable way to detect all the broken cases.
696  */
697 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
698 {
699 	clear_cpu_cap(c, X86_FEATURE_NOPL);
700 }
701 
702 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
703 {
704 	c->extended_cpuid_level = 0;
705 
706 	if (!have_cpuid_p())
707 		identify_cpu_without_cpuid(c);
708 
709 	/* cyrix could have cpuid enabled via c_identify()*/
710 	if (!have_cpuid_p())
711 		return;
712 
713 	cpu_detect(c);
714 
715 	get_cpu_vendor(c);
716 
717 	get_cpu_cap(c);
718 
719 	if (c->cpuid_level >= 0x00000001) {
720 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
721 #ifdef CONFIG_X86_32
722 # ifdef CONFIG_X86_HT
723 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
724 # else
725 		c->apicid = c->initial_apicid;
726 # endif
727 #endif
728 
729 #ifdef CONFIG_X86_HT
730 		c->phys_proc_id = c->initial_apicid;
731 #endif
732 	}
733 
734 	get_model_name(c); /* Default name */
735 
736 	init_scattered_cpuid_features(c);
737 	detect_nopl(c);
738 }
739 
740 /*
741  * This does the hard work of actually picking apart the CPU stuff...
742  */
743 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
744 {
745 	int i;
746 
747 	c->loops_per_jiffy = loops_per_jiffy;
748 	c->x86_cache_size = -1;
749 	c->x86_vendor = X86_VENDOR_UNKNOWN;
750 	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
751 	c->x86_vendor_id[0] = '\0'; /* Unset */
752 	c->x86_model_id[0] = '\0';  /* Unset */
753 	c->x86_max_cores = 1;
754 	c->x86_coreid_bits = 0;
755 #ifdef CONFIG_X86_64
756 	c->x86_clflush_size = 64;
757 	c->x86_phys_bits = 36;
758 	c->x86_virt_bits = 48;
759 #else
760 	c->cpuid_level = -1;	/* CPUID not detected */
761 	c->x86_clflush_size = 32;
762 	c->x86_phys_bits = 32;
763 	c->x86_virt_bits = 32;
764 #endif
765 	c->x86_cache_alignment = c->x86_clflush_size;
766 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
767 
768 	generic_identify(c);
769 
770 	if (this_cpu->c_identify)
771 		this_cpu->c_identify(c);
772 
773 	/* Clear/Set all flags overriden by options, after probe */
774 	for (i = 0; i < NCAPINTS; i++) {
775 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
776 		c->x86_capability[i] |= cpu_caps_set[i];
777 	}
778 
779 #ifdef CONFIG_X86_64
780 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
781 #endif
782 
783 	/*
784 	 * Vendor-specific initialization.  In this section we
785 	 * canonicalize the feature flags, meaning if there are
786 	 * features a certain CPU supports which CPUID doesn't
787 	 * tell us, CPUID claiming incorrect flags, or other bugs,
788 	 * we handle them here.
789 	 *
790 	 * At the end of this section, c->x86_capability better
791 	 * indicate the features this CPU genuinely supports!
792 	 */
793 	if (this_cpu->c_init)
794 		this_cpu->c_init(c);
795 
796 	/* Disable the PN if appropriate */
797 	squash_the_stupid_serial_number(c);
798 
799 	/*
800 	 * The vendor-specific functions might have changed features.
801 	 * Now we do "generic changes."
802 	 */
803 
804 	/* Filter out anything that depends on CPUID levels we don't have */
805 	filter_cpuid_features(c, true);
806 
807 	/* If the model name is still unset, do table lookup. */
808 	if (!c->x86_model_id[0]) {
809 		const char *p;
810 		p = table_lookup_model(c);
811 		if (p)
812 			strcpy(c->x86_model_id, p);
813 		else
814 			/* Last resort... */
815 			sprintf(c->x86_model_id, "%02x/%02x",
816 				c->x86, c->x86_model);
817 	}
818 
819 #ifdef CONFIG_X86_64
820 	detect_ht(c);
821 #endif
822 
823 	init_hypervisor(c);
824 
825 	/*
826 	 * Clear/Set all flags overriden by options, need do it
827 	 * before following smp all cpus cap AND.
828 	 */
829 	for (i = 0; i < NCAPINTS; i++) {
830 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
831 		c->x86_capability[i] |= cpu_caps_set[i];
832 	}
833 
834 	/*
835 	 * On SMP, boot_cpu_data holds the common feature set between
836 	 * all CPUs; so make sure that we indicate which features are
837 	 * common between the CPUs.  The first time this routine gets
838 	 * executed, c == &boot_cpu_data.
839 	 */
840 	if (c != &boot_cpu_data) {
841 		/* AND the already accumulated flags with these */
842 		for (i = 0; i < NCAPINTS; i++)
843 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
844 	}
845 
846 #ifdef CONFIG_X86_MCE
847 	/* Init Machine Check Exception if available. */
848 	mcheck_init(c);
849 #endif
850 
851 	select_idle_routine(c);
852 
853 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
854 	numa_add_cpu(smp_processor_id());
855 #endif
856 }
857 
858 #ifdef CONFIG_X86_64
859 static void vgetcpu_set_mode(void)
860 {
861 	if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
862 		vgetcpu_mode = VGETCPU_RDTSCP;
863 	else
864 		vgetcpu_mode = VGETCPU_LSL;
865 }
866 #endif
867 
868 void __init identify_boot_cpu(void)
869 {
870 	identify_cpu(&boot_cpu_data);
871 	init_c1e_mask();
872 #ifdef CONFIG_X86_32
873 	sysenter_setup();
874 	enable_sep_cpu();
875 #else
876 	vgetcpu_set_mode();
877 #endif
878 	init_hw_perf_counters();
879 }
880 
881 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
882 {
883 	BUG_ON(c == &boot_cpu_data);
884 	identify_cpu(c);
885 #ifdef CONFIG_X86_32
886 	enable_sep_cpu();
887 #endif
888 	mtrr_ap_init();
889 }
890 
891 struct msr_range {
892 	unsigned	min;
893 	unsigned	max;
894 };
895 
896 static const struct msr_range msr_range_array[] __cpuinitconst = {
897 	{ 0x00000000, 0x00000418},
898 	{ 0xc0000000, 0xc000040b},
899 	{ 0xc0010000, 0xc0010142},
900 	{ 0xc0011000, 0xc001103b},
901 };
902 
903 static void __cpuinit print_cpu_msr(void)
904 {
905 	unsigned index_min, index_max;
906 	unsigned index;
907 	u64 val;
908 	int i;
909 
910 	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
911 		index_min = msr_range_array[i].min;
912 		index_max = msr_range_array[i].max;
913 
914 		for (index = index_min; index < index_max; index++) {
915 			if (rdmsrl_amd_safe(index, &val))
916 				continue;
917 			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
918 		}
919 	}
920 }
921 
922 static int show_msr __cpuinitdata;
923 
924 static __init int setup_show_msr(char *arg)
925 {
926 	int num;
927 
928 	get_option(&arg, &num);
929 
930 	if (num > 0)
931 		show_msr = num;
932 	return 1;
933 }
934 __setup("show_msr=", setup_show_msr);
935 
936 static __init int setup_noclflush(char *arg)
937 {
938 	setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
939 	return 1;
940 }
941 __setup("noclflush", setup_noclflush);
942 
943 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
944 {
945 	const char *vendor = NULL;
946 
947 	if (c->x86_vendor < X86_VENDOR_NUM) {
948 		vendor = this_cpu->c_vendor;
949 	} else {
950 		if (c->cpuid_level >= 0)
951 			vendor = c->x86_vendor_id;
952 	}
953 
954 	if (vendor && !strstr(c->x86_model_id, vendor))
955 		printk(KERN_CONT "%s ", vendor);
956 
957 	if (c->x86_model_id[0])
958 		printk(KERN_CONT "%s", c->x86_model_id);
959 	else
960 		printk(KERN_CONT "%d86", c->x86);
961 
962 	if (c->x86_mask || c->cpuid_level >= 0)
963 		printk(KERN_CONT " stepping %02x\n", c->x86_mask);
964 	else
965 		printk(KERN_CONT "\n");
966 
967 #ifdef CONFIG_SMP
968 	if (c->cpu_index < show_msr)
969 		print_cpu_msr();
970 #else
971 	if (show_msr)
972 		print_cpu_msr();
973 #endif
974 }
975 
976 static __init int setup_disablecpuid(char *arg)
977 {
978 	int bit;
979 
980 	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
981 		setup_clear_cpu_cap(bit);
982 	else
983 		return 0;
984 
985 	return 1;
986 }
987 __setup("clearcpuid=", setup_disablecpuid);
988 
989 #ifdef CONFIG_X86_64
990 struct desc_ptr idt_descr = { 256 * 16 - 1, (unsigned long) idt_table };
991 
992 DEFINE_PER_CPU_FIRST(union irq_stack_union,
993 		     irq_stack_union) __aligned(PAGE_SIZE);
994 
995 DEFINE_PER_CPU(char *, irq_stack_ptr) =
996 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
997 
998 DEFINE_PER_CPU(unsigned long, kernel_stack) =
999 	(unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
1000 EXPORT_PER_CPU_SYMBOL(kernel_stack);
1001 
1002 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
1003 
1004 /*
1005  * Special IST stacks which the CPU switches to when it calls
1006  * an IST-marked descriptor entry. Up to 7 stacks (hardware
1007  * limit), all of them are 4K, except the debug stack which
1008  * is 8K.
1009  */
1010 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1011 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1012 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1013 };
1014 
1015 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1016 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ])
1017 	__aligned(PAGE_SIZE);
1018 
1019 /* May not be marked __init: used by software suspend */
1020 void syscall_init(void)
1021 {
1022 	/*
1023 	 * LSTAR and STAR live in a bit strange symbiosis.
1024 	 * They both write to the same internal register. STAR allows to
1025 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1026 	 */
1027 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1028 	wrmsrl(MSR_LSTAR, system_call);
1029 	wrmsrl(MSR_CSTAR, ignore_sysret);
1030 
1031 #ifdef CONFIG_IA32_EMULATION
1032 	syscall32_cpu_init();
1033 #endif
1034 
1035 	/* Flags to clear on syscall */
1036 	wrmsrl(MSR_SYSCALL_MASK,
1037 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
1038 }
1039 
1040 unsigned long kernel_eflags;
1041 
1042 /*
1043  * Copies of the original ist values from the tss are only accessed during
1044  * debugging, no special alignment required.
1045  */
1046 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1047 
1048 #else	/* CONFIG_X86_64 */
1049 
1050 #ifdef CONFIG_CC_STACKPROTECTOR
1051 DEFINE_PER_CPU(unsigned long, stack_canary);
1052 #endif
1053 
1054 /* Make sure %fs and %gs are initialized properly in idle threads */
1055 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
1056 {
1057 	memset(regs, 0, sizeof(struct pt_regs));
1058 	regs->fs = __KERNEL_PERCPU;
1059 	regs->gs = __KERNEL_STACK_CANARY;
1060 
1061 	return regs;
1062 }
1063 #endif	/* CONFIG_X86_64 */
1064 
1065 /*
1066  * Clear all 6 debug registers:
1067  */
1068 static void clear_all_debug_regs(void)
1069 {
1070 	int i;
1071 
1072 	for (i = 0; i < 8; i++) {
1073 		/* Ignore db4, db5 */
1074 		if ((i == 4) || (i == 5))
1075 			continue;
1076 
1077 		set_debugreg(0, i);
1078 	}
1079 }
1080 
1081 /*
1082  * cpu_init() initializes state that is per-CPU. Some data is already
1083  * initialized (naturally) in the bootstrap process, such as the GDT
1084  * and IDT. We reload them nevertheless, this function acts as a
1085  * 'CPU state barrier', nothing should get across.
1086  * A lot of state is already set up in PDA init for 64 bit
1087  */
1088 #ifdef CONFIG_X86_64
1089 
1090 void __cpuinit cpu_init(void)
1091 {
1092 	struct orig_ist *orig_ist;
1093 	struct task_struct *me;
1094 	struct tss_struct *t;
1095 	unsigned long v;
1096 	int cpu;
1097 	int i;
1098 
1099 	cpu = stack_smp_processor_id();
1100 	t = &per_cpu(init_tss, cpu);
1101 	orig_ist = &per_cpu(orig_ist, cpu);
1102 
1103 #ifdef CONFIG_NUMA
1104 	if (cpu != 0 && percpu_read(node_number) == 0 &&
1105 	    cpu_to_node(cpu) != NUMA_NO_NODE)
1106 		percpu_write(node_number, cpu_to_node(cpu));
1107 #endif
1108 
1109 	me = current;
1110 
1111 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1112 		panic("CPU#%d already initialized!\n", cpu);
1113 
1114 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1115 
1116 	clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1117 
1118 	/*
1119 	 * Initialize the per-CPU GDT with the boot GDT,
1120 	 * and set up the GDT descriptor:
1121 	 */
1122 
1123 	switch_to_new_gdt(cpu);
1124 	loadsegment(fs, 0);
1125 
1126 	load_idt((const struct desc_ptr *)&idt_descr);
1127 
1128 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1129 	syscall_init();
1130 
1131 	wrmsrl(MSR_FS_BASE, 0);
1132 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1133 	barrier();
1134 
1135 	check_efer();
1136 	if (cpu != 0)
1137 		enable_x2apic();
1138 
1139 	/*
1140 	 * set up and load the per-CPU TSS
1141 	 */
1142 	if (!orig_ist->ist[0]) {
1143 		char *estacks = per_cpu(exception_stacks, cpu);
1144 
1145 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1146 			estacks += exception_stack_sizes[v];
1147 			orig_ist->ist[v] = t->x86_tss.ist[v] =
1148 					(unsigned long)estacks;
1149 		}
1150 	}
1151 
1152 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1153 
1154 	/*
1155 	 * <= is required because the CPU will access up to
1156 	 * 8 bits beyond the end of the IO permission bitmap.
1157 	 */
1158 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1159 		t->io_bitmap[i] = ~0UL;
1160 
1161 	atomic_inc(&init_mm.mm_count);
1162 	me->active_mm = &init_mm;
1163 	BUG_ON(me->mm);
1164 	enter_lazy_tlb(&init_mm, me);
1165 
1166 	load_sp0(t, &current->thread);
1167 	set_tss_desc(cpu, t);
1168 	load_TR_desc();
1169 	load_LDT(&init_mm.context);
1170 
1171 #ifdef CONFIG_KGDB
1172 	/*
1173 	 * If the kgdb is connected no debug regs should be altered.  This
1174 	 * is only applicable when KGDB and a KGDB I/O module are built
1175 	 * into the kernel and you are using early debugging with
1176 	 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1177 	 */
1178 	if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1179 		arch_kgdb_ops.correct_hw_break();
1180 	else
1181 #endif
1182 		clear_all_debug_regs();
1183 
1184 	fpu_init();
1185 
1186 	raw_local_save_flags(kernel_eflags);
1187 
1188 	if (is_uv_system())
1189 		uv_cpu_init();
1190 }
1191 
1192 #else
1193 
1194 void __cpuinit cpu_init(void)
1195 {
1196 	int cpu = smp_processor_id();
1197 	struct task_struct *curr = current;
1198 	struct tss_struct *t = &per_cpu(init_tss, cpu);
1199 	struct thread_struct *thread = &curr->thread;
1200 
1201 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1202 		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1203 		for (;;)
1204 			local_irq_enable();
1205 	}
1206 
1207 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1208 
1209 	if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1210 		clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1211 
1212 	load_idt(&idt_descr);
1213 	switch_to_new_gdt(cpu);
1214 
1215 	/*
1216 	 * Set up and load the per-CPU TSS and LDT
1217 	 */
1218 	atomic_inc(&init_mm.mm_count);
1219 	curr->active_mm = &init_mm;
1220 	BUG_ON(curr->mm);
1221 	enter_lazy_tlb(&init_mm, curr);
1222 
1223 	load_sp0(t, thread);
1224 	set_tss_desc(cpu, t);
1225 	load_TR_desc();
1226 	load_LDT(&init_mm.context);
1227 
1228 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1229 
1230 #ifdef CONFIG_DOUBLEFAULT
1231 	/* Set up doublefault TSS pointer in the GDT */
1232 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1233 #endif
1234 
1235 	clear_all_debug_regs();
1236 
1237 	/*
1238 	 * Force FPU initialization:
1239 	 */
1240 	if (cpu_has_xsave)
1241 		current_thread_info()->status = TS_XSAVE;
1242 	else
1243 		current_thread_info()->status = 0;
1244 	clear_used_math();
1245 	mxcsr_feature_mask_init();
1246 
1247 	/*
1248 	 * Boot processor to setup the FP and extended state context info.
1249 	 */
1250 	if (smp_processor_id() == boot_cpu_id)
1251 		init_thread_xstate();
1252 
1253 	xsave_init();
1254 }
1255 #endif
1256