xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision fd589a8f)
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 <linux/topology.h>
22 #include <linux/cpumask.h>
23 #include <asm/pgtable.h>
24 #include <asm/atomic.h>
25 #include <asm/proto.h>
26 #include <asm/setup.h>
27 #include <asm/apic.h>
28 #include <asm/desc.h>
29 #include <asm/i387.h>
30 #include <asm/mtrr.h>
31 #include <linux/numa.h>
32 #include <asm/asm.h>
33 #include <asm/cpu.h>
34 #include <asm/mce.h>
35 #include <asm/msr.h>
36 #include <asm/pat.h>
37 
38 #ifdef CONFIG_X86_LOCAL_APIC
39 #include <asm/uv/uv.h>
40 #endif
41 
42 #include "cpu.h"
43 
44 /* all of these masks are initialized in setup_cpu_local_masks() */
45 cpumask_var_t cpu_initialized_mask;
46 cpumask_var_t cpu_callout_mask;
47 cpumask_var_t cpu_callin_mask;
48 
49 /* representing cpus for which sibling maps can be computed */
50 cpumask_var_t cpu_sibling_setup_mask;
51 
52 /* correctly size the local cpu masks */
53 void __init setup_cpu_local_masks(void)
54 {
55 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
56 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
57 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
58 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
59 }
60 
61 static void __cpuinit default_init(struct cpuinfo_x86 *c)
62 {
63 #ifdef CONFIG_X86_64
64 	display_cacheinfo(c);
65 #else
66 	/* Not much we can do here... */
67 	/* Check if at least it has cpuid */
68 	if (c->cpuid_level == -1) {
69 		/* No cpuid. It must be an ancient CPU */
70 		if (c->x86 == 4)
71 			strcpy(c->x86_model_id, "486");
72 		else if (c->x86 == 3)
73 			strcpy(c->x86_model_id, "386");
74 	}
75 #endif
76 }
77 
78 static const struct cpu_dev __cpuinitconst default_cpu = {
79 	.c_init		= default_init,
80 	.c_vendor	= "Unknown",
81 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
82 };
83 
84 static const struct cpu_dev *this_cpu __cpuinitdata = &default_cpu;
85 
86 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
87 #ifdef CONFIG_X86_64
88 	/*
89 	 * We need valid kernel segments for data and code in long mode too
90 	 * IRET will check the segment types  kkeil 2000/10/28
91 	 * Also sysret mandates a special GDT layout
92 	 *
93 	 * TLS descriptors are currently at a different place compared to i386.
94 	 * Hopefully nobody expects them at a fixed place (Wine?)
95 	 */
96 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
97 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
98 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
99 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
100 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
101 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
102 #else
103 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
104 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
105 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
106 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
107 	/*
108 	 * Segments used for calling PnP BIOS have byte granularity.
109 	 * They code segments and data segments have fixed 64k limits,
110 	 * the transfer segment sizes are set at run time.
111 	 */
112 	/* 32-bit code */
113 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
114 	/* 16-bit code */
115 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
116 	/* 16-bit data */
117 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
118 	/* 16-bit data */
119 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
120 	/* 16-bit data */
121 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
122 	/*
123 	 * The APM segments have byte granularity and their bases
124 	 * are set at run time.  All have 64k limits.
125 	 */
126 	/* 32-bit code */
127 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
128 	/* 16-bit code */
129 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
130 	/* data */
131 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
132 
133 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
134 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
135 	GDT_STACK_CANARY_INIT
136 #endif
137 } };
138 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
139 
140 static int __init x86_xsave_setup(char *s)
141 {
142 	setup_clear_cpu_cap(X86_FEATURE_XSAVE);
143 	return 1;
144 }
145 __setup("noxsave", x86_xsave_setup);
146 
147 #ifdef CONFIG_X86_32
148 static int cachesize_override __cpuinitdata = -1;
149 static int disable_x86_serial_nr __cpuinitdata = 1;
150 
151 static int __init cachesize_setup(char *str)
152 {
153 	get_option(&str, &cachesize_override);
154 	return 1;
155 }
156 __setup("cachesize=", cachesize_setup);
157 
158 static int __init x86_fxsr_setup(char *s)
159 {
160 	setup_clear_cpu_cap(X86_FEATURE_FXSR);
161 	setup_clear_cpu_cap(X86_FEATURE_XMM);
162 	return 1;
163 }
164 __setup("nofxsr", x86_fxsr_setup);
165 
166 static int __init x86_sep_setup(char *s)
167 {
168 	setup_clear_cpu_cap(X86_FEATURE_SEP);
169 	return 1;
170 }
171 __setup("nosep", x86_sep_setup);
172 
173 /* Standard macro to see if a specific flag is changeable */
174 static inline int flag_is_changeable_p(u32 flag)
175 {
176 	u32 f1, f2;
177 
178 	/*
179 	 * Cyrix and IDT cpus allow disabling of CPUID
180 	 * so the code below may return different results
181 	 * when it is executed before and after enabling
182 	 * the CPUID. Add "volatile" to not allow gcc to
183 	 * optimize the subsequent calls to this function.
184 	 */
185 	asm volatile ("pushfl		\n\t"
186 		      "pushfl		\n\t"
187 		      "popl %0		\n\t"
188 		      "movl %0, %1	\n\t"
189 		      "xorl %2, %0	\n\t"
190 		      "pushl %0		\n\t"
191 		      "popfl		\n\t"
192 		      "pushfl		\n\t"
193 		      "popl %0		\n\t"
194 		      "popfl		\n\t"
195 
196 		      : "=&r" (f1), "=&r" (f2)
197 		      : "ir" (flag));
198 
199 	return ((f1^f2) & flag) != 0;
200 }
201 
202 /* Probe for the CPUID instruction */
203 static int __cpuinit have_cpuid_p(void)
204 {
205 	return flag_is_changeable_p(X86_EFLAGS_ID);
206 }
207 
208 static void __cpuinit squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
209 {
210 	unsigned long lo, hi;
211 
212 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
213 		return;
214 
215 	/* Disable processor serial number: */
216 
217 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
218 	lo |= 0x200000;
219 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
220 
221 	printk(KERN_NOTICE "CPU serial number disabled.\n");
222 	clear_cpu_cap(c, X86_FEATURE_PN);
223 
224 	/* Disabling the serial number may affect the cpuid level */
225 	c->cpuid_level = cpuid_eax(0);
226 }
227 
228 static int __init x86_serial_nr_setup(char *s)
229 {
230 	disable_x86_serial_nr = 0;
231 	return 1;
232 }
233 __setup("serialnumber", x86_serial_nr_setup);
234 #else
235 static inline int flag_is_changeable_p(u32 flag)
236 {
237 	return 1;
238 }
239 /* Probe for the CPUID instruction */
240 static inline int have_cpuid_p(void)
241 {
242 	return 1;
243 }
244 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
245 {
246 }
247 #endif
248 
249 /*
250  * Some CPU features depend on higher CPUID levels, which may not always
251  * be available due to CPUID level capping or broken virtualization
252  * software.  Add those features to this table to auto-disable them.
253  */
254 struct cpuid_dependent_feature {
255 	u32 feature;
256 	u32 level;
257 };
258 
259 static const struct cpuid_dependent_feature __cpuinitconst
260 cpuid_dependent_features[] = {
261 	{ X86_FEATURE_MWAIT,		0x00000005 },
262 	{ X86_FEATURE_DCA,		0x00000009 },
263 	{ X86_FEATURE_XSAVE,		0x0000000d },
264 	{ 0, 0 }
265 };
266 
267 static void __cpuinit filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
268 {
269 	const struct cpuid_dependent_feature *df;
270 
271 	for (df = cpuid_dependent_features; df->feature; df++) {
272 
273 		if (!cpu_has(c, df->feature))
274 			continue;
275 		/*
276 		 * Note: cpuid_level is set to -1 if unavailable, but
277 		 * extended_extended_level is set to 0 if unavailable
278 		 * and the legitimate extended levels are all negative
279 		 * when signed; hence the weird messing around with
280 		 * signs here...
281 		 */
282 		if (!((s32)df->level < 0 ?
283 		     (u32)df->level > (u32)c->extended_cpuid_level :
284 		     (s32)df->level > (s32)c->cpuid_level))
285 			continue;
286 
287 		clear_cpu_cap(c, df->feature);
288 		if (!warn)
289 			continue;
290 
291 		printk(KERN_WARNING
292 		       "CPU: CPU feature %s disabled, no CPUID level 0x%x\n",
293 				x86_cap_flags[df->feature], df->level);
294 	}
295 }
296 
297 /*
298  * Naming convention should be: <Name> [(<Codename>)]
299  * This table only is used unless init_<vendor>() below doesn't set it;
300  * in particular, if CPUID levels 0x80000002..4 are supported, this
301  * isn't used
302  */
303 
304 /* Look up CPU names by table lookup. */
305 static const char *__cpuinit table_lookup_model(struct cpuinfo_x86 *c)
306 {
307 	const struct cpu_model_info *info;
308 
309 	if (c->x86_model >= 16)
310 		return NULL;	/* Range check */
311 
312 	if (!this_cpu)
313 		return NULL;
314 
315 	info = this_cpu->c_models;
316 
317 	while (info && info->family) {
318 		if (info->family == c->x86)
319 			return info->model_names[c->x86_model];
320 		info++;
321 	}
322 	return NULL;		/* Not found */
323 }
324 
325 __u32 cpu_caps_cleared[NCAPINTS] __cpuinitdata;
326 __u32 cpu_caps_set[NCAPINTS] __cpuinitdata;
327 
328 void load_percpu_segment(int cpu)
329 {
330 #ifdef CONFIG_X86_32
331 	loadsegment(fs, __KERNEL_PERCPU);
332 #else
333 	loadsegment(gs, 0);
334 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
335 #endif
336 	load_stack_canary_segment();
337 }
338 
339 /*
340  * Current gdt points %fs at the "master" per-cpu area: after this,
341  * it's on the real one.
342  */
343 void switch_to_new_gdt(int cpu)
344 {
345 	struct desc_ptr gdt_descr;
346 
347 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
348 	gdt_descr.size = GDT_SIZE - 1;
349 	load_gdt(&gdt_descr);
350 	/* Reload the per-cpu base */
351 
352 	load_percpu_segment(cpu);
353 }
354 
355 static const struct cpu_dev *__cpuinitdata cpu_devs[X86_VENDOR_NUM] = {};
356 
357 static void __cpuinit get_model_name(struct cpuinfo_x86 *c)
358 {
359 	unsigned int *v;
360 	char *p, *q;
361 
362 	if (c->extended_cpuid_level < 0x80000004)
363 		return;
364 
365 	v = (unsigned int *)c->x86_model_id;
366 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
367 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
368 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
369 	c->x86_model_id[48] = 0;
370 
371 	/*
372 	 * Intel chips right-justify this string for some dumb reason;
373 	 * undo that brain damage:
374 	 */
375 	p = q = &c->x86_model_id[0];
376 	while (*p == ' ')
377 		p++;
378 	if (p != q) {
379 		while (*p)
380 			*q++ = *p++;
381 		while (q <= &c->x86_model_id[48])
382 			*q++ = '\0';	/* Zero-pad the rest */
383 	}
384 }
385 
386 void __cpuinit display_cacheinfo(struct cpuinfo_x86 *c)
387 {
388 	unsigned int n, dummy, ebx, ecx, edx, l2size;
389 
390 	n = c->extended_cpuid_level;
391 
392 	if (n >= 0x80000005) {
393 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
394 		printk(KERN_INFO "CPU: L1 I Cache: %dK (%d bytes/line), D cache %dK (%d bytes/line)\n",
395 				edx>>24, edx&0xFF, ecx>>24, ecx&0xFF);
396 		c->x86_cache_size = (ecx>>24) + (edx>>24);
397 #ifdef CONFIG_X86_64
398 		/* On K8 L1 TLB is inclusive, so don't count it */
399 		c->x86_tlbsize = 0;
400 #endif
401 	}
402 
403 	if (n < 0x80000006)	/* Some chips just has a large L1. */
404 		return;
405 
406 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
407 	l2size = ecx >> 16;
408 
409 #ifdef CONFIG_X86_64
410 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
411 #else
412 	/* do processor-specific cache resizing */
413 	if (this_cpu->c_size_cache)
414 		l2size = this_cpu->c_size_cache(c, l2size);
415 
416 	/* Allow user to override all this if necessary. */
417 	if (cachesize_override != -1)
418 		l2size = cachesize_override;
419 
420 	if (l2size == 0)
421 		return;		/* Again, no L2 cache is possible */
422 #endif
423 
424 	c->x86_cache_size = l2size;
425 
426 	printk(KERN_INFO "CPU: L2 Cache: %dK (%d bytes/line)\n",
427 			l2size, ecx & 0xFF);
428 }
429 
430 void __cpuinit detect_ht(struct cpuinfo_x86 *c)
431 {
432 #ifdef CONFIG_X86_HT
433 	u32 eax, ebx, ecx, edx;
434 	int index_msb, core_bits;
435 
436 	if (!cpu_has(c, X86_FEATURE_HT))
437 		return;
438 
439 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
440 		goto out;
441 
442 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
443 		return;
444 
445 	cpuid(1, &eax, &ebx, &ecx, &edx);
446 
447 	smp_num_siblings = (ebx & 0xff0000) >> 16;
448 
449 	if (smp_num_siblings == 1) {
450 		printk(KERN_INFO  "CPU: Hyper-Threading is disabled\n");
451 		goto out;
452 	}
453 
454 	if (smp_num_siblings <= 1)
455 		goto out;
456 
457 	if (smp_num_siblings > nr_cpu_ids) {
458 		pr_warning("CPU: Unsupported number of siblings %d",
459 			   smp_num_siblings);
460 		smp_num_siblings = 1;
461 		return;
462 	}
463 
464 	index_msb = get_count_order(smp_num_siblings);
465 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
466 
467 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
468 
469 	index_msb = get_count_order(smp_num_siblings);
470 
471 	core_bits = get_count_order(c->x86_max_cores);
472 
473 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
474 				       ((1 << core_bits) - 1);
475 
476 out:
477 	if ((c->x86_max_cores * smp_num_siblings) > 1) {
478 		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
479 		       c->phys_proc_id);
480 		printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
481 		       c->cpu_core_id);
482 	}
483 #endif
484 }
485 
486 static void __cpuinit get_cpu_vendor(struct cpuinfo_x86 *c)
487 {
488 	char *v = c->x86_vendor_id;
489 	int i;
490 
491 	for (i = 0; i < X86_VENDOR_NUM; i++) {
492 		if (!cpu_devs[i])
493 			break;
494 
495 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
496 		    (cpu_devs[i]->c_ident[1] &&
497 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
498 
499 			this_cpu = cpu_devs[i];
500 			c->x86_vendor = this_cpu->c_x86_vendor;
501 			return;
502 		}
503 	}
504 
505 	printk_once(KERN_ERR
506 			"CPU: vendor_id '%s' unknown, using generic init.\n" \
507 			"CPU: Your system may be unstable.\n", v);
508 
509 	c->x86_vendor = X86_VENDOR_UNKNOWN;
510 	this_cpu = &default_cpu;
511 }
512 
513 void __cpuinit cpu_detect(struct cpuinfo_x86 *c)
514 {
515 	/* Get vendor name */
516 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
517 	      (unsigned int *)&c->x86_vendor_id[0],
518 	      (unsigned int *)&c->x86_vendor_id[8],
519 	      (unsigned int *)&c->x86_vendor_id[4]);
520 
521 	c->x86 = 4;
522 	/* Intel-defined flags: level 0x00000001 */
523 	if (c->cpuid_level >= 0x00000001) {
524 		u32 junk, tfms, cap0, misc;
525 
526 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
527 		c->x86 = (tfms >> 8) & 0xf;
528 		c->x86_model = (tfms >> 4) & 0xf;
529 		c->x86_mask = tfms & 0xf;
530 
531 		if (c->x86 == 0xf)
532 			c->x86 += (tfms >> 20) & 0xff;
533 		if (c->x86 >= 0x6)
534 			c->x86_model += ((tfms >> 16) & 0xf) << 4;
535 
536 		if (cap0 & (1<<19)) {
537 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
538 			c->x86_cache_alignment = c->x86_clflush_size;
539 		}
540 	}
541 }
542 
543 static void __cpuinit get_cpu_cap(struct cpuinfo_x86 *c)
544 {
545 	u32 tfms, xlvl;
546 	u32 ebx;
547 
548 	/* Intel-defined flags: level 0x00000001 */
549 	if (c->cpuid_level >= 0x00000001) {
550 		u32 capability, excap;
551 
552 		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
553 		c->x86_capability[0] = capability;
554 		c->x86_capability[4] = excap;
555 	}
556 
557 	/* AMD-defined flags: level 0x80000001 */
558 	xlvl = cpuid_eax(0x80000000);
559 	c->extended_cpuid_level = xlvl;
560 
561 	if ((xlvl & 0xffff0000) == 0x80000000) {
562 		if (xlvl >= 0x80000001) {
563 			c->x86_capability[1] = cpuid_edx(0x80000001);
564 			c->x86_capability[6] = cpuid_ecx(0x80000001);
565 		}
566 	}
567 
568 	if (c->extended_cpuid_level >= 0x80000008) {
569 		u32 eax = cpuid_eax(0x80000008);
570 
571 		c->x86_virt_bits = (eax >> 8) & 0xff;
572 		c->x86_phys_bits = eax & 0xff;
573 	}
574 #ifdef CONFIG_X86_32
575 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
576 		c->x86_phys_bits = 36;
577 #endif
578 
579 	if (c->extended_cpuid_level >= 0x80000007)
580 		c->x86_power = cpuid_edx(0x80000007);
581 
582 }
583 
584 static void __cpuinit identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
585 {
586 #ifdef CONFIG_X86_32
587 	int i;
588 
589 	/*
590 	 * First of all, decide if this is a 486 or higher
591 	 * It's a 486 if we can modify the AC flag
592 	 */
593 	if (flag_is_changeable_p(X86_EFLAGS_AC))
594 		c->x86 = 4;
595 	else
596 		c->x86 = 3;
597 
598 	for (i = 0; i < X86_VENDOR_NUM; i++)
599 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
600 			c->x86_vendor_id[0] = 0;
601 			cpu_devs[i]->c_identify(c);
602 			if (c->x86_vendor_id[0]) {
603 				get_cpu_vendor(c);
604 				break;
605 			}
606 		}
607 #endif
608 }
609 
610 /*
611  * Do minimum CPU detection early.
612  * Fields really needed: vendor, cpuid_level, family, model, mask,
613  * cache alignment.
614  * The others are not touched to avoid unwanted side effects.
615  *
616  * WARNING: this function is only called on the BP.  Don't add code here
617  * that is supposed to run on all CPUs.
618  */
619 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
620 {
621 #ifdef CONFIG_X86_64
622 	c->x86_clflush_size = 64;
623 	c->x86_phys_bits = 36;
624 	c->x86_virt_bits = 48;
625 #else
626 	c->x86_clflush_size = 32;
627 	c->x86_phys_bits = 32;
628 	c->x86_virt_bits = 32;
629 #endif
630 	c->x86_cache_alignment = c->x86_clflush_size;
631 
632 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
633 	c->extended_cpuid_level = 0;
634 
635 	if (!have_cpuid_p())
636 		identify_cpu_without_cpuid(c);
637 
638 	/* cyrix could have cpuid enabled via c_identify()*/
639 	if (!have_cpuid_p())
640 		return;
641 
642 	cpu_detect(c);
643 
644 	get_cpu_vendor(c);
645 
646 	get_cpu_cap(c);
647 
648 	if (this_cpu->c_early_init)
649 		this_cpu->c_early_init(c);
650 
651 #ifdef CONFIG_SMP
652 	c->cpu_index = boot_cpu_id;
653 #endif
654 	filter_cpuid_features(c, false);
655 }
656 
657 void __init early_cpu_init(void)
658 {
659 	const struct cpu_dev *const *cdev;
660 	int count = 0;
661 
662 	printk(KERN_INFO "KERNEL supported cpus:\n");
663 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
664 		const struct cpu_dev *cpudev = *cdev;
665 		unsigned int j;
666 
667 		if (count >= X86_VENDOR_NUM)
668 			break;
669 		cpu_devs[count] = cpudev;
670 		count++;
671 
672 		for (j = 0; j < 2; j++) {
673 			if (!cpudev->c_ident[j])
674 				continue;
675 			printk(KERN_INFO "  %s %s\n", cpudev->c_vendor,
676 				cpudev->c_ident[j]);
677 		}
678 	}
679 
680 	early_identify_cpu(&boot_cpu_data);
681 }
682 
683 /*
684  * The NOPL instruction is supposed to exist on all CPUs with
685  * family >= 6; unfortunately, that's not true in practice because
686  * of early VIA chips and (more importantly) broken virtualizers that
687  * are not easy to detect.  In the latter case it doesn't even *fail*
688  * reliably, so probing for it doesn't even work.  Disable it completely
689  * unless we can find a reliable way to detect all the broken cases.
690  */
691 static void __cpuinit detect_nopl(struct cpuinfo_x86 *c)
692 {
693 	clear_cpu_cap(c, X86_FEATURE_NOPL);
694 }
695 
696 static void __cpuinit generic_identify(struct cpuinfo_x86 *c)
697 {
698 	c->extended_cpuid_level = 0;
699 
700 	if (!have_cpuid_p())
701 		identify_cpu_without_cpuid(c);
702 
703 	/* cyrix could have cpuid enabled via c_identify()*/
704 	if (!have_cpuid_p())
705 		return;
706 
707 	cpu_detect(c);
708 
709 	get_cpu_vendor(c);
710 
711 	get_cpu_cap(c);
712 
713 	if (c->cpuid_level >= 0x00000001) {
714 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
715 #ifdef CONFIG_X86_32
716 # ifdef CONFIG_X86_HT
717 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
718 # else
719 		c->apicid = c->initial_apicid;
720 # endif
721 #endif
722 
723 #ifdef CONFIG_X86_HT
724 		c->phys_proc_id = c->initial_apicid;
725 #endif
726 	}
727 
728 	get_model_name(c); /* Default name */
729 
730 	init_scattered_cpuid_features(c);
731 	detect_nopl(c);
732 }
733 
734 /*
735  * This does the hard work of actually picking apart the CPU stuff...
736  */
737 static void __cpuinit identify_cpu(struct cpuinfo_x86 *c)
738 {
739 	int i;
740 
741 	c->loops_per_jiffy = loops_per_jiffy;
742 	c->x86_cache_size = -1;
743 	c->x86_vendor = X86_VENDOR_UNKNOWN;
744 	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
745 	c->x86_vendor_id[0] = '\0'; /* Unset */
746 	c->x86_model_id[0] = '\0';  /* Unset */
747 	c->x86_max_cores = 1;
748 	c->x86_coreid_bits = 0;
749 #ifdef CONFIG_X86_64
750 	c->x86_clflush_size = 64;
751 	c->x86_phys_bits = 36;
752 	c->x86_virt_bits = 48;
753 #else
754 	c->cpuid_level = -1;	/* CPUID not detected */
755 	c->x86_clflush_size = 32;
756 	c->x86_phys_bits = 32;
757 	c->x86_virt_bits = 32;
758 #endif
759 	c->x86_cache_alignment = c->x86_clflush_size;
760 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
761 
762 	generic_identify(c);
763 
764 	if (this_cpu->c_identify)
765 		this_cpu->c_identify(c);
766 
767 	/* Clear/Set all flags overriden by options, after probe */
768 	for (i = 0; i < NCAPINTS; i++) {
769 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
770 		c->x86_capability[i] |= cpu_caps_set[i];
771 	}
772 
773 #ifdef CONFIG_X86_64
774 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
775 #endif
776 
777 	/*
778 	 * Vendor-specific initialization.  In this section we
779 	 * canonicalize the feature flags, meaning if there are
780 	 * features a certain CPU supports which CPUID doesn't
781 	 * tell us, CPUID claiming incorrect flags, or other bugs,
782 	 * we handle them here.
783 	 *
784 	 * At the end of this section, c->x86_capability better
785 	 * indicate the features this CPU genuinely supports!
786 	 */
787 	if (this_cpu->c_init)
788 		this_cpu->c_init(c);
789 
790 	/* Disable the PN if appropriate */
791 	squash_the_stupid_serial_number(c);
792 
793 	/*
794 	 * The vendor-specific functions might have changed features.
795 	 * Now we do "generic changes."
796 	 */
797 
798 	/* Filter out anything that depends on CPUID levels we don't have */
799 	filter_cpuid_features(c, true);
800 
801 	/* If the model name is still unset, do table lookup. */
802 	if (!c->x86_model_id[0]) {
803 		const char *p;
804 		p = table_lookup_model(c);
805 		if (p)
806 			strcpy(c->x86_model_id, p);
807 		else
808 			/* Last resort... */
809 			sprintf(c->x86_model_id, "%02x/%02x",
810 				c->x86, c->x86_model);
811 	}
812 
813 #ifdef CONFIG_X86_64
814 	detect_ht(c);
815 #endif
816 
817 	init_hypervisor(c);
818 
819 	/*
820 	 * Clear/Set all flags overriden by options, need do it
821 	 * before following smp all cpus cap AND.
822 	 */
823 	for (i = 0; i < NCAPINTS; i++) {
824 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
825 		c->x86_capability[i] |= cpu_caps_set[i];
826 	}
827 
828 	/*
829 	 * On SMP, boot_cpu_data holds the common feature set between
830 	 * all CPUs; so make sure that we indicate which features are
831 	 * common between the CPUs.  The first time this routine gets
832 	 * executed, c == &boot_cpu_data.
833 	 */
834 	if (c != &boot_cpu_data) {
835 		/* AND the already accumulated flags with these */
836 		for (i = 0; i < NCAPINTS; i++)
837 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
838 	}
839 
840 #ifdef CONFIG_X86_MCE
841 	/* Init Machine Check Exception if available. */
842 	mcheck_init(c);
843 #endif
844 
845 	select_idle_routine(c);
846 
847 #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
848 	numa_add_cpu(smp_processor_id());
849 #endif
850 }
851 
852 #ifdef CONFIG_X86_64
853 static void vgetcpu_set_mode(void)
854 {
855 	if (cpu_has(&boot_cpu_data, X86_FEATURE_RDTSCP))
856 		vgetcpu_mode = VGETCPU_RDTSCP;
857 	else
858 		vgetcpu_mode = VGETCPU_LSL;
859 }
860 #endif
861 
862 void __init identify_boot_cpu(void)
863 {
864 	identify_cpu(&boot_cpu_data);
865 	init_c1e_mask();
866 #ifdef CONFIG_X86_32
867 	sysenter_setup();
868 	enable_sep_cpu();
869 #else
870 	vgetcpu_set_mode();
871 #endif
872 	init_hw_perf_counters();
873 }
874 
875 void __cpuinit identify_secondary_cpu(struct cpuinfo_x86 *c)
876 {
877 	BUG_ON(c == &boot_cpu_data);
878 	identify_cpu(c);
879 #ifdef CONFIG_X86_32
880 	enable_sep_cpu();
881 #endif
882 	mtrr_ap_init();
883 }
884 
885 struct msr_range {
886 	unsigned	min;
887 	unsigned	max;
888 };
889 
890 static const struct msr_range msr_range_array[] __cpuinitconst = {
891 	{ 0x00000000, 0x00000418},
892 	{ 0xc0000000, 0xc000040b},
893 	{ 0xc0010000, 0xc0010142},
894 	{ 0xc0011000, 0xc001103b},
895 };
896 
897 static void __cpuinit print_cpu_msr(void)
898 {
899 	unsigned index_min, index_max;
900 	unsigned index;
901 	u64 val;
902 	int i;
903 
904 	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
905 		index_min = msr_range_array[i].min;
906 		index_max = msr_range_array[i].max;
907 
908 		for (index = index_min; index < index_max; index++) {
909 			if (rdmsrl_amd_safe(index, &val))
910 				continue;
911 			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
912 		}
913 	}
914 }
915 
916 static int show_msr __cpuinitdata;
917 
918 static __init int setup_show_msr(char *arg)
919 {
920 	int num;
921 
922 	get_option(&arg, &num);
923 
924 	if (num > 0)
925 		show_msr = num;
926 	return 1;
927 }
928 __setup("show_msr=", setup_show_msr);
929 
930 static __init int setup_noclflush(char *arg)
931 {
932 	setup_clear_cpu_cap(X86_FEATURE_CLFLSH);
933 	return 1;
934 }
935 __setup("noclflush", setup_noclflush);
936 
937 void __cpuinit print_cpu_info(struct cpuinfo_x86 *c)
938 {
939 	const char *vendor = NULL;
940 
941 	if (c->x86_vendor < X86_VENDOR_NUM) {
942 		vendor = this_cpu->c_vendor;
943 	} else {
944 		if (c->cpuid_level >= 0)
945 			vendor = c->x86_vendor_id;
946 	}
947 
948 	if (vendor && !strstr(c->x86_model_id, vendor))
949 		printk(KERN_CONT "%s ", vendor);
950 
951 	if (c->x86_model_id[0])
952 		printk(KERN_CONT "%s", c->x86_model_id);
953 	else
954 		printk(KERN_CONT "%d86", c->x86);
955 
956 	if (c->x86_mask || c->cpuid_level >= 0)
957 		printk(KERN_CONT " stepping %02x\n", c->x86_mask);
958 	else
959 		printk(KERN_CONT "\n");
960 
961 #ifdef CONFIG_SMP
962 	if (c->cpu_index < show_msr)
963 		print_cpu_msr();
964 #else
965 	if (show_msr)
966 		print_cpu_msr();
967 #endif
968 }
969 
970 static __init int setup_disablecpuid(char *arg)
971 {
972 	int bit;
973 
974 	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
975 		setup_clear_cpu_cap(bit);
976 	else
977 		return 0;
978 
979 	return 1;
980 }
981 __setup("clearcpuid=", setup_disablecpuid);
982 
983 #ifdef CONFIG_X86_64
984 struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
985 
986 DEFINE_PER_CPU_FIRST(union irq_stack_union,
987 		     irq_stack_union) __aligned(PAGE_SIZE);
988 
989 /*
990  * The following four percpu variables are hot.  Align current_task to
991  * cacheline size such that all four fall in the same cacheline.
992  */
993 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
994 	&init_task;
995 EXPORT_PER_CPU_SYMBOL(current_task);
996 
997 DEFINE_PER_CPU(unsigned long, kernel_stack) =
998 	(unsigned long)&init_thread_union - KERNEL_STACK_OFFSET + THREAD_SIZE;
999 EXPORT_PER_CPU_SYMBOL(kernel_stack);
1000 
1001 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1002 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1003 
1004 DEFINE_PER_CPU(unsigned int, irq_count) = -1;
1005 
1006 /*
1007  * Special IST stacks which the CPU switches to when it calls
1008  * an IST-marked descriptor entry. Up to 7 stacks (hardware
1009  * limit), all of them are 4K, except the debug stack which
1010  * is 8K.
1011  */
1012 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1013 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1014 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1015 };
1016 
1017 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1018 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1019 
1020 /* May not be marked __init: used by software suspend */
1021 void syscall_init(void)
1022 {
1023 	/*
1024 	 * LSTAR and STAR live in a bit strange symbiosis.
1025 	 * They both write to the same internal register. STAR allows to
1026 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1027 	 */
1028 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1029 	wrmsrl(MSR_LSTAR, system_call);
1030 	wrmsrl(MSR_CSTAR, ignore_sysret);
1031 
1032 #ifdef CONFIG_IA32_EMULATION
1033 	syscall32_cpu_init();
1034 #endif
1035 
1036 	/* Flags to clear on syscall */
1037 	wrmsrl(MSR_SYSCALL_MASK,
1038 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|X86_EFLAGS_IOPL);
1039 }
1040 
1041 unsigned long kernel_eflags;
1042 
1043 /*
1044  * Copies of the original ist values from the tss are only accessed during
1045  * debugging, no special alignment required.
1046  */
1047 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1048 
1049 #else	/* CONFIG_X86_64 */
1050 
1051 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1052 EXPORT_PER_CPU_SYMBOL(current_task);
1053 
1054 #ifdef CONFIG_CC_STACKPROTECTOR
1055 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1056 #endif
1057 
1058 /* Make sure %fs and %gs are initialized properly in idle threads */
1059 struct pt_regs * __cpuinit idle_regs(struct pt_regs *regs)
1060 {
1061 	memset(regs, 0, sizeof(struct pt_regs));
1062 	regs->fs = __KERNEL_PERCPU;
1063 	regs->gs = __KERNEL_STACK_CANARY;
1064 
1065 	return regs;
1066 }
1067 #endif	/* CONFIG_X86_64 */
1068 
1069 /*
1070  * Clear all 6 debug registers:
1071  */
1072 static void clear_all_debug_regs(void)
1073 {
1074 	int i;
1075 
1076 	for (i = 0; i < 8; i++) {
1077 		/* Ignore db4, db5 */
1078 		if ((i == 4) || (i == 5))
1079 			continue;
1080 
1081 		set_debugreg(0, i);
1082 	}
1083 }
1084 
1085 /*
1086  * cpu_init() initializes state that is per-CPU. Some data is already
1087  * initialized (naturally) in the bootstrap process, such as the GDT
1088  * and IDT. We reload them nevertheless, this function acts as a
1089  * 'CPU state barrier', nothing should get across.
1090  * A lot of state is already set up in PDA init for 64 bit
1091  */
1092 #ifdef CONFIG_X86_64
1093 
1094 void __cpuinit cpu_init(void)
1095 {
1096 	struct orig_ist *orig_ist;
1097 	struct task_struct *me;
1098 	struct tss_struct *t;
1099 	unsigned long v;
1100 	int cpu;
1101 	int i;
1102 
1103 	cpu = stack_smp_processor_id();
1104 	t = &per_cpu(init_tss, cpu);
1105 	orig_ist = &per_cpu(orig_ist, cpu);
1106 
1107 #ifdef CONFIG_NUMA
1108 	if (cpu != 0 && percpu_read(node_number) == 0 &&
1109 	    cpu_to_node(cpu) != NUMA_NO_NODE)
1110 		percpu_write(node_number, cpu_to_node(cpu));
1111 #endif
1112 
1113 	me = current;
1114 
1115 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask))
1116 		panic("CPU#%d already initialized!\n", cpu);
1117 
1118 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1119 
1120 	clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1121 
1122 	/*
1123 	 * Initialize the per-CPU GDT with the boot GDT,
1124 	 * and set up the GDT descriptor:
1125 	 */
1126 
1127 	switch_to_new_gdt(cpu);
1128 	loadsegment(fs, 0);
1129 
1130 	load_idt((const struct desc_ptr *)&idt_descr);
1131 
1132 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1133 	syscall_init();
1134 
1135 	wrmsrl(MSR_FS_BASE, 0);
1136 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1137 	barrier();
1138 
1139 	check_efer();
1140 	if (cpu != 0)
1141 		enable_x2apic();
1142 
1143 	/*
1144 	 * set up and load the per-CPU TSS
1145 	 */
1146 	if (!orig_ist->ist[0]) {
1147 		char *estacks = per_cpu(exception_stacks, cpu);
1148 
1149 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1150 			estacks += exception_stack_sizes[v];
1151 			orig_ist->ist[v] = t->x86_tss.ist[v] =
1152 					(unsigned long)estacks;
1153 		}
1154 	}
1155 
1156 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1157 
1158 	/*
1159 	 * <= is required because the CPU will access up to
1160 	 * 8 bits beyond the end of the IO permission bitmap.
1161 	 */
1162 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1163 		t->io_bitmap[i] = ~0UL;
1164 
1165 	atomic_inc(&init_mm.mm_count);
1166 	me->active_mm = &init_mm;
1167 	BUG_ON(me->mm);
1168 	enter_lazy_tlb(&init_mm, me);
1169 
1170 	load_sp0(t, &current->thread);
1171 	set_tss_desc(cpu, t);
1172 	load_TR_desc();
1173 	load_LDT(&init_mm.context);
1174 
1175 #ifdef CONFIG_KGDB
1176 	/*
1177 	 * If the kgdb is connected no debug regs should be altered.  This
1178 	 * is only applicable when KGDB and a KGDB I/O module are built
1179 	 * into the kernel and you are using early debugging with
1180 	 * kgdbwait. KGDB will control the kernel HW breakpoint registers.
1181 	 */
1182 	if (kgdb_connected && arch_kgdb_ops.correct_hw_break)
1183 		arch_kgdb_ops.correct_hw_break();
1184 	else
1185 #endif
1186 		clear_all_debug_regs();
1187 
1188 	fpu_init();
1189 
1190 	raw_local_save_flags(kernel_eflags);
1191 
1192 	if (is_uv_system())
1193 		uv_cpu_init();
1194 }
1195 
1196 #else
1197 
1198 void __cpuinit cpu_init(void)
1199 {
1200 	int cpu = smp_processor_id();
1201 	struct task_struct *curr = current;
1202 	struct tss_struct *t = &per_cpu(init_tss, cpu);
1203 	struct thread_struct *thread = &curr->thread;
1204 
1205 	if (cpumask_test_and_set_cpu(cpu, cpu_initialized_mask)) {
1206 		printk(KERN_WARNING "CPU#%d already initialized!\n", cpu);
1207 		for (;;)
1208 			local_irq_enable();
1209 	}
1210 
1211 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1212 
1213 	if (cpu_has_vme || cpu_has_tsc || cpu_has_de)
1214 		clear_in_cr4(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1215 
1216 	load_idt(&idt_descr);
1217 	switch_to_new_gdt(cpu);
1218 
1219 	/*
1220 	 * Set up and load the per-CPU TSS and LDT
1221 	 */
1222 	atomic_inc(&init_mm.mm_count);
1223 	curr->active_mm = &init_mm;
1224 	BUG_ON(curr->mm);
1225 	enter_lazy_tlb(&init_mm, curr);
1226 
1227 	load_sp0(t, thread);
1228 	set_tss_desc(cpu, t);
1229 	load_TR_desc();
1230 	load_LDT(&init_mm.context);
1231 
1232 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1233 
1234 #ifdef CONFIG_DOUBLEFAULT
1235 	/* Set up doublefault TSS pointer in the GDT */
1236 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1237 #endif
1238 
1239 	clear_all_debug_regs();
1240 
1241 	/*
1242 	 * Force FPU initialization:
1243 	 */
1244 	if (cpu_has_xsave)
1245 		current_thread_info()->status = TS_XSAVE;
1246 	else
1247 		current_thread_info()->status = 0;
1248 	clear_used_math();
1249 	mxcsr_feature_mask_init();
1250 
1251 	/*
1252 	 * Boot processor to setup the FP and extended state context info.
1253 	 */
1254 	if (smp_processor_id() == boot_cpu_id)
1255 		init_thread_xstate();
1256 
1257 	xsave_init();
1258 }
1259 #endif
1260