xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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/ctype.h>
9 #include <linux/delay.h>
10 #include <linux/sched.h>
11 #include <linux/init.h>
12 #include <linux/kprobes.h>
13 #include <linux/kgdb.h>
14 #include <linux/smp.h>
15 #include <linux/io.h>
16 #include <linux/syscore_ops.h>
17 
18 #include <asm/stackprotector.h>
19 #include <asm/perf_event.h>
20 #include <asm/mmu_context.h>
21 #include <asm/archrandom.h>
22 #include <asm/hypervisor.h>
23 #include <asm/processor.h>
24 #include <asm/tlbflush.h>
25 #include <asm/debugreg.h>
26 #include <asm/sections.h>
27 #include <asm/vsyscall.h>
28 #include <linux/topology.h>
29 #include <linux/cpumask.h>
30 #include <asm/pgtable.h>
31 #include <linux/atomic.h>
32 #include <asm/proto.h>
33 #include <asm/setup.h>
34 #include <asm/apic.h>
35 #include <asm/desc.h>
36 #include <asm/fpu/internal.h>
37 #include <asm/mtrr.h>
38 #include <linux/numa.h>
39 #include <asm/asm.h>
40 #include <asm/cpu.h>
41 #include <asm/mce.h>
42 #include <asm/msr.h>
43 #include <asm/pat.h>
44 #include <asm/microcode.h>
45 #include <asm/microcode_intel.h>
46 
47 #ifdef CONFIG_X86_LOCAL_APIC
48 #include <asm/uv/uv.h>
49 #endif
50 
51 #include "cpu.h"
52 
53 /* all of these masks are initialized in setup_cpu_local_masks() */
54 cpumask_var_t cpu_initialized_mask;
55 cpumask_var_t cpu_callout_mask;
56 cpumask_var_t cpu_callin_mask;
57 
58 /* representing cpus for which sibling maps can be computed */
59 cpumask_var_t cpu_sibling_setup_mask;
60 
61 /* correctly size the local cpu masks */
62 void __init setup_cpu_local_masks(void)
63 {
64 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
65 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
66 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
67 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
68 }
69 
70 static void default_init(struct cpuinfo_x86 *c)
71 {
72 #ifdef CONFIG_X86_64
73 	cpu_detect_cache_sizes(c);
74 #else
75 	/* Not much we can do here... */
76 	/* Check if at least it has cpuid */
77 	if (c->cpuid_level == -1) {
78 		/* No cpuid. It must be an ancient CPU */
79 		if (c->x86 == 4)
80 			strcpy(c->x86_model_id, "486");
81 		else if (c->x86 == 3)
82 			strcpy(c->x86_model_id, "386");
83 	}
84 #endif
85 }
86 
87 static const struct cpu_dev default_cpu = {
88 	.c_init		= default_init,
89 	.c_vendor	= "Unknown",
90 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
91 };
92 
93 static const struct cpu_dev *this_cpu = &default_cpu;
94 
95 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
96 #ifdef CONFIG_X86_64
97 	/*
98 	 * We need valid kernel segments for data and code in long mode too
99 	 * IRET will check the segment types  kkeil 2000/10/28
100 	 * Also sysret mandates a special GDT layout
101 	 *
102 	 * TLS descriptors are currently at a different place compared to i386.
103 	 * Hopefully nobody expects them at a fixed place (Wine?)
104 	 */
105 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
106 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
107 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
108 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
109 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
110 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
111 #else
112 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
113 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
114 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
115 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
116 	/*
117 	 * Segments used for calling PnP BIOS have byte granularity.
118 	 * They code segments and data segments have fixed 64k limits,
119 	 * the transfer segment sizes are set at run time.
120 	 */
121 	/* 32-bit code */
122 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
123 	/* 16-bit code */
124 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
125 	/* 16-bit data */
126 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
127 	/* 16-bit data */
128 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
129 	/* 16-bit data */
130 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
131 	/*
132 	 * The APM segments have byte granularity and their bases
133 	 * are set at run time.  All have 64k limits.
134 	 */
135 	/* 32-bit code */
136 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
137 	/* 16-bit code */
138 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
139 	/* data */
140 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
141 
142 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
143 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
144 	GDT_STACK_CANARY_INIT
145 #endif
146 } };
147 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
148 
149 static int __init x86_mpx_setup(char *s)
150 {
151 	/* require an exact match without trailing characters */
152 	if (strlen(s))
153 		return 0;
154 
155 	/* do not emit a message if the feature is not present */
156 	if (!boot_cpu_has(X86_FEATURE_MPX))
157 		return 1;
158 
159 	setup_clear_cpu_cap(X86_FEATURE_MPX);
160 	pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
161 	return 1;
162 }
163 __setup("nompx", x86_mpx_setup);
164 
165 #ifdef CONFIG_X86_32
166 static int cachesize_override = -1;
167 static int disable_x86_serial_nr = 1;
168 
169 static int __init cachesize_setup(char *str)
170 {
171 	get_option(&str, &cachesize_override);
172 	return 1;
173 }
174 __setup("cachesize=", cachesize_setup);
175 
176 static int __init x86_sep_setup(char *s)
177 {
178 	setup_clear_cpu_cap(X86_FEATURE_SEP);
179 	return 1;
180 }
181 __setup("nosep", x86_sep_setup);
182 
183 /* Standard macro to see if a specific flag is changeable */
184 static inline int flag_is_changeable_p(u32 flag)
185 {
186 	u32 f1, f2;
187 
188 	/*
189 	 * Cyrix and IDT cpus allow disabling of CPUID
190 	 * so the code below may return different results
191 	 * when it is executed before and after enabling
192 	 * the CPUID. Add "volatile" to not allow gcc to
193 	 * optimize the subsequent calls to this function.
194 	 */
195 	asm volatile ("pushfl		\n\t"
196 		      "pushfl		\n\t"
197 		      "popl %0		\n\t"
198 		      "movl %0, %1	\n\t"
199 		      "xorl %2, %0	\n\t"
200 		      "pushl %0		\n\t"
201 		      "popfl		\n\t"
202 		      "pushfl		\n\t"
203 		      "popl %0		\n\t"
204 		      "popfl		\n\t"
205 
206 		      : "=&r" (f1), "=&r" (f2)
207 		      : "ir" (flag));
208 
209 	return ((f1^f2) & flag) != 0;
210 }
211 
212 /* Probe for the CPUID instruction */
213 int have_cpuid_p(void)
214 {
215 	return flag_is_changeable_p(X86_EFLAGS_ID);
216 }
217 
218 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
219 {
220 	unsigned long lo, hi;
221 
222 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
223 		return;
224 
225 	/* Disable processor serial number: */
226 
227 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
228 	lo |= 0x200000;
229 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
230 
231 	printk(KERN_NOTICE "CPU serial number disabled.\n");
232 	clear_cpu_cap(c, X86_FEATURE_PN);
233 
234 	/* Disabling the serial number may affect the cpuid level */
235 	c->cpuid_level = cpuid_eax(0);
236 }
237 
238 static int __init x86_serial_nr_setup(char *s)
239 {
240 	disable_x86_serial_nr = 0;
241 	return 1;
242 }
243 __setup("serialnumber", x86_serial_nr_setup);
244 #else
245 static inline int flag_is_changeable_p(u32 flag)
246 {
247 	return 1;
248 }
249 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
250 {
251 }
252 #endif
253 
254 static __init int setup_disable_smep(char *arg)
255 {
256 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
257 	return 1;
258 }
259 __setup("nosmep", setup_disable_smep);
260 
261 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
262 {
263 	if (cpu_has(c, X86_FEATURE_SMEP))
264 		cr4_set_bits(X86_CR4_SMEP);
265 }
266 
267 static __init int setup_disable_smap(char *arg)
268 {
269 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
270 	return 1;
271 }
272 __setup("nosmap", setup_disable_smap);
273 
274 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
275 {
276 	unsigned long eflags;
277 
278 	/* This should have been cleared long ago */
279 	raw_local_save_flags(eflags);
280 	BUG_ON(eflags & X86_EFLAGS_AC);
281 
282 	if (cpu_has(c, X86_FEATURE_SMAP)) {
283 #ifdef CONFIG_X86_SMAP
284 		cr4_set_bits(X86_CR4_SMAP);
285 #else
286 		cr4_clear_bits(X86_CR4_SMAP);
287 #endif
288 	}
289 }
290 
291 /*
292  * Some CPU features depend on higher CPUID levels, which may not always
293  * be available due to CPUID level capping or broken virtualization
294  * software.  Add those features to this table to auto-disable them.
295  */
296 struct cpuid_dependent_feature {
297 	u32 feature;
298 	u32 level;
299 };
300 
301 static const struct cpuid_dependent_feature
302 cpuid_dependent_features[] = {
303 	{ X86_FEATURE_MWAIT,		0x00000005 },
304 	{ X86_FEATURE_DCA,		0x00000009 },
305 	{ X86_FEATURE_XSAVE,		0x0000000d },
306 	{ 0, 0 }
307 };
308 
309 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
310 {
311 	const struct cpuid_dependent_feature *df;
312 
313 	for (df = cpuid_dependent_features; df->feature; df++) {
314 
315 		if (!cpu_has(c, df->feature))
316 			continue;
317 		/*
318 		 * Note: cpuid_level is set to -1 if unavailable, but
319 		 * extended_extended_level is set to 0 if unavailable
320 		 * and the legitimate extended levels are all negative
321 		 * when signed; hence the weird messing around with
322 		 * signs here...
323 		 */
324 		if (!((s32)df->level < 0 ?
325 		     (u32)df->level > (u32)c->extended_cpuid_level :
326 		     (s32)df->level > (s32)c->cpuid_level))
327 			continue;
328 
329 		clear_cpu_cap(c, df->feature);
330 		if (!warn)
331 			continue;
332 
333 		printk(KERN_WARNING
334 		       "CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
335 				x86_cap_flag(df->feature), df->level);
336 	}
337 }
338 
339 /*
340  * Naming convention should be: <Name> [(<Codename>)]
341  * This table only is used unless init_<vendor>() below doesn't set it;
342  * in particular, if CPUID levels 0x80000002..4 are supported, this
343  * isn't used
344  */
345 
346 /* Look up CPU names by table lookup. */
347 static const char *table_lookup_model(struct cpuinfo_x86 *c)
348 {
349 #ifdef CONFIG_X86_32
350 	const struct legacy_cpu_model_info *info;
351 
352 	if (c->x86_model >= 16)
353 		return NULL;	/* Range check */
354 
355 	if (!this_cpu)
356 		return NULL;
357 
358 	info = this_cpu->legacy_models;
359 
360 	while (info->family) {
361 		if (info->family == c->x86)
362 			return info->model_names[c->x86_model];
363 		info++;
364 	}
365 #endif
366 	return NULL;		/* Not found */
367 }
368 
369 __u32 cpu_caps_cleared[NCAPINTS];
370 __u32 cpu_caps_set[NCAPINTS];
371 
372 void load_percpu_segment(int cpu)
373 {
374 #ifdef CONFIG_X86_32
375 	loadsegment(fs, __KERNEL_PERCPU);
376 #else
377 	loadsegment(gs, 0);
378 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
379 #endif
380 	load_stack_canary_segment();
381 }
382 
383 /*
384  * Current gdt points %fs at the "master" per-cpu area: after this,
385  * it's on the real one.
386  */
387 void switch_to_new_gdt(int cpu)
388 {
389 	struct desc_ptr gdt_descr;
390 
391 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
392 	gdt_descr.size = GDT_SIZE - 1;
393 	load_gdt(&gdt_descr);
394 	/* Reload the per-cpu base */
395 
396 	load_percpu_segment(cpu);
397 }
398 
399 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
400 
401 static void get_model_name(struct cpuinfo_x86 *c)
402 {
403 	unsigned int *v;
404 	char *p, *q, *s;
405 
406 	if (c->extended_cpuid_level < 0x80000004)
407 		return;
408 
409 	v = (unsigned int *)c->x86_model_id;
410 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
411 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
412 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
413 	c->x86_model_id[48] = 0;
414 
415 	/* Trim whitespace */
416 	p = q = s = &c->x86_model_id[0];
417 
418 	while (*p == ' ')
419 		p++;
420 
421 	while (*p) {
422 		/* Note the last non-whitespace index */
423 		if (!isspace(*p))
424 			s = q;
425 
426 		*q++ = *p++;
427 	}
428 
429 	*(s + 1) = '\0';
430 }
431 
432 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
433 {
434 	unsigned int n, dummy, ebx, ecx, edx, l2size;
435 
436 	n = c->extended_cpuid_level;
437 
438 	if (n >= 0x80000005) {
439 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
440 		c->x86_cache_size = (ecx>>24) + (edx>>24);
441 #ifdef CONFIG_X86_64
442 		/* On K8 L1 TLB is inclusive, so don't count it */
443 		c->x86_tlbsize = 0;
444 #endif
445 	}
446 
447 	if (n < 0x80000006)	/* Some chips just has a large L1. */
448 		return;
449 
450 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
451 	l2size = ecx >> 16;
452 
453 #ifdef CONFIG_X86_64
454 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
455 #else
456 	/* do processor-specific cache resizing */
457 	if (this_cpu->legacy_cache_size)
458 		l2size = this_cpu->legacy_cache_size(c, l2size);
459 
460 	/* Allow user to override all this if necessary. */
461 	if (cachesize_override != -1)
462 		l2size = cachesize_override;
463 
464 	if (l2size == 0)
465 		return;		/* Again, no L2 cache is possible */
466 #endif
467 
468 	c->x86_cache_size = l2size;
469 }
470 
471 u16 __read_mostly tlb_lli_4k[NR_INFO];
472 u16 __read_mostly tlb_lli_2m[NR_INFO];
473 u16 __read_mostly tlb_lli_4m[NR_INFO];
474 u16 __read_mostly tlb_lld_4k[NR_INFO];
475 u16 __read_mostly tlb_lld_2m[NR_INFO];
476 u16 __read_mostly tlb_lld_4m[NR_INFO];
477 u16 __read_mostly tlb_lld_1g[NR_INFO];
478 
479 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
480 {
481 	if (this_cpu->c_detect_tlb)
482 		this_cpu->c_detect_tlb(c);
483 
484 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
485 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
486 		tlb_lli_4m[ENTRIES]);
487 
488 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
489 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
490 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
491 }
492 
493 void detect_ht(struct cpuinfo_x86 *c)
494 {
495 #ifdef CONFIG_SMP
496 	u32 eax, ebx, ecx, edx;
497 	int index_msb, core_bits;
498 	static bool printed;
499 
500 	if (!cpu_has(c, X86_FEATURE_HT))
501 		return;
502 
503 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
504 		goto out;
505 
506 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
507 		return;
508 
509 	cpuid(1, &eax, &ebx, &ecx, &edx);
510 
511 	smp_num_siblings = (ebx & 0xff0000) >> 16;
512 
513 	if (smp_num_siblings == 1) {
514 		printk_once(KERN_INFO "CPU0: Hyper-Threading is disabled\n");
515 		goto out;
516 	}
517 
518 	if (smp_num_siblings <= 1)
519 		goto out;
520 
521 	index_msb = get_count_order(smp_num_siblings);
522 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
523 
524 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
525 
526 	index_msb = get_count_order(smp_num_siblings);
527 
528 	core_bits = get_count_order(c->x86_max_cores);
529 
530 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
531 				       ((1 << core_bits) - 1);
532 
533 out:
534 	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
535 		printk(KERN_INFO  "CPU: Physical Processor ID: %d\n",
536 		       c->phys_proc_id);
537 		printk(KERN_INFO  "CPU: Processor Core ID: %d\n",
538 		       c->cpu_core_id);
539 		printed = 1;
540 	}
541 #endif
542 }
543 
544 static void get_cpu_vendor(struct cpuinfo_x86 *c)
545 {
546 	char *v = c->x86_vendor_id;
547 	int i;
548 
549 	for (i = 0; i < X86_VENDOR_NUM; i++) {
550 		if (!cpu_devs[i])
551 			break;
552 
553 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
554 		    (cpu_devs[i]->c_ident[1] &&
555 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
556 
557 			this_cpu = cpu_devs[i];
558 			c->x86_vendor = this_cpu->c_x86_vendor;
559 			return;
560 		}
561 	}
562 
563 	printk_once(KERN_ERR
564 			"CPU: vendor_id '%s' unknown, using generic init.\n" \
565 			"CPU: Your system may be unstable.\n", v);
566 
567 	c->x86_vendor = X86_VENDOR_UNKNOWN;
568 	this_cpu = &default_cpu;
569 }
570 
571 void cpu_detect(struct cpuinfo_x86 *c)
572 {
573 	/* Get vendor name */
574 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
575 	      (unsigned int *)&c->x86_vendor_id[0],
576 	      (unsigned int *)&c->x86_vendor_id[8],
577 	      (unsigned int *)&c->x86_vendor_id[4]);
578 
579 	c->x86 = 4;
580 	/* Intel-defined flags: level 0x00000001 */
581 	if (c->cpuid_level >= 0x00000001) {
582 		u32 junk, tfms, cap0, misc;
583 
584 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
585 		c->x86 = (tfms >> 8) & 0xf;
586 		c->x86_model = (tfms >> 4) & 0xf;
587 		c->x86_mask = tfms & 0xf;
588 
589 		if (c->x86 == 0xf)
590 			c->x86 += (tfms >> 20) & 0xff;
591 		if (c->x86 >= 0x6)
592 			c->x86_model += ((tfms >> 16) & 0xf) << 4;
593 
594 		if (cap0 & (1<<19)) {
595 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
596 			c->x86_cache_alignment = c->x86_clflush_size;
597 		}
598 	}
599 }
600 
601 void get_cpu_cap(struct cpuinfo_x86 *c)
602 {
603 	u32 tfms, xlvl;
604 	u32 ebx;
605 
606 	/* Intel-defined flags: level 0x00000001 */
607 	if (c->cpuid_level >= 0x00000001) {
608 		u32 capability, excap;
609 
610 		cpuid(0x00000001, &tfms, &ebx, &excap, &capability);
611 		c->x86_capability[0] = capability;
612 		c->x86_capability[4] = excap;
613 	}
614 
615 	/* Additional Intel-defined flags: level 0x00000007 */
616 	if (c->cpuid_level >= 0x00000007) {
617 		u32 eax, ebx, ecx, edx;
618 
619 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
620 
621 		c->x86_capability[9] = ebx;
622 	}
623 
624 	/* Extended state features: level 0x0000000d */
625 	if (c->cpuid_level >= 0x0000000d) {
626 		u32 eax, ebx, ecx, edx;
627 
628 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
629 
630 		c->x86_capability[10] = eax;
631 	}
632 
633 	/* Additional Intel-defined flags: level 0x0000000F */
634 	if (c->cpuid_level >= 0x0000000F) {
635 		u32 eax, ebx, ecx, edx;
636 
637 		/* QoS sub-leaf, EAX=0Fh, ECX=0 */
638 		cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx);
639 		c->x86_capability[11] = edx;
640 		if (cpu_has(c, X86_FEATURE_CQM_LLC)) {
641 			/* will be overridden if occupancy monitoring exists */
642 			c->x86_cache_max_rmid = ebx;
643 
644 			/* QoS sub-leaf, EAX=0Fh, ECX=1 */
645 			cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx);
646 			c->x86_capability[12] = edx;
647 			if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) {
648 				c->x86_cache_max_rmid = ecx;
649 				c->x86_cache_occ_scale = ebx;
650 			}
651 		} else {
652 			c->x86_cache_max_rmid = -1;
653 			c->x86_cache_occ_scale = -1;
654 		}
655 	}
656 
657 	/* AMD-defined flags: level 0x80000001 */
658 	xlvl = cpuid_eax(0x80000000);
659 	c->extended_cpuid_level = xlvl;
660 
661 	if ((xlvl & 0xffff0000) == 0x80000000) {
662 		if (xlvl >= 0x80000001) {
663 			c->x86_capability[1] = cpuid_edx(0x80000001);
664 			c->x86_capability[6] = cpuid_ecx(0x80000001);
665 		}
666 	}
667 
668 	if (c->extended_cpuid_level >= 0x80000008) {
669 		u32 eax = cpuid_eax(0x80000008);
670 
671 		c->x86_virt_bits = (eax >> 8) & 0xff;
672 		c->x86_phys_bits = eax & 0xff;
673 	}
674 #ifdef CONFIG_X86_32
675 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
676 		c->x86_phys_bits = 36;
677 #endif
678 
679 	if (c->extended_cpuid_level >= 0x80000007)
680 		c->x86_power = cpuid_edx(0x80000007);
681 
682 	init_scattered_cpuid_features(c);
683 }
684 
685 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
686 {
687 #ifdef CONFIG_X86_32
688 	int i;
689 
690 	/*
691 	 * First of all, decide if this is a 486 or higher
692 	 * It's a 486 if we can modify the AC flag
693 	 */
694 	if (flag_is_changeable_p(X86_EFLAGS_AC))
695 		c->x86 = 4;
696 	else
697 		c->x86 = 3;
698 
699 	for (i = 0; i < X86_VENDOR_NUM; i++)
700 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
701 			c->x86_vendor_id[0] = 0;
702 			cpu_devs[i]->c_identify(c);
703 			if (c->x86_vendor_id[0]) {
704 				get_cpu_vendor(c);
705 				break;
706 			}
707 		}
708 #endif
709 }
710 
711 /*
712  * Do minimum CPU detection early.
713  * Fields really needed: vendor, cpuid_level, family, model, mask,
714  * cache alignment.
715  * The others are not touched to avoid unwanted side effects.
716  *
717  * WARNING: this function is only called on the BP.  Don't add code here
718  * that is supposed to run on all CPUs.
719  */
720 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
721 {
722 #ifdef CONFIG_X86_64
723 	c->x86_clflush_size = 64;
724 	c->x86_phys_bits = 36;
725 	c->x86_virt_bits = 48;
726 #else
727 	c->x86_clflush_size = 32;
728 	c->x86_phys_bits = 32;
729 	c->x86_virt_bits = 32;
730 #endif
731 	c->x86_cache_alignment = c->x86_clflush_size;
732 
733 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
734 	c->extended_cpuid_level = 0;
735 
736 	if (!have_cpuid_p())
737 		identify_cpu_without_cpuid(c);
738 
739 	/* cyrix could have cpuid enabled via c_identify()*/
740 	if (!have_cpuid_p())
741 		return;
742 
743 	cpu_detect(c);
744 	get_cpu_vendor(c);
745 	get_cpu_cap(c);
746 
747 	if (this_cpu->c_early_init)
748 		this_cpu->c_early_init(c);
749 
750 	c->cpu_index = 0;
751 	filter_cpuid_features(c, false);
752 
753 	if (this_cpu->c_bsp_init)
754 		this_cpu->c_bsp_init(c);
755 
756 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
757 	fpu__init_system(c);
758 }
759 
760 void __init early_cpu_init(void)
761 {
762 	const struct cpu_dev *const *cdev;
763 	int count = 0;
764 
765 #ifdef CONFIG_PROCESSOR_SELECT
766 	printk(KERN_INFO "KERNEL supported cpus:\n");
767 #endif
768 
769 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
770 		const struct cpu_dev *cpudev = *cdev;
771 
772 		if (count >= X86_VENDOR_NUM)
773 			break;
774 		cpu_devs[count] = cpudev;
775 		count++;
776 
777 #ifdef CONFIG_PROCESSOR_SELECT
778 		{
779 			unsigned int j;
780 
781 			for (j = 0; j < 2; j++) {
782 				if (!cpudev->c_ident[j])
783 					continue;
784 				printk(KERN_INFO "  %s %s\n", cpudev->c_vendor,
785 					cpudev->c_ident[j]);
786 			}
787 		}
788 #endif
789 	}
790 	early_identify_cpu(&boot_cpu_data);
791 }
792 
793 /*
794  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
795  * unfortunately, that's not true in practice because of early VIA
796  * chips and (more importantly) broken virtualizers that are not easy
797  * to detect. In the latter case it doesn't even *fail* reliably, so
798  * probing for it doesn't even work. Disable it completely on 32-bit
799  * unless we can find a reliable way to detect all the broken cases.
800  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
801  */
802 static void detect_nopl(struct cpuinfo_x86 *c)
803 {
804 #ifdef CONFIG_X86_32
805 	clear_cpu_cap(c, X86_FEATURE_NOPL);
806 #else
807 	set_cpu_cap(c, X86_FEATURE_NOPL);
808 #endif
809 }
810 
811 static void generic_identify(struct cpuinfo_x86 *c)
812 {
813 	c->extended_cpuid_level = 0;
814 
815 	if (!have_cpuid_p())
816 		identify_cpu_without_cpuid(c);
817 
818 	/* cyrix could have cpuid enabled via c_identify()*/
819 	if (!have_cpuid_p())
820 		return;
821 
822 	cpu_detect(c);
823 
824 	get_cpu_vendor(c);
825 
826 	get_cpu_cap(c);
827 
828 	if (c->cpuid_level >= 0x00000001) {
829 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
830 #ifdef CONFIG_X86_32
831 # ifdef CONFIG_SMP
832 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
833 # else
834 		c->apicid = c->initial_apicid;
835 # endif
836 #endif
837 		c->phys_proc_id = c->initial_apicid;
838 	}
839 
840 	get_model_name(c); /* Default name */
841 
842 	detect_nopl(c);
843 }
844 
845 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
846 {
847 	/*
848 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
849 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
850 	 * in case CQM bits really aren't there in this CPU.
851 	 */
852 	if (c != &boot_cpu_data) {
853 		boot_cpu_data.x86_cache_max_rmid =
854 			min(boot_cpu_data.x86_cache_max_rmid,
855 			    c->x86_cache_max_rmid);
856 	}
857 }
858 
859 /*
860  * This does the hard work of actually picking apart the CPU stuff...
861  */
862 static void identify_cpu(struct cpuinfo_x86 *c)
863 {
864 	int i;
865 
866 	c->loops_per_jiffy = loops_per_jiffy;
867 	c->x86_cache_size = -1;
868 	c->x86_vendor = X86_VENDOR_UNKNOWN;
869 	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
870 	c->x86_vendor_id[0] = '\0'; /* Unset */
871 	c->x86_model_id[0] = '\0';  /* Unset */
872 	c->x86_max_cores = 1;
873 	c->x86_coreid_bits = 0;
874 #ifdef CONFIG_X86_64
875 	c->x86_clflush_size = 64;
876 	c->x86_phys_bits = 36;
877 	c->x86_virt_bits = 48;
878 #else
879 	c->cpuid_level = -1;	/* CPUID not detected */
880 	c->x86_clflush_size = 32;
881 	c->x86_phys_bits = 32;
882 	c->x86_virt_bits = 32;
883 #endif
884 	c->x86_cache_alignment = c->x86_clflush_size;
885 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
886 
887 	generic_identify(c);
888 
889 	if (this_cpu->c_identify)
890 		this_cpu->c_identify(c);
891 
892 	/* Clear/Set all flags overriden by options, after probe */
893 	for (i = 0; i < NCAPINTS; i++) {
894 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
895 		c->x86_capability[i] |= cpu_caps_set[i];
896 	}
897 
898 #ifdef CONFIG_X86_64
899 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
900 #endif
901 
902 	/*
903 	 * Vendor-specific initialization.  In this section we
904 	 * canonicalize the feature flags, meaning if there are
905 	 * features a certain CPU supports which CPUID doesn't
906 	 * tell us, CPUID claiming incorrect flags, or other bugs,
907 	 * we handle them here.
908 	 *
909 	 * At the end of this section, c->x86_capability better
910 	 * indicate the features this CPU genuinely supports!
911 	 */
912 	if (this_cpu->c_init)
913 		this_cpu->c_init(c);
914 
915 	/* Disable the PN if appropriate */
916 	squash_the_stupid_serial_number(c);
917 
918 	/* Set up SMEP/SMAP */
919 	setup_smep(c);
920 	setup_smap(c);
921 
922 	/*
923 	 * The vendor-specific functions might have changed features.
924 	 * Now we do "generic changes."
925 	 */
926 
927 	/* Filter out anything that depends on CPUID levels we don't have */
928 	filter_cpuid_features(c, true);
929 
930 	/* If the model name is still unset, do table lookup. */
931 	if (!c->x86_model_id[0]) {
932 		const char *p;
933 		p = table_lookup_model(c);
934 		if (p)
935 			strcpy(c->x86_model_id, p);
936 		else
937 			/* Last resort... */
938 			sprintf(c->x86_model_id, "%02x/%02x",
939 				c->x86, c->x86_model);
940 	}
941 
942 #ifdef CONFIG_X86_64
943 	detect_ht(c);
944 #endif
945 
946 	init_hypervisor(c);
947 	x86_init_rdrand(c);
948 	x86_init_cache_qos(c);
949 
950 	/*
951 	 * Clear/Set all flags overriden by options, need do it
952 	 * before following smp all cpus cap AND.
953 	 */
954 	for (i = 0; i < NCAPINTS; i++) {
955 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
956 		c->x86_capability[i] |= cpu_caps_set[i];
957 	}
958 
959 	/*
960 	 * On SMP, boot_cpu_data holds the common feature set between
961 	 * all CPUs; so make sure that we indicate which features are
962 	 * common between the CPUs.  The first time this routine gets
963 	 * executed, c == &boot_cpu_data.
964 	 */
965 	if (c != &boot_cpu_data) {
966 		/* AND the already accumulated flags with these */
967 		for (i = 0; i < NCAPINTS; i++)
968 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
969 
970 		/* OR, i.e. replicate the bug flags */
971 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
972 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
973 	}
974 
975 	/* Init Machine Check Exception if available. */
976 	mcheck_cpu_init(c);
977 
978 	select_idle_routine(c);
979 
980 #ifdef CONFIG_NUMA
981 	numa_add_cpu(smp_processor_id());
982 #endif
983 }
984 
985 /*
986  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
987  * on 32-bit kernels:
988  */
989 #ifdef CONFIG_X86_32
990 void enable_sep_cpu(void)
991 {
992 	struct tss_struct *tss;
993 	int cpu;
994 
995 	cpu = get_cpu();
996 	tss = &per_cpu(cpu_tss, cpu);
997 
998 	if (!boot_cpu_has(X86_FEATURE_SEP))
999 		goto out;
1000 
1001 	/*
1002 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1003 	 * see the big comment in struct x86_hw_tss's definition.
1004 	 */
1005 
1006 	tss->x86_tss.ss1 = __KERNEL_CS;
1007 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1008 
1009 	wrmsr(MSR_IA32_SYSENTER_ESP,
1010 	      (unsigned long)tss + offsetofend(struct tss_struct, SYSENTER_stack),
1011 	      0);
1012 
1013 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1014 
1015 out:
1016 	put_cpu();
1017 }
1018 #endif
1019 
1020 void __init identify_boot_cpu(void)
1021 {
1022 	identify_cpu(&boot_cpu_data);
1023 	init_amd_e400_c1e_mask();
1024 #ifdef CONFIG_X86_32
1025 	sysenter_setup();
1026 	enable_sep_cpu();
1027 #endif
1028 	cpu_detect_tlb(&boot_cpu_data);
1029 }
1030 
1031 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1032 {
1033 	BUG_ON(c == &boot_cpu_data);
1034 	identify_cpu(c);
1035 #ifdef CONFIG_X86_32
1036 	enable_sep_cpu();
1037 #endif
1038 	mtrr_ap_init();
1039 }
1040 
1041 struct msr_range {
1042 	unsigned	min;
1043 	unsigned	max;
1044 };
1045 
1046 static const struct msr_range msr_range_array[] = {
1047 	{ 0x00000000, 0x00000418},
1048 	{ 0xc0000000, 0xc000040b},
1049 	{ 0xc0010000, 0xc0010142},
1050 	{ 0xc0011000, 0xc001103b},
1051 };
1052 
1053 static void __print_cpu_msr(void)
1054 {
1055 	unsigned index_min, index_max;
1056 	unsigned index;
1057 	u64 val;
1058 	int i;
1059 
1060 	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
1061 		index_min = msr_range_array[i].min;
1062 		index_max = msr_range_array[i].max;
1063 
1064 		for (index = index_min; index < index_max; index++) {
1065 			if (rdmsrl_safe(index, &val))
1066 				continue;
1067 			printk(KERN_INFO " MSR%08x: %016llx\n", index, val);
1068 		}
1069 	}
1070 }
1071 
1072 static int show_msr;
1073 
1074 static __init int setup_show_msr(char *arg)
1075 {
1076 	int num;
1077 
1078 	get_option(&arg, &num);
1079 
1080 	if (num > 0)
1081 		show_msr = num;
1082 	return 1;
1083 }
1084 __setup("show_msr=", setup_show_msr);
1085 
1086 static __init int setup_noclflush(char *arg)
1087 {
1088 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1089 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1090 	return 1;
1091 }
1092 __setup("noclflush", setup_noclflush);
1093 
1094 void print_cpu_info(struct cpuinfo_x86 *c)
1095 {
1096 	const char *vendor = NULL;
1097 
1098 	if (c->x86_vendor < X86_VENDOR_NUM) {
1099 		vendor = this_cpu->c_vendor;
1100 	} else {
1101 		if (c->cpuid_level >= 0)
1102 			vendor = c->x86_vendor_id;
1103 	}
1104 
1105 	if (vendor && !strstr(c->x86_model_id, vendor))
1106 		printk(KERN_CONT "%s ", vendor);
1107 
1108 	if (c->x86_model_id[0])
1109 		printk(KERN_CONT "%s", c->x86_model_id);
1110 	else
1111 		printk(KERN_CONT "%d86", c->x86);
1112 
1113 	printk(KERN_CONT " (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1114 
1115 	if (c->x86_mask || c->cpuid_level >= 0)
1116 		printk(KERN_CONT ", stepping: 0x%x)\n", c->x86_mask);
1117 	else
1118 		printk(KERN_CONT ")\n");
1119 
1120 	print_cpu_msr(c);
1121 }
1122 
1123 void print_cpu_msr(struct cpuinfo_x86 *c)
1124 {
1125 	if (c->cpu_index < show_msr)
1126 		__print_cpu_msr();
1127 }
1128 
1129 static __init int setup_disablecpuid(char *arg)
1130 {
1131 	int bit;
1132 
1133 	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1134 		setup_clear_cpu_cap(bit);
1135 	else
1136 		return 0;
1137 
1138 	return 1;
1139 }
1140 __setup("clearcpuid=", setup_disablecpuid);
1141 
1142 #ifdef CONFIG_X86_64
1143 struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
1144 struct desc_ptr debug_idt_descr = { NR_VECTORS * 16 - 1,
1145 				    (unsigned long) debug_idt_table };
1146 
1147 DEFINE_PER_CPU_FIRST(union irq_stack_union,
1148 		     irq_stack_union) __aligned(PAGE_SIZE) __visible;
1149 
1150 /*
1151  * The following percpu variables are hot.  Align current_task to
1152  * cacheline size such that they fall in the same cacheline.
1153  */
1154 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1155 	&init_task;
1156 EXPORT_PER_CPU_SYMBOL(current_task);
1157 
1158 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1159 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1160 
1161 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1162 
1163 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1164 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1165 
1166 /*
1167  * Special IST stacks which the CPU switches to when it calls
1168  * an IST-marked descriptor entry. Up to 7 stacks (hardware
1169  * limit), all of them are 4K, except the debug stack which
1170  * is 8K.
1171  */
1172 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1173 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1174 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1175 };
1176 
1177 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1178 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1179 
1180 /* May not be marked __init: used by software suspend */
1181 void syscall_init(void)
1182 {
1183 	/*
1184 	 * LSTAR and STAR live in a bit strange symbiosis.
1185 	 * They both write to the same internal register. STAR allows to
1186 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1187 	 */
1188 	wrmsrl(MSR_STAR,  ((u64)__USER32_CS)<<48  | ((u64)__KERNEL_CS)<<32);
1189 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1190 
1191 #ifdef CONFIG_IA32_EMULATION
1192 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1193 	/*
1194 	 * This only works on Intel CPUs.
1195 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1196 	 * This does not cause SYSENTER to jump to the wrong location, because
1197 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1198 	 */
1199 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1200 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1201 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1202 #else
1203 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1204 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1205 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1206 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1207 #endif
1208 
1209 	/* Flags to clear on syscall */
1210 	wrmsrl(MSR_SYSCALL_MASK,
1211 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1212 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1213 }
1214 
1215 /*
1216  * Copies of the original ist values from the tss are only accessed during
1217  * debugging, no special alignment required.
1218  */
1219 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1220 
1221 static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1222 DEFINE_PER_CPU(int, debug_stack_usage);
1223 
1224 int is_debug_stack(unsigned long addr)
1225 {
1226 	return __this_cpu_read(debug_stack_usage) ||
1227 		(addr <= __this_cpu_read(debug_stack_addr) &&
1228 		 addr > (__this_cpu_read(debug_stack_addr) - DEBUG_STKSZ));
1229 }
1230 NOKPROBE_SYMBOL(is_debug_stack);
1231 
1232 DEFINE_PER_CPU(u32, debug_idt_ctr);
1233 
1234 void debug_stack_set_zero(void)
1235 {
1236 	this_cpu_inc(debug_idt_ctr);
1237 	load_current_idt();
1238 }
1239 NOKPROBE_SYMBOL(debug_stack_set_zero);
1240 
1241 void debug_stack_reset(void)
1242 {
1243 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1244 		return;
1245 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1246 		load_current_idt();
1247 }
1248 NOKPROBE_SYMBOL(debug_stack_reset);
1249 
1250 #else	/* CONFIG_X86_64 */
1251 
1252 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1253 EXPORT_PER_CPU_SYMBOL(current_task);
1254 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1255 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1256 
1257 /*
1258  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1259  * the top of the kernel stack.  Use an extra percpu variable to track the
1260  * top of the kernel stack directly.
1261  */
1262 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1263 	(unsigned long)&init_thread_union + THREAD_SIZE;
1264 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1265 
1266 #ifdef CONFIG_CC_STACKPROTECTOR
1267 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1268 #endif
1269 
1270 #endif	/* CONFIG_X86_64 */
1271 
1272 /*
1273  * Clear all 6 debug registers:
1274  */
1275 static void clear_all_debug_regs(void)
1276 {
1277 	int i;
1278 
1279 	for (i = 0; i < 8; i++) {
1280 		/* Ignore db4, db5 */
1281 		if ((i == 4) || (i == 5))
1282 			continue;
1283 
1284 		set_debugreg(0, i);
1285 	}
1286 }
1287 
1288 #ifdef CONFIG_KGDB
1289 /*
1290  * Restore debug regs if using kgdbwait and you have a kernel debugger
1291  * connection established.
1292  */
1293 static void dbg_restore_debug_regs(void)
1294 {
1295 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1296 		arch_kgdb_ops.correct_hw_break();
1297 }
1298 #else /* ! CONFIG_KGDB */
1299 #define dbg_restore_debug_regs()
1300 #endif /* ! CONFIG_KGDB */
1301 
1302 static void wait_for_master_cpu(int cpu)
1303 {
1304 #ifdef CONFIG_SMP
1305 	/*
1306 	 * wait for ACK from master CPU before continuing
1307 	 * with AP initialization
1308 	 */
1309 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1310 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1311 		cpu_relax();
1312 #endif
1313 }
1314 
1315 /*
1316  * cpu_init() initializes state that is per-CPU. Some data is already
1317  * initialized (naturally) in the bootstrap process, such as the GDT
1318  * and IDT. We reload them nevertheless, this function acts as a
1319  * 'CPU state barrier', nothing should get across.
1320  * A lot of state is already set up in PDA init for 64 bit
1321  */
1322 #ifdef CONFIG_X86_64
1323 
1324 void cpu_init(void)
1325 {
1326 	struct orig_ist *oist;
1327 	struct task_struct *me;
1328 	struct tss_struct *t;
1329 	unsigned long v;
1330 	int cpu = stack_smp_processor_id();
1331 	int i;
1332 
1333 	wait_for_master_cpu(cpu);
1334 
1335 	/*
1336 	 * Initialize the CR4 shadow before doing anything that could
1337 	 * try to read it.
1338 	 */
1339 	cr4_init_shadow();
1340 
1341 	/*
1342 	 * Load microcode on this cpu if a valid microcode is available.
1343 	 * This is early microcode loading procedure.
1344 	 */
1345 	load_ucode_ap();
1346 
1347 	t = &per_cpu(cpu_tss, cpu);
1348 	oist = &per_cpu(orig_ist, cpu);
1349 
1350 #ifdef CONFIG_NUMA
1351 	if (this_cpu_read(numa_node) == 0 &&
1352 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1353 		set_numa_node(early_cpu_to_node(cpu));
1354 #endif
1355 
1356 	me = current;
1357 
1358 	pr_debug("Initializing CPU#%d\n", cpu);
1359 
1360 	cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1361 
1362 	/*
1363 	 * Initialize the per-CPU GDT with the boot GDT,
1364 	 * and set up the GDT descriptor:
1365 	 */
1366 
1367 	switch_to_new_gdt(cpu);
1368 	loadsegment(fs, 0);
1369 
1370 	load_current_idt();
1371 
1372 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1373 	syscall_init();
1374 
1375 	wrmsrl(MSR_FS_BASE, 0);
1376 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1377 	barrier();
1378 
1379 	x86_configure_nx();
1380 	x2apic_setup();
1381 
1382 	/*
1383 	 * set up and load the per-CPU TSS
1384 	 */
1385 	if (!oist->ist[0]) {
1386 		char *estacks = per_cpu(exception_stacks, cpu);
1387 
1388 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1389 			estacks += exception_stack_sizes[v];
1390 			oist->ist[v] = t->x86_tss.ist[v] =
1391 					(unsigned long)estacks;
1392 			if (v == DEBUG_STACK-1)
1393 				per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1394 		}
1395 	}
1396 
1397 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1398 
1399 	/*
1400 	 * <= is required because the CPU will access up to
1401 	 * 8 bits beyond the end of the IO permission bitmap.
1402 	 */
1403 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1404 		t->io_bitmap[i] = ~0UL;
1405 
1406 	atomic_inc(&init_mm.mm_count);
1407 	me->active_mm = &init_mm;
1408 	BUG_ON(me->mm);
1409 	enter_lazy_tlb(&init_mm, me);
1410 
1411 	load_sp0(t, &current->thread);
1412 	set_tss_desc(cpu, t);
1413 	load_TR_desc();
1414 	load_mm_ldt(&init_mm);
1415 
1416 	clear_all_debug_regs();
1417 	dbg_restore_debug_regs();
1418 
1419 	fpu__init_cpu();
1420 
1421 	if (is_uv_system())
1422 		uv_cpu_init();
1423 }
1424 
1425 #else
1426 
1427 void cpu_init(void)
1428 {
1429 	int cpu = smp_processor_id();
1430 	struct task_struct *curr = current;
1431 	struct tss_struct *t = &per_cpu(cpu_tss, cpu);
1432 	struct thread_struct *thread = &curr->thread;
1433 
1434 	wait_for_master_cpu(cpu);
1435 
1436 	/*
1437 	 * Initialize the CR4 shadow before doing anything that could
1438 	 * try to read it.
1439 	 */
1440 	cr4_init_shadow();
1441 
1442 	show_ucode_info_early();
1443 
1444 	printk(KERN_INFO "Initializing CPU#%d\n", cpu);
1445 
1446 	if (cpu_feature_enabled(X86_FEATURE_VME) || cpu_has_tsc || cpu_has_de)
1447 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1448 
1449 	load_current_idt();
1450 	switch_to_new_gdt(cpu);
1451 
1452 	/*
1453 	 * Set up and load the per-CPU TSS and LDT
1454 	 */
1455 	atomic_inc(&init_mm.mm_count);
1456 	curr->active_mm = &init_mm;
1457 	BUG_ON(curr->mm);
1458 	enter_lazy_tlb(&init_mm, curr);
1459 
1460 	load_sp0(t, thread);
1461 	set_tss_desc(cpu, t);
1462 	load_TR_desc();
1463 	load_mm_ldt(&init_mm);
1464 
1465 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1466 
1467 #ifdef CONFIG_DOUBLEFAULT
1468 	/* Set up doublefault TSS pointer in the GDT */
1469 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1470 #endif
1471 
1472 	clear_all_debug_regs();
1473 	dbg_restore_debug_regs();
1474 
1475 	fpu__init_cpu();
1476 }
1477 #endif
1478 
1479 #ifdef CONFIG_X86_DEBUG_STATIC_CPU_HAS
1480 void warn_pre_alternatives(void)
1481 {
1482 	WARN(1, "You're using static_cpu_has before alternatives have run!\n");
1483 }
1484 EXPORT_SYMBOL_GPL(warn_pre_alternatives);
1485 #endif
1486 
1487 inline bool __static_cpu_has_safe(u16 bit)
1488 {
1489 	return boot_cpu_has(bit);
1490 }
1491 EXPORT_SYMBOL_GPL(__static_cpu_has_safe);
1492 
1493 static void bsp_resume(void)
1494 {
1495 	if (this_cpu->c_bsp_resume)
1496 		this_cpu->c_bsp_resume(&boot_cpu_data);
1497 }
1498 
1499 static struct syscore_ops cpu_syscore_ops = {
1500 	.resume		= bsp_resume,
1501 };
1502 
1503 static int __init init_cpu_syscore(void)
1504 {
1505 	register_syscore_ops(&cpu_syscore_ops);
1506 	return 0;
1507 }
1508 core_initcall(init_cpu_syscore);
1509