xref: /linux/arch/x86/kernel/cpu/common.c (revision 5e4e38446a62a4f50d77b0dd11d4b379dee08988)
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 static int __init x86_noinvpcid_setup(char *s)
166 {
167 	/* noinvpcid doesn't accept parameters */
168 	if (s)
169 		return -EINVAL;
170 
171 	/* do not emit a message if the feature is not present */
172 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
173 		return 0;
174 
175 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
176 	pr_info("noinvpcid: INVPCID feature disabled\n");
177 	return 0;
178 }
179 early_param("noinvpcid", x86_noinvpcid_setup);
180 
181 #ifdef CONFIG_X86_32
182 static int cachesize_override = -1;
183 static int disable_x86_serial_nr = 1;
184 
185 static int __init cachesize_setup(char *str)
186 {
187 	get_option(&str, &cachesize_override);
188 	return 1;
189 }
190 __setup("cachesize=", cachesize_setup);
191 
192 static int __init x86_sep_setup(char *s)
193 {
194 	setup_clear_cpu_cap(X86_FEATURE_SEP);
195 	return 1;
196 }
197 __setup("nosep", x86_sep_setup);
198 
199 /* Standard macro to see if a specific flag is changeable */
200 static inline int flag_is_changeable_p(u32 flag)
201 {
202 	u32 f1, f2;
203 
204 	/*
205 	 * Cyrix and IDT cpus allow disabling of CPUID
206 	 * so the code below may return different results
207 	 * when it is executed before and after enabling
208 	 * the CPUID. Add "volatile" to not allow gcc to
209 	 * optimize the subsequent calls to this function.
210 	 */
211 	asm volatile ("pushfl		\n\t"
212 		      "pushfl		\n\t"
213 		      "popl %0		\n\t"
214 		      "movl %0, %1	\n\t"
215 		      "xorl %2, %0	\n\t"
216 		      "pushl %0		\n\t"
217 		      "popfl		\n\t"
218 		      "pushfl		\n\t"
219 		      "popl %0		\n\t"
220 		      "popfl		\n\t"
221 
222 		      : "=&r" (f1), "=&r" (f2)
223 		      : "ir" (flag));
224 
225 	return ((f1^f2) & flag) != 0;
226 }
227 
228 /* Probe for the CPUID instruction */
229 int have_cpuid_p(void)
230 {
231 	return flag_is_changeable_p(X86_EFLAGS_ID);
232 }
233 
234 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
235 {
236 	unsigned long lo, hi;
237 
238 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
239 		return;
240 
241 	/* Disable processor serial number: */
242 
243 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
244 	lo |= 0x200000;
245 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
246 
247 	pr_notice("CPU serial number disabled.\n");
248 	clear_cpu_cap(c, X86_FEATURE_PN);
249 
250 	/* Disabling the serial number may affect the cpuid level */
251 	c->cpuid_level = cpuid_eax(0);
252 }
253 
254 static int __init x86_serial_nr_setup(char *s)
255 {
256 	disable_x86_serial_nr = 0;
257 	return 1;
258 }
259 __setup("serialnumber", x86_serial_nr_setup);
260 #else
261 static inline int flag_is_changeable_p(u32 flag)
262 {
263 	return 1;
264 }
265 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
266 {
267 }
268 #endif
269 
270 static __init int setup_disable_smep(char *arg)
271 {
272 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
273 	return 1;
274 }
275 __setup("nosmep", setup_disable_smep);
276 
277 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
278 {
279 	if (cpu_has(c, X86_FEATURE_SMEP))
280 		cr4_set_bits(X86_CR4_SMEP);
281 }
282 
283 static __init int setup_disable_smap(char *arg)
284 {
285 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
286 	return 1;
287 }
288 __setup("nosmap", setup_disable_smap);
289 
290 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
291 {
292 	unsigned long eflags = native_save_fl();
293 
294 	/* This should have been cleared long ago */
295 	BUG_ON(eflags & X86_EFLAGS_AC);
296 
297 	if (cpu_has(c, X86_FEATURE_SMAP)) {
298 #ifdef CONFIG_X86_SMAP
299 		cr4_set_bits(X86_CR4_SMAP);
300 #else
301 		cr4_clear_bits(X86_CR4_SMAP);
302 #endif
303 	}
304 }
305 
306 /*
307  * Some CPU features depend on higher CPUID levels, which may not always
308  * be available due to CPUID level capping or broken virtualization
309  * software.  Add those features to this table to auto-disable them.
310  */
311 struct cpuid_dependent_feature {
312 	u32 feature;
313 	u32 level;
314 };
315 
316 static const struct cpuid_dependent_feature
317 cpuid_dependent_features[] = {
318 	{ X86_FEATURE_MWAIT,		0x00000005 },
319 	{ X86_FEATURE_DCA,		0x00000009 },
320 	{ X86_FEATURE_XSAVE,		0x0000000d },
321 	{ 0, 0 }
322 };
323 
324 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
325 {
326 	const struct cpuid_dependent_feature *df;
327 
328 	for (df = cpuid_dependent_features; df->feature; df++) {
329 
330 		if (!cpu_has(c, df->feature))
331 			continue;
332 		/*
333 		 * Note: cpuid_level is set to -1 if unavailable, but
334 		 * extended_extended_level is set to 0 if unavailable
335 		 * and the legitimate extended levels are all negative
336 		 * when signed; hence the weird messing around with
337 		 * signs here...
338 		 */
339 		if (!((s32)df->level < 0 ?
340 		     (u32)df->level > (u32)c->extended_cpuid_level :
341 		     (s32)df->level > (s32)c->cpuid_level))
342 			continue;
343 
344 		clear_cpu_cap(c, df->feature);
345 		if (!warn)
346 			continue;
347 
348 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
349 			x86_cap_flag(df->feature), df->level);
350 	}
351 }
352 
353 /*
354  * Naming convention should be: <Name> [(<Codename>)]
355  * This table only is used unless init_<vendor>() below doesn't set it;
356  * in particular, if CPUID levels 0x80000002..4 are supported, this
357  * isn't used
358  */
359 
360 /* Look up CPU names by table lookup. */
361 static const char *table_lookup_model(struct cpuinfo_x86 *c)
362 {
363 #ifdef CONFIG_X86_32
364 	const struct legacy_cpu_model_info *info;
365 
366 	if (c->x86_model >= 16)
367 		return NULL;	/* Range check */
368 
369 	if (!this_cpu)
370 		return NULL;
371 
372 	info = this_cpu->legacy_models;
373 
374 	while (info->family) {
375 		if (info->family == c->x86)
376 			return info->model_names[c->x86_model];
377 		info++;
378 	}
379 #endif
380 	return NULL;		/* Not found */
381 }
382 
383 __u32 cpu_caps_cleared[NCAPINTS];
384 __u32 cpu_caps_set[NCAPINTS];
385 
386 void load_percpu_segment(int cpu)
387 {
388 #ifdef CONFIG_X86_32
389 	loadsegment(fs, __KERNEL_PERCPU);
390 #else
391 	loadsegment(gs, 0);
392 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
393 #endif
394 	load_stack_canary_segment();
395 }
396 
397 /*
398  * Current gdt points %fs at the "master" per-cpu area: after this,
399  * it's on the real one.
400  */
401 void switch_to_new_gdt(int cpu)
402 {
403 	struct desc_ptr gdt_descr;
404 
405 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
406 	gdt_descr.size = GDT_SIZE - 1;
407 	load_gdt(&gdt_descr);
408 	/* Reload the per-cpu base */
409 
410 	load_percpu_segment(cpu);
411 }
412 
413 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
414 
415 static void get_model_name(struct cpuinfo_x86 *c)
416 {
417 	unsigned int *v;
418 	char *p, *q, *s;
419 
420 	if (c->extended_cpuid_level < 0x80000004)
421 		return;
422 
423 	v = (unsigned int *)c->x86_model_id;
424 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
425 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
426 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
427 	c->x86_model_id[48] = 0;
428 
429 	/* Trim whitespace */
430 	p = q = s = &c->x86_model_id[0];
431 
432 	while (*p == ' ')
433 		p++;
434 
435 	while (*p) {
436 		/* Note the last non-whitespace index */
437 		if (!isspace(*p))
438 			s = q;
439 
440 		*q++ = *p++;
441 	}
442 
443 	*(s + 1) = '\0';
444 }
445 
446 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
447 {
448 	unsigned int n, dummy, ebx, ecx, edx, l2size;
449 
450 	n = c->extended_cpuid_level;
451 
452 	if (n >= 0x80000005) {
453 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
454 		c->x86_cache_size = (ecx>>24) + (edx>>24);
455 #ifdef CONFIG_X86_64
456 		/* On K8 L1 TLB is inclusive, so don't count it */
457 		c->x86_tlbsize = 0;
458 #endif
459 	}
460 
461 	if (n < 0x80000006)	/* Some chips just has a large L1. */
462 		return;
463 
464 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
465 	l2size = ecx >> 16;
466 
467 #ifdef CONFIG_X86_64
468 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
469 #else
470 	/* do processor-specific cache resizing */
471 	if (this_cpu->legacy_cache_size)
472 		l2size = this_cpu->legacy_cache_size(c, l2size);
473 
474 	/* Allow user to override all this if necessary. */
475 	if (cachesize_override != -1)
476 		l2size = cachesize_override;
477 
478 	if (l2size == 0)
479 		return;		/* Again, no L2 cache is possible */
480 #endif
481 
482 	c->x86_cache_size = l2size;
483 }
484 
485 u16 __read_mostly tlb_lli_4k[NR_INFO];
486 u16 __read_mostly tlb_lli_2m[NR_INFO];
487 u16 __read_mostly tlb_lli_4m[NR_INFO];
488 u16 __read_mostly tlb_lld_4k[NR_INFO];
489 u16 __read_mostly tlb_lld_2m[NR_INFO];
490 u16 __read_mostly tlb_lld_4m[NR_INFO];
491 u16 __read_mostly tlb_lld_1g[NR_INFO];
492 
493 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
494 {
495 	if (this_cpu->c_detect_tlb)
496 		this_cpu->c_detect_tlb(c);
497 
498 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
499 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
500 		tlb_lli_4m[ENTRIES]);
501 
502 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
503 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
504 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
505 }
506 
507 void detect_ht(struct cpuinfo_x86 *c)
508 {
509 #ifdef CONFIG_SMP
510 	u32 eax, ebx, ecx, edx;
511 	int index_msb, core_bits;
512 	static bool printed;
513 
514 	if (!cpu_has(c, X86_FEATURE_HT))
515 		return;
516 
517 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
518 		goto out;
519 
520 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
521 		return;
522 
523 	cpuid(1, &eax, &ebx, &ecx, &edx);
524 
525 	smp_num_siblings = (ebx & 0xff0000) >> 16;
526 
527 	if (smp_num_siblings == 1) {
528 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
529 		goto out;
530 	}
531 
532 	if (smp_num_siblings <= 1)
533 		goto out;
534 
535 	index_msb = get_count_order(smp_num_siblings);
536 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
537 
538 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
539 
540 	index_msb = get_count_order(smp_num_siblings);
541 
542 	core_bits = get_count_order(c->x86_max_cores);
543 
544 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
545 				       ((1 << core_bits) - 1);
546 
547 out:
548 	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
549 		pr_info("CPU: Physical Processor ID: %d\n",
550 			c->phys_proc_id);
551 		pr_info("CPU: Processor Core ID: %d\n",
552 			c->cpu_core_id);
553 		printed = 1;
554 	}
555 #endif
556 }
557 
558 static void get_cpu_vendor(struct cpuinfo_x86 *c)
559 {
560 	char *v = c->x86_vendor_id;
561 	int i;
562 
563 	for (i = 0; i < X86_VENDOR_NUM; i++) {
564 		if (!cpu_devs[i])
565 			break;
566 
567 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
568 		    (cpu_devs[i]->c_ident[1] &&
569 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
570 
571 			this_cpu = cpu_devs[i];
572 			c->x86_vendor = this_cpu->c_x86_vendor;
573 			return;
574 		}
575 	}
576 
577 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
578 		    "CPU: Your system may be unstable.\n", v);
579 
580 	c->x86_vendor = X86_VENDOR_UNKNOWN;
581 	this_cpu = &default_cpu;
582 }
583 
584 void cpu_detect(struct cpuinfo_x86 *c)
585 {
586 	/* Get vendor name */
587 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
588 	      (unsigned int *)&c->x86_vendor_id[0],
589 	      (unsigned int *)&c->x86_vendor_id[8],
590 	      (unsigned int *)&c->x86_vendor_id[4]);
591 
592 	c->x86 = 4;
593 	/* Intel-defined flags: level 0x00000001 */
594 	if (c->cpuid_level >= 0x00000001) {
595 		u32 junk, tfms, cap0, misc;
596 
597 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
598 		c->x86		= x86_family(tfms);
599 		c->x86_model	= x86_model(tfms);
600 		c->x86_mask	= x86_stepping(tfms);
601 
602 		if (cap0 & (1<<19)) {
603 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
604 			c->x86_cache_alignment = c->x86_clflush_size;
605 		}
606 	}
607 }
608 
609 void get_cpu_cap(struct cpuinfo_x86 *c)
610 {
611 	u32 eax, ebx, ecx, edx;
612 
613 	/* Intel-defined flags: level 0x00000001 */
614 	if (c->cpuid_level >= 0x00000001) {
615 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
616 
617 		c->x86_capability[CPUID_1_ECX] = ecx;
618 		c->x86_capability[CPUID_1_EDX] = edx;
619 	}
620 
621 	/* Additional Intel-defined flags: level 0x00000007 */
622 	if (c->cpuid_level >= 0x00000007) {
623 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
624 
625 		c->x86_capability[CPUID_7_0_EBX] = ebx;
626 
627 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
628 	}
629 
630 	/* Extended state features: level 0x0000000d */
631 	if (c->cpuid_level >= 0x0000000d) {
632 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
633 
634 		c->x86_capability[CPUID_D_1_EAX] = eax;
635 	}
636 
637 	/* Additional Intel-defined flags: level 0x0000000F */
638 	if (c->cpuid_level >= 0x0000000F) {
639 
640 		/* QoS sub-leaf, EAX=0Fh, ECX=0 */
641 		cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx);
642 		c->x86_capability[CPUID_F_0_EDX] = edx;
643 
644 		if (cpu_has(c, X86_FEATURE_CQM_LLC)) {
645 			/* will be overridden if occupancy monitoring exists */
646 			c->x86_cache_max_rmid = ebx;
647 
648 			/* QoS sub-leaf, EAX=0Fh, ECX=1 */
649 			cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx);
650 			c->x86_capability[CPUID_F_1_EDX] = edx;
651 
652 			if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) {
653 				c->x86_cache_max_rmid = ecx;
654 				c->x86_cache_occ_scale = ebx;
655 			}
656 		} else {
657 			c->x86_cache_max_rmid = -1;
658 			c->x86_cache_occ_scale = -1;
659 		}
660 	}
661 
662 	/* AMD-defined flags: level 0x80000001 */
663 	eax = cpuid_eax(0x80000000);
664 	c->extended_cpuid_level = eax;
665 
666 	if ((eax & 0xffff0000) == 0x80000000) {
667 		if (eax >= 0x80000001) {
668 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
669 
670 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
671 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
672 		}
673 	}
674 
675 	if (c->extended_cpuid_level >= 0x80000008) {
676 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
677 
678 		c->x86_virt_bits = (eax >> 8) & 0xff;
679 		c->x86_phys_bits = eax & 0xff;
680 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
681 	}
682 #ifdef CONFIG_X86_32
683 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
684 		c->x86_phys_bits = 36;
685 #endif
686 
687 	if (c->extended_cpuid_level >= 0x80000007)
688 		c->x86_power = cpuid_edx(0x80000007);
689 
690 	if (c->extended_cpuid_level >= 0x8000000a)
691 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
692 
693 	init_scattered_cpuid_features(c);
694 }
695 
696 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
697 {
698 #ifdef CONFIG_X86_32
699 	int i;
700 
701 	/*
702 	 * First of all, decide if this is a 486 or higher
703 	 * It's a 486 if we can modify the AC flag
704 	 */
705 	if (flag_is_changeable_p(X86_EFLAGS_AC))
706 		c->x86 = 4;
707 	else
708 		c->x86 = 3;
709 
710 	for (i = 0; i < X86_VENDOR_NUM; i++)
711 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
712 			c->x86_vendor_id[0] = 0;
713 			cpu_devs[i]->c_identify(c);
714 			if (c->x86_vendor_id[0]) {
715 				get_cpu_vendor(c);
716 				break;
717 			}
718 		}
719 #endif
720 }
721 
722 /*
723  * Do minimum CPU detection early.
724  * Fields really needed: vendor, cpuid_level, family, model, mask,
725  * cache alignment.
726  * The others are not touched to avoid unwanted side effects.
727  *
728  * WARNING: this function is only called on the BP.  Don't add code here
729  * that is supposed to run on all CPUs.
730  */
731 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
732 {
733 #ifdef CONFIG_X86_64
734 	c->x86_clflush_size = 64;
735 	c->x86_phys_bits = 36;
736 	c->x86_virt_bits = 48;
737 #else
738 	c->x86_clflush_size = 32;
739 	c->x86_phys_bits = 32;
740 	c->x86_virt_bits = 32;
741 #endif
742 	c->x86_cache_alignment = c->x86_clflush_size;
743 
744 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
745 	c->extended_cpuid_level = 0;
746 
747 	if (!have_cpuid_p())
748 		identify_cpu_without_cpuid(c);
749 
750 	/* cyrix could have cpuid enabled via c_identify()*/
751 	if (!have_cpuid_p())
752 		return;
753 
754 	cpu_detect(c);
755 	get_cpu_vendor(c);
756 	get_cpu_cap(c);
757 
758 	if (this_cpu->c_early_init)
759 		this_cpu->c_early_init(c);
760 
761 	c->cpu_index = 0;
762 	filter_cpuid_features(c, false);
763 
764 	if (this_cpu->c_bsp_init)
765 		this_cpu->c_bsp_init(c);
766 
767 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
768 	fpu__init_system(c);
769 }
770 
771 void __init early_cpu_init(void)
772 {
773 	const struct cpu_dev *const *cdev;
774 	int count = 0;
775 
776 #ifdef CONFIG_PROCESSOR_SELECT
777 	pr_info("KERNEL supported cpus:\n");
778 #endif
779 
780 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
781 		const struct cpu_dev *cpudev = *cdev;
782 
783 		if (count >= X86_VENDOR_NUM)
784 			break;
785 		cpu_devs[count] = cpudev;
786 		count++;
787 
788 #ifdef CONFIG_PROCESSOR_SELECT
789 		{
790 			unsigned int j;
791 
792 			for (j = 0; j < 2; j++) {
793 				if (!cpudev->c_ident[j])
794 					continue;
795 				pr_info("  %s %s\n", cpudev->c_vendor,
796 					cpudev->c_ident[j]);
797 			}
798 		}
799 #endif
800 	}
801 	early_identify_cpu(&boot_cpu_data);
802 }
803 
804 /*
805  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
806  * unfortunately, that's not true in practice because of early VIA
807  * chips and (more importantly) broken virtualizers that are not easy
808  * to detect. In the latter case it doesn't even *fail* reliably, so
809  * probing for it doesn't even work. Disable it completely on 32-bit
810  * unless we can find a reliable way to detect all the broken cases.
811  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
812  */
813 static void detect_nopl(struct cpuinfo_x86 *c)
814 {
815 #ifdef CONFIG_X86_32
816 	clear_cpu_cap(c, X86_FEATURE_NOPL);
817 #else
818 	set_cpu_cap(c, X86_FEATURE_NOPL);
819 #endif
820 
821 	/*
822 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
823 	 * systems that run Linux at CPL > 0 may or may not have the
824 	 * issue, but, even if they have the issue, there's absolutely
825 	 * nothing we can do about it because we can't use the real IRET
826 	 * instruction.
827 	 *
828 	 * NB: For the time being, only 32-bit kernels support
829 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
830 	 * whether to apply espfix using paravirt hooks.  If any
831 	 * non-paravirt system ever shows up that does *not* have the
832 	 * ESPFIX issue, we can change this.
833 	 */
834 #ifdef CONFIG_X86_32
835 #ifdef CONFIG_PARAVIRT
836 	do {
837 		extern void native_iret(void);
838 		if (pv_cpu_ops.iret == native_iret)
839 			set_cpu_bug(c, X86_BUG_ESPFIX);
840 	} while (0);
841 #else
842 	set_cpu_bug(c, X86_BUG_ESPFIX);
843 #endif
844 #endif
845 }
846 
847 static void generic_identify(struct cpuinfo_x86 *c)
848 {
849 	c->extended_cpuid_level = 0;
850 
851 	if (!have_cpuid_p())
852 		identify_cpu_without_cpuid(c);
853 
854 	/* cyrix could have cpuid enabled via c_identify()*/
855 	if (!have_cpuid_p())
856 		return;
857 
858 	cpu_detect(c);
859 
860 	get_cpu_vendor(c);
861 
862 	get_cpu_cap(c);
863 
864 	if (c->cpuid_level >= 0x00000001) {
865 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
866 #ifdef CONFIG_X86_32
867 # ifdef CONFIG_SMP
868 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
869 # else
870 		c->apicid = c->initial_apicid;
871 # endif
872 #endif
873 		c->phys_proc_id = c->initial_apicid;
874 	}
875 
876 	get_model_name(c); /* Default name */
877 
878 	detect_nopl(c);
879 }
880 
881 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
882 {
883 	/*
884 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
885 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
886 	 * in case CQM bits really aren't there in this CPU.
887 	 */
888 	if (c != &boot_cpu_data) {
889 		boot_cpu_data.x86_cache_max_rmid =
890 			min(boot_cpu_data.x86_cache_max_rmid,
891 			    c->x86_cache_max_rmid);
892 	}
893 }
894 
895 /*
896  * This does the hard work of actually picking apart the CPU stuff...
897  */
898 static void identify_cpu(struct cpuinfo_x86 *c)
899 {
900 	int i;
901 
902 	c->loops_per_jiffy = loops_per_jiffy;
903 	c->x86_cache_size = -1;
904 	c->x86_vendor = X86_VENDOR_UNKNOWN;
905 	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
906 	c->x86_vendor_id[0] = '\0'; /* Unset */
907 	c->x86_model_id[0] = '\0';  /* Unset */
908 	c->x86_max_cores = 1;
909 	c->x86_coreid_bits = 0;
910 #ifdef CONFIG_X86_64
911 	c->x86_clflush_size = 64;
912 	c->x86_phys_bits = 36;
913 	c->x86_virt_bits = 48;
914 #else
915 	c->cpuid_level = -1;	/* CPUID not detected */
916 	c->x86_clflush_size = 32;
917 	c->x86_phys_bits = 32;
918 	c->x86_virt_bits = 32;
919 #endif
920 	c->x86_cache_alignment = c->x86_clflush_size;
921 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
922 
923 	generic_identify(c);
924 
925 	if (this_cpu->c_identify)
926 		this_cpu->c_identify(c);
927 
928 	/* Clear/Set all flags overriden by options, after probe */
929 	for (i = 0; i < NCAPINTS; i++) {
930 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
931 		c->x86_capability[i] |= cpu_caps_set[i];
932 	}
933 
934 #ifdef CONFIG_X86_64
935 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
936 #endif
937 
938 	/*
939 	 * Vendor-specific initialization.  In this section we
940 	 * canonicalize the feature flags, meaning if there are
941 	 * features a certain CPU supports which CPUID doesn't
942 	 * tell us, CPUID claiming incorrect flags, or other bugs,
943 	 * we handle them here.
944 	 *
945 	 * At the end of this section, c->x86_capability better
946 	 * indicate the features this CPU genuinely supports!
947 	 */
948 	if (this_cpu->c_init)
949 		this_cpu->c_init(c);
950 
951 	/* Disable the PN if appropriate */
952 	squash_the_stupid_serial_number(c);
953 
954 	/* Set up SMEP/SMAP */
955 	setup_smep(c);
956 	setup_smap(c);
957 
958 	/*
959 	 * The vendor-specific functions might have changed features.
960 	 * Now we do "generic changes."
961 	 */
962 
963 	/* Filter out anything that depends on CPUID levels we don't have */
964 	filter_cpuid_features(c, true);
965 
966 	/* If the model name is still unset, do table lookup. */
967 	if (!c->x86_model_id[0]) {
968 		const char *p;
969 		p = table_lookup_model(c);
970 		if (p)
971 			strcpy(c->x86_model_id, p);
972 		else
973 			/* Last resort... */
974 			sprintf(c->x86_model_id, "%02x/%02x",
975 				c->x86, c->x86_model);
976 	}
977 
978 #ifdef CONFIG_X86_64
979 	detect_ht(c);
980 #endif
981 
982 	init_hypervisor(c);
983 	x86_init_rdrand(c);
984 	x86_init_cache_qos(c);
985 
986 	/*
987 	 * Clear/Set all flags overriden by options, need do it
988 	 * before following smp all cpus cap AND.
989 	 */
990 	for (i = 0; i < NCAPINTS; i++) {
991 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
992 		c->x86_capability[i] |= cpu_caps_set[i];
993 	}
994 
995 	/*
996 	 * On SMP, boot_cpu_data holds the common feature set between
997 	 * all CPUs; so make sure that we indicate which features are
998 	 * common between the CPUs.  The first time this routine gets
999 	 * executed, c == &boot_cpu_data.
1000 	 */
1001 	if (c != &boot_cpu_data) {
1002 		/* AND the already accumulated flags with these */
1003 		for (i = 0; i < NCAPINTS; i++)
1004 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1005 
1006 		/* OR, i.e. replicate the bug flags */
1007 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1008 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1009 	}
1010 
1011 	/* Init Machine Check Exception if available. */
1012 	mcheck_cpu_init(c);
1013 
1014 	select_idle_routine(c);
1015 
1016 #ifdef CONFIG_NUMA
1017 	numa_add_cpu(smp_processor_id());
1018 #endif
1019 	/* The boot/hotplug time assigment got cleared, restore it */
1020 	c->logical_proc_id = topology_phys_to_logical_pkg(c->phys_proc_id);
1021 }
1022 
1023 /*
1024  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1025  * on 32-bit kernels:
1026  */
1027 #ifdef CONFIG_X86_32
1028 void enable_sep_cpu(void)
1029 {
1030 	struct tss_struct *tss;
1031 	int cpu;
1032 
1033 	cpu = get_cpu();
1034 	tss = &per_cpu(cpu_tss, cpu);
1035 
1036 	if (!boot_cpu_has(X86_FEATURE_SEP))
1037 		goto out;
1038 
1039 	/*
1040 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1041 	 * see the big comment in struct x86_hw_tss's definition.
1042 	 */
1043 
1044 	tss->x86_tss.ss1 = __KERNEL_CS;
1045 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1046 
1047 	wrmsr(MSR_IA32_SYSENTER_ESP,
1048 	      (unsigned long)tss + offsetofend(struct tss_struct, SYSENTER_stack),
1049 	      0);
1050 
1051 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1052 
1053 out:
1054 	put_cpu();
1055 }
1056 #endif
1057 
1058 void __init identify_boot_cpu(void)
1059 {
1060 	identify_cpu(&boot_cpu_data);
1061 	init_amd_e400_c1e_mask();
1062 #ifdef CONFIG_X86_32
1063 	sysenter_setup();
1064 	enable_sep_cpu();
1065 #endif
1066 	cpu_detect_tlb(&boot_cpu_data);
1067 }
1068 
1069 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1070 {
1071 	BUG_ON(c == &boot_cpu_data);
1072 	identify_cpu(c);
1073 #ifdef CONFIG_X86_32
1074 	enable_sep_cpu();
1075 #endif
1076 	mtrr_ap_init();
1077 }
1078 
1079 struct msr_range {
1080 	unsigned	min;
1081 	unsigned	max;
1082 };
1083 
1084 static const struct msr_range msr_range_array[] = {
1085 	{ 0x00000000, 0x00000418},
1086 	{ 0xc0000000, 0xc000040b},
1087 	{ 0xc0010000, 0xc0010142},
1088 	{ 0xc0011000, 0xc001103b},
1089 };
1090 
1091 static void __print_cpu_msr(void)
1092 {
1093 	unsigned index_min, index_max;
1094 	unsigned index;
1095 	u64 val;
1096 	int i;
1097 
1098 	for (i = 0; i < ARRAY_SIZE(msr_range_array); i++) {
1099 		index_min = msr_range_array[i].min;
1100 		index_max = msr_range_array[i].max;
1101 
1102 		for (index = index_min; index < index_max; index++) {
1103 			if (rdmsrl_safe(index, &val))
1104 				continue;
1105 			pr_info(" MSR%08x: %016llx\n", index, val);
1106 		}
1107 	}
1108 }
1109 
1110 static int show_msr;
1111 
1112 static __init int setup_show_msr(char *arg)
1113 {
1114 	int num;
1115 
1116 	get_option(&arg, &num);
1117 
1118 	if (num > 0)
1119 		show_msr = num;
1120 	return 1;
1121 }
1122 __setup("show_msr=", setup_show_msr);
1123 
1124 static __init int setup_noclflush(char *arg)
1125 {
1126 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1127 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1128 	return 1;
1129 }
1130 __setup("noclflush", setup_noclflush);
1131 
1132 void print_cpu_info(struct cpuinfo_x86 *c)
1133 {
1134 	const char *vendor = NULL;
1135 
1136 	if (c->x86_vendor < X86_VENDOR_NUM) {
1137 		vendor = this_cpu->c_vendor;
1138 	} else {
1139 		if (c->cpuid_level >= 0)
1140 			vendor = c->x86_vendor_id;
1141 	}
1142 
1143 	if (vendor && !strstr(c->x86_model_id, vendor))
1144 		pr_cont("%s ", vendor);
1145 
1146 	if (c->x86_model_id[0])
1147 		pr_cont("%s", c->x86_model_id);
1148 	else
1149 		pr_cont("%d86", c->x86);
1150 
1151 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1152 
1153 	if (c->x86_mask || c->cpuid_level >= 0)
1154 		pr_cont(", stepping: 0x%x)\n", c->x86_mask);
1155 	else
1156 		pr_cont(")\n");
1157 
1158 	print_cpu_msr(c);
1159 }
1160 
1161 void print_cpu_msr(struct cpuinfo_x86 *c)
1162 {
1163 	if (c->cpu_index < show_msr)
1164 		__print_cpu_msr();
1165 }
1166 
1167 static __init int setup_disablecpuid(char *arg)
1168 {
1169 	int bit;
1170 
1171 	if (get_option(&arg, &bit) && bit < NCAPINTS*32)
1172 		setup_clear_cpu_cap(bit);
1173 	else
1174 		return 0;
1175 
1176 	return 1;
1177 }
1178 __setup("clearcpuid=", setup_disablecpuid);
1179 
1180 #ifdef CONFIG_X86_64
1181 struct desc_ptr idt_descr = { NR_VECTORS * 16 - 1, (unsigned long) idt_table };
1182 struct desc_ptr debug_idt_descr = { NR_VECTORS * 16 - 1,
1183 				    (unsigned long) debug_idt_table };
1184 
1185 DEFINE_PER_CPU_FIRST(union irq_stack_union,
1186 		     irq_stack_union) __aligned(PAGE_SIZE) __visible;
1187 
1188 /*
1189  * The following percpu variables are hot.  Align current_task to
1190  * cacheline size such that they fall in the same cacheline.
1191  */
1192 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1193 	&init_task;
1194 EXPORT_PER_CPU_SYMBOL(current_task);
1195 
1196 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1197 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE - 64;
1198 
1199 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1200 
1201 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1202 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1203 
1204 /*
1205  * Special IST stacks which the CPU switches to when it calls
1206  * an IST-marked descriptor entry. Up to 7 stacks (hardware
1207  * limit), all of them are 4K, except the debug stack which
1208  * is 8K.
1209  */
1210 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1211 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1212 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1213 };
1214 
1215 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1216 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1217 
1218 /* May not be marked __init: used by software suspend */
1219 void syscall_init(void)
1220 {
1221 	/*
1222 	 * LSTAR and STAR live in a bit strange symbiosis.
1223 	 * They both write to the same internal register. STAR allows to
1224 	 * set CS/DS but only a 32bit target. LSTAR sets the 64bit rip.
1225 	 */
1226 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1227 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1228 
1229 #ifdef CONFIG_IA32_EMULATION
1230 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1231 	/*
1232 	 * This only works on Intel CPUs.
1233 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1234 	 * This does not cause SYSENTER to jump to the wrong location, because
1235 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1236 	 */
1237 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1238 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1239 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1240 #else
1241 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1242 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1243 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1244 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1245 #endif
1246 
1247 	/* Flags to clear on syscall */
1248 	wrmsrl(MSR_SYSCALL_MASK,
1249 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1250 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1251 }
1252 
1253 /*
1254  * Copies of the original ist values from the tss are only accessed during
1255  * debugging, no special alignment required.
1256  */
1257 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1258 
1259 static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1260 DEFINE_PER_CPU(int, debug_stack_usage);
1261 
1262 int is_debug_stack(unsigned long addr)
1263 {
1264 	return __this_cpu_read(debug_stack_usage) ||
1265 		(addr <= __this_cpu_read(debug_stack_addr) &&
1266 		 addr > (__this_cpu_read(debug_stack_addr) - DEBUG_STKSZ));
1267 }
1268 NOKPROBE_SYMBOL(is_debug_stack);
1269 
1270 DEFINE_PER_CPU(u32, debug_idt_ctr);
1271 
1272 void debug_stack_set_zero(void)
1273 {
1274 	this_cpu_inc(debug_idt_ctr);
1275 	load_current_idt();
1276 }
1277 NOKPROBE_SYMBOL(debug_stack_set_zero);
1278 
1279 void debug_stack_reset(void)
1280 {
1281 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1282 		return;
1283 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1284 		load_current_idt();
1285 }
1286 NOKPROBE_SYMBOL(debug_stack_reset);
1287 
1288 #else	/* CONFIG_X86_64 */
1289 
1290 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1291 EXPORT_PER_CPU_SYMBOL(current_task);
1292 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1293 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1294 
1295 /*
1296  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1297  * the top of the kernel stack.  Use an extra percpu variable to track the
1298  * top of the kernel stack directly.
1299  */
1300 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1301 	(unsigned long)&init_thread_union + THREAD_SIZE;
1302 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1303 
1304 #ifdef CONFIG_CC_STACKPROTECTOR
1305 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1306 #endif
1307 
1308 #endif	/* CONFIG_X86_64 */
1309 
1310 /*
1311  * Clear all 6 debug registers:
1312  */
1313 static void clear_all_debug_regs(void)
1314 {
1315 	int i;
1316 
1317 	for (i = 0; i < 8; i++) {
1318 		/* Ignore db4, db5 */
1319 		if ((i == 4) || (i == 5))
1320 			continue;
1321 
1322 		set_debugreg(0, i);
1323 	}
1324 }
1325 
1326 #ifdef CONFIG_KGDB
1327 /*
1328  * Restore debug regs if using kgdbwait and you have a kernel debugger
1329  * connection established.
1330  */
1331 static void dbg_restore_debug_regs(void)
1332 {
1333 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1334 		arch_kgdb_ops.correct_hw_break();
1335 }
1336 #else /* ! CONFIG_KGDB */
1337 #define dbg_restore_debug_regs()
1338 #endif /* ! CONFIG_KGDB */
1339 
1340 static void wait_for_master_cpu(int cpu)
1341 {
1342 #ifdef CONFIG_SMP
1343 	/*
1344 	 * wait for ACK from master CPU before continuing
1345 	 * with AP initialization
1346 	 */
1347 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1348 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1349 		cpu_relax();
1350 #endif
1351 }
1352 
1353 /*
1354  * cpu_init() initializes state that is per-CPU. Some data is already
1355  * initialized (naturally) in the bootstrap process, such as the GDT
1356  * and IDT. We reload them nevertheless, this function acts as a
1357  * 'CPU state barrier', nothing should get across.
1358  * A lot of state is already set up in PDA init for 64 bit
1359  */
1360 #ifdef CONFIG_X86_64
1361 
1362 void cpu_init(void)
1363 {
1364 	struct orig_ist *oist;
1365 	struct task_struct *me;
1366 	struct tss_struct *t;
1367 	unsigned long v;
1368 	int cpu = stack_smp_processor_id();
1369 	int i;
1370 
1371 	wait_for_master_cpu(cpu);
1372 
1373 	/*
1374 	 * Initialize the CR4 shadow before doing anything that could
1375 	 * try to read it.
1376 	 */
1377 	cr4_init_shadow();
1378 
1379 	/*
1380 	 * Load microcode on this cpu if a valid microcode is available.
1381 	 * This is early microcode loading procedure.
1382 	 */
1383 	load_ucode_ap();
1384 
1385 	t = &per_cpu(cpu_tss, cpu);
1386 	oist = &per_cpu(orig_ist, cpu);
1387 
1388 #ifdef CONFIG_NUMA
1389 	if (this_cpu_read(numa_node) == 0 &&
1390 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1391 		set_numa_node(early_cpu_to_node(cpu));
1392 #endif
1393 
1394 	me = current;
1395 
1396 	pr_debug("Initializing CPU#%d\n", cpu);
1397 
1398 	cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1399 
1400 	/*
1401 	 * Initialize the per-CPU GDT with the boot GDT,
1402 	 * and set up the GDT descriptor:
1403 	 */
1404 
1405 	switch_to_new_gdt(cpu);
1406 	loadsegment(fs, 0);
1407 
1408 	load_current_idt();
1409 
1410 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1411 	syscall_init();
1412 
1413 	wrmsrl(MSR_FS_BASE, 0);
1414 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1415 	barrier();
1416 
1417 	x86_configure_nx();
1418 	x2apic_setup();
1419 
1420 	/*
1421 	 * set up and load the per-CPU TSS
1422 	 */
1423 	if (!oist->ist[0]) {
1424 		char *estacks = per_cpu(exception_stacks, cpu);
1425 
1426 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1427 			estacks += exception_stack_sizes[v];
1428 			oist->ist[v] = t->x86_tss.ist[v] =
1429 					(unsigned long)estacks;
1430 			if (v == DEBUG_STACK-1)
1431 				per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1432 		}
1433 	}
1434 
1435 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1436 
1437 	/*
1438 	 * <= is required because the CPU will access up to
1439 	 * 8 bits beyond the end of the IO permission bitmap.
1440 	 */
1441 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1442 		t->io_bitmap[i] = ~0UL;
1443 
1444 	atomic_inc(&init_mm.mm_count);
1445 	me->active_mm = &init_mm;
1446 	BUG_ON(me->mm);
1447 	enter_lazy_tlb(&init_mm, me);
1448 
1449 	load_sp0(t, &current->thread);
1450 	set_tss_desc(cpu, t);
1451 	load_TR_desc();
1452 	load_mm_ldt(&init_mm);
1453 
1454 	clear_all_debug_regs();
1455 	dbg_restore_debug_regs();
1456 
1457 	fpu__init_cpu();
1458 
1459 	if (is_uv_system())
1460 		uv_cpu_init();
1461 }
1462 
1463 #else
1464 
1465 void cpu_init(void)
1466 {
1467 	int cpu = smp_processor_id();
1468 	struct task_struct *curr = current;
1469 	struct tss_struct *t = &per_cpu(cpu_tss, cpu);
1470 	struct thread_struct *thread = &curr->thread;
1471 
1472 	wait_for_master_cpu(cpu);
1473 
1474 	/*
1475 	 * Initialize the CR4 shadow before doing anything that could
1476 	 * try to read it.
1477 	 */
1478 	cr4_init_shadow();
1479 
1480 	show_ucode_info_early();
1481 
1482 	pr_info("Initializing CPU#%d\n", cpu);
1483 
1484 	if (cpu_feature_enabled(X86_FEATURE_VME) ||
1485 	    cpu_has_tsc ||
1486 	    boot_cpu_has(X86_FEATURE_DE))
1487 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1488 
1489 	load_current_idt();
1490 	switch_to_new_gdt(cpu);
1491 
1492 	/*
1493 	 * Set up and load the per-CPU TSS and LDT
1494 	 */
1495 	atomic_inc(&init_mm.mm_count);
1496 	curr->active_mm = &init_mm;
1497 	BUG_ON(curr->mm);
1498 	enter_lazy_tlb(&init_mm, curr);
1499 
1500 	load_sp0(t, thread);
1501 	set_tss_desc(cpu, t);
1502 	load_TR_desc();
1503 	load_mm_ldt(&init_mm);
1504 
1505 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1506 
1507 #ifdef CONFIG_DOUBLEFAULT
1508 	/* Set up doublefault TSS pointer in the GDT */
1509 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1510 #endif
1511 
1512 	clear_all_debug_regs();
1513 	dbg_restore_debug_regs();
1514 
1515 	fpu__init_cpu();
1516 }
1517 #endif
1518 
1519 static void bsp_resume(void)
1520 {
1521 	if (this_cpu->c_bsp_resume)
1522 		this_cpu->c_bsp_resume(&boot_cpu_data);
1523 }
1524 
1525 static struct syscore_ops cpu_syscore_ops = {
1526 	.resume		= bsp_resume,
1527 };
1528 
1529 static int __init init_cpu_syscore(void)
1530 {
1531 	register_syscore_ops(&cpu_syscore_ops);
1532 	return 0;
1533 }
1534 core_initcall(init_cpu_syscore);
1535