xref: /linux/arch/x86/kernel/cpu/common.c (revision a16cb91ad9c4bb959e7d1be69d83d8c24a88e9f2)
1 #include <linux/bootmem.h>
2 #include <linux/linkage.h>
3 #include <linux/bitops.h>
4 #include <linux/kernel.h>
5 #include <linux/export.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/mm.h>
11 #include <linux/sched/clock.h>
12 #include <linux/sched/task.h>
13 #include <linux/init.h>
14 #include <linux/kprobes.h>
15 #include <linux/kgdb.h>
16 #include <linux/smp.h>
17 #include <linux/io.h>
18 #include <linux/syscore_ops.h>
19 
20 #include <asm/stackprotector.h>
21 #include <asm/perf_event.h>
22 #include <asm/mmu_context.h>
23 #include <asm/archrandom.h>
24 #include <asm/hypervisor.h>
25 #include <asm/processor.h>
26 #include <asm/tlbflush.h>
27 #include <asm/debugreg.h>
28 #include <asm/sections.h>
29 #include <asm/vsyscall.h>
30 #include <linux/topology.h>
31 #include <linux/cpumask.h>
32 #include <asm/pgtable.h>
33 #include <linux/atomic.h>
34 #include <asm/proto.h>
35 #include <asm/setup.h>
36 #include <asm/apic.h>
37 #include <asm/desc.h>
38 #include <asm/fpu/internal.h>
39 #include <asm/mtrr.h>
40 #include <asm/hwcap2.h>
41 #include <linux/numa.h>
42 #include <asm/asm.h>
43 #include <asm/bugs.h>
44 #include <asm/cpu.h>
45 #include <asm/mce.h>
46 #include <asm/msr.h>
47 #include <asm/pat.h>
48 #include <asm/microcode.h>
49 #include <asm/microcode_intel.h>
50 
51 #ifdef CONFIG_X86_LOCAL_APIC
52 #include <asm/uv/uv.h>
53 #endif
54 
55 #include "cpu.h"
56 
57 u32 elf_hwcap2 __read_mostly;
58 
59 /* all of these masks are initialized in setup_cpu_local_masks() */
60 cpumask_var_t cpu_initialized_mask;
61 cpumask_var_t cpu_callout_mask;
62 cpumask_var_t cpu_callin_mask;
63 
64 /* representing cpus for which sibling maps can be computed */
65 cpumask_var_t cpu_sibling_setup_mask;
66 
67 /* correctly size the local cpu masks */
68 void __init setup_cpu_local_masks(void)
69 {
70 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
71 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
72 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
73 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
74 }
75 
76 static void default_init(struct cpuinfo_x86 *c)
77 {
78 #ifdef CONFIG_X86_64
79 	cpu_detect_cache_sizes(c);
80 #else
81 	/* Not much we can do here... */
82 	/* Check if at least it has cpuid */
83 	if (c->cpuid_level == -1) {
84 		/* No cpuid. It must be an ancient CPU */
85 		if (c->x86 == 4)
86 			strcpy(c->x86_model_id, "486");
87 		else if (c->x86 == 3)
88 			strcpy(c->x86_model_id, "386");
89 	}
90 #endif
91 	clear_sched_clock_stable();
92 }
93 
94 static const struct cpu_dev default_cpu = {
95 	.c_init		= default_init,
96 	.c_vendor	= "Unknown",
97 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
98 };
99 
100 static const struct cpu_dev *this_cpu = &default_cpu;
101 
102 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
103 #ifdef CONFIG_X86_64
104 	/*
105 	 * We need valid kernel segments for data and code in long mode too
106 	 * IRET will check the segment types  kkeil 2000/10/28
107 	 * Also sysret mandates a special GDT layout
108 	 *
109 	 * TLS descriptors are currently at a different place compared to i386.
110 	 * Hopefully nobody expects them at a fixed place (Wine?)
111 	 */
112 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
113 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
114 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
115 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
116 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
117 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
118 #else
119 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
120 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
121 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
122 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
123 	/*
124 	 * Segments used for calling PnP BIOS have byte granularity.
125 	 * They code segments and data segments have fixed 64k limits,
126 	 * the transfer segment sizes are set at run time.
127 	 */
128 	/* 32-bit code */
129 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
130 	/* 16-bit code */
131 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
132 	/* 16-bit data */
133 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
134 	/* 16-bit data */
135 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
136 	/* 16-bit data */
137 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
138 	/*
139 	 * The APM segments have byte granularity and their bases
140 	 * are set at run time.  All have 64k limits.
141 	 */
142 	/* 32-bit code */
143 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
144 	/* 16-bit code */
145 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
146 	/* data */
147 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
148 
149 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
150 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
151 	GDT_STACK_CANARY_INIT
152 #endif
153 } };
154 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
155 
156 static int __init x86_mpx_setup(char *s)
157 {
158 	/* require an exact match without trailing characters */
159 	if (strlen(s))
160 		return 0;
161 
162 	/* do not emit a message if the feature is not present */
163 	if (!boot_cpu_has(X86_FEATURE_MPX))
164 		return 1;
165 
166 	setup_clear_cpu_cap(X86_FEATURE_MPX);
167 	pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
168 	return 1;
169 }
170 __setup("nompx", x86_mpx_setup);
171 
172 static int __init x86_noinvpcid_setup(char *s)
173 {
174 	/* noinvpcid doesn't accept parameters */
175 	if (s)
176 		return -EINVAL;
177 
178 	/* do not emit a message if the feature is not present */
179 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
180 		return 0;
181 
182 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
183 	pr_info("noinvpcid: INVPCID feature disabled\n");
184 	return 0;
185 }
186 early_param("noinvpcid", x86_noinvpcid_setup);
187 
188 #ifdef CONFIG_X86_32
189 static int cachesize_override = -1;
190 static int disable_x86_serial_nr = 1;
191 
192 static int __init cachesize_setup(char *str)
193 {
194 	get_option(&str, &cachesize_override);
195 	return 1;
196 }
197 __setup("cachesize=", cachesize_setup);
198 
199 static int __init x86_sep_setup(char *s)
200 {
201 	setup_clear_cpu_cap(X86_FEATURE_SEP);
202 	return 1;
203 }
204 __setup("nosep", x86_sep_setup);
205 
206 /* Standard macro to see if a specific flag is changeable */
207 static inline int flag_is_changeable_p(u32 flag)
208 {
209 	u32 f1, f2;
210 
211 	/*
212 	 * Cyrix and IDT cpus allow disabling of CPUID
213 	 * so the code below may return different results
214 	 * when it is executed before and after enabling
215 	 * the CPUID. Add "volatile" to not allow gcc to
216 	 * optimize the subsequent calls to this function.
217 	 */
218 	asm volatile ("pushfl		\n\t"
219 		      "pushfl		\n\t"
220 		      "popl %0		\n\t"
221 		      "movl %0, %1	\n\t"
222 		      "xorl %2, %0	\n\t"
223 		      "pushl %0		\n\t"
224 		      "popfl		\n\t"
225 		      "pushfl		\n\t"
226 		      "popl %0		\n\t"
227 		      "popfl		\n\t"
228 
229 		      : "=&r" (f1), "=&r" (f2)
230 		      : "ir" (flag));
231 
232 	return ((f1^f2) & flag) != 0;
233 }
234 
235 /* Probe for the CPUID instruction */
236 int have_cpuid_p(void)
237 {
238 	return flag_is_changeable_p(X86_EFLAGS_ID);
239 }
240 
241 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
242 {
243 	unsigned long lo, hi;
244 
245 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
246 		return;
247 
248 	/* Disable processor serial number: */
249 
250 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
251 	lo |= 0x200000;
252 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
253 
254 	pr_notice("CPU serial number disabled.\n");
255 	clear_cpu_cap(c, X86_FEATURE_PN);
256 
257 	/* Disabling the serial number may affect the cpuid level */
258 	c->cpuid_level = cpuid_eax(0);
259 }
260 
261 static int __init x86_serial_nr_setup(char *s)
262 {
263 	disable_x86_serial_nr = 0;
264 	return 1;
265 }
266 __setup("serialnumber", x86_serial_nr_setup);
267 #else
268 static inline int flag_is_changeable_p(u32 flag)
269 {
270 	return 1;
271 }
272 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
273 {
274 }
275 #endif
276 
277 static __init int setup_disable_smep(char *arg)
278 {
279 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
280 	/* Check for things that depend on SMEP being enabled: */
281 	check_mpx_erratum(&boot_cpu_data);
282 	return 1;
283 }
284 __setup("nosmep", setup_disable_smep);
285 
286 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
287 {
288 	if (cpu_has(c, X86_FEATURE_SMEP))
289 		cr4_set_bits(X86_CR4_SMEP);
290 }
291 
292 static __init int setup_disable_smap(char *arg)
293 {
294 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
295 	return 1;
296 }
297 __setup("nosmap", setup_disable_smap);
298 
299 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
300 {
301 	unsigned long eflags = native_save_fl();
302 
303 	/* This should have been cleared long ago */
304 	BUG_ON(eflags & X86_EFLAGS_AC);
305 
306 	if (cpu_has(c, X86_FEATURE_SMAP)) {
307 #ifdef CONFIG_X86_SMAP
308 		cr4_set_bits(X86_CR4_SMAP);
309 #else
310 		cr4_clear_bits(X86_CR4_SMAP);
311 #endif
312 	}
313 }
314 
315 /*
316  * Protection Keys are not available in 32-bit mode.
317  */
318 static bool pku_disabled;
319 
320 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
321 {
322 	/* check the boot processor, plus compile options for PKU: */
323 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
324 		return;
325 	/* checks the actual processor's cpuid bits: */
326 	if (!cpu_has(c, X86_FEATURE_PKU))
327 		return;
328 	if (pku_disabled)
329 		return;
330 
331 	cr4_set_bits(X86_CR4_PKE);
332 	/*
333 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
334 	 * cpuid bit to be set.  We need to ensure that we
335 	 * update that bit in this CPU's "cpu_info".
336 	 */
337 	get_cpu_cap(c);
338 }
339 
340 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
341 static __init int setup_disable_pku(char *arg)
342 {
343 	/*
344 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
345 	 * runtime checks are against OSPKE so clearing the
346 	 * bit does nothing.
347 	 *
348 	 * This way, we will see "pku" in cpuinfo, but not
349 	 * "ospke", which is exactly what we want.  It shows
350 	 * that the CPU has PKU, but the OS has not enabled it.
351 	 * This happens to be exactly how a system would look
352 	 * if we disabled the config option.
353 	 */
354 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
355 	pku_disabled = true;
356 	return 1;
357 }
358 __setup("nopku", setup_disable_pku);
359 #endif /* CONFIG_X86_64 */
360 
361 /*
362  * Some CPU features depend on higher CPUID levels, which may not always
363  * be available due to CPUID level capping or broken virtualization
364  * software.  Add those features to this table to auto-disable them.
365  */
366 struct cpuid_dependent_feature {
367 	u32 feature;
368 	u32 level;
369 };
370 
371 static const struct cpuid_dependent_feature
372 cpuid_dependent_features[] = {
373 	{ X86_FEATURE_MWAIT,		0x00000005 },
374 	{ X86_FEATURE_DCA,		0x00000009 },
375 	{ X86_FEATURE_XSAVE,		0x0000000d },
376 	{ 0, 0 }
377 };
378 
379 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
380 {
381 	const struct cpuid_dependent_feature *df;
382 
383 	for (df = cpuid_dependent_features; df->feature; df++) {
384 
385 		if (!cpu_has(c, df->feature))
386 			continue;
387 		/*
388 		 * Note: cpuid_level is set to -1 if unavailable, but
389 		 * extended_extended_level is set to 0 if unavailable
390 		 * and the legitimate extended levels are all negative
391 		 * when signed; hence the weird messing around with
392 		 * signs here...
393 		 */
394 		if (!((s32)df->level < 0 ?
395 		     (u32)df->level > (u32)c->extended_cpuid_level :
396 		     (s32)df->level > (s32)c->cpuid_level))
397 			continue;
398 
399 		clear_cpu_cap(c, df->feature);
400 		if (!warn)
401 			continue;
402 
403 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
404 			x86_cap_flag(df->feature), df->level);
405 	}
406 }
407 
408 /*
409  * Naming convention should be: <Name> [(<Codename>)]
410  * This table only is used unless init_<vendor>() below doesn't set it;
411  * in particular, if CPUID levels 0x80000002..4 are supported, this
412  * isn't used
413  */
414 
415 /* Look up CPU names by table lookup. */
416 static const char *table_lookup_model(struct cpuinfo_x86 *c)
417 {
418 #ifdef CONFIG_X86_32
419 	const struct legacy_cpu_model_info *info;
420 
421 	if (c->x86_model >= 16)
422 		return NULL;	/* Range check */
423 
424 	if (!this_cpu)
425 		return NULL;
426 
427 	info = this_cpu->legacy_models;
428 
429 	while (info->family) {
430 		if (info->family == c->x86)
431 			return info->model_names[c->x86_model];
432 		info++;
433 	}
434 #endif
435 	return NULL;		/* Not found */
436 }
437 
438 __u32 cpu_caps_cleared[NCAPINTS];
439 __u32 cpu_caps_set[NCAPINTS];
440 
441 void load_percpu_segment(int cpu)
442 {
443 #ifdef CONFIG_X86_32
444 	loadsegment(fs, __KERNEL_PERCPU);
445 #else
446 	__loadsegment_simple(gs, 0);
447 	wrmsrl(MSR_GS_BASE, (unsigned long)per_cpu(irq_stack_union.gs_base, cpu));
448 #endif
449 	load_stack_canary_segment();
450 }
451 
452 /*
453  * Current gdt points %fs at the "master" per-cpu area: after this,
454  * it's on the real one.
455  */
456 void switch_to_new_gdt(int cpu)
457 {
458 	struct desc_ptr gdt_descr;
459 
460 	gdt_descr.address = (long)get_cpu_gdt_table(cpu);
461 	gdt_descr.size = GDT_SIZE - 1;
462 	load_gdt(&gdt_descr);
463 	/* Reload the per-cpu base */
464 
465 	load_percpu_segment(cpu);
466 }
467 
468 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
469 
470 static void get_model_name(struct cpuinfo_x86 *c)
471 {
472 	unsigned int *v;
473 	char *p, *q, *s;
474 
475 	if (c->extended_cpuid_level < 0x80000004)
476 		return;
477 
478 	v = (unsigned int *)c->x86_model_id;
479 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
480 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
481 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
482 	c->x86_model_id[48] = 0;
483 
484 	/* Trim whitespace */
485 	p = q = s = &c->x86_model_id[0];
486 
487 	while (*p == ' ')
488 		p++;
489 
490 	while (*p) {
491 		/* Note the last non-whitespace index */
492 		if (!isspace(*p))
493 			s = q;
494 
495 		*q++ = *p++;
496 	}
497 
498 	*(s + 1) = '\0';
499 }
500 
501 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
502 {
503 	unsigned int n, dummy, ebx, ecx, edx, l2size;
504 
505 	n = c->extended_cpuid_level;
506 
507 	if (n >= 0x80000005) {
508 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
509 		c->x86_cache_size = (ecx>>24) + (edx>>24);
510 #ifdef CONFIG_X86_64
511 		/* On K8 L1 TLB is inclusive, so don't count it */
512 		c->x86_tlbsize = 0;
513 #endif
514 	}
515 
516 	if (n < 0x80000006)	/* Some chips just has a large L1. */
517 		return;
518 
519 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
520 	l2size = ecx >> 16;
521 
522 #ifdef CONFIG_X86_64
523 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
524 #else
525 	/* do processor-specific cache resizing */
526 	if (this_cpu->legacy_cache_size)
527 		l2size = this_cpu->legacy_cache_size(c, l2size);
528 
529 	/* Allow user to override all this if necessary. */
530 	if (cachesize_override != -1)
531 		l2size = cachesize_override;
532 
533 	if (l2size == 0)
534 		return;		/* Again, no L2 cache is possible */
535 #endif
536 
537 	c->x86_cache_size = l2size;
538 }
539 
540 u16 __read_mostly tlb_lli_4k[NR_INFO];
541 u16 __read_mostly tlb_lli_2m[NR_INFO];
542 u16 __read_mostly tlb_lli_4m[NR_INFO];
543 u16 __read_mostly tlb_lld_4k[NR_INFO];
544 u16 __read_mostly tlb_lld_2m[NR_INFO];
545 u16 __read_mostly tlb_lld_4m[NR_INFO];
546 u16 __read_mostly tlb_lld_1g[NR_INFO];
547 
548 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
549 {
550 	if (this_cpu->c_detect_tlb)
551 		this_cpu->c_detect_tlb(c);
552 
553 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
554 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
555 		tlb_lli_4m[ENTRIES]);
556 
557 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
558 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
559 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
560 }
561 
562 void detect_ht(struct cpuinfo_x86 *c)
563 {
564 #ifdef CONFIG_SMP
565 	u32 eax, ebx, ecx, edx;
566 	int index_msb, core_bits;
567 	static bool printed;
568 
569 	if (!cpu_has(c, X86_FEATURE_HT))
570 		return;
571 
572 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
573 		goto out;
574 
575 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
576 		return;
577 
578 	cpuid(1, &eax, &ebx, &ecx, &edx);
579 
580 	smp_num_siblings = (ebx & 0xff0000) >> 16;
581 
582 	if (smp_num_siblings == 1) {
583 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
584 		goto out;
585 	}
586 
587 	if (smp_num_siblings <= 1)
588 		goto out;
589 
590 	index_msb = get_count_order(smp_num_siblings);
591 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
592 
593 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
594 
595 	index_msb = get_count_order(smp_num_siblings);
596 
597 	core_bits = get_count_order(c->x86_max_cores);
598 
599 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
600 				       ((1 << core_bits) - 1);
601 
602 out:
603 	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
604 		pr_info("CPU: Physical Processor ID: %d\n",
605 			c->phys_proc_id);
606 		pr_info("CPU: Processor Core ID: %d\n",
607 			c->cpu_core_id);
608 		printed = 1;
609 	}
610 #endif
611 }
612 
613 static void get_cpu_vendor(struct cpuinfo_x86 *c)
614 {
615 	char *v = c->x86_vendor_id;
616 	int i;
617 
618 	for (i = 0; i < X86_VENDOR_NUM; i++) {
619 		if (!cpu_devs[i])
620 			break;
621 
622 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
623 		    (cpu_devs[i]->c_ident[1] &&
624 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
625 
626 			this_cpu = cpu_devs[i];
627 			c->x86_vendor = this_cpu->c_x86_vendor;
628 			return;
629 		}
630 	}
631 
632 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
633 		    "CPU: Your system may be unstable.\n", v);
634 
635 	c->x86_vendor = X86_VENDOR_UNKNOWN;
636 	this_cpu = &default_cpu;
637 }
638 
639 void cpu_detect(struct cpuinfo_x86 *c)
640 {
641 	/* Get vendor name */
642 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
643 	      (unsigned int *)&c->x86_vendor_id[0],
644 	      (unsigned int *)&c->x86_vendor_id[8],
645 	      (unsigned int *)&c->x86_vendor_id[4]);
646 
647 	c->x86 = 4;
648 	/* Intel-defined flags: level 0x00000001 */
649 	if (c->cpuid_level >= 0x00000001) {
650 		u32 junk, tfms, cap0, misc;
651 
652 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
653 		c->x86		= x86_family(tfms);
654 		c->x86_model	= x86_model(tfms);
655 		c->x86_mask	= x86_stepping(tfms);
656 
657 		if (cap0 & (1<<19)) {
658 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
659 			c->x86_cache_alignment = c->x86_clflush_size;
660 		}
661 	}
662 }
663 
664 static void apply_forced_caps(struct cpuinfo_x86 *c)
665 {
666 	int i;
667 
668 	for (i = 0; i < NCAPINTS; i++) {
669 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
670 		c->x86_capability[i] |= cpu_caps_set[i];
671 	}
672 }
673 
674 void get_cpu_cap(struct cpuinfo_x86 *c)
675 {
676 	u32 eax, ebx, ecx, edx;
677 
678 	/* Intel-defined flags: level 0x00000001 */
679 	if (c->cpuid_level >= 0x00000001) {
680 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
681 
682 		c->x86_capability[CPUID_1_ECX] = ecx;
683 		c->x86_capability[CPUID_1_EDX] = edx;
684 	}
685 
686 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
687 	if (c->cpuid_level >= 0x00000006)
688 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
689 
690 	/* Additional Intel-defined flags: level 0x00000007 */
691 	if (c->cpuid_level >= 0x00000007) {
692 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
693 		c->x86_capability[CPUID_7_0_EBX] = ebx;
694 		c->x86_capability[CPUID_7_ECX] = ecx;
695 	}
696 
697 	/* Extended state features: level 0x0000000d */
698 	if (c->cpuid_level >= 0x0000000d) {
699 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
700 
701 		c->x86_capability[CPUID_D_1_EAX] = eax;
702 	}
703 
704 	/* Additional Intel-defined flags: level 0x0000000F */
705 	if (c->cpuid_level >= 0x0000000F) {
706 
707 		/* QoS sub-leaf, EAX=0Fh, ECX=0 */
708 		cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx);
709 		c->x86_capability[CPUID_F_0_EDX] = edx;
710 
711 		if (cpu_has(c, X86_FEATURE_CQM_LLC)) {
712 			/* will be overridden if occupancy monitoring exists */
713 			c->x86_cache_max_rmid = ebx;
714 
715 			/* QoS sub-leaf, EAX=0Fh, ECX=1 */
716 			cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx);
717 			c->x86_capability[CPUID_F_1_EDX] = edx;
718 
719 			if ((cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) ||
720 			      ((cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL)) ||
721 			       (cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)))) {
722 				c->x86_cache_max_rmid = ecx;
723 				c->x86_cache_occ_scale = ebx;
724 			}
725 		} else {
726 			c->x86_cache_max_rmid = -1;
727 			c->x86_cache_occ_scale = -1;
728 		}
729 	}
730 
731 	/* AMD-defined flags: level 0x80000001 */
732 	eax = cpuid_eax(0x80000000);
733 	c->extended_cpuid_level = eax;
734 
735 	if ((eax & 0xffff0000) == 0x80000000) {
736 		if (eax >= 0x80000001) {
737 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
738 
739 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
740 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
741 		}
742 	}
743 
744 	if (c->extended_cpuid_level >= 0x80000007) {
745 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
746 
747 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
748 		c->x86_power = edx;
749 	}
750 
751 	if (c->extended_cpuid_level >= 0x80000008) {
752 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
753 
754 		c->x86_virt_bits = (eax >> 8) & 0xff;
755 		c->x86_phys_bits = eax & 0xff;
756 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
757 	}
758 #ifdef CONFIG_X86_32
759 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
760 		c->x86_phys_bits = 36;
761 #endif
762 
763 	if (c->extended_cpuid_level >= 0x8000000a)
764 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
765 
766 	init_scattered_cpuid_features(c);
767 
768 	/*
769 	 * Clear/Set all flags overridden by options, after probe.
770 	 * This needs to happen each time we re-probe, which may happen
771 	 * several times during CPU initialization.
772 	 */
773 	apply_forced_caps(c);
774 }
775 
776 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
777 {
778 #ifdef CONFIG_X86_32
779 	int i;
780 
781 	/*
782 	 * First of all, decide if this is a 486 or higher
783 	 * It's a 486 if we can modify the AC flag
784 	 */
785 	if (flag_is_changeable_p(X86_EFLAGS_AC))
786 		c->x86 = 4;
787 	else
788 		c->x86 = 3;
789 
790 	for (i = 0; i < X86_VENDOR_NUM; i++)
791 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
792 			c->x86_vendor_id[0] = 0;
793 			cpu_devs[i]->c_identify(c);
794 			if (c->x86_vendor_id[0]) {
795 				get_cpu_vendor(c);
796 				break;
797 			}
798 		}
799 #endif
800 }
801 
802 /*
803  * Do minimum CPU detection early.
804  * Fields really needed: vendor, cpuid_level, family, model, mask,
805  * cache alignment.
806  * The others are not touched to avoid unwanted side effects.
807  *
808  * WARNING: this function is only called on the BP.  Don't add code here
809  * that is supposed to run on all CPUs.
810  */
811 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
812 {
813 #ifdef CONFIG_X86_64
814 	c->x86_clflush_size = 64;
815 	c->x86_phys_bits = 36;
816 	c->x86_virt_bits = 48;
817 #else
818 	c->x86_clflush_size = 32;
819 	c->x86_phys_bits = 32;
820 	c->x86_virt_bits = 32;
821 #endif
822 	c->x86_cache_alignment = c->x86_clflush_size;
823 
824 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
825 	c->extended_cpuid_level = 0;
826 
827 	/* cyrix could have cpuid enabled via c_identify()*/
828 	if (have_cpuid_p()) {
829 		cpu_detect(c);
830 		get_cpu_vendor(c);
831 		get_cpu_cap(c);
832 		setup_force_cpu_cap(X86_FEATURE_CPUID);
833 
834 		if (this_cpu->c_early_init)
835 			this_cpu->c_early_init(c);
836 
837 		c->cpu_index = 0;
838 		filter_cpuid_features(c, false);
839 
840 		if (this_cpu->c_bsp_init)
841 			this_cpu->c_bsp_init(c);
842 	} else {
843 		identify_cpu_without_cpuid(c);
844 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
845 	}
846 
847 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
848 	fpu__init_system(c);
849 }
850 
851 void __init early_cpu_init(void)
852 {
853 	const struct cpu_dev *const *cdev;
854 	int count = 0;
855 
856 #ifdef CONFIG_PROCESSOR_SELECT
857 	pr_info("KERNEL supported cpus:\n");
858 #endif
859 
860 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
861 		const struct cpu_dev *cpudev = *cdev;
862 
863 		if (count >= X86_VENDOR_NUM)
864 			break;
865 		cpu_devs[count] = cpudev;
866 		count++;
867 
868 #ifdef CONFIG_PROCESSOR_SELECT
869 		{
870 			unsigned int j;
871 
872 			for (j = 0; j < 2; j++) {
873 				if (!cpudev->c_ident[j])
874 					continue;
875 				pr_info("  %s %s\n", cpudev->c_vendor,
876 					cpudev->c_ident[j]);
877 			}
878 		}
879 #endif
880 	}
881 	early_identify_cpu(&boot_cpu_data);
882 }
883 
884 /*
885  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
886  * unfortunately, that's not true in practice because of early VIA
887  * chips and (more importantly) broken virtualizers that are not easy
888  * to detect. In the latter case it doesn't even *fail* reliably, so
889  * probing for it doesn't even work. Disable it completely on 32-bit
890  * unless we can find a reliable way to detect all the broken cases.
891  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
892  */
893 static void detect_nopl(struct cpuinfo_x86 *c)
894 {
895 #ifdef CONFIG_X86_32
896 	clear_cpu_cap(c, X86_FEATURE_NOPL);
897 #else
898 	set_cpu_cap(c, X86_FEATURE_NOPL);
899 #endif
900 }
901 
902 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
903 {
904 #ifdef CONFIG_X86_64
905 	/*
906 	 * Empirically, writing zero to a segment selector on AMD does
907 	 * not clear the base, whereas writing zero to a segment
908 	 * selector on Intel does clear the base.  Intel's behavior
909 	 * allows slightly faster context switches in the common case
910 	 * where GS is unused by the prev and next threads.
911 	 *
912 	 * Since neither vendor documents this anywhere that I can see,
913 	 * detect it directly instead of hardcoding the choice by
914 	 * vendor.
915 	 *
916 	 * I've designated AMD's behavior as the "bug" because it's
917 	 * counterintuitive and less friendly.
918 	 */
919 
920 	unsigned long old_base, tmp;
921 	rdmsrl(MSR_FS_BASE, old_base);
922 	wrmsrl(MSR_FS_BASE, 1);
923 	loadsegment(fs, 0);
924 	rdmsrl(MSR_FS_BASE, tmp);
925 	if (tmp != 0)
926 		set_cpu_bug(c, X86_BUG_NULL_SEG);
927 	wrmsrl(MSR_FS_BASE, old_base);
928 #endif
929 }
930 
931 static void generic_identify(struct cpuinfo_x86 *c)
932 {
933 	c->extended_cpuid_level = 0;
934 
935 	if (!have_cpuid_p())
936 		identify_cpu_without_cpuid(c);
937 
938 	/* cyrix could have cpuid enabled via c_identify()*/
939 	if (!have_cpuid_p())
940 		return;
941 
942 	cpu_detect(c);
943 
944 	get_cpu_vendor(c);
945 
946 	get_cpu_cap(c);
947 
948 	if (c->cpuid_level >= 0x00000001) {
949 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
950 #ifdef CONFIG_X86_32
951 # ifdef CONFIG_SMP
952 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
953 # else
954 		c->apicid = c->initial_apicid;
955 # endif
956 #endif
957 		c->phys_proc_id = c->initial_apicid;
958 	}
959 
960 	get_model_name(c); /* Default name */
961 
962 	detect_nopl(c);
963 
964 	detect_null_seg_behavior(c);
965 
966 	/*
967 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
968 	 * systems that run Linux at CPL > 0 may or may not have the
969 	 * issue, but, even if they have the issue, there's absolutely
970 	 * nothing we can do about it because we can't use the real IRET
971 	 * instruction.
972 	 *
973 	 * NB: For the time being, only 32-bit kernels support
974 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
975 	 * whether to apply espfix using paravirt hooks.  If any
976 	 * non-paravirt system ever shows up that does *not* have the
977 	 * ESPFIX issue, we can change this.
978 	 */
979 #ifdef CONFIG_X86_32
980 # ifdef CONFIG_PARAVIRT
981 	do {
982 		extern void native_iret(void);
983 		if (pv_cpu_ops.iret == native_iret)
984 			set_cpu_bug(c, X86_BUG_ESPFIX);
985 	} while (0);
986 # else
987 	set_cpu_bug(c, X86_BUG_ESPFIX);
988 # endif
989 #endif
990 }
991 
992 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
993 {
994 	/*
995 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
996 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
997 	 * in case CQM bits really aren't there in this CPU.
998 	 */
999 	if (c != &boot_cpu_data) {
1000 		boot_cpu_data.x86_cache_max_rmid =
1001 			min(boot_cpu_data.x86_cache_max_rmid,
1002 			    c->x86_cache_max_rmid);
1003 	}
1004 }
1005 
1006 /*
1007  * Validate that ACPI/mptables have the same information about the
1008  * effective APIC id and update the package map.
1009  */
1010 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1011 {
1012 #ifdef CONFIG_SMP
1013 	unsigned int apicid, cpu = smp_processor_id();
1014 
1015 	apicid = apic->cpu_present_to_apicid(cpu);
1016 
1017 	if (apicid != c->apicid) {
1018 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1019 		       cpu, apicid, c->initial_apicid);
1020 	}
1021 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1022 #else
1023 	c->logical_proc_id = 0;
1024 #endif
1025 }
1026 
1027 /*
1028  * This does the hard work of actually picking apart the CPU stuff...
1029  */
1030 static void identify_cpu(struct cpuinfo_x86 *c)
1031 {
1032 	int i;
1033 
1034 	c->loops_per_jiffy = loops_per_jiffy;
1035 	c->x86_cache_size = -1;
1036 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1037 	c->x86_model = c->x86_mask = 0;	/* So far unknown... */
1038 	c->x86_vendor_id[0] = '\0'; /* Unset */
1039 	c->x86_model_id[0] = '\0';  /* Unset */
1040 	c->x86_max_cores = 1;
1041 	c->x86_coreid_bits = 0;
1042 	c->cu_id = 0xff;
1043 #ifdef CONFIG_X86_64
1044 	c->x86_clflush_size = 64;
1045 	c->x86_phys_bits = 36;
1046 	c->x86_virt_bits = 48;
1047 #else
1048 	c->cpuid_level = -1;	/* CPUID not detected */
1049 	c->x86_clflush_size = 32;
1050 	c->x86_phys_bits = 32;
1051 	c->x86_virt_bits = 32;
1052 #endif
1053 	c->x86_cache_alignment = c->x86_clflush_size;
1054 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
1055 
1056 	generic_identify(c);
1057 
1058 	if (this_cpu->c_identify)
1059 		this_cpu->c_identify(c);
1060 
1061 	/* Clear/Set all flags overridden by options, after probe */
1062 	apply_forced_caps(c);
1063 
1064 #ifdef CONFIG_X86_64
1065 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1066 #endif
1067 
1068 	/*
1069 	 * Vendor-specific initialization.  In this section we
1070 	 * canonicalize the feature flags, meaning if there are
1071 	 * features a certain CPU supports which CPUID doesn't
1072 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1073 	 * we handle them here.
1074 	 *
1075 	 * At the end of this section, c->x86_capability better
1076 	 * indicate the features this CPU genuinely supports!
1077 	 */
1078 	if (this_cpu->c_init)
1079 		this_cpu->c_init(c);
1080 	else
1081 		clear_sched_clock_stable();
1082 
1083 	/* Disable the PN if appropriate */
1084 	squash_the_stupid_serial_number(c);
1085 
1086 	/* Set up SMEP/SMAP */
1087 	setup_smep(c);
1088 	setup_smap(c);
1089 
1090 	/*
1091 	 * The vendor-specific functions might have changed features.
1092 	 * Now we do "generic changes."
1093 	 */
1094 
1095 	/* Filter out anything that depends on CPUID levels we don't have */
1096 	filter_cpuid_features(c, true);
1097 
1098 	/* If the model name is still unset, do table lookup. */
1099 	if (!c->x86_model_id[0]) {
1100 		const char *p;
1101 		p = table_lookup_model(c);
1102 		if (p)
1103 			strcpy(c->x86_model_id, p);
1104 		else
1105 			/* Last resort... */
1106 			sprintf(c->x86_model_id, "%02x/%02x",
1107 				c->x86, c->x86_model);
1108 	}
1109 
1110 #ifdef CONFIG_X86_64
1111 	detect_ht(c);
1112 #endif
1113 
1114 	init_hypervisor(c);
1115 	x86_init_rdrand(c);
1116 	x86_init_cache_qos(c);
1117 	setup_pku(c);
1118 
1119 	/*
1120 	 * Clear/Set all flags overridden by options, need do it
1121 	 * before following smp all cpus cap AND.
1122 	 */
1123 	apply_forced_caps(c);
1124 
1125 	/*
1126 	 * On SMP, boot_cpu_data holds the common feature set between
1127 	 * all CPUs; so make sure that we indicate which features are
1128 	 * common between the CPUs.  The first time this routine gets
1129 	 * executed, c == &boot_cpu_data.
1130 	 */
1131 	if (c != &boot_cpu_data) {
1132 		/* AND the already accumulated flags with these */
1133 		for (i = 0; i < NCAPINTS; i++)
1134 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1135 
1136 		/* OR, i.e. replicate the bug flags */
1137 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1138 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1139 	}
1140 
1141 	/* Init Machine Check Exception if available. */
1142 	mcheck_cpu_init(c);
1143 
1144 	select_idle_routine(c);
1145 
1146 #ifdef CONFIG_NUMA
1147 	numa_add_cpu(smp_processor_id());
1148 #endif
1149 }
1150 
1151 /*
1152  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1153  * on 32-bit kernels:
1154  */
1155 #ifdef CONFIG_X86_32
1156 void enable_sep_cpu(void)
1157 {
1158 	struct tss_struct *tss;
1159 	int cpu;
1160 
1161 	if (!boot_cpu_has(X86_FEATURE_SEP))
1162 		return;
1163 
1164 	cpu = get_cpu();
1165 	tss = &per_cpu(cpu_tss, cpu);
1166 
1167 	/*
1168 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1169 	 * see the big comment in struct x86_hw_tss's definition.
1170 	 */
1171 
1172 	tss->x86_tss.ss1 = __KERNEL_CS;
1173 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1174 
1175 	wrmsr(MSR_IA32_SYSENTER_ESP,
1176 	      (unsigned long)tss + offsetofend(struct tss_struct, SYSENTER_stack),
1177 	      0);
1178 
1179 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1180 
1181 	put_cpu();
1182 }
1183 #endif
1184 
1185 void __init identify_boot_cpu(void)
1186 {
1187 	identify_cpu(&boot_cpu_data);
1188 #ifdef CONFIG_X86_32
1189 	sysenter_setup();
1190 	enable_sep_cpu();
1191 #endif
1192 	cpu_detect_tlb(&boot_cpu_data);
1193 }
1194 
1195 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1196 {
1197 	BUG_ON(c == &boot_cpu_data);
1198 	identify_cpu(c);
1199 #ifdef CONFIG_X86_32
1200 	enable_sep_cpu();
1201 #endif
1202 	mtrr_ap_init();
1203 	validate_apic_and_package_id(c);
1204 }
1205 
1206 static __init int setup_noclflush(char *arg)
1207 {
1208 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1209 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1210 	return 1;
1211 }
1212 __setup("noclflush", setup_noclflush);
1213 
1214 void print_cpu_info(struct cpuinfo_x86 *c)
1215 {
1216 	const char *vendor = NULL;
1217 
1218 	if (c->x86_vendor < X86_VENDOR_NUM) {
1219 		vendor = this_cpu->c_vendor;
1220 	} else {
1221 		if (c->cpuid_level >= 0)
1222 			vendor = c->x86_vendor_id;
1223 	}
1224 
1225 	if (vendor && !strstr(c->x86_model_id, vendor))
1226 		pr_cont("%s ", vendor);
1227 
1228 	if (c->x86_model_id[0])
1229 		pr_cont("%s", c->x86_model_id);
1230 	else
1231 		pr_cont("%d86", c->x86);
1232 
1233 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1234 
1235 	if (c->x86_mask || c->cpuid_level >= 0)
1236 		pr_cont(", stepping: 0x%x)\n", c->x86_mask);
1237 	else
1238 		pr_cont(")\n");
1239 }
1240 
1241 static __init int setup_disablecpuid(char *arg)
1242 {
1243 	int bit;
1244 
1245 	if (get_option(&arg, &bit) && bit >= 0 && bit < NCAPINTS * 32)
1246 		setup_clear_cpu_cap(bit);
1247 	else
1248 		return 0;
1249 
1250 	return 1;
1251 }
1252 __setup("clearcpuid=", setup_disablecpuid);
1253 
1254 #ifdef CONFIG_X86_64
1255 struct desc_ptr idt_descr __ro_after_init = {
1256 	.size = NR_VECTORS * 16 - 1,
1257 	.address = (unsigned long) idt_table,
1258 };
1259 const struct desc_ptr debug_idt_descr = {
1260 	.size = NR_VECTORS * 16 - 1,
1261 	.address = (unsigned long) debug_idt_table,
1262 };
1263 
1264 DEFINE_PER_CPU_FIRST(union irq_stack_union,
1265 		     irq_stack_union) __aligned(PAGE_SIZE) __visible;
1266 
1267 /*
1268  * The following percpu variables are hot.  Align current_task to
1269  * cacheline size such that they fall in the same cacheline.
1270  */
1271 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1272 	&init_task;
1273 EXPORT_PER_CPU_SYMBOL(current_task);
1274 
1275 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1276 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE;
1277 
1278 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1279 
1280 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1281 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1282 
1283 /*
1284  * Special IST stacks which the CPU switches to when it calls
1285  * an IST-marked descriptor entry. Up to 7 stacks (hardware
1286  * limit), all of them are 4K, except the debug stack which
1287  * is 8K.
1288  */
1289 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
1290 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
1291 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
1292 };
1293 
1294 static DEFINE_PER_CPU_PAGE_ALIGNED(char, exception_stacks
1295 	[(N_EXCEPTION_STACKS - 1) * EXCEPTION_STKSZ + DEBUG_STKSZ]);
1296 
1297 /* May not be marked __init: used by software suspend */
1298 void syscall_init(void)
1299 {
1300 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1301 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1302 
1303 #ifdef CONFIG_IA32_EMULATION
1304 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1305 	/*
1306 	 * This only works on Intel CPUs.
1307 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1308 	 * This does not cause SYSENTER to jump to the wrong location, because
1309 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1310 	 */
1311 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1312 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1313 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1314 #else
1315 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1316 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1317 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1318 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1319 #endif
1320 
1321 	/* Flags to clear on syscall */
1322 	wrmsrl(MSR_SYSCALL_MASK,
1323 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1324 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1325 }
1326 
1327 /*
1328  * Copies of the original ist values from the tss are only accessed during
1329  * debugging, no special alignment required.
1330  */
1331 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1332 
1333 static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1334 DEFINE_PER_CPU(int, debug_stack_usage);
1335 
1336 int is_debug_stack(unsigned long addr)
1337 {
1338 	return __this_cpu_read(debug_stack_usage) ||
1339 		(addr <= __this_cpu_read(debug_stack_addr) &&
1340 		 addr > (__this_cpu_read(debug_stack_addr) - DEBUG_STKSZ));
1341 }
1342 NOKPROBE_SYMBOL(is_debug_stack);
1343 
1344 DEFINE_PER_CPU(u32, debug_idt_ctr);
1345 
1346 void debug_stack_set_zero(void)
1347 {
1348 	this_cpu_inc(debug_idt_ctr);
1349 	load_current_idt();
1350 }
1351 NOKPROBE_SYMBOL(debug_stack_set_zero);
1352 
1353 void debug_stack_reset(void)
1354 {
1355 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1356 		return;
1357 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1358 		load_current_idt();
1359 }
1360 NOKPROBE_SYMBOL(debug_stack_reset);
1361 
1362 #else	/* CONFIG_X86_64 */
1363 
1364 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1365 EXPORT_PER_CPU_SYMBOL(current_task);
1366 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1367 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1368 
1369 /*
1370  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1371  * the top of the kernel stack.  Use an extra percpu variable to track the
1372  * top of the kernel stack directly.
1373  */
1374 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1375 	(unsigned long)&init_thread_union + THREAD_SIZE;
1376 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1377 
1378 #ifdef CONFIG_CC_STACKPROTECTOR
1379 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1380 #endif
1381 
1382 #endif	/* CONFIG_X86_64 */
1383 
1384 /*
1385  * Clear all 6 debug registers:
1386  */
1387 static void clear_all_debug_regs(void)
1388 {
1389 	int i;
1390 
1391 	for (i = 0; i < 8; i++) {
1392 		/* Ignore db4, db5 */
1393 		if ((i == 4) || (i == 5))
1394 			continue;
1395 
1396 		set_debugreg(0, i);
1397 	}
1398 }
1399 
1400 #ifdef CONFIG_KGDB
1401 /*
1402  * Restore debug regs if using kgdbwait and you have a kernel debugger
1403  * connection established.
1404  */
1405 static void dbg_restore_debug_regs(void)
1406 {
1407 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1408 		arch_kgdb_ops.correct_hw_break();
1409 }
1410 #else /* ! CONFIG_KGDB */
1411 #define dbg_restore_debug_regs()
1412 #endif /* ! CONFIG_KGDB */
1413 
1414 static void wait_for_master_cpu(int cpu)
1415 {
1416 #ifdef CONFIG_SMP
1417 	/*
1418 	 * wait for ACK from master CPU before continuing
1419 	 * with AP initialization
1420 	 */
1421 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1422 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1423 		cpu_relax();
1424 #endif
1425 }
1426 
1427 /*
1428  * cpu_init() initializes state that is per-CPU. Some data is already
1429  * initialized (naturally) in the bootstrap process, such as the GDT
1430  * and IDT. We reload them nevertheless, this function acts as a
1431  * 'CPU state barrier', nothing should get across.
1432  * A lot of state is already set up in PDA init for 64 bit
1433  */
1434 #ifdef CONFIG_X86_64
1435 
1436 void cpu_init(void)
1437 {
1438 	struct orig_ist *oist;
1439 	struct task_struct *me;
1440 	struct tss_struct *t;
1441 	unsigned long v;
1442 	int cpu = raw_smp_processor_id();
1443 	int i;
1444 
1445 	wait_for_master_cpu(cpu);
1446 
1447 	/*
1448 	 * Initialize the CR4 shadow before doing anything that could
1449 	 * try to read it.
1450 	 */
1451 	cr4_init_shadow();
1452 
1453 	if (cpu)
1454 		load_ucode_ap();
1455 
1456 	t = &per_cpu(cpu_tss, cpu);
1457 	oist = &per_cpu(orig_ist, cpu);
1458 
1459 #ifdef CONFIG_NUMA
1460 	if (this_cpu_read(numa_node) == 0 &&
1461 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1462 		set_numa_node(early_cpu_to_node(cpu));
1463 #endif
1464 
1465 	me = current;
1466 
1467 	pr_debug("Initializing CPU#%d\n", cpu);
1468 
1469 	cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1470 
1471 	/*
1472 	 * Initialize the per-CPU GDT with the boot GDT,
1473 	 * and set up the GDT descriptor:
1474 	 */
1475 
1476 	switch_to_new_gdt(cpu);
1477 	loadsegment(fs, 0);
1478 
1479 	load_current_idt();
1480 
1481 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1482 	syscall_init();
1483 
1484 	wrmsrl(MSR_FS_BASE, 0);
1485 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1486 	barrier();
1487 
1488 	x86_configure_nx();
1489 	x2apic_setup();
1490 
1491 	/*
1492 	 * set up and load the per-CPU TSS
1493 	 */
1494 	if (!oist->ist[0]) {
1495 		char *estacks = per_cpu(exception_stacks, cpu);
1496 
1497 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1498 			estacks += exception_stack_sizes[v];
1499 			oist->ist[v] = t->x86_tss.ist[v] =
1500 					(unsigned long)estacks;
1501 			if (v == DEBUG_STACK-1)
1502 				per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1503 		}
1504 	}
1505 
1506 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1507 
1508 	/*
1509 	 * <= is required because the CPU will access up to
1510 	 * 8 bits beyond the end of the IO permission bitmap.
1511 	 */
1512 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1513 		t->io_bitmap[i] = ~0UL;
1514 
1515 	mmgrab(&init_mm);
1516 	me->active_mm = &init_mm;
1517 	BUG_ON(me->mm);
1518 	enter_lazy_tlb(&init_mm, me);
1519 
1520 	load_sp0(t, &current->thread);
1521 	set_tss_desc(cpu, t);
1522 	load_TR_desc();
1523 	load_mm_ldt(&init_mm);
1524 
1525 	clear_all_debug_regs();
1526 	dbg_restore_debug_regs();
1527 
1528 	fpu__init_cpu();
1529 
1530 	if (is_uv_system())
1531 		uv_cpu_init();
1532 }
1533 
1534 #else
1535 
1536 void cpu_init(void)
1537 {
1538 	int cpu = smp_processor_id();
1539 	struct task_struct *curr = current;
1540 	struct tss_struct *t = &per_cpu(cpu_tss, cpu);
1541 	struct thread_struct *thread = &curr->thread;
1542 
1543 	wait_for_master_cpu(cpu);
1544 
1545 	/*
1546 	 * Initialize the CR4 shadow before doing anything that could
1547 	 * try to read it.
1548 	 */
1549 	cr4_init_shadow();
1550 
1551 	show_ucode_info_early();
1552 
1553 	pr_info("Initializing CPU#%d\n", cpu);
1554 
1555 	if (cpu_feature_enabled(X86_FEATURE_VME) ||
1556 	    boot_cpu_has(X86_FEATURE_TSC) ||
1557 	    boot_cpu_has(X86_FEATURE_DE))
1558 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1559 
1560 	load_current_idt();
1561 	switch_to_new_gdt(cpu);
1562 
1563 	/*
1564 	 * Set up and load the per-CPU TSS and LDT
1565 	 */
1566 	mmgrab(&init_mm);
1567 	curr->active_mm = &init_mm;
1568 	BUG_ON(curr->mm);
1569 	enter_lazy_tlb(&init_mm, curr);
1570 
1571 	load_sp0(t, thread);
1572 	set_tss_desc(cpu, t);
1573 	load_TR_desc();
1574 	load_mm_ldt(&init_mm);
1575 
1576 	t->x86_tss.io_bitmap_base = offsetof(struct tss_struct, io_bitmap);
1577 
1578 #ifdef CONFIG_DOUBLEFAULT
1579 	/* Set up doublefault TSS pointer in the GDT */
1580 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1581 #endif
1582 
1583 	clear_all_debug_regs();
1584 	dbg_restore_debug_regs();
1585 
1586 	fpu__init_cpu();
1587 }
1588 #endif
1589 
1590 static void bsp_resume(void)
1591 {
1592 	if (this_cpu->c_bsp_resume)
1593 		this_cpu->c_bsp_resume(&boot_cpu_data);
1594 }
1595 
1596 static struct syscore_ops cpu_syscore_ops = {
1597 	.resume		= bsp_resume,
1598 };
1599 
1600 static int __init init_cpu_syscore(void)
1601 {
1602 	register_syscore_ops(&cpu_syscore_ops);
1603 	return 0;
1604 }
1605 core_initcall(init_cpu_syscore);
1606