xref: /linux/arch/x86/kernel/cpu/common.c (revision 9f3926e08c26607a0dd5b1bc8a8aa1d03f72fcdc)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* cpu_feature_enabled() cannot be used this early */
3 #define USE_EARLY_PGTABLE_L5
4 
5 #include <linux/memblock.h>
6 #include <linux/linkage.h>
7 #include <linux/bitops.h>
8 #include <linux/kernel.h>
9 #include <linux/export.h>
10 #include <linux/percpu.h>
11 #include <linux/string.h>
12 #include <linux/ctype.h>
13 #include <linux/delay.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/clock.h>
16 #include <linux/sched/task.h>
17 #include <linux/init.h>
18 #include <linux/kprobes.h>
19 #include <linux/kgdb.h>
20 #include <linux/smp.h>
21 #include <linux/io.h>
22 #include <linux/syscore_ops.h>
23 
24 #include <asm/stackprotector.h>
25 #include <asm/perf_event.h>
26 #include <asm/mmu_context.h>
27 #include <asm/archrandom.h>
28 #include <asm/hypervisor.h>
29 #include <asm/processor.h>
30 #include <asm/tlbflush.h>
31 #include <asm/debugreg.h>
32 #include <asm/sections.h>
33 #include <asm/vsyscall.h>
34 #include <linux/topology.h>
35 #include <linux/cpumask.h>
36 #include <asm/pgtable.h>
37 #include <linux/atomic.h>
38 #include <asm/proto.h>
39 #include <asm/setup.h>
40 #include <asm/apic.h>
41 #include <asm/desc.h>
42 #include <asm/fpu/internal.h>
43 #include <asm/mtrr.h>
44 #include <asm/hwcap2.h>
45 #include <linux/numa.h>
46 #include <asm/asm.h>
47 #include <asm/bugs.h>
48 #include <asm/cpu.h>
49 #include <asm/mce.h>
50 #include <asm/msr.h>
51 #include <asm/pat.h>
52 #include <asm/microcode.h>
53 #include <asm/microcode_intel.h>
54 #include <asm/intel-family.h>
55 #include <asm/cpu_device_id.h>
56 
57 #ifdef CONFIG_X86_LOCAL_APIC
58 #include <asm/uv/uv.h>
59 #endif
60 
61 #include "cpu.h"
62 
63 u32 elf_hwcap2 __read_mostly;
64 
65 /* all of these masks are initialized in setup_cpu_local_masks() */
66 cpumask_var_t cpu_initialized_mask;
67 cpumask_var_t cpu_callout_mask;
68 cpumask_var_t cpu_callin_mask;
69 
70 /* representing cpus for which sibling maps can be computed */
71 cpumask_var_t cpu_sibling_setup_mask;
72 
73 /* Number of siblings per CPU package */
74 int smp_num_siblings = 1;
75 EXPORT_SYMBOL(smp_num_siblings);
76 
77 /* Last level cache ID of each logical CPU */
78 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
79 
80 /* correctly size the local cpu masks */
81 void __init setup_cpu_local_masks(void)
82 {
83 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
84 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
85 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
86 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
87 }
88 
89 static void default_init(struct cpuinfo_x86 *c)
90 {
91 #ifdef CONFIG_X86_64
92 	cpu_detect_cache_sizes(c);
93 #else
94 	/* Not much we can do here... */
95 	/* Check if at least it has cpuid */
96 	if (c->cpuid_level == -1) {
97 		/* No cpuid. It must be an ancient CPU */
98 		if (c->x86 == 4)
99 			strcpy(c->x86_model_id, "486");
100 		else if (c->x86 == 3)
101 			strcpy(c->x86_model_id, "386");
102 	}
103 #endif
104 }
105 
106 static const struct cpu_dev default_cpu = {
107 	.c_init		= default_init,
108 	.c_vendor	= "Unknown",
109 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
110 };
111 
112 static const struct cpu_dev *this_cpu = &default_cpu;
113 
114 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
115 #ifdef CONFIG_X86_64
116 	/*
117 	 * We need valid kernel segments for data and code in long mode too
118 	 * IRET will check the segment types  kkeil 2000/10/28
119 	 * Also sysret mandates a special GDT layout
120 	 *
121 	 * TLS descriptors are currently at a different place compared to i386.
122 	 * Hopefully nobody expects them at a fixed place (Wine?)
123 	 */
124 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
125 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
126 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
127 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
128 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
129 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
130 #else
131 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
132 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
133 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
134 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
135 	/*
136 	 * Segments used for calling PnP BIOS have byte granularity.
137 	 * They code segments and data segments have fixed 64k limits,
138 	 * the transfer segment sizes are set at run time.
139 	 */
140 	/* 32-bit code */
141 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
142 	/* 16-bit code */
143 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
144 	/* 16-bit data */
145 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
146 	/* 16-bit data */
147 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
148 	/* 16-bit data */
149 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
150 	/*
151 	 * The APM segments have byte granularity and their bases
152 	 * are set at run time.  All have 64k limits.
153 	 */
154 	/* 32-bit code */
155 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
156 	/* 16-bit code */
157 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
158 	/* data */
159 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
160 
161 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
162 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
163 	GDT_STACK_CANARY_INIT
164 #endif
165 } };
166 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
167 
168 static int __init x86_mpx_setup(char *s)
169 {
170 	/* require an exact match without trailing characters */
171 	if (strlen(s))
172 		return 0;
173 
174 	/* do not emit a message if the feature is not present */
175 	if (!boot_cpu_has(X86_FEATURE_MPX))
176 		return 1;
177 
178 	setup_clear_cpu_cap(X86_FEATURE_MPX);
179 	pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
180 	return 1;
181 }
182 __setup("nompx", x86_mpx_setup);
183 
184 #ifdef CONFIG_X86_64
185 static int __init x86_nopcid_setup(char *s)
186 {
187 	/* nopcid doesn't accept parameters */
188 	if (s)
189 		return -EINVAL;
190 
191 	/* do not emit a message if the feature is not present */
192 	if (!boot_cpu_has(X86_FEATURE_PCID))
193 		return 0;
194 
195 	setup_clear_cpu_cap(X86_FEATURE_PCID);
196 	pr_info("nopcid: PCID feature disabled\n");
197 	return 0;
198 }
199 early_param("nopcid", x86_nopcid_setup);
200 #endif
201 
202 static int __init x86_noinvpcid_setup(char *s)
203 {
204 	/* noinvpcid doesn't accept parameters */
205 	if (s)
206 		return -EINVAL;
207 
208 	/* do not emit a message if the feature is not present */
209 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
210 		return 0;
211 
212 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
213 	pr_info("noinvpcid: INVPCID feature disabled\n");
214 	return 0;
215 }
216 early_param("noinvpcid", x86_noinvpcid_setup);
217 
218 #ifdef CONFIG_X86_32
219 static int cachesize_override = -1;
220 static int disable_x86_serial_nr = 1;
221 
222 static int __init cachesize_setup(char *str)
223 {
224 	get_option(&str, &cachesize_override);
225 	return 1;
226 }
227 __setup("cachesize=", cachesize_setup);
228 
229 static int __init x86_sep_setup(char *s)
230 {
231 	setup_clear_cpu_cap(X86_FEATURE_SEP);
232 	return 1;
233 }
234 __setup("nosep", x86_sep_setup);
235 
236 /* Standard macro to see if a specific flag is changeable */
237 static inline int flag_is_changeable_p(u32 flag)
238 {
239 	u32 f1, f2;
240 
241 	/*
242 	 * Cyrix and IDT cpus allow disabling of CPUID
243 	 * so the code below may return different results
244 	 * when it is executed before and after enabling
245 	 * the CPUID. Add "volatile" to not allow gcc to
246 	 * optimize the subsequent calls to this function.
247 	 */
248 	asm volatile ("pushfl		\n\t"
249 		      "pushfl		\n\t"
250 		      "popl %0		\n\t"
251 		      "movl %0, %1	\n\t"
252 		      "xorl %2, %0	\n\t"
253 		      "pushl %0		\n\t"
254 		      "popfl		\n\t"
255 		      "pushfl		\n\t"
256 		      "popl %0		\n\t"
257 		      "popfl		\n\t"
258 
259 		      : "=&r" (f1), "=&r" (f2)
260 		      : "ir" (flag));
261 
262 	return ((f1^f2) & flag) != 0;
263 }
264 
265 /* Probe for the CPUID instruction */
266 int have_cpuid_p(void)
267 {
268 	return flag_is_changeable_p(X86_EFLAGS_ID);
269 }
270 
271 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
272 {
273 	unsigned long lo, hi;
274 
275 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
276 		return;
277 
278 	/* Disable processor serial number: */
279 
280 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
281 	lo |= 0x200000;
282 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
283 
284 	pr_notice("CPU serial number disabled.\n");
285 	clear_cpu_cap(c, X86_FEATURE_PN);
286 
287 	/* Disabling the serial number may affect the cpuid level */
288 	c->cpuid_level = cpuid_eax(0);
289 }
290 
291 static int __init x86_serial_nr_setup(char *s)
292 {
293 	disable_x86_serial_nr = 0;
294 	return 1;
295 }
296 __setup("serialnumber", x86_serial_nr_setup);
297 #else
298 static inline int flag_is_changeable_p(u32 flag)
299 {
300 	return 1;
301 }
302 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
303 {
304 }
305 #endif
306 
307 static __init int setup_disable_smep(char *arg)
308 {
309 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
310 	/* Check for things that depend on SMEP being enabled: */
311 	check_mpx_erratum(&boot_cpu_data);
312 	return 1;
313 }
314 __setup("nosmep", setup_disable_smep);
315 
316 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
317 {
318 	if (cpu_has(c, X86_FEATURE_SMEP))
319 		cr4_set_bits(X86_CR4_SMEP);
320 }
321 
322 static __init int setup_disable_smap(char *arg)
323 {
324 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
325 	return 1;
326 }
327 __setup("nosmap", setup_disable_smap);
328 
329 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
330 {
331 	unsigned long eflags = native_save_fl();
332 
333 	/* This should have been cleared long ago */
334 	BUG_ON(eflags & X86_EFLAGS_AC);
335 
336 	if (cpu_has(c, X86_FEATURE_SMAP)) {
337 #ifdef CONFIG_X86_SMAP
338 		cr4_set_bits(X86_CR4_SMAP);
339 #else
340 		cr4_clear_bits(X86_CR4_SMAP);
341 #endif
342 	}
343 }
344 
345 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
346 {
347 	/* Check the boot processor, plus build option for UMIP. */
348 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
349 		goto out;
350 
351 	/* Check the current processor's cpuid bits. */
352 	if (!cpu_has(c, X86_FEATURE_UMIP))
353 		goto out;
354 
355 	cr4_set_bits(X86_CR4_UMIP);
356 
357 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
358 
359 	return;
360 
361 out:
362 	/*
363 	 * Make sure UMIP is disabled in case it was enabled in a
364 	 * previous boot (e.g., via kexec).
365 	 */
366 	cr4_clear_bits(X86_CR4_UMIP);
367 }
368 
369 static __init int x86_nofsgsbase_setup(char *arg)
370 {
371 	/* Require an exact match without trailing characters. */
372 	if (strlen(arg))
373 		return 0;
374 
375 	/* Do not emit a message if the feature is not present. */
376 	if (!boot_cpu_has(X86_FEATURE_FSGSBASE))
377 		return 1;
378 
379 	setup_clear_cpu_cap(X86_FEATURE_FSGSBASE);
380 	pr_info("FSGSBASE disabled via kernel command line\n");
381 	return 1;
382 }
383 __setup("nofsgsbase", x86_nofsgsbase_setup);
384 
385 /*
386  * Protection Keys are not available in 32-bit mode.
387  */
388 static bool pku_disabled;
389 
390 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
391 {
392 	struct pkru_state *pk;
393 
394 	/* check the boot processor, plus compile options for PKU: */
395 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
396 		return;
397 	/* checks the actual processor's cpuid bits: */
398 	if (!cpu_has(c, X86_FEATURE_PKU))
399 		return;
400 	if (pku_disabled)
401 		return;
402 
403 	cr4_set_bits(X86_CR4_PKE);
404 	pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU);
405 	if (pk)
406 		pk->pkru = init_pkru_value;
407 	/*
408 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
409 	 * cpuid bit to be set.  We need to ensure that we
410 	 * update that bit in this CPU's "cpu_info".
411 	 */
412 	get_cpu_cap(c);
413 }
414 
415 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
416 static __init int setup_disable_pku(char *arg)
417 {
418 	/*
419 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
420 	 * runtime checks are against OSPKE so clearing the
421 	 * bit does nothing.
422 	 *
423 	 * This way, we will see "pku" in cpuinfo, but not
424 	 * "ospke", which is exactly what we want.  It shows
425 	 * that the CPU has PKU, but the OS has not enabled it.
426 	 * This happens to be exactly how a system would look
427 	 * if we disabled the config option.
428 	 */
429 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
430 	pku_disabled = true;
431 	return 1;
432 }
433 __setup("nopku", setup_disable_pku);
434 #endif /* CONFIG_X86_64 */
435 
436 /*
437  * Some CPU features depend on higher CPUID levels, which may not always
438  * be available due to CPUID level capping or broken virtualization
439  * software.  Add those features to this table to auto-disable them.
440  */
441 struct cpuid_dependent_feature {
442 	u32 feature;
443 	u32 level;
444 };
445 
446 static const struct cpuid_dependent_feature
447 cpuid_dependent_features[] = {
448 	{ X86_FEATURE_MWAIT,		0x00000005 },
449 	{ X86_FEATURE_DCA,		0x00000009 },
450 	{ X86_FEATURE_XSAVE,		0x0000000d },
451 	{ 0, 0 }
452 };
453 
454 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
455 {
456 	const struct cpuid_dependent_feature *df;
457 
458 	for (df = cpuid_dependent_features; df->feature; df++) {
459 
460 		if (!cpu_has(c, df->feature))
461 			continue;
462 		/*
463 		 * Note: cpuid_level is set to -1 if unavailable, but
464 		 * extended_extended_level is set to 0 if unavailable
465 		 * and the legitimate extended levels are all negative
466 		 * when signed; hence the weird messing around with
467 		 * signs here...
468 		 */
469 		if (!((s32)df->level < 0 ?
470 		     (u32)df->level > (u32)c->extended_cpuid_level :
471 		     (s32)df->level > (s32)c->cpuid_level))
472 			continue;
473 
474 		clear_cpu_cap(c, df->feature);
475 		if (!warn)
476 			continue;
477 
478 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
479 			x86_cap_flag(df->feature), df->level);
480 	}
481 }
482 
483 /*
484  * Naming convention should be: <Name> [(<Codename>)]
485  * This table only is used unless init_<vendor>() below doesn't set it;
486  * in particular, if CPUID levels 0x80000002..4 are supported, this
487  * isn't used
488  */
489 
490 /* Look up CPU names by table lookup. */
491 static const char *table_lookup_model(struct cpuinfo_x86 *c)
492 {
493 #ifdef CONFIG_X86_32
494 	const struct legacy_cpu_model_info *info;
495 
496 	if (c->x86_model >= 16)
497 		return NULL;	/* Range check */
498 
499 	if (!this_cpu)
500 		return NULL;
501 
502 	info = this_cpu->legacy_models;
503 
504 	while (info->family) {
505 		if (info->family == c->x86)
506 			return info->model_names[c->x86_model];
507 		info++;
508 	}
509 #endif
510 	return NULL;		/* Not found */
511 }
512 
513 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
514 __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
515 
516 void load_percpu_segment(int cpu)
517 {
518 #ifdef CONFIG_X86_32
519 	loadsegment(fs, __KERNEL_PERCPU);
520 #else
521 	__loadsegment_simple(gs, 0);
522 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
523 #endif
524 	load_stack_canary_segment();
525 }
526 
527 #ifdef CONFIG_X86_32
528 /* The 32-bit entry code needs to find cpu_entry_area. */
529 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
530 #endif
531 
532 /* Load the original GDT from the per-cpu structure */
533 void load_direct_gdt(int cpu)
534 {
535 	struct desc_ptr gdt_descr;
536 
537 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
538 	gdt_descr.size = GDT_SIZE - 1;
539 	load_gdt(&gdt_descr);
540 }
541 EXPORT_SYMBOL_GPL(load_direct_gdt);
542 
543 /* Load a fixmap remapping of the per-cpu GDT */
544 void load_fixmap_gdt(int cpu)
545 {
546 	struct desc_ptr gdt_descr;
547 
548 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
549 	gdt_descr.size = GDT_SIZE - 1;
550 	load_gdt(&gdt_descr);
551 }
552 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
553 
554 /*
555  * Current gdt points %fs at the "master" per-cpu area: after this,
556  * it's on the real one.
557  */
558 void switch_to_new_gdt(int cpu)
559 {
560 	/* Load the original GDT */
561 	load_direct_gdt(cpu);
562 	/* Reload the per-cpu base */
563 	load_percpu_segment(cpu);
564 }
565 
566 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
567 
568 static void get_model_name(struct cpuinfo_x86 *c)
569 {
570 	unsigned int *v;
571 	char *p, *q, *s;
572 
573 	if (c->extended_cpuid_level < 0x80000004)
574 		return;
575 
576 	v = (unsigned int *)c->x86_model_id;
577 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
578 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
579 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
580 	c->x86_model_id[48] = 0;
581 
582 	/* Trim whitespace */
583 	p = q = s = &c->x86_model_id[0];
584 
585 	while (*p == ' ')
586 		p++;
587 
588 	while (*p) {
589 		/* Note the last non-whitespace index */
590 		if (!isspace(*p))
591 			s = q;
592 
593 		*q++ = *p++;
594 	}
595 
596 	*(s + 1) = '\0';
597 }
598 
599 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
600 {
601 	unsigned int eax, ebx, ecx, edx;
602 
603 	c->x86_max_cores = 1;
604 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
605 		return;
606 
607 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
608 	if (eax & 0x1f)
609 		c->x86_max_cores = (eax >> 26) + 1;
610 }
611 
612 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
613 {
614 	unsigned int n, dummy, ebx, ecx, edx, l2size;
615 
616 	n = c->extended_cpuid_level;
617 
618 	if (n >= 0x80000005) {
619 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
620 		c->x86_cache_size = (ecx>>24) + (edx>>24);
621 #ifdef CONFIG_X86_64
622 		/* On K8 L1 TLB is inclusive, so don't count it */
623 		c->x86_tlbsize = 0;
624 #endif
625 	}
626 
627 	if (n < 0x80000006)	/* Some chips just has a large L1. */
628 		return;
629 
630 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
631 	l2size = ecx >> 16;
632 
633 #ifdef CONFIG_X86_64
634 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
635 #else
636 	/* do processor-specific cache resizing */
637 	if (this_cpu->legacy_cache_size)
638 		l2size = this_cpu->legacy_cache_size(c, l2size);
639 
640 	/* Allow user to override all this if necessary. */
641 	if (cachesize_override != -1)
642 		l2size = cachesize_override;
643 
644 	if (l2size == 0)
645 		return;		/* Again, no L2 cache is possible */
646 #endif
647 
648 	c->x86_cache_size = l2size;
649 }
650 
651 u16 __read_mostly tlb_lli_4k[NR_INFO];
652 u16 __read_mostly tlb_lli_2m[NR_INFO];
653 u16 __read_mostly tlb_lli_4m[NR_INFO];
654 u16 __read_mostly tlb_lld_4k[NR_INFO];
655 u16 __read_mostly tlb_lld_2m[NR_INFO];
656 u16 __read_mostly tlb_lld_4m[NR_INFO];
657 u16 __read_mostly tlb_lld_1g[NR_INFO];
658 
659 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
660 {
661 	if (this_cpu->c_detect_tlb)
662 		this_cpu->c_detect_tlb(c);
663 
664 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
665 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
666 		tlb_lli_4m[ENTRIES]);
667 
668 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
669 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
670 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
671 }
672 
673 int detect_ht_early(struct cpuinfo_x86 *c)
674 {
675 #ifdef CONFIG_SMP
676 	u32 eax, ebx, ecx, edx;
677 
678 	if (!cpu_has(c, X86_FEATURE_HT))
679 		return -1;
680 
681 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
682 		return -1;
683 
684 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
685 		return -1;
686 
687 	cpuid(1, &eax, &ebx, &ecx, &edx);
688 
689 	smp_num_siblings = (ebx & 0xff0000) >> 16;
690 	if (smp_num_siblings == 1)
691 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
692 #endif
693 	return 0;
694 }
695 
696 void detect_ht(struct cpuinfo_x86 *c)
697 {
698 #ifdef CONFIG_SMP
699 	int index_msb, core_bits;
700 
701 	if (detect_ht_early(c) < 0)
702 		return;
703 
704 	index_msb = get_count_order(smp_num_siblings);
705 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
706 
707 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
708 
709 	index_msb = get_count_order(smp_num_siblings);
710 
711 	core_bits = get_count_order(c->x86_max_cores);
712 
713 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
714 				       ((1 << core_bits) - 1);
715 #endif
716 }
717 
718 static void get_cpu_vendor(struct cpuinfo_x86 *c)
719 {
720 	char *v = c->x86_vendor_id;
721 	int i;
722 
723 	for (i = 0; i < X86_VENDOR_NUM; i++) {
724 		if (!cpu_devs[i])
725 			break;
726 
727 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
728 		    (cpu_devs[i]->c_ident[1] &&
729 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
730 
731 			this_cpu = cpu_devs[i];
732 			c->x86_vendor = this_cpu->c_x86_vendor;
733 			return;
734 		}
735 	}
736 
737 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
738 		    "CPU: Your system may be unstable.\n", v);
739 
740 	c->x86_vendor = X86_VENDOR_UNKNOWN;
741 	this_cpu = &default_cpu;
742 }
743 
744 void cpu_detect(struct cpuinfo_x86 *c)
745 {
746 	/* Get vendor name */
747 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
748 	      (unsigned int *)&c->x86_vendor_id[0],
749 	      (unsigned int *)&c->x86_vendor_id[8],
750 	      (unsigned int *)&c->x86_vendor_id[4]);
751 
752 	c->x86 = 4;
753 	/* Intel-defined flags: level 0x00000001 */
754 	if (c->cpuid_level >= 0x00000001) {
755 		u32 junk, tfms, cap0, misc;
756 
757 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
758 		c->x86		= x86_family(tfms);
759 		c->x86_model	= x86_model(tfms);
760 		c->x86_stepping	= x86_stepping(tfms);
761 
762 		if (cap0 & (1<<19)) {
763 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
764 			c->x86_cache_alignment = c->x86_clflush_size;
765 		}
766 	}
767 }
768 
769 static void apply_forced_caps(struct cpuinfo_x86 *c)
770 {
771 	int i;
772 
773 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
774 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
775 		c->x86_capability[i] |= cpu_caps_set[i];
776 	}
777 }
778 
779 static void init_speculation_control(struct cpuinfo_x86 *c)
780 {
781 	/*
782 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
783 	 * and they also have a different bit for STIBP support. Also,
784 	 * a hypervisor might have set the individual AMD bits even on
785 	 * Intel CPUs, for finer-grained selection of what's available.
786 	 */
787 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
788 		set_cpu_cap(c, X86_FEATURE_IBRS);
789 		set_cpu_cap(c, X86_FEATURE_IBPB);
790 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
791 	}
792 
793 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
794 		set_cpu_cap(c, X86_FEATURE_STIBP);
795 
796 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
797 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
798 		set_cpu_cap(c, X86_FEATURE_SSBD);
799 
800 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
801 		set_cpu_cap(c, X86_FEATURE_IBRS);
802 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
803 	}
804 
805 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
806 		set_cpu_cap(c, X86_FEATURE_IBPB);
807 
808 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
809 		set_cpu_cap(c, X86_FEATURE_STIBP);
810 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
811 	}
812 
813 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
814 		set_cpu_cap(c, X86_FEATURE_SSBD);
815 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
816 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
817 	}
818 }
819 
820 static void init_cqm(struct cpuinfo_x86 *c)
821 {
822 	if (!cpu_has(c, X86_FEATURE_CQM_LLC)) {
823 		c->x86_cache_max_rmid  = -1;
824 		c->x86_cache_occ_scale = -1;
825 		return;
826 	}
827 
828 	/* will be overridden if occupancy monitoring exists */
829 	c->x86_cache_max_rmid = cpuid_ebx(0xf);
830 
831 	if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC) ||
832 	    cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL) ||
833 	    cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)) {
834 		u32 eax, ebx, ecx, edx;
835 
836 		/* QoS sub-leaf, EAX=0Fh, ECX=1 */
837 		cpuid_count(0xf, 1, &eax, &ebx, &ecx, &edx);
838 
839 		c->x86_cache_max_rmid  = ecx;
840 		c->x86_cache_occ_scale = ebx;
841 	}
842 }
843 
844 void get_cpu_cap(struct cpuinfo_x86 *c)
845 {
846 	u32 eax, ebx, ecx, edx;
847 
848 	/* Intel-defined flags: level 0x00000001 */
849 	if (c->cpuid_level >= 0x00000001) {
850 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
851 
852 		c->x86_capability[CPUID_1_ECX] = ecx;
853 		c->x86_capability[CPUID_1_EDX] = edx;
854 	}
855 
856 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
857 	if (c->cpuid_level >= 0x00000006)
858 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
859 
860 	/* Additional Intel-defined flags: level 0x00000007 */
861 	if (c->cpuid_level >= 0x00000007) {
862 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
863 		c->x86_capability[CPUID_7_0_EBX] = ebx;
864 		c->x86_capability[CPUID_7_ECX] = ecx;
865 		c->x86_capability[CPUID_7_EDX] = edx;
866 
867 		/* Check valid sub-leaf index before accessing it */
868 		if (eax >= 1) {
869 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
870 			c->x86_capability[CPUID_7_1_EAX] = eax;
871 		}
872 	}
873 
874 	/* Extended state features: level 0x0000000d */
875 	if (c->cpuid_level >= 0x0000000d) {
876 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
877 
878 		c->x86_capability[CPUID_D_1_EAX] = eax;
879 	}
880 
881 	/* AMD-defined flags: level 0x80000001 */
882 	eax = cpuid_eax(0x80000000);
883 	c->extended_cpuid_level = eax;
884 
885 	if ((eax & 0xffff0000) == 0x80000000) {
886 		if (eax >= 0x80000001) {
887 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
888 
889 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
890 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
891 		}
892 	}
893 
894 	if (c->extended_cpuid_level >= 0x80000007) {
895 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
896 
897 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
898 		c->x86_power = edx;
899 	}
900 
901 	if (c->extended_cpuid_level >= 0x80000008) {
902 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
903 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
904 	}
905 
906 	if (c->extended_cpuid_level >= 0x8000000a)
907 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
908 
909 	init_scattered_cpuid_features(c);
910 	init_speculation_control(c);
911 	init_cqm(c);
912 
913 	/*
914 	 * Clear/Set all flags overridden by options, after probe.
915 	 * This needs to happen each time we re-probe, which may happen
916 	 * several times during CPU initialization.
917 	 */
918 	apply_forced_caps(c);
919 }
920 
921 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
922 {
923 	u32 eax, ebx, ecx, edx;
924 
925 	if (c->extended_cpuid_level >= 0x80000008) {
926 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
927 
928 		c->x86_virt_bits = (eax >> 8) & 0xff;
929 		c->x86_phys_bits = eax & 0xff;
930 	}
931 #ifdef CONFIG_X86_32
932 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
933 		c->x86_phys_bits = 36;
934 #endif
935 	c->x86_cache_bits = c->x86_phys_bits;
936 }
937 
938 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
939 {
940 #ifdef CONFIG_X86_32
941 	int i;
942 
943 	/*
944 	 * First of all, decide if this is a 486 or higher
945 	 * It's a 486 if we can modify the AC flag
946 	 */
947 	if (flag_is_changeable_p(X86_EFLAGS_AC))
948 		c->x86 = 4;
949 	else
950 		c->x86 = 3;
951 
952 	for (i = 0; i < X86_VENDOR_NUM; i++)
953 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
954 			c->x86_vendor_id[0] = 0;
955 			cpu_devs[i]->c_identify(c);
956 			if (c->x86_vendor_id[0]) {
957 				get_cpu_vendor(c);
958 				break;
959 			}
960 		}
961 #endif
962 }
963 
964 #define NO_SPECULATION	BIT(0)
965 #define NO_MELTDOWN	BIT(1)
966 #define NO_SSB		BIT(2)
967 #define NO_L1TF		BIT(3)
968 #define NO_MDS		BIT(4)
969 #define MSBDS_ONLY	BIT(5)
970 
971 #define VULNWL(_vendor, _family, _model, _whitelist)	\
972 	{ X86_VENDOR_##_vendor, _family, _model, X86_FEATURE_ANY, _whitelist }
973 
974 #define VULNWL_INTEL(model, whitelist)		\
975 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
976 
977 #define VULNWL_AMD(family, whitelist)		\
978 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
979 
980 #define VULNWL_HYGON(family, whitelist)		\
981 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
982 
983 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
984 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
985 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
986 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
987 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
988 
989 	/* Intel Family 6 */
990 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION),
991 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION),
992 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION),
993 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION),
994 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION),
995 
996 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY),
997 	VULNWL_INTEL(ATOM_SILVERMONT_X,		NO_SSB | NO_L1TF | MSBDS_ONLY),
998 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY),
999 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY),
1000 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY),
1001 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY),
1002 
1003 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1004 
1005 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY),
1006 
1007 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF),
1008 	VULNWL_INTEL(ATOM_GOLDMONT_X,		NO_MDS | NO_L1TF),
1009 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF),
1010 
1011 	/* AMD Family 0xf - 0x12 */
1012 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
1013 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
1014 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
1015 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
1016 
1017 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1018 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS),
1019 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS),
1020 	{}
1021 };
1022 
1023 static bool __init cpu_matches(unsigned long which)
1024 {
1025 	const struct x86_cpu_id *m = x86_match_cpu(cpu_vuln_whitelist);
1026 
1027 	return m && !!(m->driver_data & which);
1028 }
1029 
1030 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1031 {
1032 	u64 ia32_cap = 0;
1033 
1034 	if (cpu_matches(NO_SPECULATION))
1035 		return;
1036 
1037 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1038 	setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1039 
1040 	if (cpu_has(c, X86_FEATURE_ARCH_CAPABILITIES))
1041 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1042 
1043 	if (!cpu_matches(NO_SSB) && !(ia32_cap & ARCH_CAP_SSB_NO) &&
1044 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1045 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1046 
1047 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1048 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1049 
1050 	if (!cpu_matches(NO_MDS) && !(ia32_cap & ARCH_CAP_MDS_NO)) {
1051 		setup_force_cpu_bug(X86_BUG_MDS);
1052 		if (cpu_matches(MSBDS_ONLY))
1053 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1054 	}
1055 
1056 	if (cpu_matches(NO_MELTDOWN))
1057 		return;
1058 
1059 	/* Rogue Data Cache Load? No! */
1060 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1061 		return;
1062 
1063 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1064 
1065 	if (cpu_matches(NO_L1TF))
1066 		return;
1067 
1068 	setup_force_cpu_bug(X86_BUG_L1TF);
1069 }
1070 
1071 /*
1072  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1073  * unfortunately, that's not true in practice because of early VIA
1074  * chips and (more importantly) broken virtualizers that are not easy
1075  * to detect. In the latter case it doesn't even *fail* reliably, so
1076  * probing for it doesn't even work. Disable it completely on 32-bit
1077  * unless we can find a reliable way to detect all the broken cases.
1078  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1079  */
1080 static void detect_nopl(void)
1081 {
1082 #ifdef CONFIG_X86_32
1083 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1084 #else
1085 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1086 #endif
1087 }
1088 
1089 /*
1090  * Do minimum CPU detection early.
1091  * Fields really needed: vendor, cpuid_level, family, model, mask,
1092  * cache alignment.
1093  * The others are not touched to avoid unwanted side effects.
1094  *
1095  * WARNING: this function is only called on the boot CPU.  Don't add code
1096  * here that is supposed to run on all CPUs.
1097  */
1098 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1099 {
1100 #ifdef CONFIG_X86_64
1101 	c->x86_clflush_size = 64;
1102 	c->x86_phys_bits = 36;
1103 	c->x86_virt_bits = 48;
1104 #else
1105 	c->x86_clflush_size = 32;
1106 	c->x86_phys_bits = 32;
1107 	c->x86_virt_bits = 32;
1108 #endif
1109 	c->x86_cache_alignment = c->x86_clflush_size;
1110 
1111 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1112 	c->extended_cpuid_level = 0;
1113 
1114 	if (!have_cpuid_p())
1115 		identify_cpu_without_cpuid(c);
1116 
1117 	/* cyrix could have cpuid enabled via c_identify()*/
1118 	if (have_cpuid_p()) {
1119 		cpu_detect(c);
1120 		get_cpu_vendor(c);
1121 		get_cpu_cap(c);
1122 		get_cpu_address_sizes(c);
1123 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1124 
1125 		if (this_cpu->c_early_init)
1126 			this_cpu->c_early_init(c);
1127 
1128 		c->cpu_index = 0;
1129 		filter_cpuid_features(c, false);
1130 
1131 		if (this_cpu->c_bsp_init)
1132 			this_cpu->c_bsp_init(c);
1133 	} else {
1134 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1135 	}
1136 
1137 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1138 
1139 	cpu_set_bug_bits(c);
1140 
1141 	fpu__init_system(c);
1142 
1143 #ifdef CONFIG_X86_32
1144 	/*
1145 	 * Regardless of whether PCID is enumerated, the SDM says
1146 	 * that it can't be enabled in 32-bit mode.
1147 	 */
1148 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1149 #endif
1150 
1151 	/*
1152 	 * Later in the boot process pgtable_l5_enabled() relies on
1153 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1154 	 * enabled by this point we need to clear the feature bit to avoid
1155 	 * false-positives at the later stage.
1156 	 *
1157 	 * pgtable_l5_enabled() can be false here for several reasons:
1158 	 *  - 5-level paging is disabled compile-time;
1159 	 *  - it's 32-bit kernel;
1160 	 *  - machine doesn't support 5-level paging;
1161 	 *  - user specified 'no5lvl' in kernel command line.
1162 	 */
1163 	if (!pgtable_l5_enabled())
1164 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1165 
1166 	detect_nopl();
1167 }
1168 
1169 void __init early_cpu_init(void)
1170 {
1171 	const struct cpu_dev *const *cdev;
1172 	int count = 0;
1173 
1174 #ifdef CONFIG_PROCESSOR_SELECT
1175 	pr_info("KERNEL supported cpus:\n");
1176 #endif
1177 
1178 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1179 		const struct cpu_dev *cpudev = *cdev;
1180 
1181 		if (count >= X86_VENDOR_NUM)
1182 			break;
1183 		cpu_devs[count] = cpudev;
1184 		count++;
1185 
1186 #ifdef CONFIG_PROCESSOR_SELECT
1187 		{
1188 			unsigned int j;
1189 
1190 			for (j = 0; j < 2; j++) {
1191 				if (!cpudev->c_ident[j])
1192 					continue;
1193 				pr_info("  %s %s\n", cpudev->c_vendor,
1194 					cpudev->c_ident[j]);
1195 			}
1196 		}
1197 #endif
1198 	}
1199 	early_identify_cpu(&boot_cpu_data);
1200 }
1201 
1202 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1203 {
1204 #ifdef CONFIG_X86_64
1205 	/*
1206 	 * Empirically, writing zero to a segment selector on AMD does
1207 	 * not clear the base, whereas writing zero to a segment
1208 	 * selector on Intel does clear the base.  Intel's behavior
1209 	 * allows slightly faster context switches in the common case
1210 	 * where GS is unused by the prev and next threads.
1211 	 *
1212 	 * Since neither vendor documents this anywhere that I can see,
1213 	 * detect it directly instead of hardcoding the choice by
1214 	 * vendor.
1215 	 *
1216 	 * I've designated AMD's behavior as the "bug" because it's
1217 	 * counterintuitive and less friendly.
1218 	 */
1219 
1220 	unsigned long old_base, tmp;
1221 	rdmsrl(MSR_FS_BASE, old_base);
1222 	wrmsrl(MSR_FS_BASE, 1);
1223 	loadsegment(fs, 0);
1224 	rdmsrl(MSR_FS_BASE, tmp);
1225 	if (tmp != 0)
1226 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1227 	wrmsrl(MSR_FS_BASE, old_base);
1228 #endif
1229 }
1230 
1231 static void generic_identify(struct cpuinfo_x86 *c)
1232 {
1233 	c->extended_cpuid_level = 0;
1234 
1235 	if (!have_cpuid_p())
1236 		identify_cpu_without_cpuid(c);
1237 
1238 	/* cyrix could have cpuid enabled via c_identify()*/
1239 	if (!have_cpuid_p())
1240 		return;
1241 
1242 	cpu_detect(c);
1243 
1244 	get_cpu_vendor(c);
1245 
1246 	get_cpu_cap(c);
1247 
1248 	get_cpu_address_sizes(c);
1249 
1250 	if (c->cpuid_level >= 0x00000001) {
1251 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1252 #ifdef CONFIG_X86_32
1253 # ifdef CONFIG_SMP
1254 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1255 # else
1256 		c->apicid = c->initial_apicid;
1257 # endif
1258 #endif
1259 		c->phys_proc_id = c->initial_apicid;
1260 	}
1261 
1262 	get_model_name(c); /* Default name */
1263 
1264 	detect_null_seg_behavior(c);
1265 
1266 	/*
1267 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1268 	 * systems that run Linux at CPL > 0 may or may not have the
1269 	 * issue, but, even if they have the issue, there's absolutely
1270 	 * nothing we can do about it because we can't use the real IRET
1271 	 * instruction.
1272 	 *
1273 	 * NB: For the time being, only 32-bit kernels support
1274 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1275 	 * whether to apply espfix using paravirt hooks.  If any
1276 	 * non-paravirt system ever shows up that does *not* have the
1277 	 * ESPFIX issue, we can change this.
1278 	 */
1279 #ifdef CONFIG_X86_32
1280 # ifdef CONFIG_PARAVIRT_XXL
1281 	do {
1282 		extern void native_iret(void);
1283 		if (pv_ops.cpu.iret == native_iret)
1284 			set_cpu_bug(c, X86_BUG_ESPFIX);
1285 	} while (0);
1286 # else
1287 	set_cpu_bug(c, X86_BUG_ESPFIX);
1288 # endif
1289 #endif
1290 }
1291 
1292 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
1293 {
1294 	/*
1295 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
1296 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
1297 	 * in case CQM bits really aren't there in this CPU.
1298 	 */
1299 	if (c != &boot_cpu_data) {
1300 		boot_cpu_data.x86_cache_max_rmid =
1301 			min(boot_cpu_data.x86_cache_max_rmid,
1302 			    c->x86_cache_max_rmid);
1303 	}
1304 }
1305 
1306 /*
1307  * Validate that ACPI/mptables have the same information about the
1308  * effective APIC id and update the package map.
1309  */
1310 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1311 {
1312 #ifdef CONFIG_SMP
1313 	unsigned int apicid, cpu = smp_processor_id();
1314 
1315 	apicid = apic->cpu_present_to_apicid(cpu);
1316 
1317 	if (apicid != c->apicid) {
1318 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1319 		       cpu, apicid, c->initial_apicid);
1320 	}
1321 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1322 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1323 #else
1324 	c->logical_proc_id = 0;
1325 #endif
1326 }
1327 
1328 /*
1329  * This does the hard work of actually picking apart the CPU stuff...
1330  */
1331 static void identify_cpu(struct cpuinfo_x86 *c)
1332 {
1333 	int i;
1334 
1335 	c->loops_per_jiffy = loops_per_jiffy;
1336 	c->x86_cache_size = 0;
1337 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1338 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1339 	c->x86_vendor_id[0] = '\0'; /* Unset */
1340 	c->x86_model_id[0] = '\0';  /* Unset */
1341 	c->x86_max_cores = 1;
1342 	c->x86_coreid_bits = 0;
1343 	c->cu_id = 0xff;
1344 #ifdef CONFIG_X86_64
1345 	c->x86_clflush_size = 64;
1346 	c->x86_phys_bits = 36;
1347 	c->x86_virt_bits = 48;
1348 #else
1349 	c->cpuid_level = -1;	/* CPUID not detected */
1350 	c->x86_clflush_size = 32;
1351 	c->x86_phys_bits = 32;
1352 	c->x86_virt_bits = 32;
1353 #endif
1354 	c->x86_cache_alignment = c->x86_clflush_size;
1355 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1356 
1357 	generic_identify(c);
1358 
1359 	if (this_cpu->c_identify)
1360 		this_cpu->c_identify(c);
1361 
1362 	/* Clear/Set all flags overridden by options, after probe */
1363 	apply_forced_caps(c);
1364 
1365 #ifdef CONFIG_X86_64
1366 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1367 #endif
1368 
1369 	/*
1370 	 * Vendor-specific initialization.  In this section we
1371 	 * canonicalize the feature flags, meaning if there are
1372 	 * features a certain CPU supports which CPUID doesn't
1373 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1374 	 * we handle them here.
1375 	 *
1376 	 * At the end of this section, c->x86_capability better
1377 	 * indicate the features this CPU genuinely supports!
1378 	 */
1379 	if (this_cpu->c_init)
1380 		this_cpu->c_init(c);
1381 
1382 	/* Disable the PN if appropriate */
1383 	squash_the_stupid_serial_number(c);
1384 
1385 	/* Set up SMEP/SMAP/UMIP */
1386 	setup_smep(c);
1387 	setup_smap(c);
1388 	setup_umip(c);
1389 
1390 	/* Enable FSGSBASE instructions if available. */
1391 	if (cpu_has(c, X86_FEATURE_FSGSBASE)) {
1392 		cr4_set_bits(X86_CR4_FSGSBASE);
1393 		elf_hwcap2 |= HWCAP2_FSGSBASE;
1394 	}
1395 
1396 	/*
1397 	 * The vendor-specific functions might have changed features.
1398 	 * Now we do "generic changes."
1399 	 */
1400 
1401 	/* Filter out anything that depends on CPUID levels we don't have */
1402 	filter_cpuid_features(c, true);
1403 
1404 	/* If the model name is still unset, do table lookup. */
1405 	if (!c->x86_model_id[0]) {
1406 		const char *p;
1407 		p = table_lookup_model(c);
1408 		if (p)
1409 			strcpy(c->x86_model_id, p);
1410 		else
1411 			/* Last resort... */
1412 			sprintf(c->x86_model_id, "%02x/%02x",
1413 				c->x86, c->x86_model);
1414 	}
1415 
1416 #ifdef CONFIG_X86_64
1417 	detect_ht(c);
1418 #endif
1419 
1420 	x86_init_rdrand(c);
1421 	x86_init_cache_qos(c);
1422 	setup_pku(c);
1423 
1424 	/*
1425 	 * Clear/Set all flags overridden by options, need do it
1426 	 * before following smp all cpus cap AND.
1427 	 */
1428 	apply_forced_caps(c);
1429 
1430 	/*
1431 	 * On SMP, boot_cpu_data holds the common feature set between
1432 	 * all CPUs; so make sure that we indicate which features are
1433 	 * common between the CPUs.  The first time this routine gets
1434 	 * executed, c == &boot_cpu_data.
1435 	 */
1436 	if (c != &boot_cpu_data) {
1437 		/* AND the already accumulated flags with these */
1438 		for (i = 0; i < NCAPINTS; i++)
1439 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1440 
1441 		/* OR, i.e. replicate the bug flags */
1442 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1443 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1444 	}
1445 
1446 	/* Init Machine Check Exception if available. */
1447 	mcheck_cpu_init(c);
1448 
1449 	select_idle_routine(c);
1450 
1451 #ifdef CONFIG_NUMA
1452 	numa_add_cpu(smp_processor_id());
1453 #endif
1454 }
1455 
1456 /*
1457  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1458  * on 32-bit kernels:
1459  */
1460 #ifdef CONFIG_X86_32
1461 void enable_sep_cpu(void)
1462 {
1463 	struct tss_struct *tss;
1464 	int cpu;
1465 
1466 	if (!boot_cpu_has(X86_FEATURE_SEP))
1467 		return;
1468 
1469 	cpu = get_cpu();
1470 	tss = &per_cpu(cpu_tss_rw, cpu);
1471 
1472 	/*
1473 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1474 	 * see the big comment in struct x86_hw_tss's definition.
1475 	 */
1476 
1477 	tss->x86_tss.ss1 = __KERNEL_CS;
1478 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1479 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1480 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1481 
1482 	put_cpu();
1483 }
1484 #endif
1485 
1486 void __init identify_boot_cpu(void)
1487 {
1488 	identify_cpu(&boot_cpu_data);
1489 #ifdef CONFIG_X86_32
1490 	sysenter_setup();
1491 	enable_sep_cpu();
1492 #endif
1493 	cpu_detect_tlb(&boot_cpu_data);
1494 }
1495 
1496 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1497 {
1498 	BUG_ON(c == &boot_cpu_data);
1499 	identify_cpu(c);
1500 #ifdef CONFIG_X86_32
1501 	enable_sep_cpu();
1502 #endif
1503 	mtrr_ap_init();
1504 	validate_apic_and_package_id(c);
1505 	x86_spec_ctrl_setup_ap();
1506 }
1507 
1508 static __init int setup_noclflush(char *arg)
1509 {
1510 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1511 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1512 	return 1;
1513 }
1514 __setup("noclflush", setup_noclflush);
1515 
1516 void print_cpu_info(struct cpuinfo_x86 *c)
1517 {
1518 	const char *vendor = NULL;
1519 
1520 	if (c->x86_vendor < X86_VENDOR_NUM) {
1521 		vendor = this_cpu->c_vendor;
1522 	} else {
1523 		if (c->cpuid_level >= 0)
1524 			vendor = c->x86_vendor_id;
1525 	}
1526 
1527 	if (vendor && !strstr(c->x86_model_id, vendor))
1528 		pr_cont("%s ", vendor);
1529 
1530 	if (c->x86_model_id[0])
1531 		pr_cont("%s", c->x86_model_id);
1532 	else
1533 		pr_cont("%d86", c->x86);
1534 
1535 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1536 
1537 	if (c->x86_stepping || c->cpuid_level >= 0)
1538 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1539 	else
1540 		pr_cont(")\n");
1541 }
1542 
1543 /*
1544  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1545  * But we need to keep a dummy __setup around otherwise it would
1546  * show up as an environment variable for init.
1547  */
1548 static __init int setup_clearcpuid(char *arg)
1549 {
1550 	return 1;
1551 }
1552 __setup("clearcpuid=", setup_clearcpuid);
1553 
1554 #ifdef CONFIG_X86_64
1555 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1556 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1557 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1558 
1559 /*
1560  * The following percpu variables are hot.  Align current_task to
1561  * cacheline size such that they fall in the same cacheline.
1562  */
1563 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1564 	&init_task;
1565 EXPORT_PER_CPU_SYMBOL(current_task);
1566 
1567 DEFINE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
1568 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1569 
1570 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1571 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1572 
1573 /* May not be marked __init: used by software suspend */
1574 void syscall_init(void)
1575 {
1576 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1577 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1578 
1579 #ifdef CONFIG_IA32_EMULATION
1580 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1581 	/*
1582 	 * This only works on Intel CPUs.
1583 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1584 	 * This does not cause SYSENTER to jump to the wrong location, because
1585 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1586 	 */
1587 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1588 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1589 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1590 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1591 #else
1592 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1593 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1594 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1595 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1596 #endif
1597 
1598 	/* Flags to clear on syscall */
1599 	wrmsrl(MSR_SYSCALL_MASK,
1600 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1601 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1602 }
1603 
1604 DEFINE_PER_CPU(int, debug_stack_usage);
1605 DEFINE_PER_CPU(u32, debug_idt_ctr);
1606 
1607 void debug_stack_set_zero(void)
1608 {
1609 	this_cpu_inc(debug_idt_ctr);
1610 	load_current_idt();
1611 }
1612 NOKPROBE_SYMBOL(debug_stack_set_zero);
1613 
1614 void debug_stack_reset(void)
1615 {
1616 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1617 		return;
1618 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1619 		load_current_idt();
1620 }
1621 NOKPROBE_SYMBOL(debug_stack_reset);
1622 
1623 #else	/* CONFIG_X86_64 */
1624 
1625 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1626 EXPORT_PER_CPU_SYMBOL(current_task);
1627 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1628 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1629 
1630 /*
1631  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1632  * the top of the kernel stack.  Use an extra percpu variable to track the
1633  * top of the kernel stack directly.
1634  */
1635 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1636 	(unsigned long)&init_thread_union + THREAD_SIZE;
1637 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1638 
1639 #ifdef CONFIG_STACKPROTECTOR
1640 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1641 #endif
1642 
1643 #endif	/* CONFIG_X86_64 */
1644 
1645 /*
1646  * Clear all 6 debug registers:
1647  */
1648 static void clear_all_debug_regs(void)
1649 {
1650 	int i;
1651 
1652 	for (i = 0; i < 8; i++) {
1653 		/* Ignore db4, db5 */
1654 		if ((i == 4) || (i == 5))
1655 			continue;
1656 
1657 		set_debugreg(0, i);
1658 	}
1659 }
1660 
1661 #ifdef CONFIG_KGDB
1662 /*
1663  * Restore debug regs if using kgdbwait and you have a kernel debugger
1664  * connection established.
1665  */
1666 static void dbg_restore_debug_regs(void)
1667 {
1668 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1669 		arch_kgdb_ops.correct_hw_break();
1670 }
1671 #else /* ! CONFIG_KGDB */
1672 #define dbg_restore_debug_regs()
1673 #endif /* ! CONFIG_KGDB */
1674 
1675 static void wait_for_master_cpu(int cpu)
1676 {
1677 #ifdef CONFIG_SMP
1678 	/*
1679 	 * wait for ACK from master CPU before continuing
1680 	 * with AP initialization
1681 	 */
1682 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1683 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1684 		cpu_relax();
1685 #endif
1686 }
1687 
1688 #ifdef CONFIG_X86_64
1689 static void setup_getcpu(int cpu)
1690 {
1691 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1692 	struct desc_struct d = { };
1693 
1694 	if (boot_cpu_has(X86_FEATURE_RDTSCP))
1695 		write_rdtscp_aux(cpudata);
1696 
1697 	/* Store CPU and node number in limit. */
1698 	d.limit0 = cpudata;
1699 	d.limit1 = cpudata >> 16;
1700 
1701 	d.type = 5;		/* RO data, expand down, accessed */
1702 	d.dpl = 3;		/* Visible to user code */
1703 	d.s = 1;		/* Not a system segment */
1704 	d.p = 1;		/* Present */
1705 	d.d = 1;		/* 32-bit */
1706 
1707 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1708 }
1709 #endif
1710 
1711 /*
1712  * cpu_init() initializes state that is per-CPU. Some data is already
1713  * initialized (naturally) in the bootstrap process, such as the GDT
1714  * and IDT. We reload them nevertheless, this function acts as a
1715  * 'CPU state barrier', nothing should get across.
1716  */
1717 #ifdef CONFIG_X86_64
1718 
1719 void cpu_init(void)
1720 {
1721 	int cpu = raw_smp_processor_id();
1722 	struct task_struct *me;
1723 	struct tss_struct *t;
1724 	int i;
1725 
1726 	wait_for_master_cpu(cpu);
1727 
1728 	/*
1729 	 * Initialize the CR4 shadow before doing anything that could
1730 	 * try to read it.
1731 	 */
1732 	cr4_init_shadow();
1733 
1734 	if (cpu)
1735 		load_ucode_ap();
1736 
1737 	t = &per_cpu(cpu_tss_rw, cpu);
1738 
1739 #ifdef CONFIG_NUMA
1740 	if (this_cpu_read(numa_node) == 0 &&
1741 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1742 		set_numa_node(early_cpu_to_node(cpu));
1743 #endif
1744 	setup_getcpu(cpu);
1745 
1746 	me = current;
1747 
1748 	pr_debug("Initializing CPU#%d\n", cpu);
1749 
1750 	cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1751 
1752 	/*
1753 	 * Initialize the per-CPU GDT with the boot GDT,
1754 	 * and set up the GDT descriptor:
1755 	 */
1756 
1757 	switch_to_new_gdt(cpu);
1758 	loadsegment(fs, 0);
1759 
1760 	load_current_idt();
1761 
1762 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1763 	syscall_init();
1764 
1765 	wrmsrl(MSR_FS_BASE, 0);
1766 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1767 	barrier();
1768 
1769 	x86_configure_nx();
1770 	x2apic_setup();
1771 
1772 	/*
1773 	 * set up and load the per-CPU TSS
1774 	 */
1775 	if (!t->x86_tss.ist[0]) {
1776 		t->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1777 		t->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1778 		t->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1779 		t->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1780 	}
1781 
1782 	t->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
1783 
1784 	/*
1785 	 * <= is required because the CPU will access up to
1786 	 * 8 bits beyond the end of the IO permission bitmap.
1787 	 */
1788 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1789 		t->io_bitmap[i] = ~0UL;
1790 
1791 	mmgrab(&init_mm);
1792 	me->active_mm = &init_mm;
1793 	BUG_ON(me->mm);
1794 	initialize_tlbstate_and_flush();
1795 	enter_lazy_tlb(&init_mm, me);
1796 
1797 	/*
1798 	 * Initialize the TSS.  sp0 points to the entry trampoline stack
1799 	 * regardless of what task is running.
1800 	 */
1801 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1802 	load_TR_desc();
1803 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1804 
1805 	load_mm_ldt(&init_mm);
1806 
1807 	clear_all_debug_regs();
1808 	dbg_restore_debug_regs();
1809 
1810 	fpu__init_cpu();
1811 
1812 	if (is_uv_system())
1813 		uv_cpu_init();
1814 
1815 	load_fixmap_gdt(cpu);
1816 }
1817 
1818 #else
1819 
1820 void cpu_init(void)
1821 {
1822 	int cpu = smp_processor_id();
1823 	struct task_struct *curr = current;
1824 	struct tss_struct *t = &per_cpu(cpu_tss_rw, cpu);
1825 
1826 	wait_for_master_cpu(cpu);
1827 
1828 	/*
1829 	 * Initialize the CR4 shadow before doing anything that could
1830 	 * try to read it.
1831 	 */
1832 	cr4_init_shadow();
1833 
1834 	show_ucode_info_early();
1835 
1836 	pr_info("Initializing CPU#%d\n", cpu);
1837 
1838 	if (cpu_feature_enabled(X86_FEATURE_VME) ||
1839 	    boot_cpu_has(X86_FEATURE_TSC) ||
1840 	    boot_cpu_has(X86_FEATURE_DE))
1841 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1842 
1843 	load_current_idt();
1844 	switch_to_new_gdt(cpu);
1845 
1846 	/*
1847 	 * Set up and load the per-CPU TSS and LDT
1848 	 */
1849 	mmgrab(&init_mm);
1850 	curr->active_mm = &init_mm;
1851 	BUG_ON(curr->mm);
1852 	initialize_tlbstate_and_flush();
1853 	enter_lazy_tlb(&init_mm, curr);
1854 
1855 	/*
1856 	 * Initialize the TSS.  sp0 points to the entry trampoline stack
1857 	 * regardless of what task is running.
1858 	 */
1859 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1860 	load_TR_desc();
1861 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1862 
1863 	load_mm_ldt(&init_mm);
1864 
1865 	t->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
1866 
1867 #ifdef CONFIG_DOUBLEFAULT
1868 	/* Set up doublefault TSS pointer in the GDT */
1869 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1870 #endif
1871 
1872 	clear_all_debug_regs();
1873 	dbg_restore_debug_regs();
1874 
1875 	fpu__init_cpu();
1876 
1877 	load_fixmap_gdt(cpu);
1878 }
1879 #endif
1880 
1881 /*
1882  * The microcode loader calls this upon late microcode load to recheck features,
1883  * only when microcode has been updated. Caller holds microcode_mutex and CPU
1884  * hotplug lock.
1885  */
1886 void microcode_check(void)
1887 {
1888 	struct cpuinfo_x86 info;
1889 
1890 	perf_check_microcode();
1891 
1892 	/* Reload CPUID max function as it might've changed. */
1893 	info.cpuid_level = cpuid_eax(0);
1894 
1895 	/*
1896 	 * Copy all capability leafs to pick up the synthetic ones so that
1897 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
1898 	 * get overwritten in get_cpu_cap().
1899 	 */
1900 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
1901 
1902 	get_cpu_cap(&info);
1903 
1904 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
1905 		return;
1906 
1907 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
1908 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
1909 }
1910