xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 1a59d1b8)
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 /*
370  * Protection Keys are not available in 32-bit mode.
371  */
372 static bool pku_disabled;
373 
374 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
375 {
376 	struct pkru_state *pk;
377 
378 	/* check the boot processor, plus compile options for PKU: */
379 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
380 		return;
381 	/* checks the actual processor's cpuid bits: */
382 	if (!cpu_has(c, X86_FEATURE_PKU))
383 		return;
384 	if (pku_disabled)
385 		return;
386 
387 	cr4_set_bits(X86_CR4_PKE);
388 	pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU);
389 	if (pk)
390 		pk->pkru = init_pkru_value;
391 	/*
392 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
393 	 * cpuid bit to be set.  We need to ensure that we
394 	 * update that bit in this CPU's "cpu_info".
395 	 */
396 	get_cpu_cap(c);
397 }
398 
399 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
400 static __init int setup_disable_pku(char *arg)
401 {
402 	/*
403 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
404 	 * runtime checks are against OSPKE so clearing the
405 	 * bit does nothing.
406 	 *
407 	 * This way, we will see "pku" in cpuinfo, but not
408 	 * "ospke", which is exactly what we want.  It shows
409 	 * that the CPU has PKU, but the OS has not enabled it.
410 	 * This happens to be exactly how a system would look
411 	 * if we disabled the config option.
412 	 */
413 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
414 	pku_disabled = true;
415 	return 1;
416 }
417 __setup("nopku", setup_disable_pku);
418 #endif /* CONFIG_X86_64 */
419 
420 /*
421  * Some CPU features depend on higher CPUID levels, which may not always
422  * be available due to CPUID level capping or broken virtualization
423  * software.  Add those features to this table to auto-disable them.
424  */
425 struct cpuid_dependent_feature {
426 	u32 feature;
427 	u32 level;
428 };
429 
430 static const struct cpuid_dependent_feature
431 cpuid_dependent_features[] = {
432 	{ X86_FEATURE_MWAIT,		0x00000005 },
433 	{ X86_FEATURE_DCA,		0x00000009 },
434 	{ X86_FEATURE_XSAVE,		0x0000000d },
435 	{ 0, 0 }
436 };
437 
438 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
439 {
440 	const struct cpuid_dependent_feature *df;
441 
442 	for (df = cpuid_dependent_features; df->feature; df++) {
443 
444 		if (!cpu_has(c, df->feature))
445 			continue;
446 		/*
447 		 * Note: cpuid_level is set to -1 if unavailable, but
448 		 * extended_extended_level is set to 0 if unavailable
449 		 * and the legitimate extended levels are all negative
450 		 * when signed; hence the weird messing around with
451 		 * signs here...
452 		 */
453 		if (!((s32)df->level < 0 ?
454 		     (u32)df->level > (u32)c->extended_cpuid_level :
455 		     (s32)df->level > (s32)c->cpuid_level))
456 			continue;
457 
458 		clear_cpu_cap(c, df->feature);
459 		if (!warn)
460 			continue;
461 
462 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
463 			x86_cap_flag(df->feature), df->level);
464 	}
465 }
466 
467 /*
468  * Naming convention should be: <Name> [(<Codename>)]
469  * This table only is used unless init_<vendor>() below doesn't set it;
470  * in particular, if CPUID levels 0x80000002..4 are supported, this
471  * isn't used
472  */
473 
474 /* Look up CPU names by table lookup. */
475 static const char *table_lookup_model(struct cpuinfo_x86 *c)
476 {
477 #ifdef CONFIG_X86_32
478 	const struct legacy_cpu_model_info *info;
479 
480 	if (c->x86_model >= 16)
481 		return NULL;	/* Range check */
482 
483 	if (!this_cpu)
484 		return NULL;
485 
486 	info = this_cpu->legacy_models;
487 
488 	while (info->family) {
489 		if (info->family == c->x86)
490 			return info->model_names[c->x86_model];
491 		info++;
492 	}
493 #endif
494 	return NULL;		/* Not found */
495 }
496 
497 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
498 __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
499 
500 void load_percpu_segment(int cpu)
501 {
502 #ifdef CONFIG_X86_32
503 	loadsegment(fs, __KERNEL_PERCPU);
504 #else
505 	__loadsegment_simple(gs, 0);
506 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
507 #endif
508 	load_stack_canary_segment();
509 }
510 
511 #ifdef CONFIG_X86_32
512 /* The 32-bit entry code needs to find cpu_entry_area. */
513 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
514 #endif
515 
516 /* Load the original GDT from the per-cpu structure */
517 void load_direct_gdt(int cpu)
518 {
519 	struct desc_ptr gdt_descr;
520 
521 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
522 	gdt_descr.size = GDT_SIZE - 1;
523 	load_gdt(&gdt_descr);
524 }
525 EXPORT_SYMBOL_GPL(load_direct_gdt);
526 
527 /* Load a fixmap remapping of the per-cpu GDT */
528 void load_fixmap_gdt(int cpu)
529 {
530 	struct desc_ptr gdt_descr;
531 
532 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
533 	gdt_descr.size = GDT_SIZE - 1;
534 	load_gdt(&gdt_descr);
535 }
536 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
537 
538 /*
539  * Current gdt points %fs at the "master" per-cpu area: after this,
540  * it's on the real one.
541  */
542 void switch_to_new_gdt(int cpu)
543 {
544 	/* Load the original GDT */
545 	load_direct_gdt(cpu);
546 	/* Reload the per-cpu base */
547 	load_percpu_segment(cpu);
548 }
549 
550 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
551 
552 static void get_model_name(struct cpuinfo_x86 *c)
553 {
554 	unsigned int *v;
555 	char *p, *q, *s;
556 
557 	if (c->extended_cpuid_level < 0x80000004)
558 		return;
559 
560 	v = (unsigned int *)c->x86_model_id;
561 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
562 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
563 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
564 	c->x86_model_id[48] = 0;
565 
566 	/* Trim whitespace */
567 	p = q = s = &c->x86_model_id[0];
568 
569 	while (*p == ' ')
570 		p++;
571 
572 	while (*p) {
573 		/* Note the last non-whitespace index */
574 		if (!isspace(*p))
575 			s = q;
576 
577 		*q++ = *p++;
578 	}
579 
580 	*(s + 1) = '\0';
581 }
582 
583 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
584 {
585 	unsigned int eax, ebx, ecx, edx;
586 
587 	c->x86_max_cores = 1;
588 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
589 		return;
590 
591 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
592 	if (eax & 0x1f)
593 		c->x86_max_cores = (eax >> 26) + 1;
594 }
595 
596 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
597 {
598 	unsigned int n, dummy, ebx, ecx, edx, l2size;
599 
600 	n = c->extended_cpuid_level;
601 
602 	if (n >= 0x80000005) {
603 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
604 		c->x86_cache_size = (ecx>>24) + (edx>>24);
605 #ifdef CONFIG_X86_64
606 		/* On K8 L1 TLB is inclusive, so don't count it */
607 		c->x86_tlbsize = 0;
608 #endif
609 	}
610 
611 	if (n < 0x80000006)	/* Some chips just has a large L1. */
612 		return;
613 
614 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
615 	l2size = ecx >> 16;
616 
617 #ifdef CONFIG_X86_64
618 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
619 #else
620 	/* do processor-specific cache resizing */
621 	if (this_cpu->legacy_cache_size)
622 		l2size = this_cpu->legacy_cache_size(c, l2size);
623 
624 	/* Allow user to override all this if necessary. */
625 	if (cachesize_override != -1)
626 		l2size = cachesize_override;
627 
628 	if (l2size == 0)
629 		return;		/* Again, no L2 cache is possible */
630 #endif
631 
632 	c->x86_cache_size = l2size;
633 }
634 
635 u16 __read_mostly tlb_lli_4k[NR_INFO];
636 u16 __read_mostly tlb_lli_2m[NR_INFO];
637 u16 __read_mostly tlb_lli_4m[NR_INFO];
638 u16 __read_mostly tlb_lld_4k[NR_INFO];
639 u16 __read_mostly tlb_lld_2m[NR_INFO];
640 u16 __read_mostly tlb_lld_4m[NR_INFO];
641 u16 __read_mostly tlb_lld_1g[NR_INFO];
642 
643 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
644 {
645 	if (this_cpu->c_detect_tlb)
646 		this_cpu->c_detect_tlb(c);
647 
648 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
649 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
650 		tlb_lli_4m[ENTRIES]);
651 
652 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
653 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
654 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
655 }
656 
657 int detect_ht_early(struct cpuinfo_x86 *c)
658 {
659 #ifdef CONFIG_SMP
660 	u32 eax, ebx, ecx, edx;
661 
662 	if (!cpu_has(c, X86_FEATURE_HT))
663 		return -1;
664 
665 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
666 		return -1;
667 
668 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
669 		return -1;
670 
671 	cpuid(1, &eax, &ebx, &ecx, &edx);
672 
673 	smp_num_siblings = (ebx & 0xff0000) >> 16;
674 	if (smp_num_siblings == 1)
675 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
676 #endif
677 	return 0;
678 }
679 
680 void detect_ht(struct cpuinfo_x86 *c)
681 {
682 #ifdef CONFIG_SMP
683 	int index_msb, core_bits;
684 
685 	if (detect_ht_early(c) < 0)
686 		return;
687 
688 	index_msb = get_count_order(smp_num_siblings);
689 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
690 
691 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
692 
693 	index_msb = get_count_order(smp_num_siblings);
694 
695 	core_bits = get_count_order(c->x86_max_cores);
696 
697 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
698 				       ((1 << core_bits) - 1);
699 #endif
700 }
701 
702 static void get_cpu_vendor(struct cpuinfo_x86 *c)
703 {
704 	char *v = c->x86_vendor_id;
705 	int i;
706 
707 	for (i = 0; i < X86_VENDOR_NUM; i++) {
708 		if (!cpu_devs[i])
709 			break;
710 
711 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
712 		    (cpu_devs[i]->c_ident[1] &&
713 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
714 
715 			this_cpu = cpu_devs[i];
716 			c->x86_vendor = this_cpu->c_x86_vendor;
717 			return;
718 		}
719 	}
720 
721 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
722 		    "CPU: Your system may be unstable.\n", v);
723 
724 	c->x86_vendor = X86_VENDOR_UNKNOWN;
725 	this_cpu = &default_cpu;
726 }
727 
728 void cpu_detect(struct cpuinfo_x86 *c)
729 {
730 	/* Get vendor name */
731 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
732 	      (unsigned int *)&c->x86_vendor_id[0],
733 	      (unsigned int *)&c->x86_vendor_id[8],
734 	      (unsigned int *)&c->x86_vendor_id[4]);
735 
736 	c->x86 = 4;
737 	/* Intel-defined flags: level 0x00000001 */
738 	if (c->cpuid_level >= 0x00000001) {
739 		u32 junk, tfms, cap0, misc;
740 
741 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
742 		c->x86		= x86_family(tfms);
743 		c->x86_model	= x86_model(tfms);
744 		c->x86_stepping	= x86_stepping(tfms);
745 
746 		if (cap0 & (1<<19)) {
747 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
748 			c->x86_cache_alignment = c->x86_clflush_size;
749 		}
750 	}
751 }
752 
753 static void apply_forced_caps(struct cpuinfo_x86 *c)
754 {
755 	int i;
756 
757 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
758 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
759 		c->x86_capability[i] |= cpu_caps_set[i];
760 	}
761 }
762 
763 static void init_speculation_control(struct cpuinfo_x86 *c)
764 {
765 	/*
766 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
767 	 * and they also have a different bit for STIBP support. Also,
768 	 * a hypervisor might have set the individual AMD bits even on
769 	 * Intel CPUs, for finer-grained selection of what's available.
770 	 */
771 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
772 		set_cpu_cap(c, X86_FEATURE_IBRS);
773 		set_cpu_cap(c, X86_FEATURE_IBPB);
774 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
775 	}
776 
777 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
778 		set_cpu_cap(c, X86_FEATURE_STIBP);
779 
780 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
781 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
782 		set_cpu_cap(c, X86_FEATURE_SSBD);
783 
784 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
785 		set_cpu_cap(c, X86_FEATURE_IBRS);
786 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
787 	}
788 
789 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
790 		set_cpu_cap(c, X86_FEATURE_IBPB);
791 
792 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
793 		set_cpu_cap(c, X86_FEATURE_STIBP);
794 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
795 	}
796 
797 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
798 		set_cpu_cap(c, X86_FEATURE_SSBD);
799 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
800 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
801 	}
802 }
803 
804 void get_cpu_cap(struct cpuinfo_x86 *c)
805 {
806 	u32 eax, ebx, ecx, edx;
807 
808 	/* Intel-defined flags: level 0x00000001 */
809 	if (c->cpuid_level >= 0x00000001) {
810 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
811 
812 		c->x86_capability[CPUID_1_ECX] = ecx;
813 		c->x86_capability[CPUID_1_EDX] = edx;
814 	}
815 
816 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
817 	if (c->cpuid_level >= 0x00000006)
818 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
819 
820 	/* Additional Intel-defined flags: level 0x00000007 */
821 	if (c->cpuid_level >= 0x00000007) {
822 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
823 		c->x86_capability[CPUID_7_0_EBX] = ebx;
824 		c->x86_capability[CPUID_7_ECX] = ecx;
825 		c->x86_capability[CPUID_7_EDX] = edx;
826 	}
827 
828 	/* Extended state features: level 0x0000000d */
829 	if (c->cpuid_level >= 0x0000000d) {
830 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
831 
832 		c->x86_capability[CPUID_D_1_EAX] = eax;
833 	}
834 
835 	/* Additional Intel-defined flags: level 0x0000000F */
836 	if (c->cpuid_level >= 0x0000000F) {
837 
838 		/* QoS sub-leaf, EAX=0Fh, ECX=0 */
839 		cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx);
840 		c->x86_capability[CPUID_F_0_EDX] = edx;
841 
842 		if (cpu_has(c, X86_FEATURE_CQM_LLC)) {
843 			/* will be overridden if occupancy monitoring exists */
844 			c->x86_cache_max_rmid = ebx;
845 
846 			/* QoS sub-leaf, EAX=0Fh, ECX=1 */
847 			cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx);
848 			c->x86_capability[CPUID_F_1_EDX] = edx;
849 
850 			if ((cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) ||
851 			      ((cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL)) ||
852 			       (cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)))) {
853 				c->x86_cache_max_rmid = ecx;
854 				c->x86_cache_occ_scale = ebx;
855 			}
856 		} else {
857 			c->x86_cache_max_rmid = -1;
858 			c->x86_cache_occ_scale = -1;
859 		}
860 	}
861 
862 	/* AMD-defined flags: level 0x80000001 */
863 	eax = cpuid_eax(0x80000000);
864 	c->extended_cpuid_level = eax;
865 
866 	if ((eax & 0xffff0000) == 0x80000000) {
867 		if (eax >= 0x80000001) {
868 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
869 
870 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
871 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
872 		}
873 	}
874 
875 	if (c->extended_cpuid_level >= 0x80000007) {
876 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
877 
878 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
879 		c->x86_power = edx;
880 	}
881 
882 	if (c->extended_cpuid_level >= 0x80000008) {
883 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
884 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
885 	}
886 
887 	if (c->extended_cpuid_level >= 0x8000000a)
888 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
889 
890 	init_scattered_cpuid_features(c);
891 	init_speculation_control(c);
892 
893 	/*
894 	 * Clear/Set all flags overridden by options, after probe.
895 	 * This needs to happen each time we re-probe, which may happen
896 	 * several times during CPU initialization.
897 	 */
898 	apply_forced_caps(c);
899 }
900 
901 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
902 {
903 	u32 eax, ebx, ecx, edx;
904 
905 	if (c->extended_cpuid_level >= 0x80000008) {
906 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
907 
908 		c->x86_virt_bits = (eax >> 8) & 0xff;
909 		c->x86_phys_bits = eax & 0xff;
910 	}
911 #ifdef CONFIG_X86_32
912 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
913 		c->x86_phys_bits = 36;
914 #endif
915 	c->x86_cache_bits = c->x86_phys_bits;
916 }
917 
918 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
919 {
920 #ifdef CONFIG_X86_32
921 	int i;
922 
923 	/*
924 	 * First of all, decide if this is a 486 or higher
925 	 * It's a 486 if we can modify the AC flag
926 	 */
927 	if (flag_is_changeable_p(X86_EFLAGS_AC))
928 		c->x86 = 4;
929 	else
930 		c->x86 = 3;
931 
932 	for (i = 0; i < X86_VENDOR_NUM; i++)
933 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
934 			c->x86_vendor_id[0] = 0;
935 			cpu_devs[i]->c_identify(c);
936 			if (c->x86_vendor_id[0]) {
937 				get_cpu_vendor(c);
938 				break;
939 			}
940 		}
941 #endif
942 }
943 
944 #define NO_SPECULATION	BIT(0)
945 #define NO_MELTDOWN	BIT(1)
946 #define NO_SSB		BIT(2)
947 #define NO_L1TF		BIT(3)
948 #define NO_MDS		BIT(4)
949 #define MSBDS_ONLY	BIT(5)
950 
951 #define VULNWL(_vendor, _family, _model, _whitelist)	\
952 	{ X86_VENDOR_##_vendor, _family, _model, X86_FEATURE_ANY, _whitelist }
953 
954 #define VULNWL_INTEL(model, whitelist)		\
955 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
956 
957 #define VULNWL_AMD(family, whitelist)		\
958 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
959 
960 #define VULNWL_HYGON(family, whitelist)		\
961 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
962 
963 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
964 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
965 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
966 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
967 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
968 
969 	/* Intel Family 6 */
970 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION),
971 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION),
972 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION),
973 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION),
974 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION),
975 
976 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY),
977 	VULNWL_INTEL(ATOM_SILVERMONT_X,		NO_SSB | NO_L1TF | MSBDS_ONLY),
978 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY),
979 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY),
980 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY),
981 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY),
982 
983 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
984 
985 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY),
986 
987 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF),
988 	VULNWL_INTEL(ATOM_GOLDMONT_X,		NO_MDS | NO_L1TF),
989 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF),
990 
991 	/* AMD Family 0xf - 0x12 */
992 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
993 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
994 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
995 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
996 
997 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
998 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS),
999 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS),
1000 	{}
1001 };
1002 
1003 static bool __init cpu_matches(unsigned long which)
1004 {
1005 	const struct x86_cpu_id *m = x86_match_cpu(cpu_vuln_whitelist);
1006 
1007 	return m && !!(m->driver_data & which);
1008 }
1009 
1010 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1011 {
1012 	u64 ia32_cap = 0;
1013 
1014 	if (cpu_matches(NO_SPECULATION))
1015 		return;
1016 
1017 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1018 	setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1019 
1020 	if (cpu_has(c, X86_FEATURE_ARCH_CAPABILITIES))
1021 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1022 
1023 	if (!cpu_matches(NO_SSB) && !(ia32_cap & ARCH_CAP_SSB_NO) &&
1024 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1025 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1026 
1027 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1028 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1029 
1030 	if (!cpu_matches(NO_MDS) && !(ia32_cap & ARCH_CAP_MDS_NO)) {
1031 		setup_force_cpu_bug(X86_BUG_MDS);
1032 		if (cpu_matches(MSBDS_ONLY))
1033 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1034 	}
1035 
1036 	if (cpu_matches(NO_MELTDOWN))
1037 		return;
1038 
1039 	/* Rogue Data Cache Load? No! */
1040 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1041 		return;
1042 
1043 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1044 
1045 	if (cpu_matches(NO_L1TF))
1046 		return;
1047 
1048 	setup_force_cpu_bug(X86_BUG_L1TF);
1049 }
1050 
1051 /*
1052  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1053  * unfortunately, that's not true in practice because of early VIA
1054  * chips and (more importantly) broken virtualizers that are not easy
1055  * to detect. In the latter case it doesn't even *fail* reliably, so
1056  * probing for it doesn't even work. Disable it completely on 32-bit
1057  * unless we can find a reliable way to detect all the broken cases.
1058  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1059  */
1060 static void detect_nopl(void)
1061 {
1062 #ifdef CONFIG_X86_32
1063 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1064 #else
1065 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1066 #endif
1067 }
1068 
1069 /*
1070  * Do minimum CPU detection early.
1071  * Fields really needed: vendor, cpuid_level, family, model, mask,
1072  * cache alignment.
1073  * The others are not touched to avoid unwanted side effects.
1074  *
1075  * WARNING: this function is only called on the boot CPU.  Don't add code
1076  * here that is supposed to run on all CPUs.
1077  */
1078 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1079 {
1080 #ifdef CONFIG_X86_64
1081 	c->x86_clflush_size = 64;
1082 	c->x86_phys_bits = 36;
1083 	c->x86_virt_bits = 48;
1084 #else
1085 	c->x86_clflush_size = 32;
1086 	c->x86_phys_bits = 32;
1087 	c->x86_virt_bits = 32;
1088 #endif
1089 	c->x86_cache_alignment = c->x86_clflush_size;
1090 
1091 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1092 	c->extended_cpuid_level = 0;
1093 
1094 	if (!have_cpuid_p())
1095 		identify_cpu_without_cpuid(c);
1096 
1097 	/* cyrix could have cpuid enabled via c_identify()*/
1098 	if (have_cpuid_p()) {
1099 		cpu_detect(c);
1100 		get_cpu_vendor(c);
1101 		get_cpu_cap(c);
1102 		get_cpu_address_sizes(c);
1103 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1104 
1105 		if (this_cpu->c_early_init)
1106 			this_cpu->c_early_init(c);
1107 
1108 		c->cpu_index = 0;
1109 		filter_cpuid_features(c, false);
1110 
1111 		if (this_cpu->c_bsp_init)
1112 			this_cpu->c_bsp_init(c);
1113 	} else {
1114 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1115 	}
1116 
1117 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1118 
1119 	cpu_set_bug_bits(c);
1120 
1121 	fpu__init_system(c);
1122 
1123 #ifdef CONFIG_X86_32
1124 	/*
1125 	 * Regardless of whether PCID is enumerated, the SDM says
1126 	 * that it can't be enabled in 32-bit mode.
1127 	 */
1128 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1129 #endif
1130 
1131 	/*
1132 	 * Later in the boot process pgtable_l5_enabled() relies on
1133 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1134 	 * enabled by this point we need to clear the feature bit to avoid
1135 	 * false-positives at the later stage.
1136 	 *
1137 	 * pgtable_l5_enabled() can be false here for several reasons:
1138 	 *  - 5-level paging is disabled compile-time;
1139 	 *  - it's 32-bit kernel;
1140 	 *  - machine doesn't support 5-level paging;
1141 	 *  - user specified 'no5lvl' in kernel command line.
1142 	 */
1143 	if (!pgtable_l5_enabled())
1144 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1145 
1146 	detect_nopl();
1147 }
1148 
1149 void __init early_cpu_init(void)
1150 {
1151 	const struct cpu_dev *const *cdev;
1152 	int count = 0;
1153 
1154 #ifdef CONFIG_PROCESSOR_SELECT
1155 	pr_info("KERNEL supported cpus:\n");
1156 #endif
1157 
1158 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1159 		const struct cpu_dev *cpudev = *cdev;
1160 
1161 		if (count >= X86_VENDOR_NUM)
1162 			break;
1163 		cpu_devs[count] = cpudev;
1164 		count++;
1165 
1166 #ifdef CONFIG_PROCESSOR_SELECT
1167 		{
1168 			unsigned int j;
1169 
1170 			for (j = 0; j < 2; j++) {
1171 				if (!cpudev->c_ident[j])
1172 					continue;
1173 				pr_info("  %s %s\n", cpudev->c_vendor,
1174 					cpudev->c_ident[j]);
1175 			}
1176 		}
1177 #endif
1178 	}
1179 	early_identify_cpu(&boot_cpu_data);
1180 }
1181 
1182 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1183 {
1184 #ifdef CONFIG_X86_64
1185 	/*
1186 	 * Empirically, writing zero to a segment selector on AMD does
1187 	 * not clear the base, whereas writing zero to a segment
1188 	 * selector on Intel does clear the base.  Intel's behavior
1189 	 * allows slightly faster context switches in the common case
1190 	 * where GS is unused by the prev and next threads.
1191 	 *
1192 	 * Since neither vendor documents this anywhere that I can see,
1193 	 * detect it directly instead of hardcoding the choice by
1194 	 * vendor.
1195 	 *
1196 	 * I've designated AMD's behavior as the "bug" because it's
1197 	 * counterintuitive and less friendly.
1198 	 */
1199 
1200 	unsigned long old_base, tmp;
1201 	rdmsrl(MSR_FS_BASE, old_base);
1202 	wrmsrl(MSR_FS_BASE, 1);
1203 	loadsegment(fs, 0);
1204 	rdmsrl(MSR_FS_BASE, tmp);
1205 	if (tmp != 0)
1206 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1207 	wrmsrl(MSR_FS_BASE, old_base);
1208 #endif
1209 }
1210 
1211 static void generic_identify(struct cpuinfo_x86 *c)
1212 {
1213 	c->extended_cpuid_level = 0;
1214 
1215 	if (!have_cpuid_p())
1216 		identify_cpu_without_cpuid(c);
1217 
1218 	/* cyrix could have cpuid enabled via c_identify()*/
1219 	if (!have_cpuid_p())
1220 		return;
1221 
1222 	cpu_detect(c);
1223 
1224 	get_cpu_vendor(c);
1225 
1226 	get_cpu_cap(c);
1227 
1228 	get_cpu_address_sizes(c);
1229 
1230 	if (c->cpuid_level >= 0x00000001) {
1231 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1232 #ifdef CONFIG_X86_32
1233 # ifdef CONFIG_SMP
1234 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1235 # else
1236 		c->apicid = c->initial_apicid;
1237 # endif
1238 #endif
1239 		c->phys_proc_id = c->initial_apicid;
1240 	}
1241 
1242 	get_model_name(c); /* Default name */
1243 
1244 	detect_null_seg_behavior(c);
1245 
1246 	/*
1247 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1248 	 * systems that run Linux at CPL > 0 may or may not have the
1249 	 * issue, but, even if they have the issue, there's absolutely
1250 	 * nothing we can do about it because we can't use the real IRET
1251 	 * instruction.
1252 	 *
1253 	 * NB: For the time being, only 32-bit kernels support
1254 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1255 	 * whether to apply espfix using paravirt hooks.  If any
1256 	 * non-paravirt system ever shows up that does *not* have the
1257 	 * ESPFIX issue, we can change this.
1258 	 */
1259 #ifdef CONFIG_X86_32
1260 # ifdef CONFIG_PARAVIRT_XXL
1261 	do {
1262 		extern void native_iret(void);
1263 		if (pv_ops.cpu.iret == native_iret)
1264 			set_cpu_bug(c, X86_BUG_ESPFIX);
1265 	} while (0);
1266 # else
1267 	set_cpu_bug(c, X86_BUG_ESPFIX);
1268 # endif
1269 #endif
1270 }
1271 
1272 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
1273 {
1274 	/*
1275 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
1276 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
1277 	 * in case CQM bits really aren't there in this CPU.
1278 	 */
1279 	if (c != &boot_cpu_data) {
1280 		boot_cpu_data.x86_cache_max_rmid =
1281 			min(boot_cpu_data.x86_cache_max_rmid,
1282 			    c->x86_cache_max_rmid);
1283 	}
1284 }
1285 
1286 /*
1287  * Validate that ACPI/mptables have the same information about the
1288  * effective APIC id and update the package map.
1289  */
1290 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1291 {
1292 #ifdef CONFIG_SMP
1293 	unsigned int apicid, cpu = smp_processor_id();
1294 
1295 	apicid = apic->cpu_present_to_apicid(cpu);
1296 
1297 	if (apicid != c->apicid) {
1298 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1299 		       cpu, apicid, c->initial_apicid);
1300 	}
1301 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1302 #else
1303 	c->logical_proc_id = 0;
1304 #endif
1305 }
1306 
1307 /*
1308  * This does the hard work of actually picking apart the CPU stuff...
1309  */
1310 static void identify_cpu(struct cpuinfo_x86 *c)
1311 {
1312 	int i;
1313 
1314 	c->loops_per_jiffy = loops_per_jiffy;
1315 	c->x86_cache_size = 0;
1316 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1317 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1318 	c->x86_vendor_id[0] = '\0'; /* Unset */
1319 	c->x86_model_id[0] = '\0';  /* Unset */
1320 	c->x86_max_cores = 1;
1321 	c->x86_coreid_bits = 0;
1322 	c->cu_id = 0xff;
1323 #ifdef CONFIG_X86_64
1324 	c->x86_clflush_size = 64;
1325 	c->x86_phys_bits = 36;
1326 	c->x86_virt_bits = 48;
1327 #else
1328 	c->cpuid_level = -1;	/* CPUID not detected */
1329 	c->x86_clflush_size = 32;
1330 	c->x86_phys_bits = 32;
1331 	c->x86_virt_bits = 32;
1332 #endif
1333 	c->x86_cache_alignment = c->x86_clflush_size;
1334 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1335 
1336 	generic_identify(c);
1337 
1338 	if (this_cpu->c_identify)
1339 		this_cpu->c_identify(c);
1340 
1341 	/* Clear/Set all flags overridden by options, after probe */
1342 	apply_forced_caps(c);
1343 
1344 #ifdef CONFIG_X86_64
1345 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1346 #endif
1347 
1348 	/*
1349 	 * Vendor-specific initialization.  In this section we
1350 	 * canonicalize the feature flags, meaning if there are
1351 	 * features a certain CPU supports which CPUID doesn't
1352 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1353 	 * we handle them here.
1354 	 *
1355 	 * At the end of this section, c->x86_capability better
1356 	 * indicate the features this CPU genuinely supports!
1357 	 */
1358 	if (this_cpu->c_init)
1359 		this_cpu->c_init(c);
1360 
1361 	/* Disable the PN if appropriate */
1362 	squash_the_stupid_serial_number(c);
1363 
1364 	/* Set up SMEP/SMAP/UMIP */
1365 	setup_smep(c);
1366 	setup_smap(c);
1367 	setup_umip(c);
1368 
1369 	/*
1370 	 * The vendor-specific functions might have changed features.
1371 	 * Now we do "generic changes."
1372 	 */
1373 
1374 	/* Filter out anything that depends on CPUID levels we don't have */
1375 	filter_cpuid_features(c, true);
1376 
1377 	/* If the model name is still unset, do table lookup. */
1378 	if (!c->x86_model_id[0]) {
1379 		const char *p;
1380 		p = table_lookup_model(c);
1381 		if (p)
1382 			strcpy(c->x86_model_id, p);
1383 		else
1384 			/* Last resort... */
1385 			sprintf(c->x86_model_id, "%02x/%02x",
1386 				c->x86, c->x86_model);
1387 	}
1388 
1389 #ifdef CONFIG_X86_64
1390 	detect_ht(c);
1391 #endif
1392 
1393 	x86_init_rdrand(c);
1394 	x86_init_cache_qos(c);
1395 	setup_pku(c);
1396 
1397 	/*
1398 	 * Clear/Set all flags overridden by options, need do it
1399 	 * before following smp all cpus cap AND.
1400 	 */
1401 	apply_forced_caps(c);
1402 
1403 	/*
1404 	 * On SMP, boot_cpu_data holds the common feature set between
1405 	 * all CPUs; so make sure that we indicate which features are
1406 	 * common between the CPUs.  The first time this routine gets
1407 	 * executed, c == &boot_cpu_data.
1408 	 */
1409 	if (c != &boot_cpu_data) {
1410 		/* AND the already accumulated flags with these */
1411 		for (i = 0; i < NCAPINTS; i++)
1412 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1413 
1414 		/* OR, i.e. replicate the bug flags */
1415 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1416 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1417 	}
1418 
1419 	/* Init Machine Check Exception if available. */
1420 	mcheck_cpu_init(c);
1421 
1422 	select_idle_routine(c);
1423 
1424 #ifdef CONFIG_NUMA
1425 	numa_add_cpu(smp_processor_id());
1426 #endif
1427 }
1428 
1429 /*
1430  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1431  * on 32-bit kernels:
1432  */
1433 #ifdef CONFIG_X86_32
1434 void enable_sep_cpu(void)
1435 {
1436 	struct tss_struct *tss;
1437 	int cpu;
1438 
1439 	if (!boot_cpu_has(X86_FEATURE_SEP))
1440 		return;
1441 
1442 	cpu = get_cpu();
1443 	tss = &per_cpu(cpu_tss_rw, cpu);
1444 
1445 	/*
1446 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1447 	 * see the big comment in struct x86_hw_tss's definition.
1448 	 */
1449 
1450 	tss->x86_tss.ss1 = __KERNEL_CS;
1451 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1452 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1453 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1454 
1455 	put_cpu();
1456 }
1457 #endif
1458 
1459 void __init identify_boot_cpu(void)
1460 {
1461 	identify_cpu(&boot_cpu_data);
1462 #ifdef CONFIG_X86_32
1463 	sysenter_setup();
1464 	enable_sep_cpu();
1465 #endif
1466 	cpu_detect_tlb(&boot_cpu_data);
1467 }
1468 
1469 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1470 {
1471 	BUG_ON(c == &boot_cpu_data);
1472 	identify_cpu(c);
1473 #ifdef CONFIG_X86_32
1474 	enable_sep_cpu();
1475 #endif
1476 	mtrr_ap_init();
1477 	validate_apic_and_package_id(c);
1478 	x86_spec_ctrl_setup_ap();
1479 }
1480 
1481 static __init int setup_noclflush(char *arg)
1482 {
1483 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1484 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1485 	return 1;
1486 }
1487 __setup("noclflush", setup_noclflush);
1488 
1489 void print_cpu_info(struct cpuinfo_x86 *c)
1490 {
1491 	const char *vendor = NULL;
1492 
1493 	if (c->x86_vendor < X86_VENDOR_NUM) {
1494 		vendor = this_cpu->c_vendor;
1495 	} else {
1496 		if (c->cpuid_level >= 0)
1497 			vendor = c->x86_vendor_id;
1498 	}
1499 
1500 	if (vendor && !strstr(c->x86_model_id, vendor))
1501 		pr_cont("%s ", vendor);
1502 
1503 	if (c->x86_model_id[0])
1504 		pr_cont("%s", c->x86_model_id);
1505 	else
1506 		pr_cont("%d86", c->x86);
1507 
1508 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1509 
1510 	if (c->x86_stepping || c->cpuid_level >= 0)
1511 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1512 	else
1513 		pr_cont(")\n");
1514 }
1515 
1516 /*
1517  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1518  * But we need to keep a dummy __setup around otherwise it would
1519  * show up as an environment variable for init.
1520  */
1521 static __init int setup_clearcpuid(char *arg)
1522 {
1523 	return 1;
1524 }
1525 __setup("clearcpuid=", setup_clearcpuid);
1526 
1527 #ifdef CONFIG_X86_64
1528 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1529 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1530 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1531 
1532 /*
1533  * The following percpu variables are hot.  Align current_task to
1534  * cacheline size such that they fall in the same cacheline.
1535  */
1536 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1537 	&init_task;
1538 EXPORT_PER_CPU_SYMBOL(current_task);
1539 
1540 DEFINE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
1541 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1542 
1543 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1544 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1545 
1546 /* May not be marked __init: used by software suspend */
1547 void syscall_init(void)
1548 {
1549 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1550 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1551 
1552 #ifdef CONFIG_IA32_EMULATION
1553 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1554 	/*
1555 	 * This only works on Intel CPUs.
1556 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1557 	 * This does not cause SYSENTER to jump to the wrong location, because
1558 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1559 	 */
1560 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1561 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1562 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1563 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1564 #else
1565 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1566 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1567 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1568 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1569 #endif
1570 
1571 	/* Flags to clear on syscall */
1572 	wrmsrl(MSR_SYSCALL_MASK,
1573 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1574 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1575 }
1576 
1577 DEFINE_PER_CPU(int, debug_stack_usage);
1578 DEFINE_PER_CPU(u32, debug_idt_ctr);
1579 
1580 void debug_stack_set_zero(void)
1581 {
1582 	this_cpu_inc(debug_idt_ctr);
1583 	load_current_idt();
1584 }
1585 NOKPROBE_SYMBOL(debug_stack_set_zero);
1586 
1587 void debug_stack_reset(void)
1588 {
1589 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1590 		return;
1591 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1592 		load_current_idt();
1593 }
1594 NOKPROBE_SYMBOL(debug_stack_reset);
1595 
1596 #else	/* CONFIG_X86_64 */
1597 
1598 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1599 EXPORT_PER_CPU_SYMBOL(current_task);
1600 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1601 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1602 
1603 /*
1604  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1605  * the top of the kernel stack.  Use an extra percpu variable to track the
1606  * top of the kernel stack directly.
1607  */
1608 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1609 	(unsigned long)&init_thread_union + THREAD_SIZE;
1610 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1611 
1612 #ifdef CONFIG_STACKPROTECTOR
1613 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1614 #endif
1615 
1616 #endif	/* CONFIG_X86_64 */
1617 
1618 /*
1619  * Clear all 6 debug registers:
1620  */
1621 static void clear_all_debug_regs(void)
1622 {
1623 	int i;
1624 
1625 	for (i = 0; i < 8; i++) {
1626 		/* Ignore db4, db5 */
1627 		if ((i == 4) || (i == 5))
1628 			continue;
1629 
1630 		set_debugreg(0, i);
1631 	}
1632 }
1633 
1634 #ifdef CONFIG_KGDB
1635 /*
1636  * Restore debug regs if using kgdbwait and you have a kernel debugger
1637  * connection established.
1638  */
1639 static void dbg_restore_debug_regs(void)
1640 {
1641 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1642 		arch_kgdb_ops.correct_hw_break();
1643 }
1644 #else /* ! CONFIG_KGDB */
1645 #define dbg_restore_debug_regs()
1646 #endif /* ! CONFIG_KGDB */
1647 
1648 static void wait_for_master_cpu(int cpu)
1649 {
1650 #ifdef CONFIG_SMP
1651 	/*
1652 	 * wait for ACK from master CPU before continuing
1653 	 * with AP initialization
1654 	 */
1655 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1656 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1657 		cpu_relax();
1658 #endif
1659 }
1660 
1661 #ifdef CONFIG_X86_64
1662 static void setup_getcpu(int cpu)
1663 {
1664 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1665 	struct desc_struct d = { };
1666 
1667 	if (boot_cpu_has(X86_FEATURE_RDTSCP))
1668 		write_rdtscp_aux(cpudata);
1669 
1670 	/* Store CPU and node number in limit. */
1671 	d.limit0 = cpudata;
1672 	d.limit1 = cpudata >> 16;
1673 
1674 	d.type = 5;		/* RO data, expand down, accessed */
1675 	d.dpl = 3;		/* Visible to user code */
1676 	d.s = 1;		/* Not a system segment */
1677 	d.p = 1;		/* Present */
1678 	d.d = 1;		/* 32-bit */
1679 
1680 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1681 }
1682 #endif
1683 
1684 /*
1685  * cpu_init() initializes state that is per-CPU. Some data is already
1686  * initialized (naturally) in the bootstrap process, such as the GDT
1687  * and IDT. We reload them nevertheless, this function acts as a
1688  * 'CPU state barrier', nothing should get across.
1689  */
1690 #ifdef CONFIG_X86_64
1691 
1692 void cpu_init(void)
1693 {
1694 	int cpu = raw_smp_processor_id();
1695 	struct task_struct *me;
1696 	struct tss_struct *t;
1697 	int i;
1698 
1699 	wait_for_master_cpu(cpu);
1700 
1701 	/*
1702 	 * Initialize the CR4 shadow before doing anything that could
1703 	 * try to read it.
1704 	 */
1705 	cr4_init_shadow();
1706 
1707 	if (cpu)
1708 		load_ucode_ap();
1709 
1710 	t = &per_cpu(cpu_tss_rw, cpu);
1711 
1712 #ifdef CONFIG_NUMA
1713 	if (this_cpu_read(numa_node) == 0 &&
1714 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1715 		set_numa_node(early_cpu_to_node(cpu));
1716 #endif
1717 	setup_getcpu(cpu);
1718 
1719 	me = current;
1720 
1721 	pr_debug("Initializing CPU#%d\n", cpu);
1722 
1723 	cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1724 
1725 	/*
1726 	 * Initialize the per-CPU GDT with the boot GDT,
1727 	 * and set up the GDT descriptor:
1728 	 */
1729 
1730 	switch_to_new_gdt(cpu);
1731 	loadsegment(fs, 0);
1732 
1733 	load_current_idt();
1734 
1735 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1736 	syscall_init();
1737 
1738 	wrmsrl(MSR_FS_BASE, 0);
1739 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1740 	barrier();
1741 
1742 	x86_configure_nx();
1743 	x2apic_setup();
1744 
1745 	/*
1746 	 * set up and load the per-CPU TSS
1747 	 */
1748 	if (!t->x86_tss.ist[0]) {
1749 		t->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1750 		t->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1751 		t->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1752 		t->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1753 	}
1754 
1755 	t->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
1756 
1757 	/*
1758 	 * <= is required because the CPU will access up to
1759 	 * 8 bits beyond the end of the IO permission bitmap.
1760 	 */
1761 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1762 		t->io_bitmap[i] = ~0UL;
1763 
1764 	mmgrab(&init_mm);
1765 	me->active_mm = &init_mm;
1766 	BUG_ON(me->mm);
1767 	initialize_tlbstate_and_flush();
1768 	enter_lazy_tlb(&init_mm, me);
1769 
1770 	/*
1771 	 * Initialize the TSS.  sp0 points to the entry trampoline stack
1772 	 * regardless of what task is running.
1773 	 */
1774 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1775 	load_TR_desc();
1776 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1777 
1778 	load_mm_ldt(&init_mm);
1779 
1780 	clear_all_debug_regs();
1781 	dbg_restore_debug_regs();
1782 
1783 	fpu__init_cpu();
1784 
1785 	if (is_uv_system())
1786 		uv_cpu_init();
1787 
1788 	load_fixmap_gdt(cpu);
1789 }
1790 
1791 #else
1792 
1793 void cpu_init(void)
1794 {
1795 	int cpu = smp_processor_id();
1796 	struct task_struct *curr = current;
1797 	struct tss_struct *t = &per_cpu(cpu_tss_rw, cpu);
1798 
1799 	wait_for_master_cpu(cpu);
1800 
1801 	/*
1802 	 * Initialize the CR4 shadow before doing anything that could
1803 	 * try to read it.
1804 	 */
1805 	cr4_init_shadow();
1806 
1807 	show_ucode_info_early();
1808 
1809 	pr_info("Initializing CPU#%d\n", cpu);
1810 
1811 	if (cpu_feature_enabled(X86_FEATURE_VME) ||
1812 	    boot_cpu_has(X86_FEATURE_TSC) ||
1813 	    boot_cpu_has(X86_FEATURE_DE))
1814 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1815 
1816 	load_current_idt();
1817 	switch_to_new_gdt(cpu);
1818 
1819 	/*
1820 	 * Set up and load the per-CPU TSS and LDT
1821 	 */
1822 	mmgrab(&init_mm);
1823 	curr->active_mm = &init_mm;
1824 	BUG_ON(curr->mm);
1825 	initialize_tlbstate_and_flush();
1826 	enter_lazy_tlb(&init_mm, curr);
1827 
1828 	/*
1829 	 * Initialize the TSS.  sp0 points to the entry trampoline stack
1830 	 * regardless of what task is running.
1831 	 */
1832 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1833 	load_TR_desc();
1834 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1835 
1836 	load_mm_ldt(&init_mm);
1837 
1838 	t->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
1839 
1840 #ifdef CONFIG_DOUBLEFAULT
1841 	/* Set up doublefault TSS pointer in the GDT */
1842 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1843 #endif
1844 
1845 	clear_all_debug_regs();
1846 	dbg_restore_debug_regs();
1847 
1848 	fpu__init_cpu();
1849 
1850 	load_fixmap_gdt(cpu);
1851 }
1852 #endif
1853 
1854 /*
1855  * The microcode loader calls this upon late microcode load to recheck features,
1856  * only when microcode has been updated. Caller holds microcode_mutex and CPU
1857  * hotplug lock.
1858  */
1859 void microcode_check(void)
1860 {
1861 	struct cpuinfo_x86 info;
1862 
1863 	perf_check_microcode();
1864 
1865 	/* Reload CPUID max function as it might've changed. */
1866 	info.cpuid_level = cpuid_eax(0);
1867 
1868 	/*
1869 	 * Copy all capability leafs to pick up the synthetic ones so that
1870 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
1871 	 * get overwritten in get_cpu_cap().
1872 	 */
1873 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
1874 
1875 	get_cpu_cap(&info);
1876 
1877 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
1878 		return;
1879 
1880 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
1881 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
1882 }
1883