xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision b830f94f)
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 DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
370 static unsigned long cr4_pinned_bits __ro_after_init;
371 
372 void native_write_cr0(unsigned long val)
373 {
374 	unsigned long bits_missing = 0;
375 
376 set_register:
377 	asm volatile("mov %0,%%cr0": "+r" (val), "+m" (__force_order));
378 
379 	if (static_branch_likely(&cr_pinning)) {
380 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
381 			bits_missing = X86_CR0_WP;
382 			val |= bits_missing;
383 			goto set_register;
384 		}
385 		/* Warn after we've set the missing bits. */
386 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
387 	}
388 }
389 EXPORT_SYMBOL(native_write_cr0);
390 
391 void native_write_cr4(unsigned long val)
392 {
393 	unsigned long bits_missing = 0;
394 
395 set_register:
396 	asm volatile("mov %0,%%cr4": "+r" (val), "+m" (cr4_pinned_bits));
397 
398 	if (static_branch_likely(&cr_pinning)) {
399 		if (unlikely((val & cr4_pinned_bits) != cr4_pinned_bits)) {
400 			bits_missing = ~val & cr4_pinned_bits;
401 			val |= bits_missing;
402 			goto set_register;
403 		}
404 		/* Warn after we've set the missing bits. */
405 		WARN_ONCE(bits_missing, "CR4 bits went missing: %lx!?\n",
406 			  bits_missing);
407 	}
408 }
409 EXPORT_SYMBOL(native_write_cr4);
410 
411 void cr4_init(void)
412 {
413 	unsigned long cr4 = __read_cr4();
414 
415 	if (boot_cpu_has(X86_FEATURE_PCID))
416 		cr4 |= X86_CR4_PCIDE;
417 	if (static_branch_likely(&cr_pinning))
418 		cr4 |= cr4_pinned_bits;
419 
420 	__write_cr4(cr4);
421 
422 	/* Initialize cr4 shadow for this CPU. */
423 	this_cpu_write(cpu_tlbstate.cr4, cr4);
424 }
425 
426 /*
427  * Once CPU feature detection is finished (and boot params have been
428  * parsed), record any of the sensitive CR bits that are set, and
429  * enable CR pinning.
430  */
431 static void __init setup_cr_pinning(void)
432 {
433 	unsigned long mask;
434 
435 	mask = (X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP);
436 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & mask;
437 	static_key_enable(&cr_pinning.key);
438 }
439 
440 /*
441  * Protection Keys are not available in 32-bit mode.
442  */
443 static bool pku_disabled;
444 
445 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
446 {
447 	struct pkru_state *pk;
448 
449 	/* check the boot processor, plus compile options for PKU: */
450 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
451 		return;
452 	/* checks the actual processor's cpuid bits: */
453 	if (!cpu_has(c, X86_FEATURE_PKU))
454 		return;
455 	if (pku_disabled)
456 		return;
457 
458 	cr4_set_bits(X86_CR4_PKE);
459 	pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU);
460 	if (pk)
461 		pk->pkru = init_pkru_value;
462 	/*
463 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
464 	 * cpuid bit to be set.  We need to ensure that we
465 	 * update that bit in this CPU's "cpu_info".
466 	 */
467 	get_cpu_cap(c);
468 }
469 
470 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
471 static __init int setup_disable_pku(char *arg)
472 {
473 	/*
474 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
475 	 * runtime checks are against OSPKE so clearing the
476 	 * bit does nothing.
477 	 *
478 	 * This way, we will see "pku" in cpuinfo, but not
479 	 * "ospke", which is exactly what we want.  It shows
480 	 * that the CPU has PKU, but the OS has not enabled it.
481 	 * This happens to be exactly how a system would look
482 	 * if we disabled the config option.
483 	 */
484 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
485 	pku_disabled = true;
486 	return 1;
487 }
488 __setup("nopku", setup_disable_pku);
489 #endif /* CONFIG_X86_64 */
490 
491 /*
492  * Some CPU features depend on higher CPUID levels, which may not always
493  * be available due to CPUID level capping or broken virtualization
494  * software.  Add those features to this table to auto-disable them.
495  */
496 struct cpuid_dependent_feature {
497 	u32 feature;
498 	u32 level;
499 };
500 
501 static const struct cpuid_dependent_feature
502 cpuid_dependent_features[] = {
503 	{ X86_FEATURE_MWAIT,		0x00000005 },
504 	{ X86_FEATURE_DCA,		0x00000009 },
505 	{ X86_FEATURE_XSAVE,		0x0000000d },
506 	{ 0, 0 }
507 };
508 
509 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
510 {
511 	const struct cpuid_dependent_feature *df;
512 
513 	for (df = cpuid_dependent_features; df->feature; df++) {
514 
515 		if (!cpu_has(c, df->feature))
516 			continue;
517 		/*
518 		 * Note: cpuid_level is set to -1 if unavailable, but
519 		 * extended_extended_level is set to 0 if unavailable
520 		 * and the legitimate extended levels are all negative
521 		 * when signed; hence the weird messing around with
522 		 * signs here...
523 		 */
524 		if (!((s32)df->level < 0 ?
525 		     (u32)df->level > (u32)c->extended_cpuid_level :
526 		     (s32)df->level > (s32)c->cpuid_level))
527 			continue;
528 
529 		clear_cpu_cap(c, df->feature);
530 		if (!warn)
531 			continue;
532 
533 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
534 			x86_cap_flag(df->feature), df->level);
535 	}
536 }
537 
538 /*
539  * Naming convention should be: <Name> [(<Codename>)]
540  * This table only is used unless init_<vendor>() below doesn't set it;
541  * in particular, if CPUID levels 0x80000002..4 are supported, this
542  * isn't used
543  */
544 
545 /* Look up CPU names by table lookup. */
546 static const char *table_lookup_model(struct cpuinfo_x86 *c)
547 {
548 #ifdef CONFIG_X86_32
549 	const struct legacy_cpu_model_info *info;
550 
551 	if (c->x86_model >= 16)
552 		return NULL;	/* Range check */
553 
554 	if (!this_cpu)
555 		return NULL;
556 
557 	info = this_cpu->legacy_models;
558 
559 	while (info->family) {
560 		if (info->family == c->x86)
561 			return info->model_names[c->x86_model];
562 		info++;
563 	}
564 #endif
565 	return NULL;		/* Not found */
566 }
567 
568 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
569 __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
570 
571 void load_percpu_segment(int cpu)
572 {
573 #ifdef CONFIG_X86_32
574 	loadsegment(fs, __KERNEL_PERCPU);
575 #else
576 	__loadsegment_simple(gs, 0);
577 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
578 #endif
579 	load_stack_canary_segment();
580 }
581 
582 #ifdef CONFIG_X86_32
583 /* The 32-bit entry code needs to find cpu_entry_area. */
584 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
585 #endif
586 
587 /* Load the original GDT from the per-cpu structure */
588 void load_direct_gdt(int cpu)
589 {
590 	struct desc_ptr gdt_descr;
591 
592 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
593 	gdt_descr.size = GDT_SIZE - 1;
594 	load_gdt(&gdt_descr);
595 }
596 EXPORT_SYMBOL_GPL(load_direct_gdt);
597 
598 /* Load a fixmap remapping of the per-cpu GDT */
599 void load_fixmap_gdt(int cpu)
600 {
601 	struct desc_ptr gdt_descr;
602 
603 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
604 	gdt_descr.size = GDT_SIZE - 1;
605 	load_gdt(&gdt_descr);
606 }
607 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
608 
609 /*
610  * Current gdt points %fs at the "master" per-cpu area: after this,
611  * it's on the real one.
612  */
613 void switch_to_new_gdt(int cpu)
614 {
615 	/* Load the original GDT */
616 	load_direct_gdt(cpu);
617 	/* Reload the per-cpu base */
618 	load_percpu_segment(cpu);
619 }
620 
621 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
622 
623 static void get_model_name(struct cpuinfo_x86 *c)
624 {
625 	unsigned int *v;
626 	char *p, *q, *s;
627 
628 	if (c->extended_cpuid_level < 0x80000004)
629 		return;
630 
631 	v = (unsigned int *)c->x86_model_id;
632 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
633 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
634 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
635 	c->x86_model_id[48] = 0;
636 
637 	/* Trim whitespace */
638 	p = q = s = &c->x86_model_id[0];
639 
640 	while (*p == ' ')
641 		p++;
642 
643 	while (*p) {
644 		/* Note the last non-whitespace index */
645 		if (!isspace(*p))
646 			s = q;
647 
648 		*q++ = *p++;
649 	}
650 
651 	*(s + 1) = '\0';
652 }
653 
654 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
655 {
656 	unsigned int eax, ebx, ecx, edx;
657 
658 	c->x86_max_cores = 1;
659 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
660 		return;
661 
662 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
663 	if (eax & 0x1f)
664 		c->x86_max_cores = (eax >> 26) + 1;
665 }
666 
667 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
668 {
669 	unsigned int n, dummy, ebx, ecx, edx, l2size;
670 
671 	n = c->extended_cpuid_level;
672 
673 	if (n >= 0x80000005) {
674 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
675 		c->x86_cache_size = (ecx>>24) + (edx>>24);
676 #ifdef CONFIG_X86_64
677 		/* On K8 L1 TLB is inclusive, so don't count it */
678 		c->x86_tlbsize = 0;
679 #endif
680 	}
681 
682 	if (n < 0x80000006)	/* Some chips just has a large L1. */
683 		return;
684 
685 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
686 	l2size = ecx >> 16;
687 
688 #ifdef CONFIG_X86_64
689 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
690 #else
691 	/* do processor-specific cache resizing */
692 	if (this_cpu->legacy_cache_size)
693 		l2size = this_cpu->legacy_cache_size(c, l2size);
694 
695 	/* Allow user to override all this if necessary. */
696 	if (cachesize_override != -1)
697 		l2size = cachesize_override;
698 
699 	if (l2size == 0)
700 		return;		/* Again, no L2 cache is possible */
701 #endif
702 
703 	c->x86_cache_size = l2size;
704 }
705 
706 u16 __read_mostly tlb_lli_4k[NR_INFO];
707 u16 __read_mostly tlb_lli_2m[NR_INFO];
708 u16 __read_mostly tlb_lli_4m[NR_INFO];
709 u16 __read_mostly tlb_lld_4k[NR_INFO];
710 u16 __read_mostly tlb_lld_2m[NR_INFO];
711 u16 __read_mostly tlb_lld_4m[NR_INFO];
712 u16 __read_mostly tlb_lld_1g[NR_INFO];
713 
714 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
715 {
716 	if (this_cpu->c_detect_tlb)
717 		this_cpu->c_detect_tlb(c);
718 
719 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
720 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
721 		tlb_lli_4m[ENTRIES]);
722 
723 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
724 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
725 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
726 }
727 
728 int detect_ht_early(struct cpuinfo_x86 *c)
729 {
730 #ifdef CONFIG_SMP
731 	u32 eax, ebx, ecx, edx;
732 
733 	if (!cpu_has(c, X86_FEATURE_HT))
734 		return -1;
735 
736 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
737 		return -1;
738 
739 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
740 		return -1;
741 
742 	cpuid(1, &eax, &ebx, &ecx, &edx);
743 
744 	smp_num_siblings = (ebx & 0xff0000) >> 16;
745 	if (smp_num_siblings == 1)
746 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
747 #endif
748 	return 0;
749 }
750 
751 void detect_ht(struct cpuinfo_x86 *c)
752 {
753 #ifdef CONFIG_SMP
754 	int index_msb, core_bits;
755 
756 	if (detect_ht_early(c) < 0)
757 		return;
758 
759 	index_msb = get_count_order(smp_num_siblings);
760 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
761 
762 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
763 
764 	index_msb = get_count_order(smp_num_siblings);
765 
766 	core_bits = get_count_order(c->x86_max_cores);
767 
768 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
769 				       ((1 << core_bits) - 1);
770 #endif
771 }
772 
773 static void get_cpu_vendor(struct cpuinfo_x86 *c)
774 {
775 	char *v = c->x86_vendor_id;
776 	int i;
777 
778 	for (i = 0; i < X86_VENDOR_NUM; i++) {
779 		if (!cpu_devs[i])
780 			break;
781 
782 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
783 		    (cpu_devs[i]->c_ident[1] &&
784 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
785 
786 			this_cpu = cpu_devs[i];
787 			c->x86_vendor = this_cpu->c_x86_vendor;
788 			return;
789 		}
790 	}
791 
792 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
793 		    "CPU: Your system may be unstable.\n", v);
794 
795 	c->x86_vendor = X86_VENDOR_UNKNOWN;
796 	this_cpu = &default_cpu;
797 }
798 
799 void cpu_detect(struct cpuinfo_x86 *c)
800 {
801 	/* Get vendor name */
802 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
803 	      (unsigned int *)&c->x86_vendor_id[0],
804 	      (unsigned int *)&c->x86_vendor_id[8],
805 	      (unsigned int *)&c->x86_vendor_id[4]);
806 
807 	c->x86 = 4;
808 	/* Intel-defined flags: level 0x00000001 */
809 	if (c->cpuid_level >= 0x00000001) {
810 		u32 junk, tfms, cap0, misc;
811 
812 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
813 		c->x86		= x86_family(tfms);
814 		c->x86_model	= x86_model(tfms);
815 		c->x86_stepping	= x86_stepping(tfms);
816 
817 		if (cap0 & (1<<19)) {
818 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
819 			c->x86_cache_alignment = c->x86_clflush_size;
820 		}
821 	}
822 }
823 
824 static void apply_forced_caps(struct cpuinfo_x86 *c)
825 {
826 	int i;
827 
828 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
829 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
830 		c->x86_capability[i] |= cpu_caps_set[i];
831 	}
832 }
833 
834 static void init_speculation_control(struct cpuinfo_x86 *c)
835 {
836 	/*
837 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
838 	 * and they also have a different bit for STIBP support. Also,
839 	 * a hypervisor might have set the individual AMD bits even on
840 	 * Intel CPUs, for finer-grained selection of what's available.
841 	 */
842 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
843 		set_cpu_cap(c, X86_FEATURE_IBRS);
844 		set_cpu_cap(c, X86_FEATURE_IBPB);
845 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
846 	}
847 
848 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
849 		set_cpu_cap(c, X86_FEATURE_STIBP);
850 
851 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
852 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
853 		set_cpu_cap(c, X86_FEATURE_SSBD);
854 
855 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
856 		set_cpu_cap(c, X86_FEATURE_IBRS);
857 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
858 	}
859 
860 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
861 		set_cpu_cap(c, X86_FEATURE_IBPB);
862 
863 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
864 		set_cpu_cap(c, X86_FEATURE_STIBP);
865 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
866 	}
867 
868 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
869 		set_cpu_cap(c, X86_FEATURE_SSBD);
870 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
871 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
872 	}
873 }
874 
875 static void init_cqm(struct cpuinfo_x86 *c)
876 {
877 	if (!cpu_has(c, X86_FEATURE_CQM_LLC)) {
878 		c->x86_cache_max_rmid  = -1;
879 		c->x86_cache_occ_scale = -1;
880 		return;
881 	}
882 
883 	/* will be overridden if occupancy monitoring exists */
884 	c->x86_cache_max_rmid = cpuid_ebx(0xf);
885 
886 	if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC) ||
887 	    cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL) ||
888 	    cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)) {
889 		u32 eax, ebx, ecx, edx;
890 
891 		/* QoS sub-leaf, EAX=0Fh, ECX=1 */
892 		cpuid_count(0xf, 1, &eax, &ebx, &ecx, &edx);
893 
894 		c->x86_cache_max_rmid  = ecx;
895 		c->x86_cache_occ_scale = ebx;
896 	}
897 }
898 
899 void get_cpu_cap(struct cpuinfo_x86 *c)
900 {
901 	u32 eax, ebx, ecx, edx;
902 
903 	/* Intel-defined flags: level 0x00000001 */
904 	if (c->cpuid_level >= 0x00000001) {
905 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
906 
907 		c->x86_capability[CPUID_1_ECX] = ecx;
908 		c->x86_capability[CPUID_1_EDX] = edx;
909 	}
910 
911 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
912 	if (c->cpuid_level >= 0x00000006)
913 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
914 
915 	/* Additional Intel-defined flags: level 0x00000007 */
916 	if (c->cpuid_level >= 0x00000007) {
917 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
918 		c->x86_capability[CPUID_7_0_EBX] = ebx;
919 		c->x86_capability[CPUID_7_ECX] = ecx;
920 		c->x86_capability[CPUID_7_EDX] = edx;
921 
922 		/* Check valid sub-leaf index before accessing it */
923 		if (eax >= 1) {
924 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
925 			c->x86_capability[CPUID_7_1_EAX] = eax;
926 		}
927 	}
928 
929 	/* Extended state features: level 0x0000000d */
930 	if (c->cpuid_level >= 0x0000000d) {
931 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
932 
933 		c->x86_capability[CPUID_D_1_EAX] = eax;
934 	}
935 
936 	/* AMD-defined flags: level 0x80000001 */
937 	eax = cpuid_eax(0x80000000);
938 	c->extended_cpuid_level = eax;
939 
940 	if ((eax & 0xffff0000) == 0x80000000) {
941 		if (eax >= 0x80000001) {
942 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
943 
944 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
945 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
946 		}
947 	}
948 
949 	if (c->extended_cpuid_level >= 0x80000007) {
950 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
951 
952 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
953 		c->x86_power = edx;
954 	}
955 
956 	if (c->extended_cpuid_level >= 0x80000008) {
957 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
958 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
959 	}
960 
961 	if (c->extended_cpuid_level >= 0x8000000a)
962 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
963 
964 	init_scattered_cpuid_features(c);
965 	init_speculation_control(c);
966 	init_cqm(c);
967 
968 	/*
969 	 * Clear/Set all flags overridden by options, after probe.
970 	 * This needs to happen each time we re-probe, which may happen
971 	 * several times during CPU initialization.
972 	 */
973 	apply_forced_caps(c);
974 }
975 
976 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
977 {
978 	u32 eax, ebx, ecx, edx;
979 
980 	if (c->extended_cpuid_level >= 0x80000008) {
981 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
982 
983 		c->x86_virt_bits = (eax >> 8) & 0xff;
984 		c->x86_phys_bits = eax & 0xff;
985 	}
986 #ifdef CONFIG_X86_32
987 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
988 		c->x86_phys_bits = 36;
989 #endif
990 	c->x86_cache_bits = c->x86_phys_bits;
991 }
992 
993 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
994 {
995 #ifdef CONFIG_X86_32
996 	int i;
997 
998 	/*
999 	 * First of all, decide if this is a 486 or higher
1000 	 * It's a 486 if we can modify the AC flag
1001 	 */
1002 	if (flag_is_changeable_p(X86_EFLAGS_AC))
1003 		c->x86 = 4;
1004 	else
1005 		c->x86 = 3;
1006 
1007 	for (i = 0; i < X86_VENDOR_NUM; i++)
1008 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
1009 			c->x86_vendor_id[0] = 0;
1010 			cpu_devs[i]->c_identify(c);
1011 			if (c->x86_vendor_id[0]) {
1012 				get_cpu_vendor(c);
1013 				break;
1014 			}
1015 		}
1016 #endif
1017 }
1018 
1019 #define NO_SPECULATION	BIT(0)
1020 #define NO_MELTDOWN	BIT(1)
1021 #define NO_SSB		BIT(2)
1022 #define NO_L1TF		BIT(3)
1023 #define NO_MDS		BIT(4)
1024 #define MSBDS_ONLY	BIT(5)
1025 
1026 #define VULNWL(_vendor, _family, _model, _whitelist)	\
1027 	{ X86_VENDOR_##_vendor, _family, _model, X86_FEATURE_ANY, _whitelist }
1028 
1029 #define VULNWL_INTEL(model, whitelist)		\
1030 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1031 
1032 #define VULNWL_AMD(family, whitelist)		\
1033 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1034 
1035 #define VULNWL_HYGON(family, whitelist)		\
1036 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1037 
1038 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1039 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1040 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1041 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1042 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1043 
1044 	/* Intel Family 6 */
1045 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION),
1046 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION),
1047 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION),
1048 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION),
1049 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION),
1050 
1051 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY),
1052 	VULNWL_INTEL(ATOM_SILVERMONT_X,		NO_SSB | NO_L1TF | MSBDS_ONLY),
1053 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY),
1054 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY),
1055 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY),
1056 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY),
1057 
1058 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1059 
1060 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY),
1061 
1062 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF),
1063 	VULNWL_INTEL(ATOM_GOLDMONT_X,		NO_MDS | NO_L1TF),
1064 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF),
1065 
1066 	/* AMD Family 0xf - 0x12 */
1067 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
1068 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
1069 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
1070 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS),
1071 
1072 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1073 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS),
1074 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS),
1075 	{}
1076 };
1077 
1078 static bool __init cpu_matches(unsigned long which)
1079 {
1080 	const struct x86_cpu_id *m = x86_match_cpu(cpu_vuln_whitelist);
1081 
1082 	return m && !!(m->driver_data & which);
1083 }
1084 
1085 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1086 {
1087 	u64 ia32_cap = 0;
1088 
1089 	if (cpu_matches(NO_SPECULATION))
1090 		return;
1091 
1092 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1093 	setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1094 
1095 	if (cpu_has(c, X86_FEATURE_ARCH_CAPABILITIES))
1096 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1097 
1098 	if (!cpu_matches(NO_SSB) && !(ia32_cap & ARCH_CAP_SSB_NO) &&
1099 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1100 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1101 
1102 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1103 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1104 
1105 	if (!cpu_matches(NO_MDS) && !(ia32_cap & ARCH_CAP_MDS_NO)) {
1106 		setup_force_cpu_bug(X86_BUG_MDS);
1107 		if (cpu_matches(MSBDS_ONLY))
1108 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1109 	}
1110 
1111 	if (cpu_matches(NO_MELTDOWN))
1112 		return;
1113 
1114 	/* Rogue Data Cache Load? No! */
1115 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1116 		return;
1117 
1118 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1119 
1120 	if (cpu_matches(NO_L1TF))
1121 		return;
1122 
1123 	setup_force_cpu_bug(X86_BUG_L1TF);
1124 }
1125 
1126 /*
1127  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1128  * unfortunately, that's not true in practice because of early VIA
1129  * chips and (more importantly) broken virtualizers that are not easy
1130  * to detect. In the latter case it doesn't even *fail* reliably, so
1131  * probing for it doesn't even work. Disable it completely on 32-bit
1132  * unless we can find a reliable way to detect all the broken cases.
1133  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1134  */
1135 static void detect_nopl(void)
1136 {
1137 #ifdef CONFIG_X86_32
1138 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1139 #else
1140 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1141 #endif
1142 }
1143 
1144 /*
1145  * Do minimum CPU detection early.
1146  * Fields really needed: vendor, cpuid_level, family, model, mask,
1147  * cache alignment.
1148  * The others are not touched to avoid unwanted side effects.
1149  *
1150  * WARNING: this function is only called on the boot CPU.  Don't add code
1151  * here that is supposed to run on all CPUs.
1152  */
1153 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1154 {
1155 #ifdef CONFIG_X86_64
1156 	c->x86_clflush_size = 64;
1157 	c->x86_phys_bits = 36;
1158 	c->x86_virt_bits = 48;
1159 #else
1160 	c->x86_clflush_size = 32;
1161 	c->x86_phys_bits = 32;
1162 	c->x86_virt_bits = 32;
1163 #endif
1164 	c->x86_cache_alignment = c->x86_clflush_size;
1165 
1166 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1167 	c->extended_cpuid_level = 0;
1168 
1169 	if (!have_cpuid_p())
1170 		identify_cpu_without_cpuid(c);
1171 
1172 	/* cyrix could have cpuid enabled via c_identify()*/
1173 	if (have_cpuid_p()) {
1174 		cpu_detect(c);
1175 		get_cpu_vendor(c);
1176 		get_cpu_cap(c);
1177 		get_cpu_address_sizes(c);
1178 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1179 
1180 		if (this_cpu->c_early_init)
1181 			this_cpu->c_early_init(c);
1182 
1183 		c->cpu_index = 0;
1184 		filter_cpuid_features(c, false);
1185 
1186 		if (this_cpu->c_bsp_init)
1187 			this_cpu->c_bsp_init(c);
1188 	} else {
1189 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1190 	}
1191 
1192 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1193 
1194 	cpu_set_bug_bits(c);
1195 
1196 	fpu__init_system(c);
1197 
1198 #ifdef CONFIG_X86_32
1199 	/*
1200 	 * Regardless of whether PCID is enumerated, the SDM says
1201 	 * that it can't be enabled in 32-bit mode.
1202 	 */
1203 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1204 #endif
1205 
1206 	/*
1207 	 * Later in the boot process pgtable_l5_enabled() relies on
1208 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1209 	 * enabled by this point we need to clear the feature bit to avoid
1210 	 * false-positives at the later stage.
1211 	 *
1212 	 * pgtable_l5_enabled() can be false here for several reasons:
1213 	 *  - 5-level paging is disabled compile-time;
1214 	 *  - it's 32-bit kernel;
1215 	 *  - machine doesn't support 5-level paging;
1216 	 *  - user specified 'no5lvl' in kernel command line.
1217 	 */
1218 	if (!pgtable_l5_enabled())
1219 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1220 
1221 	detect_nopl();
1222 }
1223 
1224 void __init early_cpu_init(void)
1225 {
1226 	const struct cpu_dev *const *cdev;
1227 	int count = 0;
1228 
1229 #ifdef CONFIG_PROCESSOR_SELECT
1230 	pr_info("KERNEL supported cpus:\n");
1231 #endif
1232 
1233 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1234 		const struct cpu_dev *cpudev = *cdev;
1235 
1236 		if (count >= X86_VENDOR_NUM)
1237 			break;
1238 		cpu_devs[count] = cpudev;
1239 		count++;
1240 
1241 #ifdef CONFIG_PROCESSOR_SELECT
1242 		{
1243 			unsigned int j;
1244 
1245 			for (j = 0; j < 2; j++) {
1246 				if (!cpudev->c_ident[j])
1247 					continue;
1248 				pr_info("  %s %s\n", cpudev->c_vendor,
1249 					cpudev->c_ident[j]);
1250 			}
1251 		}
1252 #endif
1253 	}
1254 	early_identify_cpu(&boot_cpu_data);
1255 }
1256 
1257 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1258 {
1259 #ifdef CONFIG_X86_64
1260 	/*
1261 	 * Empirically, writing zero to a segment selector on AMD does
1262 	 * not clear the base, whereas writing zero to a segment
1263 	 * selector on Intel does clear the base.  Intel's behavior
1264 	 * allows slightly faster context switches in the common case
1265 	 * where GS is unused by the prev and next threads.
1266 	 *
1267 	 * Since neither vendor documents this anywhere that I can see,
1268 	 * detect it directly instead of hardcoding the choice by
1269 	 * vendor.
1270 	 *
1271 	 * I've designated AMD's behavior as the "bug" because it's
1272 	 * counterintuitive and less friendly.
1273 	 */
1274 
1275 	unsigned long old_base, tmp;
1276 	rdmsrl(MSR_FS_BASE, old_base);
1277 	wrmsrl(MSR_FS_BASE, 1);
1278 	loadsegment(fs, 0);
1279 	rdmsrl(MSR_FS_BASE, tmp);
1280 	if (tmp != 0)
1281 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1282 	wrmsrl(MSR_FS_BASE, old_base);
1283 #endif
1284 }
1285 
1286 static void generic_identify(struct cpuinfo_x86 *c)
1287 {
1288 	c->extended_cpuid_level = 0;
1289 
1290 	if (!have_cpuid_p())
1291 		identify_cpu_without_cpuid(c);
1292 
1293 	/* cyrix could have cpuid enabled via c_identify()*/
1294 	if (!have_cpuid_p())
1295 		return;
1296 
1297 	cpu_detect(c);
1298 
1299 	get_cpu_vendor(c);
1300 
1301 	get_cpu_cap(c);
1302 
1303 	get_cpu_address_sizes(c);
1304 
1305 	if (c->cpuid_level >= 0x00000001) {
1306 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1307 #ifdef CONFIG_X86_32
1308 # ifdef CONFIG_SMP
1309 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1310 # else
1311 		c->apicid = c->initial_apicid;
1312 # endif
1313 #endif
1314 		c->phys_proc_id = c->initial_apicid;
1315 	}
1316 
1317 	get_model_name(c); /* Default name */
1318 
1319 	detect_null_seg_behavior(c);
1320 
1321 	/*
1322 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1323 	 * systems that run Linux at CPL > 0 may or may not have the
1324 	 * issue, but, even if they have the issue, there's absolutely
1325 	 * nothing we can do about it because we can't use the real IRET
1326 	 * instruction.
1327 	 *
1328 	 * NB: For the time being, only 32-bit kernels support
1329 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1330 	 * whether to apply espfix using paravirt hooks.  If any
1331 	 * non-paravirt system ever shows up that does *not* have the
1332 	 * ESPFIX issue, we can change this.
1333 	 */
1334 #ifdef CONFIG_X86_32
1335 # ifdef CONFIG_PARAVIRT_XXL
1336 	do {
1337 		extern void native_iret(void);
1338 		if (pv_ops.cpu.iret == native_iret)
1339 			set_cpu_bug(c, X86_BUG_ESPFIX);
1340 	} while (0);
1341 # else
1342 	set_cpu_bug(c, X86_BUG_ESPFIX);
1343 # endif
1344 #endif
1345 }
1346 
1347 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
1348 {
1349 	/*
1350 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
1351 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
1352 	 * in case CQM bits really aren't there in this CPU.
1353 	 */
1354 	if (c != &boot_cpu_data) {
1355 		boot_cpu_data.x86_cache_max_rmid =
1356 			min(boot_cpu_data.x86_cache_max_rmid,
1357 			    c->x86_cache_max_rmid);
1358 	}
1359 }
1360 
1361 /*
1362  * Validate that ACPI/mptables have the same information about the
1363  * effective APIC id and update the package map.
1364  */
1365 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1366 {
1367 #ifdef CONFIG_SMP
1368 	unsigned int apicid, cpu = smp_processor_id();
1369 
1370 	apicid = apic->cpu_present_to_apicid(cpu);
1371 
1372 	if (apicid != c->apicid) {
1373 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1374 		       cpu, apicid, c->initial_apicid);
1375 	}
1376 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1377 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1378 #else
1379 	c->logical_proc_id = 0;
1380 #endif
1381 }
1382 
1383 /*
1384  * This does the hard work of actually picking apart the CPU stuff...
1385  */
1386 static void identify_cpu(struct cpuinfo_x86 *c)
1387 {
1388 	int i;
1389 
1390 	c->loops_per_jiffy = loops_per_jiffy;
1391 	c->x86_cache_size = 0;
1392 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1393 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1394 	c->x86_vendor_id[0] = '\0'; /* Unset */
1395 	c->x86_model_id[0] = '\0';  /* Unset */
1396 	c->x86_max_cores = 1;
1397 	c->x86_coreid_bits = 0;
1398 	c->cu_id = 0xff;
1399 #ifdef CONFIG_X86_64
1400 	c->x86_clflush_size = 64;
1401 	c->x86_phys_bits = 36;
1402 	c->x86_virt_bits = 48;
1403 #else
1404 	c->cpuid_level = -1;	/* CPUID not detected */
1405 	c->x86_clflush_size = 32;
1406 	c->x86_phys_bits = 32;
1407 	c->x86_virt_bits = 32;
1408 #endif
1409 	c->x86_cache_alignment = c->x86_clflush_size;
1410 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1411 
1412 	generic_identify(c);
1413 
1414 	if (this_cpu->c_identify)
1415 		this_cpu->c_identify(c);
1416 
1417 	/* Clear/Set all flags overridden by options, after probe */
1418 	apply_forced_caps(c);
1419 
1420 #ifdef CONFIG_X86_64
1421 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1422 #endif
1423 
1424 	/*
1425 	 * Vendor-specific initialization.  In this section we
1426 	 * canonicalize the feature flags, meaning if there are
1427 	 * features a certain CPU supports which CPUID doesn't
1428 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1429 	 * we handle them here.
1430 	 *
1431 	 * At the end of this section, c->x86_capability better
1432 	 * indicate the features this CPU genuinely supports!
1433 	 */
1434 	if (this_cpu->c_init)
1435 		this_cpu->c_init(c);
1436 
1437 	/* Disable the PN if appropriate */
1438 	squash_the_stupid_serial_number(c);
1439 
1440 	/* Set up SMEP/SMAP/UMIP */
1441 	setup_smep(c);
1442 	setup_smap(c);
1443 	setup_umip(c);
1444 
1445 	/*
1446 	 * The vendor-specific functions might have changed features.
1447 	 * Now we do "generic changes."
1448 	 */
1449 
1450 	/* Filter out anything that depends on CPUID levels we don't have */
1451 	filter_cpuid_features(c, true);
1452 
1453 	/* If the model name is still unset, do table lookup. */
1454 	if (!c->x86_model_id[0]) {
1455 		const char *p;
1456 		p = table_lookup_model(c);
1457 		if (p)
1458 			strcpy(c->x86_model_id, p);
1459 		else
1460 			/* Last resort... */
1461 			sprintf(c->x86_model_id, "%02x/%02x",
1462 				c->x86, c->x86_model);
1463 	}
1464 
1465 #ifdef CONFIG_X86_64
1466 	detect_ht(c);
1467 #endif
1468 
1469 	x86_init_rdrand(c);
1470 	x86_init_cache_qos(c);
1471 	setup_pku(c);
1472 
1473 	/*
1474 	 * Clear/Set all flags overridden by options, need do it
1475 	 * before following smp all cpus cap AND.
1476 	 */
1477 	apply_forced_caps(c);
1478 
1479 	/*
1480 	 * On SMP, boot_cpu_data holds the common feature set between
1481 	 * all CPUs; so make sure that we indicate which features are
1482 	 * common between the CPUs.  The first time this routine gets
1483 	 * executed, c == &boot_cpu_data.
1484 	 */
1485 	if (c != &boot_cpu_data) {
1486 		/* AND the already accumulated flags with these */
1487 		for (i = 0; i < NCAPINTS; i++)
1488 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1489 
1490 		/* OR, i.e. replicate the bug flags */
1491 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1492 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1493 	}
1494 
1495 	/* Init Machine Check Exception if available. */
1496 	mcheck_cpu_init(c);
1497 
1498 	select_idle_routine(c);
1499 
1500 #ifdef CONFIG_NUMA
1501 	numa_add_cpu(smp_processor_id());
1502 #endif
1503 }
1504 
1505 /*
1506  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1507  * on 32-bit kernels:
1508  */
1509 #ifdef CONFIG_X86_32
1510 void enable_sep_cpu(void)
1511 {
1512 	struct tss_struct *tss;
1513 	int cpu;
1514 
1515 	if (!boot_cpu_has(X86_FEATURE_SEP))
1516 		return;
1517 
1518 	cpu = get_cpu();
1519 	tss = &per_cpu(cpu_tss_rw, cpu);
1520 
1521 	/*
1522 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1523 	 * see the big comment in struct x86_hw_tss's definition.
1524 	 */
1525 
1526 	tss->x86_tss.ss1 = __KERNEL_CS;
1527 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1528 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1529 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1530 
1531 	put_cpu();
1532 }
1533 #endif
1534 
1535 void __init identify_boot_cpu(void)
1536 {
1537 	identify_cpu(&boot_cpu_data);
1538 #ifdef CONFIG_X86_32
1539 	sysenter_setup();
1540 	enable_sep_cpu();
1541 #endif
1542 	cpu_detect_tlb(&boot_cpu_data);
1543 	setup_cr_pinning();
1544 }
1545 
1546 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1547 {
1548 	BUG_ON(c == &boot_cpu_data);
1549 	identify_cpu(c);
1550 #ifdef CONFIG_X86_32
1551 	enable_sep_cpu();
1552 #endif
1553 	mtrr_ap_init();
1554 	validate_apic_and_package_id(c);
1555 	x86_spec_ctrl_setup_ap();
1556 }
1557 
1558 static __init int setup_noclflush(char *arg)
1559 {
1560 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1561 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1562 	return 1;
1563 }
1564 __setup("noclflush", setup_noclflush);
1565 
1566 void print_cpu_info(struct cpuinfo_x86 *c)
1567 {
1568 	const char *vendor = NULL;
1569 
1570 	if (c->x86_vendor < X86_VENDOR_NUM) {
1571 		vendor = this_cpu->c_vendor;
1572 	} else {
1573 		if (c->cpuid_level >= 0)
1574 			vendor = c->x86_vendor_id;
1575 	}
1576 
1577 	if (vendor && !strstr(c->x86_model_id, vendor))
1578 		pr_cont("%s ", vendor);
1579 
1580 	if (c->x86_model_id[0])
1581 		pr_cont("%s", c->x86_model_id);
1582 	else
1583 		pr_cont("%d86", c->x86);
1584 
1585 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1586 
1587 	if (c->x86_stepping || c->cpuid_level >= 0)
1588 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1589 	else
1590 		pr_cont(")\n");
1591 }
1592 
1593 /*
1594  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1595  * But we need to keep a dummy __setup around otherwise it would
1596  * show up as an environment variable for init.
1597  */
1598 static __init int setup_clearcpuid(char *arg)
1599 {
1600 	return 1;
1601 }
1602 __setup("clearcpuid=", setup_clearcpuid);
1603 
1604 #ifdef CONFIG_X86_64
1605 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1606 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1607 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1608 
1609 /*
1610  * The following percpu variables are hot.  Align current_task to
1611  * cacheline size such that they fall in the same cacheline.
1612  */
1613 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1614 	&init_task;
1615 EXPORT_PER_CPU_SYMBOL(current_task);
1616 
1617 DEFINE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
1618 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1619 
1620 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1621 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1622 
1623 /* May not be marked __init: used by software suspend */
1624 void syscall_init(void)
1625 {
1626 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1627 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1628 
1629 #ifdef CONFIG_IA32_EMULATION
1630 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1631 	/*
1632 	 * This only works on Intel CPUs.
1633 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1634 	 * This does not cause SYSENTER to jump to the wrong location, because
1635 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1636 	 */
1637 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1638 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1639 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1640 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1641 #else
1642 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1643 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1644 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1645 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1646 #endif
1647 
1648 	/* Flags to clear on syscall */
1649 	wrmsrl(MSR_SYSCALL_MASK,
1650 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1651 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1652 }
1653 
1654 DEFINE_PER_CPU(int, debug_stack_usage);
1655 DEFINE_PER_CPU(u32, debug_idt_ctr);
1656 
1657 void debug_stack_set_zero(void)
1658 {
1659 	this_cpu_inc(debug_idt_ctr);
1660 	load_current_idt();
1661 }
1662 NOKPROBE_SYMBOL(debug_stack_set_zero);
1663 
1664 void debug_stack_reset(void)
1665 {
1666 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1667 		return;
1668 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1669 		load_current_idt();
1670 }
1671 NOKPROBE_SYMBOL(debug_stack_reset);
1672 
1673 #else	/* CONFIG_X86_64 */
1674 
1675 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1676 EXPORT_PER_CPU_SYMBOL(current_task);
1677 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1678 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1679 
1680 /*
1681  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1682  * the top of the kernel stack.  Use an extra percpu variable to track the
1683  * top of the kernel stack directly.
1684  */
1685 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1686 	(unsigned long)&init_thread_union + THREAD_SIZE;
1687 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1688 
1689 #ifdef CONFIG_STACKPROTECTOR
1690 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1691 #endif
1692 
1693 #endif	/* CONFIG_X86_64 */
1694 
1695 /*
1696  * Clear all 6 debug registers:
1697  */
1698 static void clear_all_debug_regs(void)
1699 {
1700 	int i;
1701 
1702 	for (i = 0; i < 8; i++) {
1703 		/* Ignore db4, db5 */
1704 		if ((i == 4) || (i == 5))
1705 			continue;
1706 
1707 		set_debugreg(0, i);
1708 	}
1709 }
1710 
1711 #ifdef CONFIG_KGDB
1712 /*
1713  * Restore debug regs if using kgdbwait and you have a kernel debugger
1714  * connection established.
1715  */
1716 static void dbg_restore_debug_regs(void)
1717 {
1718 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1719 		arch_kgdb_ops.correct_hw_break();
1720 }
1721 #else /* ! CONFIG_KGDB */
1722 #define dbg_restore_debug_regs()
1723 #endif /* ! CONFIG_KGDB */
1724 
1725 static void wait_for_master_cpu(int cpu)
1726 {
1727 #ifdef CONFIG_SMP
1728 	/*
1729 	 * wait for ACK from master CPU before continuing
1730 	 * with AP initialization
1731 	 */
1732 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1733 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1734 		cpu_relax();
1735 #endif
1736 }
1737 
1738 #ifdef CONFIG_X86_64
1739 static void setup_getcpu(int cpu)
1740 {
1741 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1742 	struct desc_struct d = { };
1743 
1744 	if (boot_cpu_has(X86_FEATURE_RDTSCP))
1745 		write_rdtscp_aux(cpudata);
1746 
1747 	/* Store CPU and node number in limit. */
1748 	d.limit0 = cpudata;
1749 	d.limit1 = cpudata >> 16;
1750 
1751 	d.type = 5;		/* RO data, expand down, accessed */
1752 	d.dpl = 3;		/* Visible to user code */
1753 	d.s = 1;		/* Not a system segment */
1754 	d.p = 1;		/* Present */
1755 	d.d = 1;		/* 32-bit */
1756 
1757 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1758 }
1759 #endif
1760 
1761 /*
1762  * cpu_init() initializes state that is per-CPU. Some data is already
1763  * initialized (naturally) in the bootstrap process, such as the GDT
1764  * and IDT. We reload them nevertheless, this function acts as a
1765  * 'CPU state barrier', nothing should get across.
1766  */
1767 #ifdef CONFIG_X86_64
1768 
1769 void cpu_init(void)
1770 {
1771 	int cpu = raw_smp_processor_id();
1772 	struct task_struct *me;
1773 	struct tss_struct *t;
1774 	int i;
1775 
1776 	wait_for_master_cpu(cpu);
1777 
1778 	if (cpu)
1779 		load_ucode_ap();
1780 
1781 	t = &per_cpu(cpu_tss_rw, cpu);
1782 
1783 #ifdef CONFIG_NUMA
1784 	if (this_cpu_read(numa_node) == 0 &&
1785 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1786 		set_numa_node(early_cpu_to_node(cpu));
1787 #endif
1788 	setup_getcpu(cpu);
1789 
1790 	me = current;
1791 
1792 	pr_debug("Initializing CPU#%d\n", cpu);
1793 
1794 	cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1795 
1796 	/*
1797 	 * Initialize the per-CPU GDT with the boot GDT,
1798 	 * and set up the GDT descriptor:
1799 	 */
1800 
1801 	switch_to_new_gdt(cpu);
1802 	loadsegment(fs, 0);
1803 
1804 	load_current_idt();
1805 
1806 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1807 	syscall_init();
1808 
1809 	wrmsrl(MSR_FS_BASE, 0);
1810 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1811 	barrier();
1812 
1813 	x86_configure_nx();
1814 	x2apic_setup();
1815 
1816 	/*
1817 	 * set up and load the per-CPU TSS
1818 	 */
1819 	if (!t->x86_tss.ist[0]) {
1820 		t->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1821 		t->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1822 		t->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1823 		t->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1824 	}
1825 
1826 	t->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
1827 
1828 	/*
1829 	 * <= is required because the CPU will access up to
1830 	 * 8 bits beyond the end of the IO permission bitmap.
1831 	 */
1832 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1833 		t->io_bitmap[i] = ~0UL;
1834 
1835 	mmgrab(&init_mm);
1836 	me->active_mm = &init_mm;
1837 	BUG_ON(me->mm);
1838 	initialize_tlbstate_and_flush();
1839 	enter_lazy_tlb(&init_mm, me);
1840 
1841 	/*
1842 	 * Initialize the TSS.  sp0 points to the entry trampoline stack
1843 	 * regardless of what task is running.
1844 	 */
1845 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1846 	load_TR_desc();
1847 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1848 
1849 	load_mm_ldt(&init_mm);
1850 
1851 	clear_all_debug_regs();
1852 	dbg_restore_debug_regs();
1853 
1854 	fpu__init_cpu();
1855 
1856 	if (is_uv_system())
1857 		uv_cpu_init();
1858 
1859 	load_fixmap_gdt(cpu);
1860 }
1861 
1862 #else
1863 
1864 void cpu_init(void)
1865 {
1866 	int cpu = smp_processor_id();
1867 	struct task_struct *curr = current;
1868 	struct tss_struct *t = &per_cpu(cpu_tss_rw, cpu);
1869 
1870 	wait_for_master_cpu(cpu);
1871 
1872 	show_ucode_info_early();
1873 
1874 	pr_info("Initializing CPU#%d\n", cpu);
1875 
1876 	if (cpu_feature_enabled(X86_FEATURE_VME) ||
1877 	    boot_cpu_has(X86_FEATURE_TSC) ||
1878 	    boot_cpu_has(X86_FEATURE_DE))
1879 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1880 
1881 	load_current_idt();
1882 	switch_to_new_gdt(cpu);
1883 
1884 	/*
1885 	 * Set up and load the per-CPU TSS and LDT
1886 	 */
1887 	mmgrab(&init_mm);
1888 	curr->active_mm = &init_mm;
1889 	BUG_ON(curr->mm);
1890 	initialize_tlbstate_and_flush();
1891 	enter_lazy_tlb(&init_mm, curr);
1892 
1893 	/*
1894 	 * Initialize the TSS.  sp0 points to the entry trampoline stack
1895 	 * regardless of what task is running.
1896 	 */
1897 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1898 	load_TR_desc();
1899 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1900 
1901 	load_mm_ldt(&init_mm);
1902 
1903 	t->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
1904 
1905 #ifdef CONFIG_DOUBLEFAULT
1906 	/* Set up doublefault TSS pointer in the GDT */
1907 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1908 #endif
1909 
1910 	clear_all_debug_regs();
1911 	dbg_restore_debug_regs();
1912 
1913 	fpu__init_cpu();
1914 
1915 	load_fixmap_gdt(cpu);
1916 }
1917 #endif
1918 
1919 /*
1920  * The microcode loader calls this upon late microcode load to recheck features,
1921  * only when microcode has been updated. Caller holds microcode_mutex and CPU
1922  * hotplug lock.
1923  */
1924 void microcode_check(void)
1925 {
1926 	struct cpuinfo_x86 info;
1927 
1928 	perf_check_microcode();
1929 
1930 	/* Reload CPUID max function as it might've changed. */
1931 	info.cpuid_level = cpuid_eax(0);
1932 
1933 	/*
1934 	 * Copy all capability leafs to pick up the synthetic ones so that
1935 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
1936 	 * get overwritten in get_cpu_cap().
1937 	 */
1938 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
1939 
1940 	get_cpu_cap(&info);
1941 
1942 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
1943 		return;
1944 
1945 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
1946 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
1947 }
1948