xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 89df62c3)
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/sched/smt.h>
18 #include <linux/init.h>
19 #include <linux/kprobes.h>
20 #include <linux/kgdb.h>
21 #include <linux/smp.h>
22 #include <linux/io.h>
23 #include <linux/syscore_ops.h>
24 #include <linux/pgtable.h>
25 #include <linux/stackprotector.h>
26 
27 #include <asm/cmdline.h>
28 #include <asm/perf_event.h>
29 #include <asm/mmu_context.h>
30 #include <asm/doublefault.h>
31 #include <asm/archrandom.h>
32 #include <asm/hypervisor.h>
33 #include <asm/processor.h>
34 #include <asm/tlbflush.h>
35 #include <asm/debugreg.h>
36 #include <asm/sections.h>
37 #include <asm/vsyscall.h>
38 #include <linux/topology.h>
39 #include <linux/cpumask.h>
40 #include <linux/atomic.h>
41 #include <asm/proto.h>
42 #include <asm/setup.h>
43 #include <asm/apic.h>
44 #include <asm/desc.h>
45 #include <asm/fpu/api.h>
46 #include <asm/mtrr.h>
47 #include <asm/hwcap2.h>
48 #include <linux/numa.h>
49 #include <asm/numa.h>
50 #include <asm/asm.h>
51 #include <asm/bugs.h>
52 #include <asm/cpu.h>
53 #include <asm/mce.h>
54 #include <asm/msr.h>
55 #include <asm/cacheinfo.h>
56 #include <asm/memtype.h>
57 #include <asm/microcode.h>
58 #include <asm/microcode_intel.h>
59 #include <asm/intel-family.h>
60 #include <asm/cpu_device_id.h>
61 #include <asm/uv/uv.h>
62 #include <asm/sigframe.h>
63 #include <asm/traps.h>
64 #include <asm/sev.h>
65 
66 #include "cpu.h"
67 
68 u32 elf_hwcap2 __read_mostly;
69 
70 /* all of these masks are initialized in setup_cpu_local_masks() */
71 cpumask_var_t cpu_initialized_mask;
72 cpumask_var_t cpu_callout_mask;
73 cpumask_var_t cpu_callin_mask;
74 
75 /* representing cpus for which sibling maps can be computed */
76 cpumask_var_t cpu_sibling_setup_mask;
77 
78 /* Number of siblings per CPU package */
79 int smp_num_siblings = 1;
80 EXPORT_SYMBOL(smp_num_siblings);
81 
82 /* Last level cache ID of each logical CPU */
83 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
84 
85 u16 get_llc_id(unsigned int cpu)
86 {
87 	return per_cpu(cpu_llc_id, cpu);
88 }
89 EXPORT_SYMBOL_GPL(get_llc_id);
90 
91 /* L2 cache ID of each logical CPU */
92 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_l2c_id) = BAD_APICID;
93 
94 static struct ppin_info {
95 	int	feature;
96 	int	msr_ppin_ctl;
97 	int	msr_ppin;
98 } ppin_info[] = {
99 	[X86_VENDOR_INTEL] = {
100 		.feature = X86_FEATURE_INTEL_PPIN,
101 		.msr_ppin_ctl = MSR_PPIN_CTL,
102 		.msr_ppin = MSR_PPIN
103 	},
104 	[X86_VENDOR_AMD] = {
105 		.feature = X86_FEATURE_AMD_PPIN,
106 		.msr_ppin_ctl = MSR_AMD_PPIN_CTL,
107 		.msr_ppin = MSR_AMD_PPIN
108 	},
109 };
110 
111 static const struct x86_cpu_id ppin_cpuids[] = {
112 	X86_MATCH_FEATURE(X86_FEATURE_AMD_PPIN, &ppin_info[X86_VENDOR_AMD]),
113 	X86_MATCH_FEATURE(X86_FEATURE_INTEL_PPIN, &ppin_info[X86_VENDOR_INTEL]),
114 
115 	/* Legacy models without CPUID enumeration */
116 	X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE_X, &ppin_info[X86_VENDOR_INTEL]),
117 	X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, &ppin_info[X86_VENDOR_INTEL]),
118 	X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_D, &ppin_info[X86_VENDOR_INTEL]),
119 	X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, &ppin_info[X86_VENDOR_INTEL]),
120 	X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_X, &ppin_info[X86_VENDOR_INTEL]),
121 	X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_X, &ppin_info[X86_VENDOR_INTEL]),
122 	X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_D, &ppin_info[X86_VENDOR_INTEL]),
123 	X86_MATCH_INTEL_FAM6_MODEL(SAPPHIRERAPIDS_X, &ppin_info[X86_VENDOR_INTEL]),
124 	X86_MATCH_INTEL_FAM6_MODEL(EMERALDRAPIDS_X, &ppin_info[X86_VENDOR_INTEL]),
125 	X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNL, &ppin_info[X86_VENDOR_INTEL]),
126 	X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNM, &ppin_info[X86_VENDOR_INTEL]),
127 
128 	{}
129 };
130 
131 static void ppin_init(struct cpuinfo_x86 *c)
132 {
133 	const struct x86_cpu_id *id;
134 	unsigned long long val;
135 	struct ppin_info *info;
136 
137 	id = x86_match_cpu(ppin_cpuids);
138 	if (!id)
139 		return;
140 
141 	/*
142 	 * Testing the presence of the MSR is not enough. Need to check
143 	 * that the PPIN_CTL allows reading of the PPIN.
144 	 */
145 	info = (struct ppin_info *)id->driver_data;
146 
147 	if (rdmsrl_safe(info->msr_ppin_ctl, &val))
148 		goto clear_ppin;
149 
150 	if ((val & 3UL) == 1UL) {
151 		/* PPIN locked in disabled mode */
152 		goto clear_ppin;
153 	}
154 
155 	/* If PPIN is disabled, try to enable */
156 	if (!(val & 2UL)) {
157 		wrmsrl_safe(info->msr_ppin_ctl,  val | 2UL);
158 		rdmsrl_safe(info->msr_ppin_ctl, &val);
159 	}
160 
161 	/* Is the enable bit set? */
162 	if (val & 2UL) {
163 		c->ppin = __rdmsr(info->msr_ppin);
164 		set_cpu_cap(c, info->feature);
165 		return;
166 	}
167 
168 clear_ppin:
169 	clear_cpu_cap(c, info->feature);
170 }
171 
172 /* correctly size the local cpu masks */
173 void __init setup_cpu_local_masks(void)
174 {
175 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
176 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
177 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
178 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
179 }
180 
181 static void default_init(struct cpuinfo_x86 *c)
182 {
183 #ifdef CONFIG_X86_64
184 	cpu_detect_cache_sizes(c);
185 #else
186 	/* Not much we can do here... */
187 	/* Check if at least it has cpuid */
188 	if (c->cpuid_level == -1) {
189 		/* No cpuid. It must be an ancient CPU */
190 		if (c->x86 == 4)
191 			strcpy(c->x86_model_id, "486");
192 		else if (c->x86 == 3)
193 			strcpy(c->x86_model_id, "386");
194 	}
195 #endif
196 }
197 
198 static const struct cpu_dev default_cpu = {
199 	.c_init		= default_init,
200 	.c_vendor	= "Unknown",
201 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
202 };
203 
204 static const struct cpu_dev *this_cpu = &default_cpu;
205 
206 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
207 #ifdef CONFIG_X86_64
208 	/*
209 	 * We need valid kernel segments for data and code in long mode too
210 	 * IRET will check the segment types  kkeil 2000/10/28
211 	 * Also sysret mandates a special GDT layout
212 	 *
213 	 * TLS descriptors are currently at a different place compared to i386.
214 	 * Hopefully nobody expects them at a fixed place (Wine?)
215 	 */
216 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
217 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
218 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
219 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
220 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
221 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
222 #else
223 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
224 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
225 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
226 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
227 	/*
228 	 * Segments used for calling PnP BIOS have byte granularity.
229 	 * They code segments and data segments have fixed 64k limits,
230 	 * the transfer segment sizes are set at run time.
231 	 */
232 	/* 32-bit code */
233 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
234 	/* 16-bit code */
235 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
236 	/* 16-bit data */
237 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
238 	/* 16-bit data */
239 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
240 	/* 16-bit data */
241 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
242 	/*
243 	 * The APM segments have byte granularity and their bases
244 	 * are set at run time.  All have 64k limits.
245 	 */
246 	/* 32-bit code */
247 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
248 	/* 16-bit code */
249 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
250 	/* data */
251 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
252 
253 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
254 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
255 #endif
256 } };
257 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
258 
259 #ifdef CONFIG_X86_64
260 static int __init x86_nopcid_setup(char *s)
261 {
262 	/* nopcid doesn't accept parameters */
263 	if (s)
264 		return -EINVAL;
265 
266 	/* do not emit a message if the feature is not present */
267 	if (!boot_cpu_has(X86_FEATURE_PCID))
268 		return 0;
269 
270 	setup_clear_cpu_cap(X86_FEATURE_PCID);
271 	pr_info("nopcid: PCID feature disabled\n");
272 	return 0;
273 }
274 early_param("nopcid", x86_nopcid_setup);
275 #endif
276 
277 static int __init x86_noinvpcid_setup(char *s)
278 {
279 	/* noinvpcid doesn't accept parameters */
280 	if (s)
281 		return -EINVAL;
282 
283 	/* do not emit a message if the feature is not present */
284 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
285 		return 0;
286 
287 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
288 	pr_info("noinvpcid: INVPCID feature disabled\n");
289 	return 0;
290 }
291 early_param("noinvpcid", x86_noinvpcid_setup);
292 
293 #ifdef CONFIG_X86_32
294 static int cachesize_override = -1;
295 static int disable_x86_serial_nr = 1;
296 
297 static int __init cachesize_setup(char *str)
298 {
299 	get_option(&str, &cachesize_override);
300 	return 1;
301 }
302 __setup("cachesize=", cachesize_setup);
303 
304 /* Standard macro to see if a specific flag is changeable */
305 static inline int flag_is_changeable_p(u32 flag)
306 {
307 	u32 f1, f2;
308 
309 	/*
310 	 * Cyrix and IDT cpus allow disabling of CPUID
311 	 * so the code below may return different results
312 	 * when it is executed before and after enabling
313 	 * the CPUID. Add "volatile" to not allow gcc to
314 	 * optimize the subsequent calls to this function.
315 	 */
316 	asm volatile ("pushfl		\n\t"
317 		      "pushfl		\n\t"
318 		      "popl %0		\n\t"
319 		      "movl %0, %1	\n\t"
320 		      "xorl %2, %0	\n\t"
321 		      "pushl %0		\n\t"
322 		      "popfl		\n\t"
323 		      "pushfl		\n\t"
324 		      "popl %0		\n\t"
325 		      "popfl		\n\t"
326 
327 		      : "=&r" (f1), "=&r" (f2)
328 		      : "ir" (flag));
329 
330 	return ((f1^f2) & flag) != 0;
331 }
332 
333 /* Probe for the CPUID instruction */
334 int have_cpuid_p(void)
335 {
336 	return flag_is_changeable_p(X86_EFLAGS_ID);
337 }
338 
339 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
340 {
341 	unsigned long lo, hi;
342 
343 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
344 		return;
345 
346 	/* Disable processor serial number: */
347 
348 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
349 	lo |= 0x200000;
350 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
351 
352 	pr_notice("CPU serial number disabled.\n");
353 	clear_cpu_cap(c, X86_FEATURE_PN);
354 
355 	/* Disabling the serial number may affect the cpuid level */
356 	c->cpuid_level = cpuid_eax(0);
357 }
358 
359 static int __init x86_serial_nr_setup(char *s)
360 {
361 	disable_x86_serial_nr = 0;
362 	return 1;
363 }
364 __setup("serialnumber", x86_serial_nr_setup);
365 #else
366 static inline int flag_is_changeable_p(u32 flag)
367 {
368 	return 1;
369 }
370 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
371 {
372 }
373 #endif
374 
375 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
376 {
377 	if (cpu_has(c, X86_FEATURE_SMEP))
378 		cr4_set_bits(X86_CR4_SMEP);
379 }
380 
381 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
382 {
383 	unsigned long eflags = native_save_fl();
384 
385 	/* This should have been cleared long ago */
386 	BUG_ON(eflags & X86_EFLAGS_AC);
387 
388 	if (cpu_has(c, X86_FEATURE_SMAP))
389 		cr4_set_bits(X86_CR4_SMAP);
390 }
391 
392 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
393 {
394 	/* Check the boot processor, plus build option for UMIP. */
395 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
396 		goto out;
397 
398 	/* Check the current processor's cpuid bits. */
399 	if (!cpu_has(c, X86_FEATURE_UMIP))
400 		goto out;
401 
402 	cr4_set_bits(X86_CR4_UMIP);
403 
404 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
405 
406 	return;
407 
408 out:
409 	/*
410 	 * Make sure UMIP is disabled in case it was enabled in a
411 	 * previous boot (e.g., via kexec).
412 	 */
413 	cr4_clear_bits(X86_CR4_UMIP);
414 }
415 
416 /* These bits should not change their value after CPU init is finished. */
417 static const unsigned long cr4_pinned_mask =
418 	X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP |
419 	X86_CR4_FSGSBASE | X86_CR4_CET;
420 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
421 static unsigned long cr4_pinned_bits __ro_after_init;
422 
423 void native_write_cr0(unsigned long val)
424 {
425 	unsigned long bits_missing = 0;
426 
427 set_register:
428 	asm volatile("mov %0,%%cr0": "+r" (val) : : "memory");
429 
430 	if (static_branch_likely(&cr_pinning)) {
431 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
432 			bits_missing = X86_CR0_WP;
433 			val |= bits_missing;
434 			goto set_register;
435 		}
436 		/* Warn after we've set the missing bits. */
437 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
438 	}
439 }
440 EXPORT_SYMBOL(native_write_cr0);
441 
442 void __no_profile native_write_cr4(unsigned long val)
443 {
444 	unsigned long bits_changed = 0;
445 
446 set_register:
447 	asm volatile("mov %0,%%cr4": "+r" (val) : : "memory");
448 
449 	if (static_branch_likely(&cr_pinning)) {
450 		if (unlikely((val & cr4_pinned_mask) != cr4_pinned_bits)) {
451 			bits_changed = (val & cr4_pinned_mask) ^ cr4_pinned_bits;
452 			val = (val & ~cr4_pinned_mask) | cr4_pinned_bits;
453 			goto set_register;
454 		}
455 		/* Warn after we've corrected the changed bits. */
456 		WARN_ONCE(bits_changed, "pinned CR4 bits changed: 0x%lx!?\n",
457 			  bits_changed);
458 	}
459 }
460 #if IS_MODULE(CONFIG_LKDTM)
461 EXPORT_SYMBOL_GPL(native_write_cr4);
462 #endif
463 
464 void cr4_update_irqsoff(unsigned long set, unsigned long clear)
465 {
466 	unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
467 
468 	lockdep_assert_irqs_disabled();
469 
470 	newval = (cr4 & ~clear) | set;
471 	if (newval != cr4) {
472 		this_cpu_write(cpu_tlbstate.cr4, newval);
473 		__write_cr4(newval);
474 	}
475 }
476 EXPORT_SYMBOL(cr4_update_irqsoff);
477 
478 /* Read the CR4 shadow. */
479 unsigned long cr4_read_shadow(void)
480 {
481 	return this_cpu_read(cpu_tlbstate.cr4);
482 }
483 EXPORT_SYMBOL_GPL(cr4_read_shadow);
484 
485 void cr4_init(void)
486 {
487 	unsigned long cr4 = __read_cr4();
488 
489 	if (boot_cpu_has(X86_FEATURE_PCID))
490 		cr4 |= X86_CR4_PCIDE;
491 	if (static_branch_likely(&cr_pinning))
492 		cr4 = (cr4 & ~cr4_pinned_mask) | cr4_pinned_bits;
493 
494 	__write_cr4(cr4);
495 
496 	/* Initialize cr4 shadow for this CPU. */
497 	this_cpu_write(cpu_tlbstate.cr4, cr4);
498 }
499 
500 /*
501  * Once CPU feature detection is finished (and boot params have been
502  * parsed), record any of the sensitive CR bits that are set, and
503  * enable CR pinning.
504  */
505 static void __init setup_cr_pinning(void)
506 {
507 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & cr4_pinned_mask;
508 	static_key_enable(&cr_pinning.key);
509 }
510 
511 static __init int x86_nofsgsbase_setup(char *arg)
512 {
513 	/* Require an exact match without trailing characters. */
514 	if (strlen(arg))
515 		return 0;
516 
517 	/* Do not emit a message if the feature is not present. */
518 	if (!boot_cpu_has(X86_FEATURE_FSGSBASE))
519 		return 1;
520 
521 	setup_clear_cpu_cap(X86_FEATURE_FSGSBASE);
522 	pr_info("FSGSBASE disabled via kernel command line\n");
523 	return 1;
524 }
525 __setup("nofsgsbase", x86_nofsgsbase_setup);
526 
527 /*
528  * Protection Keys are not available in 32-bit mode.
529  */
530 static bool pku_disabled;
531 
532 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
533 {
534 	if (c == &boot_cpu_data) {
535 		if (pku_disabled || !cpu_feature_enabled(X86_FEATURE_PKU))
536 			return;
537 		/*
538 		 * Setting CR4.PKE will cause the X86_FEATURE_OSPKE cpuid
539 		 * bit to be set.  Enforce it.
540 		 */
541 		setup_force_cpu_cap(X86_FEATURE_OSPKE);
542 
543 	} else if (!cpu_feature_enabled(X86_FEATURE_OSPKE)) {
544 		return;
545 	}
546 
547 	cr4_set_bits(X86_CR4_PKE);
548 	/* Load the default PKRU value */
549 	pkru_write_default();
550 }
551 
552 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
553 static __init int setup_disable_pku(char *arg)
554 {
555 	/*
556 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
557 	 * runtime checks are against OSPKE so clearing the
558 	 * bit does nothing.
559 	 *
560 	 * This way, we will see "pku" in cpuinfo, but not
561 	 * "ospke", which is exactly what we want.  It shows
562 	 * that the CPU has PKU, but the OS has not enabled it.
563 	 * This happens to be exactly how a system would look
564 	 * if we disabled the config option.
565 	 */
566 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
567 	pku_disabled = true;
568 	return 1;
569 }
570 __setup("nopku", setup_disable_pku);
571 #endif
572 
573 #ifdef CONFIG_X86_KERNEL_IBT
574 
575 __noendbr u64 ibt_save(bool disable)
576 {
577 	u64 msr = 0;
578 
579 	if (cpu_feature_enabled(X86_FEATURE_IBT)) {
580 		rdmsrl(MSR_IA32_S_CET, msr);
581 		if (disable)
582 			wrmsrl(MSR_IA32_S_CET, msr & ~CET_ENDBR_EN);
583 	}
584 
585 	return msr;
586 }
587 
588 __noendbr void ibt_restore(u64 save)
589 {
590 	u64 msr;
591 
592 	if (cpu_feature_enabled(X86_FEATURE_IBT)) {
593 		rdmsrl(MSR_IA32_S_CET, msr);
594 		msr &= ~CET_ENDBR_EN;
595 		msr |= (save & CET_ENDBR_EN);
596 		wrmsrl(MSR_IA32_S_CET, msr);
597 	}
598 }
599 
600 #endif
601 
602 static __always_inline void setup_cet(struct cpuinfo_x86 *c)
603 {
604 	u64 msr = CET_ENDBR_EN;
605 
606 	if (!HAS_KERNEL_IBT ||
607 	    !cpu_feature_enabled(X86_FEATURE_IBT))
608 		return;
609 
610 	wrmsrl(MSR_IA32_S_CET, msr);
611 	cr4_set_bits(X86_CR4_CET);
612 
613 	if (!ibt_selftest()) {
614 		pr_err("IBT selftest: Failed!\n");
615 		wrmsrl(MSR_IA32_S_CET, 0);
616 		setup_clear_cpu_cap(X86_FEATURE_IBT);
617 		return;
618 	}
619 }
620 
621 __noendbr void cet_disable(void)
622 {
623 	if (cpu_feature_enabled(X86_FEATURE_IBT))
624 		wrmsrl(MSR_IA32_S_CET, 0);
625 }
626 
627 /*
628  * Some CPU features depend on higher CPUID levels, which may not always
629  * be available due to CPUID level capping or broken virtualization
630  * software.  Add those features to this table to auto-disable them.
631  */
632 struct cpuid_dependent_feature {
633 	u32 feature;
634 	u32 level;
635 };
636 
637 static const struct cpuid_dependent_feature
638 cpuid_dependent_features[] = {
639 	{ X86_FEATURE_MWAIT,		0x00000005 },
640 	{ X86_FEATURE_DCA,		0x00000009 },
641 	{ X86_FEATURE_XSAVE,		0x0000000d },
642 	{ 0, 0 }
643 };
644 
645 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
646 {
647 	const struct cpuid_dependent_feature *df;
648 
649 	for (df = cpuid_dependent_features; df->feature; df++) {
650 
651 		if (!cpu_has(c, df->feature))
652 			continue;
653 		/*
654 		 * Note: cpuid_level is set to -1 if unavailable, but
655 		 * extended_extended_level is set to 0 if unavailable
656 		 * and the legitimate extended levels are all negative
657 		 * when signed; hence the weird messing around with
658 		 * signs here...
659 		 */
660 		if (!((s32)df->level < 0 ?
661 		     (u32)df->level > (u32)c->extended_cpuid_level :
662 		     (s32)df->level > (s32)c->cpuid_level))
663 			continue;
664 
665 		clear_cpu_cap(c, df->feature);
666 		if (!warn)
667 			continue;
668 
669 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
670 			x86_cap_flag(df->feature), df->level);
671 	}
672 }
673 
674 /*
675  * Naming convention should be: <Name> [(<Codename>)]
676  * This table only is used unless init_<vendor>() below doesn't set it;
677  * in particular, if CPUID levels 0x80000002..4 are supported, this
678  * isn't used
679  */
680 
681 /* Look up CPU names by table lookup. */
682 static const char *table_lookup_model(struct cpuinfo_x86 *c)
683 {
684 #ifdef CONFIG_X86_32
685 	const struct legacy_cpu_model_info *info;
686 
687 	if (c->x86_model >= 16)
688 		return NULL;	/* Range check */
689 
690 	if (!this_cpu)
691 		return NULL;
692 
693 	info = this_cpu->legacy_models;
694 
695 	while (info->family) {
696 		if (info->family == c->x86)
697 			return info->model_names[c->x86_model];
698 		info++;
699 	}
700 #endif
701 	return NULL;		/* Not found */
702 }
703 
704 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
705 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
706 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
707 
708 #ifdef CONFIG_X86_32
709 /* The 32-bit entry code needs to find cpu_entry_area. */
710 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
711 #endif
712 
713 /* Load the original GDT from the per-cpu structure */
714 void load_direct_gdt(int cpu)
715 {
716 	struct desc_ptr gdt_descr;
717 
718 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
719 	gdt_descr.size = GDT_SIZE - 1;
720 	load_gdt(&gdt_descr);
721 }
722 EXPORT_SYMBOL_GPL(load_direct_gdt);
723 
724 /* Load a fixmap remapping of the per-cpu GDT */
725 void load_fixmap_gdt(int cpu)
726 {
727 	struct desc_ptr gdt_descr;
728 
729 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
730 	gdt_descr.size = GDT_SIZE - 1;
731 	load_gdt(&gdt_descr);
732 }
733 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
734 
735 /**
736  * switch_gdt_and_percpu_base - Switch to direct GDT and runtime per CPU base
737  * @cpu:	The CPU number for which this is invoked
738  *
739  * Invoked during early boot to switch from early GDT and early per CPU to
740  * the direct GDT and the runtime per CPU area. On 32-bit the percpu base
741  * switch is implicit by loading the direct GDT. On 64bit this requires
742  * to update GSBASE.
743  */
744 void __init switch_gdt_and_percpu_base(int cpu)
745 {
746 	load_direct_gdt(cpu);
747 
748 #ifdef CONFIG_X86_64
749 	/*
750 	 * No need to load %gs. It is already correct.
751 	 *
752 	 * Writing %gs on 64bit would zero GSBASE which would make any per
753 	 * CPU operation up to the point of the wrmsrl() fault.
754 	 *
755 	 * Set GSBASE to the new offset. Until the wrmsrl() happens the
756 	 * early mapping is still valid. That means the GSBASE update will
757 	 * lose any prior per CPU data which was not copied over in
758 	 * setup_per_cpu_areas().
759 	 *
760 	 * This works even with stackprotector enabled because the
761 	 * per CPU stack canary is 0 in both per CPU areas.
762 	 */
763 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
764 #else
765 	/*
766 	 * %fs is already set to __KERNEL_PERCPU, but after switching GDT
767 	 * it is required to load FS again so that the 'hidden' part is
768 	 * updated from the new GDT. Up to this point the early per CPU
769 	 * translation is active. Any content of the early per CPU data
770 	 * which was not copied over in setup_per_cpu_areas() is lost.
771 	 */
772 	loadsegment(fs, __KERNEL_PERCPU);
773 #endif
774 }
775 
776 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
777 
778 static void get_model_name(struct cpuinfo_x86 *c)
779 {
780 	unsigned int *v;
781 	char *p, *q, *s;
782 
783 	if (c->extended_cpuid_level < 0x80000004)
784 		return;
785 
786 	v = (unsigned int *)c->x86_model_id;
787 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
788 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
789 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
790 	c->x86_model_id[48] = 0;
791 
792 	/* Trim whitespace */
793 	p = q = s = &c->x86_model_id[0];
794 
795 	while (*p == ' ')
796 		p++;
797 
798 	while (*p) {
799 		/* Note the last non-whitespace index */
800 		if (!isspace(*p))
801 			s = q;
802 
803 		*q++ = *p++;
804 	}
805 
806 	*(s + 1) = '\0';
807 }
808 
809 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
810 {
811 	unsigned int eax, ebx, ecx, edx;
812 
813 	c->x86_max_cores = 1;
814 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
815 		return;
816 
817 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
818 	if (eax & 0x1f)
819 		c->x86_max_cores = (eax >> 26) + 1;
820 }
821 
822 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
823 {
824 	unsigned int n, dummy, ebx, ecx, edx, l2size;
825 
826 	n = c->extended_cpuid_level;
827 
828 	if (n >= 0x80000005) {
829 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
830 		c->x86_cache_size = (ecx>>24) + (edx>>24);
831 #ifdef CONFIG_X86_64
832 		/* On K8 L1 TLB is inclusive, so don't count it */
833 		c->x86_tlbsize = 0;
834 #endif
835 	}
836 
837 	if (n < 0x80000006)	/* Some chips just has a large L1. */
838 		return;
839 
840 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
841 	l2size = ecx >> 16;
842 
843 #ifdef CONFIG_X86_64
844 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
845 #else
846 	/* do processor-specific cache resizing */
847 	if (this_cpu->legacy_cache_size)
848 		l2size = this_cpu->legacy_cache_size(c, l2size);
849 
850 	/* Allow user to override all this if necessary. */
851 	if (cachesize_override != -1)
852 		l2size = cachesize_override;
853 
854 	if (l2size == 0)
855 		return;		/* Again, no L2 cache is possible */
856 #endif
857 
858 	c->x86_cache_size = l2size;
859 }
860 
861 u16 __read_mostly tlb_lli_4k[NR_INFO];
862 u16 __read_mostly tlb_lli_2m[NR_INFO];
863 u16 __read_mostly tlb_lli_4m[NR_INFO];
864 u16 __read_mostly tlb_lld_4k[NR_INFO];
865 u16 __read_mostly tlb_lld_2m[NR_INFO];
866 u16 __read_mostly tlb_lld_4m[NR_INFO];
867 u16 __read_mostly tlb_lld_1g[NR_INFO];
868 
869 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
870 {
871 	if (this_cpu->c_detect_tlb)
872 		this_cpu->c_detect_tlb(c);
873 
874 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
875 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
876 		tlb_lli_4m[ENTRIES]);
877 
878 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
879 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
880 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
881 }
882 
883 int detect_ht_early(struct cpuinfo_x86 *c)
884 {
885 #ifdef CONFIG_SMP
886 	u32 eax, ebx, ecx, edx;
887 
888 	if (!cpu_has(c, X86_FEATURE_HT))
889 		return -1;
890 
891 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
892 		return -1;
893 
894 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
895 		return -1;
896 
897 	cpuid(1, &eax, &ebx, &ecx, &edx);
898 
899 	smp_num_siblings = (ebx & 0xff0000) >> 16;
900 	if (smp_num_siblings == 1)
901 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
902 #endif
903 	return 0;
904 }
905 
906 void detect_ht(struct cpuinfo_x86 *c)
907 {
908 #ifdef CONFIG_SMP
909 	int index_msb, core_bits;
910 
911 	if (detect_ht_early(c) < 0)
912 		return;
913 
914 	index_msb = get_count_order(smp_num_siblings);
915 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
916 
917 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
918 
919 	index_msb = get_count_order(smp_num_siblings);
920 
921 	core_bits = get_count_order(c->x86_max_cores);
922 
923 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
924 				       ((1 << core_bits) - 1);
925 #endif
926 }
927 
928 static void get_cpu_vendor(struct cpuinfo_x86 *c)
929 {
930 	char *v = c->x86_vendor_id;
931 	int i;
932 
933 	for (i = 0; i < X86_VENDOR_NUM; i++) {
934 		if (!cpu_devs[i])
935 			break;
936 
937 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
938 		    (cpu_devs[i]->c_ident[1] &&
939 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
940 
941 			this_cpu = cpu_devs[i];
942 			c->x86_vendor = this_cpu->c_x86_vendor;
943 			return;
944 		}
945 	}
946 
947 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
948 		    "CPU: Your system may be unstable.\n", v);
949 
950 	c->x86_vendor = X86_VENDOR_UNKNOWN;
951 	this_cpu = &default_cpu;
952 }
953 
954 void cpu_detect(struct cpuinfo_x86 *c)
955 {
956 	/* Get vendor name */
957 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
958 	      (unsigned int *)&c->x86_vendor_id[0],
959 	      (unsigned int *)&c->x86_vendor_id[8],
960 	      (unsigned int *)&c->x86_vendor_id[4]);
961 
962 	c->x86 = 4;
963 	/* Intel-defined flags: level 0x00000001 */
964 	if (c->cpuid_level >= 0x00000001) {
965 		u32 junk, tfms, cap0, misc;
966 
967 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
968 		c->x86		= x86_family(tfms);
969 		c->x86_model	= x86_model(tfms);
970 		c->x86_stepping	= x86_stepping(tfms);
971 
972 		if (cap0 & (1<<19)) {
973 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
974 			c->x86_cache_alignment = c->x86_clflush_size;
975 		}
976 	}
977 }
978 
979 static void apply_forced_caps(struct cpuinfo_x86 *c)
980 {
981 	int i;
982 
983 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
984 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
985 		c->x86_capability[i] |= cpu_caps_set[i];
986 	}
987 }
988 
989 static void init_speculation_control(struct cpuinfo_x86 *c)
990 {
991 	/*
992 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
993 	 * and they also have a different bit for STIBP support. Also,
994 	 * a hypervisor might have set the individual AMD bits even on
995 	 * Intel CPUs, for finer-grained selection of what's available.
996 	 */
997 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
998 		set_cpu_cap(c, X86_FEATURE_IBRS);
999 		set_cpu_cap(c, X86_FEATURE_IBPB);
1000 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1001 	}
1002 
1003 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
1004 		set_cpu_cap(c, X86_FEATURE_STIBP);
1005 
1006 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
1007 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
1008 		set_cpu_cap(c, X86_FEATURE_SSBD);
1009 
1010 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
1011 		set_cpu_cap(c, X86_FEATURE_IBRS);
1012 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1013 	}
1014 
1015 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
1016 		set_cpu_cap(c, X86_FEATURE_IBPB);
1017 
1018 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
1019 		set_cpu_cap(c, X86_FEATURE_STIBP);
1020 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1021 	}
1022 
1023 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
1024 		set_cpu_cap(c, X86_FEATURE_SSBD);
1025 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1026 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
1027 	}
1028 }
1029 
1030 void get_cpu_cap(struct cpuinfo_x86 *c)
1031 {
1032 	u32 eax, ebx, ecx, edx;
1033 
1034 	/* Intel-defined flags: level 0x00000001 */
1035 	if (c->cpuid_level >= 0x00000001) {
1036 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
1037 
1038 		c->x86_capability[CPUID_1_ECX] = ecx;
1039 		c->x86_capability[CPUID_1_EDX] = edx;
1040 	}
1041 
1042 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
1043 	if (c->cpuid_level >= 0x00000006)
1044 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
1045 
1046 	/* Additional Intel-defined flags: level 0x00000007 */
1047 	if (c->cpuid_level >= 0x00000007) {
1048 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
1049 		c->x86_capability[CPUID_7_0_EBX] = ebx;
1050 		c->x86_capability[CPUID_7_ECX] = ecx;
1051 		c->x86_capability[CPUID_7_EDX] = edx;
1052 
1053 		/* Check valid sub-leaf index before accessing it */
1054 		if (eax >= 1) {
1055 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
1056 			c->x86_capability[CPUID_7_1_EAX] = eax;
1057 		}
1058 	}
1059 
1060 	/* Extended state features: level 0x0000000d */
1061 	if (c->cpuid_level >= 0x0000000d) {
1062 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
1063 
1064 		c->x86_capability[CPUID_D_1_EAX] = eax;
1065 	}
1066 
1067 	/* AMD-defined flags: level 0x80000001 */
1068 	eax = cpuid_eax(0x80000000);
1069 	c->extended_cpuid_level = eax;
1070 
1071 	if ((eax & 0xffff0000) == 0x80000000) {
1072 		if (eax >= 0x80000001) {
1073 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
1074 
1075 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
1076 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
1077 		}
1078 	}
1079 
1080 	if (c->extended_cpuid_level >= 0x80000007) {
1081 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
1082 
1083 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
1084 		c->x86_power = edx;
1085 	}
1086 
1087 	if (c->extended_cpuid_level >= 0x80000008) {
1088 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
1089 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
1090 	}
1091 
1092 	if (c->extended_cpuid_level >= 0x8000000a)
1093 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
1094 
1095 	if (c->extended_cpuid_level >= 0x8000001f)
1096 		c->x86_capability[CPUID_8000_001F_EAX] = cpuid_eax(0x8000001f);
1097 
1098 	if (c->extended_cpuid_level >= 0x80000021)
1099 		c->x86_capability[CPUID_8000_0021_EAX] = cpuid_eax(0x80000021);
1100 
1101 	init_scattered_cpuid_features(c);
1102 	init_speculation_control(c);
1103 
1104 	/*
1105 	 * Clear/Set all flags overridden by options, after probe.
1106 	 * This needs to happen each time we re-probe, which may happen
1107 	 * several times during CPU initialization.
1108 	 */
1109 	apply_forced_caps(c);
1110 }
1111 
1112 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
1113 {
1114 	u32 eax, ebx, ecx, edx;
1115 
1116 	if (c->extended_cpuid_level >= 0x80000008) {
1117 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
1118 
1119 		c->x86_virt_bits = (eax >> 8) & 0xff;
1120 		c->x86_phys_bits = eax & 0xff;
1121 	}
1122 #ifdef CONFIG_X86_32
1123 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
1124 		c->x86_phys_bits = 36;
1125 #endif
1126 	c->x86_cache_bits = c->x86_phys_bits;
1127 }
1128 
1129 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
1130 {
1131 #ifdef CONFIG_X86_32
1132 	int i;
1133 
1134 	/*
1135 	 * First of all, decide if this is a 486 or higher
1136 	 * It's a 486 if we can modify the AC flag
1137 	 */
1138 	if (flag_is_changeable_p(X86_EFLAGS_AC))
1139 		c->x86 = 4;
1140 	else
1141 		c->x86 = 3;
1142 
1143 	for (i = 0; i < X86_VENDOR_NUM; i++)
1144 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
1145 			c->x86_vendor_id[0] = 0;
1146 			cpu_devs[i]->c_identify(c);
1147 			if (c->x86_vendor_id[0]) {
1148 				get_cpu_vendor(c);
1149 				break;
1150 			}
1151 		}
1152 #endif
1153 }
1154 
1155 #define NO_SPECULATION		BIT(0)
1156 #define NO_MELTDOWN		BIT(1)
1157 #define NO_SSB			BIT(2)
1158 #define NO_L1TF			BIT(3)
1159 #define NO_MDS			BIT(4)
1160 #define MSBDS_ONLY		BIT(5)
1161 #define NO_SWAPGS		BIT(6)
1162 #define NO_ITLB_MULTIHIT	BIT(7)
1163 #define NO_SPECTRE_V2		BIT(8)
1164 #define NO_MMIO			BIT(9)
1165 #define NO_EIBRS_PBRSB		BIT(10)
1166 
1167 #define VULNWL(vendor, family, model, whitelist)	\
1168 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist)
1169 
1170 #define VULNWL_INTEL(model, whitelist)		\
1171 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1172 
1173 #define VULNWL_AMD(family, whitelist)		\
1174 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1175 
1176 #define VULNWL_HYGON(family, whitelist)		\
1177 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1178 
1179 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1180 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1181 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1182 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1183 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1184 	VULNWL(VORTEX,	5, X86_MODEL_ANY,	NO_SPECULATION),
1185 	VULNWL(VORTEX,	6, X86_MODEL_ANY,	NO_SPECULATION),
1186 
1187 	/* Intel Family 6 */
1188 	VULNWL_INTEL(TIGERLAKE,			NO_MMIO),
1189 	VULNWL_INTEL(TIGERLAKE_L,		NO_MMIO),
1190 	VULNWL_INTEL(ALDERLAKE,			NO_MMIO),
1191 	VULNWL_INTEL(ALDERLAKE_L,		NO_MMIO),
1192 
1193 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1194 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION | NO_ITLB_MULTIHIT),
1195 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1196 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1197 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1198 
1199 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1200 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1201 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1202 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1203 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1204 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1205 
1206 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1207 
1208 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1209 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1210 
1211 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1212 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1213 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB),
1214 
1215 	/*
1216 	 * Technically, swapgs isn't serializing on AMD (despite it previously
1217 	 * being documented as such in the APM).  But according to AMD, %gs is
1218 	 * updated non-speculatively, and the issuing of %gs-relative memory
1219 	 * operands will be blocked until the %gs update completes, which is
1220 	 * good enough for our purposes.
1221 	 */
1222 
1223 	VULNWL_INTEL(ATOM_TREMONT,		NO_EIBRS_PBRSB),
1224 	VULNWL_INTEL(ATOM_TREMONT_L,		NO_EIBRS_PBRSB),
1225 	VULNWL_INTEL(ATOM_TREMONT_D,		NO_ITLB_MULTIHIT | NO_EIBRS_PBRSB),
1226 
1227 	/* AMD Family 0xf - 0x12 */
1228 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1229 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1230 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1231 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1232 
1233 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1234 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB),
1235 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB),
1236 
1237 	/* Zhaoxin Family 7 */
1238 	VULNWL(CENTAUR,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO),
1239 	VULNWL(ZHAOXIN,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO),
1240 	{}
1241 };
1242 
1243 #define VULNBL(vendor, family, model, blacklist)	\
1244 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, blacklist)
1245 
1246 #define VULNBL_INTEL_STEPPINGS(model, steppings, issues)		   \
1247 	X86_MATCH_VENDOR_FAM_MODEL_STEPPINGS_FEATURE(INTEL, 6,		   \
1248 					    INTEL_FAM6_##model, steppings, \
1249 					    X86_FEATURE_ANY, issues)
1250 
1251 #define VULNBL_AMD(family, blacklist)		\
1252 	VULNBL(AMD, family, X86_MODEL_ANY, blacklist)
1253 
1254 #define VULNBL_HYGON(family, blacklist)		\
1255 	VULNBL(HYGON, family, X86_MODEL_ANY, blacklist)
1256 
1257 #define SRBDS		BIT(0)
1258 /* CPU is affected by X86_BUG_MMIO_STALE_DATA */
1259 #define MMIO		BIT(1)
1260 /* CPU is affected by Shared Buffers Data Sampling (SBDS), a variant of X86_BUG_MMIO_STALE_DATA */
1261 #define MMIO_SBDS	BIT(2)
1262 /* CPU is affected by RETbleed, speculating where you would not expect it */
1263 #define RETBLEED	BIT(3)
1264 /* CPU is affected by SMT (cross-thread) return predictions */
1265 #define SMT_RSB		BIT(4)
1266 
1267 static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = {
1268 	VULNBL_INTEL_STEPPINGS(IVYBRIDGE,	X86_STEPPING_ANY,		SRBDS),
1269 	VULNBL_INTEL_STEPPINGS(HASWELL,		X86_STEPPING_ANY,		SRBDS),
1270 	VULNBL_INTEL_STEPPINGS(HASWELL_L,	X86_STEPPING_ANY,		SRBDS),
1271 	VULNBL_INTEL_STEPPINGS(HASWELL_G,	X86_STEPPING_ANY,		SRBDS),
1272 	VULNBL_INTEL_STEPPINGS(HASWELL_X,	X86_STEPPING_ANY,		MMIO),
1273 	VULNBL_INTEL_STEPPINGS(BROADWELL_D,	X86_STEPPING_ANY,		MMIO),
1274 	VULNBL_INTEL_STEPPINGS(BROADWELL_G,	X86_STEPPING_ANY,		SRBDS),
1275 	VULNBL_INTEL_STEPPINGS(BROADWELL_X,	X86_STEPPING_ANY,		MMIO),
1276 	VULNBL_INTEL_STEPPINGS(BROADWELL,	X86_STEPPING_ANY,		SRBDS),
1277 	VULNBL_INTEL_STEPPINGS(SKYLAKE_L,	X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1278 	VULNBL_INTEL_STEPPINGS(SKYLAKE_X,	X86_STEPPING_ANY,		MMIO | RETBLEED),
1279 	VULNBL_INTEL_STEPPINGS(SKYLAKE,		X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1280 	VULNBL_INTEL_STEPPINGS(KABYLAKE_L,	X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1281 	VULNBL_INTEL_STEPPINGS(KABYLAKE,	X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1282 	VULNBL_INTEL_STEPPINGS(CANNONLAKE_L,	X86_STEPPING_ANY,		RETBLEED),
1283 	VULNBL_INTEL_STEPPINGS(ICELAKE_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1284 	VULNBL_INTEL_STEPPINGS(ICELAKE_D,	X86_STEPPING_ANY,		MMIO),
1285 	VULNBL_INTEL_STEPPINGS(ICELAKE_X,	X86_STEPPING_ANY,		MMIO),
1286 	VULNBL_INTEL_STEPPINGS(COMETLAKE,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1287 	VULNBL_INTEL_STEPPINGS(COMETLAKE_L,	X86_STEPPINGS(0x0, 0x0),	MMIO | RETBLEED),
1288 	VULNBL_INTEL_STEPPINGS(COMETLAKE_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1289 	VULNBL_INTEL_STEPPINGS(LAKEFIELD,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1290 	VULNBL_INTEL_STEPPINGS(ROCKETLAKE,	X86_STEPPING_ANY,		MMIO | RETBLEED),
1291 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS),
1292 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_D,	X86_STEPPING_ANY,		MMIO),
1293 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS),
1294 
1295 	VULNBL_AMD(0x15, RETBLEED),
1296 	VULNBL_AMD(0x16, RETBLEED),
1297 	VULNBL_AMD(0x17, RETBLEED | SMT_RSB),
1298 	VULNBL_HYGON(0x18, RETBLEED | SMT_RSB),
1299 	{}
1300 };
1301 
1302 static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which)
1303 {
1304 	const struct x86_cpu_id *m = x86_match_cpu(table);
1305 
1306 	return m && !!(m->driver_data & which);
1307 }
1308 
1309 u64 x86_read_arch_cap_msr(void)
1310 {
1311 	u64 ia32_cap = 0;
1312 
1313 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1314 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1315 
1316 	return ia32_cap;
1317 }
1318 
1319 static bool arch_cap_mmio_immune(u64 ia32_cap)
1320 {
1321 	return (ia32_cap & ARCH_CAP_FBSDP_NO &&
1322 		ia32_cap & ARCH_CAP_PSDP_NO &&
1323 		ia32_cap & ARCH_CAP_SBDR_SSDP_NO);
1324 }
1325 
1326 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1327 {
1328 	u64 ia32_cap = x86_read_arch_cap_msr();
1329 
1330 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1331 	if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) &&
1332 	    !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO))
1333 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1334 
1335 	if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION))
1336 		return;
1337 
1338 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1339 
1340 	if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2))
1341 		setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1342 
1343 	if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) &&
1344 	    !(ia32_cap & ARCH_CAP_SSB_NO) &&
1345 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1346 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1347 
1348 	/*
1349 	 * AMD's AutoIBRS is equivalent to Intel's eIBRS - use the Intel feature
1350 	 * flag and protect from vendor-specific bugs via the whitelist.
1351 	 */
1352 	if ((ia32_cap & ARCH_CAP_IBRS_ALL) || cpu_has(c, X86_FEATURE_AUTOIBRS)) {
1353 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1354 		if (!cpu_matches(cpu_vuln_whitelist, NO_EIBRS_PBRSB) &&
1355 		    !(ia32_cap & ARCH_CAP_PBRSB_NO))
1356 			setup_force_cpu_bug(X86_BUG_EIBRS_PBRSB);
1357 	}
1358 
1359 	if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) &&
1360 	    !(ia32_cap & ARCH_CAP_MDS_NO)) {
1361 		setup_force_cpu_bug(X86_BUG_MDS);
1362 		if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY))
1363 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1364 	}
1365 
1366 	if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS))
1367 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1368 
1369 	/*
1370 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1371 	 *	- TSX is supported or
1372 	 *	- TSX_CTRL is present
1373 	 *
1374 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1375 	 * the kernel boot e.g. kexec.
1376 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1377 	 * update is not present or running as guest that don't get TSX_CTRL.
1378 	 */
1379 	if (!(ia32_cap & ARCH_CAP_TAA_NO) &&
1380 	    (cpu_has(c, X86_FEATURE_RTM) ||
1381 	     (ia32_cap & ARCH_CAP_TSX_CTRL_MSR)))
1382 		setup_force_cpu_bug(X86_BUG_TAA);
1383 
1384 	/*
1385 	 * SRBDS affects CPUs which support RDRAND or RDSEED and are listed
1386 	 * in the vulnerability blacklist.
1387 	 *
1388 	 * Some of the implications and mitigation of Shared Buffers Data
1389 	 * Sampling (SBDS) are similar to SRBDS. Give SBDS same treatment as
1390 	 * SRBDS.
1391 	 */
1392 	if ((cpu_has(c, X86_FEATURE_RDRAND) ||
1393 	     cpu_has(c, X86_FEATURE_RDSEED)) &&
1394 	    cpu_matches(cpu_vuln_blacklist, SRBDS | MMIO_SBDS))
1395 		    setup_force_cpu_bug(X86_BUG_SRBDS);
1396 
1397 	/*
1398 	 * Processor MMIO Stale Data bug enumeration
1399 	 *
1400 	 * Affected CPU list is generally enough to enumerate the vulnerability,
1401 	 * but for virtualization case check for ARCH_CAP MSR bits also, VMM may
1402 	 * not want the guest to enumerate the bug.
1403 	 *
1404 	 * Set X86_BUG_MMIO_UNKNOWN for CPUs that are neither in the blacklist,
1405 	 * nor in the whitelist and also don't enumerate MSR ARCH_CAP MMIO bits.
1406 	 */
1407 	if (!arch_cap_mmio_immune(ia32_cap)) {
1408 		if (cpu_matches(cpu_vuln_blacklist, MMIO))
1409 			setup_force_cpu_bug(X86_BUG_MMIO_STALE_DATA);
1410 		else if (!cpu_matches(cpu_vuln_whitelist, NO_MMIO))
1411 			setup_force_cpu_bug(X86_BUG_MMIO_UNKNOWN);
1412 	}
1413 
1414 	if (!cpu_has(c, X86_FEATURE_BTC_NO)) {
1415 		if (cpu_matches(cpu_vuln_blacklist, RETBLEED) || (ia32_cap & ARCH_CAP_RSBA))
1416 			setup_force_cpu_bug(X86_BUG_RETBLEED);
1417 	}
1418 
1419 	if (cpu_matches(cpu_vuln_blacklist, SMT_RSB))
1420 		setup_force_cpu_bug(X86_BUG_SMT_RSB);
1421 
1422 	if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN))
1423 		return;
1424 
1425 	/* Rogue Data Cache Load? No! */
1426 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1427 		return;
1428 
1429 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1430 
1431 	if (cpu_matches(cpu_vuln_whitelist, NO_L1TF))
1432 		return;
1433 
1434 	setup_force_cpu_bug(X86_BUG_L1TF);
1435 }
1436 
1437 /*
1438  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1439  * unfortunately, that's not true in practice because of early VIA
1440  * chips and (more importantly) broken virtualizers that are not easy
1441  * to detect. In the latter case it doesn't even *fail* reliably, so
1442  * probing for it doesn't even work. Disable it completely on 32-bit
1443  * unless we can find a reliable way to detect all the broken cases.
1444  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1445  */
1446 static void detect_nopl(void)
1447 {
1448 #ifdef CONFIG_X86_32
1449 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1450 #else
1451 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1452 #endif
1453 }
1454 
1455 /*
1456  * We parse cpu parameters early because fpu__init_system() is executed
1457  * before parse_early_param().
1458  */
1459 static void __init cpu_parse_early_param(void)
1460 {
1461 	char arg[128];
1462 	char *argptr = arg, *opt;
1463 	int arglen, taint = 0;
1464 
1465 #ifdef CONFIG_X86_32
1466 	if (cmdline_find_option_bool(boot_command_line, "no387"))
1467 #ifdef CONFIG_MATH_EMULATION
1468 		setup_clear_cpu_cap(X86_FEATURE_FPU);
1469 #else
1470 		pr_err("Option 'no387' required CONFIG_MATH_EMULATION enabled.\n");
1471 #endif
1472 
1473 	if (cmdline_find_option_bool(boot_command_line, "nofxsr"))
1474 		setup_clear_cpu_cap(X86_FEATURE_FXSR);
1475 #endif
1476 
1477 	if (cmdline_find_option_bool(boot_command_line, "noxsave"))
1478 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
1479 
1480 	if (cmdline_find_option_bool(boot_command_line, "noxsaveopt"))
1481 		setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
1482 
1483 	if (cmdline_find_option_bool(boot_command_line, "noxsaves"))
1484 		setup_clear_cpu_cap(X86_FEATURE_XSAVES);
1485 
1486 	arglen = cmdline_find_option(boot_command_line, "clearcpuid", arg, sizeof(arg));
1487 	if (arglen <= 0)
1488 		return;
1489 
1490 	pr_info("Clearing CPUID bits:");
1491 
1492 	while (argptr) {
1493 		bool found __maybe_unused = false;
1494 		unsigned int bit;
1495 
1496 		opt = strsep(&argptr, ",");
1497 
1498 		/*
1499 		 * Handle naked numbers first for feature flags which don't
1500 		 * have names.
1501 		 */
1502 		if (!kstrtouint(opt, 10, &bit)) {
1503 			if (bit < NCAPINTS * 32) {
1504 
1505 #ifdef CONFIG_X86_FEATURE_NAMES
1506 				/* empty-string, i.e., ""-defined feature flags */
1507 				if (!x86_cap_flags[bit])
1508 					pr_cont(" " X86_CAP_FMT_NUM, x86_cap_flag_num(bit));
1509 				else
1510 #endif
1511 					pr_cont(" " X86_CAP_FMT, x86_cap_flag(bit));
1512 
1513 				setup_clear_cpu_cap(bit);
1514 				taint++;
1515 			}
1516 			/*
1517 			 * The assumption is that there are no feature names with only
1518 			 * numbers in the name thus go to the next argument.
1519 			 */
1520 			continue;
1521 		}
1522 
1523 #ifdef CONFIG_X86_FEATURE_NAMES
1524 		for (bit = 0; bit < 32 * NCAPINTS; bit++) {
1525 			if (!x86_cap_flag(bit))
1526 				continue;
1527 
1528 			if (strcmp(x86_cap_flag(bit), opt))
1529 				continue;
1530 
1531 			pr_cont(" %s", opt);
1532 			setup_clear_cpu_cap(bit);
1533 			taint++;
1534 			found = true;
1535 			break;
1536 		}
1537 
1538 		if (!found)
1539 			pr_cont(" (unknown: %s)", opt);
1540 #endif
1541 	}
1542 	pr_cont("\n");
1543 
1544 	if (taint)
1545 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_STILL_OK);
1546 }
1547 
1548 /*
1549  * Do minimum CPU detection early.
1550  * Fields really needed: vendor, cpuid_level, family, model, mask,
1551  * cache alignment.
1552  * The others are not touched to avoid unwanted side effects.
1553  *
1554  * WARNING: this function is only called on the boot CPU.  Don't add code
1555  * here that is supposed to run on all CPUs.
1556  */
1557 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1558 {
1559 #ifdef CONFIG_X86_64
1560 	c->x86_clflush_size = 64;
1561 	c->x86_phys_bits = 36;
1562 	c->x86_virt_bits = 48;
1563 #else
1564 	c->x86_clflush_size = 32;
1565 	c->x86_phys_bits = 32;
1566 	c->x86_virt_bits = 32;
1567 #endif
1568 	c->x86_cache_alignment = c->x86_clflush_size;
1569 
1570 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1571 	c->extended_cpuid_level = 0;
1572 
1573 	if (!have_cpuid_p())
1574 		identify_cpu_without_cpuid(c);
1575 
1576 	/* cyrix could have cpuid enabled via c_identify()*/
1577 	if (have_cpuid_p()) {
1578 		cpu_detect(c);
1579 		get_cpu_vendor(c);
1580 		get_cpu_cap(c);
1581 		get_cpu_address_sizes(c);
1582 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1583 		cpu_parse_early_param();
1584 
1585 		if (this_cpu->c_early_init)
1586 			this_cpu->c_early_init(c);
1587 
1588 		c->cpu_index = 0;
1589 		filter_cpuid_features(c, false);
1590 
1591 		if (this_cpu->c_bsp_init)
1592 			this_cpu->c_bsp_init(c);
1593 	} else {
1594 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1595 	}
1596 
1597 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1598 
1599 	cpu_set_bug_bits(c);
1600 
1601 	sld_setup(c);
1602 
1603 	fpu__init_system(c);
1604 
1605 	init_sigframe_size();
1606 
1607 #ifdef CONFIG_X86_32
1608 	/*
1609 	 * Regardless of whether PCID is enumerated, the SDM says
1610 	 * that it can't be enabled in 32-bit mode.
1611 	 */
1612 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1613 #endif
1614 
1615 	/*
1616 	 * Later in the boot process pgtable_l5_enabled() relies on
1617 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1618 	 * enabled by this point we need to clear the feature bit to avoid
1619 	 * false-positives at the later stage.
1620 	 *
1621 	 * pgtable_l5_enabled() can be false here for several reasons:
1622 	 *  - 5-level paging is disabled compile-time;
1623 	 *  - it's 32-bit kernel;
1624 	 *  - machine doesn't support 5-level paging;
1625 	 *  - user specified 'no5lvl' in kernel command line.
1626 	 */
1627 	if (!pgtable_l5_enabled())
1628 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1629 
1630 	detect_nopl();
1631 }
1632 
1633 void __init early_cpu_init(void)
1634 {
1635 	const struct cpu_dev *const *cdev;
1636 	int count = 0;
1637 
1638 #ifdef CONFIG_PROCESSOR_SELECT
1639 	pr_info("KERNEL supported cpus:\n");
1640 #endif
1641 
1642 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1643 		const struct cpu_dev *cpudev = *cdev;
1644 
1645 		if (count >= X86_VENDOR_NUM)
1646 			break;
1647 		cpu_devs[count] = cpudev;
1648 		count++;
1649 
1650 #ifdef CONFIG_PROCESSOR_SELECT
1651 		{
1652 			unsigned int j;
1653 
1654 			for (j = 0; j < 2; j++) {
1655 				if (!cpudev->c_ident[j])
1656 					continue;
1657 				pr_info("  %s %s\n", cpudev->c_vendor,
1658 					cpudev->c_ident[j]);
1659 			}
1660 		}
1661 #endif
1662 	}
1663 	early_identify_cpu(&boot_cpu_data);
1664 }
1665 
1666 static bool detect_null_seg_behavior(void)
1667 {
1668 	/*
1669 	 * Empirically, writing zero to a segment selector on AMD does
1670 	 * not clear the base, whereas writing zero to a segment
1671 	 * selector on Intel does clear the base.  Intel's behavior
1672 	 * allows slightly faster context switches in the common case
1673 	 * where GS is unused by the prev and next threads.
1674 	 *
1675 	 * Since neither vendor documents this anywhere that I can see,
1676 	 * detect it directly instead of hard-coding the choice by
1677 	 * vendor.
1678 	 *
1679 	 * I've designated AMD's behavior as the "bug" because it's
1680 	 * counterintuitive and less friendly.
1681 	 */
1682 
1683 	unsigned long old_base, tmp;
1684 	rdmsrl(MSR_FS_BASE, old_base);
1685 	wrmsrl(MSR_FS_BASE, 1);
1686 	loadsegment(fs, 0);
1687 	rdmsrl(MSR_FS_BASE, tmp);
1688 	wrmsrl(MSR_FS_BASE, old_base);
1689 	return tmp == 0;
1690 }
1691 
1692 void check_null_seg_clears_base(struct cpuinfo_x86 *c)
1693 {
1694 	/* BUG_NULL_SEG is only relevant with 64bit userspace */
1695 	if (!IS_ENABLED(CONFIG_X86_64))
1696 		return;
1697 
1698 	if (cpu_has(c, X86_FEATURE_NULL_SEL_CLR_BASE))
1699 		return;
1700 
1701 	/*
1702 	 * CPUID bit above wasn't set. If this kernel is still running
1703 	 * as a HV guest, then the HV has decided not to advertize
1704 	 * that CPUID bit for whatever reason.	For example, one
1705 	 * member of the migration pool might be vulnerable.  Which
1706 	 * means, the bug is present: set the BUG flag and return.
1707 	 */
1708 	if (cpu_has(c, X86_FEATURE_HYPERVISOR)) {
1709 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1710 		return;
1711 	}
1712 
1713 	/*
1714 	 * Zen2 CPUs also have this behaviour, but no CPUID bit.
1715 	 * 0x18 is the respective family for Hygon.
1716 	 */
1717 	if ((c->x86 == 0x17 || c->x86 == 0x18) &&
1718 	    detect_null_seg_behavior())
1719 		return;
1720 
1721 	/* All the remaining ones are affected */
1722 	set_cpu_bug(c, X86_BUG_NULL_SEG);
1723 }
1724 
1725 static void generic_identify(struct cpuinfo_x86 *c)
1726 {
1727 	c->extended_cpuid_level = 0;
1728 
1729 	if (!have_cpuid_p())
1730 		identify_cpu_without_cpuid(c);
1731 
1732 	/* cyrix could have cpuid enabled via c_identify()*/
1733 	if (!have_cpuid_p())
1734 		return;
1735 
1736 	cpu_detect(c);
1737 
1738 	get_cpu_vendor(c);
1739 
1740 	get_cpu_cap(c);
1741 
1742 	get_cpu_address_sizes(c);
1743 
1744 	if (c->cpuid_level >= 0x00000001) {
1745 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1746 #ifdef CONFIG_X86_32
1747 # ifdef CONFIG_SMP
1748 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1749 # else
1750 		c->apicid = c->initial_apicid;
1751 # endif
1752 #endif
1753 		c->phys_proc_id = c->initial_apicid;
1754 	}
1755 
1756 	get_model_name(c); /* Default name */
1757 
1758 	/*
1759 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1760 	 * systems that run Linux at CPL > 0 may or may not have the
1761 	 * issue, but, even if they have the issue, there's absolutely
1762 	 * nothing we can do about it because we can't use the real IRET
1763 	 * instruction.
1764 	 *
1765 	 * NB: For the time being, only 32-bit kernels support
1766 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1767 	 * whether to apply espfix using paravirt hooks.  If any
1768 	 * non-paravirt system ever shows up that does *not* have the
1769 	 * ESPFIX issue, we can change this.
1770 	 */
1771 #ifdef CONFIG_X86_32
1772 	set_cpu_bug(c, X86_BUG_ESPFIX);
1773 #endif
1774 }
1775 
1776 /*
1777  * Validate that ACPI/mptables have the same information about the
1778  * effective APIC id and update the package map.
1779  */
1780 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1781 {
1782 #ifdef CONFIG_SMP
1783 	unsigned int apicid, cpu = smp_processor_id();
1784 
1785 	apicid = apic->cpu_present_to_apicid(cpu);
1786 
1787 	if (apicid != c->apicid) {
1788 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1789 		       cpu, apicid, c->initial_apicid);
1790 	}
1791 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1792 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1793 #else
1794 	c->logical_proc_id = 0;
1795 #endif
1796 }
1797 
1798 /*
1799  * This does the hard work of actually picking apart the CPU stuff...
1800  */
1801 static void identify_cpu(struct cpuinfo_x86 *c)
1802 {
1803 	int i;
1804 
1805 	c->loops_per_jiffy = loops_per_jiffy;
1806 	c->x86_cache_size = 0;
1807 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1808 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1809 	c->x86_vendor_id[0] = '\0'; /* Unset */
1810 	c->x86_model_id[0] = '\0';  /* Unset */
1811 	c->x86_max_cores = 1;
1812 	c->x86_coreid_bits = 0;
1813 	c->cu_id = 0xff;
1814 #ifdef CONFIG_X86_64
1815 	c->x86_clflush_size = 64;
1816 	c->x86_phys_bits = 36;
1817 	c->x86_virt_bits = 48;
1818 #else
1819 	c->cpuid_level = -1;	/* CPUID not detected */
1820 	c->x86_clflush_size = 32;
1821 	c->x86_phys_bits = 32;
1822 	c->x86_virt_bits = 32;
1823 #endif
1824 	c->x86_cache_alignment = c->x86_clflush_size;
1825 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1826 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
1827 	memset(&c->vmx_capability, 0, sizeof(c->vmx_capability));
1828 #endif
1829 
1830 	generic_identify(c);
1831 
1832 	if (this_cpu->c_identify)
1833 		this_cpu->c_identify(c);
1834 
1835 	/* Clear/Set all flags overridden by options, after probe */
1836 	apply_forced_caps(c);
1837 
1838 #ifdef CONFIG_X86_64
1839 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1840 #endif
1841 
1842 	/*
1843 	 * Vendor-specific initialization.  In this section we
1844 	 * canonicalize the feature flags, meaning if there are
1845 	 * features a certain CPU supports which CPUID doesn't
1846 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1847 	 * we handle them here.
1848 	 *
1849 	 * At the end of this section, c->x86_capability better
1850 	 * indicate the features this CPU genuinely supports!
1851 	 */
1852 	if (this_cpu->c_init)
1853 		this_cpu->c_init(c);
1854 
1855 	/* Disable the PN if appropriate */
1856 	squash_the_stupid_serial_number(c);
1857 
1858 	/* Set up SMEP/SMAP/UMIP */
1859 	setup_smep(c);
1860 	setup_smap(c);
1861 	setup_umip(c);
1862 
1863 	/* Enable FSGSBASE instructions if available. */
1864 	if (cpu_has(c, X86_FEATURE_FSGSBASE)) {
1865 		cr4_set_bits(X86_CR4_FSGSBASE);
1866 		elf_hwcap2 |= HWCAP2_FSGSBASE;
1867 	}
1868 
1869 	/*
1870 	 * The vendor-specific functions might have changed features.
1871 	 * Now we do "generic changes."
1872 	 */
1873 
1874 	/* Filter out anything that depends on CPUID levels we don't have */
1875 	filter_cpuid_features(c, true);
1876 
1877 	/* If the model name is still unset, do table lookup. */
1878 	if (!c->x86_model_id[0]) {
1879 		const char *p;
1880 		p = table_lookup_model(c);
1881 		if (p)
1882 			strcpy(c->x86_model_id, p);
1883 		else
1884 			/* Last resort... */
1885 			sprintf(c->x86_model_id, "%02x/%02x",
1886 				c->x86, c->x86_model);
1887 	}
1888 
1889 #ifdef CONFIG_X86_64
1890 	detect_ht(c);
1891 #endif
1892 
1893 	x86_init_rdrand(c);
1894 	setup_pku(c);
1895 	setup_cet(c);
1896 
1897 	/*
1898 	 * Clear/Set all flags overridden by options, need do it
1899 	 * before following smp all cpus cap AND.
1900 	 */
1901 	apply_forced_caps(c);
1902 
1903 	/*
1904 	 * On SMP, boot_cpu_data holds the common feature set between
1905 	 * all CPUs; so make sure that we indicate which features are
1906 	 * common between the CPUs.  The first time this routine gets
1907 	 * executed, c == &boot_cpu_data.
1908 	 */
1909 	if (c != &boot_cpu_data) {
1910 		/* AND the already accumulated flags with these */
1911 		for (i = 0; i < NCAPINTS; i++)
1912 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1913 
1914 		/* OR, i.e. replicate the bug flags */
1915 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1916 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1917 	}
1918 
1919 	ppin_init(c);
1920 
1921 	/* Init Machine Check Exception if available. */
1922 	mcheck_cpu_init(c);
1923 
1924 	select_idle_routine(c);
1925 
1926 #ifdef CONFIG_NUMA
1927 	numa_add_cpu(smp_processor_id());
1928 #endif
1929 }
1930 
1931 /*
1932  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1933  * on 32-bit kernels:
1934  */
1935 #ifdef CONFIG_X86_32
1936 void enable_sep_cpu(void)
1937 {
1938 	struct tss_struct *tss;
1939 	int cpu;
1940 
1941 	if (!boot_cpu_has(X86_FEATURE_SEP))
1942 		return;
1943 
1944 	cpu = get_cpu();
1945 	tss = &per_cpu(cpu_tss_rw, cpu);
1946 
1947 	/*
1948 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1949 	 * see the big comment in struct x86_hw_tss's definition.
1950 	 */
1951 
1952 	tss->x86_tss.ss1 = __KERNEL_CS;
1953 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1954 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1955 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1956 
1957 	put_cpu();
1958 }
1959 #endif
1960 
1961 void __init identify_boot_cpu(void)
1962 {
1963 	identify_cpu(&boot_cpu_data);
1964 	if (HAS_KERNEL_IBT && cpu_feature_enabled(X86_FEATURE_IBT))
1965 		pr_info("CET detected: Indirect Branch Tracking enabled\n");
1966 #ifdef CONFIG_X86_32
1967 	enable_sep_cpu();
1968 #endif
1969 	cpu_detect_tlb(&boot_cpu_data);
1970 	setup_cr_pinning();
1971 
1972 	tsx_init();
1973 	lkgs_init();
1974 }
1975 
1976 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1977 {
1978 	BUG_ON(c == &boot_cpu_data);
1979 	identify_cpu(c);
1980 #ifdef CONFIG_X86_32
1981 	enable_sep_cpu();
1982 #endif
1983 	validate_apic_and_package_id(c);
1984 	x86_spec_ctrl_setup_ap();
1985 	update_srbds_msr();
1986 
1987 	tsx_ap_init();
1988 }
1989 
1990 void print_cpu_info(struct cpuinfo_x86 *c)
1991 {
1992 	const char *vendor = NULL;
1993 
1994 	if (c->x86_vendor < X86_VENDOR_NUM) {
1995 		vendor = this_cpu->c_vendor;
1996 	} else {
1997 		if (c->cpuid_level >= 0)
1998 			vendor = c->x86_vendor_id;
1999 	}
2000 
2001 	if (vendor && !strstr(c->x86_model_id, vendor))
2002 		pr_cont("%s ", vendor);
2003 
2004 	if (c->x86_model_id[0])
2005 		pr_cont("%s", c->x86_model_id);
2006 	else
2007 		pr_cont("%d86", c->x86);
2008 
2009 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
2010 
2011 	if (c->x86_stepping || c->cpuid_level >= 0)
2012 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
2013 	else
2014 		pr_cont(")\n");
2015 }
2016 
2017 /*
2018  * clearcpuid= was already parsed in cpu_parse_early_param().  This dummy
2019  * function prevents it from becoming an environment variable for init.
2020  */
2021 static __init int setup_clearcpuid(char *arg)
2022 {
2023 	return 1;
2024 }
2025 __setup("clearcpuid=", setup_clearcpuid);
2026 
2027 DEFINE_PER_CPU_ALIGNED(struct pcpu_hot, pcpu_hot) = {
2028 	.current_task	= &init_task,
2029 	.preempt_count	= INIT_PREEMPT_COUNT,
2030 	.top_of_stack	= TOP_OF_INIT_STACK,
2031 };
2032 EXPORT_PER_CPU_SYMBOL(pcpu_hot);
2033 
2034 #ifdef CONFIG_X86_64
2035 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
2036 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
2037 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
2038 
2039 static void wrmsrl_cstar(unsigned long val)
2040 {
2041 	/*
2042 	 * Intel CPUs do not support 32-bit SYSCALL. Writing to MSR_CSTAR
2043 	 * is so far ignored by the CPU, but raises a #VE trap in a TDX
2044 	 * guest. Avoid the pointless write on all Intel CPUs.
2045 	 */
2046 	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
2047 		wrmsrl(MSR_CSTAR, val);
2048 }
2049 
2050 /* May not be marked __init: used by software suspend */
2051 void syscall_init(void)
2052 {
2053 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
2054 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
2055 
2056 #ifdef CONFIG_IA32_EMULATION
2057 	wrmsrl_cstar((unsigned long)entry_SYSCALL_compat);
2058 	/*
2059 	 * This only works on Intel CPUs.
2060 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
2061 	 * This does not cause SYSENTER to jump to the wrong location, because
2062 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
2063 	 */
2064 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
2065 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
2066 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
2067 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
2068 #else
2069 	wrmsrl_cstar((unsigned long)ignore_sysret);
2070 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
2071 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
2072 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
2073 #endif
2074 
2075 	/*
2076 	 * Flags to clear on syscall; clear as much as possible
2077 	 * to minimize user space-kernel interference.
2078 	 */
2079 	wrmsrl(MSR_SYSCALL_MASK,
2080 	       X86_EFLAGS_CF|X86_EFLAGS_PF|X86_EFLAGS_AF|
2081 	       X86_EFLAGS_ZF|X86_EFLAGS_SF|X86_EFLAGS_TF|
2082 	       X86_EFLAGS_IF|X86_EFLAGS_DF|X86_EFLAGS_OF|
2083 	       X86_EFLAGS_IOPL|X86_EFLAGS_NT|X86_EFLAGS_RF|
2084 	       X86_EFLAGS_AC|X86_EFLAGS_ID);
2085 }
2086 
2087 #else	/* CONFIG_X86_64 */
2088 
2089 #ifdef CONFIG_STACKPROTECTOR
2090 DEFINE_PER_CPU(unsigned long, __stack_chk_guard);
2091 EXPORT_PER_CPU_SYMBOL(__stack_chk_guard);
2092 #endif
2093 
2094 #endif	/* CONFIG_X86_64 */
2095 
2096 /*
2097  * Clear all 6 debug registers:
2098  */
2099 static void clear_all_debug_regs(void)
2100 {
2101 	int i;
2102 
2103 	for (i = 0; i < 8; i++) {
2104 		/* Ignore db4, db5 */
2105 		if ((i == 4) || (i == 5))
2106 			continue;
2107 
2108 		set_debugreg(0, i);
2109 	}
2110 }
2111 
2112 #ifdef CONFIG_KGDB
2113 /*
2114  * Restore debug regs if using kgdbwait and you have a kernel debugger
2115  * connection established.
2116  */
2117 static void dbg_restore_debug_regs(void)
2118 {
2119 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
2120 		arch_kgdb_ops.correct_hw_break();
2121 }
2122 #else /* ! CONFIG_KGDB */
2123 #define dbg_restore_debug_regs()
2124 #endif /* ! CONFIG_KGDB */
2125 
2126 static void wait_for_master_cpu(int cpu)
2127 {
2128 #ifdef CONFIG_SMP
2129 	/*
2130 	 * wait for ACK from master CPU before continuing
2131 	 * with AP initialization
2132 	 */
2133 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
2134 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
2135 		cpu_relax();
2136 #endif
2137 }
2138 
2139 static inline void setup_getcpu(int cpu)
2140 {
2141 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
2142 	struct desc_struct d = { };
2143 
2144 	if (boot_cpu_has(X86_FEATURE_RDTSCP) || boot_cpu_has(X86_FEATURE_RDPID))
2145 		wrmsr(MSR_TSC_AUX, cpudata, 0);
2146 
2147 	/* Store CPU and node number in limit. */
2148 	d.limit0 = cpudata;
2149 	d.limit1 = cpudata >> 16;
2150 
2151 	d.type = 5;		/* RO data, expand down, accessed */
2152 	d.dpl = 3;		/* Visible to user code */
2153 	d.s = 1;		/* Not a system segment */
2154 	d.p = 1;		/* Present */
2155 	d.d = 1;		/* 32-bit */
2156 
2157 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
2158 }
2159 
2160 #ifdef CONFIG_X86_64
2161 static inline void ucode_cpu_init(int cpu)
2162 {
2163 	if (cpu)
2164 		load_ucode_ap();
2165 }
2166 
2167 static inline void tss_setup_ist(struct tss_struct *tss)
2168 {
2169 	/* Set up the per-CPU TSS IST stacks */
2170 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
2171 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
2172 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
2173 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
2174 	/* Only mapped when SEV-ES is active */
2175 	tss->x86_tss.ist[IST_INDEX_VC] = __this_cpu_ist_top_va(VC);
2176 }
2177 
2178 #else /* CONFIG_X86_64 */
2179 
2180 static inline void ucode_cpu_init(int cpu)
2181 {
2182 	show_ucode_info_early();
2183 }
2184 
2185 static inline void tss_setup_ist(struct tss_struct *tss) { }
2186 
2187 #endif /* !CONFIG_X86_64 */
2188 
2189 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
2190 {
2191 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
2192 
2193 #ifdef CONFIG_X86_IOPL_IOPERM
2194 	tss->io_bitmap.prev_max = 0;
2195 	tss->io_bitmap.prev_sequence = 0;
2196 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
2197 	/*
2198 	 * Invalidate the extra array entry past the end of the all
2199 	 * permission bitmap as required by the hardware.
2200 	 */
2201 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
2202 #endif
2203 }
2204 
2205 /*
2206  * Setup everything needed to handle exceptions from the IDT, including the IST
2207  * exceptions which use paranoid_entry().
2208  */
2209 void cpu_init_exception_handling(void)
2210 {
2211 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
2212 	int cpu = raw_smp_processor_id();
2213 
2214 	/* paranoid_entry() gets the CPU number from the GDT */
2215 	setup_getcpu(cpu);
2216 
2217 	/* IST vectors need TSS to be set up. */
2218 	tss_setup_ist(tss);
2219 	tss_setup_io_bitmap(tss);
2220 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
2221 
2222 	load_TR_desc();
2223 
2224 	/* GHCB needs to be setup to handle #VC. */
2225 	setup_ghcb();
2226 
2227 	/* Finally load the IDT */
2228 	load_current_idt();
2229 }
2230 
2231 /*
2232  * cpu_init() initializes state that is per-CPU. Some data is already
2233  * initialized (naturally) in the bootstrap process, such as the GDT.  We
2234  * reload it nevertheless, this function acts as a 'CPU state barrier',
2235  * nothing should get across.
2236  */
2237 void cpu_init(void)
2238 {
2239 	struct task_struct *cur = current;
2240 	int cpu = raw_smp_processor_id();
2241 
2242 	wait_for_master_cpu(cpu);
2243 
2244 	ucode_cpu_init(cpu);
2245 
2246 #ifdef CONFIG_NUMA
2247 	if (this_cpu_read(numa_node) == 0 &&
2248 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
2249 		set_numa_node(early_cpu_to_node(cpu));
2250 #endif
2251 	pr_debug("Initializing CPU#%d\n", cpu);
2252 
2253 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
2254 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
2255 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
2256 
2257 	if (IS_ENABLED(CONFIG_X86_64)) {
2258 		loadsegment(fs, 0);
2259 		memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
2260 		syscall_init();
2261 
2262 		wrmsrl(MSR_FS_BASE, 0);
2263 		wrmsrl(MSR_KERNEL_GS_BASE, 0);
2264 		barrier();
2265 
2266 		x2apic_setup();
2267 	}
2268 
2269 	mmgrab(&init_mm);
2270 	cur->active_mm = &init_mm;
2271 	BUG_ON(cur->mm);
2272 	initialize_tlbstate_and_flush();
2273 	enter_lazy_tlb(&init_mm, cur);
2274 
2275 	/*
2276 	 * sp0 points to the entry trampoline stack regardless of what task
2277 	 * is running.
2278 	 */
2279 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
2280 
2281 	load_mm_ldt(&init_mm);
2282 
2283 	clear_all_debug_regs();
2284 	dbg_restore_debug_regs();
2285 
2286 	doublefault_init_cpu_tss();
2287 
2288 	fpu__init_cpu();
2289 
2290 	if (is_uv_system())
2291 		uv_cpu_init();
2292 
2293 	load_fixmap_gdt(cpu);
2294 }
2295 
2296 #ifdef CONFIG_SMP
2297 void cpu_init_secondary(void)
2298 {
2299 	/*
2300 	 * Relies on the BP having set-up the IDT tables, which are loaded
2301 	 * on this CPU in cpu_init_exception_handling().
2302 	 */
2303 	cpu_init_exception_handling();
2304 	cpu_init();
2305 }
2306 #endif
2307 
2308 #ifdef CONFIG_MICROCODE_LATE_LOADING
2309 /**
2310  * store_cpu_caps() - Store a snapshot of CPU capabilities
2311  * @curr_info: Pointer where to store it
2312  *
2313  * Returns: None
2314  */
2315 void store_cpu_caps(struct cpuinfo_x86 *curr_info)
2316 {
2317 	/* Reload CPUID max function as it might've changed. */
2318 	curr_info->cpuid_level = cpuid_eax(0);
2319 
2320 	/* Copy all capability leafs and pick up the synthetic ones. */
2321 	memcpy(&curr_info->x86_capability, &boot_cpu_data.x86_capability,
2322 	       sizeof(curr_info->x86_capability));
2323 
2324 	/* Get the hardware CPUID leafs */
2325 	get_cpu_cap(curr_info);
2326 }
2327 
2328 /**
2329  * microcode_check() - Check if any CPU capabilities changed after an update.
2330  * @prev_info:	CPU capabilities stored before an update.
2331  *
2332  * The microcode loader calls this upon late microcode load to recheck features,
2333  * only when microcode has been updated. Caller holds microcode_mutex and CPU
2334  * hotplug lock.
2335  *
2336  * Return: None
2337  */
2338 void microcode_check(struct cpuinfo_x86 *prev_info)
2339 {
2340 	struct cpuinfo_x86 curr_info;
2341 
2342 	perf_check_microcode();
2343 
2344 	store_cpu_caps(&curr_info);
2345 
2346 	if (!memcmp(&prev_info->x86_capability, &curr_info.x86_capability,
2347 		    sizeof(prev_info->x86_capability)))
2348 		return;
2349 
2350 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
2351 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
2352 }
2353 #endif
2354 
2355 /*
2356  * Invoked from core CPU hotplug code after hotplug operations
2357  */
2358 void arch_smt_update(void)
2359 {
2360 	/* Handle the speculative execution misfeatures */
2361 	cpu_bugs_smt_update();
2362 	/* Check whether IPI broadcasting can be enabled */
2363 	apic_smt_update();
2364 }
2365