xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 71de0a05)
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(XEON_PHI_KNL, &ppin_info[X86_VENDOR_INTEL]),
125 	X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNM, &ppin_info[X86_VENDOR_INTEL]),
126 
127 	{}
128 };
129 
130 static void ppin_init(struct cpuinfo_x86 *c)
131 {
132 	const struct x86_cpu_id *id;
133 	unsigned long long val;
134 	struct ppin_info *info;
135 
136 	id = x86_match_cpu(ppin_cpuids);
137 	if (!id)
138 		return;
139 
140 	/*
141 	 * Testing the presence of the MSR is not enough. Need to check
142 	 * that the PPIN_CTL allows reading of the PPIN.
143 	 */
144 	info = (struct ppin_info *)id->driver_data;
145 
146 	if (rdmsrl_safe(info->msr_ppin_ctl, &val))
147 		goto clear_ppin;
148 
149 	if ((val & 3UL) == 1UL) {
150 		/* PPIN locked in disabled mode */
151 		goto clear_ppin;
152 	}
153 
154 	/* If PPIN is disabled, try to enable */
155 	if (!(val & 2UL)) {
156 		wrmsrl_safe(info->msr_ppin_ctl,  val | 2UL);
157 		rdmsrl_safe(info->msr_ppin_ctl, &val);
158 	}
159 
160 	/* Is the enable bit set? */
161 	if (val & 2UL) {
162 		c->ppin = __rdmsr(info->msr_ppin);
163 		set_cpu_cap(c, info->feature);
164 		return;
165 	}
166 
167 clear_ppin:
168 	clear_cpu_cap(c, info->feature);
169 }
170 
171 /* correctly size the local cpu masks */
172 void __init setup_cpu_local_masks(void)
173 {
174 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
175 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
176 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
177 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
178 }
179 
180 static void default_init(struct cpuinfo_x86 *c)
181 {
182 #ifdef CONFIG_X86_64
183 	cpu_detect_cache_sizes(c);
184 #else
185 	/* Not much we can do here... */
186 	/* Check if at least it has cpuid */
187 	if (c->cpuid_level == -1) {
188 		/* No cpuid. It must be an ancient CPU */
189 		if (c->x86 == 4)
190 			strcpy(c->x86_model_id, "486");
191 		else if (c->x86 == 3)
192 			strcpy(c->x86_model_id, "386");
193 	}
194 #endif
195 }
196 
197 static const struct cpu_dev default_cpu = {
198 	.c_init		= default_init,
199 	.c_vendor	= "Unknown",
200 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
201 };
202 
203 static const struct cpu_dev *this_cpu = &default_cpu;
204 
205 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
206 #ifdef CONFIG_X86_64
207 	/*
208 	 * We need valid kernel segments for data and code in long mode too
209 	 * IRET will check the segment types  kkeil 2000/10/28
210 	 * Also sysret mandates a special GDT layout
211 	 *
212 	 * TLS descriptors are currently at a different place compared to i386.
213 	 * Hopefully nobody expects them at a fixed place (Wine?)
214 	 */
215 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
216 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
217 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
218 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
219 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
220 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
221 #else
222 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
223 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
224 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
225 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
226 	/*
227 	 * Segments used for calling PnP BIOS have byte granularity.
228 	 * They code segments and data segments have fixed 64k limits,
229 	 * the transfer segment sizes are set at run time.
230 	 */
231 	/* 32-bit code */
232 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
233 	/* 16-bit code */
234 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
235 	/* 16-bit data */
236 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
237 	/* 16-bit data */
238 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
239 	/* 16-bit data */
240 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
241 	/*
242 	 * The APM segments have byte granularity and their bases
243 	 * are set at run time.  All have 64k limits.
244 	 */
245 	/* 32-bit code */
246 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
247 	/* 16-bit code */
248 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
249 	/* data */
250 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
251 
252 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
253 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
254 #endif
255 } };
256 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
257 
258 #ifdef CONFIG_X86_64
259 static int __init x86_nopcid_setup(char *s)
260 {
261 	/* nopcid doesn't accept parameters */
262 	if (s)
263 		return -EINVAL;
264 
265 	/* do not emit a message if the feature is not present */
266 	if (!boot_cpu_has(X86_FEATURE_PCID))
267 		return 0;
268 
269 	setup_clear_cpu_cap(X86_FEATURE_PCID);
270 	pr_info("nopcid: PCID feature disabled\n");
271 	return 0;
272 }
273 early_param("nopcid", x86_nopcid_setup);
274 #endif
275 
276 static int __init x86_noinvpcid_setup(char *s)
277 {
278 	/* noinvpcid doesn't accept parameters */
279 	if (s)
280 		return -EINVAL;
281 
282 	/* do not emit a message if the feature is not present */
283 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
284 		return 0;
285 
286 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
287 	pr_info("noinvpcid: INVPCID feature disabled\n");
288 	return 0;
289 }
290 early_param("noinvpcid", x86_noinvpcid_setup);
291 
292 #ifdef CONFIG_X86_32
293 static int cachesize_override = -1;
294 static int disable_x86_serial_nr = 1;
295 
296 static int __init cachesize_setup(char *str)
297 {
298 	get_option(&str, &cachesize_override);
299 	return 1;
300 }
301 __setup("cachesize=", cachesize_setup);
302 
303 /* Standard macro to see if a specific flag is changeable */
304 static inline int flag_is_changeable_p(u32 flag)
305 {
306 	u32 f1, f2;
307 
308 	/*
309 	 * Cyrix and IDT cpus allow disabling of CPUID
310 	 * so the code below may return different results
311 	 * when it is executed before and after enabling
312 	 * the CPUID. Add "volatile" to not allow gcc to
313 	 * optimize the subsequent calls to this function.
314 	 */
315 	asm volatile ("pushfl		\n\t"
316 		      "pushfl		\n\t"
317 		      "popl %0		\n\t"
318 		      "movl %0, %1	\n\t"
319 		      "xorl %2, %0	\n\t"
320 		      "pushl %0		\n\t"
321 		      "popfl		\n\t"
322 		      "pushfl		\n\t"
323 		      "popl %0		\n\t"
324 		      "popfl		\n\t"
325 
326 		      : "=&r" (f1), "=&r" (f2)
327 		      : "ir" (flag));
328 
329 	return ((f1^f2) & flag) != 0;
330 }
331 
332 /* Probe for the CPUID instruction */
333 int have_cpuid_p(void)
334 {
335 	return flag_is_changeable_p(X86_EFLAGS_ID);
336 }
337 
338 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
339 {
340 	unsigned long lo, hi;
341 
342 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
343 		return;
344 
345 	/* Disable processor serial number: */
346 
347 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
348 	lo |= 0x200000;
349 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
350 
351 	pr_notice("CPU serial number disabled.\n");
352 	clear_cpu_cap(c, X86_FEATURE_PN);
353 
354 	/* Disabling the serial number may affect the cpuid level */
355 	c->cpuid_level = cpuid_eax(0);
356 }
357 
358 static int __init x86_serial_nr_setup(char *s)
359 {
360 	disable_x86_serial_nr = 0;
361 	return 1;
362 }
363 __setup("serialnumber", x86_serial_nr_setup);
364 #else
365 static inline int flag_is_changeable_p(u32 flag)
366 {
367 	return 1;
368 }
369 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
370 {
371 }
372 #endif
373 
374 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
375 {
376 	if (cpu_has(c, X86_FEATURE_SMEP))
377 		cr4_set_bits(X86_CR4_SMEP);
378 }
379 
380 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
381 {
382 	unsigned long eflags = native_save_fl();
383 
384 	/* This should have been cleared long ago */
385 	BUG_ON(eflags & X86_EFLAGS_AC);
386 
387 	if (cpu_has(c, X86_FEATURE_SMAP))
388 		cr4_set_bits(X86_CR4_SMAP);
389 }
390 
391 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
392 {
393 	/* Check the boot processor, plus build option for UMIP. */
394 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
395 		goto out;
396 
397 	/* Check the current processor's cpuid bits. */
398 	if (!cpu_has(c, X86_FEATURE_UMIP))
399 		goto out;
400 
401 	cr4_set_bits(X86_CR4_UMIP);
402 
403 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
404 
405 	return;
406 
407 out:
408 	/*
409 	 * Make sure UMIP is disabled in case it was enabled in a
410 	 * previous boot (e.g., via kexec).
411 	 */
412 	cr4_clear_bits(X86_CR4_UMIP);
413 }
414 
415 /* These bits should not change their value after CPU init is finished. */
416 static const unsigned long cr4_pinned_mask =
417 	X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP |
418 	X86_CR4_FSGSBASE | X86_CR4_CET;
419 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
420 static unsigned long cr4_pinned_bits __ro_after_init;
421 
422 void native_write_cr0(unsigned long val)
423 {
424 	unsigned long bits_missing = 0;
425 
426 set_register:
427 	asm volatile("mov %0,%%cr0": "+r" (val) : : "memory");
428 
429 	if (static_branch_likely(&cr_pinning)) {
430 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
431 			bits_missing = X86_CR0_WP;
432 			val |= bits_missing;
433 			goto set_register;
434 		}
435 		/* Warn after we've set the missing bits. */
436 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
437 	}
438 }
439 EXPORT_SYMBOL(native_write_cr0);
440 
441 void __no_profile native_write_cr4(unsigned long val)
442 {
443 	unsigned long bits_changed = 0;
444 
445 set_register:
446 	asm volatile("mov %0,%%cr4": "+r" (val) : : "memory");
447 
448 	if (static_branch_likely(&cr_pinning)) {
449 		if (unlikely((val & cr4_pinned_mask) != cr4_pinned_bits)) {
450 			bits_changed = (val & cr4_pinned_mask) ^ cr4_pinned_bits;
451 			val = (val & ~cr4_pinned_mask) | cr4_pinned_bits;
452 			goto set_register;
453 		}
454 		/* Warn after we've corrected the changed bits. */
455 		WARN_ONCE(bits_changed, "pinned CR4 bits changed: 0x%lx!?\n",
456 			  bits_changed);
457 	}
458 }
459 #if IS_MODULE(CONFIG_LKDTM)
460 EXPORT_SYMBOL_GPL(native_write_cr4);
461 #endif
462 
463 void cr4_update_irqsoff(unsigned long set, unsigned long clear)
464 {
465 	unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
466 
467 	lockdep_assert_irqs_disabled();
468 
469 	newval = (cr4 & ~clear) | set;
470 	if (newval != cr4) {
471 		this_cpu_write(cpu_tlbstate.cr4, newval);
472 		__write_cr4(newval);
473 	}
474 }
475 EXPORT_SYMBOL(cr4_update_irqsoff);
476 
477 /* Read the CR4 shadow. */
478 unsigned long cr4_read_shadow(void)
479 {
480 	return this_cpu_read(cpu_tlbstate.cr4);
481 }
482 EXPORT_SYMBOL_GPL(cr4_read_shadow);
483 
484 void cr4_init(void)
485 {
486 	unsigned long cr4 = __read_cr4();
487 
488 	if (boot_cpu_has(X86_FEATURE_PCID))
489 		cr4 |= X86_CR4_PCIDE;
490 	if (static_branch_likely(&cr_pinning))
491 		cr4 = (cr4 & ~cr4_pinned_mask) | cr4_pinned_bits;
492 
493 	__write_cr4(cr4);
494 
495 	/* Initialize cr4 shadow for this CPU. */
496 	this_cpu_write(cpu_tlbstate.cr4, cr4);
497 }
498 
499 /*
500  * Once CPU feature detection is finished (and boot params have been
501  * parsed), record any of the sensitive CR bits that are set, and
502  * enable CR pinning.
503  */
504 static void __init setup_cr_pinning(void)
505 {
506 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & cr4_pinned_mask;
507 	static_key_enable(&cr_pinning.key);
508 }
509 
510 static __init int x86_nofsgsbase_setup(char *arg)
511 {
512 	/* Require an exact match without trailing characters. */
513 	if (strlen(arg))
514 		return 0;
515 
516 	/* Do not emit a message if the feature is not present. */
517 	if (!boot_cpu_has(X86_FEATURE_FSGSBASE))
518 		return 1;
519 
520 	setup_clear_cpu_cap(X86_FEATURE_FSGSBASE);
521 	pr_info("FSGSBASE disabled via kernel command line\n");
522 	return 1;
523 }
524 __setup("nofsgsbase", x86_nofsgsbase_setup);
525 
526 /*
527  * Protection Keys are not available in 32-bit mode.
528  */
529 static bool pku_disabled;
530 
531 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
532 {
533 	if (c == &boot_cpu_data) {
534 		if (pku_disabled || !cpu_feature_enabled(X86_FEATURE_PKU))
535 			return;
536 		/*
537 		 * Setting CR4.PKE will cause the X86_FEATURE_OSPKE cpuid
538 		 * bit to be set.  Enforce it.
539 		 */
540 		setup_force_cpu_cap(X86_FEATURE_OSPKE);
541 
542 	} else if (!cpu_feature_enabled(X86_FEATURE_OSPKE)) {
543 		return;
544 	}
545 
546 	cr4_set_bits(X86_CR4_PKE);
547 	/* Load the default PKRU value */
548 	pkru_write_default();
549 }
550 
551 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
552 static __init int setup_disable_pku(char *arg)
553 {
554 	/*
555 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
556 	 * runtime checks are against OSPKE so clearing the
557 	 * bit does nothing.
558 	 *
559 	 * This way, we will see "pku" in cpuinfo, but not
560 	 * "ospke", which is exactly what we want.  It shows
561 	 * that the CPU has PKU, but the OS has not enabled it.
562 	 * This happens to be exactly how a system would look
563 	 * if we disabled the config option.
564 	 */
565 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
566 	pku_disabled = true;
567 	return 1;
568 }
569 __setup("nopku", setup_disable_pku);
570 #endif
571 
572 #ifdef CONFIG_X86_KERNEL_IBT
573 
574 __noendbr u64 ibt_save(bool disable)
575 {
576 	u64 msr = 0;
577 
578 	if (cpu_feature_enabled(X86_FEATURE_IBT)) {
579 		rdmsrl(MSR_IA32_S_CET, msr);
580 		if (disable)
581 			wrmsrl(MSR_IA32_S_CET, msr & ~CET_ENDBR_EN);
582 	}
583 
584 	return msr;
585 }
586 
587 __noendbr void ibt_restore(u64 save)
588 {
589 	u64 msr;
590 
591 	if (cpu_feature_enabled(X86_FEATURE_IBT)) {
592 		rdmsrl(MSR_IA32_S_CET, msr);
593 		msr &= ~CET_ENDBR_EN;
594 		msr |= (save & CET_ENDBR_EN);
595 		wrmsrl(MSR_IA32_S_CET, msr);
596 	}
597 }
598 
599 #endif
600 
601 static __always_inline void setup_cet(struct cpuinfo_x86 *c)
602 {
603 	u64 msr = CET_ENDBR_EN;
604 
605 	if (!HAS_KERNEL_IBT ||
606 	    !cpu_feature_enabled(X86_FEATURE_IBT))
607 		return;
608 
609 	wrmsrl(MSR_IA32_S_CET, msr);
610 	cr4_set_bits(X86_CR4_CET);
611 
612 	if (!ibt_selftest()) {
613 		pr_err("IBT selftest: Failed!\n");
614 		wrmsrl(MSR_IA32_S_CET, 0);
615 		setup_clear_cpu_cap(X86_FEATURE_IBT);
616 		return;
617 	}
618 }
619 
620 __noendbr void cet_disable(void)
621 {
622 	if (cpu_feature_enabled(X86_FEATURE_IBT))
623 		wrmsrl(MSR_IA32_S_CET, 0);
624 }
625 
626 /*
627  * Some CPU features depend on higher CPUID levels, which may not always
628  * be available due to CPUID level capping or broken virtualization
629  * software.  Add those features to this table to auto-disable them.
630  */
631 struct cpuid_dependent_feature {
632 	u32 feature;
633 	u32 level;
634 };
635 
636 static const struct cpuid_dependent_feature
637 cpuid_dependent_features[] = {
638 	{ X86_FEATURE_MWAIT,		0x00000005 },
639 	{ X86_FEATURE_DCA,		0x00000009 },
640 	{ X86_FEATURE_XSAVE,		0x0000000d },
641 	{ 0, 0 }
642 };
643 
644 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
645 {
646 	const struct cpuid_dependent_feature *df;
647 
648 	for (df = cpuid_dependent_features; df->feature; df++) {
649 
650 		if (!cpu_has(c, df->feature))
651 			continue;
652 		/*
653 		 * Note: cpuid_level is set to -1 if unavailable, but
654 		 * extended_extended_level is set to 0 if unavailable
655 		 * and the legitimate extended levels are all negative
656 		 * when signed; hence the weird messing around with
657 		 * signs here...
658 		 */
659 		if (!((s32)df->level < 0 ?
660 		     (u32)df->level > (u32)c->extended_cpuid_level :
661 		     (s32)df->level > (s32)c->cpuid_level))
662 			continue;
663 
664 		clear_cpu_cap(c, df->feature);
665 		if (!warn)
666 			continue;
667 
668 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
669 			x86_cap_flag(df->feature), df->level);
670 	}
671 }
672 
673 /*
674  * Naming convention should be: <Name> [(<Codename>)]
675  * This table only is used unless init_<vendor>() below doesn't set it;
676  * in particular, if CPUID levels 0x80000002..4 are supported, this
677  * isn't used
678  */
679 
680 /* Look up CPU names by table lookup. */
681 static const char *table_lookup_model(struct cpuinfo_x86 *c)
682 {
683 #ifdef CONFIG_X86_32
684 	const struct legacy_cpu_model_info *info;
685 
686 	if (c->x86_model >= 16)
687 		return NULL;	/* Range check */
688 
689 	if (!this_cpu)
690 		return NULL;
691 
692 	info = this_cpu->legacy_models;
693 
694 	while (info->family) {
695 		if (info->family == c->x86)
696 			return info->model_names[c->x86_model];
697 		info++;
698 	}
699 #endif
700 	return NULL;		/* Not found */
701 }
702 
703 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
704 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
705 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
706 
707 #ifdef CONFIG_X86_32
708 /* The 32-bit entry code needs to find cpu_entry_area. */
709 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
710 #endif
711 
712 /* Load the original GDT from the per-cpu structure */
713 void load_direct_gdt(int cpu)
714 {
715 	struct desc_ptr gdt_descr;
716 
717 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
718 	gdt_descr.size = GDT_SIZE - 1;
719 	load_gdt(&gdt_descr);
720 }
721 EXPORT_SYMBOL_GPL(load_direct_gdt);
722 
723 /* Load a fixmap remapping of the per-cpu GDT */
724 void load_fixmap_gdt(int cpu)
725 {
726 	struct desc_ptr gdt_descr;
727 
728 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
729 	gdt_descr.size = GDT_SIZE - 1;
730 	load_gdt(&gdt_descr);
731 }
732 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
733 
734 /**
735  * switch_gdt_and_percpu_base - Switch to direct GDT and runtime per CPU base
736  * @cpu:	The CPU number for which this is invoked
737  *
738  * Invoked during early boot to switch from early GDT and early per CPU to
739  * the direct GDT and the runtime per CPU area. On 32-bit the percpu base
740  * switch is implicit by loading the direct GDT. On 64bit this requires
741  * to update GSBASE.
742  */
743 void __init switch_gdt_and_percpu_base(int cpu)
744 {
745 	load_direct_gdt(cpu);
746 
747 #ifdef CONFIG_X86_64
748 	/*
749 	 * No need to load %gs. It is already correct.
750 	 *
751 	 * Writing %gs on 64bit would zero GSBASE which would make any per
752 	 * CPU operation up to the point of the wrmsrl() fault.
753 	 *
754 	 * Set GSBASE to the new offset. Until the wrmsrl() happens the
755 	 * early mapping is still valid. That means the GSBASE update will
756 	 * lose any prior per CPU data which was not copied over in
757 	 * setup_per_cpu_areas().
758 	 *
759 	 * This works even with stackprotector enabled because the
760 	 * per CPU stack canary is 0 in both per CPU areas.
761 	 */
762 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
763 #else
764 	/*
765 	 * %fs is already set to __KERNEL_PERCPU, but after switching GDT
766 	 * it is required to load FS again so that the 'hidden' part is
767 	 * updated from the new GDT. Up to this point the early per CPU
768 	 * translation is active. Any content of the early per CPU data
769 	 * which was not copied over in setup_per_cpu_areas() is lost.
770 	 */
771 	loadsegment(fs, __KERNEL_PERCPU);
772 #endif
773 }
774 
775 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
776 
777 static void get_model_name(struct cpuinfo_x86 *c)
778 {
779 	unsigned int *v;
780 	char *p, *q, *s;
781 
782 	if (c->extended_cpuid_level < 0x80000004)
783 		return;
784 
785 	v = (unsigned int *)c->x86_model_id;
786 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
787 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
788 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
789 	c->x86_model_id[48] = 0;
790 
791 	/* Trim whitespace */
792 	p = q = s = &c->x86_model_id[0];
793 
794 	while (*p == ' ')
795 		p++;
796 
797 	while (*p) {
798 		/* Note the last non-whitespace index */
799 		if (!isspace(*p))
800 			s = q;
801 
802 		*q++ = *p++;
803 	}
804 
805 	*(s + 1) = '\0';
806 }
807 
808 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
809 {
810 	unsigned int eax, ebx, ecx, edx;
811 
812 	c->x86_max_cores = 1;
813 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
814 		return;
815 
816 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
817 	if (eax & 0x1f)
818 		c->x86_max_cores = (eax >> 26) + 1;
819 }
820 
821 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
822 {
823 	unsigned int n, dummy, ebx, ecx, edx, l2size;
824 
825 	n = c->extended_cpuid_level;
826 
827 	if (n >= 0x80000005) {
828 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
829 		c->x86_cache_size = (ecx>>24) + (edx>>24);
830 #ifdef CONFIG_X86_64
831 		/* On K8 L1 TLB is inclusive, so don't count it */
832 		c->x86_tlbsize = 0;
833 #endif
834 	}
835 
836 	if (n < 0x80000006)	/* Some chips just has a large L1. */
837 		return;
838 
839 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
840 	l2size = ecx >> 16;
841 
842 #ifdef CONFIG_X86_64
843 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
844 #else
845 	/* do processor-specific cache resizing */
846 	if (this_cpu->legacy_cache_size)
847 		l2size = this_cpu->legacy_cache_size(c, l2size);
848 
849 	/* Allow user to override all this if necessary. */
850 	if (cachesize_override != -1)
851 		l2size = cachesize_override;
852 
853 	if (l2size == 0)
854 		return;		/* Again, no L2 cache is possible */
855 #endif
856 
857 	c->x86_cache_size = l2size;
858 }
859 
860 u16 __read_mostly tlb_lli_4k[NR_INFO];
861 u16 __read_mostly tlb_lli_2m[NR_INFO];
862 u16 __read_mostly tlb_lli_4m[NR_INFO];
863 u16 __read_mostly tlb_lld_4k[NR_INFO];
864 u16 __read_mostly tlb_lld_2m[NR_INFO];
865 u16 __read_mostly tlb_lld_4m[NR_INFO];
866 u16 __read_mostly tlb_lld_1g[NR_INFO];
867 
868 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
869 {
870 	if (this_cpu->c_detect_tlb)
871 		this_cpu->c_detect_tlb(c);
872 
873 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
874 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
875 		tlb_lli_4m[ENTRIES]);
876 
877 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
878 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
879 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
880 }
881 
882 int detect_ht_early(struct cpuinfo_x86 *c)
883 {
884 #ifdef CONFIG_SMP
885 	u32 eax, ebx, ecx, edx;
886 
887 	if (!cpu_has(c, X86_FEATURE_HT))
888 		return -1;
889 
890 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
891 		return -1;
892 
893 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
894 		return -1;
895 
896 	cpuid(1, &eax, &ebx, &ecx, &edx);
897 
898 	smp_num_siblings = (ebx & 0xff0000) >> 16;
899 	if (smp_num_siblings == 1)
900 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
901 #endif
902 	return 0;
903 }
904 
905 void detect_ht(struct cpuinfo_x86 *c)
906 {
907 #ifdef CONFIG_SMP
908 	int index_msb, core_bits;
909 
910 	if (detect_ht_early(c) < 0)
911 		return;
912 
913 	index_msb = get_count_order(smp_num_siblings);
914 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
915 
916 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
917 
918 	index_msb = get_count_order(smp_num_siblings);
919 
920 	core_bits = get_count_order(c->x86_max_cores);
921 
922 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
923 				       ((1 << core_bits) - 1);
924 #endif
925 }
926 
927 static void get_cpu_vendor(struct cpuinfo_x86 *c)
928 {
929 	char *v = c->x86_vendor_id;
930 	int i;
931 
932 	for (i = 0; i < X86_VENDOR_NUM; i++) {
933 		if (!cpu_devs[i])
934 			break;
935 
936 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
937 		    (cpu_devs[i]->c_ident[1] &&
938 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
939 
940 			this_cpu = cpu_devs[i];
941 			c->x86_vendor = this_cpu->c_x86_vendor;
942 			return;
943 		}
944 	}
945 
946 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
947 		    "CPU: Your system may be unstable.\n", v);
948 
949 	c->x86_vendor = X86_VENDOR_UNKNOWN;
950 	this_cpu = &default_cpu;
951 }
952 
953 void cpu_detect(struct cpuinfo_x86 *c)
954 {
955 	/* Get vendor name */
956 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
957 	      (unsigned int *)&c->x86_vendor_id[0],
958 	      (unsigned int *)&c->x86_vendor_id[8],
959 	      (unsigned int *)&c->x86_vendor_id[4]);
960 
961 	c->x86 = 4;
962 	/* Intel-defined flags: level 0x00000001 */
963 	if (c->cpuid_level >= 0x00000001) {
964 		u32 junk, tfms, cap0, misc;
965 
966 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
967 		c->x86		= x86_family(tfms);
968 		c->x86_model	= x86_model(tfms);
969 		c->x86_stepping	= x86_stepping(tfms);
970 
971 		if (cap0 & (1<<19)) {
972 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
973 			c->x86_cache_alignment = c->x86_clflush_size;
974 		}
975 	}
976 }
977 
978 static void apply_forced_caps(struct cpuinfo_x86 *c)
979 {
980 	int i;
981 
982 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
983 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
984 		c->x86_capability[i] |= cpu_caps_set[i];
985 	}
986 }
987 
988 static void init_speculation_control(struct cpuinfo_x86 *c)
989 {
990 	/*
991 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
992 	 * and they also have a different bit for STIBP support. Also,
993 	 * a hypervisor might have set the individual AMD bits even on
994 	 * Intel CPUs, for finer-grained selection of what's available.
995 	 */
996 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
997 		set_cpu_cap(c, X86_FEATURE_IBRS);
998 		set_cpu_cap(c, X86_FEATURE_IBPB);
999 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1000 	}
1001 
1002 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
1003 		set_cpu_cap(c, X86_FEATURE_STIBP);
1004 
1005 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
1006 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
1007 		set_cpu_cap(c, X86_FEATURE_SSBD);
1008 
1009 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
1010 		set_cpu_cap(c, X86_FEATURE_IBRS);
1011 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1012 	}
1013 
1014 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
1015 		set_cpu_cap(c, X86_FEATURE_IBPB);
1016 
1017 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
1018 		set_cpu_cap(c, X86_FEATURE_STIBP);
1019 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1020 	}
1021 
1022 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
1023 		set_cpu_cap(c, X86_FEATURE_SSBD);
1024 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1025 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
1026 	}
1027 }
1028 
1029 void get_cpu_cap(struct cpuinfo_x86 *c)
1030 {
1031 	u32 eax, ebx, ecx, edx;
1032 
1033 	/* Intel-defined flags: level 0x00000001 */
1034 	if (c->cpuid_level >= 0x00000001) {
1035 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
1036 
1037 		c->x86_capability[CPUID_1_ECX] = ecx;
1038 		c->x86_capability[CPUID_1_EDX] = edx;
1039 	}
1040 
1041 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
1042 	if (c->cpuid_level >= 0x00000006)
1043 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
1044 
1045 	/* Additional Intel-defined flags: level 0x00000007 */
1046 	if (c->cpuid_level >= 0x00000007) {
1047 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
1048 		c->x86_capability[CPUID_7_0_EBX] = ebx;
1049 		c->x86_capability[CPUID_7_ECX] = ecx;
1050 		c->x86_capability[CPUID_7_EDX] = edx;
1051 
1052 		/* Check valid sub-leaf index before accessing it */
1053 		if (eax >= 1) {
1054 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
1055 			c->x86_capability[CPUID_7_1_EAX] = eax;
1056 		}
1057 	}
1058 
1059 	/* Extended state features: level 0x0000000d */
1060 	if (c->cpuid_level >= 0x0000000d) {
1061 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
1062 
1063 		c->x86_capability[CPUID_D_1_EAX] = eax;
1064 	}
1065 
1066 	/* AMD-defined flags: level 0x80000001 */
1067 	eax = cpuid_eax(0x80000000);
1068 	c->extended_cpuid_level = eax;
1069 
1070 	if ((eax & 0xffff0000) == 0x80000000) {
1071 		if (eax >= 0x80000001) {
1072 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
1073 
1074 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
1075 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
1076 		}
1077 	}
1078 
1079 	if (c->extended_cpuid_level >= 0x80000007) {
1080 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
1081 
1082 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
1083 		c->x86_power = edx;
1084 	}
1085 
1086 	if (c->extended_cpuid_level >= 0x80000008) {
1087 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
1088 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
1089 	}
1090 
1091 	if (c->extended_cpuid_level >= 0x8000000a)
1092 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
1093 
1094 	if (c->extended_cpuid_level >= 0x8000001f)
1095 		c->x86_capability[CPUID_8000_001F_EAX] = cpuid_eax(0x8000001f);
1096 
1097 	if (c->extended_cpuid_level >= 0x80000021)
1098 		c->x86_capability[CPUID_8000_0021_EAX] = cpuid_eax(0x80000021);
1099 
1100 	init_scattered_cpuid_features(c);
1101 	init_speculation_control(c);
1102 
1103 	/*
1104 	 * Clear/Set all flags overridden by options, after probe.
1105 	 * This needs to happen each time we re-probe, which may happen
1106 	 * several times during CPU initialization.
1107 	 */
1108 	apply_forced_caps(c);
1109 }
1110 
1111 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
1112 {
1113 	u32 eax, ebx, ecx, edx;
1114 
1115 	if (c->extended_cpuid_level >= 0x80000008) {
1116 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
1117 
1118 		c->x86_virt_bits = (eax >> 8) & 0xff;
1119 		c->x86_phys_bits = eax & 0xff;
1120 	}
1121 #ifdef CONFIG_X86_32
1122 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
1123 		c->x86_phys_bits = 36;
1124 #endif
1125 	c->x86_cache_bits = c->x86_phys_bits;
1126 }
1127 
1128 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
1129 {
1130 #ifdef CONFIG_X86_32
1131 	int i;
1132 
1133 	/*
1134 	 * First of all, decide if this is a 486 or higher
1135 	 * It's a 486 if we can modify the AC flag
1136 	 */
1137 	if (flag_is_changeable_p(X86_EFLAGS_AC))
1138 		c->x86 = 4;
1139 	else
1140 		c->x86 = 3;
1141 
1142 	for (i = 0; i < X86_VENDOR_NUM; i++)
1143 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
1144 			c->x86_vendor_id[0] = 0;
1145 			cpu_devs[i]->c_identify(c);
1146 			if (c->x86_vendor_id[0]) {
1147 				get_cpu_vendor(c);
1148 				break;
1149 			}
1150 		}
1151 #endif
1152 }
1153 
1154 #define NO_SPECULATION		BIT(0)
1155 #define NO_MELTDOWN		BIT(1)
1156 #define NO_SSB			BIT(2)
1157 #define NO_L1TF			BIT(3)
1158 #define NO_MDS			BIT(4)
1159 #define MSBDS_ONLY		BIT(5)
1160 #define NO_SWAPGS		BIT(6)
1161 #define NO_ITLB_MULTIHIT	BIT(7)
1162 #define NO_SPECTRE_V2		BIT(8)
1163 #define NO_MMIO			BIT(9)
1164 #define NO_EIBRS_PBRSB		BIT(10)
1165 
1166 #define VULNWL(vendor, family, model, whitelist)	\
1167 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist)
1168 
1169 #define VULNWL_INTEL(model, whitelist)		\
1170 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1171 
1172 #define VULNWL_AMD(family, whitelist)		\
1173 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1174 
1175 #define VULNWL_HYGON(family, whitelist)		\
1176 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1177 
1178 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1179 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1180 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1181 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1182 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1183 	VULNWL(VORTEX,	5, X86_MODEL_ANY,	NO_SPECULATION),
1184 	VULNWL(VORTEX,	6, X86_MODEL_ANY,	NO_SPECULATION),
1185 
1186 	/* Intel Family 6 */
1187 	VULNWL_INTEL(TIGERLAKE,			NO_MMIO),
1188 	VULNWL_INTEL(TIGERLAKE_L,		NO_MMIO),
1189 	VULNWL_INTEL(ALDERLAKE,			NO_MMIO),
1190 	VULNWL_INTEL(ALDERLAKE_L,		NO_MMIO),
1191 
1192 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1193 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION | NO_ITLB_MULTIHIT),
1194 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1195 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1196 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1197 
1198 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1199 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1200 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1201 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1202 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1203 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1204 
1205 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1206 
1207 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1208 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1209 
1210 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1211 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1212 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB),
1213 
1214 	/*
1215 	 * Technically, swapgs isn't serializing on AMD (despite it previously
1216 	 * being documented as such in the APM).  But according to AMD, %gs is
1217 	 * updated non-speculatively, and the issuing of %gs-relative memory
1218 	 * operands will be blocked until the %gs update completes, which is
1219 	 * good enough for our purposes.
1220 	 */
1221 
1222 	VULNWL_INTEL(ATOM_TREMONT,		NO_EIBRS_PBRSB),
1223 	VULNWL_INTEL(ATOM_TREMONT_L,		NO_EIBRS_PBRSB),
1224 	VULNWL_INTEL(ATOM_TREMONT_D,		NO_ITLB_MULTIHIT | NO_EIBRS_PBRSB),
1225 
1226 	/* AMD Family 0xf - 0x12 */
1227 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1228 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1229 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1230 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1231 
1232 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1233 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB),
1234 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB),
1235 
1236 	/* Zhaoxin Family 7 */
1237 	VULNWL(CENTAUR,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO),
1238 	VULNWL(ZHAOXIN,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO),
1239 	{}
1240 };
1241 
1242 #define VULNBL(vendor, family, model, blacklist)	\
1243 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, blacklist)
1244 
1245 #define VULNBL_INTEL_STEPPINGS(model, steppings, issues)		   \
1246 	X86_MATCH_VENDOR_FAM_MODEL_STEPPINGS_FEATURE(INTEL, 6,		   \
1247 					    INTEL_FAM6_##model, steppings, \
1248 					    X86_FEATURE_ANY, issues)
1249 
1250 #define VULNBL_AMD(family, blacklist)		\
1251 	VULNBL(AMD, family, X86_MODEL_ANY, blacklist)
1252 
1253 #define VULNBL_HYGON(family, blacklist)		\
1254 	VULNBL(HYGON, family, X86_MODEL_ANY, blacklist)
1255 
1256 #define SRBDS		BIT(0)
1257 /* CPU is affected by X86_BUG_MMIO_STALE_DATA */
1258 #define MMIO		BIT(1)
1259 /* CPU is affected by Shared Buffers Data Sampling (SBDS), a variant of X86_BUG_MMIO_STALE_DATA */
1260 #define MMIO_SBDS	BIT(2)
1261 /* CPU is affected by RETbleed, speculating where you would not expect it */
1262 #define RETBLEED	BIT(3)
1263 /* CPU is affected by SMT (cross-thread) return predictions */
1264 #define SMT_RSB		BIT(4)
1265 
1266 static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = {
1267 	VULNBL_INTEL_STEPPINGS(IVYBRIDGE,	X86_STEPPING_ANY,		SRBDS),
1268 	VULNBL_INTEL_STEPPINGS(HASWELL,		X86_STEPPING_ANY,		SRBDS),
1269 	VULNBL_INTEL_STEPPINGS(HASWELL_L,	X86_STEPPING_ANY,		SRBDS),
1270 	VULNBL_INTEL_STEPPINGS(HASWELL_G,	X86_STEPPING_ANY,		SRBDS),
1271 	VULNBL_INTEL_STEPPINGS(HASWELL_X,	X86_STEPPING_ANY,		MMIO),
1272 	VULNBL_INTEL_STEPPINGS(BROADWELL_D,	X86_STEPPING_ANY,		MMIO),
1273 	VULNBL_INTEL_STEPPINGS(BROADWELL_G,	X86_STEPPING_ANY,		SRBDS),
1274 	VULNBL_INTEL_STEPPINGS(BROADWELL_X,	X86_STEPPING_ANY,		MMIO),
1275 	VULNBL_INTEL_STEPPINGS(BROADWELL,	X86_STEPPING_ANY,		SRBDS),
1276 	VULNBL_INTEL_STEPPINGS(SKYLAKE_L,	X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1277 	VULNBL_INTEL_STEPPINGS(SKYLAKE_X,	X86_STEPPING_ANY,		MMIO | RETBLEED),
1278 	VULNBL_INTEL_STEPPINGS(SKYLAKE,		X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1279 	VULNBL_INTEL_STEPPINGS(KABYLAKE_L,	X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1280 	VULNBL_INTEL_STEPPINGS(KABYLAKE,	X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1281 	VULNBL_INTEL_STEPPINGS(CANNONLAKE_L,	X86_STEPPING_ANY,		RETBLEED),
1282 	VULNBL_INTEL_STEPPINGS(ICELAKE_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1283 	VULNBL_INTEL_STEPPINGS(ICELAKE_D,	X86_STEPPING_ANY,		MMIO),
1284 	VULNBL_INTEL_STEPPINGS(ICELAKE_X,	X86_STEPPING_ANY,		MMIO),
1285 	VULNBL_INTEL_STEPPINGS(COMETLAKE,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1286 	VULNBL_INTEL_STEPPINGS(COMETLAKE_L,	X86_STEPPINGS(0x0, 0x0),	MMIO | RETBLEED),
1287 	VULNBL_INTEL_STEPPINGS(COMETLAKE_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1288 	VULNBL_INTEL_STEPPINGS(LAKEFIELD,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1289 	VULNBL_INTEL_STEPPINGS(ROCKETLAKE,	X86_STEPPING_ANY,		MMIO | RETBLEED),
1290 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS),
1291 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_D,	X86_STEPPING_ANY,		MMIO),
1292 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS),
1293 
1294 	VULNBL_AMD(0x15, RETBLEED),
1295 	VULNBL_AMD(0x16, RETBLEED),
1296 	VULNBL_AMD(0x17, RETBLEED | SMT_RSB),
1297 	VULNBL_HYGON(0x18, RETBLEED | SMT_RSB),
1298 	{}
1299 };
1300 
1301 static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which)
1302 {
1303 	const struct x86_cpu_id *m = x86_match_cpu(table);
1304 
1305 	return m && !!(m->driver_data & which);
1306 }
1307 
1308 u64 x86_read_arch_cap_msr(void)
1309 {
1310 	u64 ia32_cap = 0;
1311 
1312 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1313 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1314 
1315 	return ia32_cap;
1316 }
1317 
1318 static bool arch_cap_mmio_immune(u64 ia32_cap)
1319 {
1320 	return (ia32_cap & ARCH_CAP_FBSDP_NO &&
1321 		ia32_cap & ARCH_CAP_PSDP_NO &&
1322 		ia32_cap & ARCH_CAP_SBDR_SSDP_NO);
1323 }
1324 
1325 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1326 {
1327 	u64 ia32_cap = x86_read_arch_cap_msr();
1328 
1329 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1330 	if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) &&
1331 	    !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO))
1332 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1333 
1334 	if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION))
1335 		return;
1336 
1337 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1338 
1339 	if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2))
1340 		setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1341 
1342 	if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) &&
1343 	    !(ia32_cap & ARCH_CAP_SSB_NO) &&
1344 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1345 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1346 
1347 	/*
1348 	 * AMD's AutoIBRS is equivalent to Intel's eIBRS - use the Intel feature
1349 	 * flag and protect from vendor-specific bugs via the whitelist.
1350 	 */
1351 	if ((ia32_cap & ARCH_CAP_IBRS_ALL) || cpu_has(c, X86_FEATURE_AUTOIBRS)) {
1352 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1353 		if (!cpu_matches(cpu_vuln_whitelist, NO_EIBRS_PBRSB) &&
1354 		    !(ia32_cap & ARCH_CAP_PBRSB_NO))
1355 			setup_force_cpu_bug(X86_BUG_EIBRS_PBRSB);
1356 	}
1357 
1358 	if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) &&
1359 	    !(ia32_cap & ARCH_CAP_MDS_NO)) {
1360 		setup_force_cpu_bug(X86_BUG_MDS);
1361 		if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY))
1362 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1363 	}
1364 
1365 	if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS))
1366 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1367 
1368 	/*
1369 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1370 	 *	- TSX is supported or
1371 	 *	- TSX_CTRL is present
1372 	 *
1373 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1374 	 * the kernel boot e.g. kexec.
1375 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1376 	 * update is not present or running as guest that don't get TSX_CTRL.
1377 	 */
1378 	if (!(ia32_cap & ARCH_CAP_TAA_NO) &&
1379 	    (cpu_has(c, X86_FEATURE_RTM) ||
1380 	     (ia32_cap & ARCH_CAP_TSX_CTRL_MSR)))
1381 		setup_force_cpu_bug(X86_BUG_TAA);
1382 
1383 	/*
1384 	 * SRBDS affects CPUs which support RDRAND or RDSEED and are listed
1385 	 * in the vulnerability blacklist.
1386 	 *
1387 	 * Some of the implications and mitigation of Shared Buffers Data
1388 	 * Sampling (SBDS) are similar to SRBDS. Give SBDS same treatment as
1389 	 * SRBDS.
1390 	 */
1391 	if ((cpu_has(c, X86_FEATURE_RDRAND) ||
1392 	     cpu_has(c, X86_FEATURE_RDSEED)) &&
1393 	    cpu_matches(cpu_vuln_blacklist, SRBDS | MMIO_SBDS))
1394 		    setup_force_cpu_bug(X86_BUG_SRBDS);
1395 
1396 	/*
1397 	 * Processor MMIO Stale Data bug enumeration
1398 	 *
1399 	 * Affected CPU list is generally enough to enumerate the vulnerability,
1400 	 * but for virtualization case check for ARCH_CAP MSR bits also, VMM may
1401 	 * not want the guest to enumerate the bug.
1402 	 *
1403 	 * Set X86_BUG_MMIO_UNKNOWN for CPUs that are neither in the blacklist,
1404 	 * nor in the whitelist and also don't enumerate MSR ARCH_CAP MMIO bits.
1405 	 */
1406 	if (!arch_cap_mmio_immune(ia32_cap)) {
1407 		if (cpu_matches(cpu_vuln_blacklist, MMIO))
1408 			setup_force_cpu_bug(X86_BUG_MMIO_STALE_DATA);
1409 		else if (!cpu_matches(cpu_vuln_whitelist, NO_MMIO))
1410 			setup_force_cpu_bug(X86_BUG_MMIO_UNKNOWN);
1411 	}
1412 
1413 	if (!cpu_has(c, X86_FEATURE_BTC_NO)) {
1414 		if (cpu_matches(cpu_vuln_blacklist, RETBLEED) || (ia32_cap & ARCH_CAP_RSBA))
1415 			setup_force_cpu_bug(X86_BUG_RETBLEED);
1416 	}
1417 
1418 	if (cpu_matches(cpu_vuln_blacklist, SMT_RSB))
1419 		setup_force_cpu_bug(X86_BUG_SMT_RSB);
1420 
1421 	if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN))
1422 		return;
1423 
1424 	/* Rogue Data Cache Load? No! */
1425 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1426 		return;
1427 
1428 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1429 
1430 	if (cpu_matches(cpu_vuln_whitelist, NO_L1TF))
1431 		return;
1432 
1433 	setup_force_cpu_bug(X86_BUG_L1TF);
1434 }
1435 
1436 /*
1437  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1438  * unfortunately, that's not true in practice because of early VIA
1439  * chips and (more importantly) broken virtualizers that are not easy
1440  * to detect. In the latter case it doesn't even *fail* reliably, so
1441  * probing for it doesn't even work. Disable it completely on 32-bit
1442  * unless we can find a reliable way to detect all the broken cases.
1443  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1444  */
1445 static void detect_nopl(void)
1446 {
1447 #ifdef CONFIG_X86_32
1448 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1449 #else
1450 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1451 #endif
1452 }
1453 
1454 /*
1455  * We parse cpu parameters early because fpu__init_system() is executed
1456  * before parse_early_param().
1457  */
1458 static void __init cpu_parse_early_param(void)
1459 {
1460 	char arg[128];
1461 	char *argptr = arg, *opt;
1462 	int arglen, taint = 0;
1463 
1464 #ifdef CONFIG_X86_32
1465 	if (cmdline_find_option_bool(boot_command_line, "no387"))
1466 #ifdef CONFIG_MATH_EMULATION
1467 		setup_clear_cpu_cap(X86_FEATURE_FPU);
1468 #else
1469 		pr_err("Option 'no387' required CONFIG_MATH_EMULATION enabled.\n");
1470 #endif
1471 
1472 	if (cmdline_find_option_bool(boot_command_line, "nofxsr"))
1473 		setup_clear_cpu_cap(X86_FEATURE_FXSR);
1474 #endif
1475 
1476 	if (cmdline_find_option_bool(boot_command_line, "noxsave"))
1477 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
1478 
1479 	if (cmdline_find_option_bool(boot_command_line, "noxsaveopt"))
1480 		setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
1481 
1482 	if (cmdline_find_option_bool(boot_command_line, "noxsaves"))
1483 		setup_clear_cpu_cap(X86_FEATURE_XSAVES);
1484 
1485 	arglen = cmdline_find_option(boot_command_line, "clearcpuid", arg, sizeof(arg));
1486 	if (arglen <= 0)
1487 		return;
1488 
1489 	pr_info("Clearing CPUID bits:");
1490 
1491 	while (argptr) {
1492 		bool found __maybe_unused = false;
1493 		unsigned int bit;
1494 
1495 		opt = strsep(&argptr, ",");
1496 
1497 		/*
1498 		 * Handle naked numbers first for feature flags which don't
1499 		 * have names.
1500 		 */
1501 		if (!kstrtouint(opt, 10, &bit)) {
1502 			if (bit < NCAPINTS * 32) {
1503 
1504 #ifdef CONFIG_X86_FEATURE_NAMES
1505 				/* empty-string, i.e., ""-defined feature flags */
1506 				if (!x86_cap_flags[bit])
1507 					pr_cont(" " X86_CAP_FMT_NUM, x86_cap_flag_num(bit));
1508 				else
1509 #endif
1510 					pr_cont(" " X86_CAP_FMT, x86_cap_flag(bit));
1511 
1512 				setup_clear_cpu_cap(bit);
1513 				taint++;
1514 			}
1515 			/*
1516 			 * The assumption is that there are no feature names with only
1517 			 * numbers in the name thus go to the next argument.
1518 			 */
1519 			continue;
1520 		}
1521 
1522 #ifdef CONFIG_X86_FEATURE_NAMES
1523 		for (bit = 0; bit < 32 * NCAPINTS; bit++) {
1524 			if (!x86_cap_flag(bit))
1525 				continue;
1526 
1527 			if (strcmp(x86_cap_flag(bit), opt))
1528 				continue;
1529 
1530 			pr_cont(" %s", opt);
1531 			setup_clear_cpu_cap(bit);
1532 			taint++;
1533 			found = true;
1534 			break;
1535 		}
1536 
1537 		if (!found)
1538 			pr_cont(" (unknown: %s)", opt);
1539 #endif
1540 	}
1541 	pr_cont("\n");
1542 
1543 	if (taint)
1544 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_STILL_OK);
1545 }
1546 
1547 /*
1548  * Do minimum CPU detection early.
1549  * Fields really needed: vendor, cpuid_level, family, model, mask,
1550  * cache alignment.
1551  * The others are not touched to avoid unwanted side effects.
1552  *
1553  * WARNING: this function is only called on the boot CPU.  Don't add code
1554  * here that is supposed to run on all CPUs.
1555  */
1556 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1557 {
1558 #ifdef CONFIG_X86_64
1559 	c->x86_clflush_size = 64;
1560 	c->x86_phys_bits = 36;
1561 	c->x86_virt_bits = 48;
1562 #else
1563 	c->x86_clflush_size = 32;
1564 	c->x86_phys_bits = 32;
1565 	c->x86_virt_bits = 32;
1566 #endif
1567 	c->x86_cache_alignment = c->x86_clflush_size;
1568 
1569 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1570 	c->extended_cpuid_level = 0;
1571 
1572 	if (!have_cpuid_p())
1573 		identify_cpu_without_cpuid(c);
1574 
1575 	/* cyrix could have cpuid enabled via c_identify()*/
1576 	if (have_cpuid_p()) {
1577 		cpu_detect(c);
1578 		get_cpu_vendor(c);
1579 		get_cpu_cap(c);
1580 		get_cpu_address_sizes(c);
1581 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1582 		cpu_parse_early_param();
1583 
1584 		if (this_cpu->c_early_init)
1585 			this_cpu->c_early_init(c);
1586 
1587 		c->cpu_index = 0;
1588 		filter_cpuid_features(c, false);
1589 
1590 		if (this_cpu->c_bsp_init)
1591 			this_cpu->c_bsp_init(c);
1592 	} else {
1593 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1594 	}
1595 
1596 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1597 
1598 	cpu_set_bug_bits(c);
1599 
1600 	sld_setup(c);
1601 
1602 	fpu__init_system(c);
1603 
1604 	init_sigframe_size();
1605 
1606 #ifdef CONFIG_X86_32
1607 	/*
1608 	 * Regardless of whether PCID is enumerated, the SDM says
1609 	 * that it can't be enabled in 32-bit mode.
1610 	 */
1611 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1612 #endif
1613 
1614 	/*
1615 	 * Later in the boot process pgtable_l5_enabled() relies on
1616 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1617 	 * enabled by this point we need to clear the feature bit to avoid
1618 	 * false-positives at the later stage.
1619 	 *
1620 	 * pgtable_l5_enabled() can be false here for several reasons:
1621 	 *  - 5-level paging is disabled compile-time;
1622 	 *  - it's 32-bit kernel;
1623 	 *  - machine doesn't support 5-level paging;
1624 	 *  - user specified 'no5lvl' in kernel command line.
1625 	 */
1626 	if (!pgtable_l5_enabled())
1627 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1628 
1629 	detect_nopl();
1630 }
1631 
1632 void __init early_cpu_init(void)
1633 {
1634 	const struct cpu_dev *const *cdev;
1635 	int count = 0;
1636 
1637 #ifdef CONFIG_PROCESSOR_SELECT
1638 	pr_info("KERNEL supported cpus:\n");
1639 #endif
1640 
1641 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1642 		const struct cpu_dev *cpudev = *cdev;
1643 
1644 		if (count >= X86_VENDOR_NUM)
1645 			break;
1646 		cpu_devs[count] = cpudev;
1647 		count++;
1648 
1649 #ifdef CONFIG_PROCESSOR_SELECT
1650 		{
1651 			unsigned int j;
1652 
1653 			for (j = 0; j < 2; j++) {
1654 				if (!cpudev->c_ident[j])
1655 					continue;
1656 				pr_info("  %s %s\n", cpudev->c_vendor,
1657 					cpudev->c_ident[j]);
1658 			}
1659 		}
1660 #endif
1661 	}
1662 	early_identify_cpu(&boot_cpu_data);
1663 }
1664 
1665 static bool detect_null_seg_behavior(void)
1666 {
1667 	/*
1668 	 * Empirically, writing zero to a segment selector on AMD does
1669 	 * not clear the base, whereas writing zero to a segment
1670 	 * selector on Intel does clear the base.  Intel's behavior
1671 	 * allows slightly faster context switches in the common case
1672 	 * where GS is unused by the prev and next threads.
1673 	 *
1674 	 * Since neither vendor documents this anywhere that I can see,
1675 	 * detect it directly instead of hard-coding the choice by
1676 	 * vendor.
1677 	 *
1678 	 * I've designated AMD's behavior as the "bug" because it's
1679 	 * counterintuitive and less friendly.
1680 	 */
1681 
1682 	unsigned long old_base, tmp;
1683 	rdmsrl(MSR_FS_BASE, old_base);
1684 	wrmsrl(MSR_FS_BASE, 1);
1685 	loadsegment(fs, 0);
1686 	rdmsrl(MSR_FS_BASE, tmp);
1687 	wrmsrl(MSR_FS_BASE, old_base);
1688 	return tmp == 0;
1689 }
1690 
1691 void check_null_seg_clears_base(struct cpuinfo_x86 *c)
1692 {
1693 	/* BUG_NULL_SEG is only relevant with 64bit userspace */
1694 	if (!IS_ENABLED(CONFIG_X86_64))
1695 		return;
1696 
1697 	if (cpu_has(c, X86_FEATURE_NULL_SEL_CLR_BASE))
1698 		return;
1699 
1700 	/*
1701 	 * CPUID bit above wasn't set. If this kernel is still running
1702 	 * as a HV guest, then the HV has decided not to advertize
1703 	 * that CPUID bit for whatever reason.	For example, one
1704 	 * member of the migration pool might be vulnerable.  Which
1705 	 * means, the bug is present: set the BUG flag and return.
1706 	 */
1707 	if (cpu_has(c, X86_FEATURE_HYPERVISOR)) {
1708 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1709 		return;
1710 	}
1711 
1712 	/*
1713 	 * Zen2 CPUs also have this behaviour, but no CPUID bit.
1714 	 * 0x18 is the respective family for Hygon.
1715 	 */
1716 	if ((c->x86 == 0x17 || c->x86 == 0x18) &&
1717 	    detect_null_seg_behavior())
1718 		return;
1719 
1720 	/* All the remaining ones are affected */
1721 	set_cpu_bug(c, X86_BUG_NULL_SEG);
1722 }
1723 
1724 static void generic_identify(struct cpuinfo_x86 *c)
1725 {
1726 	c->extended_cpuid_level = 0;
1727 
1728 	if (!have_cpuid_p())
1729 		identify_cpu_without_cpuid(c);
1730 
1731 	/* cyrix could have cpuid enabled via c_identify()*/
1732 	if (!have_cpuid_p())
1733 		return;
1734 
1735 	cpu_detect(c);
1736 
1737 	get_cpu_vendor(c);
1738 
1739 	get_cpu_cap(c);
1740 
1741 	get_cpu_address_sizes(c);
1742 
1743 	if (c->cpuid_level >= 0x00000001) {
1744 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1745 #ifdef CONFIG_X86_32
1746 # ifdef CONFIG_SMP
1747 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1748 # else
1749 		c->apicid = c->initial_apicid;
1750 # endif
1751 #endif
1752 		c->phys_proc_id = c->initial_apicid;
1753 	}
1754 
1755 	get_model_name(c); /* Default name */
1756 
1757 	/*
1758 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1759 	 * systems that run Linux at CPL > 0 may or may not have the
1760 	 * issue, but, even if they have the issue, there's absolutely
1761 	 * nothing we can do about it because we can't use the real IRET
1762 	 * instruction.
1763 	 *
1764 	 * NB: For the time being, only 32-bit kernels support
1765 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1766 	 * whether to apply espfix using paravirt hooks.  If any
1767 	 * non-paravirt system ever shows up that does *not* have the
1768 	 * ESPFIX issue, we can change this.
1769 	 */
1770 #ifdef CONFIG_X86_32
1771 	set_cpu_bug(c, X86_BUG_ESPFIX);
1772 #endif
1773 }
1774 
1775 /*
1776  * Validate that ACPI/mptables have the same information about the
1777  * effective APIC id and update the package map.
1778  */
1779 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1780 {
1781 #ifdef CONFIG_SMP
1782 	unsigned int apicid, cpu = smp_processor_id();
1783 
1784 	apicid = apic->cpu_present_to_apicid(cpu);
1785 
1786 	if (apicid != c->apicid) {
1787 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1788 		       cpu, apicid, c->initial_apicid);
1789 	}
1790 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1791 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1792 #else
1793 	c->logical_proc_id = 0;
1794 #endif
1795 }
1796 
1797 /*
1798  * This does the hard work of actually picking apart the CPU stuff...
1799  */
1800 static void identify_cpu(struct cpuinfo_x86 *c)
1801 {
1802 	int i;
1803 
1804 	c->loops_per_jiffy = loops_per_jiffy;
1805 	c->x86_cache_size = 0;
1806 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1807 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1808 	c->x86_vendor_id[0] = '\0'; /* Unset */
1809 	c->x86_model_id[0] = '\0';  /* Unset */
1810 	c->x86_max_cores = 1;
1811 	c->x86_coreid_bits = 0;
1812 	c->cu_id = 0xff;
1813 #ifdef CONFIG_X86_64
1814 	c->x86_clflush_size = 64;
1815 	c->x86_phys_bits = 36;
1816 	c->x86_virt_bits = 48;
1817 #else
1818 	c->cpuid_level = -1;	/* CPUID not detected */
1819 	c->x86_clflush_size = 32;
1820 	c->x86_phys_bits = 32;
1821 	c->x86_virt_bits = 32;
1822 #endif
1823 	c->x86_cache_alignment = c->x86_clflush_size;
1824 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1825 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
1826 	memset(&c->vmx_capability, 0, sizeof(c->vmx_capability));
1827 #endif
1828 
1829 	generic_identify(c);
1830 
1831 	if (this_cpu->c_identify)
1832 		this_cpu->c_identify(c);
1833 
1834 	/* Clear/Set all flags overridden by options, after probe */
1835 	apply_forced_caps(c);
1836 
1837 #ifdef CONFIG_X86_64
1838 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1839 #endif
1840 
1841 	/*
1842 	 * Vendor-specific initialization.  In this section we
1843 	 * canonicalize the feature flags, meaning if there are
1844 	 * features a certain CPU supports which CPUID doesn't
1845 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1846 	 * we handle them here.
1847 	 *
1848 	 * At the end of this section, c->x86_capability better
1849 	 * indicate the features this CPU genuinely supports!
1850 	 */
1851 	if (this_cpu->c_init)
1852 		this_cpu->c_init(c);
1853 
1854 	/* Disable the PN if appropriate */
1855 	squash_the_stupid_serial_number(c);
1856 
1857 	/* Set up SMEP/SMAP/UMIP */
1858 	setup_smep(c);
1859 	setup_smap(c);
1860 	setup_umip(c);
1861 
1862 	/* Enable FSGSBASE instructions if available. */
1863 	if (cpu_has(c, X86_FEATURE_FSGSBASE)) {
1864 		cr4_set_bits(X86_CR4_FSGSBASE);
1865 		elf_hwcap2 |= HWCAP2_FSGSBASE;
1866 	}
1867 
1868 	/*
1869 	 * The vendor-specific functions might have changed features.
1870 	 * Now we do "generic changes."
1871 	 */
1872 
1873 	/* Filter out anything that depends on CPUID levels we don't have */
1874 	filter_cpuid_features(c, true);
1875 
1876 	/* If the model name is still unset, do table lookup. */
1877 	if (!c->x86_model_id[0]) {
1878 		const char *p;
1879 		p = table_lookup_model(c);
1880 		if (p)
1881 			strcpy(c->x86_model_id, p);
1882 		else
1883 			/* Last resort... */
1884 			sprintf(c->x86_model_id, "%02x/%02x",
1885 				c->x86, c->x86_model);
1886 	}
1887 
1888 #ifdef CONFIG_X86_64
1889 	detect_ht(c);
1890 #endif
1891 
1892 	x86_init_rdrand(c);
1893 	setup_pku(c);
1894 	setup_cet(c);
1895 
1896 	/*
1897 	 * Clear/Set all flags overridden by options, need do it
1898 	 * before following smp all cpus cap AND.
1899 	 */
1900 	apply_forced_caps(c);
1901 
1902 	/*
1903 	 * On SMP, boot_cpu_data holds the common feature set between
1904 	 * all CPUs; so make sure that we indicate which features are
1905 	 * common between the CPUs.  The first time this routine gets
1906 	 * executed, c == &boot_cpu_data.
1907 	 */
1908 	if (c != &boot_cpu_data) {
1909 		/* AND the already accumulated flags with these */
1910 		for (i = 0; i < NCAPINTS; i++)
1911 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1912 
1913 		/* OR, i.e. replicate the bug flags */
1914 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1915 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1916 	}
1917 
1918 	ppin_init(c);
1919 
1920 	/* Init Machine Check Exception if available. */
1921 	mcheck_cpu_init(c);
1922 
1923 	select_idle_routine(c);
1924 
1925 #ifdef CONFIG_NUMA
1926 	numa_add_cpu(smp_processor_id());
1927 #endif
1928 }
1929 
1930 /*
1931  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1932  * on 32-bit kernels:
1933  */
1934 #ifdef CONFIG_X86_32
1935 void enable_sep_cpu(void)
1936 {
1937 	struct tss_struct *tss;
1938 	int cpu;
1939 
1940 	if (!boot_cpu_has(X86_FEATURE_SEP))
1941 		return;
1942 
1943 	cpu = get_cpu();
1944 	tss = &per_cpu(cpu_tss_rw, cpu);
1945 
1946 	/*
1947 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1948 	 * see the big comment in struct x86_hw_tss's definition.
1949 	 */
1950 
1951 	tss->x86_tss.ss1 = __KERNEL_CS;
1952 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1953 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1954 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1955 
1956 	put_cpu();
1957 }
1958 #endif
1959 
1960 void __init identify_boot_cpu(void)
1961 {
1962 	identify_cpu(&boot_cpu_data);
1963 	if (HAS_KERNEL_IBT && cpu_feature_enabled(X86_FEATURE_IBT))
1964 		pr_info("CET detected: Indirect Branch Tracking enabled\n");
1965 #ifdef CONFIG_X86_32
1966 	enable_sep_cpu();
1967 #endif
1968 	cpu_detect_tlb(&boot_cpu_data);
1969 	setup_cr_pinning();
1970 
1971 	tsx_init();
1972 	lkgs_init();
1973 }
1974 
1975 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1976 {
1977 	BUG_ON(c == &boot_cpu_data);
1978 	identify_cpu(c);
1979 #ifdef CONFIG_X86_32
1980 	enable_sep_cpu();
1981 #endif
1982 	validate_apic_and_package_id(c);
1983 	x86_spec_ctrl_setup_ap();
1984 	update_srbds_msr();
1985 
1986 	tsx_ap_init();
1987 }
1988 
1989 void print_cpu_info(struct cpuinfo_x86 *c)
1990 {
1991 	const char *vendor = NULL;
1992 
1993 	if (c->x86_vendor < X86_VENDOR_NUM) {
1994 		vendor = this_cpu->c_vendor;
1995 	} else {
1996 		if (c->cpuid_level >= 0)
1997 			vendor = c->x86_vendor_id;
1998 	}
1999 
2000 	if (vendor && !strstr(c->x86_model_id, vendor))
2001 		pr_cont("%s ", vendor);
2002 
2003 	if (c->x86_model_id[0])
2004 		pr_cont("%s", c->x86_model_id);
2005 	else
2006 		pr_cont("%d86", c->x86);
2007 
2008 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
2009 
2010 	if (c->x86_stepping || c->cpuid_level >= 0)
2011 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
2012 	else
2013 		pr_cont(")\n");
2014 }
2015 
2016 /*
2017  * clearcpuid= was already parsed in cpu_parse_early_param().  This dummy
2018  * function prevents it from becoming an environment variable for init.
2019  */
2020 static __init int setup_clearcpuid(char *arg)
2021 {
2022 	return 1;
2023 }
2024 __setup("clearcpuid=", setup_clearcpuid);
2025 
2026 DEFINE_PER_CPU_ALIGNED(struct pcpu_hot, pcpu_hot) = {
2027 	.current_task	= &init_task,
2028 	.preempt_count	= INIT_PREEMPT_COUNT,
2029 	.top_of_stack	= TOP_OF_INIT_STACK,
2030 };
2031 EXPORT_PER_CPU_SYMBOL(pcpu_hot);
2032 
2033 #ifdef CONFIG_X86_64
2034 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
2035 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
2036 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
2037 
2038 static void wrmsrl_cstar(unsigned long val)
2039 {
2040 	/*
2041 	 * Intel CPUs do not support 32-bit SYSCALL. Writing to MSR_CSTAR
2042 	 * is so far ignored by the CPU, but raises a #VE trap in a TDX
2043 	 * guest. Avoid the pointless write on all Intel CPUs.
2044 	 */
2045 	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
2046 		wrmsrl(MSR_CSTAR, val);
2047 }
2048 
2049 /* May not be marked __init: used by software suspend */
2050 void syscall_init(void)
2051 {
2052 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
2053 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
2054 
2055 #ifdef CONFIG_IA32_EMULATION
2056 	wrmsrl_cstar((unsigned long)entry_SYSCALL_compat);
2057 	/*
2058 	 * This only works on Intel CPUs.
2059 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
2060 	 * This does not cause SYSENTER to jump to the wrong location, because
2061 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
2062 	 */
2063 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
2064 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
2065 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
2066 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
2067 #else
2068 	wrmsrl_cstar((unsigned long)ignore_sysret);
2069 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
2070 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
2071 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
2072 #endif
2073 
2074 	/*
2075 	 * Flags to clear on syscall; clear as much as possible
2076 	 * to minimize user space-kernel interference.
2077 	 */
2078 	wrmsrl(MSR_SYSCALL_MASK,
2079 	       X86_EFLAGS_CF|X86_EFLAGS_PF|X86_EFLAGS_AF|
2080 	       X86_EFLAGS_ZF|X86_EFLAGS_SF|X86_EFLAGS_TF|
2081 	       X86_EFLAGS_IF|X86_EFLAGS_DF|X86_EFLAGS_OF|
2082 	       X86_EFLAGS_IOPL|X86_EFLAGS_NT|X86_EFLAGS_RF|
2083 	       X86_EFLAGS_AC|X86_EFLAGS_ID);
2084 }
2085 
2086 #else	/* CONFIG_X86_64 */
2087 
2088 #ifdef CONFIG_STACKPROTECTOR
2089 DEFINE_PER_CPU(unsigned long, __stack_chk_guard);
2090 EXPORT_PER_CPU_SYMBOL(__stack_chk_guard);
2091 #endif
2092 
2093 #endif	/* CONFIG_X86_64 */
2094 
2095 /*
2096  * Clear all 6 debug registers:
2097  */
2098 static void clear_all_debug_regs(void)
2099 {
2100 	int i;
2101 
2102 	for (i = 0; i < 8; i++) {
2103 		/* Ignore db4, db5 */
2104 		if ((i == 4) || (i == 5))
2105 			continue;
2106 
2107 		set_debugreg(0, i);
2108 	}
2109 }
2110 
2111 #ifdef CONFIG_KGDB
2112 /*
2113  * Restore debug regs if using kgdbwait and you have a kernel debugger
2114  * connection established.
2115  */
2116 static void dbg_restore_debug_regs(void)
2117 {
2118 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
2119 		arch_kgdb_ops.correct_hw_break();
2120 }
2121 #else /* ! CONFIG_KGDB */
2122 #define dbg_restore_debug_regs()
2123 #endif /* ! CONFIG_KGDB */
2124 
2125 static void wait_for_master_cpu(int cpu)
2126 {
2127 #ifdef CONFIG_SMP
2128 	/*
2129 	 * wait for ACK from master CPU before continuing
2130 	 * with AP initialization
2131 	 */
2132 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
2133 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
2134 		cpu_relax();
2135 #endif
2136 }
2137 
2138 static inline void setup_getcpu(int cpu)
2139 {
2140 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
2141 	struct desc_struct d = { };
2142 
2143 	if (boot_cpu_has(X86_FEATURE_RDTSCP) || boot_cpu_has(X86_FEATURE_RDPID))
2144 		wrmsr(MSR_TSC_AUX, cpudata, 0);
2145 
2146 	/* Store CPU and node number in limit. */
2147 	d.limit0 = cpudata;
2148 	d.limit1 = cpudata >> 16;
2149 
2150 	d.type = 5;		/* RO data, expand down, accessed */
2151 	d.dpl = 3;		/* Visible to user code */
2152 	d.s = 1;		/* Not a system segment */
2153 	d.p = 1;		/* Present */
2154 	d.d = 1;		/* 32-bit */
2155 
2156 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
2157 }
2158 
2159 #ifdef CONFIG_X86_64
2160 static inline void ucode_cpu_init(int cpu)
2161 {
2162 	if (cpu)
2163 		load_ucode_ap();
2164 }
2165 
2166 static inline void tss_setup_ist(struct tss_struct *tss)
2167 {
2168 	/* Set up the per-CPU TSS IST stacks */
2169 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
2170 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
2171 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
2172 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
2173 	/* Only mapped when SEV-ES is active */
2174 	tss->x86_tss.ist[IST_INDEX_VC] = __this_cpu_ist_top_va(VC);
2175 }
2176 
2177 #else /* CONFIG_X86_64 */
2178 
2179 static inline void ucode_cpu_init(int cpu)
2180 {
2181 	show_ucode_info_early();
2182 }
2183 
2184 static inline void tss_setup_ist(struct tss_struct *tss) { }
2185 
2186 #endif /* !CONFIG_X86_64 */
2187 
2188 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
2189 {
2190 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
2191 
2192 #ifdef CONFIG_X86_IOPL_IOPERM
2193 	tss->io_bitmap.prev_max = 0;
2194 	tss->io_bitmap.prev_sequence = 0;
2195 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
2196 	/*
2197 	 * Invalidate the extra array entry past the end of the all
2198 	 * permission bitmap as required by the hardware.
2199 	 */
2200 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
2201 #endif
2202 }
2203 
2204 /*
2205  * Setup everything needed to handle exceptions from the IDT, including the IST
2206  * exceptions which use paranoid_entry().
2207  */
2208 void cpu_init_exception_handling(void)
2209 {
2210 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
2211 	int cpu = raw_smp_processor_id();
2212 
2213 	/* paranoid_entry() gets the CPU number from the GDT */
2214 	setup_getcpu(cpu);
2215 
2216 	/* IST vectors need TSS to be set up. */
2217 	tss_setup_ist(tss);
2218 	tss_setup_io_bitmap(tss);
2219 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
2220 
2221 	load_TR_desc();
2222 
2223 	/* GHCB needs to be setup to handle #VC. */
2224 	setup_ghcb();
2225 
2226 	/* Finally load the IDT */
2227 	load_current_idt();
2228 }
2229 
2230 /*
2231  * cpu_init() initializes state that is per-CPU. Some data is already
2232  * initialized (naturally) in the bootstrap process, such as the GDT.  We
2233  * reload it nevertheless, this function acts as a 'CPU state barrier',
2234  * nothing should get across.
2235  */
2236 void cpu_init(void)
2237 {
2238 	struct task_struct *cur = current;
2239 	int cpu = raw_smp_processor_id();
2240 
2241 	wait_for_master_cpu(cpu);
2242 
2243 	ucode_cpu_init(cpu);
2244 
2245 #ifdef CONFIG_NUMA
2246 	if (this_cpu_read(numa_node) == 0 &&
2247 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
2248 		set_numa_node(early_cpu_to_node(cpu));
2249 #endif
2250 	pr_debug("Initializing CPU#%d\n", cpu);
2251 
2252 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
2253 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
2254 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
2255 
2256 	if (IS_ENABLED(CONFIG_X86_64)) {
2257 		loadsegment(fs, 0);
2258 		memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
2259 		syscall_init();
2260 
2261 		wrmsrl(MSR_FS_BASE, 0);
2262 		wrmsrl(MSR_KERNEL_GS_BASE, 0);
2263 		barrier();
2264 
2265 		x2apic_setup();
2266 	}
2267 
2268 	mmgrab(&init_mm);
2269 	cur->active_mm = &init_mm;
2270 	BUG_ON(cur->mm);
2271 	initialize_tlbstate_and_flush();
2272 	enter_lazy_tlb(&init_mm, cur);
2273 
2274 	/*
2275 	 * sp0 points to the entry trampoline stack regardless of what task
2276 	 * is running.
2277 	 */
2278 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
2279 
2280 	load_mm_ldt(&init_mm);
2281 
2282 	clear_all_debug_regs();
2283 	dbg_restore_debug_regs();
2284 
2285 	doublefault_init_cpu_tss();
2286 
2287 	fpu__init_cpu();
2288 
2289 	if (is_uv_system())
2290 		uv_cpu_init();
2291 
2292 	load_fixmap_gdt(cpu);
2293 }
2294 
2295 #ifdef CONFIG_SMP
2296 void cpu_init_secondary(void)
2297 {
2298 	/*
2299 	 * Relies on the BP having set-up the IDT tables, which are loaded
2300 	 * on this CPU in cpu_init_exception_handling().
2301 	 */
2302 	cpu_init_exception_handling();
2303 	cpu_init();
2304 }
2305 #endif
2306 
2307 #ifdef CONFIG_MICROCODE_LATE_LOADING
2308 /**
2309  * store_cpu_caps() - Store a snapshot of CPU capabilities
2310  * @curr_info: Pointer where to store it
2311  *
2312  * Returns: None
2313  */
2314 void store_cpu_caps(struct cpuinfo_x86 *curr_info)
2315 {
2316 	/* Reload CPUID max function as it might've changed. */
2317 	curr_info->cpuid_level = cpuid_eax(0);
2318 
2319 	/* Copy all capability leafs and pick up the synthetic ones. */
2320 	memcpy(&curr_info->x86_capability, &boot_cpu_data.x86_capability,
2321 	       sizeof(curr_info->x86_capability));
2322 
2323 	/* Get the hardware CPUID leafs */
2324 	get_cpu_cap(curr_info);
2325 }
2326 
2327 /**
2328  * microcode_check() - Check if any CPU capabilities changed after an update.
2329  * @prev_info:	CPU capabilities stored before an update.
2330  *
2331  * The microcode loader calls this upon late microcode load to recheck features,
2332  * only when microcode has been updated. Caller holds microcode_mutex and CPU
2333  * hotplug lock.
2334  *
2335  * Return: None
2336  */
2337 void microcode_check(struct cpuinfo_x86 *prev_info)
2338 {
2339 	struct cpuinfo_x86 curr_info;
2340 
2341 	perf_check_microcode();
2342 
2343 	store_cpu_caps(&curr_info);
2344 
2345 	if (!memcmp(&prev_info->x86_capability, &curr_info.x86_capability,
2346 		    sizeof(prev_info->x86_capability)))
2347 		return;
2348 
2349 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
2350 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
2351 }
2352 #endif
2353 
2354 /*
2355  * Invoked from core CPU hotplug code after hotplug operations
2356  */
2357 void arch_smt_update(void)
2358 {
2359 	/* Handle the speculative execution misfeatures */
2360 	cpu_bugs_smt_update();
2361 	/* Check whether IPI broadcasting can be enabled */
2362 	apic_smt_update();
2363 }
2364