xref: /linux/arch/x86/kernel/cpu/common.c (revision 2fc60e2ff972d3dca836bff0b08cbe503c4ca1ce)
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 /* CONFIG_X86_64 */
571 
572 #ifdef CONFIG_X86_KERNEL_IBT
573 
574 __noendbr u64 ibt_save(void)
575 {
576 	u64 msr = 0;
577 
578 	if (cpu_feature_enabled(X86_FEATURE_IBT)) {
579 		rdmsrl(MSR_IA32_S_CET, msr);
580 		wrmsrl(MSR_IA32_S_CET, msr & ~CET_ENDBR_EN);
581 	}
582 
583 	return msr;
584 }
585 
586 __noendbr void ibt_restore(u64 save)
587 {
588 	u64 msr;
589 
590 	if (cpu_feature_enabled(X86_FEATURE_IBT)) {
591 		rdmsrl(MSR_IA32_S_CET, msr);
592 		msr &= ~CET_ENDBR_EN;
593 		msr |= (save & CET_ENDBR_EN);
594 		wrmsrl(MSR_IA32_S_CET, msr);
595 	}
596 }
597 
598 #endif
599 
600 static __always_inline void setup_cet(struct cpuinfo_x86 *c)
601 {
602 	u64 msr = CET_ENDBR_EN;
603 
604 	if (!HAS_KERNEL_IBT ||
605 	    !cpu_feature_enabled(X86_FEATURE_IBT))
606 		return;
607 
608 	wrmsrl(MSR_IA32_S_CET, msr);
609 	cr4_set_bits(X86_CR4_CET);
610 
611 	if (!ibt_selftest()) {
612 		pr_err("IBT selftest: Failed!\n");
613 		setup_clear_cpu_cap(X86_FEATURE_IBT);
614 		return;
615 	}
616 }
617 
618 __noendbr void cet_disable(void)
619 {
620 	if (cpu_feature_enabled(X86_FEATURE_IBT))
621 		wrmsrl(MSR_IA32_S_CET, 0);
622 }
623 
624 /*
625  * Some CPU features depend on higher CPUID levels, which may not always
626  * be available due to CPUID level capping or broken virtualization
627  * software.  Add those features to this table to auto-disable them.
628  */
629 struct cpuid_dependent_feature {
630 	u32 feature;
631 	u32 level;
632 };
633 
634 static const struct cpuid_dependent_feature
635 cpuid_dependent_features[] = {
636 	{ X86_FEATURE_MWAIT,		0x00000005 },
637 	{ X86_FEATURE_DCA,		0x00000009 },
638 	{ X86_FEATURE_XSAVE,		0x0000000d },
639 	{ 0, 0 }
640 };
641 
642 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
643 {
644 	const struct cpuid_dependent_feature *df;
645 
646 	for (df = cpuid_dependent_features; df->feature; df++) {
647 
648 		if (!cpu_has(c, df->feature))
649 			continue;
650 		/*
651 		 * Note: cpuid_level is set to -1 if unavailable, but
652 		 * extended_extended_level is set to 0 if unavailable
653 		 * and the legitimate extended levels are all negative
654 		 * when signed; hence the weird messing around with
655 		 * signs here...
656 		 */
657 		if (!((s32)df->level < 0 ?
658 		     (u32)df->level > (u32)c->extended_cpuid_level :
659 		     (s32)df->level > (s32)c->cpuid_level))
660 			continue;
661 
662 		clear_cpu_cap(c, df->feature);
663 		if (!warn)
664 			continue;
665 
666 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
667 			x86_cap_flag(df->feature), df->level);
668 	}
669 }
670 
671 /*
672  * Naming convention should be: <Name> [(<Codename>)]
673  * This table only is used unless init_<vendor>() below doesn't set it;
674  * in particular, if CPUID levels 0x80000002..4 are supported, this
675  * isn't used
676  */
677 
678 /* Look up CPU names by table lookup. */
679 static const char *table_lookup_model(struct cpuinfo_x86 *c)
680 {
681 #ifdef CONFIG_X86_32
682 	const struct legacy_cpu_model_info *info;
683 
684 	if (c->x86_model >= 16)
685 		return NULL;	/* Range check */
686 
687 	if (!this_cpu)
688 		return NULL;
689 
690 	info = this_cpu->legacy_models;
691 
692 	while (info->family) {
693 		if (info->family == c->x86)
694 			return info->model_names[c->x86_model];
695 		info++;
696 	}
697 #endif
698 	return NULL;		/* Not found */
699 }
700 
701 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
702 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
703 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
704 
705 void load_percpu_segment(int cpu)
706 {
707 #ifdef CONFIG_X86_32
708 	loadsegment(fs, __KERNEL_PERCPU);
709 #else
710 	__loadsegment_simple(gs, 0);
711 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
712 #endif
713 }
714 
715 #ifdef CONFIG_X86_32
716 /* The 32-bit entry code needs to find cpu_entry_area. */
717 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
718 #endif
719 
720 /* Load the original GDT from the per-cpu structure */
721 void load_direct_gdt(int cpu)
722 {
723 	struct desc_ptr gdt_descr;
724 
725 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
726 	gdt_descr.size = GDT_SIZE - 1;
727 	load_gdt(&gdt_descr);
728 }
729 EXPORT_SYMBOL_GPL(load_direct_gdt);
730 
731 /* Load a fixmap remapping of the per-cpu GDT */
732 void load_fixmap_gdt(int cpu)
733 {
734 	struct desc_ptr gdt_descr;
735 
736 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
737 	gdt_descr.size = GDT_SIZE - 1;
738 	load_gdt(&gdt_descr);
739 }
740 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
741 
742 /*
743  * Current gdt points %fs at the "master" per-cpu area: after this,
744  * it's on the real one.
745  */
746 void switch_to_new_gdt(int cpu)
747 {
748 	/* Load the original GDT */
749 	load_direct_gdt(cpu);
750 	/* Reload the per-cpu base */
751 	load_percpu_segment(cpu);
752 }
753 
754 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
755 
756 static void get_model_name(struct cpuinfo_x86 *c)
757 {
758 	unsigned int *v;
759 	char *p, *q, *s;
760 
761 	if (c->extended_cpuid_level < 0x80000004)
762 		return;
763 
764 	v = (unsigned int *)c->x86_model_id;
765 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
766 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
767 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
768 	c->x86_model_id[48] = 0;
769 
770 	/* Trim whitespace */
771 	p = q = s = &c->x86_model_id[0];
772 
773 	while (*p == ' ')
774 		p++;
775 
776 	while (*p) {
777 		/* Note the last non-whitespace index */
778 		if (!isspace(*p))
779 			s = q;
780 
781 		*q++ = *p++;
782 	}
783 
784 	*(s + 1) = '\0';
785 }
786 
787 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
788 {
789 	unsigned int eax, ebx, ecx, edx;
790 
791 	c->x86_max_cores = 1;
792 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
793 		return;
794 
795 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
796 	if (eax & 0x1f)
797 		c->x86_max_cores = (eax >> 26) + 1;
798 }
799 
800 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
801 {
802 	unsigned int n, dummy, ebx, ecx, edx, l2size;
803 
804 	n = c->extended_cpuid_level;
805 
806 	if (n >= 0x80000005) {
807 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
808 		c->x86_cache_size = (ecx>>24) + (edx>>24);
809 #ifdef CONFIG_X86_64
810 		/* On K8 L1 TLB is inclusive, so don't count it */
811 		c->x86_tlbsize = 0;
812 #endif
813 	}
814 
815 	if (n < 0x80000006)	/* Some chips just has a large L1. */
816 		return;
817 
818 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
819 	l2size = ecx >> 16;
820 
821 #ifdef CONFIG_X86_64
822 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
823 #else
824 	/* do processor-specific cache resizing */
825 	if (this_cpu->legacy_cache_size)
826 		l2size = this_cpu->legacy_cache_size(c, l2size);
827 
828 	/* Allow user to override all this if necessary. */
829 	if (cachesize_override != -1)
830 		l2size = cachesize_override;
831 
832 	if (l2size == 0)
833 		return;		/* Again, no L2 cache is possible */
834 #endif
835 
836 	c->x86_cache_size = l2size;
837 }
838 
839 u16 __read_mostly tlb_lli_4k[NR_INFO];
840 u16 __read_mostly tlb_lli_2m[NR_INFO];
841 u16 __read_mostly tlb_lli_4m[NR_INFO];
842 u16 __read_mostly tlb_lld_4k[NR_INFO];
843 u16 __read_mostly tlb_lld_2m[NR_INFO];
844 u16 __read_mostly tlb_lld_4m[NR_INFO];
845 u16 __read_mostly tlb_lld_1g[NR_INFO];
846 
847 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
848 {
849 	if (this_cpu->c_detect_tlb)
850 		this_cpu->c_detect_tlb(c);
851 
852 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
853 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
854 		tlb_lli_4m[ENTRIES]);
855 
856 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
857 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
858 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
859 }
860 
861 int detect_ht_early(struct cpuinfo_x86 *c)
862 {
863 #ifdef CONFIG_SMP
864 	u32 eax, ebx, ecx, edx;
865 
866 	if (!cpu_has(c, X86_FEATURE_HT))
867 		return -1;
868 
869 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
870 		return -1;
871 
872 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
873 		return -1;
874 
875 	cpuid(1, &eax, &ebx, &ecx, &edx);
876 
877 	smp_num_siblings = (ebx & 0xff0000) >> 16;
878 	if (smp_num_siblings == 1)
879 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
880 #endif
881 	return 0;
882 }
883 
884 void detect_ht(struct cpuinfo_x86 *c)
885 {
886 #ifdef CONFIG_SMP
887 	int index_msb, core_bits;
888 
889 	if (detect_ht_early(c) < 0)
890 		return;
891 
892 	index_msb = get_count_order(smp_num_siblings);
893 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
894 
895 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
896 
897 	index_msb = get_count_order(smp_num_siblings);
898 
899 	core_bits = get_count_order(c->x86_max_cores);
900 
901 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
902 				       ((1 << core_bits) - 1);
903 #endif
904 }
905 
906 static void get_cpu_vendor(struct cpuinfo_x86 *c)
907 {
908 	char *v = c->x86_vendor_id;
909 	int i;
910 
911 	for (i = 0; i < X86_VENDOR_NUM; i++) {
912 		if (!cpu_devs[i])
913 			break;
914 
915 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
916 		    (cpu_devs[i]->c_ident[1] &&
917 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
918 
919 			this_cpu = cpu_devs[i];
920 			c->x86_vendor = this_cpu->c_x86_vendor;
921 			return;
922 		}
923 	}
924 
925 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
926 		    "CPU: Your system may be unstable.\n", v);
927 
928 	c->x86_vendor = X86_VENDOR_UNKNOWN;
929 	this_cpu = &default_cpu;
930 }
931 
932 void cpu_detect(struct cpuinfo_x86 *c)
933 {
934 	/* Get vendor name */
935 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
936 	      (unsigned int *)&c->x86_vendor_id[0],
937 	      (unsigned int *)&c->x86_vendor_id[8],
938 	      (unsigned int *)&c->x86_vendor_id[4]);
939 
940 	c->x86 = 4;
941 	/* Intel-defined flags: level 0x00000001 */
942 	if (c->cpuid_level >= 0x00000001) {
943 		u32 junk, tfms, cap0, misc;
944 
945 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
946 		c->x86		= x86_family(tfms);
947 		c->x86_model	= x86_model(tfms);
948 		c->x86_stepping	= x86_stepping(tfms);
949 
950 		if (cap0 & (1<<19)) {
951 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
952 			c->x86_cache_alignment = c->x86_clflush_size;
953 		}
954 	}
955 }
956 
957 static void apply_forced_caps(struct cpuinfo_x86 *c)
958 {
959 	int i;
960 
961 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
962 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
963 		c->x86_capability[i] |= cpu_caps_set[i];
964 	}
965 }
966 
967 static void init_speculation_control(struct cpuinfo_x86 *c)
968 {
969 	/*
970 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
971 	 * and they also have a different bit for STIBP support. Also,
972 	 * a hypervisor might have set the individual AMD bits even on
973 	 * Intel CPUs, for finer-grained selection of what's available.
974 	 */
975 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
976 		set_cpu_cap(c, X86_FEATURE_IBRS);
977 		set_cpu_cap(c, X86_FEATURE_IBPB);
978 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
979 	}
980 
981 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
982 		set_cpu_cap(c, X86_FEATURE_STIBP);
983 
984 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
985 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
986 		set_cpu_cap(c, X86_FEATURE_SSBD);
987 
988 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
989 		set_cpu_cap(c, X86_FEATURE_IBRS);
990 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
991 	}
992 
993 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
994 		set_cpu_cap(c, X86_FEATURE_IBPB);
995 
996 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
997 		set_cpu_cap(c, X86_FEATURE_STIBP);
998 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
999 	}
1000 
1001 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
1002 		set_cpu_cap(c, X86_FEATURE_SSBD);
1003 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1004 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
1005 	}
1006 }
1007 
1008 void get_cpu_cap(struct cpuinfo_x86 *c)
1009 {
1010 	u32 eax, ebx, ecx, edx;
1011 
1012 	/* Intel-defined flags: level 0x00000001 */
1013 	if (c->cpuid_level >= 0x00000001) {
1014 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
1015 
1016 		c->x86_capability[CPUID_1_ECX] = ecx;
1017 		c->x86_capability[CPUID_1_EDX] = edx;
1018 	}
1019 
1020 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
1021 	if (c->cpuid_level >= 0x00000006)
1022 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
1023 
1024 	/* Additional Intel-defined flags: level 0x00000007 */
1025 	if (c->cpuid_level >= 0x00000007) {
1026 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
1027 		c->x86_capability[CPUID_7_0_EBX] = ebx;
1028 		c->x86_capability[CPUID_7_ECX] = ecx;
1029 		c->x86_capability[CPUID_7_EDX] = edx;
1030 
1031 		/* Check valid sub-leaf index before accessing it */
1032 		if (eax >= 1) {
1033 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
1034 			c->x86_capability[CPUID_7_1_EAX] = eax;
1035 		}
1036 	}
1037 
1038 	/* Extended state features: level 0x0000000d */
1039 	if (c->cpuid_level >= 0x0000000d) {
1040 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
1041 
1042 		c->x86_capability[CPUID_D_1_EAX] = eax;
1043 	}
1044 
1045 	/* AMD-defined flags: level 0x80000001 */
1046 	eax = cpuid_eax(0x80000000);
1047 	c->extended_cpuid_level = eax;
1048 
1049 	if ((eax & 0xffff0000) == 0x80000000) {
1050 		if (eax >= 0x80000001) {
1051 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
1052 
1053 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
1054 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
1055 		}
1056 	}
1057 
1058 	if (c->extended_cpuid_level >= 0x80000007) {
1059 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
1060 
1061 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
1062 		c->x86_power = edx;
1063 	}
1064 
1065 	if (c->extended_cpuid_level >= 0x80000008) {
1066 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
1067 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
1068 	}
1069 
1070 	if (c->extended_cpuid_level >= 0x8000000a)
1071 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
1072 
1073 	if (c->extended_cpuid_level >= 0x8000001f)
1074 		c->x86_capability[CPUID_8000_001F_EAX] = cpuid_eax(0x8000001f);
1075 
1076 	init_scattered_cpuid_features(c);
1077 	init_speculation_control(c);
1078 
1079 	/*
1080 	 * Clear/Set all flags overridden by options, after probe.
1081 	 * This needs to happen each time we re-probe, which may happen
1082 	 * several times during CPU initialization.
1083 	 */
1084 	apply_forced_caps(c);
1085 }
1086 
1087 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
1088 {
1089 	u32 eax, ebx, ecx, edx;
1090 
1091 	if (c->extended_cpuid_level >= 0x80000008) {
1092 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
1093 
1094 		c->x86_virt_bits = (eax >> 8) & 0xff;
1095 		c->x86_phys_bits = eax & 0xff;
1096 	}
1097 #ifdef CONFIG_X86_32
1098 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
1099 		c->x86_phys_bits = 36;
1100 #endif
1101 	c->x86_cache_bits = c->x86_phys_bits;
1102 }
1103 
1104 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
1105 {
1106 #ifdef CONFIG_X86_32
1107 	int i;
1108 
1109 	/*
1110 	 * First of all, decide if this is a 486 or higher
1111 	 * It's a 486 if we can modify the AC flag
1112 	 */
1113 	if (flag_is_changeable_p(X86_EFLAGS_AC))
1114 		c->x86 = 4;
1115 	else
1116 		c->x86 = 3;
1117 
1118 	for (i = 0; i < X86_VENDOR_NUM; i++)
1119 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
1120 			c->x86_vendor_id[0] = 0;
1121 			cpu_devs[i]->c_identify(c);
1122 			if (c->x86_vendor_id[0]) {
1123 				get_cpu_vendor(c);
1124 				break;
1125 			}
1126 		}
1127 #endif
1128 }
1129 
1130 #define NO_SPECULATION		BIT(0)
1131 #define NO_MELTDOWN		BIT(1)
1132 #define NO_SSB			BIT(2)
1133 #define NO_L1TF			BIT(3)
1134 #define NO_MDS			BIT(4)
1135 #define MSBDS_ONLY		BIT(5)
1136 #define NO_SWAPGS		BIT(6)
1137 #define NO_ITLB_MULTIHIT	BIT(7)
1138 #define NO_SPECTRE_V2		BIT(8)
1139 #define NO_MMIO			BIT(9)
1140 #define NO_EIBRS_PBRSB		BIT(10)
1141 
1142 #define VULNWL(vendor, family, model, whitelist)	\
1143 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist)
1144 
1145 #define VULNWL_INTEL(model, whitelist)		\
1146 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1147 
1148 #define VULNWL_AMD(family, whitelist)		\
1149 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1150 
1151 #define VULNWL_HYGON(family, whitelist)		\
1152 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1153 
1154 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1155 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1156 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1157 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1158 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1159 	VULNWL(VORTEX,	5, X86_MODEL_ANY,	NO_SPECULATION),
1160 	VULNWL(VORTEX,	6, X86_MODEL_ANY,	NO_SPECULATION),
1161 
1162 	/* Intel Family 6 */
1163 	VULNWL_INTEL(TIGERLAKE,			NO_MMIO),
1164 	VULNWL_INTEL(TIGERLAKE_L,		NO_MMIO),
1165 	VULNWL_INTEL(ALDERLAKE,			NO_MMIO),
1166 	VULNWL_INTEL(ALDERLAKE_L,		NO_MMIO),
1167 
1168 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1169 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION | NO_ITLB_MULTIHIT),
1170 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1171 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1172 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1173 
1174 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1175 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1176 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1177 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1178 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1179 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1180 
1181 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1182 
1183 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1184 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1185 
1186 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1187 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1188 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB),
1189 
1190 	/*
1191 	 * Technically, swapgs isn't serializing on AMD (despite it previously
1192 	 * being documented as such in the APM).  But according to AMD, %gs is
1193 	 * updated non-speculatively, and the issuing of %gs-relative memory
1194 	 * operands will be blocked until the %gs update completes, which is
1195 	 * good enough for our purposes.
1196 	 */
1197 
1198 	VULNWL_INTEL(ATOM_TREMONT,		NO_EIBRS_PBRSB),
1199 	VULNWL_INTEL(ATOM_TREMONT_L,		NO_EIBRS_PBRSB),
1200 	VULNWL_INTEL(ATOM_TREMONT_D,		NO_ITLB_MULTIHIT | NO_EIBRS_PBRSB),
1201 
1202 	/* AMD Family 0xf - 0x12 */
1203 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1204 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1205 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1206 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1207 
1208 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1209 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1210 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1211 
1212 	/* Zhaoxin Family 7 */
1213 	VULNWL(CENTAUR,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO),
1214 	VULNWL(ZHAOXIN,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO),
1215 	{}
1216 };
1217 
1218 #define VULNBL(vendor, family, model, blacklist)	\
1219 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, blacklist)
1220 
1221 #define VULNBL_INTEL_STEPPINGS(model, steppings, issues)		   \
1222 	X86_MATCH_VENDOR_FAM_MODEL_STEPPINGS_FEATURE(INTEL, 6,		   \
1223 					    INTEL_FAM6_##model, steppings, \
1224 					    X86_FEATURE_ANY, issues)
1225 
1226 #define VULNBL_AMD(family, blacklist)		\
1227 	VULNBL(AMD, family, X86_MODEL_ANY, blacklist)
1228 
1229 #define VULNBL_HYGON(family, blacklist)		\
1230 	VULNBL(HYGON, family, X86_MODEL_ANY, blacklist)
1231 
1232 #define SRBDS		BIT(0)
1233 /* CPU is affected by X86_BUG_MMIO_STALE_DATA */
1234 #define MMIO		BIT(1)
1235 /* CPU is affected by Shared Buffers Data Sampling (SBDS), a variant of X86_BUG_MMIO_STALE_DATA */
1236 #define MMIO_SBDS	BIT(2)
1237 /* CPU is affected by RETbleed, speculating where you would not expect it */
1238 #define RETBLEED	BIT(3)
1239 
1240 static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = {
1241 	VULNBL_INTEL_STEPPINGS(IVYBRIDGE,	X86_STEPPING_ANY,		SRBDS),
1242 	VULNBL_INTEL_STEPPINGS(HASWELL,		X86_STEPPING_ANY,		SRBDS),
1243 	VULNBL_INTEL_STEPPINGS(HASWELL_L,	X86_STEPPING_ANY,		SRBDS),
1244 	VULNBL_INTEL_STEPPINGS(HASWELL_G,	X86_STEPPING_ANY,		SRBDS),
1245 	VULNBL_INTEL_STEPPINGS(HASWELL_X,	X86_STEPPING_ANY,		MMIO),
1246 	VULNBL_INTEL_STEPPINGS(BROADWELL_D,	X86_STEPPING_ANY,		MMIO),
1247 	VULNBL_INTEL_STEPPINGS(BROADWELL_G,	X86_STEPPING_ANY,		SRBDS),
1248 	VULNBL_INTEL_STEPPINGS(BROADWELL_X,	X86_STEPPING_ANY,		MMIO),
1249 	VULNBL_INTEL_STEPPINGS(BROADWELL,	X86_STEPPING_ANY,		SRBDS),
1250 	VULNBL_INTEL_STEPPINGS(SKYLAKE_L,	X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1251 	VULNBL_INTEL_STEPPINGS(SKYLAKE_X,	X86_STEPPING_ANY,		MMIO | RETBLEED),
1252 	VULNBL_INTEL_STEPPINGS(SKYLAKE,		X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1253 	VULNBL_INTEL_STEPPINGS(KABYLAKE_L,	X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1254 	VULNBL_INTEL_STEPPINGS(KABYLAKE,	X86_STEPPING_ANY,		SRBDS | MMIO | RETBLEED),
1255 	VULNBL_INTEL_STEPPINGS(CANNONLAKE_L,	X86_STEPPING_ANY,		RETBLEED),
1256 	VULNBL_INTEL_STEPPINGS(ICELAKE_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1257 	VULNBL_INTEL_STEPPINGS(ICELAKE_D,	X86_STEPPING_ANY,		MMIO),
1258 	VULNBL_INTEL_STEPPINGS(ICELAKE_X,	X86_STEPPING_ANY,		MMIO),
1259 	VULNBL_INTEL_STEPPINGS(COMETLAKE,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1260 	VULNBL_INTEL_STEPPINGS(COMETLAKE_L,	X86_STEPPINGS(0x0, 0x0),	MMIO | RETBLEED),
1261 	VULNBL_INTEL_STEPPINGS(COMETLAKE_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1262 	VULNBL_INTEL_STEPPINGS(LAKEFIELD,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1263 	VULNBL_INTEL_STEPPINGS(ROCKETLAKE,	X86_STEPPING_ANY,		MMIO | RETBLEED),
1264 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS),
1265 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_D,	X86_STEPPING_ANY,		MMIO),
1266 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS),
1267 
1268 	VULNBL_AMD(0x15, RETBLEED),
1269 	VULNBL_AMD(0x16, RETBLEED),
1270 	VULNBL_AMD(0x17, RETBLEED),
1271 	VULNBL_HYGON(0x18, RETBLEED),
1272 	{}
1273 };
1274 
1275 static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which)
1276 {
1277 	const struct x86_cpu_id *m = x86_match_cpu(table);
1278 
1279 	return m && !!(m->driver_data & which);
1280 }
1281 
1282 u64 x86_read_arch_cap_msr(void)
1283 {
1284 	u64 ia32_cap = 0;
1285 
1286 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1287 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1288 
1289 	return ia32_cap;
1290 }
1291 
1292 static bool arch_cap_mmio_immune(u64 ia32_cap)
1293 {
1294 	return (ia32_cap & ARCH_CAP_FBSDP_NO &&
1295 		ia32_cap & ARCH_CAP_PSDP_NO &&
1296 		ia32_cap & ARCH_CAP_SBDR_SSDP_NO);
1297 }
1298 
1299 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1300 {
1301 	u64 ia32_cap = x86_read_arch_cap_msr();
1302 
1303 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1304 	if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) &&
1305 	    !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO))
1306 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1307 
1308 	if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION))
1309 		return;
1310 
1311 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1312 
1313 	if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2))
1314 		setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1315 
1316 	if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) &&
1317 	    !(ia32_cap & ARCH_CAP_SSB_NO) &&
1318 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1319 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1320 
1321 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1322 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1323 
1324 	if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) &&
1325 	    !(ia32_cap & ARCH_CAP_MDS_NO)) {
1326 		setup_force_cpu_bug(X86_BUG_MDS);
1327 		if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY))
1328 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1329 	}
1330 
1331 	if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS))
1332 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1333 
1334 	/*
1335 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1336 	 *	- TSX is supported or
1337 	 *	- TSX_CTRL is present
1338 	 *
1339 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1340 	 * the kernel boot e.g. kexec.
1341 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1342 	 * update is not present or running as guest that don't get TSX_CTRL.
1343 	 */
1344 	if (!(ia32_cap & ARCH_CAP_TAA_NO) &&
1345 	    (cpu_has(c, X86_FEATURE_RTM) ||
1346 	     (ia32_cap & ARCH_CAP_TSX_CTRL_MSR)))
1347 		setup_force_cpu_bug(X86_BUG_TAA);
1348 
1349 	/*
1350 	 * SRBDS affects CPUs which support RDRAND or RDSEED and are listed
1351 	 * in the vulnerability blacklist.
1352 	 *
1353 	 * Some of the implications and mitigation of Shared Buffers Data
1354 	 * Sampling (SBDS) are similar to SRBDS. Give SBDS same treatment as
1355 	 * SRBDS.
1356 	 */
1357 	if ((cpu_has(c, X86_FEATURE_RDRAND) ||
1358 	     cpu_has(c, X86_FEATURE_RDSEED)) &&
1359 	    cpu_matches(cpu_vuln_blacklist, SRBDS | MMIO_SBDS))
1360 		    setup_force_cpu_bug(X86_BUG_SRBDS);
1361 
1362 	/*
1363 	 * Processor MMIO Stale Data bug enumeration
1364 	 *
1365 	 * Affected CPU list is generally enough to enumerate the vulnerability,
1366 	 * but for virtualization case check for ARCH_CAP MSR bits also, VMM may
1367 	 * not want the guest to enumerate the bug.
1368 	 *
1369 	 * Set X86_BUG_MMIO_UNKNOWN for CPUs that are neither in the blacklist,
1370 	 * nor in the whitelist and also don't enumerate MSR ARCH_CAP MMIO bits.
1371 	 */
1372 	if (!arch_cap_mmio_immune(ia32_cap)) {
1373 		if (cpu_matches(cpu_vuln_blacklist, MMIO))
1374 			setup_force_cpu_bug(X86_BUG_MMIO_STALE_DATA);
1375 		else if (!cpu_matches(cpu_vuln_whitelist, NO_MMIO))
1376 			setup_force_cpu_bug(X86_BUG_MMIO_UNKNOWN);
1377 	}
1378 
1379 	if (!cpu_has(c, X86_FEATURE_BTC_NO)) {
1380 		if (cpu_matches(cpu_vuln_blacklist, RETBLEED) || (ia32_cap & ARCH_CAP_RSBA))
1381 			setup_force_cpu_bug(X86_BUG_RETBLEED);
1382 	}
1383 
1384 	if (cpu_has(c, X86_FEATURE_IBRS_ENHANCED) &&
1385 	    !cpu_matches(cpu_vuln_whitelist, NO_EIBRS_PBRSB) &&
1386 	    !(ia32_cap & ARCH_CAP_PBRSB_NO))
1387 		setup_force_cpu_bug(X86_BUG_EIBRS_PBRSB);
1388 
1389 	if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN))
1390 		return;
1391 
1392 	/* Rogue Data Cache Load? No! */
1393 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1394 		return;
1395 
1396 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1397 
1398 	if (cpu_matches(cpu_vuln_whitelist, NO_L1TF))
1399 		return;
1400 
1401 	setup_force_cpu_bug(X86_BUG_L1TF);
1402 }
1403 
1404 /*
1405  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1406  * unfortunately, that's not true in practice because of early VIA
1407  * chips and (more importantly) broken virtualizers that are not easy
1408  * to detect. In the latter case it doesn't even *fail* reliably, so
1409  * probing for it doesn't even work. Disable it completely on 32-bit
1410  * unless we can find a reliable way to detect all the broken cases.
1411  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1412  */
1413 static void detect_nopl(void)
1414 {
1415 #ifdef CONFIG_X86_32
1416 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1417 #else
1418 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1419 #endif
1420 }
1421 
1422 /*
1423  * We parse cpu parameters early because fpu__init_system() is executed
1424  * before parse_early_param().
1425  */
1426 static void __init cpu_parse_early_param(void)
1427 {
1428 	char arg[128];
1429 	char *argptr = arg, *opt;
1430 	int arglen, taint = 0;
1431 
1432 #ifdef CONFIG_X86_32
1433 	if (cmdline_find_option_bool(boot_command_line, "no387"))
1434 #ifdef CONFIG_MATH_EMULATION
1435 		setup_clear_cpu_cap(X86_FEATURE_FPU);
1436 #else
1437 		pr_err("Option 'no387' required CONFIG_MATH_EMULATION enabled.\n");
1438 #endif
1439 
1440 	if (cmdline_find_option_bool(boot_command_line, "nofxsr"))
1441 		setup_clear_cpu_cap(X86_FEATURE_FXSR);
1442 #endif
1443 
1444 	if (cmdline_find_option_bool(boot_command_line, "noxsave"))
1445 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
1446 
1447 	if (cmdline_find_option_bool(boot_command_line, "noxsaveopt"))
1448 		setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
1449 
1450 	if (cmdline_find_option_bool(boot_command_line, "noxsaves"))
1451 		setup_clear_cpu_cap(X86_FEATURE_XSAVES);
1452 
1453 	arglen = cmdline_find_option(boot_command_line, "clearcpuid", arg, sizeof(arg));
1454 	if (arglen <= 0)
1455 		return;
1456 
1457 	pr_info("Clearing CPUID bits:");
1458 
1459 	while (argptr) {
1460 		bool found __maybe_unused = false;
1461 		unsigned int bit;
1462 
1463 		opt = strsep(&argptr, ",");
1464 
1465 		/*
1466 		 * Handle naked numbers first for feature flags which don't
1467 		 * have names.
1468 		 */
1469 		if (!kstrtouint(opt, 10, &bit)) {
1470 			if (bit < NCAPINTS * 32) {
1471 
1472 #ifdef CONFIG_X86_FEATURE_NAMES
1473 				/* empty-string, i.e., ""-defined feature flags */
1474 				if (!x86_cap_flags[bit])
1475 					pr_cont(" " X86_CAP_FMT_NUM, x86_cap_flag_num(bit));
1476 				else
1477 #endif
1478 					pr_cont(" " X86_CAP_FMT, x86_cap_flag(bit));
1479 
1480 				setup_clear_cpu_cap(bit);
1481 				taint++;
1482 			}
1483 			/*
1484 			 * The assumption is that there are no feature names with only
1485 			 * numbers in the name thus go to the next argument.
1486 			 */
1487 			continue;
1488 		}
1489 
1490 #ifdef CONFIG_X86_FEATURE_NAMES
1491 		for (bit = 0; bit < 32 * NCAPINTS; bit++) {
1492 			if (!x86_cap_flag(bit))
1493 				continue;
1494 
1495 			if (strcmp(x86_cap_flag(bit), opt))
1496 				continue;
1497 
1498 			pr_cont(" %s", opt);
1499 			setup_clear_cpu_cap(bit);
1500 			taint++;
1501 			found = true;
1502 			break;
1503 		}
1504 
1505 		if (!found)
1506 			pr_cont(" (unknown: %s)", opt);
1507 #endif
1508 	}
1509 	pr_cont("\n");
1510 
1511 	if (taint)
1512 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_STILL_OK);
1513 }
1514 
1515 /*
1516  * Do minimum CPU detection early.
1517  * Fields really needed: vendor, cpuid_level, family, model, mask,
1518  * cache alignment.
1519  * The others are not touched to avoid unwanted side effects.
1520  *
1521  * WARNING: this function is only called on the boot CPU.  Don't add code
1522  * here that is supposed to run on all CPUs.
1523  */
1524 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1525 {
1526 #ifdef CONFIG_X86_64
1527 	c->x86_clflush_size = 64;
1528 	c->x86_phys_bits = 36;
1529 	c->x86_virt_bits = 48;
1530 #else
1531 	c->x86_clflush_size = 32;
1532 	c->x86_phys_bits = 32;
1533 	c->x86_virt_bits = 32;
1534 #endif
1535 	c->x86_cache_alignment = c->x86_clflush_size;
1536 
1537 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1538 	c->extended_cpuid_level = 0;
1539 
1540 	if (!have_cpuid_p())
1541 		identify_cpu_without_cpuid(c);
1542 
1543 	/* cyrix could have cpuid enabled via c_identify()*/
1544 	if (have_cpuid_p()) {
1545 		cpu_detect(c);
1546 		get_cpu_vendor(c);
1547 		get_cpu_cap(c);
1548 		get_cpu_address_sizes(c);
1549 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1550 		cpu_parse_early_param();
1551 
1552 		if (this_cpu->c_early_init)
1553 			this_cpu->c_early_init(c);
1554 
1555 		c->cpu_index = 0;
1556 		filter_cpuid_features(c, false);
1557 
1558 		if (this_cpu->c_bsp_init)
1559 			this_cpu->c_bsp_init(c);
1560 	} else {
1561 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1562 	}
1563 
1564 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1565 
1566 	cpu_set_bug_bits(c);
1567 
1568 	sld_setup(c);
1569 
1570 	fpu__init_system(c);
1571 
1572 	init_sigframe_size();
1573 
1574 #ifdef CONFIG_X86_32
1575 	/*
1576 	 * Regardless of whether PCID is enumerated, the SDM says
1577 	 * that it can't be enabled in 32-bit mode.
1578 	 */
1579 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1580 #endif
1581 
1582 	/*
1583 	 * Later in the boot process pgtable_l5_enabled() relies on
1584 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1585 	 * enabled by this point we need to clear the feature bit to avoid
1586 	 * false-positives at the later stage.
1587 	 *
1588 	 * pgtable_l5_enabled() can be false here for several reasons:
1589 	 *  - 5-level paging is disabled compile-time;
1590 	 *  - it's 32-bit kernel;
1591 	 *  - machine doesn't support 5-level paging;
1592 	 *  - user specified 'no5lvl' in kernel command line.
1593 	 */
1594 	if (!pgtable_l5_enabled())
1595 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1596 
1597 	detect_nopl();
1598 }
1599 
1600 void __init early_cpu_init(void)
1601 {
1602 	const struct cpu_dev *const *cdev;
1603 	int count = 0;
1604 
1605 #ifdef CONFIG_PROCESSOR_SELECT
1606 	pr_info("KERNEL supported cpus:\n");
1607 #endif
1608 
1609 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1610 		const struct cpu_dev *cpudev = *cdev;
1611 
1612 		if (count >= X86_VENDOR_NUM)
1613 			break;
1614 		cpu_devs[count] = cpudev;
1615 		count++;
1616 
1617 #ifdef CONFIG_PROCESSOR_SELECT
1618 		{
1619 			unsigned int j;
1620 
1621 			for (j = 0; j < 2; j++) {
1622 				if (!cpudev->c_ident[j])
1623 					continue;
1624 				pr_info("  %s %s\n", cpudev->c_vendor,
1625 					cpudev->c_ident[j]);
1626 			}
1627 		}
1628 #endif
1629 	}
1630 	early_identify_cpu(&boot_cpu_data);
1631 }
1632 
1633 static bool detect_null_seg_behavior(void)
1634 {
1635 	/*
1636 	 * Empirically, writing zero to a segment selector on AMD does
1637 	 * not clear the base, whereas writing zero to a segment
1638 	 * selector on Intel does clear the base.  Intel's behavior
1639 	 * allows slightly faster context switches in the common case
1640 	 * where GS is unused by the prev and next threads.
1641 	 *
1642 	 * Since neither vendor documents this anywhere that I can see,
1643 	 * detect it directly instead of hard-coding the choice by
1644 	 * vendor.
1645 	 *
1646 	 * I've designated AMD's behavior as the "bug" because it's
1647 	 * counterintuitive and less friendly.
1648 	 */
1649 
1650 	unsigned long old_base, tmp;
1651 	rdmsrl(MSR_FS_BASE, old_base);
1652 	wrmsrl(MSR_FS_BASE, 1);
1653 	loadsegment(fs, 0);
1654 	rdmsrl(MSR_FS_BASE, tmp);
1655 	wrmsrl(MSR_FS_BASE, old_base);
1656 	return tmp == 0;
1657 }
1658 
1659 void check_null_seg_clears_base(struct cpuinfo_x86 *c)
1660 {
1661 	/* BUG_NULL_SEG is only relevant with 64bit userspace */
1662 	if (!IS_ENABLED(CONFIG_X86_64))
1663 		return;
1664 
1665 	/* Zen3 CPUs advertise Null Selector Clears Base in CPUID. */
1666 	if (c->extended_cpuid_level >= 0x80000021 &&
1667 	    cpuid_eax(0x80000021) & BIT(6))
1668 		return;
1669 
1670 	/*
1671 	 * CPUID bit above wasn't set. If this kernel is still running
1672 	 * as a HV guest, then the HV has decided not to advertize
1673 	 * that CPUID bit for whatever reason.	For example, one
1674 	 * member of the migration pool might be vulnerable.  Which
1675 	 * means, the bug is present: set the BUG flag and return.
1676 	 */
1677 	if (cpu_has(c, X86_FEATURE_HYPERVISOR)) {
1678 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1679 		return;
1680 	}
1681 
1682 	/*
1683 	 * Zen2 CPUs also have this behaviour, but no CPUID bit.
1684 	 * 0x18 is the respective family for Hygon.
1685 	 */
1686 	if ((c->x86 == 0x17 || c->x86 == 0x18) &&
1687 	    detect_null_seg_behavior())
1688 		return;
1689 
1690 	/* All the remaining ones are affected */
1691 	set_cpu_bug(c, X86_BUG_NULL_SEG);
1692 }
1693 
1694 static void generic_identify(struct cpuinfo_x86 *c)
1695 {
1696 	c->extended_cpuid_level = 0;
1697 
1698 	if (!have_cpuid_p())
1699 		identify_cpu_without_cpuid(c);
1700 
1701 	/* cyrix could have cpuid enabled via c_identify()*/
1702 	if (!have_cpuid_p())
1703 		return;
1704 
1705 	cpu_detect(c);
1706 
1707 	get_cpu_vendor(c);
1708 
1709 	get_cpu_cap(c);
1710 
1711 	get_cpu_address_sizes(c);
1712 
1713 	if (c->cpuid_level >= 0x00000001) {
1714 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1715 #ifdef CONFIG_X86_32
1716 # ifdef CONFIG_SMP
1717 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1718 # else
1719 		c->apicid = c->initial_apicid;
1720 # endif
1721 #endif
1722 		c->phys_proc_id = c->initial_apicid;
1723 	}
1724 
1725 	get_model_name(c); /* Default name */
1726 
1727 	/*
1728 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1729 	 * systems that run Linux at CPL > 0 may or may not have the
1730 	 * issue, but, even if they have the issue, there's absolutely
1731 	 * nothing we can do about it because we can't use the real IRET
1732 	 * instruction.
1733 	 *
1734 	 * NB: For the time being, only 32-bit kernels support
1735 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1736 	 * whether to apply espfix using paravirt hooks.  If any
1737 	 * non-paravirt system ever shows up that does *not* have the
1738 	 * ESPFIX issue, we can change this.
1739 	 */
1740 #ifdef CONFIG_X86_32
1741 	set_cpu_bug(c, X86_BUG_ESPFIX);
1742 #endif
1743 }
1744 
1745 /*
1746  * Validate that ACPI/mptables have the same information about the
1747  * effective APIC id and update the package map.
1748  */
1749 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1750 {
1751 #ifdef CONFIG_SMP
1752 	unsigned int apicid, cpu = smp_processor_id();
1753 
1754 	apicid = apic->cpu_present_to_apicid(cpu);
1755 
1756 	if (apicid != c->apicid) {
1757 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1758 		       cpu, apicid, c->initial_apicid);
1759 	}
1760 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1761 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1762 #else
1763 	c->logical_proc_id = 0;
1764 #endif
1765 }
1766 
1767 /*
1768  * This does the hard work of actually picking apart the CPU stuff...
1769  */
1770 static void identify_cpu(struct cpuinfo_x86 *c)
1771 {
1772 	int i;
1773 
1774 	c->loops_per_jiffy = loops_per_jiffy;
1775 	c->x86_cache_size = 0;
1776 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1777 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1778 	c->x86_vendor_id[0] = '\0'; /* Unset */
1779 	c->x86_model_id[0] = '\0';  /* Unset */
1780 	c->x86_max_cores = 1;
1781 	c->x86_coreid_bits = 0;
1782 	c->cu_id = 0xff;
1783 #ifdef CONFIG_X86_64
1784 	c->x86_clflush_size = 64;
1785 	c->x86_phys_bits = 36;
1786 	c->x86_virt_bits = 48;
1787 #else
1788 	c->cpuid_level = -1;	/* CPUID not detected */
1789 	c->x86_clflush_size = 32;
1790 	c->x86_phys_bits = 32;
1791 	c->x86_virt_bits = 32;
1792 #endif
1793 	c->x86_cache_alignment = c->x86_clflush_size;
1794 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1795 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
1796 	memset(&c->vmx_capability, 0, sizeof(c->vmx_capability));
1797 #endif
1798 
1799 	generic_identify(c);
1800 
1801 	if (this_cpu->c_identify)
1802 		this_cpu->c_identify(c);
1803 
1804 	/* Clear/Set all flags overridden by options, after probe */
1805 	apply_forced_caps(c);
1806 
1807 #ifdef CONFIG_X86_64
1808 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1809 #endif
1810 
1811 	/*
1812 	 * Vendor-specific initialization.  In this section we
1813 	 * canonicalize the feature flags, meaning if there are
1814 	 * features a certain CPU supports which CPUID doesn't
1815 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1816 	 * we handle them here.
1817 	 *
1818 	 * At the end of this section, c->x86_capability better
1819 	 * indicate the features this CPU genuinely supports!
1820 	 */
1821 	if (this_cpu->c_init)
1822 		this_cpu->c_init(c);
1823 
1824 	/* Disable the PN if appropriate */
1825 	squash_the_stupid_serial_number(c);
1826 
1827 	/* Set up SMEP/SMAP/UMIP */
1828 	setup_smep(c);
1829 	setup_smap(c);
1830 	setup_umip(c);
1831 
1832 	/* Enable FSGSBASE instructions if available. */
1833 	if (cpu_has(c, X86_FEATURE_FSGSBASE)) {
1834 		cr4_set_bits(X86_CR4_FSGSBASE);
1835 		elf_hwcap2 |= HWCAP2_FSGSBASE;
1836 	}
1837 
1838 	/*
1839 	 * The vendor-specific functions might have changed features.
1840 	 * Now we do "generic changes."
1841 	 */
1842 
1843 	/* Filter out anything that depends on CPUID levels we don't have */
1844 	filter_cpuid_features(c, true);
1845 
1846 	/* If the model name is still unset, do table lookup. */
1847 	if (!c->x86_model_id[0]) {
1848 		const char *p;
1849 		p = table_lookup_model(c);
1850 		if (p)
1851 			strcpy(c->x86_model_id, p);
1852 		else
1853 			/* Last resort... */
1854 			sprintf(c->x86_model_id, "%02x/%02x",
1855 				c->x86, c->x86_model);
1856 	}
1857 
1858 #ifdef CONFIG_X86_64
1859 	detect_ht(c);
1860 #endif
1861 
1862 	x86_init_rdrand(c);
1863 	setup_pku(c);
1864 	setup_cet(c);
1865 
1866 	/*
1867 	 * Clear/Set all flags overridden by options, need do it
1868 	 * before following smp all cpus cap AND.
1869 	 */
1870 	apply_forced_caps(c);
1871 
1872 	/*
1873 	 * On SMP, boot_cpu_data holds the common feature set between
1874 	 * all CPUs; so make sure that we indicate which features are
1875 	 * common between the CPUs.  The first time this routine gets
1876 	 * executed, c == &boot_cpu_data.
1877 	 */
1878 	if (c != &boot_cpu_data) {
1879 		/* AND the already accumulated flags with these */
1880 		for (i = 0; i < NCAPINTS; i++)
1881 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1882 
1883 		/* OR, i.e. replicate the bug flags */
1884 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1885 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1886 	}
1887 
1888 	ppin_init(c);
1889 
1890 	/* Init Machine Check Exception if available. */
1891 	mcheck_cpu_init(c);
1892 
1893 	select_idle_routine(c);
1894 
1895 #ifdef CONFIG_NUMA
1896 	numa_add_cpu(smp_processor_id());
1897 #endif
1898 }
1899 
1900 /*
1901  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1902  * on 32-bit kernels:
1903  */
1904 #ifdef CONFIG_X86_32
1905 void enable_sep_cpu(void)
1906 {
1907 	struct tss_struct *tss;
1908 	int cpu;
1909 
1910 	if (!boot_cpu_has(X86_FEATURE_SEP))
1911 		return;
1912 
1913 	cpu = get_cpu();
1914 	tss = &per_cpu(cpu_tss_rw, cpu);
1915 
1916 	/*
1917 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1918 	 * see the big comment in struct x86_hw_tss's definition.
1919 	 */
1920 
1921 	tss->x86_tss.ss1 = __KERNEL_CS;
1922 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1923 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1924 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1925 
1926 	put_cpu();
1927 }
1928 #endif
1929 
1930 void __init identify_boot_cpu(void)
1931 {
1932 	identify_cpu(&boot_cpu_data);
1933 	if (HAS_KERNEL_IBT && cpu_feature_enabled(X86_FEATURE_IBT))
1934 		pr_info("CET detected: Indirect Branch Tracking enabled\n");
1935 #ifdef CONFIG_X86_32
1936 	sysenter_setup();
1937 	enable_sep_cpu();
1938 #endif
1939 	cpu_detect_tlb(&boot_cpu_data);
1940 	setup_cr_pinning();
1941 
1942 	tsx_init();
1943 }
1944 
1945 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1946 {
1947 	BUG_ON(c == &boot_cpu_data);
1948 	identify_cpu(c);
1949 #ifdef CONFIG_X86_32
1950 	enable_sep_cpu();
1951 #endif
1952 	validate_apic_and_package_id(c);
1953 	x86_spec_ctrl_setup_ap();
1954 	update_srbds_msr();
1955 
1956 	tsx_ap_init();
1957 }
1958 
1959 void print_cpu_info(struct cpuinfo_x86 *c)
1960 {
1961 	const char *vendor = NULL;
1962 
1963 	if (c->x86_vendor < X86_VENDOR_NUM) {
1964 		vendor = this_cpu->c_vendor;
1965 	} else {
1966 		if (c->cpuid_level >= 0)
1967 			vendor = c->x86_vendor_id;
1968 	}
1969 
1970 	if (vendor && !strstr(c->x86_model_id, vendor))
1971 		pr_cont("%s ", vendor);
1972 
1973 	if (c->x86_model_id[0])
1974 		pr_cont("%s", c->x86_model_id);
1975 	else
1976 		pr_cont("%d86", c->x86);
1977 
1978 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1979 
1980 	if (c->x86_stepping || c->cpuid_level >= 0)
1981 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1982 	else
1983 		pr_cont(")\n");
1984 }
1985 
1986 /*
1987  * clearcpuid= was already parsed in cpu_parse_early_param().  This dummy
1988  * function prevents it from becoming an environment variable for init.
1989  */
1990 static __init int setup_clearcpuid(char *arg)
1991 {
1992 	return 1;
1993 }
1994 __setup("clearcpuid=", setup_clearcpuid);
1995 
1996 #ifdef CONFIG_X86_64
1997 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1998 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1999 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
2000 
2001 /*
2002  * The following percpu variables are hot.  Align current_task to
2003  * cacheline size such that they fall in the same cacheline.
2004  */
2005 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
2006 	&init_task;
2007 EXPORT_PER_CPU_SYMBOL(current_task);
2008 
2009 DEFINE_PER_CPU(void *, hardirq_stack_ptr);
2010 DEFINE_PER_CPU(bool, hardirq_stack_inuse);
2011 
2012 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
2013 EXPORT_PER_CPU_SYMBOL(__preempt_count);
2014 
2015 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) = TOP_OF_INIT_STACK;
2016 
2017 static void wrmsrl_cstar(unsigned long val)
2018 {
2019 	/*
2020 	 * Intel CPUs do not support 32-bit SYSCALL. Writing to MSR_CSTAR
2021 	 * is so far ignored by the CPU, but raises a #VE trap in a TDX
2022 	 * guest. Avoid the pointless write on all Intel CPUs.
2023 	 */
2024 	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
2025 		wrmsrl(MSR_CSTAR, val);
2026 }
2027 
2028 /* May not be marked __init: used by software suspend */
2029 void syscall_init(void)
2030 {
2031 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
2032 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
2033 
2034 #ifdef CONFIG_IA32_EMULATION
2035 	wrmsrl_cstar((unsigned long)entry_SYSCALL_compat);
2036 	/*
2037 	 * This only works on Intel CPUs.
2038 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
2039 	 * This does not cause SYSENTER to jump to the wrong location, because
2040 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
2041 	 */
2042 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
2043 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
2044 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
2045 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
2046 #else
2047 	wrmsrl_cstar((unsigned long)ignore_sysret);
2048 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
2049 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
2050 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
2051 #endif
2052 
2053 	/*
2054 	 * Flags to clear on syscall; clear as much as possible
2055 	 * to minimize user space-kernel interference.
2056 	 */
2057 	wrmsrl(MSR_SYSCALL_MASK,
2058 	       X86_EFLAGS_CF|X86_EFLAGS_PF|X86_EFLAGS_AF|
2059 	       X86_EFLAGS_ZF|X86_EFLAGS_SF|X86_EFLAGS_TF|
2060 	       X86_EFLAGS_IF|X86_EFLAGS_DF|X86_EFLAGS_OF|
2061 	       X86_EFLAGS_IOPL|X86_EFLAGS_NT|X86_EFLAGS_RF|
2062 	       X86_EFLAGS_AC|X86_EFLAGS_ID);
2063 }
2064 
2065 #else	/* CONFIG_X86_64 */
2066 
2067 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
2068 EXPORT_PER_CPU_SYMBOL(current_task);
2069 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
2070 EXPORT_PER_CPU_SYMBOL(__preempt_count);
2071 
2072 /*
2073  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
2074  * the top of the kernel stack.  Use an extra percpu variable to track the
2075  * top of the kernel stack directly.
2076  */
2077 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
2078 	(unsigned long)&init_thread_union + THREAD_SIZE;
2079 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
2080 
2081 #ifdef CONFIG_STACKPROTECTOR
2082 DEFINE_PER_CPU(unsigned long, __stack_chk_guard);
2083 EXPORT_PER_CPU_SYMBOL(__stack_chk_guard);
2084 #endif
2085 
2086 #endif	/* CONFIG_X86_64 */
2087 
2088 /*
2089  * Clear all 6 debug registers:
2090  */
2091 static void clear_all_debug_regs(void)
2092 {
2093 	int i;
2094 
2095 	for (i = 0; i < 8; i++) {
2096 		/* Ignore db4, db5 */
2097 		if ((i == 4) || (i == 5))
2098 			continue;
2099 
2100 		set_debugreg(0, i);
2101 	}
2102 }
2103 
2104 #ifdef CONFIG_KGDB
2105 /*
2106  * Restore debug regs if using kgdbwait and you have a kernel debugger
2107  * connection established.
2108  */
2109 static void dbg_restore_debug_regs(void)
2110 {
2111 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
2112 		arch_kgdb_ops.correct_hw_break();
2113 }
2114 #else /* ! CONFIG_KGDB */
2115 #define dbg_restore_debug_regs()
2116 #endif /* ! CONFIG_KGDB */
2117 
2118 static void wait_for_master_cpu(int cpu)
2119 {
2120 #ifdef CONFIG_SMP
2121 	/*
2122 	 * wait for ACK from master CPU before continuing
2123 	 * with AP initialization
2124 	 */
2125 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
2126 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
2127 		cpu_relax();
2128 #endif
2129 }
2130 
2131 #ifdef CONFIG_X86_64
2132 static inline void setup_getcpu(int cpu)
2133 {
2134 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
2135 	struct desc_struct d = { };
2136 
2137 	if (boot_cpu_has(X86_FEATURE_RDTSCP) || boot_cpu_has(X86_FEATURE_RDPID))
2138 		wrmsr(MSR_TSC_AUX, cpudata, 0);
2139 
2140 	/* Store CPU and node number in limit. */
2141 	d.limit0 = cpudata;
2142 	d.limit1 = cpudata >> 16;
2143 
2144 	d.type = 5;		/* RO data, expand down, accessed */
2145 	d.dpl = 3;		/* Visible to user code */
2146 	d.s = 1;		/* Not a system segment */
2147 	d.p = 1;		/* Present */
2148 	d.d = 1;		/* 32-bit */
2149 
2150 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
2151 }
2152 
2153 static inline void ucode_cpu_init(int cpu)
2154 {
2155 	if (cpu)
2156 		load_ucode_ap();
2157 }
2158 
2159 static inline void tss_setup_ist(struct tss_struct *tss)
2160 {
2161 	/* Set up the per-CPU TSS IST stacks */
2162 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
2163 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
2164 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
2165 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
2166 	/* Only mapped when SEV-ES is active */
2167 	tss->x86_tss.ist[IST_INDEX_VC] = __this_cpu_ist_top_va(VC);
2168 }
2169 
2170 #else /* CONFIG_X86_64 */
2171 
2172 static inline void setup_getcpu(int cpu) { }
2173 
2174 static inline void ucode_cpu_init(int cpu)
2175 {
2176 	show_ucode_info_early();
2177 }
2178 
2179 static inline void tss_setup_ist(struct tss_struct *tss) { }
2180 
2181 #endif /* !CONFIG_X86_64 */
2182 
2183 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
2184 {
2185 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
2186 
2187 #ifdef CONFIG_X86_IOPL_IOPERM
2188 	tss->io_bitmap.prev_max = 0;
2189 	tss->io_bitmap.prev_sequence = 0;
2190 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
2191 	/*
2192 	 * Invalidate the extra array entry past the end of the all
2193 	 * permission bitmap as required by the hardware.
2194 	 */
2195 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
2196 #endif
2197 }
2198 
2199 /*
2200  * Setup everything needed to handle exceptions from the IDT, including the IST
2201  * exceptions which use paranoid_entry().
2202  */
2203 void cpu_init_exception_handling(void)
2204 {
2205 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
2206 	int cpu = raw_smp_processor_id();
2207 
2208 	/* paranoid_entry() gets the CPU number from the GDT */
2209 	setup_getcpu(cpu);
2210 
2211 	/* IST vectors need TSS to be set up. */
2212 	tss_setup_ist(tss);
2213 	tss_setup_io_bitmap(tss);
2214 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
2215 
2216 	load_TR_desc();
2217 
2218 	/* GHCB needs to be setup to handle #VC. */
2219 	setup_ghcb();
2220 
2221 	/* Finally load the IDT */
2222 	load_current_idt();
2223 }
2224 
2225 /*
2226  * cpu_init() initializes state that is per-CPU. Some data is already
2227  * initialized (naturally) in the bootstrap process, such as the GDT.  We
2228  * reload it nevertheless, this function acts as a 'CPU state barrier',
2229  * nothing should get across.
2230  */
2231 void cpu_init(void)
2232 {
2233 	struct task_struct *cur = current;
2234 	int cpu = raw_smp_processor_id();
2235 
2236 	wait_for_master_cpu(cpu);
2237 
2238 	ucode_cpu_init(cpu);
2239 
2240 #ifdef CONFIG_NUMA
2241 	if (this_cpu_read(numa_node) == 0 &&
2242 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
2243 		set_numa_node(early_cpu_to_node(cpu));
2244 #endif
2245 	pr_debug("Initializing CPU#%d\n", cpu);
2246 
2247 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
2248 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
2249 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
2250 
2251 	/*
2252 	 * Initialize the per-CPU GDT with the boot GDT,
2253 	 * and set up the GDT descriptor:
2254 	 */
2255 	switch_to_new_gdt(cpu);
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  * The microcode loader calls this upon late microcode load to recheck features,
2311  * only when microcode has been updated. Caller holds microcode_mutex and CPU
2312  * hotplug lock.
2313  */
2314 void microcode_check(void)
2315 {
2316 	struct cpuinfo_x86 info;
2317 
2318 	perf_check_microcode();
2319 
2320 	/* Reload CPUID max function as it might've changed. */
2321 	info.cpuid_level = cpuid_eax(0);
2322 
2323 	/*
2324 	 * Copy all capability leafs to pick up the synthetic ones so that
2325 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
2326 	 * get overwritten in get_cpu_cap().
2327 	 */
2328 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
2329 
2330 	get_cpu_cap(&info);
2331 
2332 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
2333 		return;
2334 
2335 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
2336 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
2337 }
2338 #endif
2339 
2340 /*
2341  * Invoked from core CPU hotplug code after hotplug operations
2342  */
2343 void arch_smt_update(void)
2344 {
2345 	/* Handle the speculative execution misfeatures */
2346 	cpu_bugs_smt_update();
2347 	/* Check whether IPI broadcasting can be enabled */
2348 	apic_smt_update();
2349 }
2350