xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 2762b3cc)
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/mem_encrypt.h>
22 #include <linux/smp.h>
23 #include <linux/cpu.h>
24 #include <linux/io.h>
25 #include <linux/syscore_ops.h>
26 #include <linux/pgtable.h>
27 #include <linux/stackprotector.h>
28 #include <linux/utsname.h>
29 
30 #include <asm/alternative.h>
31 #include <asm/cmdline.h>
32 #include <asm/perf_event.h>
33 #include <asm/mmu_context.h>
34 #include <asm/doublefault.h>
35 #include <asm/archrandom.h>
36 #include <asm/hypervisor.h>
37 #include <asm/processor.h>
38 #include <asm/tlbflush.h>
39 #include <asm/debugreg.h>
40 #include <asm/sections.h>
41 #include <asm/vsyscall.h>
42 #include <linux/topology.h>
43 #include <linux/cpumask.h>
44 #include <linux/atomic.h>
45 #include <asm/proto.h>
46 #include <asm/setup.h>
47 #include <asm/apic.h>
48 #include <asm/desc.h>
49 #include <asm/fpu/api.h>
50 #include <asm/mtrr.h>
51 #include <asm/hwcap2.h>
52 #include <linux/numa.h>
53 #include <asm/numa.h>
54 #include <asm/asm.h>
55 #include <asm/bugs.h>
56 #include <asm/cpu.h>
57 #include <asm/mce.h>
58 #include <asm/msr.h>
59 #include <asm/cacheinfo.h>
60 #include <asm/memtype.h>
61 #include <asm/microcode.h>
62 #include <asm/intel-family.h>
63 #include <asm/cpu_device_id.h>
64 #include <asm/uv/uv.h>
65 #include <asm/set_memory.h>
66 #include <asm/traps.h>
67 #include <asm/sev.h>
68 
69 #include "cpu.h"
70 
71 u32 elf_hwcap2 __read_mostly;
72 
73 /* Number of siblings per CPU package */
74 int smp_num_siblings = 1;
75 EXPORT_SYMBOL(smp_num_siblings);
76 
77 /* Last level cache ID of each logical CPU */
78 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
79 
80 u16 get_llc_id(unsigned int cpu)
81 {
82 	return per_cpu(cpu_llc_id, cpu);
83 }
84 EXPORT_SYMBOL_GPL(get_llc_id);
85 
86 /* L2 cache ID of each logical CPU */
87 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_l2c_id) = BAD_APICID;
88 
89 static struct ppin_info {
90 	int	feature;
91 	int	msr_ppin_ctl;
92 	int	msr_ppin;
93 } ppin_info[] = {
94 	[X86_VENDOR_INTEL] = {
95 		.feature = X86_FEATURE_INTEL_PPIN,
96 		.msr_ppin_ctl = MSR_PPIN_CTL,
97 		.msr_ppin = MSR_PPIN
98 	},
99 	[X86_VENDOR_AMD] = {
100 		.feature = X86_FEATURE_AMD_PPIN,
101 		.msr_ppin_ctl = MSR_AMD_PPIN_CTL,
102 		.msr_ppin = MSR_AMD_PPIN
103 	},
104 };
105 
106 static const struct x86_cpu_id ppin_cpuids[] = {
107 	X86_MATCH_FEATURE(X86_FEATURE_AMD_PPIN, &ppin_info[X86_VENDOR_AMD]),
108 	X86_MATCH_FEATURE(X86_FEATURE_INTEL_PPIN, &ppin_info[X86_VENDOR_INTEL]),
109 
110 	/* Legacy models without CPUID enumeration */
111 	X86_MATCH_INTEL_FAM6_MODEL(IVYBRIDGE_X, &ppin_info[X86_VENDOR_INTEL]),
112 	X86_MATCH_INTEL_FAM6_MODEL(HASWELL_X, &ppin_info[X86_VENDOR_INTEL]),
113 	X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_D, &ppin_info[X86_VENDOR_INTEL]),
114 	X86_MATCH_INTEL_FAM6_MODEL(BROADWELL_X, &ppin_info[X86_VENDOR_INTEL]),
115 	X86_MATCH_INTEL_FAM6_MODEL(SKYLAKE_X, &ppin_info[X86_VENDOR_INTEL]),
116 	X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_X, &ppin_info[X86_VENDOR_INTEL]),
117 	X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_D, &ppin_info[X86_VENDOR_INTEL]),
118 	X86_MATCH_INTEL_FAM6_MODEL(SAPPHIRERAPIDS_X, &ppin_info[X86_VENDOR_INTEL]),
119 	X86_MATCH_INTEL_FAM6_MODEL(EMERALDRAPIDS_X, &ppin_info[X86_VENDOR_INTEL]),
120 	X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNL, &ppin_info[X86_VENDOR_INTEL]),
121 	X86_MATCH_INTEL_FAM6_MODEL(XEON_PHI_KNM, &ppin_info[X86_VENDOR_INTEL]),
122 
123 	{}
124 };
125 
126 static void ppin_init(struct cpuinfo_x86 *c)
127 {
128 	const struct x86_cpu_id *id;
129 	unsigned long long val;
130 	struct ppin_info *info;
131 
132 	id = x86_match_cpu(ppin_cpuids);
133 	if (!id)
134 		return;
135 
136 	/*
137 	 * Testing the presence of the MSR is not enough. Need to check
138 	 * that the PPIN_CTL allows reading of the PPIN.
139 	 */
140 	info = (struct ppin_info *)id->driver_data;
141 
142 	if (rdmsrl_safe(info->msr_ppin_ctl, &val))
143 		goto clear_ppin;
144 
145 	if ((val & 3UL) == 1UL) {
146 		/* PPIN locked in disabled mode */
147 		goto clear_ppin;
148 	}
149 
150 	/* If PPIN is disabled, try to enable */
151 	if (!(val & 2UL)) {
152 		wrmsrl_safe(info->msr_ppin_ctl,  val | 2UL);
153 		rdmsrl_safe(info->msr_ppin_ctl, &val);
154 	}
155 
156 	/* Is the enable bit set? */
157 	if (val & 2UL) {
158 		c->ppin = __rdmsr(info->msr_ppin);
159 		set_cpu_cap(c, info->feature);
160 		return;
161 	}
162 
163 clear_ppin:
164 	clear_cpu_cap(c, info->feature);
165 }
166 
167 static void default_init(struct cpuinfo_x86 *c)
168 {
169 #ifdef CONFIG_X86_64
170 	cpu_detect_cache_sizes(c);
171 #else
172 	/* Not much we can do here... */
173 	/* Check if at least it has cpuid */
174 	if (c->cpuid_level == -1) {
175 		/* No cpuid. It must be an ancient CPU */
176 		if (c->x86 == 4)
177 			strcpy(c->x86_model_id, "486");
178 		else if (c->x86 == 3)
179 			strcpy(c->x86_model_id, "386");
180 	}
181 #endif
182 }
183 
184 static const struct cpu_dev default_cpu = {
185 	.c_init		= default_init,
186 	.c_vendor	= "Unknown",
187 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
188 };
189 
190 static const struct cpu_dev *this_cpu = &default_cpu;
191 
192 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
193 #ifdef CONFIG_X86_64
194 	/*
195 	 * We need valid kernel segments for data and code in long mode too
196 	 * IRET will check the segment types  kkeil 2000/10/28
197 	 * Also sysret mandates a special GDT layout
198 	 *
199 	 * TLS descriptors are currently at a different place compared to i386.
200 	 * Hopefully nobody expects them at a fixed place (Wine?)
201 	 */
202 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
203 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
204 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
205 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
206 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
207 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
208 #else
209 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
210 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
211 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
212 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
213 	/*
214 	 * Segments used for calling PnP BIOS have byte granularity.
215 	 * They code segments and data segments have fixed 64k limits,
216 	 * the transfer segment sizes are set at run time.
217 	 */
218 	/* 32-bit code */
219 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
220 	/* 16-bit code */
221 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
222 	/* 16-bit data */
223 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
224 	/* 16-bit data */
225 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
226 	/* 16-bit data */
227 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
228 	/*
229 	 * The APM segments have byte granularity and their bases
230 	 * are set at run time.  All have 64k limits.
231 	 */
232 	/* 32-bit code */
233 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
234 	/* 16-bit code */
235 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
236 	/* data */
237 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
238 
239 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
240 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
241 #endif
242 } };
243 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
244 
245 #ifdef CONFIG_X86_64
246 static int __init x86_nopcid_setup(char *s)
247 {
248 	/* nopcid doesn't accept parameters */
249 	if (s)
250 		return -EINVAL;
251 
252 	/* do not emit a message if the feature is not present */
253 	if (!boot_cpu_has(X86_FEATURE_PCID))
254 		return 0;
255 
256 	setup_clear_cpu_cap(X86_FEATURE_PCID);
257 	pr_info("nopcid: PCID feature disabled\n");
258 	return 0;
259 }
260 early_param("nopcid", x86_nopcid_setup);
261 #endif
262 
263 static int __init x86_noinvpcid_setup(char *s)
264 {
265 	/* noinvpcid doesn't accept parameters */
266 	if (s)
267 		return -EINVAL;
268 
269 	/* do not emit a message if the feature is not present */
270 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
271 		return 0;
272 
273 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
274 	pr_info("noinvpcid: INVPCID feature disabled\n");
275 	return 0;
276 }
277 early_param("noinvpcid", x86_noinvpcid_setup);
278 
279 #ifdef CONFIG_X86_32
280 static int cachesize_override = -1;
281 static int disable_x86_serial_nr = 1;
282 
283 static int __init cachesize_setup(char *str)
284 {
285 	get_option(&str, &cachesize_override);
286 	return 1;
287 }
288 __setup("cachesize=", cachesize_setup);
289 
290 /* Standard macro to see if a specific flag is changeable */
291 static inline int flag_is_changeable_p(u32 flag)
292 {
293 	u32 f1, f2;
294 
295 	/*
296 	 * Cyrix and IDT cpus allow disabling of CPUID
297 	 * so the code below may return different results
298 	 * when it is executed before and after enabling
299 	 * the CPUID. Add "volatile" to not allow gcc to
300 	 * optimize the subsequent calls to this function.
301 	 */
302 	asm volatile ("pushfl		\n\t"
303 		      "pushfl		\n\t"
304 		      "popl %0		\n\t"
305 		      "movl %0, %1	\n\t"
306 		      "xorl %2, %0	\n\t"
307 		      "pushl %0		\n\t"
308 		      "popfl		\n\t"
309 		      "pushfl		\n\t"
310 		      "popl %0		\n\t"
311 		      "popfl		\n\t"
312 
313 		      : "=&r" (f1), "=&r" (f2)
314 		      : "ir" (flag));
315 
316 	return ((f1^f2) & flag) != 0;
317 }
318 
319 /* Probe for the CPUID instruction */
320 int have_cpuid_p(void)
321 {
322 	return flag_is_changeable_p(X86_EFLAGS_ID);
323 }
324 
325 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
326 {
327 	unsigned long lo, hi;
328 
329 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
330 		return;
331 
332 	/* Disable processor serial number: */
333 
334 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
335 	lo |= 0x200000;
336 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
337 
338 	pr_notice("CPU serial number disabled.\n");
339 	clear_cpu_cap(c, X86_FEATURE_PN);
340 
341 	/* Disabling the serial number may affect the cpuid level */
342 	c->cpuid_level = cpuid_eax(0);
343 }
344 
345 static int __init x86_serial_nr_setup(char *s)
346 {
347 	disable_x86_serial_nr = 0;
348 	return 1;
349 }
350 __setup("serialnumber", x86_serial_nr_setup);
351 #else
352 static inline int flag_is_changeable_p(u32 flag)
353 {
354 	return 1;
355 }
356 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
357 {
358 }
359 #endif
360 
361 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
362 {
363 	if (cpu_has(c, X86_FEATURE_SMEP))
364 		cr4_set_bits(X86_CR4_SMEP);
365 }
366 
367 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
368 {
369 	unsigned long eflags = native_save_fl();
370 
371 	/* This should have been cleared long ago */
372 	BUG_ON(eflags & X86_EFLAGS_AC);
373 
374 	if (cpu_has(c, X86_FEATURE_SMAP))
375 		cr4_set_bits(X86_CR4_SMAP);
376 }
377 
378 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
379 {
380 	/* Check the boot processor, plus build option for UMIP. */
381 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
382 		goto out;
383 
384 	/* Check the current processor's cpuid bits. */
385 	if (!cpu_has(c, X86_FEATURE_UMIP))
386 		goto out;
387 
388 	cr4_set_bits(X86_CR4_UMIP);
389 
390 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
391 
392 	return;
393 
394 out:
395 	/*
396 	 * Make sure UMIP is disabled in case it was enabled in a
397 	 * previous boot (e.g., via kexec).
398 	 */
399 	cr4_clear_bits(X86_CR4_UMIP);
400 }
401 
402 /* These bits should not change their value after CPU init is finished. */
403 static const unsigned long cr4_pinned_mask =
404 	X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP |
405 	X86_CR4_FSGSBASE | X86_CR4_CET;
406 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
407 static unsigned long cr4_pinned_bits __ro_after_init;
408 
409 void native_write_cr0(unsigned long val)
410 {
411 	unsigned long bits_missing = 0;
412 
413 set_register:
414 	asm volatile("mov %0,%%cr0": "+r" (val) : : "memory");
415 
416 	if (static_branch_likely(&cr_pinning)) {
417 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
418 			bits_missing = X86_CR0_WP;
419 			val |= bits_missing;
420 			goto set_register;
421 		}
422 		/* Warn after we've set the missing bits. */
423 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
424 	}
425 }
426 EXPORT_SYMBOL(native_write_cr0);
427 
428 void __no_profile native_write_cr4(unsigned long val)
429 {
430 	unsigned long bits_changed = 0;
431 
432 set_register:
433 	asm volatile("mov %0,%%cr4": "+r" (val) : : "memory");
434 
435 	if (static_branch_likely(&cr_pinning)) {
436 		if (unlikely((val & cr4_pinned_mask) != cr4_pinned_bits)) {
437 			bits_changed = (val & cr4_pinned_mask) ^ cr4_pinned_bits;
438 			val = (val & ~cr4_pinned_mask) | cr4_pinned_bits;
439 			goto set_register;
440 		}
441 		/* Warn after we've corrected the changed bits. */
442 		WARN_ONCE(bits_changed, "pinned CR4 bits changed: 0x%lx!?\n",
443 			  bits_changed);
444 	}
445 }
446 #if IS_MODULE(CONFIG_LKDTM)
447 EXPORT_SYMBOL_GPL(native_write_cr4);
448 #endif
449 
450 void cr4_update_irqsoff(unsigned long set, unsigned long clear)
451 {
452 	unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
453 
454 	lockdep_assert_irqs_disabled();
455 
456 	newval = (cr4 & ~clear) | set;
457 	if (newval != cr4) {
458 		this_cpu_write(cpu_tlbstate.cr4, newval);
459 		__write_cr4(newval);
460 	}
461 }
462 EXPORT_SYMBOL(cr4_update_irqsoff);
463 
464 /* Read the CR4 shadow. */
465 unsigned long cr4_read_shadow(void)
466 {
467 	return this_cpu_read(cpu_tlbstate.cr4);
468 }
469 EXPORT_SYMBOL_GPL(cr4_read_shadow);
470 
471 void cr4_init(void)
472 {
473 	unsigned long cr4 = __read_cr4();
474 
475 	if (boot_cpu_has(X86_FEATURE_PCID))
476 		cr4 |= X86_CR4_PCIDE;
477 	if (static_branch_likely(&cr_pinning))
478 		cr4 = (cr4 & ~cr4_pinned_mask) | cr4_pinned_bits;
479 
480 	__write_cr4(cr4);
481 
482 	/* Initialize cr4 shadow for this CPU. */
483 	this_cpu_write(cpu_tlbstate.cr4, cr4);
484 }
485 
486 /*
487  * Once CPU feature detection is finished (and boot params have been
488  * parsed), record any of the sensitive CR bits that are set, and
489  * enable CR pinning.
490  */
491 static void __init setup_cr_pinning(void)
492 {
493 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & cr4_pinned_mask;
494 	static_key_enable(&cr_pinning.key);
495 }
496 
497 static __init int x86_nofsgsbase_setup(char *arg)
498 {
499 	/* Require an exact match without trailing characters. */
500 	if (strlen(arg))
501 		return 0;
502 
503 	/* Do not emit a message if the feature is not present. */
504 	if (!boot_cpu_has(X86_FEATURE_FSGSBASE))
505 		return 1;
506 
507 	setup_clear_cpu_cap(X86_FEATURE_FSGSBASE);
508 	pr_info("FSGSBASE disabled via kernel command line\n");
509 	return 1;
510 }
511 __setup("nofsgsbase", x86_nofsgsbase_setup);
512 
513 /*
514  * Protection Keys are not available in 32-bit mode.
515  */
516 static bool pku_disabled;
517 
518 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
519 {
520 	if (c == &boot_cpu_data) {
521 		if (pku_disabled || !cpu_feature_enabled(X86_FEATURE_PKU))
522 			return;
523 		/*
524 		 * Setting CR4.PKE will cause the X86_FEATURE_OSPKE cpuid
525 		 * bit to be set.  Enforce it.
526 		 */
527 		setup_force_cpu_cap(X86_FEATURE_OSPKE);
528 
529 	} else if (!cpu_feature_enabled(X86_FEATURE_OSPKE)) {
530 		return;
531 	}
532 
533 	cr4_set_bits(X86_CR4_PKE);
534 	/* Load the default PKRU value */
535 	pkru_write_default();
536 }
537 
538 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
539 static __init int setup_disable_pku(char *arg)
540 {
541 	/*
542 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
543 	 * runtime checks are against OSPKE so clearing the
544 	 * bit does nothing.
545 	 *
546 	 * This way, we will see "pku" in cpuinfo, but not
547 	 * "ospke", which is exactly what we want.  It shows
548 	 * that the CPU has PKU, but the OS has not enabled it.
549 	 * This happens to be exactly how a system would look
550 	 * if we disabled the config option.
551 	 */
552 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
553 	pku_disabled = true;
554 	return 1;
555 }
556 __setup("nopku", setup_disable_pku);
557 #endif
558 
559 #ifdef CONFIG_X86_KERNEL_IBT
560 
561 __noendbr u64 ibt_save(bool disable)
562 {
563 	u64 msr = 0;
564 
565 	if (cpu_feature_enabled(X86_FEATURE_IBT)) {
566 		rdmsrl(MSR_IA32_S_CET, msr);
567 		if (disable)
568 			wrmsrl(MSR_IA32_S_CET, msr & ~CET_ENDBR_EN);
569 	}
570 
571 	return msr;
572 }
573 
574 __noendbr void ibt_restore(u64 save)
575 {
576 	u64 msr;
577 
578 	if (cpu_feature_enabled(X86_FEATURE_IBT)) {
579 		rdmsrl(MSR_IA32_S_CET, msr);
580 		msr &= ~CET_ENDBR_EN;
581 		msr |= (save & CET_ENDBR_EN);
582 		wrmsrl(MSR_IA32_S_CET, msr);
583 	}
584 }
585 
586 #endif
587 
588 static __always_inline void setup_cet(struct cpuinfo_x86 *c)
589 {
590 	bool user_shstk, kernel_ibt;
591 
592 	if (!IS_ENABLED(CONFIG_X86_CET))
593 		return;
594 
595 	kernel_ibt = HAS_KERNEL_IBT && cpu_feature_enabled(X86_FEATURE_IBT);
596 	user_shstk = cpu_feature_enabled(X86_FEATURE_SHSTK) &&
597 		     IS_ENABLED(CONFIG_X86_USER_SHADOW_STACK);
598 
599 	if (!kernel_ibt && !user_shstk)
600 		return;
601 
602 	if (user_shstk)
603 		set_cpu_cap(c, X86_FEATURE_USER_SHSTK);
604 
605 	if (kernel_ibt)
606 		wrmsrl(MSR_IA32_S_CET, CET_ENDBR_EN);
607 	else
608 		wrmsrl(MSR_IA32_S_CET, 0);
609 
610 	cr4_set_bits(X86_CR4_CET);
611 
612 	if (kernel_ibt && 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 	}
617 }
618 
619 __noendbr void cet_disable(void)
620 {
621 	if (!(cpu_feature_enabled(X86_FEATURE_IBT) ||
622 	      cpu_feature_enabled(X86_FEATURE_SHSTK)))
623 		return;
624 
625 	wrmsrl(MSR_IA32_S_CET, 0);
626 	wrmsrl(MSR_IA32_U_CET, 0);
627 }
628 
629 /*
630  * Some CPU features depend on higher CPUID levels, which may not always
631  * be available due to CPUID level capping or broken virtualization
632  * software.  Add those features to this table to auto-disable them.
633  */
634 struct cpuid_dependent_feature {
635 	u32 feature;
636 	u32 level;
637 };
638 
639 static const struct cpuid_dependent_feature
640 cpuid_dependent_features[] = {
641 	{ X86_FEATURE_MWAIT,		0x00000005 },
642 	{ X86_FEATURE_DCA,		0x00000009 },
643 	{ X86_FEATURE_XSAVE,		0x0000000d },
644 	{ 0, 0 }
645 };
646 
647 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
648 {
649 	const struct cpuid_dependent_feature *df;
650 
651 	for (df = cpuid_dependent_features; df->feature; df++) {
652 
653 		if (!cpu_has(c, df->feature))
654 			continue;
655 		/*
656 		 * Note: cpuid_level is set to -1 if unavailable, but
657 		 * extended_extended_level is set to 0 if unavailable
658 		 * and the legitimate extended levels are all negative
659 		 * when signed; hence the weird messing around with
660 		 * signs here...
661 		 */
662 		if (!((s32)df->level < 0 ?
663 		     (u32)df->level > (u32)c->extended_cpuid_level :
664 		     (s32)df->level > (s32)c->cpuid_level))
665 			continue;
666 
667 		clear_cpu_cap(c, df->feature);
668 		if (!warn)
669 			continue;
670 
671 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
672 			x86_cap_flag(df->feature), df->level);
673 	}
674 }
675 
676 /*
677  * Naming convention should be: <Name> [(<Codename>)]
678  * This table only is used unless init_<vendor>() below doesn't set it;
679  * in particular, if CPUID levels 0x80000002..4 are supported, this
680  * isn't used
681  */
682 
683 /* Look up CPU names by table lookup. */
684 static const char *table_lookup_model(struct cpuinfo_x86 *c)
685 {
686 #ifdef CONFIG_X86_32
687 	const struct legacy_cpu_model_info *info;
688 
689 	if (c->x86_model >= 16)
690 		return NULL;	/* Range check */
691 
692 	if (!this_cpu)
693 		return NULL;
694 
695 	info = this_cpu->legacy_models;
696 
697 	while (info->family) {
698 		if (info->family == c->x86)
699 			return info->model_names[c->x86_model];
700 		info++;
701 	}
702 #endif
703 	return NULL;		/* Not found */
704 }
705 
706 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
707 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
708 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
709 
710 #ifdef CONFIG_X86_32
711 /* The 32-bit entry code needs to find cpu_entry_area. */
712 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
713 #endif
714 
715 /* Load the original GDT from the per-cpu structure */
716 void load_direct_gdt(int cpu)
717 {
718 	struct desc_ptr gdt_descr;
719 
720 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
721 	gdt_descr.size = GDT_SIZE - 1;
722 	load_gdt(&gdt_descr);
723 }
724 EXPORT_SYMBOL_GPL(load_direct_gdt);
725 
726 /* Load a fixmap remapping of the per-cpu GDT */
727 void load_fixmap_gdt(int cpu)
728 {
729 	struct desc_ptr gdt_descr;
730 
731 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
732 	gdt_descr.size = GDT_SIZE - 1;
733 	load_gdt(&gdt_descr);
734 }
735 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
736 
737 /**
738  * switch_gdt_and_percpu_base - Switch to direct GDT and runtime per CPU base
739  * @cpu:	The CPU number for which this is invoked
740  *
741  * Invoked during early boot to switch from early GDT and early per CPU to
742  * the direct GDT and the runtime per CPU area. On 32-bit the percpu base
743  * switch is implicit by loading the direct GDT. On 64bit this requires
744  * to update GSBASE.
745  */
746 void __init switch_gdt_and_percpu_base(int cpu)
747 {
748 	load_direct_gdt(cpu);
749 
750 #ifdef CONFIG_X86_64
751 	/*
752 	 * No need to load %gs. It is already correct.
753 	 *
754 	 * Writing %gs on 64bit would zero GSBASE which would make any per
755 	 * CPU operation up to the point of the wrmsrl() fault.
756 	 *
757 	 * Set GSBASE to the new offset. Until the wrmsrl() happens the
758 	 * early mapping is still valid. That means the GSBASE update will
759 	 * lose any prior per CPU data which was not copied over in
760 	 * setup_per_cpu_areas().
761 	 *
762 	 * This works even with stackprotector enabled because the
763 	 * per CPU stack canary is 0 in both per CPU areas.
764 	 */
765 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
766 #else
767 	/*
768 	 * %fs is already set to __KERNEL_PERCPU, but after switching GDT
769 	 * it is required to load FS again so that the 'hidden' part is
770 	 * updated from the new GDT. Up to this point the early per CPU
771 	 * translation is active. Any content of the early per CPU data
772 	 * which was not copied over in setup_per_cpu_areas() is lost.
773 	 */
774 	loadsegment(fs, __KERNEL_PERCPU);
775 #endif
776 }
777 
778 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
779 
780 static void get_model_name(struct cpuinfo_x86 *c)
781 {
782 	unsigned int *v;
783 	char *p, *q, *s;
784 
785 	if (c->extended_cpuid_level < 0x80000004)
786 		return;
787 
788 	v = (unsigned int *)c->x86_model_id;
789 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
790 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
791 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
792 	c->x86_model_id[48] = 0;
793 
794 	/* Trim whitespace */
795 	p = q = s = &c->x86_model_id[0];
796 
797 	while (*p == ' ')
798 		p++;
799 
800 	while (*p) {
801 		/* Note the last non-whitespace index */
802 		if (!isspace(*p))
803 			s = q;
804 
805 		*q++ = *p++;
806 	}
807 
808 	*(s + 1) = '\0';
809 }
810 
811 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
812 {
813 	unsigned int eax, ebx, ecx, edx;
814 
815 	c->x86_max_cores = 1;
816 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
817 		return;
818 
819 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
820 	if (eax & 0x1f)
821 		c->x86_max_cores = (eax >> 26) + 1;
822 }
823 
824 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
825 {
826 	unsigned int n, dummy, ebx, ecx, edx, l2size;
827 
828 	n = c->extended_cpuid_level;
829 
830 	if (n >= 0x80000005) {
831 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
832 		c->x86_cache_size = (ecx>>24) + (edx>>24);
833 #ifdef CONFIG_X86_64
834 		/* On K8 L1 TLB is inclusive, so don't count it */
835 		c->x86_tlbsize = 0;
836 #endif
837 	}
838 
839 	if (n < 0x80000006)	/* Some chips just has a large L1. */
840 		return;
841 
842 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
843 	l2size = ecx >> 16;
844 
845 #ifdef CONFIG_X86_64
846 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
847 #else
848 	/* do processor-specific cache resizing */
849 	if (this_cpu->legacy_cache_size)
850 		l2size = this_cpu->legacy_cache_size(c, l2size);
851 
852 	/* Allow user to override all this if necessary. */
853 	if (cachesize_override != -1)
854 		l2size = cachesize_override;
855 
856 	if (l2size == 0)
857 		return;		/* Again, no L2 cache is possible */
858 #endif
859 
860 	c->x86_cache_size = l2size;
861 }
862 
863 u16 __read_mostly tlb_lli_4k[NR_INFO];
864 u16 __read_mostly tlb_lli_2m[NR_INFO];
865 u16 __read_mostly tlb_lli_4m[NR_INFO];
866 u16 __read_mostly tlb_lld_4k[NR_INFO];
867 u16 __read_mostly tlb_lld_2m[NR_INFO];
868 u16 __read_mostly tlb_lld_4m[NR_INFO];
869 u16 __read_mostly tlb_lld_1g[NR_INFO];
870 
871 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
872 {
873 	if (this_cpu->c_detect_tlb)
874 		this_cpu->c_detect_tlb(c);
875 
876 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
877 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
878 		tlb_lli_4m[ENTRIES]);
879 
880 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
881 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
882 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
883 }
884 
885 int detect_ht_early(struct cpuinfo_x86 *c)
886 {
887 #ifdef CONFIG_SMP
888 	u32 eax, ebx, ecx, edx;
889 
890 	if (!cpu_has(c, X86_FEATURE_HT))
891 		return -1;
892 
893 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
894 		return -1;
895 
896 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
897 		return -1;
898 
899 	cpuid(1, &eax, &ebx, &ecx, &edx);
900 
901 	smp_num_siblings = (ebx & 0xff0000) >> 16;
902 	if (smp_num_siblings == 1)
903 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
904 #endif
905 	return 0;
906 }
907 
908 void detect_ht(struct cpuinfo_x86 *c)
909 {
910 #ifdef CONFIG_SMP
911 	int index_msb, core_bits;
912 
913 	if (detect_ht_early(c) < 0)
914 		return;
915 
916 	index_msb = get_count_order(smp_num_siblings);
917 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
918 
919 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
920 
921 	index_msb = get_count_order(smp_num_siblings);
922 
923 	core_bits = get_count_order(c->x86_max_cores);
924 
925 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
926 				       ((1 << core_bits) - 1);
927 #endif
928 }
929 
930 static void get_cpu_vendor(struct cpuinfo_x86 *c)
931 {
932 	char *v = c->x86_vendor_id;
933 	int i;
934 
935 	for (i = 0; i < X86_VENDOR_NUM; i++) {
936 		if (!cpu_devs[i])
937 			break;
938 
939 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
940 		    (cpu_devs[i]->c_ident[1] &&
941 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
942 
943 			this_cpu = cpu_devs[i];
944 			c->x86_vendor = this_cpu->c_x86_vendor;
945 			return;
946 		}
947 	}
948 
949 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
950 		    "CPU: Your system may be unstable.\n", v);
951 
952 	c->x86_vendor = X86_VENDOR_UNKNOWN;
953 	this_cpu = &default_cpu;
954 }
955 
956 void cpu_detect(struct cpuinfo_x86 *c)
957 {
958 	/* Get vendor name */
959 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
960 	      (unsigned int *)&c->x86_vendor_id[0],
961 	      (unsigned int *)&c->x86_vendor_id[8],
962 	      (unsigned int *)&c->x86_vendor_id[4]);
963 
964 	c->x86 = 4;
965 	/* Intel-defined flags: level 0x00000001 */
966 	if (c->cpuid_level >= 0x00000001) {
967 		u32 junk, tfms, cap0, misc;
968 
969 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
970 		c->x86		= x86_family(tfms);
971 		c->x86_model	= x86_model(tfms);
972 		c->x86_stepping	= x86_stepping(tfms);
973 
974 		if (cap0 & (1<<19)) {
975 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
976 			c->x86_cache_alignment = c->x86_clflush_size;
977 		}
978 	}
979 }
980 
981 static void apply_forced_caps(struct cpuinfo_x86 *c)
982 {
983 	int i;
984 
985 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
986 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
987 		c->x86_capability[i] |= cpu_caps_set[i];
988 	}
989 }
990 
991 static void init_speculation_control(struct cpuinfo_x86 *c)
992 {
993 	/*
994 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
995 	 * and they also have a different bit for STIBP support. Also,
996 	 * a hypervisor might have set the individual AMD bits even on
997 	 * Intel CPUs, for finer-grained selection of what's available.
998 	 */
999 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
1000 		set_cpu_cap(c, X86_FEATURE_IBRS);
1001 		set_cpu_cap(c, X86_FEATURE_IBPB);
1002 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1003 	}
1004 
1005 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
1006 		set_cpu_cap(c, X86_FEATURE_STIBP);
1007 
1008 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
1009 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
1010 		set_cpu_cap(c, X86_FEATURE_SSBD);
1011 
1012 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
1013 		set_cpu_cap(c, X86_FEATURE_IBRS);
1014 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1015 	}
1016 
1017 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
1018 		set_cpu_cap(c, X86_FEATURE_IBPB);
1019 
1020 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
1021 		set_cpu_cap(c, X86_FEATURE_STIBP);
1022 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1023 	}
1024 
1025 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
1026 		set_cpu_cap(c, X86_FEATURE_SSBD);
1027 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
1028 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
1029 	}
1030 }
1031 
1032 void get_cpu_cap(struct cpuinfo_x86 *c)
1033 {
1034 	u32 eax, ebx, ecx, edx;
1035 
1036 	/* Intel-defined flags: level 0x00000001 */
1037 	if (c->cpuid_level >= 0x00000001) {
1038 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
1039 
1040 		c->x86_capability[CPUID_1_ECX] = ecx;
1041 		c->x86_capability[CPUID_1_EDX] = edx;
1042 	}
1043 
1044 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
1045 	if (c->cpuid_level >= 0x00000006)
1046 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
1047 
1048 	/* Additional Intel-defined flags: level 0x00000007 */
1049 	if (c->cpuid_level >= 0x00000007) {
1050 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
1051 		c->x86_capability[CPUID_7_0_EBX] = ebx;
1052 		c->x86_capability[CPUID_7_ECX] = ecx;
1053 		c->x86_capability[CPUID_7_EDX] = edx;
1054 
1055 		/* Check valid sub-leaf index before accessing it */
1056 		if (eax >= 1) {
1057 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
1058 			c->x86_capability[CPUID_7_1_EAX] = eax;
1059 		}
1060 	}
1061 
1062 	/* Extended state features: level 0x0000000d */
1063 	if (c->cpuid_level >= 0x0000000d) {
1064 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
1065 
1066 		c->x86_capability[CPUID_D_1_EAX] = eax;
1067 	}
1068 
1069 	/* AMD-defined flags: level 0x80000001 */
1070 	eax = cpuid_eax(0x80000000);
1071 	c->extended_cpuid_level = eax;
1072 
1073 	if ((eax & 0xffff0000) == 0x80000000) {
1074 		if (eax >= 0x80000001) {
1075 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
1076 
1077 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
1078 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
1079 		}
1080 	}
1081 
1082 	if (c->extended_cpuid_level >= 0x80000007) {
1083 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
1084 
1085 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
1086 		c->x86_power = edx;
1087 	}
1088 
1089 	if (c->extended_cpuid_level >= 0x80000008) {
1090 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
1091 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
1092 	}
1093 
1094 	if (c->extended_cpuid_level >= 0x8000000a)
1095 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
1096 
1097 	if (c->extended_cpuid_level >= 0x8000001f)
1098 		c->x86_capability[CPUID_8000_001F_EAX] = cpuid_eax(0x8000001f);
1099 
1100 	if (c->extended_cpuid_level >= 0x80000021)
1101 		c->x86_capability[CPUID_8000_0021_EAX] = cpuid_eax(0x80000021);
1102 
1103 	init_scattered_cpuid_features(c);
1104 	init_speculation_control(c);
1105 
1106 	/*
1107 	 * Clear/Set all flags overridden by options, after probe.
1108 	 * This needs to happen each time we re-probe, which may happen
1109 	 * several times during CPU initialization.
1110 	 */
1111 	apply_forced_caps(c);
1112 }
1113 
1114 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
1115 {
1116 	u32 eax, ebx, ecx, edx;
1117 
1118 	if (c->extended_cpuid_level >= 0x80000008) {
1119 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
1120 
1121 		c->x86_virt_bits = (eax >> 8) & 0xff;
1122 		c->x86_phys_bits = eax & 0xff;
1123 	}
1124 #ifdef CONFIG_X86_32
1125 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
1126 		c->x86_phys_bits = 36;
1127 #endif
1128 	c->x86_cache_bits = c->x86_phys_bits;
1129 }
1130 
1131 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
1132 {
1133 #ifdef CONFIG_X86_32
1134 	int i;
1135 
1136 	/*
1137 	 * First of all, decide if this is a 486 or higher
1138 	 * It's a 486 if we can modify the AC flag
1139 	 */
1140 	if (flag_is_changeable_p(X86_EFLAGS_AC))
1141 		c->x86 = 4;
1142 	else
1143 		c->x86 = 3;
1144 
1145 	for (i = 0; i < X86_VENDOR_NUM; i++)
1146 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
1147 			c->x86_vendor_id[0] = 0;
1148 			cpu_devs[i]->c_identify(c);
1149 			if (c->x86_vendor_id[0]) {
1150 				get_cpu_vendor(c);
1151 				break;
1152 			}
1153 		}
1154 #endif
1155 }
1156 
1157 #define NO_SPECULATION		BIT(0)
1158 #define NO_MELTDOWN		BIT(1)
1159 #define NO_SSB			BIT(2)
1160 #define NO_L1TF			BIT(3)
1161 #define NO_MDS			BIT(4)
1162 #define MSBDS_ONLY		BIT(5)
1163 #define NO_SWAPGS		BIT(6)
1164 #define NO_ITLB_MULTIHIT	BIT(7)
1165 #define NO_SPECTRE_V2		BIT(8)
1166 #define NO_MMIO			BIT(9)
1167 #define NO_EIBRS_PBRSB		BIT(10)
1168 #define NO_BHI			BIT(11)
1169 
1170 #define VULNWL(vendor, family, model, whitelist)	\
1171 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist)
1172 
1173 #define VULNWL_INTEL(model, whitelist)		\
1174 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1175 
1176 #define VULNWL_AMD(family, whitelist)		\
1177 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1178 
1179 #define VULNWL_HYGON(family, whitelist)		\
1180 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1181 
1182 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1183 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1184 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1185 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1186 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1187 	VULNWL(VORTEX,	5, X86_MODEL_ANY,	NO_SPECULATION),
1188 	VULNWL(VORTEX,	6, X86_MODEL_ANY,	NO_SPECULATION),
1189 
1190 	/* Intel Family 6 */
1191 	VULNWL_INTEL(TIGERLAKE,			NO_MMIO),
1192 	VULNWL_INTEL(TIGERLAKE_L,		NO_MMIO),
1193 	VULNWL_INTEL(ALDERLAKE,			NO_MMIO),
1194 	VULNWL_INTEL(ALDERLAKE_L,		NO_MMIO),
1195 
1196 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1197 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION | NO_ITLB_MULTIHIT),
1198 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1199 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1200 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1201 
1202 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1203 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1204 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1205 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1206 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1207 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1208 
1209 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1210 
1211 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1212 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1213 
1214 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1215 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO),
1216 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB),
1217 
1218 	/*
1219 	 * Technically, swapgs isn't serializing on AMD (despite it previously
1220 	 * being documented as such in the APM).  But according to AMD, %gs is
1221 	 * updated non-speculatively, and the issuing of %gs-relative memory
1222 	 * operands will be blocked until the %gs update completes, which is
1223 	 * good enough for our purposes.
1224 	 */
1225 
1226 	VULNWL_INTEL(ATOM_TREMONT,		NO_EIBRS_PBRSB),
1227 	VULNWL_INTEL(ATOM_TREMONT_L,		NO_EIBRS_PBRSB),
1228 	VULNWL_INTEL(ATOM_TREMONT_D,		NO_ITLB_MULTIHIT | NO_EIBRS_PBRSB),
1229 
1230 	/* AMD Family 0xf - 0x12 */
1231 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_BHI),
1232 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_BHI),
1233 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_BHI),
1234 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_BHI),
1235 
1236 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1237 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB | NO_BHI),
1238 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT | NO_MMIO | NO_EIBRS_PBRSB | NO_BHI),
1239 
1240 	/* Zhaoxin Family 7 */
1241 	VULNWL(CENTAUR,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO | NO_BHI),
1242 	VULNWL(ZHAOXIN,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS | NO_MMIO | NO_BHI),
1243 	{}
1244 };
1245 
1246 #define VULNBL(vendor, family, model, blacklist)	\
1247 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, blacklist)
1248 
1249 #define VULNBL_INTEL_STEPPINGS(model, steppings, issues)		   \
1250 	X86_MATCH_VENDOR_FAM_MODEL_STEPPINGS_FEATURE(INTEL, 6,		   \
1251 					    INTEL_FAM6_##model, steppings, \
1252 					    X86_FEATURE_ANY, issues)
1253 
1254 #define VULNBL_AMD(family, blacklist)		\
1255 	VULNBL(AMD, family, X86_MODEL_ANY, blacklist)
1256 
1257 #define VULNBL_HYGON(family, blacklist)		\
1258 	VULNBL(HYGON, family, X86_MODEL_ANY, blacklist)
1259 
1260 #define SRBDS		BIT(0)
1261 /* CPU is affected by X86_BUG_MMIO_STALE_DATA */
1262 #define MMIO		BIT(1)
1263 /* CPU is affected by Shared Buffers Data Sampling (SBDS), a variant of X86_BUG_MMIO_STALE_DATA */
1264 #define MMIO_SBDS	BIT(2)
1265 /* CPU is affected by RETbleed, speculating where you would not expect it */
1266 #define RETBLEED	BIT(3)
1267 /* CPU is affected by SMT (cross-thread) return predictions */
1268 #define SMT_RSB		BIT(4)
1269 /* CPU is affected by SRSO */
1270 #define SRSO		BIT(5)
1271 /* CPU is affected by GDS */
1272 #define GDS		BIT(6)
1273 /* CPU is affected by Register File Data Sampling */
1274 #define RFDS		BIT(7)
1275 
1276 static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = {
1277 	VULNBL_INTEL_STEPPINGS(IVYBRIDGE,	X86_STEPPING_ANY,		SRBDS),
1278 	VULNBL_INTEL_STEPPINGS(HASWELL,		X86_STEPPING_ANY,		SRBDS),
1279 	VULNBL_INTEL_STEPPINGS(HASWELL_L,	X86_STEPPING_ANY,		SRBDS),
1280 	VULNBL_INTEL_STEPPINGS(HASWELL_G,	X86_STEPPING_ANY,		SRBDS),
1281 	VULNBL_INTEL_STEPPINGS(HASWELL_X,	X86_STEPPING_ANY,		MMIO),
1282 	VULNBL_INTEL_STEPPINGS(BROADWELL_D,	X86_STEPPING_ANY,		MMIO),
1283 	VULNBL_INTEL_STEPPINGS(BROADWELL_G,	X86_STEPPING_ANY,		SRBDS),
1284 	VULNBL_INTEL_STEPPINGS(BROADWELL_X,	X86_STEPPING_ANY,		MMIO),
1285 	VULNBL_INTEL_STEPPINGS(BROADWELL,	X86_STEPPING_ANY,		SRBDS),
1286 	VULNBL_INTEL_STEPPINGS(SKYLAKE_X,	X86_STEPPING_ANY,		MMIO | RETBLEED | GDS),
1287 	VULNBL_INTEL_STEPPINGS(SKYLAKE_L,	X86_STEPPING_ANY,		MMIO | RETBLEED | GDS | SRBDS),
1288 	VULNBL_INTEL_STEPPINGS(SKYLAKE,		X86_STEPPING_ANY,		MMIO | RETBLEED | GDS | SRBDS),
1289 	VULNBL_INTEL_STEPPINGS(KABYLAKE_L,	X86_STEPPING_ANY,		MMIO | RETBLEED | GDS | SRBDS),
1290 	VULNBL_INTEL_STEPPINGS(KABYLAKE,	X86_STEPPING_ANY,		MMIO | RETBLEED | GDS | SRBDS),
1291 	VULNBL_INTEL_STEPPINGS(CANNONLAKE_L,	X86_STEPPING_ANY,		RETBLEED),
1292 	VULNBL_INTEL_STEPPINGS(ICELAKE_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED | GDS),
1293 	VULNBL_INTEL_STEPPINGS(ICELAKE_D,	X86_STEPPING_ANY,		MMIO | GDS),
1294 	VULNBL_INTEL_STEPPINGS(ICELAKE_X,	X86_STEPPING_ANY,		MMIO | GDS),
1295 	VULNBL_INTEL_STEPPINGS(COMETLAKE,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED | GDS),
1296 	VULNBL_INTEL_STEPPINGS(COMETLAKE_L,	X86_STEPPINGS(0x0, 0x0),	MMIO | RETBLEED),
1297 	VULNBL_INTEL_STEPPINGS(COMETLAKE_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED | GDS),
1298 	VULNBL_INTEL_STEPPINGS(TIGERLAKE_L,	X86_STEPPING_ANY,		GDS),
1299 	VULNBL_INTEL_STEPPINGS(TIGERLAKE,	X86_STEPPING_ANY,		GDS),
1300 	VULNBL_INTEL_STEPPINGS(LAKEFIELD,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RETBLEED),
1301 	VULNBL_INTEL_STEPPINGS(ROCKETLAKE,	X86_STEPPING_ANY,		MMIO | RETBLEED | GDS),
1302 	VULNBL_INTEL_STEPPINGS(ALDERLAKE,	X86_STEPPING_ANY,		RFDS),
1303 	VULNBL_INTEL_STEPPINGS(ALDERLAKE_L,	X86_STEPPING_ANY,		RFDS),
1304 	VULNBL_INTEL_STEPPINGS(RAPTORLAKE,	X86_STEPPING_ANY,		RFDS),
1305 	VULNBL_INTEL_STEPPINGS(RAPTORLAKE_P,	X86_STEPPING_ANY,		RFDS),
1306 	VULNBL_INTEL_STEPPINGS(RAPTORLAKE_S,	X86_STEPPING_ANY,		RFDS),
1307 	VULNBL_INTEL_STEPPINGS(ATOM_GRACEMONT,	X86_STEPPING_ANY,		RFDS),
1308 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RFDS),
1309 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_D,	X86_STEPPING_ANY,		MMIO | RFDS),
1310 	VULNBL_INTEL_STEPPINGS(ATOM_TREMONT_L,	X86_STEPPING_ANY,		MMIO | MMIO_SBDS | RFDS),
1311 	VULNBL_INTEL_STEPPINGS(ATOM_GOLDMONT,	X86_STEPPING_ANY,		RFDS),
1312 	VULNBL_INTEL_STEPPINGS(ATOM_GOLDMONT_D,	X86_STEPPING_ANY,		RFDS),
1313 	VULNBL_INTEL_STEPPINGS(ATOM_GOLDMONT_PLUS, X86_STEPPING_ANY,		RFDS),
1314 
1315 	VULNBL_AMD(0x15, RETBLEED),
1316 	VULNBL_AMD(0x16, RETBLEED),
1317 	VULNBL_AMD(0x17, RETBLEED | SMT_RSB | SRSO),
1318 	VULNBL_HYGON(0x18, RETBLEED | SMT_RSB | SRSO),
1319 	VULNBL_AMD(0x19, SRSO),
1320 	{}
1321 };
1322 
1323 static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which)
1324 {
1325 	const struct x86_cpu_id *m = x86_match_cpu(table);
1326 
1327 	return m && !!(m->driver_data & which);
1328 }
1329 
1330 u64 x86_read_arch_cap_msr(void)
1331 {
1332 	u64 x86_arch_cap_msr = 0;
1333 
1334 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1335 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, x86_arch_cap_msr);
1336 
1337 	return x86_arch_cap_msr;
1338 }
1339 
1340 static bool arch_cap_mmio_immune(u64 x86_arch_cap_msr)
1341 {
1342 	return (x86_arch_cap_msr & ARCH_CAP_FBSDP_NO &&
1343 		x86_arch_cap_msr & ARCH_CAP_PSDP_NO &&
1344 		x86_arch_cap_msr & ARCH_CAP_SBDR_SSDP_NO);
1345 }
1346 
1347 static bool __init vulnerable_to_rfds(u64 x86_arch_cap_msr)
1348 {
1349 	/* The "immunity" bit trumps everything else: */
1350 	if (x86_arch_cap_msr & ARCH_CAP_RFDS_NO)
1351 		return false;
1352 
1353 	/*
1354 	 * VMMs set ARCH_CAP_RFDS_CLEAR for processors not in the blacklist to
1355 	 * indicate that mitigation is needed because guest is running on a
1356 	 * vulnerable hardware or may migrate to such hardware:
1357 	 */
1358 	if (x86_arch_cap_msr & ARCH_CAP_RFDS_CLEAR)
1359 		return true;
1360 
1361 	/* Only consult the blacklist when there is no enumeration: */
1362 	return cpu_matches(cpu_vuln_blacklist, RFDS);
1363 }
1364 
1365 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1366 {
1367 	u64 x86_arch_cap_msr = x86_read_arch_cap_msr();
1368 
1369 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1370 	if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) &&
1371 	    !(x86_arch_cap_msr & ARCH_CAP_PSCHANGE_MC_NO))
1372 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1373 
1374 	if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION))
1375 		return;
1376 
1377 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1378 
1379 	if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2))
1380 		setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1381 
1382 	if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) &&
1383 	    !(x86_arch_cap_msr & ARCH_CAP_SSB_NO) &&
1384 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1385 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1386 
1387 	/*
1388 	 * AMD's AutoIBRS is equivalent to Intel's eIBRS - use the Intel feature
1389 	 * flag and protect from vendor-specific bugs via the whitelist.
1390 	 */
1391 	if ((x86_arch_cap_msr & ARCH_CAP_IBRS_ALL) || cpu_has(c, X86_FEATURE_AUTOIBRS)) {
1392 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1393 		if (!cpu_matches(cpu_vuln_whitelist, NO_EIBRS_PBRSB) &&
1394 		    !(x86_arch_cap_msr & ARCH_CAP_PBRSB_NO))
1395 			setup_force_cpu_bug(X86_BUG_EIBRS_PBRSB);
1396 	}
1397 
1398 	if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) &&
1399 	    !(x86_arch_cap_msr & ARCH_CAP_MDS_NO)) {
1400 		setup_force_cpu_bug(X86_BUG_MDS);
1401 		if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY))
1402 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1403 	}
1404 
1405 	if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS))
1406 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1407 
1408 	/*
1409 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1410 	 *	- TSX is supported or
1411 	 *	- TSX_CTRL is present
1412 	 *
1413 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1414 	 * the kernel boot e.g. kexec.
1415 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1416 	 * update is not present or running as guest that don't get TSX_CTRL.
1417 	 */
1418 	if (!(x86_arch_cap_msr & ARCH_CAP_TAA_NO) &&
1419 	    (cpu_has(c, X86_FEATURE_RTM) ||
1420 	     (x86_arch_cap_msr & ARCH_CAP_TSX_CTRL_MSR)))
1421 		setup_force_cpu_bug(X86_BUG_TAA);
1422 
1423 	/*
1424 	 * SRBDS affects CPUs which support RDRAND or RDSEED and are listed
1425 	 * in the vulnerability blacklist.
1426 	 *
1427 	 * Some of the implications and mitigation of Shared Buffers Data
1428 	 * Sampling (SBDS) are similar to SRBDS. Give SBDS same treatment as
1429 	 * SRBDS.
1430 	 */
1431 	if ((cpu_has(c, X86_FEATURE_RDRAND) ||
1432 	     cpu_has(c, X86_FEATURE_RDSEED)) &&
1433 	    cpu_matches(cpu_vuln_blacklist, SRBDS | MMIO_SBDS))
1434 		    setup_force_cpu_bug(X86_BUG_SRBDS);
1435 
1436 	/*
1437 	 * Processor MMIO Stale Data bug enumeration
1438 	 *
1439 	 * Affected CPU list is generally enough to enumerate the vulnerability,
1440 	 * but for virtualization case check for ARCH_CAP MSR bits also, VMM may
1441 	 * not want the guest to enumerate the bug.
1442 	 *
1443 	 * Set X86_BUG_MMIO_UNKNOWN for CPUs that are neither in the blacklist,
1444 	 * nor in the whitelist and also don't enumerate MSR ARCH_CAP MMIO bits.
1445 	 */
1446 	if (!arch_cap_mmio_immune(x86_arch_cap_msr)) {
1447 		if (cpu_matches(cpu_vuln_blacklist, MMIO))
1448 			setup_force_cpu_bug(X86_BUG_MMIO_STALE_DATA);
1449 		else if (!cpu_matches(cpu_vuln_whitelist, NO_MMIO))
1450 			setup_force_cpu_bug(X86_BUG_MMIO_UNKNOWN);
1451 	}
1452 
1453 	if (!cpu_has(c, X86_FEATURE_BTC_NO)) {
1454 		if (cpu_matches(cpu_vuln_blacklist, RETBLEED) || (x86_arch_cap_msr & ARCH_CAP_RSBA))
1455 			setup_force_cpu_bug(X86_BUG_RETBLEED);
1456 	}
1457 
1458 	if (cpu_matches(cpu_vuln_blacklist, SMT_RSB))
1459 		setup_force_cpu_bug(X86_BUG_SMT_RSB);
1460 
1461 	if (!cpu_has(c, X86_FEATURE_SRSO_NO)) {
1462 		if (cpu_matches(cpu_vuln_blacklist, SRSO))
1463 			setup_force_cpu_bug(X86_BUG_SRSO);
1464 	}
1465 
1466 	/*
1467 	 * Check if CPU is vulnerable to GDS. If running in a virtual machine on
1468 	 * an affected processor, the VMM may have disabled the use of GATHER by
1469 	 * disabling AVX2. The only way to do this in HW is to clear XCR0[2],
1470 	 * which means that AVX will be disabled.
1471 	 */
1472 	if (cpu_matches(cpu_vuln_blacklist, GDS) && !(x86_arch_cap_msr & ARCH_CAP_GDS_NO) &&
1473 	    boot_cpu_has(X86_FEATURE_AVX))
1474 		setup_force_cpu_bug(X86_BUG_GDS);
1475 
1476 	if (vulnerable_to_rfds(x86_arch_cap_msr))
1477 		setup_force_cpu_bug(X86_BUG_RFDS);
1478 
1479 	/* When virtualized, eIBRS could be hidden, assume vulnerable */
1480 	if (!(x86_arch_cap_msr & ARCH_CAP_BHI_NO) &&
1481 	    !cpu_matches(cpu_vuln_whitelist, NO_BHI) &&
1482 	    (boot_cpu_has(X86_FEATURE_IBRS_ENHANCED) ||
1483 	     boot_cpu_has(X86_FEATURE_HYPERVISOR)))
1484 		setup_force_cpu_bug(X86_BUG_BHI);
1485 
1486 	if (cpu_has(c, X86_FEATURE_AMD_IBPB) && !cpu_has(c, X86_FEATURE_AMD_IBPB_RET))
1487 		setup_force_cpu_bug(X86_BUG_IBPB_NO_RET);
1488 
1489 	if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN))
1490 		return;
1491 
1492 	/* Rogue Data Cache Load? No! */
1493 	if (x86_arch_cap_msr & ARCH_CAP_RDCL_NO)
1494 		return;
1495 
1496 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1497 
1498 	if (cpu_matches(cpu_vuln_whitelist, NO_L1TF))
1499 		return;
1500 
1501 	setup_force_cpu_bug(X86_BUG_L1TF);
1502 }
1503 
1504 /*
1505  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1506  * unfortunately, that's not true in practice because of early VIA
1507  * chips and (more importantly) broken virtualizers that are not easy
1508  * to detect. In the latter case it doesn't even *fail* reliably, so
1509  * probing for it doesn't even work. Disable it completely on 32-bit
1510  * unless we can find a reliable way to detect all the broken cases.
1511  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1512  */
1513 static void detect_nopl(void)
1514 {
1515 #ifdef CONFIG_X86_32
1516 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1517 #else
1518 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1519 #endif
1520 }
1521 
1522 /*
1523  * We parse cpu parameters early because fpu__init_system() is executed
1524  * before parse_early_param().
1525  */
1526 static void __init cpu_parse_early_param(void)
1527 {
1528 	char arg[128];
1529 	char *argptr = arg, *opt;
1530 	int arglen, taint = 0;
1531 
1532 #ifdef CONFIG_X86_32
1533 	if (cmdline_find_option_bool(boot_command_line, "no387"))
1534 #ifdef CONFIG_MATH_EMULATION
1535 		setup_clear_cpu_cap(X86_FEATURE_FPU);
1536 #else
1537 		pr_err("Option 'no387' required CONFIG_MATH_EMULATION enabled.\n");
1538 #endif
1539 
1540 	if (cmdline_find_option_bool(boot_command_line, "nofxsr"))
1541 		setup_clear_cpu_cap(X86_FEATURE_FXSR);
1542 #endif
1543 
1544 	if (cmdline_find_option_bool(boot_command_line, "noxsave"))
1545 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
1546 
1547 	if (cmdline_find_option_bool(boot_command_line, "noxsaveopt"))
1548 		setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
1549 
1550 	if (cmdline_find_option_bool(boot_command_line, "noxsaves"))
1551 		setup_clear_cpu_cap(X86_FEATURE_XSAVES);
1552 
1553 	if (cmdline_find_option_bool(boot_command_line, "nousershstk"))
1554 		setup_clear_cpu_cap(X86_FEATURE_USER_SHSTK);
1555 
1556 	arglen = cmdline_find_option(boot_command_line, "clearcpuid", arg, sizeof(arg));
1557 	if (arglen <= 0)
1558 		return;
1559 
1560 	pr_info("Clearing CPUID bits:");
1561 
1562 	while (argptr) {
1563 		bool found __maybe_unused = false;
1564 		unsigned int bit;
1565 
1566 		opt = strsep(&argptr, ",");
1567 
1568 		/*
1569 		 * Handle naked numbers first for feature flags which don't
1570 		 * have names.
1571 		 */
1572 		if (!kstrtouint(opt, 10, &bit)) {
1573 			if (bit < NCAPINTS * 32) {
1574 
1575 				/* empty-string, i.e., ""-defined feature flags */
1576 				if (!x86_cap_flags[bit])
1577 					pr_cont(" " X86_CAP_FMT_NUM, x86_cap_flag_num(bit));
1578 				else
1579 					pr_cont(" " X86_CAP_FMT, x86_cap_flag(bit));
1580 
1581 				setup_clear_cpu_cap(bit);
1582 				taint++;
1583 			}
1584 			/*
1585 			 * The assumption is that there are no feature names with only
1586 			 * numbers in the name thus go to the next argument.
1587 			 */
1588 			continue;
1589 		}
1590 
1591 		for (bit = 0; bit < 32 * NCAPINTS; bit++) {
1592 			if (!x86_cap_flag(bit))
1593 				continue;
1594 
1595 			if (strcmp(x86_cap_flag(bit), opt))
1596 				continue;
1597 
1598 			pr_cont(" %s", opt);
1599 			setup_clear_cpu_cap(bit);
1600 			taint++;
1601 			found = true;
1602 			break;
1603 		}
1604 
1605 		if (!found)
1606 			pr_cont(" (unknown: %s)", opt);
1607 	}
1608 	pr_cont("\n");
1609 
1610 	if (taint)
1611 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_STILL_OK);
1612 }
1613 
1614 /*
1615  * Do minimum CPU detection early.
1616  * Fields really needed: vendor, cpuid_level, family, model, mask,
1617  * cache alignment.
1618  * The others are not touched to avoid unwanted side effects.
1619  *
1620  * WARNING: this function is only called on the boot CPU.  Don't add code
1621  * here that is supposed to run on all CPUs.
1622  */
1623 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1624 {
1625 #ifdef CONFIG_X86_64
1626 	c->x86_clflush_size = 64;
1627 	c->x86_phys_bits = 36;
1628 	c->x86_virt_bits = 48;
1629 #else
1630 	c->x86_clflush_size = 32;
1631 	c->x86_phys_bits = 32;
1632 	c->x86_virt_bits = 32;
1633 #endif
1634 	c->x86_cache_alignment = c->x86_clflush_size;
1635 
1636 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1637 	c->extended_cpuid_level = 0;
1638 
1639 	if (!have_cpuid_p())
1640 		identify_cpu_without_cpuid(c);
1641 
1642 	/* cyrix could have cpuid enabled via c_identify()*/
1643 	if (have_cpuid_p()) {
1644 		cpu_detect(c);
1645 		get_cpu_vendor(c);
1646 		get_cpu_cap(c);
1647 		get_cpu_address_sizes(c);
1648 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1649 		cpu_parse_early_param();
1650 
1651 		if (this_cpu->c_early_init)
1652 			this_cpu->c_early_init(c);
1653 
1654 		c->cpu_index = 0;
1655 		filter_cpuid_features(c, false);
1656 
1657 		if (this_cpu->c_bsp_init)
1658 			this_cpu->c_bsp_init(c);
1659 	} else {
1660 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1661 	}
1662 
1663 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1664 
1665 	cpu_set_bug_bits(c);
1666 
1667 	sld_setup(c);
1668 
1669 #ifdef CONFIG_X86_32
1670 	/*
1671 	 * Regardless of whether PCID is enumerated, the SDM says
1672 	 * that it can't be enabled in 32-bit mode.
1673 	 */
1674 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1675 #endif
1676 
1677 	/*
1678 	 * Later in the boot process pgtable_l5_enabled() relies on
1679 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1680 	 * enabled by this point we need to clear the feature bit to avoid
1681 	 * false-positives at the later stage.
1682 	 *
1683 	 * pgtable_l5_enabled() can be false here for several reasons:
1684 	 *  - 5-level paging is disabled compile-time;
1685 	 *  - it's 32-bit kernel;
1686 	 *  - machine doesn't support 5-level paging;
1687 	 *  - user specified 'no5lvl' in kernel command line.
1688 	 */
1689 	if (!pgtable_l5_enabled())
1690 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1691 
1692 	detect_nopl();
1693 }
1694 
1695 void __init early_cpu_init(void)
1696 {
1697 	const struct cpu_dev *const *cdev;
1698 	int count = 0;
1699 
1700 #ifdef CONFIG_PROCESSOR_SELECT
1701 	pr_info("KERNEL supported cpus:\n");
1702 #endif
1703 
1704 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1705 		const struct cpu_dev *cpudev = *cdev;
1706 
1707 		if (count >= X86_VENDOR_NUM)
1708 			break;
1709 		cpu_devs[count] = cpudev;
1710 		count++;
1711 
1712 #ifdef CONFIG_PROCESSOR_SELECT
1713 		{
1714 			unsigned int j;
1715 
1716 			for (j = 0; j < 2; j++) {
1717 				if (!cpudev->c_ident[j])
1718 					continue;
1719 				pr_info("  %s %s\n", cpudev->c_vendor,
1720 					cpudev->c_ident[j]);
1721 			}
1722 		}
1723 #endif
1724 	}
1725 	early_identify_cpu(&boot_cpu_data);
1726 }
1727 
1728 static bool detect_null_seg_behavior(void)
1729 {
1730 	/*
1731 	 * Empirically, writing zero to a segment selector on AMD does
1732 	 * not clear the base, whereas writing zero to a segment
1733 	 * selector on Intel does clear the base.  Intel's behavior
1734 	 * allows slightly faster context switches in the common case
1735 	 * where GS is unused by the prev and next threads.
1736 	 *
1737 	 * Since neither vendor documents this anywhere that I can see,
1738 	 * detect it directly instead of hard-coding the choice by
1739 	 * vendor.
1740 	 *
1741 	 * I've designated AMD's behavior as the "bug" because it's
1742 	 * counterintuitive and less friendly.
1743 	 */
1744 
1745 	unsigned long old_base, tmp;
1746 	rdmsrl(MSR_FS_BASE, old_base);
1747 	wrmsrl(MSR_FS_BASE, 1);
1748 	loadsegment(fs, 0);
1749 	rdmsrl(MSR_FS_BASE, tmp);
1750 	wrmsrl(MSR_FS_BASE, old_base);
1751 	return tmp == 0;
1752 }
1753 
1754 void check_null_seg_clears_base(struct cpuinfo_x86 *c)
1755 {
1756 	/* BUG_NULL_SEG is only relevant with 64bit userspace */
1757 	if (!IS_ENABLED(CONFIG_X86_64))
1758 		return;
1759 
1760 	if (cpu_has(c, X86_FEATURE_NULL_SEL_CLR_BASE))
1761 		return;
1762 
1763 	/*
1764 	 * CPUID bit above wasn't set. If this kernel is still running
1765 	 * as a HV guest, then the HV has decided not to advertize
1766 	 * that CPUID bit for whatever reason.	For example, one
1767 	 * member of the migration pool might be vulnerable.  Which
1768 	 * means, the bug is present: set the BUG flag and return.
1769 	 */
1770 	if (cpu_has(c, X86_FEATURE_HYPERVISOR)) {
1771 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1772 		return;
1773 	}
1774 
1775 	/*
1776 	 * Zen2 CPUs also have this behaviour, but no CPUID bit.
1777 	 * 0x18 is the respective family for Hygon.
1778 	 */
1779 	if ((c->x86 == 0x17 || c->x86 == 0x18) &&
1780 	    detect_null_seg_behavior())
1781 		return;
1782 
1783 	/* All the remaining ones are affected */
1784 	set_cpu_bug(c, X86_BUG_NULL_SEG);
1785 }
1786 
1787 static void generic_identify(struct cpuinfo_x86 *c)
1788 {
1789 	c->extended_cpuid_level = 0;
1790 
1791 	if (!have_cpuid_p())
1792 		identify_cpu_without_cpuid(c);
1793 
1794 	/* cyrix could have cpuid enabled via c_identify()*/
1795 	if (!have_cpuid_p())
1796 		return;
1797 
1798 	cpu_detect(c);
1799 
1800 	get_cpu_vendor(c);
1801 
1802 	get_cpu_cap(c);
1803 
1804 	get_cpu_address_sizes(c);
1805 
1806 	if (c->cpuid_level >= 0x00000001) {
1807 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1808 #ifdef CONFIG_X86_32
1809 # ifdef CONFIG_SMP
1810 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1811 # else
1812 		c->apicid = c->initial_apicid;
1813 # endif
1814 #endif
1815 		c->phys_proc_id = c->initial_apicid;
1816 	}
1817 
1818 	get_model_name(c); /* Default name */
1819 
1820 	/*
1821 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1822 	 * systems that run Linux at CPL > 0 may or may not have the
1823 	 * issue, but, even if they have the issue, there's absolutely
1824 	 * nothing we can do about it because we can't use the real IRET
1825 	 * instruction.
1826 	 *
1827 	 * NB: For the time being, only 32-bit kernels support
1828 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1829 	 * whether to apply espfix using paravirt hooks.  If any
1830 	 * non-paravirt system ever shows up that does *not* have the
1831 	 * ESPFIX issue, we can change this.
1832 	 */
1833 #ifdef CONFIG_X86_32
1834 	set_cpu_bug(c, X86_BUG_ESPFIX);
1835 #endif
1836 }
1837 
1838 /*
1839  * Validate that ACPI/mptables have the same information about the
1840  * effective APIC id and update the package map.
1841  */
1842 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1843 {
1844 #ifdef CONFIG_SMP
1845 	unsigned int apicid, cpu = smp_processor_id();
1846 
1847 	apicid = apic->cpu_present_to_apicid(cpu);
1848 
1849 	if (apicid != c->apicid) {
1850 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1851 		       cpu, apicid, c->initial_apicid);
1852 	}
1853 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1854 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1855 #else
1856 	c->logical_proc_id = 0;
1857 #endif
1858 }
1859 
1860 /*
1861  * This does the hard work of actually picking apart the CPU stuff...
1862  */
1863 static void identify_cpu(struct cpuinfo_x86 *c)
1864 {
1865 	int i;
1866 
1867 	c->loops_per_jiffy = loops_per_jiffy;
1868 	c->x86_cache_size = 0;
1869 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1870 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1871 	c->x86_vendor_id[0] = '\0'; /* Unset */
1872 	c->x86_model_id[0] = '\0';  /* Unset */
1873 	c->x86_max_cores = 1;
1874 	c->x86_coreid_bits = 0;
1875 	c->cu_id = 0xff;
1876 #ifdef CONFIG_X86_64
1877 	c->x86_clflush_size = 64;
1878 	c->x86_phys_bits = 36;
1879 	c->x86_virt_bits = 48;
1880 #else
1881 	c->cpuid_level = -1;	/* CPUID not detected */
1882 	c->x86_clflush_size = 32;
1883 	c->x86_phys_bits = 32;
1884 	c->x86_virt_bits = 32;
1885 #endif
1886 	c->x86_cache_alignment = c->x86_clflush_size;
1887 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1888 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
1889 	memset(&c->vmx_capability, 0, sizeof(c->vmx_capability));
1890 #endif
1891 
1892 	generic_identify(c);
1893 
1894 	if (this_cpu->c_identify)
1895 		this_cpu->c_identify(c);
1896 
1897 	/* Clear/Set all flags overridden by options, after probe */
1898 	apply_forced_caps(c);
1899 
1900 #ifdef CONFIG_X86_64
1901 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1902 #endif
1903 
1904 
1905 	/*
1906 	 * Set default APIC and TSC_DEADLINE MSR fencing flag. AMD and
1907 	 * Hygon will clear it in ->c_init() below.
1908 	 */
1909 	set_cpu_cap(c, X86_FEATURE_APIC_MSRS_FENCE);
1910 
1911 	/*
1912 	 * Vendor-specific initialization.  In this section we
1913 	 * canonicalize the feature flags, meaning if there are
1914 	 * features a certain CPU supports which CPUID doesn't
1915 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1916 	 * we handle them here.
1917 	 *
1918 	 * At the end of this section, c->x86_capability better
1919 	 * indicate the features this CPU genuinely supports!
1920 	 */
1921 	if (this_cpu->c_init)
1922 		this_cpu->c_init(c);
1923 
1924 	/* Disable the PN if appropriate */
1925 	squash_the_stupid_serial_number(c);
1926 
1927 	/* Set up SMEP/SMAP/UMIP */
1928 	setup_smep(c);
1929 	setup_smap(c);
1930 	setup_umip(c);
1931 
1932 	/* Enable FSGSBASE instructions if available. */
1933 	if (cpu_has(c, X86_FEATURE_FSGSBASE)) {
1934 		cr4_set_bits(X86_CR4_FSGSBASE);
1935 		elf_hwcap2 |= HWCAP2_FSGSBASE;
1936 	}
1937 
1938 	/*
1939 	 * The vendor-specific functions might have changed features.
1940 	 * Now we do "generic changes."
1941 	 */
1942 
1943 	/* Filter out anything that depends on CPUID levels we don't have */
1944 	filter_cpuid_features(c, true);
1945 
1946 	/* If the model name is still unset, do table lookup. */
1947 	if (!c->x86_model_id[0]) {
1948 		const char *p;
1949 		p = table_lookup_model(c);
1950 		if (p)
1951 			strcpy(c->x86_model_id, p);
1952 		else
1953 			/* Last resort... */
1954 			sprintf(c->x86_model_id, "%02x/%02x",
1955 				c->x86, c->x86_model);
1956 	}
1957 
1958 #ifdef CONFIG_X86_64
1959 	detect_ht(c);
1960 #endif
1961 
1962 	x86_init_rdrand(c);
1963 	setup_pku(c);
1964 	setup_cet(c);
1965 
1966 	/*
1967 	 * Clear/Set all flags overridden by options, need do it
1968 	 * before following smp all cpus cap AND.
1969 	 */
1970 	apply_forced_caps(c);
1971 
1972 	/*
1973 	 * On SMP, boot_cpu_data holds the common feature set between
1974 	 * all CPUs; so make sure that we indicate which features are
1975 	 * common between the CPUs.  The first time this routine gets
1976 	 * executed, c == &boot_cpu_data.
1977 	 */
1978 	if (c != &boot_cpu_data) {
1979 		/* AND the already accumulated flags with these */
1980 		for (i = 0; i < NCAPINTS; i++)
1981 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1982 
1983 		/* OR, i.e. replicate the bug flags */
1984 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1985 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1986 	}
1987 
1988 	ppin_init(c);
1989 
1990 	/* Init Machine Check Exception if available. */
1991 	mcheck_cpu_init(c);
1992 
1993 	select_idle_routine(c);
1994 
1995 #ifdef CONFIG_NUMA
1996 	numa_add_cpu(smp_processor_id());
1997 #endif
1998 }
1999 
2000 /*
2001  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
2002  * on 32-bit kernels:
2003  */
2004 #ifdef CONFIG_X86_32
2005 void enable_sep_cpu(void)
2006 {
2007 	struct tss_struct *tss;
2008 	int cpu;
2009 
2010 	if (!boot_cpu_has(X86_FEATURE_SEP))
2011 		return;
2012 
2013 	cpu = get_cpu();
2014 	tss = &per_cpu(cpu_tss_rw, cpu);
2015 
2016 	/*
2017 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
2018 	 * see the big comment in struct x86_hw_tss's definition.
2019 	 */
2020 
2021 	tss->x86_tss.ss1 = __KERNEL_CS;
2022 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
2023 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
2024 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
2025 
2026 	put_cpu();
2027 }
2028 #endif
2029 
2030 static __init void identify_boot_cpu(void)
2031 {
2032 	identify_cpu(&boot_cpu_data);
2033 	if (HAS_KERNEL_IBT && cpu_feature_enabled(X86_FEATURE_IBT))
2034 		pr_info("CET detected: Indirect Branch Tracking enabled\n");
2035 #ifdef CONFIG_X86_32
2036 	enable_sep_cpu();
2037 #endif
2038 	cpu_detect_tlb(&boot_cpu_data);
2039 	setup_cr_pinning();
2040 
2041 	tsx_init();
2042 	lkgs_init();
2043 }
2044 
2045 void identify_secondary_cpu(struct cpuinfo_x86 *c)
2046 {
2047 	BUG_ON(c == &boot_cpu_data);
2048 	identify_cpu(c);
2049 #ifdef CONFIG_X86_32
2050 	enable_sep_cpu();
2051 #endif
2052 	validate_apic_and_package_id(c);
2053 	x86_spec_ctrl_setup_ap();
2054 	update_srbds_msr();
2055 	if (boot_cpu_has_bug(X86_BUG_GDS))
2056 		update_gds_msr();
2057 
2058 	tsx_ap_init();
2059 }
2060 
2061 void print_cpu_info(struct cpuinfo_x86 *c)
2062 {
2063 	const char *vendor = NULL;
2064 
2065 	if (c->x86_vendor < X86_VENDOR_NUM) {
2066 		vendor = this_cpu->c_vendor;
2067 	} else {
2068 		if (c->cpuid_level >= 0)
2069 			vendor = c->x86_vendor_id;
2070 	}
2071 
2072 	if (vendor && !strstr(c->x86_model_id, vendor))
2073 		pr_cont("%s ", vendor);
2074 
2075 	if (c->x86_model_id[0])
2076 		pr_cont("%s", c->x86_model_id);
2077 	else
2078 		pr_cont("%d86", c->x86);
2079 
2080 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
2081 
2082 	if (c->x86_stepping || c->cpuid_level >= 0)
2083 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
2084 	else
2085 		pr_cont(")\n");
2086 }
2087 
2088 /*
2089  * clearcpuid= was already parsed in cpu_parse_early_param().  This dummy
2090  * function prevents it from becoming an environment variable for init.
2091  */
2092 static __init int setup_clearcpuid(char *arg)
2093 {
2094 	return 1;
2095 }
2096 __setup("clearcpuid=", setup_clearcpuid);
2097 
2098 DEFINE_PER_CPU_ALIGNED(struct pcpu_hot, pcpu_hot) = {
2099 	.current_task	= &init_task,
2100 	.preempt_count	= INIT_PREEMPT_COUNT,
2101 	.top_of_stack	= TOP_OF_INIT_STACK,
2102 };
2103 EXPORT_PER_CPU_SYMBOL(pcpu_hot);
2104 
2105 #ifdef CONFIG_X86_64
2106 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
2107 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
2108 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
2109 
2110 static void wrmsrl_cstar(unsigned long val)
2111 {
2112 	/*
2113 	 * Intel CPUs do not support 32-bit SYSCALL. Writing to MSR_CSTAR
2114 	 * is so far ignored by the CPU, but raises a #VE trap in a TDX
2115 	 * guest. Avoid the pointless write on all Intel CPUs.
2116 	 */
2117 	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
2118 		wrmsrl(MSR_CSTAR, val);
2119 }
2120 
2121 /* May not be marked __init: used by software suspend */
2122 void syscall_init(void)
2123 {
2124 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
2125 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
2126 
2127 #ifdef CONFIG_IA32_EMULATION
2128 	wrmsrl_cstar((unsigned long)entry_SYSCALL_compat);
2129 	/*
2130 	 * This only works on Intel CPUs.
2131 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
2132 	 * This does not cause SYSENTER to jump to the wrong location, because
2133 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
2134 	 */
2135 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
2136 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
2137 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
2138 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
2139 #else
2140 	wrmsrl_cstar((unsigned long)entry_SYSCALL32_ignore);
2141 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
2142 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
2143 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
2144 #endif
2145 
2146 	/*
2147 	 * Flags to clear on syscall; clear as much as possible
2148 	 * to minimize user space-kernel interference.
2149 	 */
2150 	wrmsrl(MSR_SYSCALL_MASK,
2151 	       X86_EFLAGS_CF|X86_EFLAGS_PF|X86_EFLAGS_AF|
2152 	       X86_EFLAGS_ZF|X86_EFLAGS_SF|X86_EFLAGS_TF|
2153 	       X86_EFLAGS_IF|X86_EFLAGS_DF|X86_EFLAGS_OF|
2154 	       X86_EFLAGS_IOPL|X86_EFLAGS_NT|X86_EFLAGS_RF|
2155 	       X86_EFLAGS_AC|X86_EFLAGS_ID);
2156 }
2157 
2158 #else	/* CONFIG_X86_64 */
2159 
2160 #ifdef CONFIG_STACKPROTECTOR
2161 DEFINE_PER_CPU(unsigned long, __stack_chk_guard);
2162 EXPORT_PER_CPU_SYMBOL(__stack_chk_guard);
2163 #endif
2164 
2165 #endif	/* CONFIG_X86_64 */
2166 
2167 /*
2168  * Clear all 6 debug registers:
2169  */
2170 static void clear_all_debug_regs(void)
2171 {
2172 	int i;
2173 
2174 	for (i = 0; i < 8; i++) {
2175 		/* Ignore db4, db5 */
2176 		if ((i == 4) || (i == 5))
2177 			continue;
2178 
2179 		set_debugreg(0, i);
2180 	}
2181 }
2182 
2183 #ifdef CONFIG_KGDB
2184 /*
2185  * Restore debug regs if using kgdbwait and you have a kernel debugger
2186  * connection established.
2187  */
2188 static void dbg_restore_debug_regs(void)
2189 {
2190 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
2191 		arch_kgdb_ops.correct_hw_break();
2192 }
2193 #else /* ! CONFIG_KGDB */
2194 #define dbg_restore_debug_regs()
2195 #endif /* ! CONFIG_KGDB */
2196 
2197 static inline void setup_getcpu(int cpu)
2198 {
2199 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
2200 	struct desc_struct d = { };
2201 
2202 	if (boot_cpu_has(X86_FEATURE_RDTSCP) || boot_cpu_has(X86_FEATURE_RDPID))
2203 		wrmsr(MSR_TSC_AUX, cpudata, 0);
2204 
2205 	/* Store CPU and node number in limit. */
2206 	d.limit0 = cpudata;
2207 	d.limit1 = cpudata >> 16;
2208 
2209 	d.type = 5;		/* RO data, expand down, accessed */
2210 	d.dpl = 3;		/* Visible to user code */
2211 	d.s = 1;		/* Not a system segment */
2212 	d.p = 1;		/* Present */
2213 	d.d = 1;		/* 32-bit */
2214 
2215 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
2216 }
2217 
2218 #ifdef CONFIG_X86_64
2219 static inline void ucode_cpu_init(int cpu) { }
2220 
2221 static inline void tss_setup_ist(struct tss_struct *tss)
2222 {
2223 	/* Set up the per-CPU TSS IST stacks */
2224 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
2225 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
2226 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
2227 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
2228 	/* Only mapped when SEV-ES is active */
2229 	tss->x86_tss.ist[IST_INDEX_VC] = __this_cpu_ist_top_va(VC);
2230 }
2231 
2232 #else /* CONFIG_X86_64 */
2233 
2234 static inline void ucode_cpu_init(int cpu)
2235 {
2236 	show_ucode_info_early();
2237 }
2238 
2239 static inline void tss_setup_ist(struct tss_struct *tss) { }
2240 
2241 #endif /* !CONFIG_X86_64 */
2242 
2243 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
2244 {
2245 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
2246 
2247 #ifdef CONFIG_X86_IOPL_IOPERM
2248 	tss->io_bitmap.prev_max = 0;
2249 	tss->io_bitmap.prev_sequence = 0;
2250 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
2251 	/*
2252 	 * Invalidate the extra array entry past the end of the all
2253 	 * permission bitmap as required by the hardware.
2254 	 */
2255 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
2256 #endif
2257 }
2258 
2259 /*
2260  * Setup everything needed to handle exceptions from the IDT, including the IST
2261  * exceptions which use paranoid_entry().
2262  */
2263 void cpu_init_exception_handling(void)
2264 {
2265 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
2266 	int cpu = raw_smp_processor_id();
2267 
2268 	/* paranoid_entry() gets the CPU number from the GDT */
2269 	setup_getcpu(cpu);
2270 
2271 	/* IST vectors need TSS to be set up. */
2272 	tss_setup_ist(tss);
2273 	tss_setup_io_bitmap(tss);
2274 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
2275 
2276 	load_TR_desc();
2277 
2278 	/* GHCB needs to be setup to handle #VC. */
2279 	setup_ghcb();
2280 
2281 	/* Finally load the IDT */
2282 	load_current_idt();
2283 }
2284 
2285 /*
2286  * cpu_init() initializes state that is per-CPU. Some data is already
2287  * initialized (naturally) in the bootstrap process, such as the GDT.  We
2288  * reload it nevertheless, this function acts as a 'CPU state barrier',
2289  * nothing should get across.
2290  */
2291 void cpu_init(void)
2292 {
2293 	struct task_struct *cur = current;
2294 	int cpu = raw_smp_processor_id();
2295 
2296 	ucode_cpu_init(cpu);
2297 
2298 #ifdef CONFIG_NUMA
2299 	if (this_cpu_read(numa_node) == 0 &&
2300 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
2301 		set_numa_node(early_cpu_to_node(cpu));
2302 #endif
2303 	pr_debug("Initializing CPU#%d\n", cpu);
2304 
2305 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
2306 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
2307 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
2308 
2309 	if (IS_ENABLED(CONFIG_X86_64)) {
2310 		loadsegment(fs, 0);
2311 		memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
2312 		syscall_init();
2313 
2314 		wrmsrl(MSR_FS_BASE, 0);
2315 		wrmsrl(MSR_KERNEL_GS_BASE, 0);
2316 		barrier();
2317 
2318 		x2apic_setup();
2319 	}
2320 
2321 	mmgrab(&init_mm);
2322 	cur->active_mm = &init_mm;
2323 	BUG_ON(cur->mm);
2324 	initialize_tlbstate_and_flush();
2325 	enter_lazy_tlb(&init_mm, cur);
2326 
2327 	/*
2328 	 * sp0 points to the entry trampoline stack regardless of what task
2329 	 * is running.
2330 	 */
2331 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
2332 
2333 	load_mm_ldt(&init_mm);
2334 
2335 	clear_all_debug_regs();
2336 	dbg_restore_debug_regs();
2337 
2338 	doublefault_init_cpu_tss();
2339 
2340 	if (is_uv_system())
2341 		uv_cpu_init();
2342 
2343 	load_fixmap_gdt(cpu);
2344 }
2345 
2346 #ifdef CONFIG_MICROCODE_LATE_LOADING
2347 /**
2348  * store_cpu_caps() - Store a snapshot of CPU capabilities
2349  * @curr_info: Pointer where to store it
2350  *
2351  * Returns: None
2352  */
2353 void store_cpu_caps(struct cpuinfo_x86 *curr_info)
2354 {
2355 	/* Reload CPUID max function as it might've changed. */
2356 	curr_info->cpuid_level = cpuid_eax(0);
2357 
2358 	/* Copy all capability leafs and pick up the synthetic ones. */
2359 	memcpy(&curr_info->x86_capability, &boot_cpu_data.x86_capability,
2360 	       sizeof(curr_info->x86_capability));
2361 
2362 	/* Get the hardware CPUID leafs */
2363 	get_cpu_cap(curr_info);
2364 }
2365 
2366 /**
2367  * microcode_check() - Check if any CPU capabilities changed after an update.
2368  * @prev_info:	CPU capabilities stored before an update.
2369  *
2370  * The microcode loader calls this upon late microcode load to recheck features,
2371  * only when microcode has been updated. Caller holds and CPU hotplug lock.
2372  *
2373  * Return: None
2374  */
2375 void microcode_check(struct cpuinfo_x86 *prev_info)
2376 {
2377 	struct cpuinfo_x86 curr_info;
2378 
2379 	perf_check_microcode();
2380 
2381 	amd_check_microcode();
2382 
2383 	store_cpu_caps(&curr_info);
2384 
2385 	if (!memcmp(&prev_info->x86_capability, &curr_info.x86_capability,
2386 		    sizeof(prev_info->x86_capability)))
2387 		return;
2388 
2389 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
2390 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
2391 }
2392 #endif
2393 
2394 /*
2395  * Invoked from core CPU hotplug code after hotplug operations
2396  */
2397 void arch_smt_update(void)
2398 {
2399 	/* Handle the speculative execution misfeatures */
2400 	cpu_bugs_smt_update();
2401 	/* Check whether IPI broadcasting can be enabled */
2402 	apic_smt_update();
2403 }
2404 
2405 void __init arch_cpu_finalize_init(void)
2406 {
2407 	identify_boot_cpu();
2408 
2409 	/*
2410 	 * identify_boot_cpu() initialized SMT support information, let the
2411 	 * core code know.
2412 	 */
2413 	cpu_smt_set_num_threads(smp_num_siblings, smp_num_siblings);
2414 
2415 	if (!IS_ENABLED(CONFIG_SMP)) {
2416 		pr_info("CPU: ");
2417 		print_cpu_info(&boot_cpu_data);
2418 	}
2419 
2420 	cpu_select_mitigations();
2421 
2422 	arch_smt_update();
2423 
2424 	if (IS_ENABLED(CONFIG_X86_32)) {
2425 		/*
2426 		 * Check whether this is a real i386 which is not longer
2427 		 * supported and fixup the utsname.
2428 		 */
2429 		if (boot_cpu_data.x86 < 4)
2430 			panic("Kernel requires i486+ for 'invlpg' and other features");
2431 
2432 		init_utsname()->machine[1] =
2433 			'0' + (boot_cpu_data.x86 > 6 ? 6 : boot_cpu_data.x86);
2434 	}
2435 
2436 	/*
2437 	 * Must be before alternatives because it might set or clear
2438 	 * feature bits.
2439 	 */
2440 	fpu__init_system();
2441 	fpu__init_cpu();
2442 
2443 	alternative_instructions();
2444 
2445 	if (IS_ENABLED(CONFIG_X86_64)) {
2446 		/*
2447 		 * Make sure the first 2MB area is not mapped by huge pages
2448 		 * There are typically fixed size MTRRs in there and overlapping
2449 		 * MTRRs into large pages causes slow downs.
2450 		 *
2451 		 * Right now we don't do that with gbpages because there seems
2452 		 * very little benefit for that case.
2453 		 */
2454 		if (!direct_gbpages)
2455 			set_memory_4k((unsigned long)__va(0), 1);
2456 	} else {
2457 		fpu__init_check_bugs();
2458 	}
2459 
2460 	/*
2461 	 * This needs to be called before any devices perform DMA
2462 	 * operations that might use the SWIOTLB bounce buffers. It will
2463 	 * mark the bounce buffers as decrypted so that their usage will
2464 	 * not cause "plain-text" data to be decrypted when accessed. It
2465 	 * must be called after late_time_init() so that Hyper-V x86/x64
2466 	 * hypercalls work when the SWIOTLB bounce buffers are decrypted.
2467 	 */
2468 	mem_encrypt_init();
2469 }
2470