xref: /linux/arch/x86/kernel/cpu/common.c (revision 8137a49e1567726eb10fcf55ad141ac19804ca6b)
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 
26 #include <asm/cmdline.h>
27 #include <asm/stackprotector.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/internal.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/memtype.h>
56 #include <asm/microcode.h>
57 #include <asm/microcode_intel.h>
58 #include <asm/intel-family.h>
59 #include <asm/cpu_device_id.h>
60 #include <asm/uv/uv.h>
61 #include <asm/sigframe.h>
62 
63 #include "cpu.h"
64 
65 u32 elf_hwcap2 __read_mostly;
66 
67 /* all of these masks are initialized in setup_cpu_local_masks() */
68 cpumask_var_t cpu_initialized_mask;
69 cpumask_var_t cpu_callout_mask;
70 cpumask_var_t cpu_callin_mask;
71 
72 /* representing cpus for which sibling maps can be computed */
73 cpumask_var_t cpu_sibling_setup_mask;
74 
75 /* Number of siblings per CPU package */
76 int smp_num_siblings = 1;
77 EXPORT_SYMBOL(smp_num_siblings);
78 
79 /* Last level cache ID of each logical CPU */
80 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
81 
82 /* correctly size the local cpu masks */
83 void __init setup_cpu_local_masks(void)
84 {
85 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
86 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
87 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
88 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
89 }
90 
91 static void default_init(struct cpuinfo_x86 *c)
92 {
93 #ifdef CONFIG_X86_64
94 	cpu_detect_cache_sizes(c);
95 #else
96 	/* Not much we can do here... */
97 	/* Check if at least it has cpuid */
98 	if (c->cpuid_level == -1) {
99 		/* No cpuid. It must be an ancient CPU */
100 		if (c->x86 == 4)
101 			strcpy(c->x86_model_id, "486");
102 		else if (c->x86 == 3)
103 			strcpy(c->x86_model_id, "386");
104 	}
105 #endif
106 }
107 
108 static const struct cpu_dev default_cpu = {
109 	.c_init		= default_init,
110 	.c_vendor	= "Unknown",
111 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
112 };
113 
114 static const struct cpu_dev *this_cpu = &default_cpu;
115 
116 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
117 #ifdef CONFIG_X86_64
118 	/*
119 	 * We need valid kernel segments for data and code in long mode too
120 	 * IRET will check the segment types  kkeil 2000/10/28
121 	 * Also sysret mandates a special GDT layout
122 	 *
123 	 * TLS descriptors are currently at a different place compared to i386.
124 	 * Hopefully nobody expects them at a fixed place (Wine?)
125 	 */
126 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
127 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
128 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
129 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
130 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
131 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
132 #else
133 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
134 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
135 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
136 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
137 	/*
138 	 * Segments used for calling PnP BIOS have byte granularity.
139 	 * They code segments and data segments have fixed 64k limits,
140 	 * the transfer segment sizes are set at run time.
141 	 */
142 	/* 32-bit code */
143 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
144 	/* 16-bit code */
145 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
146 	/* 16-bit data */
147 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
148 	/* 16-bit data */
149 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
150 	/* 16-bit data */
151 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
152 	/*
153 	 * The APM segments have byte granularity and their bases
154 	 * are set at run time.  All have 64k limits.
155 	 */
156 	/* 32-bit code */
157 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
158 	/* 16-bit code */
159 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
160 	/* data */
161 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
162 
163 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
164 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
165 #endif
166 } };
167 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
168 
169 #ifdef CONFIG_X86_64
170 static int __init x86_nopcid_setup(char *s)
171 {
172 	/* nopcid doesn't accept parameters */
173 	if (s)
174 		return -EINVAL;
175 
176 	/* do not emit a message if the feature is not present */
177 	if (!boot_cpu_has(X86_FEATURE_PCID))
178 		return 0;
179 
180 	setup_clear_cpu_cap(X86_FEATURE_PCID);
181 	pr_info("nopcid: PCID feature disabled\n");
182 	return 0;
183 }
184 early_param("nopcid", x86_nopcid_setup);
185 #endif
186 
187 static int __init x86_noinvpcid_setup(char *s)
188 {
189 	/* noinvpcid doesn't accept parameters */
190 	if (s)
191 		return -EINVAL;
192 
193 	/* do not emit a message if the feature is not present */
194 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
195 		return 0;
196 
197 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
198 	pr_info("noinvpcid: INVPCID feature disabled\n");
199 	return 0;
200 }
201 early_param("noinvpcid", x86_noinvpcid_setup);
202 
203 #ifdef CONFIG_X86_32
204 static int cachesize_override = -1;
205 static int disable_x86_serial_nr = 1;
206 
207 static int __init cachesize_setup(char *str)
208 {
209 	get_option(&str, &cachesize_override);
210 	return 1;
211 }
212 __setup("cachesize=", cachesize_setup);
213 
214 static int __init x86_sep_setup(char *s)
215 {
216 	setup_clear_cpu_cap(X86_FEATURE_SEP);
217 	return 1;
218 }
219 __setup("nosep", x86_sep_setup);
220 
221 /* Standard macro to see if a specific flag is changeable */
222 static inline int flag_is_changeable_p(u32 flag)
223 {
224 	u32 f1, f2;
225 
226 	/*
227 	 * Cyrix and IDT cpus allow disabling of CPUID
228 	 * so the code below may return different results
229 	 * when it is executed before and after enabling
230 	 * the CPUID. Add "volatile" to not allow gcc to
231 	 * optimize the subsequent calls to this function.
232 	 */
233 	asm volatile ("pushfl		\n\t"
234 		      "pushfl		\n\t"
235 		      "popl %0		\n\t"
236 		      "movl %0, %1	\n\t"
237 		      "xorl %2, %0	\n\t"
238 		      "pushl %0		\n\t"
239 		      "popfl		\n\t"
240 		      "pushfl		\n\t"
241 		      "popl %0		\n\t"
242 		      "popfl		\n\t"
243 
244 		      : "=&r" (f1), "=&r" (f2)
245 		      : "ir" (flag));
246 
247 	return ((f1^f2) & flag) != 0;
248 }
249 
250 /* Probe for the CPUID instruction */
251 int have_cpuid_p(void)
252 {
253 	return flag_is_changeable_p(X86_EFLAGS_ID);
254 }
255 
256 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
257 {
258 	unsigned long lo, hi;
259 
260 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
261 		return;
262 
263 	/* Disable processor serial number: */
264 
265 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
266 	lo |= 0x200000;
267 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
268 
269 	pr_notice("CPU serial number disabled.\n");
270 	clear_cpu_cap(c, X86_FEATURE_PN);
271 
272 	/* Disabling the serial number may affect the cpuid level */
273 	c->cpuid_level = cpuid_eax(0);
274 }
275 
276 static int __init x86_serial_nr_setup(char *s)
277 {
278 	disable_x86_serial_nr = 0;
279 	return 1;
280 }
281 __setup("serialnumber", x86_serial_nr_setup);
282 #else
283 static inline int flag_is_changeable_p(u32 flag)
284 {
285 	return 1;
286 }
287 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
288 {
289 }
290 #endif
291 
292 static __init int setup_disable_smep(char *arg)
293 {
294 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
295 	return 1;
296 }
297 __setup("nosmep", setup_disable_smep);
298 
299 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
300 {
301 	if (cpu_has(c, X86_FEATURE_SMEP))
302 		cr4_set_bits(X86_CR4_SMEP);
303 }
304 
305 static __init int setup_disable_smap(char *arg)
306 {
307 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
308 	return 1;
309 }
310 __setup("nosmap", setup_disable_smap);
311 
312 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
313 {
314 	unsigned long eflags = native_save_fl();
315 
316 	/* This should have been cleared long ago */
317 	BUG_ON(eflags & X86_EFLAGS_AC);
318 
319 	if (cpu_has(c, X86_FEATURE_SMAP)) {
320 #ifdef CONFIG_X86_SMAP
321 		cr4_set_bits(X86_CR4_SMAP);
322 #else
323 		cr4_clear_bits(X86_CR4_SMAP);
324 #endif
325 	}
326 }
327 
328 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
329 {
330 	/* Check the boot processor, plus build option for UMIP. */
331 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
332 		goto out;
333 
334 	/* Check the current processor's cpuid bits. */
335 	if (!cpu_has(c, X86_FEATURE_UMIP))
336 		goto out;
337 
338 	cr4_set_bits(X86_CR4_UMIP);
339 
340 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
341 
342 	return;
343 
344 out:
345 	/*
346 	 * Make sure UMIP is disabled in case it was enabled in a
347 	 * previous boot (e.g., via kexec).
348 	 */
349 	cr4_clear_bits(X86_CR4_UMIP);
350 }
351 
352 /* These bits should not change their value after CPU init is finished. */
353 static const unsigned long cr4_pinned_mask =
354 	X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP | X86_CR4_FSGSBASE;
355 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
356 static unsigned long cr4_pinned_bits __ro_after_init;
357 
358 void native_write_cr0(unsigned long val)
359 {
360 	unsigned long bits_missing = 0;
361 
362 set_register:
363 	asm volatile("mov %0,%%cr0": "+r" (val) : : "memory");
364 
365 	if (static_branch_likely(&cr_pinning)) {
366 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
367 			bits_missing = X86_CR0_WP;
368 			val |= bits_missing;
369 			goto set_register;
370 		}
371 		/* Warn after we've set the missing bits. */
372 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
373 	}
374 }
375 EXPORT_SYMBOL(native_write_cr0);
376 
377 void native_write_cr4(unsigned long val)
378 {
379 	unsigned long bits_changed = 0;
380 
381 set_register:
382 	asm volatile("mov %0,%%cr4": "+r" (val) : : "memory");
383 
384 	if (static_branch_likely(&cr_pinning)) {
385 		if (unlikely((val & cr4_pinned_mask) != cr4_pinned_bits)) {
386 			bits_changed = (val & cr4_pinned_mask) ^ cr4_pinned_bits;
387 			val = (val & ~cr4_pinned_mask) | cr4_pinned_bits;
388 			goto set_register;
389 		}
390 		/* Warn after we've corrected the changed bits. */
391 		WARN_ONCE(bits_changed, "pinned CR4 bits changed: 0x%lx!?\n",
392 			  bits_changed);
393 	}
394 }
395 #if IS_MODULE(CONFIG_LKDTM)
396 EXPORT_SYMBOL_GPL(native_write_cr4);
397 #endif
398 
399 void cr4_update_irqsoff(unsigned long set, unsigned long clear)
400 {
401 	unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
402 
403 	lockdep_assert_irqs_disabled();
404 
405 	newval = (cr4 & ~clear) | set;
406 	if (newval != cr4) {
407 		this_cpu_write(cpu_tlbstate.cr4, newval);
408 		__write_cr4(newval);
409 	}
410 }
411 EXPORT_SYMBOL(cr4_update_irqsoff);
412 
413 /* Read the CR4 shadow. */
414 unsigned long cr4_read_shadow(void)
415 {
416 	return this_cpu_read(cpu_tlbstate.cr4);
417 }
418 EXPORT_SYMBOL_GPL(cr4_read_shadow);
419 
420 void cr4_init(void)
421 {
422 	unsigned long cr4 = __read_cr4();
423 
424 	if (boot_cpu_has(X86_FEATURE_PCID))
425 		cr4 |= X86_CR4_PCIDE;
426 	if (static_branch_likely(&cr_pinning))
427 		cr4 = (cr4 & ~cr4_pinned_mask) | cr4_pinned_bits;
428 
429 	__write_cr4(cr4);
430 
431 	/* Initialize cr4 shadow for this CPU. */
432 	this_cpu_write(cpu_tlbstate.cr4, cr4);
433 }
434 
435 /*
436  * Once CPU feature detection is finished (and boot params have been
437  * parsed), record any of the sensitive CR bits that are set, and
438  * enable CR pinning.
439  */
440 static void __init setup_cr_pinning(void)
441 {
442 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & cr4_pinned_mask;
443 	static_key_enable(&cr_pinning.key);
444 }
445 
446 static __init int x86_nofsgsbase_setup(char *arg)
447 {
448 	/* Require an exact match without trailing characters. */
449 	if (strlen(arg))
450 		return 0;
451 
452 	/* Do not emit a message if the feature is not present. */
453 	if (!boot_cpu_has(X86_FEATURE_FSGSBASE))
454 		return 1;
455 
456 	setup_clear_cpu_cap(X86_FEATURE_FSGSBASE);
457 	pr_info("FSGSBASE disabled via kernel command line\n");
458 	return 1;
459 }
460 __setup("nofsgsbase", x86_nofsgsbase_setup);
461 
462 /*
463  * Protection Keys are not available in 32-bit mode.
464  */
465 static bool pku_disabled;
466 
467 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
468 {
469 	if (c == &boot_cpu_data) {
470 		if (pku_disabled || !cpu_feature_enabled(X86_FEATURE_PKU))
471 			return;
472 		/*
473 		 * Setting CR4.PKE will cause the X86_FEATURE_OSPKE cpuid
474 		 * bit to be set.  Enforce it.
475 		 */
476 		setup_force_cpu_cap(X86_FEATURE_OSPKE);
477 
478 	} else if (!cpu_feature_enabled(X86_FEATURE_OSPKE)) {
479 		return;
480 	}
481 
482 	cr4_set_bits(X86_CR4_PKE);
483 	/* Load the default PKRU value */
484 	pkru_write_default();
485 }
486 
487 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
488 static __init int setup_disable_pku(char *arg)
489 {
490 	/*
491 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
492 	 * runtime checks are against OSPKE so clearing the
493 	 * bit does nothing.
494 	 *
495 	 * This way, we will see "pku" in cpuinfo, but not
496 	 * "ospke", which is exactly what we want.  It shows
497 	 * that the CPU has PKU, but the OS has not enabled it.
498 	 * This happens to be exactly how a system would look
499 	 * if we disabled the config option.
500 	 */
501 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
502 	pku_disabled = true;
503 	return 1;
504 }
505 __setup("nopku", setup_disable_pku);
506 #endif /* CONFIG_X86_64 */
507 
508 /*
509  * Some CPU features depend on higher CPUID levels, which may not always
510  * be available due to CPUID level capping or broken virtualization
511  * software.  Add those features to this table to auto-disable them.
512  */
513 struct cpuid_dependent_feature {
514 	u32 feature;
515 	u32 level;
516 };
517 
518 static const struct cpuid_dependent_feature
519 cpuid_dependent_features[] = {
520 	{ X86_FEATURE_MWAIT,		0x00000005 },
521 	{ X86_FEATURE_DCA,		0x00000009 },
522 	{ X86_FEATURE_XSAVE,		0x0000000d },
523 	{ 0, 0 }
524 };
525 
526 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
527 {
528 	const struct cpuid_dependent_feature *df;
529 
530 	for (df = cpuid_dependent_features; df->feature; df++) {
531 
532 		if (!cpu_has(c, df->feature))
533 			continue;
534 		/*
535 		 * Note: cpuid_level is set to -1 if unavailable, but
536 		 * extended_extended_level is set to 0 if unavailable
537 		 * and the legitimate extended levels are all negative
538 		 * when signed; hence the weird messing around with
539 		 * signs here...
540 		 */
541 		if (!((s32)df->level < 0 ?
542 		     (u32)df->level > (u32)c->extended_cpuid_level :
543 		     (s32)df->level > (s32)c->cpuid_level))
544 			continue;
545 
546 		clear_cpu_cap(c, df->feature);
547 		if (!warn)
548 			continue;
549 
550 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
551 			x86_cap_flag(df->feature), df->level);
552 	}
553 }
554 
555 /*
556  * Naming convention should be: <Name> [(<Codename>)]
557  * This table only is used unless init_<vendor>() below doesn't set it;
558  * in particular, if CPUID levels 0x80000002..4 are supported, this
559  * isn't used
560  */
561 
562 /* Look up CPU names by table lookup. */
563 static const char *table_lookup_model(struct cpuinfo_x86 *c)
564 {
565 #ifdef CONFIG_X86_32
566 	const struct legacy_cpu_model_info *info;
567 
568 	if (c->x86_model >= 16)
569 		return NULL;	/* Range check */
570 
571 	if (!this_cpu)
572 		return NULL;
573 
574 	info = this_cpu->legacy_models;
575 
576 	while (info->family) {
577 		if (info->family == c->x86)
578 			return info->model_names[c->x86_model];
579 		info++;
580 	}
581 #endif
582 	return NULL;		/* Not found */
583 }
584 
585 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
586 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
587 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
588 
589 void load_percpu_segment(int cpu)
590 {
591 #ifdef CONFIG_X86_32
592 	loadsegment(fs, __KERNEL_PERCPU);
593 #else
594 	__loadsegment_simple(gs, 0);
595 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
596 #endif
597 }
598 
599 #ifdef CONFIG_X86_32
600 /* The 32-bit entry code needs to find cpu_entry_area. */
601 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
602 #endif
603 
604 /* Load the original GDT from the per-cpu structure */
605 void load_direct_gdt(int cpu)
606 {
607 	struct desc_ptr gdt_descr;
608 
609 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
610 	gdt_descr.size = GDT_SIZE - 1;
611 	load_gdt(&gdt_descr);
612 }
613 EXPORT_SYMBOL_GPL(load_direct_gdt);
614 
615 /* Load a fixmap remapping of the per-cpu GDT */
616 void load_fixmap_gdt(int cpu)
617 {
618 	struct desc_ptr gdt_descr;
619 
620 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
621 	gdt_descr.size = GDT_SIZE - 1;
622 	load_gdt(&gdt_descr);
623 }
624 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
625 
626 /*
627  * Current gdt points %fs at the "master" per-cpu area: after this,
628  * it's on the real one.
629  */
630 void switch_to_new_gdt(int cpu)
631 {
632 	/* Load the original GDT */
633 	load_direct_gdt(cpu);
634 	/* Reload the per-cpu base */
635 	load_percpu_segment(cpu);
636 }
637 
638 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
639 
640 static void get_model_name(struct cpuinfo_x86 *c)
641 {
642 	unsigned int *v;
643 	char *p, *q, *s;
644 
645 	if (c->extended_cpuid_level < 0x80000004)
646 		return;
647 
648 	v = (unsigned int *)c->x86_model_id;
649 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
650 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
651 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
652 	c->x86_model_id[48] = 0;
653 
654 	/* Trim whitespace */
655 	p = q = s = &c->x86_model_id[0];
656 
657 	while (*p == ' ')
658 		p++;
659 
660 	while (*p) {
661 		/* Note the last non-whitespace index */
662 		if (!isspace(*p))
663 			s = q;
664 
665 		*q++ = *p++;
666 	}
667 
668 	*(s + 1) = '\0';
669 }
670 
671 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
672 {
673 	unsigned int eax, ebx, ecx, edx;
674 
675 	c->x86_max_cores = 1;
676 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
677 		return;
678 
679 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
680 	if (eax & 0x1f)
681 		c->x86_max_cores = (eax >> 26) + 1;
682 }
683 
684 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
685 {
686 	unsigned int n, dummy, ebx, ecx, edx, l2size;
687 
688 	n = c->extended_cpuid_level;
689 
690 	if (n >= 0x80000005) {
691 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
692 		c->x86_cache_size = (ecx>>24) + (edx>>24);
693 #ifdef CONFIG_X86_64
694 		/* On K8 L1 TLB is inclusive, so don't count it */
695 		c->x86_tlbsize = 0;
696 #endif
697 	}
698 
699 	if (n < 0x80000006)	/* Some chips just has a large L1. */
700 		return;
701 
702 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
703 	l2size = ecx >> 16;
704 
705 #ifdef CONFIG_X86_64
706 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
707 #else
708 	/* do processor-specific cache resizing */
709 	if (this_cpu->legacy_cache_size)
710 		l2size = this_cpu->legacy_cache_size(c, l2size);
711 
712 	/* Allow user to override all this if necessary. */
713 	if (cachesize_override != -1)
714 		l2size = cachesize_override;
715 
716 	if (l2size == 0)
717 		return;		/* Again, no L2 cache is possible */
718 #endif
719 
720 	c->x86_cache_size = l2size;
721 }
722 
723 u16 __read_mostly tlb_lli_4k[NR_INFO];
724 u16 __read_mostly tlb_lli_2m[NR_INFO];
725 u16 __read_mostly tlb_lli_4m[NR_INFO];
726 u16 __read_mostly tlb_lld_4k[NR_INFO];
727 u16 __read_mostly tlb_lld_2m[NR_INFO];
728 u16 __read_mostly tlb_lld_4m[NR_INFO];
729 u16 __read_mostly tlb_lld_1g[NR_INFO];
730 
731 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
732 {
733 	if (this_cpu->c_detect_tlb)
734 		this_cpu->c_detect_tlb(c);
735 
736 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
737 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
738 		tlb_lli_4m[ENTRIES]);
739 
740 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
741 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
742 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
743 }
744 
745 int detect_ht_early(struct cpuinfo_x86 *c)
746 {
747 #ifdef CONFIG_SMP
748 	u32 eax, ebx, ecx, edx;
749 
750 	if (!cpu_has(c, X86_FEATURE_HT))
751 		return -1;
752 
753 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
754 		return -1;
755 
756 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
757 		return -1;
758 
759 	cpuid(1, &eax, &ebx, &ecx, &edx);
760 
761 	smp_num_siblings = (ebx & 0xff0000) >> 16;
762 	if (smp_num_siblings == 1)
763 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
764 #endif
765 	return 0;
766 }
767 
768 void detect_ht(struct cpuinfo_x86 *c)
769 {
770 #ifdef CONFIG_SMP
771 	int index_msb, core_bits;
772 
773 	if (detect_ht_early(c) < 0)
774 		return;
775 
776 	index_msb = get_count_order(smp_num_siblings);
777 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
778 
779 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
780 
781 	index_msb = get_count_order(smp_num_siblings);
782 
783 	core_bits = get_count_order(c->x86_max_cores);
784 
785 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
786 				       ((1 << core_bits) - 1);
787 #endif
788 }
789 
790 static void get_cpu_vendor(struct cpuinfo_x86 *c)
791 {
792 	char *v = c->x86_vendor_id;
793 	int i;
794 
795 	for (i = 0; i < X86_VENDOR_NUM; i++) {
796 		if (!cpu_devs[i])
797 			break;
798 
799 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
800 		    (cpu_devs[i]->c_ident[1] &&
801 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
802 
803 			this_cpu = cpu_devs[i];
804 			c->x86_vendor = this_cpu->c_x86_vendor;
805 			return;
806 		}
807 	}
808 
809 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
810 		    "CPU: Your system may be unstable.\n", v);
811 
812 	c->x86_vendor = X86_VENDOR_UNKNOWN;
813 	this_cpu = &default_cpu;
814 }
815 
816 void cpu_detect(struct cpuinfo_x86 *c)
817 {
818 	/* Get vendor name */
819 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
820 	      (unsigned int *)&c->x86_vendor_id[0],
821 	      (unsigned int *)&c->x86_vendor_id[8],
822 	      (unsigned int *)&c->x86_vendor_id[4]);
823 
824 	c->x86 = 4;
825 	/* Intel-defined flags: level 0x00000001 */
826 	if (c->cpuid_level >= 0x00000001) {
827 		u32 junk, tfms, cap0, misc;
828 
829 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
830 		c->x86		= x86_family(tfms);
831 		c->x86_model	= x86_model(tfms);
832 		c->x86_stepping	= x86_stepping(tfms);
833 
834 		if (cap0 & (1<<19)) {
835 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
836 			c->x86_cache_alignment = c->x86_clflush_size;
837 		}
838 	}
839 }
840 
841 static void apply_forced_caps(struct cpuinfo_x86 *c)
842 {
843 	int i;
844 
845 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
846 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
847 		c->x86_capability[i] |= cpu_caps_set[i];
848 	}
849 }
850 
851 static void init_speculation_control(struct cpuinfo_x86 *c)
852 {
853 	/*
854 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
855 	 * and they also have a different bit for STIBP support. Also,
856 	 * a hypervisor might have set the individual AMD bits even on
857 	 * Intel CPUs, for finer-grained selection of what's available.
858 	 */
859 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
860 		set_cpu_cap(c, X86_FEATURE_IBRS);
861 		set_cpu_cap(c, X86_FEATURE_IBPB);
862 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
863 	}
864 
865 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
866 		set_cpu_cap(c, X86_FEATURE_STIBP);
867 
868 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
869 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
870 		set_cpu_cap(c, X86_FEATURE_SSBD);
871 
872 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
873 		set_cpu_cap(c, X86_FEATURE_IBRS);
874 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
875 	}
876 
877 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
878 		set_cpu_cap(c, X86_FEATURE_IBPB);
879 
880 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
881 		set_cpu_cap(c, X86_FEATURE_STIBP);
882 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
883 	}
884 
885 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
886 		set_cpu_cap(c, X86_FEATURE_SSBD);
887 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
888 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
889 	}
890 }
891 
892 void get_cpu_cap(struct cpuinfo_x86 *c)
893 {
894 	u32 eax, ebx, ecx, edx;
895 
896 	/* Intel-defined flags: level 0x00000001 */
897 	if (c->cpuid_level >= 0x00000001) {
898 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
899 
900 		c->x86_capability[CPUID_1_ECX] = ecx;
901 		c->x86_capability[CPUID_1_EDX] = edx;
902 	}
903 
904 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
905 	if (c->cpuid_level >= 0x00000006)
906 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
907 
908 	/* Additional Intel-defined flags: level 0x00000007 */
909 	if (c->cpuid_level >= 0x00000007) {
910 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
911 		c->x86_capability[CPUID_7_0_EBX] = ebx;
912 		c->x86_capability[CPUID_7_ECX] = ecx;
913 		c->x86_capability[CPUID_7_EDX] = edx;
914 
915 		/* Check valid sub-leaf index before accessing it */
916 		if (eax >= 1) {
917 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
918 			c->x86_capability[CPUID_7_1_EAX] = eax;
919 		}
920 	}
921 
922 	/* Extended state features: level 0x0000000d */
923 	if (c->cpuid_level >= 0x0000000d) {
924 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
925 
926 		c->x86_capability[CPUID_D_1_EAX] = eax;
927 	}
928 
929 	/* AMD-defined flags: level 0x80000001 */
930 	eax = cpuid_eax(0x80000000);
931 	c->extended_cpuid_level = eax;
932 
933 	if ((eax & 0xffff0000) == 0x80000000) {
934 		if (eax >= 0x80000001) {
935 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
936 
937 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
938 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
939 		}
940 	}
941 
942 	if (c->extended_cpuid_level >= 0x80000007) {
943 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
944 
945 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
946 		c->x86_power = edx;
947 	}
948 
949 	if (c->extended_cpuid_level >= 0x80000008) {
950 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
951 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
952 	}
953 
954 	if (c->extended_cpuid_level >= 0x8000000a)
955 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
956 
957 	if (c->extended_cpuid_level >= 0x8000001f)
958 		c->x86_capability[CPUID_8000_001F_EAX] = cpuid_eax(0x8000001f);
959 
960 	init_scattered_cpuid_features(c);
961 	init_speculation_control(c);
962 
963 	/*
964 	 * Clear/Set all flags overridden by options, after probe.
965 	 * This needs to happen each time we re-probe, which may happen
966 	 * several times during CPU initialization.
967 	 */
968 	apply_forced_caps(c);
969 }
970 
971 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
972 {
973 	u32 eax, ebx, ecx, edx;
974 
975 	if (c->extended_cpuid_level >= 0x80000008) {
976 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
977 
978 		c->x86_virt_bits = (eax >> 8) & 0xff;
979 		c->x86_phys_bits = eax & 0xff;
980 	}
981 #ifdef CONFIG_X86_32
982 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
983 		c->x86_phys_bits = 36;
984 #endif
985 	c->x86_cache_bits = c->x86_phys_bits;
986 }
987 
988 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
989 {
990 #ifdef CONFIG_X86_32
991 	int i;
992 
993 	/*
994 	 * First of all, decide if this is a 486 or higher
995 	 * It's a 486 if we can modify the AC flag
996 	 */
997 	if (flag_is_changeable_p(X86_EFLAGS_AC))
998 		c->x86 = 4;
999 	else
1000 		c->x86 = 3;
1001 
1002 	for (i = 0; i < X86_VENDOR_NUM; i++)
1003 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
1004 			c->x86_vendor_id[0] = 0;
1005 			cpu_devs[i]->c_identify(c);
1006 			if (c->x86_vendor_id[0]) {
1007 				get_cpu_vendor(c);
1008 				break;
1009 			}
1010 		}
1011 #endif
1012 }
1013 
1014 #define NO_SPECULATION		BIT(0)
1015 #define NO_MELTDOWN		BIT(1)
1016 #define NO_SSB			BIT(2)
1017 #define NO_L1TF			BIT(3)
1018 #define NO_MDS			BIT(4)
1019 #define MSBDS_ONLY		BIT(5)
1020 #define NO_SWAPGS		BIT(6)
1021 #define NO_ITLB_MULTIHIT	BIT(7)
1022 #define NO_SPECTRE_V2		BIT(8)
1023 
1024 #define VULNWL(vendor, family, model, whitelist)	\
1025 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist)
1026 
1027 #define VULNWL_INTEL(model, whitelist)		\
1028 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1029 
1030 #define VULNWL_AMD(family, whitelist)		\
1031 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1032 
1033 #define VULNWL_HYGON(family, whitelist)		\
1034 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1035 
1036 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1037 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1038 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1039 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1040 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1041 
1042 	/* Intel Family 6 */
1043 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1044 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION | NO_ITLB_MULTIHIT),
1045 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1046 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1047 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1048 
1049 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1050 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1051 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1052 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1053 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1054 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1055 
1056 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1057 
1058 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1059 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1060 
1061 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1062 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1063 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1064 
1065 	/*
1066 	 * Technically, swapgs isn't serializing on AMD (despite it previously
1067 	 * being documented as such in the APM).  But according to AMD, %gs is
1068 	 * updated non-speculatively, and the issuing of %gs-relative memory
1069 	 * operands will be blocked until the %gs update completes, which is
1070 	 * good enough for our purposes.
1071 	 */
1072 
1073 	VULNWL_INTEL(ATOM_TREMONT_D,		NO_ITLB_MULTIHIT),
1074 
1075 	/* AMD Family 0xf - 0x12 */
1076 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1077 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1078 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1079 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1080 
1081 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1082 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1083 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1084 
1085 	/* Zhaoxin Family 7 */
1086 	VULNWL(CENTAUR,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1087 	VULNWL(ZHAOXIN,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1088 	{}
1089 };
1090 
1091 #define VULNBL_INTEL_STEPPINGS(model, steppings, issues)		   \
1092 	X86_MATCH_VENDOR_FAM_MODEL_STEPPINGS_FEATURE(INTEL, 6,		   \
1093 					    INTEL_FAM6_##model, steppings, \
1094 					    X86_FEATURE_ANY, issues)
1095 
1096 #define SRBDS		BIT(0)
1097 
1098 static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = {
1099 	VULNBL_INTEL_STEPPINGS(IVYBRIDGE,	X86_STEPPING_ANY,		SRBDS),
1100 	VULNBL_INTEL_STEPPINGS(HASWELL,		X86_STEPPING_ANY,		SRBDS),
1101 	VULNBL_INTEL_STEPPINGS(HASWELL_L,	X86_STEPPING_ANY,		SRBDS),
1102 	VULNBL_INTEL_STEPPINGS(HASWELL_G,	X86_STEPPING_ANY,		SRBDS),
1103 	VULNBL_INTEL_STEPPINGS(BROADWELL_G,	X86_STEPPING_ANY,		SRBDS),
1104 	VULNBL_INTEL_STEPPINGS(BROADWELL,	X86_STEPPING_ANY,		SRBDS),
1105 	VULNBL_INTEL_STEPPINGS(SKYLAKE_L,	X86_STEPPING_ANY,		SRBDS),
1106 	VULNBL_INTEL_STEPPINGS(SKYLAKE,		X86_STEPPING_ANY,		SRBDS),
1107 	VULNBL_INTEL_STEPPINGS(KABYLAKE_L,	X86_STEPPINGS(0x0, 0xC),	SRBDS),
1108 	VULNBL_INTEL_STEPPINGS(KABYLAKE,	X86_STEPPINGS(0x0, 0xD),	SRBDS),
1109 	{}
1110 };
1111 
1112 static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which)
1113 {
1114 	const struct x86_cpu_id *m = x86_match_cpu(table);
1115 
1116 	return m && !!(m->driver_data & which);
1117 }
1118 
1119 u64 x86_read_arch_cap_msr(void)
1120 {
1121 	u64 ia32_cap = 0;
1122 
1123 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1124 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1125 
1126 	return ia32_cap;
1127 }
1128 
1129 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1130 {
1131 	u64 ia32_cap = x86_read_arch_cap_msr();
1132 
1133 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1134 	if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) &&
1135 	    !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO))
1136 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1137 
1138 	if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION))
1139 		return;
1140 
1141 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1142 
1143 	if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2))
1144 		setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1145 
1146 	if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) &&
1147 	    !(ia32_cap & ARCH_CAP_SSB_NO) &&
1148 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1149 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1150 
1151 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1152 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1153 
1154 	if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) &&
1155 	    !(ia32_cap & ARCH_CAP_MDS_NO)) {
1156 		setup_force_cpu_bug(X86_BUG_MDS);
1157 		if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY))
1158 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1159 	}
1160 
1161 	if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS))
1162 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1163 
1164 	/*
1165 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1166 	 *	- TSX is supported or
1167 	 *	- TSX_CTRL is present
1168 	 *
1169 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1170 	 * the kernel boot e.g. kexec.
1171 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1172 	 * update is not present or running as guest that don't get TSX_CTRL.
1173 	 */
1174 	if (!(ia32_cap & ARCH_CAP_TAA_NO) &&
1175 	    (cpu_has(c, X86_FEATURE_RTM) ||
1176 	     (ia32_cap & ARCH_CAP_TSX_CTRL_MSR)))
1177 		setup_force_cpu_bug(X86_BUG_TAA);
1178 
1179 	/*
1180 	 * SRBDS affects CPUs which support RDRAND or RDSEED and are listed
1181 	 * in the vulnerability blacklist.
1182 	 */
1183 	if ((cpu_has(c, X86_FEATURE_RDRAND) ||
1184 	     cpu_has(c, X86_FEATURE_RDSEED)) &&
1185 	    cpu_matches(cpu_vuln_blacklist, SRBDS))
1186 		    setup_force_cpu_bug(X86_BUG_SRBDS);
1187 
1188 	if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN))
1189 		return;
1190 
1191 	/* Rogue Data Cache Load? No! */
1192 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1193 		return;
1194 
1195 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1196 
1197 	if (cpu_matches(cpu_vuln_whitelist, NO_L1TF))
1198 		return;
1199 
1200 	setup_force_cpu_bug(X86_BUG_L1TF);
1201 }
1202 
1203 /*
1204  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1205  * unfortunately, that's not true in practice because of early VIA
1206  * chips and (more importantly) broken virtualizers that are not easy
1207  * to detect. In the latter case it doesn't even *fail* reliably, so
1208  * probing for it doesn't even work. Disable it completely on 32-bit
1209  * unless we can find a reliable way to detect all the broken cases.
1210  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1211  */
1212 static void detect_nopl(void)
1213 {
1214 #ifdef CONFIG_X86_32
1215 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1216 #else
1217 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1218 #endif
1219 }
1220 
1221 /*
1222  * We parse cpu parameters early because fpu__init_system() is executed
1223  * before parse_early_param().
1224  */
1225 static void __init cpu_parse_early_param(void)
1226 {
1227 	char arg[128];
1228 	char *argptr = arg;
1229 	int arglen, res, bit;
1230 
1231 #ifdef CONFIG_X86_32
1232 	if (cmdline_find_option_bool(boot_command_line, "no387"))
1233 #ifdef CONFIG_MATH_EMULATION
1234 		setup_clear_cpu_cap(X86_FEATURE_FPU);
1235 #else
1236 		pr_err("Option 'no387' required CONFIG_MATH_EMULATION enabled.\n");
1237 #endif
1238 
1239 	if (cmdline_find_option_bool(boot_command_line, "nofxsr"))
1240 		setup_clear_cpu_cap(X86_FEATURE_FXSR);
1241 #endif
1242 
1243 	if (cmdline_find_option_bool(boot_command_line, "noxsave"))
1244 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
1245 
1246 	if (cmdline_find_option_bool(boot_command_line, "noxsaveopt"))
1247 		setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
1248 
1249 	if (cmdline_find_option_bool(boot_command_line, "noxsaves"))
1250 		setup_clear_cpu_cap(X86_FEATURE_XSAVES);
1251 
1252 	arglen = cmdline_find_option(boot_command_line, "clearcpuid", arg, sizeof(arg));
1253 	if (arglen <= 0)
1254 		return;
1255 
1256 	pr_info("Clearing CPUID bits:");
1257 	do {
1258 		res = get_option(&argptr, &bit);
1259 		if (res == 0 || res == 3)
1260 			break;
1261 
1262 		/* If the argument was too long, the last bit may be cut off */
1263 		if (res == 1 && arglen >= sizeof(arg))
1264 			break;
1265 
1266 		if (bit >= 0 && bit < NCAPINTS * 32) {
1267 			pr_cont(" " X86_CAP_FMT, x86_cap_flag(bit));
1268 			setup_clear_cpu_cap(bit);
1269 		}
1270 	} while (res == 2);
1271 	pr_cont("\n");
1272 }
1273 
1274 /*
1275  * Do minimum CPU detection early.
1276  * Fields really needed: vendor, cpuid_level, family, model, mask,
1277  * cache alignment.
1278  * The others are not touched to avoid unwanted side effects.
1279  *
1280  * WARNING: this function is only called on the boot CPU.  Don't add code
1281  * here that is supposed to run on all CPUs.
1282  */
1283 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1284 {
1285 #ifdef CONFIG_X86_64
1286 	c->x86_clflush_size = 64;
1287 	c->x86_phys_bits = 36;
1288 	c->x86_virt_bits = 48;
1289 #else
1290 	c->x86_clflush_size = 32;
1291 	c->x86_phys_bits = 32;
1292 	c->x86_virt_bits = 32;
1293 #endif
1294 	c->x86_cache_alignment = c->x86_clflush_size;
1295 
1296 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1297 	c->extended_cpuid_level = 0;
1298 
1299 	if (!have_cpuid_p())
1300 		identify_cpu_without_cpuid(c);
1301 
1302 	/* cyrix could have cpuid enabled via c_identify()*/
1303 	if (have_cpuid_p()) {
1304 		cpu_detect(c);
1305 		get_cpu_vendor(c);
1306 		get_cpu_cap(c);
1307 		get_cpu_address_sizes(c);
1308 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1309 		cpu_parse_early_param();
1310 
1311 		if (this_cpu->c_early_init)
1312 			this_cpu->c_early_init(c);
1313 
1314 		c->cpu_index = 0;
1315 		filter_cpuid_features(c, false);
1316 
1317 		if (this_cpu->c_bsp_init)
1318 			this_cpu->c_bsp_init(c);
1319 	} else {
1320 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1321 	}
1322 
1323 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1324 
1325 	cpu_set_bug_bits(c);
1326 
1327 	sld_setup(c);
1328 
1329 	fpu__init_system(c);
1330 
1331 	init_sigframe_size();
1332 
1333 #ifdef CONFIG_X86_32
1334 	/*
1335 	 * Regardless of whether PCID is enumerated, the SDM says
1336 	 * that it can't be enabled in 32-bit mode.
1337 	 */
1338 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1339 #endif
1340 
1341 	/*
1342 	 * Later in the boot process pgtable_l5_enabled() relies on
1343 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1344 	 * enabled by this point we need to clear the feature bit to avoid
1345 	 * false-positives at the later stage.
1346 	 *
1347 	 * pgtable_l5_enabled() can be false here for several reasons:
1348 	 *  - 5-level paging is disabled compile-time;
1349 	 *  - it's 32-bit kernel;
1350 	 *  - machine doesn't support 5-level paging;
1351 	 *  - user specified 'no5lvl' in kernel command line.
1352 	 */
1353 	if (!pgtable_l5_enabled())
1354 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1355 
1356 	detect_nopl();
1357 }
1358 
1359 void __init early_cpu_init(void)
1360 {
1361 	const struct cpu_dev *const *cdev;
1362 	int count = 0;
1363 
1364 #ifdef CONFIG_PROCESSOR_SELECT
1365 	pr_info("KERNEL supported cpus:\n");
1366 #endif
1367 
1368 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1369 		const struct cpu_dev *cpudev = *cdev;
1370 
1371 		if (count >= X86_VENDOR_NUM)
1372 			break;
1373 		cpu_devs[count] = cpudev;
1374 		count++;
1375 
1376 #ifdef CONFIG_PROCESSOR_SELECT
1377 		{
1378 			unsigned int j;
1379 
1380 			for (j = 0; j < 2; j++) {
1381 				if (!cpudev->c_ident[j])
1382 					continue;
1383 				pr_info("  %s %s\n", cpudev->c_vendor,
1384 					cpudev->c_ident[j]);
1385 			}
1386 		}
1387 #endif
1388 	}
1389 	early_identify_cpu(&boot_cpu_data);
1390 }
1391 
1392 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1393 {
1394 #ifdef CONFIG_X86_64
1395 	/*
1396 	 * Empirically, writing zero to a segment selector on AMD does
1397 	 * not clear the base, whereas writing zero to a segment
1398 	 * selector on Intel does clear the base.  Intel's behavior
1399 	 * allows slightly faster context switches in the common case
1400 	 * where GS is unused by the prev and next threads.
1401 	 *
1402 	 * Since neither vendor documents this anywhere that I can see,
1403 	 * detect it directly instead of hard-coding the choice by
1404 	 * vendor.
1405 	 *
1406 	 * I've designated AMD's behavior as the "bug" because it's
1407 	 * counterintuitive and less friendly.
1408 	 */
1409 
1410 	unsigned long old_base, tmp;
1411 	rdmsrl(MSR_FS_BASE, old_base);
1412 	wrmsrl(MSR_FS_BASE, 1);
1413 	loadsegment(fs, 0);
1414 	rdmsrl(MSR_FS_BASE, tmp);
1415 	if (tmp != 0)
1416 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1417 	wrmsrl(MSR_FS_BASE, old_base);
1418 #endif
1419 }
1420 
1421 static void generic_identify(struct cpuinfo_x86 *c)
1422 {
1423 	c->extended_cpuid_level = 0;
1424 
1425 	if (!have_cpuid_p())
1426 		identify_cpu_without_cpuid(c);
1427 
1428 	/* cyrix could have cpuid enabled via c_identify()*/
1429 	if (!have_cpuid_p())
1430 		return;
1431 
1432 	cpu_detect(c);
1433 
1434 	get_cpu_vendor(c);
1435 
1436 	get_cpu_cap(c);
1437 
1438 	get_cpu_address_sizes(c);
1439 
1440 	if (c->cpuid_level >= 0x00000001) {
1441 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1442 #ifdef CONFIG_X86_32
1443 # ifdef CONFIG_SMP
1444 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1445 # else
1446 		c->apicid = c->initial_apicid;
1447 # endif
1448 #endif
1449 		c->phys_proc_id = c->initial_apicid;
1450 	}
1451 
1452 	get_model_name(c); /* Default name */
1453 
1454 	detect_null_seg_behavior(c);
1455 
1456 	/*
1457 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1458 	 * systems that run Linux at CPL > 0 may or may not have the
1459 	 * issue, but, even if they have the issue, there's absolutely
1460 	 * nothing we can do about it because we can't use the real IRET
1461 	 * instruction.
1462 	 *
1463 	 * NB: For the time being, only 32-bit kernels support
1464 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1465 	 * whether to apply espfix using paravirt hooks.  If any
1466 	 * non-paravirt system ever shows up that does *not* have the
1467 	 * ESPFIX issue, we can change this.
1468 	 */
1469 #ifdef CONFIG_X86_32
1470 	set_cpu_bug(c, X86_BUG_ESPFIX);
1471 #endif
1472 }
1473 
1474 /*
1475  * Validate that ACPI/mptables have the same information about the
1476  * effective APIC id and update the package map.
1477  */
1478 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1479 {
1480 #ifdef CONFIG_SMP
1481 	unsigned int apicid, cpu = smp_processor_id();
1482 
1483 	apicid = apic->cpu_present_to_apicid(cpu);
1484 
1485 	if (apicid != c->apicid) {
1486 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1487 		       cpu, apicid, c->initial_apicid);
1488 	}
1489 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1490 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1491 #else
1492 	c->logical_proc_id = 0;
1493 #endif
1494 }
1495 
1496 /*
1497  * This does the hard work of actually picking apart the CPU stuff...
1498  */
1499 static void identify_cpu(struct cpuinfo_x86 *c)
1500 {
1501 	int i;
1502 
1503 	c->loops_per_jiffy = loops_per_jiffy;
1504 	c->x86_cache_size = 0;
1505 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1506 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1507 	c->x86_vendor_id[0] = '\0'; /* Unset */
1508 	c->x86_model_id[0] = '\0';  /* Unset */
1509 	c->x86_max_cores = 1;
1510 	c->x86_coreid_bits = 0;
1511 	c->cu_id = 0xff;
1512 #ifdef CONFIG_X86_64
1513 	c->x86_clflush_size = 64;
1514 	c->x86_phys_bits = 36;
1515 	c->x86_virt_bits = 48;
1516 #else
1517 	c->cpuid_level = -1;	/* CPUID not detected */
1518 	c->x86_clflush_size = 32;
1519 	c->x86_phys_bits = 32;
1520 	c->x86_virt_bits = 32;
1521 #endif
1522 	c->x86_cache_alignment = c->x86_clflush_size;
1523 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1524 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
1525 	memset(&c->vmx_capability, 0, sizeof(c->vmx_capability));
1526 #endif
1527 
1528 	generic_identify(c);
1529 
1530 	if (this_cpu->c_identify)
1531 		this_cpu->c_identify(c);
1532 
1533 	/* Clear/Set all flags overridden by options, after probe */
1534 	apply_forced_caps(c);
1535 
1536 #ifdef CONFIG_X86_64
1537 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1538 #endif
1539 
1540 	/*
1541 	 * Vendor-specific initialization.  In this section we
1542 	 * canonicalize the feature flags, meaning if there are
1543 	 * features a certain CPU supports which CPUID doesn't
1544 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1545 	 * we handle them here.
1546 	 *
1547 	 * At the end of this section, c->x86_capability better
1548 	 * indicate the features this CPU genuinely supports!
1549 	 */
1550 	if (this_cpu->c_init)
1551 		this_cpu->c_init(c);
1552 
1553 	/* Disable the PN if appropriate */
1554 	squash_the_stupid_serial_number(c);
1555 
1556 	/* Set up SMEP/SMAP/UMIP */
1557 	setup_smep(c);
1558 	setup_smap(c);
1559 	setup_umip(c);
1560 
1561 	/* Enable FSGSBASE instructions if available. */
1562 	if (cpu_has(c, X86_FEATURE_FSGSBASE)) {
1563 		cr4_set_bits(X86_CR4_FSGSBASE);
1564 		elf_hwcap2 |= HWCAP2_FSGSBASE;
1565 	}
1566 
1567 	/*
1568 	 * The vendor-specific functions might have changed features.
1569 	 * Now we do "generic changes."
1570 	 */
1571 
1572 	/* Filter out anything that depends on CPUID levels we don't have */
1573 	filter_cpuid_features(c, true);
1574 
1575 	/* If the model name is still unset, do table lookup. */
1576 	if (!c->x86_model_id[0]) {
1577 		const char *p;
1578 		p = table_lookup_model(c);
1579 		if (p)
1580 			strcpy(c->x86_model_id, p);
1581 		else
1582 			/* Last resort... */
1583 			sprintf(c->x86_model_id, "%02x/%02x",
1584 				c->x86, c->x86_model);
1585 	}
1586 
1587 #ifdef CONFIG_X86_64
1588 	detect_ht(c);
1589 #endif
1590 
1591 	x86_init_rdrand(c);
1592 	setup_pku(c);
1593 
1594 	/*
1595 	 * Clear/Set all flags overridden by options, need do it
1596 	 * before following smp all cpus cap AND.
1597 	 */
1598 	apply_forced_caps(c);
1599 
1600 	/*
1601 	 * On SMP, boot_cpu_data holds the common feature set between
1602 	 * all CPUs; so make sure that we indicate which features are
1603 	 * common between the CPUs.  The first time this routine gets
1604 	 * executed, c == &boot_cpu_data.
1605 	 */
1606 	if (c != &boot_cpu_data) {
1607 		/* AND the already accumulated flags with these */
1608 		for (i = 0; i < NCAPINTS; i++)
1609 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1610 
1611 		/* OR, i.e. replicate the bug flags */
1612 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1613 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1614 	}
1615 
1616 	/* Init Machine Check Exception if available. */
1617 	mcheck_cpu_init(c);
1618 
1619 	select_idle_routine(c);
1620 
1621 #ifdef CONFIG_NUMA
1622 	numa_add_cpu(smp_processor_id());
1623 #endif
1624 }
1625 
1626 /*
1627  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1628  * on 32-bit kernels:
1629  */
1630 #ifdef CONFIG_X86_32
1631 void enable_sep_cpu(void)
1632 {
1633 	struct tss_struct *tss;
1634 	int cpu;
1635 
1636 	if (!boot_cpu_has(X86_FEATURE_SEP))
1637 		return;
1638 
1639 	cpu = get_cpu();
1640 	tss = &per_cpu(cpu_tss_rw, cpu);
1641 
1642 	/*
1643 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1644 	 * see the big comment in struct x86_hw_tss's definition.
1645 	 */
1646 
1647 	tss->x86_tss.ss1 = __KERNEL_CS;
1648 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1649 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1650 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1651 
1652 	put_cpu();
1653 }
1654 #endif
1655 
1656 void __init identify_boot_cpu(void)
1657 {
1658 	identify_cpu(&boot_cpu_data);
1659 #ifdef CONFIG_X86_32
1660 	sysenter_setup();
1661 	enable_sep_cpu();
1662 #endif
1663 	cpu_detect_tlb(&boot_cpu_data);
1664 	setup_cr_pinning();
1665 
1666 	tsx_init();
1667 }
1668 
1669 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1670 {
1671 	BUG_ON(c == &boot_cpu_data);
1672 	identify_cpu(c);
1673 #ifdef CONFIG_X86_32
1674 	enable_sep_cpu();
1675 #endif
1676 	mtrr_ap_init();
1677 	validate_apic_and_package_id(c);
1678 	x86_spec_ctrl_setup_ap();
1679 	update_srbds_msr();
1680 }
1681 
1682 static __init int setup_noclflush(char *arg)
1683 {
1684 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1685 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1686 	return 1;
1687 }
1688 __setup("noclflush", setup_noclflush);
1689 
1690 void print_cpu_info(struct cpuinfo_x86 *c)
1691 {
1692 	const char *vendor = NULL;
1693 
1694 	if (c->x86_vendor < X86_VENDOR_NUM) {
1695 		vendor = this_cpu->c_vendor;
1696 	} else {
1697 		if (c->cpuid_level >= 0)
1698 			vendor = c->x86_vendor_id;
1699 	}
1700 
1701 	if (vendor && !strstr(c->x86_model_id, vendor))
1702 		pr_cont("%s ", vendor);
1703 
1704 	if (c->x86_model_id[0])
1705 		pr_cont("%s", c->x86_model_id);
1706 	else
1707 		pr_cont("%d86", c->x86);
1708 
1709 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1710 
1711 	if (c->x86_stepping || c->cpuid_level >= 0)
1712 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1713 	else
1714 		pr_cont(")\n");
1715 }
1716 
1717 /*
1718  * clearcpuid= was already parsed in cpu_parse_early_param().  This dummy
1719  * function prevents it from becoming an environment variable for init.
1720  */
1721 static __init int setup_clearcpuid(char *arg)
1722 {
1723 	return 1;
1724 }
1725 __setup("clearcpuid=", setup_clearcpuid);
1726 
1727 #ifdef CONFIG_X86_64
1728 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1729 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1730 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1731 
1732 /*
1733  * The following percpu variables are hot.  Align current_task to
1734  * cacheline size such that they fall in the same cacheline.
1735  */
1736 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1737 	&init_task;
1738 EXPORT_PER_CPU_SYMBOL(current_task);
1739 
1740 DEFINE_PER_CPU(void *, hardirq_stack_ptr);
1741 DEFINE_PER_CPU(bool, hardirq_stack_inuse);
1742 
1743 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1744 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1745 
1746 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) = TOP_OF_INIT_STACK;
1747 
1748 /* May not be marked __init: used by software suspend */
1749 void syscall_init(void)
1750 {
1751 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1752 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1753 
1754 #ifdef CONFIG_IA32_EMULATION
1755 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1756 	/*
1757 	 * This only works on Intel CPUs.
1758 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1759 	 * This does not cause SYSENTER to jump to the wrong location, because
1760 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1761 	 */
1762 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1763 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1764 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1765 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1766 #else
1767 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1768 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1769 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1770 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1771 #endif
1772 
1773 	/*
1774 	 * Flags to clear on syscall; clear as much as possible
1775 	 * to minimize user space-kernel interference.
1776 	 */
1777 	wrmsrl(MSR_SYSCALL_MASK,
1778 	       X86_EFLAGS_CF|X86_EFLAGS_PF|X86_EFLAGS_AF|
1779 	       X86_EFLAGS_ZF|X86_EFLAGS_SF|X86_EFLAGS_TF|
1780 	       X86_EFLAGS_IF|X86_EFLAGS_DF|X86_EFLAGS_OF|
1781 	       X86_EFLAGS_IOPL|X86_EFLAGS_NT|X86_EFLAGS_RF|
1782 	       X86_EFLAGS_AC|X86_EFLAGS_ID);
1783 }
1784 
1785 #else	/* CONFIG_X86_64 */
1786 
1787 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1788 EXPORT_PER_CPU_SYMBOL(current_task);
1789 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1790 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1791 
1792 /*
1793  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1794  * the top of the kernel stack.  Use an extra percpu variable to track the
1795  * top of the kernel stack directly.
1796  */
1797 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1798 	(unsigned long)&init_thread_union + THREAD_SIZE;
1799 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1800 
1801 #ifdef CONFIG_STACKPROTECTOR
1802 DEFINE_PER_CPU(unsigned long, __stack_chk_guard);
1803 EXPORT_PER_CPU_SYMBOL(__stack_chk_guard);
1804 #endif
1805 
1806 #endif	/* CONFIG_X86_64 */
1807 
1808 /*
1809  * Clear all 6 debug registers:
1810  */
1811 static void clear_all_debug_regs(void)
1812 {
1813 	int i;
1814 
1815 	for (i = 0; i < 8; i++) {
1816 		/* Ignore db4, db5 */
1817 		if ((i == 4) || (i == 5))
1818 			continue;
1819 
1820 		set_debugreg(0, i);
1821 	}
1822 }
1823 
1824 #ifdef CONFIG_KGDB
1825 /*
1826  * Restore debug regs if using kgdbwait and you have a kernel debugger
1827  * connection established.
1828  */
1829 static void dbg_restore_debug_regs(void)
1830 {
1831 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1832 		arch_kgdb_ops.correct_hw_break();
1833 }
1834 #else /* ! CONFIG_KGDB */
1835 #define dbg_restore_debug_regs()
1836 #endif /* ! CONFIG_KGDB */
1837 
1838 static void wait_for_master_cpu(int cpu)
1839 {
1840 #ifdef CONFIG_SMP
1841 	/*
1842 	 * wait for ACK from master CPU before continuing
1843 	 * with AP initialization
1844 	 */
1845 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1846 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1847 		cpu_relax();
1848 #endif
1849 }
1850 
1851 #ifdef CONFIG_X86_64
1852 static inline void setup_getcpu(int cpu)
1853 {
1854 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1855 	struct desc_struct d = { };
1856 
1857 	if (boot_cpu_has(X86_FEATURE_RDTSCP) || boot_cpu_has(X86_FEATURE_RDPID))
1858 		wrmsr(MSR_TSC_AUX, cpudata, 0);
1859 
1860 	/* Store CPU and node number in limit. */
1861 	d.limit0 = cpudata;
1862 	d.limit1 = cpudata >> 16;
1863 
1864 	d.type = 5;		/* RO data, expand down, accessed */
1865 	d.dpl = 3;		/* Visible to user code */
1866 	d.s = 1;		/* Not a system segment */
1867 	d.p = 1;		/* Present */
1868 	d.d = 1;		/* 32-bit */
1869 
1870 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1871 }
1872 
1873 static inline void ucode_cpu_init(int cpu)
1874 {
1875 	if (cpu)
1876 		load_ucode_ap();
1877 }
1878 
1879 static inline void tss_setup_ist(struct tss_struct *tss)
1880 {
1881 	/* Set up the per-CPU TSS IST stacks */
1882 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1883 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1884 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1885 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1886 	/* Only mapped when SEV-ES is active */
1887 	tss->x86_tss.ist[IST_INDEX_VC] = __this_cpu_ist_top_va(VC);
1888 }
1889 
1890 #else /* CONFIG_X86_64 */
1891 
1892 static inline void setup_getcpu(int cpu) { }
1893 
1894 static inline void ucode_cpu_init(int cpu)
1895 {
1896 	show_ucode_info_early();
1897 }
1898 
1899 static inline void tss_setup_ist(struct tss_struct *tss) { }
1900 
1901 #endif /* !CONFIG_X86_64 */
1902 
1903 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
1904 {
1905 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
1906 
1907 #ifdef CONFIG_X86_IOPL_IOPERM
1908 	tss->io_bitmap.prev_max = 0;
1909 	tss->io_bitmap.prev_sequence = 0;
1910 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
1911 	/*
1912 	 * Invalidate the extra array entry past the end of the all
1913 	 * permission bitmap as required by the hardware.
1914 	 */
1915 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
1916 #endif
1917 }
1918 
1919 /*
1920  * Setup everything needed to handle exceptions from the IDT, including the IST
1921  * exceptions which use paranoid_entry().
1922  */
1923 void cpu_init_exception_handling(void)
1924 {
1925 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
1926 	int cpu = raw_smp_processor_id();
1927 
1928 	/* paranoid_entry() gets the CPU number from the GDT */
1929 	setup_getcpu(cpu);
1930 
1931 	/* IST vectors need TSS to be set up. */
1932 	tss_setup_ist(tss);
1933 	tss_setup_io_bitmap(tss);
1934 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1935 
1936 	load_TR_desc();
1937 
1938 	/* Finally load the IDT */
1939 	load_current_idt();
1940 }
1941 
1942 /*
1943  * cpu_init() initializes state that is per-CPU. Some data is already
1944  * initialized (naturally) in the bootstrap process, such as the GDT.  We
1945  * reload it nevertheless, this function acts as a 'CPU state barrier',
1946  * nothing should get across.
1947  */
1948 void cpu_init(void)
1949 {
1950 	struct task_struct *cur = current;
1951 	int cpu = raw_smp_processor_id();
1952 
1953 	wait_for_master_cpu(cpu);
1954 
1955 	ucode_cpu_init(cpu);
1956 
1957 #ifdef CONFIG_NUMA
1958 	if (this_cpu_read(numa_node) == 0 &&
1959 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1960 		set_numa_node(early_cpu_to_node(cpu));
1961 #endif
1962 	pr_debug("Initializing CPU#%d\n", cpu);
1963 
1964 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
1965 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
1966 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1967 
1968 	/*
1969 	 * Initialize the per-CPU GDT with the boot GDT,
1970 	 * and set up the GDT descriptor:
1971 	 */
1972 	switch_to_new_gdt(cpu);
1973 
1974 	if (IS_ENABLED(CONFIG_X86_64)) {
1975 		loadsegment(fs, 0);
1976 		memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1977 		syscall_init();
1978 
1979 		wrmsrl(MSR_FS_BASE, 0);
1980 		wrmsrl(MSR_KERNEL_GS_BASE, 0);
1981 		barrier();
1982 
1983 		x2apic_setup();
1984 	}
1985 
1986 	mmgrab(&init_mm);
1987 	cur->active_mm = &init_mm;
1988 	BUG_ON(cur->mm);
1989 	initialize_tlbstate_and_flush();
1990 	enter_lazy_tlb(&init_mm, cur);
1991 
1992 	/*
1993 	 * sp0 points to the entry trampoline stack regardless of what task
1994 	 * is running.
1995 	 */
1996 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1997 
1998 	load_mm_ldt(&init_mm);
1999 
2000 	clear_all_debug_regs();
2001 	dbg_restore_debug_regs();
2002 
2003 	doublefault_init_cpu_tss();
2004 
2005 	fpu__init_cpu();
2006 
2007 	if (is_uv_system())
2008 		uv_cpu_init();
2009 
2010 	load_fixmap_gdt(cpu);
2011 }
2012 
2013 #ifdef CONFIG_SMP
2014 void cpu_init_secondary(void)
2015 {
2016 	/*
2017 	 * Relies on the BP having set-up the IDT tables, which are loaded
2018 	 * on this CPU in cpu_init_exception_handling().
2019 	 */
2020 	cpu_init_exception_handling();
2021 	cpu_init();
2022 }
2023 #endif
2024 
2025 /*
2026  * The microcode loader calls this upon late microcode load to recheck features,
2027  * only when microcode has been updated. Caller holds microcode_mutex and CPU
2028  * hotplug lock.
2029  */
2030 void microcode_check(void)
2031 {
2032 	struct cpuinfo_x86 info;
2033 
2034 	perf_check_microcode();
2035 
2036 	/* Reload CPUID max function as it might've changed. */
2037 	info.cpuid_level = cpuid_eax(0);
2038 
2039 	/*
2040 	 * Copy all capability leafs to pick up the synthetic ones so that
2041 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
2042 	 * get overwritten in get_cpu_cap().
2043 	 */
2044 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
2045 
2046 	get_cpu_cap(&info);
2047 
2048 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
2049 		return;
2050 
2051 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
2052 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
2053 }
2054 
2055 /*
2056  * Invoked from core CPU hotplug code after hotplug operations
2057  */
2058 void arch_smt_update(void)
2059 {
2060 	/* Handle the speculative execution misfeatures */
2061 	cpu_bugs_smt_update();
2062 	/* Check whether IPI broadcasting can be enabled */
2063 	apic_smt_update();
2064 }
2065