xref: /linux/arch/x86/kernel/cpu/common.c (revision c81ea7203b224cb85c5daf360556304e7212bab6)
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 
25 #include <asm/stackprotector.h>
26 #include <asm/perf_event.h>
27 #include <asm/mmu_context.h>
28 #include <asm/doublefault.h>
29 #include <asm/archrandom.h>
30 #include <asm/hypervisor.h>
31 #include <asm/processor.h>
32 #include <asm/tlbflush.h>
33 #include <asm/debugreg.h>
34 #include <asm/sections.h>
35 #include <asm/vsyscall.h>
36 #include <linux/topology.h>
37 #include <linux/cpumask.h>
38 #include <asm/pgtable.h>
39 #include <linux/atomic.h>
40 #include <asm/proto.h>
41 #include <asm/setup.h>
42 #include <asm/apic.h>
43 #include <asm/desc.h>
44 #include <asm/fpu/internal.h>
45 #include <asm/mtrr.h>
46 #include <asm/hwcap2.h>
47 #include <linux/numa.h>
48 #include <asm/asm.h>
49 #include <asm/bugs.h>
50 #include <asm/cpu.h>
51 #include <asm/mce.h>
52 #include <asm/msr.h>
53 #include <asm/memtype.h>
54 #include <asm/microcode.h>
55 #include <asm/microcode_intel.h>
56 #include <asm/intel-family.h>
57 #include <asm/cpu_device_id.h>
58 #include <asm/uv/uv.h>
59 
60 #include "cpu.h"
61 
62 u32 elf_hwcap2 __read_mostly;
63 
64 /* all of these masks are initialized in setup_cpu_local_masks() */
65 cpumask_var_t cpu_initialized_mask;
66 cpumask_var_t cpu_callout_mask;
67 cpumask_var_t cpu_callin_mask;
68 
69 /* representing cpus for which sibling maps can be computed */
70 cpumask_var_t cpu_sibling_setup_mask;
71 
72 /* Number of siblings per CPU package */
73 int smp_num_siblings = 1;
74 EXPORT_SYMBOL(smp_num_siblings);
75 
76 /* Last level cache ID of each logical CPU */
77 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
78 
79 /* correctly size the local cpu masks */
80 void __init setup_cpu_local_masks(void)
81 {
82 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
83 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
84 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
85 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
86 }
87 
88 static void default_init(struct cpuinfo_x86 *c)
89 {
90 #ifdef CONFIG_X86_64
91 	cpu_detect_cache_sizes(c);
92 #else
93 	/* Not much we can do here... */
94 	/* Check if at least it has cpuid */
95 	if (c->cpuid_level == -1) {
96 		/* No cpuid. It must be an ancient CPU */
97 		if (c->x86 == 4)
98 			strcpy(c->x86_model_id, "486");
99 		else if (c->x86 == 3)
100 			strcpy(c->x86_model_id, "386");
101 	}
102 #endif
103 }
104 
105 static const struct cpu_dev default_cpu = {
106 	.c_init		= default_init,
107 	.c_vendor	= "Unknown",
108 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
109 };
110 
111 static const struct cpu_dev *this_cpu = &default_cpu;
112 
113 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
114 #ifdef CONFIG_X86_64
115 	/*
116 	 * We need valid kernel segments for data and code in long mode too
117 	 * IRET will check the segment types  kkeil 2000/10/28
118 	 * Also sysret mandates a special GDT layout
119 	 *
120 	 * TLS descriptors are currently at a different place compared to i386.
121 	 * Hopefully nobody expects them at a fixed place (Wine?)
122 	 */
123 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
124 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
125 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
126 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
127 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
128 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
129 #else
130 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
131 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
132 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
133 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
134 	/*
135 	 * Segments used for calling PnP BIOS have byte granularity.
136 	 * They code segments and data segments have fixed 64k limits,
137 	 * the transfer segment sizes are set at run time.
138 	 */
139 	/* 32-bit code */
140 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
141 	/* 16-bit code */
142 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
143 	/* 16-bit data */
144 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
145 	/* 16-bit data */
146 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
147 	/* 16-bit data */
148 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
149 	/*
150 	 * The APM segments have byte granularity and their bases
151 	 * are set at run time.  All have 64k limits.
152 	 */
153 	/* 32-bit code */
154 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
155 	/* 16-bit code */
156 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
157 	/* data */
158 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
159 
160 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
161 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
162 	GDT_STACK_CANARY_INIT
163 #endif
164 } };
165 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
166 
167 static int __init x86_mpx_setup(char *s)
168 {
169 	/* require an exact match without trailing characters */
170 	if (strlen(s))
171 		return 0;
172 
173 	/* do not emit a message if the feature is not present */
174 	if (!boot_cpu_has(X86_FEATURE_MPX))
175 		return 1;
176 
177 	setup_clear_cpu_cap(X86_FEATURE_MPX);
178 	pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
179 	return 1;
180 }
181 __setup("nompx", x86_mpx_setup);
182 
183 #ifdef CONFIG_X86_64
184 static int __init x86_nopcid_setup(char *s)
185 {
186 	/* nopcid doesn't accept parameters */
187 	if (s)
188 		return -EINVAL;
189 
190 	/* do not emit a message if the feature is not present */
191 	if (!boot_cpu_has(X86_FEATURE_PCID))
192 		return 0;
193 
194 	setup_clear_cpu_cap(X86_FEATURE_PCID);
195 	pr_info("nopcid: PCID feature disabled\n");
196 	return 0;
197 }
198 early_param("nopcid", x86_nopcid_setup);
199 #endif
200 
201 static int __init x86_noinvpcid_setup(char *s)
202 {
203 	/* noinvpcid doesn't accept parameters */
204 	if (s)
205 		return -EINVAL;
206 
207 	/* do not emit a message if the feature is not present */
208 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
209 		return 0;
210 
211 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
212 	pr_info("noinvpcid: INVPCID feature disabled\n");
213 	return 0;
214 }
215 early_param("noinvpcid", x86_noinvpcid_setup);
216 
217 #ifdef CONFIG_X86_32
218 static int cachesize_override = -1;
219 static int disable_x86_serial_nr = 1;
220 
221 static int __init cachesize_setup(char *str)
222 {
223 	get_option(&str, &cachesize_override);
224 	return 1;
225 }
226 __setup("cachesize=", cachesize_setup);
227 
228 static int __init x86_sep_setup(char *s)
229 {
230 	setup_clear_cpu_cap(X86_FEATURE_SEP);
231 	return 1;
232 }
233 __setup("nosep", x86_sep_setup);
234 
235 /* Standard macro to see if a specific flag is changeable */
236 static inline int flag_is_changeable_p(u32 flag)
237 {
238 	u32 f1, f2;
239 
240 	/*
241 	 * Cyrix and IDT cpus allow disabling of CPUID
242 	 * so the code below may return different results
243 	 * when it is executed before and after enabling
244 	 * the CPUID. Add "volatile" to not allow gcc to
245 	 * optimize the subsequent calls to this function.
246 	 */
247 	asm volatile ("pushfl		\n\t"
248 		      "pushfl		\n\t"
249 		      "popl %0		\n\t"
250 		      "movl %0, %1	\n\t"
251 		      "xorl %2, %0	\n\t"
252 		      "pushl %0		\n\t"
253 		      "popfl		\n\t"
254 		      "pushfl		\n\t"
255 		      "popl %0		\n\t"
256 		      "popfl		\n\t"
257 
258 		      : "=&r" (f1), "=&r" (f2)
259 		      : "ir" (flag));
260 
261 	return ((f1^f2) & flag) != 0;
262 }
263 
264 /* Probe for the CPUID instruction */
265 int have_cpuid_p(void)
266 {
267 	return flag_is_changeable_p(X86_EFLAGS_ID);
268 }
269 
270 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
271 {
272 	unsigned long lo, hi;
273 
274 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
275 		return;
276 
277 	/* Disable processor serial number: */
278 
279 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
280 	lo |= 0x200000;
281 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
282 
283 	pr_notice("CPU serial number disabled.\n");
284 	clear_cpu_cap(c, X86_FEATURE_PN);
285 
286 	/* Disabling the serial number may affect the cpuid level */
287 	c->cpuid_level = cpuid_eax(0);
288 }
289 
290 static int __init x86_serial_nr_setup(char *s)
291 {
292 	disable_x86_serial_nr = 0;
293 	return 1;
294 }
295 __setup("serialnumber", x86_serial_nr_setup);
296 #else
297 static inline int flag_is_changeable_p(u32 flag)
298 {
299 	return 1;
300 }
301 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
302 {
303 }
304 #endif
305 
306 static __init int setup_disable_smep(char *arg)
307 {
308 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
309 	/* Check for things that depend on SMEP being enabled: */
310 	check_mpx_erratum(&boot_cpu_data);
311 	return 1;
312 }
313 __setup("nosmep", setup_disable_smep);
314 
315 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
316 {
317 	if (cpu_has(c, X86_FEATURE_SMEP))
318 		cr4_set_bits(X86_CR4_SMEP);
319 }
320 
321 static __init int setup_disable_smap(char *arg)
322 {
323 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
324 	return 1;
325 }
326 __setup("nosmap", setup_disable_smap);
327 
328 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
329 {
330 	unsigned long eflags = native_save_fl();
331 
332 	/* This should have been cleared long ago */
333 	BUG_ON(eflags & X86_EFLAGS_AC);
334 
335 	if (cpu_has(c, X86_FEATURE_SMAP)) {
336 #ifdef CONFIG_X86_SMAP
337 		cr4_set_bits(X86_CR4_SMAP);
338 #else
339 		cr4_clear_bits(X86_CR4_SMAP);
340 #endif
341 	}
342 }
343 
344 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
345 {
346 	/* Check the boot processor, plus build option for UMIP. */
347 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
348 		goto out;
349 
350 	/* Check the current processor's cpuid bits. */
351 	if (!cpu_has(c, X86_FEATURE_UMIP))
352 		goto out;
353 
354 	cr4_set_bits(X86_CR4_UMIP);
355 
356 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
357 
358 	return;
359 
360 out:
361 	/*
362 	 * Make sure UMIP is disabled in case it was enabled in a
363 	 * previous boot (e.g., via kexec).
364 	 */
365 	cr4_clear_bits(X86_CR4_UMIP);
366 }
367 
368 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
369 static unsigned long cr4_pinned_bits __ro_after_init;
370 
371 void native_write_cr0(unsigned long val)
372 {
373 	unsigned long bits_missing = 0;
374 
375 set_register:
376 	asm volatile("mov %0,%%cr0": "+r" (val), "+m" (__force_order));
377 
378 	if (static_branch_likely(&cr_pinning)) {
379 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
380 			bits_missing = X86_CR0_WP;
381 			val |= bits_missing;
382 			goto set_register;
383 		}
384 		/* Warn after we've set the missing bits. */
385 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
386 	}
387 }
388 EXPORT_SYMBOL(native_write_cr0);
389 
390 void native_write_cr4(unsigned long val)
391 {
392 	unsigned long bits_missing = 0;
393 
394 set_register:
395 	asm volatile("mov %0,%%cr4": "+r" (val), "+m" (cr4_pinned_bits));
396 
397 	if (static_branch_likely(&cr_pinning)) {
398 		if (unlikely((val & cr4_pinned_bits) != cr4_pinned_bits)) {
399 			bits_missing = ~val & cr4_pinned_bits;
400 			val |= bits_missing;
401 			goto set_register;
402 		}
403 		/* Warn after we've set the missing bits. */
404 		WARN_ONCE(bits_missing, "CR4 bits went missing: %lx!?\n",
405 			  bits_missing);
406 	}
407 }
408 EXPORT_SYMBOL(native_write_cr4);
409 
410 void cr4_init(void)
411 {
412 	unsigned long cr4 = __read_cr4();
413 
414 	if (boot_cpu_has(X86_FEATURE_PCID))
415 		cr4 |= X86_CR4_PCIDE;
416 	if (static_branch_likely(&cr_pinning))
417 		cr4 |= cr4_pinned_bits;
418 
419 	__write_cr4(cr4);
420 
421 	/* Initialize cr4 shadow for this CPU. */
422 	this_cpu_write(cpu_tlbstate.cr4, cr4);
423 }
424 
425 /*
426  * Once CPU feature detection is finished (and boot params have been
427  * parsed), record any of the sensitive CR bits that are set, and
428  * enable CR pinning.
429  */
430 static void __init setup_cr_pinning(void)
431 {
432 	unsigned long mask;
433 
434 	mask = (X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP);
435 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & mask;
436 	static_key_enable(&cr_pinning.key);
437 }
438 
439 /*
440  * Protection Keys are not available in 32-bit mode.
441  */
442 static bool pku_disabled;
443 
444 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
445 {
446 	struct pkru_state *pk;
447 
448 	/* check the boot processor, plus compile options for PKU: */
449 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
450 		return;
451 	/* checks the actual processor's cpuid bits: */
452 	if (!cpu_has(c, X86_FEATURE_PKU))
453 		return;
454 	if (pku_disabled)
455 		return;
456 
457 	cr4_set_bits(X86_CR4_PKE);
458 	pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU);
459 	if (pk)
460 		pk->pkru = init_pkru_value;
461 	/*
462 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
463 	 * cpuid bit to be set.  We need to ensure that we
464 	 * update that bit in this CPU's "cpu_info".
465 	 */
466 	get_cpu_cap(c);
467 }
468 
469 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
470 static __init int setup_disable_pku(char *arg)
471 {
472 	/*
473 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
474 	 * runtime checks are against OSPKE so clearing the
475 	 * bit does nothing.
476 	 *
477 	 * This way, we will see "pku" in cpuinfo, but not
478 	 * "ospke", which is exactly what we want.  It shows
479 	 * that the CPU has PKU, but the OS has not enabled it.
480 	 * This happens to be exactly how a system would look
481 	 * if we disabled the config option.
482 	 */
483 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
484 	pku_disabled = true;
485 	return 1;
486 }
487 __setup("nopku", setup_disable_pku);
488 #endif /* CONFIG_X86_64 */
489 
490 /*
491  * Some CPU features depend on higher CPUID levels, which may not always
492  * be available due to CPUID level capping or broken virtualization
493  * software.  Add those features to this table to auto-disable them.
494  */
495 struct cpuid_dependent_feature {
496 	u32 feature;
497 	u32 level;
498 };
499 
500 static const struct cpuid_dependent_feature
501 cpuid_dependent_features[] = {
502 	{ X86_FEATURE_MWAIT,		0x00000005 },
503 	{ X86_FEATURE_DCA,		0x00000009 },
504 	{ X86_FEATURE_XSAVE,		0x0000000d },
505 	{ 0, 0 }
506 };
507 
508 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
509 {
510 	const struct cpuid_dependent_feature *df;
511 
512 	for (df = cpuid_dependent_features; df->feature; df++) {
513 
514 		if (!cpu_has(c, df->feature))
515 			continue;
516 		/*
517 		 * Note: cpuid_level is set to -1 if unavailable, but
518 		 * extended_extended_level is set to 0 if unavailable
519 		 * and the legitimate extended levels are all negative
520 		 * when signed; hence the weird messing around with
521 		 * signs here...
522 		 */
523 		if (!((s32)df->level < 0 ?
524 		     (u32)df->level > (u32)c->extended_cpuid_level :
525 		     (s32)df->level > (s32)c->cpuid_level))
526 			continue;
527 
528 		clear_cpu_cap(c, df->feature);
529 		if (!warn)
530 			continue;
531 
532 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
533 			x86_cap_flag(df->feature), df->level);
534 	}
535 }
536 
537 /*
538  * Naming convention should be: <Name> [(<Codename>)]
539  * This table only is used unless init_<vendor>() below doesn't set it;
540  * in particular, if CPUID levels 0x80000002..4 are supported, this
541  * isn't used
542  */
543 
544 /* Look up CPU names by table lookup. */
545 static const char *table_lookup_model(struct cpuinfo_x86 *c)
546 {
547 #ifdef CONFIG_X86_32
548 	const struct legacy_cpu_model_info *info;
549 
550 	if (c->x86_model >= 16)
551 		return NULL;	/* Range check */
552 
553 	if (!this_cpu)
554 		return NULL;
555 
556 	info = this_cpu->legacy_models;
557 
558 	while (info->family) {
559 		if (info->family == c->x86)
560 			return info->model_names[c->x86_model];
561 		info++;
562 	}
563 #endif
564 	return NULL;		/* Not found */
565 }
566 
567 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
568 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
569 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
570 
571 void load_percpu_segment(int cpu)
572 {
573 #ifdef CONFIG_X86_32
574 	loadsegment(fs, __KERNEL_PERCPU);
575 #else
576 	__loadsegment_simple(gs, 0);
577 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
578 #endif
579 	load_stack_canary_segment();
580 }
581 
582 #ifdef CONFIG_X86_32
583 /* The 32-bit entry code needs to find cpu_entry_area. */
584 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
585 #endif
586 
587 /* Load the original GDT from the per-cpu structure */
588 void load_direct_gdt(int cpu)
589 {
590 	struct desc_ptr gdt_descr;
591 
592 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
593 	gdt_descr.size = GDT_SIZE - 1;
594 	load_gdt(&gdt_descr);
595 }
596 EXPORT_SYMBOL_GPL(load_direct_gdt);
597 
598 /* Load a fixmap remapping of the per-cpu GDT */
599 void load_fixmap_gdt(int cpu)
600 {
601 	struct desc_ptr gdt_descr;
602 
603 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
604 	gdt_descr.size = GDT_SIZE - 1;
605 	load_gdt(&gdt_descr);
606 }
607 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
608 
609 /*
610  * Current gdt points %fs at the "master" per-cpu area: after this,
611  * it's on the real one.
612  */
613 void switch_to_new_gdt(int cpu)
614 {
615 	/* Load the original GDT */
616 	load_direct_gdt(cpu);
617 	/* Reload the per-cpu base */
618 	load_percpu_segment(cpu);
619 }
620 
621 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
622 
623 static void get_model_name(struct cpuinfo_x86 *c)
624 {
625 	unsigned int *v;
626 	char *p, *q, *s;
627 
628 	if (c->extended_cpuid_level < 0x80000004)
629 		return;
630 
631 	v = (unsigned int *)c->x86_model_id;
632 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
633 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
634 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
635 	c->x86_model_id[48] = 0;
636 
637 	/* Trim whitespace */
638 	p = q = s = &c->x86_model_id[0];
639 
640 	while (*p == ' ')
641 		p++;
642 
643 	while (*p) {
644 		/* Note the last non-whitespace index */
645 		if (!isspace(*p))
646 			s = q;
647 
648 		*q++ = *p++;
649 	}
650 
651 	*(s + 1) = '\0';
652 }
653 
654 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
655 {
656 	unsigned int eax, ebx, ecx, edx;
657 
658 	c->x86_max_cores = 1;
659 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
660 		return;
661 
662 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
663 	if (eax & 0x1f)
664 		c->x86_max_cores = (eax >> 26) + 1;
665 }
666 
667 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
668 {
669 	unsigned int n, dummy, ebx, ecx, edx, l2size;
670 
671 	n = c->extended_cpuid_level;
672 
673 	if (n >= 0x80000005) {
674 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
675 		c->x86_cache_size = (ecx>>24) + (edx>>24);
676 #ifdef CONFIG_X86_64
677 		/* On K8 L1 TLB is inclusive, so don't count it */
678 		c->x86_tlbsize = 0;
679 #endif
680 	}
681 
682 	if (n < 0x80000006)	/* Some chips just has a large L1. */
683 		return;
684 
685 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
686 	l2size = ecx >> 16;
687 
688 #ifdef CONFIG_X86_64
689 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
690 #else
691 	/* do processor-specific cache resizing */
692 	if (this_cpu->legacy_cache_size)
693 		l2size = this_cpu->legacy_cache_size(c, l2size);
694 
695 	/* Allow user to override all this if necessary. */
696 	if (cachesize_override != -1)
697 		l2size = cachesize_override;
698 
699 	if (l2size == 0)
700 		return;		/* Again, no L2 cache is possible */
701 #endif
702 
703 	c->x86_cache_size = l2size;
704 }
705 
706 u16 __read_mostly tlb_lli_4k[NR_INFO];
707 u16 __read_mostly tlb_lli_2m[NR_INFO];
708 u16 __read_mostly tlb_lli_4m[NR_INFO];
709 u16 __read_mostly tlb_lld_4k[NR_INFO];
710 u16 __read_mostly tlb_lld_2m[NR_INFO];
711 u16 __read_mostly tlb_lld_4m[NR_INFO];
712 u16 __read_mostly tlb_lld_1g[NR_INFO];
713 
714 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
715 {
716 	if (this_cpu->c_detect_tlb)
717 		this_cpu->c_detect_tlb(c);
718 
719 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
720 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
721 		tlb_lli_4m[ENTRIES]);
722 
723 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
724 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
725 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
726 }
727 
728 int detect_ht_early(struct cpuinfo_x86 *c)
729 {
730 #ifdef CONFIG_SMP
731 	u32 eax, ebx, ecx, edx;
732 
733 	if (!cpu_has(c, X86_FEATURE_HT))
734 		return -1;
735 
736 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
737 		return -1;
738 
739 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
740 		return -1;
741 
742 	cpuid(1, &eax, &ebx, &ecx, &edx);
743 
744 	smp_num_siblings = (ebx & 0xff0000) >> 16;
745 	if (smp_num_siblings == 1)
746 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
747 #endif
748 	return 0;
749 }
750 
751 void detect_ht(struct cpuinfo_x86 *c)
752 {
753 #ifdef CONFIG_SMP
754 	int index_msb, core_bits;
755 
756 	if (detect_ht_early(c) < 0)
757 		return;
758 
759 	index_msb = get_count_order(smp_num_siblings);
760 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
761 
762 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
763 
764 	index_msb = get_count_order(smp_num_siblings);
765 
766 	core_bits = get_count_order(c->x86_max_cores);
767 
768 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
769 				       ((1 << core_bits) - 1);
770 #endif
771 }
772 
773 static void get_cpu_vendor(struct cpuinfo_x86 *c)
774 {
775 	char *v = c->x86_vendor_id;
776 	int i;
777 
778 	for (i = 0; i < X86_VENDOR_NUM; i++) {
779 		if (!cpu_devs[i])
780 			break;
781 
782 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
783 		    (cpu_devs[i]->c_ident[1] &&
784 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
785 
786 			this_cpu = cpu_devs[i];
787 			c->x86_vendor = this_cpu->c_x86_vendor;
788 			return;
789 		}
790 	}
791 
792 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
793 		    "CPU: Your system may be unstable.\n", v);
794 
795 	c->x86_vendor = X86_VENDOR_UNKNOWN;
796 	this_cpu = &default_cpu;
797 }
798 
799 void cpu_detect(struct cpuinfo_x86 *c)
800 {
801 	/* Get vendor name */
802 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
803 	      (unsigned int *)&c->x86_vendor_id[0],
804 	      (unsigned int *)&c->x86_vendor_id[8],
805 	      (unsigned int *)&c->x86_vendor_id[4]);
806 
807 	c->x86 = 4;
808 	/* Intel-defined flags: level 0x00000001 */
809 	if (c->cpuid_level >= 0x00000001) {
810 		u32 junk, tfms, cap0, misc;
811 
812 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
813 		c->x86		= x86_family(tfms);
814 		c->x86_model	= x86_model(tfms);
815 		c->x86_stepping	= x86_stepping(tfms);
816 
817 		if (cap0 & (1<<19)) {
818 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
819 			c->x86_cache_alignment = c->x86_clflush_size;
820 		}
821 	}
822 }
823 
824 static void apply_forced_caps(struct cpuinfo_x86 *c)
825 {
826 	int i;
827 
828 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
829 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
830 		c->x86_capability[i] |= cpu_caps_set[i];
831 	}
832 }
833 
834 static void init_speculation_control(struct cpuinfo_x86 *c)
835 {
836 	/*
837 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
838 	 * and they also have a different bit for STIBP support. Also,
839 	 * a hypervisor might have set the individual AMD bits even on
840 	 * Intel CPUs, for finer-grained selection of what's available.
841 	 */
842 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
843 		set_cpu_cap(c, X86_FEATURE_IBRS);
844 		set_cpu_cap(c, X86_FEATURE_IBPB);
845 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
846 	}
847 
848 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
849 		set_cpu_cap(c, X86_FEATURE_STIBP);
850 
851 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
852 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
853 		set_cpu_cap(c, X86_FEATURE_SSBD);
854 
855 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
856 		set_cpu_cap(c, X86_FEATURE_IBRS);
857 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
858 	}
859 
860 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
861 		set_cpu_cap(c, X86_FEATURE_IBPB);
862 
863 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
864 		set_cpu_cap(c, X86_FEATURE_STIBP);
865 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
866 	}
867 
868 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
869 		set_cpu_cap(c, X86_FEATURE_SSBD);
870 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
871 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
872 	}
873 }
874 
875 static void init_cqm(struct cpuinfo_x86 *c)
876 {
877 	if (!cpu_has(c, X86_FEATURE_CQM_LLC)) {
878 		c->x86_cache_max_rmid  = -1;
879 		c->x86_cache_occ_scale = -1;
880 		return;
881 	}
882 
883 	/* will be overridden if occupancy monitoring exists */
884 	c->x86_cache_max_rmid = cpuid_ebx(0xf);
885 
886 	if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC) ||
887 	    cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL) ||
888 	    cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)) {
889 		u32 eax, ebx, ecx, edx;
890 
891 		/* QoS sub-leaf, EAX=0Fh, ECX=1 */
892 		cpuid_count(0xf, 1, &eax, &ebx, &ecx, &edx);
893 
894 		c->x86_cache_max_rmid  = ecx;
895 		c->x86_cache_occ_scale = ebx;
896 	}
897 }
898 
899 void get_cpu_cap(struct cpuinfo_x86 *c)
900 {
901 	u32 eax, ebx, ecx, edx;
902 
903 	/* Intel-defined flags: level 0x00000001 */
904 	if (c->cpuid_level >= 0x00000001) {
905 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
906 
907 		c->x86_capability[CPUID_1_ECX] = ecx;
908 		c->x86_capability[CPUID_1_EDX] = edx;
909 	}
910 
911 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
912 	if (c->cpuid_level >= 0x00000006)
913 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
914 
915 	/* Additional Intel-defined flags: level 0x00000007 */
916 	if (c->cpuid_level >= 0x00000007) {
917 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
918 		c->x86_capability[CPUID_7_0_EBX] = ebx;
919 		c->x86_capability[CPUID_7_ECX] = ecx;
920 		c->x86_capability[CPUID_7_EDX] = edx;
921 
922 		/* Check valid sub-leaf index before accessing it */
923 		if (eax >= 1) {
924 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
925 			c->x86_capability[CPUID_7_1_EAX] = eax;
926 		}
927 	}
928 
929 	/* Extended state features: level 0x0000000d */
930 	if (c->cpuid_level >= 0x0000000d) {
931 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
932 
933 		c->x86_capability[CPUID_D_1_EAX] = eax;
934 	}
935 
936 	/* AMD-defined flags: level 0x80000001 */
937 	eax = cpuid_eax(0x80000000);
938 	c->extended_cpuid_level = eax;
939 
940 	if ((eax & 0xffff0000) == 0x80000000) {
941 		if (eax >= 0x80000001) {
942 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
943 
944 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
945 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
946 		}
947 	}
948 
949 	if (c->extended_cpuid_level >= 0x80000007) {
950 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
951 
952 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
953 		c->x86_power = edx;
954 	}
955 
956 	if (c->extended_cpuid_level >= 0x80000008) {
957 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
958 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
959 	}
960 
961 	if (c->extended_cpuid_level >= 0x8000000a)
962 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
963 
964 	init_scattered_cpuid_features(c);
965 	init_speculation_control(c);
966 	init_cqm(c);
967 
968 	/*
969 	 * Clear/Set all flags overridden by options, after probe.
970 	 * This needs to happen each time we re-probe, which may happen
971 	 * several times during CPU initialization.
972 	 */
973 	apply_forced_caps(c);
974 }
975 
976 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
977 {
978 	u32 eax, ebx, ecx, edx;
979 
980 	if (c->extended_cpuid_level >= 0x80000008) {
981 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
982 
983 		c->x86_virt_bits = (eax >> 8) & 0xff;
984 		c->x86_phys_bits = eax & 0xff;
985 	}
986 #ifdef CONFIG_X86_32
987 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
988 		c->x86_phys_bits = 36;
989 #endif
990 	c->x86_cache_bits = c->x86_phys_bits;
991 }
992 
993 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
994 {
995 #ifdef CONFIG_X86_32
996 	int i;
997 
998 	/*
999 	 * First of all, decide if this is a 486 or higher
1000 	 * It's a 486 if we can modify the AC flag
1001 	 */
1002 	if (flag_is_changeable_p(X86_EFLAGS_AC))
1003 		c->x86 = 4;
1004 	else
1005 		c->x86 = 3;
1006 
1007 	for (i = 0; i < X86_VENDOR_NUM; i++)
1008 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
1009 			c->x86_vendor_id[0] = 0;
1010 			cpu_devs[i]->c_identify(c);
1011 			if (c->x86_vendor_id[0]) {
1012 				get_cpu_vendor(c);
1013 				break;
1014 			}
1015 		}
1016 #endif
1017 }
1018 
1019 #define NO_SPECULATION		BIT(0)
1020 #define NO_MELTDOWN		BIT(1)
1021 #define NO_SSB			BIT(2)
1022 #define NO_L1TF			BIT(3)
1023 #define NO_MDS			BIT(4)
1024 #define MSBDS_ONLY		BIT(5)
1025 #define NO_SWAPGS		BIT(6)
1026 #define NO_ITLB_MULTIHIT	BIT(7)
1027 #define NO_SPECTRE_V2		BIT(8)
1028 
1029 #define VULNWL(_vendor, _family, _model, _whitelist)	\
1030 	{ X86_VENDOR_##_vendor, _family, _model, X86_FEATURE_ANY, _whitelist }
1031 
1032 #define VULNWL_INTEL(model, whitelist)		\
1033 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1034 
1035 #define VULNWL_AMD(family, whitelist)		\
1036 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1037 
1038 #define VULNWL_HYGON(family, whitelist)		\
1039 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1040 
1041 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1042 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1043 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1044 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1045 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1046 
1047 	/* Intel Family 6 */
1048 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1049 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION | NO_ITLB_MULTIHIT),
1050 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1051 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1052 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1053 
1054 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1055 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1056 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1057 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1058 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1059 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1060 
1061 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1062 
1063 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1064 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1065 
1066 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1067 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1068 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1069 
1070 	/*
1071 	 * Technically, swapgs isn't serializing on AMD (despite it previously
1072 	 * being documented as such in the APM).  But according to AMD, %gs is
1073 	 * updated non-speculatively, and the issuing of %gs-relative memory
1074 	 * operands will be blocked until the %gs update completes, which is
1075 	 * good enough for our purposes.
1076 	 */
1077 
1078 	VULNWL_INTEL(ATOM_TREMONT_D,		NO_ITLB_MULTIHIT),
1079 
1080 	/* AMD Family 0xf - 0x12 */
1081 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1082 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1083 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1084 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1085 
1086 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1087 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1088 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1089 
1090 	/* Zhaoxin Family 7 */
1091 	VULNWL(CENTAUR,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1092 	VULNWL(ZHAOXIN,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1093 	{}
1094 };
1095 
1096 static bool __init cpu_matches(unsigned long which)
1097 {
1098 	const struct x86_cpu_id *m = x86_match_cpu(cpu_vuln_whitelist);
1099 
1100 	return m && !!(m->driver_data & which);
1101 }
1102 
1103 u64 x86_read_arch_cap_msr(void)
1104 {
1105 	u64 ia32_cap = 0;
1106 
1107 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1108 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1109 
1110 	return ia32_cap;
1111 }
1112 
1113 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1114 {
1115 	u64 ia32_cap = x86_read_arch_cap_msr();
1116 
1117 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1118 	if (!cpu_matches(NO_ITLB_MULTIHIT) && !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO))
1119 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1120 
1121 	if (cpu_matches(NO_SPECULATION))
1122 		return;
1123 
1124 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1125 
1126 	if (!cpu_matches(NO_SPECTRE_V2))
1127 		setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1128 
1129 	if (!cpu_matches(NO_SSB) && !(ia32_cap & ARCH_CAP_SSB_NO) &&
1130 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1131 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1132 
1133 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1134 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1135 
1136 	if (!cpu_matches(NO_MDS) && !(ia32_cap & ARCH_CAP_MDS_NO)) {
1137 		setup_force_cpu_bug(X86_BUG_MDS);
1138 		if (cpu_matches(MSBDS_ONLY))
1139 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1140 	}
1141 
1142 	if (!cpu_matches(NO_SWAPGS))
1143 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1144 
1145 	/*
1146 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1147 	 *	- TSX is supported or
1148 	 *	- TSX_CTRL is present
1149 	 *
1150 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1151 	 * the kernel boot e.g. kexec.
1152 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1153 	 * update is not present or running as guest that don't get TSX_CTRL.
1154 	 */
1155 	if (!(ia32_cap & ARCH_CAP_TAA_NO) &&
1156 	    (cpu_has(c, X86_FEATURE_RTM) ||
1157 	     (ia32_cap & ARCH_CAP_TSX_CTRL_MSR)))
1158 		setup_force_cpu_bug(X86_BUG_TAA);
1159 
1160 	if (cpu_matches(NO_MELTDOWN))
1161 		return;
1162 
1163 	/* Rogue Data Cache Load? No! */
1164 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1165 		return;
1166 
1167 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1168 
1169 	if (cpu_matches(NO_L1TF))
1170 		return;
1171 
1172 	setup_force_cpu_bug(X86_BUG_L1TF);
1173 }
1174 
1175 /*
1176  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1177  * unfortunately, that's not true in practice because of early VIA
1178  * chips and (more importantly) broken virtualizers that are not easy
1179  * to detect. In the latter case it doesn't even *fail* reliably, so
1180  * probing for it doesn't even work. Disable it completely on 32-bit
1181  * unless we can find a reliable way to detect all the broken cases.
1182  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1183  */
1184 static void detect_nopl(void)
1185 {
1186 #ifdef CONFIG_X86_32
1187 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1188 #else
1189 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1190 #endif
1191 }
1192 
1193 /*
1194  * Do minimum CPU detection early.
1195  * Fields really needed: vendor, cpuid_level, family, model, mask,
1196  * cache alignment.
1197  * The others are not touched to avoid unwanted side effects.
1198  *
1199  * WARNING: this function is only called on the boot CPU.  Don't add code
1200  * here that is supposed to run on all CPUs.
1201  */
1202 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1203 {
1204 #ifdef CONFIG_X86_64
1205 	c->x86_clflush_size = 64;
1206 	c->x86_phys_bits = 36;
1207 	c->x86_virt_bits = 48;
1208 #else
1209 	c->x86_clflush_size = 32;
1210 	c->x86_phys_bits = 32;
1211 	c->x86_virt_bits = 32;
1212 #endif
1213 	c->x86_cache_alignment = c->x86_clflush_size;
1214 
1215 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1216 	c->extended_cpuid_level = 0;
1217 
1218 	if (!have_cpuid_p())
1219 		identify_cpu_without_cpuid(c);
1220 
1221 	/* cyrix could have cpuid enabled via c_identify()*/
1222 	if (have_cpuid_p()) {
1223 		cpu_detect(c);
1224 		get_cpu_vendor(c);
1225 		get_cpu_cap(c);
1226 		get_cpu_address_sizes(c);
1227 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1228 
1229 		if (this_cpu->c_early_init)
1230 			this_cpu->c_early_init(c);
1231 
1232 		c->cpu_index = 0;
1233 		filter_cpuid_features(c, false);
1234 
1235 		if (this_cpu->c_bsp_init)
1236 			this_cpu->c_bsp_init(c);
1237 	} else {
1238 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1239 	}
1240 
1241 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1242 
1243 	cpu_set_bug_bits(c);
1244 
1245 	fpu__init_system(c);
1246 
1247 #ifdef CONFIG_X86_32
1248 	/*
1249 	 * Regardless of whether PCID is enumerated, the SDM says
1250 	 * that it can't be enabled in 32-bit mode.
1251 	 */
1252 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1253 #endif
1254 
1255 	/*
1256 	 * Later in the boot process pgtable_l5_enabled() relies on
1257 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1258 	 * enabled by this point we need to clear the feature bit to avoid
1259 	 * false-positives at the later stage.
1260 	 *
1261 	 * pgtable_l5_enabled() can be false here for several reasons:
1262 	 *  - 5-level paging is disabled compile-time;
1263 	 *  - it's 32-bit kernel;
1264 	 *  - machine doesn't support 5-level paging;
1265 	 *  - user specified 'no5lvl' in kernel command line.
1266 	 */
1267 	if (!pgtable_l5_enabled())
1268 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1269 
1270 	detect_nopl();
1271 }
1272 
1273 void __init early_cpu_init(void)
1274 {
1275 	const struct cpu_dev *const *cdev;
1276 	int count = 0;
1277 
1278 #ifdef CONFIG_PROCESSOR_SELECT
1279 	pr_info("KERNEL supported cpus:\n");
1280 #endif
1281 
1282 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1283 		const struct cpu_dev *cpudev = *cdev;
1284 
1285 		if (count >= X86_VENDOR_NUM)
1286 			break;
1287 		cpu_devs[count] = cpudev;
1288 		count++;
1289 
1290 #ifdef CONFIG_PROCESSOR_SELECT
1291 		{
1292 			unsigned int j;
1293 
1294 			for (j = 0; j < 2; j++) {
1295 				if (!cpudev->c_ident[j])
1296 					continue;
1297 				pr_info("  %s %s\n", cpudev->c_vendor,
1298 					cpudev->c_ident[j]);
1299 			}
1300 		}
1301 #endif
1302 	}
1303 	early_identify_cpu(&boot_cpu_data);
1304 }
1305 
1306 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1307 {
1308 #ifdef CONFIG_X86_64
1309 	/*
1310 	 * Empirically, writing zero to a segment selector on AMD does
1311 	 * not clear the base, whereas writing zero to a segment
1312 	 * selector on Intel does clear the base.  Intel's behavior
1313 	 * allows slightly faster context switches in the common case
1314 	 * where GS is unused by the prev and next threads.
1315 	 *
1316 	 * Since neither vendor documents this anywhere that I can see,
1317 	 * detect it directly instead of hardcoding the choice by
1318 	 * vendor.
1319 	 *
1320 	 * I've designated AMD's behavior as the "bug" because it's
1321 	 * counterintuitive and less friendly.
1322 	 */
1323 
1324 	unsigned long old_base, tmp;
1325 	rdmsrl(MSR_FS_BASE, old_base);
1326 	wrmsrl(MSR_FS_BASE, 1);
1327 	loadsegment(fs, 0);
1328 	rdmsrl(MSR_FS_BASE, tmp);
1329 	if (tmp != 0)
1330 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1331 	wrmsrl(MSR_FS_BASE, old_base);
1332 #endif
1333 }
1334 
1335 static void generic_identify(struct cpuinfo_x86 *c)
1336 {
1337 	c->extended_cpuid_level = 0;
1338 
1339 	if (!have_cpuid_p())
1340 		identify_cpu_without_cpuid(c);
1341 
1342 	/* cyrix could have cpuid enabled via c_identify()*/
1343 	if (!have_cpuid_p())
1344 		return;
1345 
1346 	cpu_detect(c);
1347 
1348 	get_cpu_vendor(c);
1349 
1350 	get_cpu_cap(c);
1351 
1352 	get_cpu_address_sizes(c);
1353 
1354 	if (c->cpuid_level >= 0x00000001) {
1355 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1356 #ifdef CONFIG_X86_32
1357 # ifdef CONFIG_SMP
1358 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1359 # else
1360 		c->apicid = c->initial_apicid;
1361 # endif
1362 #endif
1363 		c->phys_proc_id = c->initial_apicid;
1364 	}
1365 
1366 	get_model_name(c); /* Default name */
1367 
1368 	detect_null_seg_behavior(c);
1369 
1370 	/*
1371 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1372 	 * systems that run Linux at CPL > 0 may or may not have the
1373 	 * issue, but, even if they have the issue, there's absolutely
1374 	 * nothing we can do about it because we can't use the real IRET
1375 	 * instruction.
1376 	 *
1377 	 * NB: For the time being, only 32-bit kernels support
1378 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1379 	 * whether to apply espfix using paravirt hooks.  If any
1380 	 * non-paravirt system ever shows up that does *not* have the
1381 	 * ESPFIX issue, we can change this.
1382 	 */
1383 #ifdef CONFIG_X86_32
1384 # ifdef CONFIG_PARAVIRT_XXL
1385 	do {
1386 		extern void native_iret(void);
1387 		if (pv_ops.cpu.iret == native_iret)
1388 			set_cpu_bug(c, X86_BUG_ESPFIX);
1389 	} while (0);
1390 # else
1391 	set_cpu_bug(c, X86_BUG_ESPFIX);
1392 # endif
1393 #endif
1394 }
1395 
1396 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
1397 {
1398 	/*
1399 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
1400 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
1401 	 * in case CQM bits really aren't there in this CPU.
1402 	 */
1403 	if (c != &boot_cpu_data) {
1404 		boot_cpu_data.x86_cache_max_rmid =
1405 			min(boot_cpu_data.x86_cache_max_rmid,
1406 			    c->x86_cache_max_rmid);
1407 	}
1408 }
1409 
1410 /*
1411  * Validate that ACPI/mptables have the same information about the
1412  * effective APIC id and update the package map.
1413  */
1414 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1415 {
1416 #ifdef CONFIG_SMP
1417 	unsigned int apicid, cpu = smp_processor_id();
1418 
1419 	apicid = apic->cpu_present_to_apicid(cpu);
1420 
1421 	if (apicid != c->apicid) {
1422 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1423 		       cpu, apicid, c->initial_apicid);
1424 	}
1425 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1426 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1427 #else
1428 	c->logical_proc_id = 0;
1429 #endif
1430 }
1431 
1432 /*
1433  * This does the hard work of actually picking apart the CPU stuff...
1434  */
1435 static void identify_cpu(struct cpuinfo_x86 *c)
1436 {
1437 	int i;
1438 
1439 	c->loops_per_jiffy = loops_per_jiffy;
1440 	c->x86_cache_size = 0;
1441 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1442 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1443 	c->x86_vendor_id[0] = '\0'; /* Unset */
1444 	c->x86_model_id[0] = '\0';  /* Unset */
1445 	c->x86_max_cores = 1;
1446 	c->x86_coreid_bits = 0;
1447 	c->cu_id = 0xff;
1448 #ifdef CONFIG_X86_64
1449 	c->x86_clflush_size = 64;
1450 	c->x86_phys_bits = 36;
1451 	c->x86_virt_bits = 48;
1452 #else
1453 	c->cpuid_level = -1;	/* CPUID not detected */
1454 	c->x86_clflush_size = 32;
1455 	c->x86_phys_bits = 32;
1456 	c->x86_virt_bits = 32;
1457 #endif
1458 	c->x86_cache_alignment = c->x86_clflush_size;
1459 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1460 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
1461 	memset(&c->vmx_capability, 0, sizeof(c->vmx_capability));
1462 #endif
1463 
1464 	generic_identify(c);
1465 
1466 	if (this_cpu->c_identify)
1467 		this_cpu->c_identify(c);
1468 
1469 	/* Clear/Set all flags overridden by options, after probe */
1470 	apply_forced_caps(c);
1471 
1472 #ifdef CONFIG_X86_64
1473 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1474 #endif
1475 
1476 	/*
1477 	 * Vendor-specific initialization.  In this section we
1478 	 * canonicalize the feature flags, meaning if there are
1479 	 * features a certain CPU supports which CPUID doesn't
1480 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1481 	 * we handle them here.
1482 	 *
1483 	 * At the end of this section, c->x86_capability better
1484 	 * indicate the features this CPU genuinely supports!
1485 	 */
1486 	if (this_cpu->c_init)
1487 		this_cpu->c_init(c);
1488 
1489 	/* Disable the PN if appropriate */
1490 	squash_the_stupid_serial_number(c);
1491 
1492 	/* Set up SMEP/SMAP/UMIP */
1493 	setup_smep(c);
1494 	setup_smap(c);
1495 	setup_umip(c);
1496 
1497 	/*
1498 	 * The vendor-specific functions might have changed features.
1499 	 * Now we do "generic changes."
1500 	 */
1501 
1502 	/* Filter out anything that depends on CPUID levels we don't have */
1503 	filter_cpuid_features(c, true);
1504 
1505 	/* If the model name is still unset, do table lookup. */
1506 	if (!c->x86_model_id[0]) {
1507 		const char *p;
1508 		p = table_lookup_model(c);
1509 		if (p)
1510 			strcpy(c->x86_model_id, p);
1511 		else
1512 			/* Last resort... */
1513 			sprintf(c->x86_model_id, "%02x/%02x",
1514 				c->x86, c->x86_model);
1515 	}
1516 
1517 #ifdef CONFIG_X86_64
1518 	detect_ht(c);
1519 #endif
1520 
1521 	x86_init_rdrand(c);
1522 	x86_init_cache_qos(c);
1523 	setup_pku(c);
1524 
1525 	/*
1526 	 * Clear/Set all flags overridden by options, need do it
1527 	 * before following smp all cpus cap AND.
1528 	 */
1529 	apply_forced_caps(c);
1530 
1531 	/*
1532 	 * On SMP, boot_cpu_data holds the common feature set between
1533 	 * all CPUs; so make sure that we indicate which features are
1534 	 * common between the CPUs.  The first time this routine gets
1535 	 * executed, c == &boot_cpu_data.
1536 	 */
1537 	if (c != &boot_cpu_data) {
1538 		/* AND the already accumulated flags with these */
1539 		for (i = 0; i < NCAPINTS; i++)
1540 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1541 
1542 		/* OR, i.e. replicate the bug flags */
1543 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1544 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1545 	}
1546 
1547 	/* Init Machine Check Exception if available. */
1548 	mcheck_cpu_init(c);
1549 
1550 	select_idle_routine(c);
1551 
1552 #ifdef CONFIG_NUMA
1553 	numa_add_cpu(smp_processor_id());
1554 #endif
1555 }
1556 
1557 /*
1558  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1559  * on 32-bit kernels:
1560  */
1561 #ifdef CONFIG_X86_32
1562 void enable_sep_cpu(void)
1563 {
1564 	struct tss_struct *tss;
1565 	int cpu;
1566 
1567 	if (!boot_cpu_has(X86_FEATURE_SEP))
1568 		return;
1569 
1570 	cpu = get_cpu();
1571 	tss = &per_cpu(cpu_tss_rw, cpu);
1572 
1573 	/*
1574 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1575 	 * see the big comment in struct x86_hw_tss's definition.
1576 	 */
1577 
1578 	tss->x86_tss.ss1 = __KERNEL_CS;
1579 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1580 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1581 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1582 
1583 	put_cpu();
1584 }
1585 #endif
1586 
1587 void __init identify_boot_cpu(void)
1588 {
1589 	identify_cpu(&boot_cpu_data);
1590 #ifdef CONFIG_X86_32
1591 	sysenter_setup();
1592 	enable_sep_cpu();
1593 #endif
1594 	cpu_detect_tlb(&boot_cpu_data);
1595 	setup_cr_pinning();
1596 
1597 	tsx_init();
1598 }
1599 
1600 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1601 {
1602 	BUG_ON(c == &boot_cpu_data);
1603 	identify_cpu(c);
1604 #ifdef CONFIG_X86_32
1605 	enable_sep_cpu();
1606 #endif
1607 	mtrr_ap_init();
1608 	validate_apic_and_package_id(c);
1609 	x86_spec_ctrl_setup_ap();
1610 }
1611 
1612 static __init int setup_noclflush(char *arg)
1613 {
1614 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1615 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1616 	return 1;
1617 }
1618 __setup("noclflush", setup_noclflush);
1619 
1620 void print_cpu_info(struct cpuinfo_x86 *c)
1621 {
1622 	const char *vendor = NULL;
1623 
1624 	if (c->x86_vendor < X86_VENDOR_NUM) {
1625 		vendor = this_cpu->c_vendor;
1626 	} else {
1627 		if (c->cpuid_level >= 0)
1628 			vendor = c->x86_vendor_id;
1629 	}
1630 
1631 	if (vendor && !strstr(c->x86_model_id, vendor))
1632 		pr_cont("%s ", vendor);
1633 
1634 	if (c->x86_model_id[0])
1635 		pr_cont("%s", c->x86_model_id);
1636 	else
1637 		pr_cont("%d86", c->x86);
1638 
1639 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1640 
1641 	if (c->x86_stepping || c->cpuid_level >= 0)
1642 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1643 	else
1644 		pr_cont(")\n");
1645 }
1646 
1647 /*
1648  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1649  * But we need to keep a dummy __setup around otherwise it would
1650  * show up as an environment variable for init.
1651  */
1652 static __init int setup_clearcpuid(char *arg)
1653 {
1654 	return 1;
1655 }
1656 __setup("clearcpuid=", setup_clearcpuid);
1657 
1658 #ifdef CONFIG_X86_64
1659 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1660 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1661 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1662 
1663 /*
1664  * The following percpu variables are hot.  Align current_task to
1665  * cacheline size such that they fall in the same cacheline.
1666  */
1667 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1668 	&init_task;
1669 EXPORT_PER_CPU_SYMBOL(current_task);
1670 
1671 DEFINE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
1672 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1673 
1674 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1675 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1676 
1677 /* May not be marked __init: used by software suspend */
1678 void syscall_init(void)
1679 {
1680 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1681 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1682 
1683 #ifdef CONFIG_IA32_EMULATION
1684 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1685 	/*
1686 	 * This only works on Intel CPUs.
1687 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1688 	 * This does not cause SYSENTER to jump to the wrong location, because
1689 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1690 	 */
1691 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1692 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1693 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1694 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1695 #else
1696 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1697 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1698 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1699 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1700 #endif
1701 
1702 	/* Flags to clear on syscall */
1703 	wrmsrl(MSR_SYSCALL_MASK,
1704 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1705 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1706 }
1707 
1708 DEFINE_PER_CPU(int, debug_stack_usage);
1709 DEFINE_PER_CPU(u32, debug_idt_ctr);
1710 
1711 void debug_stack_set_zero(void)
1712 {
1713 	this_cpu_inc(debug_idt_ctr);
1714 	load_current_idt();
1715 }
1716 NOKPROBE_SYMBOL(debug_stack_set_zero);
1717 
1718 void debug_stack_reset(void)
1719 {
1720 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1721 		return;
1722 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1723 		load_current_idt();
1724 }
1725 NOKPROBE_SYMBOL(debug_stack_reset);
1726 
1727 #else	/* CONFIG_X86_64 */
1728 
1729 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1730 EXPORT_PER_CPU_SYMBOL(current_task);
1731 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1732 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1733 
1734 /*
1735  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1736  * the top of the kernel stack.  Use an extra percpu variable to track the
1737  * top of the kernel stack directly.
1738  */
1739 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1740 	(unsigned long)&init_thread_union + THREAD_SIZE;
1741 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1742 
1743 #ifdef CONFIG_STACKPROTECTOR
1744 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1745 #endif
1746 
1747 #endif	/* CONFIG_X86_64 */
1748 
1749 /*
1750  * Clear all 6 debug registers:
1751  */
1752 static void clear_all_debug_regs(void)
1753 {
1754 	int i;
1755 
1756 	for (i = 0; i < 8; i++) {
1757 		/* Ignore db4, db5 */
1758 		if ((i == 4) || (i == 5))
1759 			continue;
1760 
1761 		set_debugreg(0, i);
1762 	}
1763 }
1764 
1765 #ifdef CONFIG_KGDB
1766 /*
1767  * Restore debug regs if using kgdbwait and you have a kernel debugger
1768  * connection established.
1769  */
1770 static void dbg_restore_debug_regs(void)
1771 {
1772 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1773 		arch_kgdb_ops.correct_hw_break();
1774 }
1775 #else /* ! CONFIG_KGDB */
1776 #define dbg_restore_debug_regs()
1777 #endif /* ! CONFIG_KGDB */
1778 
1779 static void wait_for_master_cpu(int cpu)
1780 {
1781 #ifdef CONFIG_SMP
1782 	/*
1783 	 * wait for ACK from master CPU before continuing
1784 	 * with AP initialization
1785 	 */
1786 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1787 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1788 		cpu_relax();
1789 #endif
1790 }
1791 
1792 #ifdef CONFIG_X86_64
1793 static inline void setup_getcpu(int cpu)
1794 {
1795 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1796 	struct desc_struct d = { };
1797 
1798 	if (boot_cpu_has(X86_FEATURE_RDTSCP))
1799 		write_rdtscp_aux(cpudata);
1800 
1801 	/* Store CPU and node number in limit. */
1802 	d.limit0 = cpudata;
1803 	d.limit1 = cpudata >> 16;
1804 
1805 	d.type = 5;		/* RO data, expand down, accessed */
1806 	d.dpl = 3;		/* Visible to user code */
1807 	d.s = 1;		/* Not a system segment */
1808 	d.p = 1;		/* Present */
1809 	d.d = 1;		/* 32-bit */
1810 
1811 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1812 }
1813 
1814 static inline void ucode_cpu_init(int cpu)
1815 {
1816 	if (cpu)
1817 		load_ucode_ap();
1818 }
1819 
1820 static inline void tss_setup_ist(struct tss_struct *tss)
1821 {
1822 	/* Set up the per-CPU TSS IST stacks */
1823 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1824 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1825 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1826 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1827 }
1828 
1829 #else /* CONFIG_X86_64 */
1830 
1831 static inline void setup_getcpu(int cpu) { }
1832 
1833 static inline void ucode_cpu_init(int cpu)
1834 {
1835 	show_ucode_info_early();
1836 }
1837 
1838 static inline void tss_setup_ist(struct tss_struct *tss) { }
1839 
1840 #endif /* !CONFIG_X86_64 */
1841 
1842 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
1843 {
1844 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
1845 
1846 #ifdef CONFIG_X86_IOPL_IOPERM
1847 	tss->io_bitmap.prev_max = 0;
1848 	tss->io_bitmap.prev_sequence = 0;
1849 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
1850 	/*
1851 	 * Invalidate the extra array entry past the end of the all
1852 	 * permission bitmap as required by the hardware.
1853 	 */
1854 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
1855 #endif
1856 }
1857 
1858 /*
1859  * cpu_init() initializes state that is per-CPU. Some data is already
1860  * initialized (naturally) in the bootstrap process, such as the GDT
1861  * and IDT. We reload them nevertheless, this function acts as a
1862  * 'CPU state barrier', nothing should get across.
1863  */
1864 void cpu_init(void)
1865 {
1866 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
1867 	struct task_struct *cur = current;
1868 	int cpu = raw_smp_processor_id();
1869 
1870 	wait_for_master_cpu(cpu);
1871 
1872 	ucode_cpu_init(cpu);
1873 
1874 #ifdef CONFIG_NUMA
1875 	if (this_cpu_read(numa_node) == 0 &&
1876 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1877 		set_numa_node(early_cpu_to_node(cpu));
1878 #endif
1879 	setup_getcpu(cpu);
1880 
1881 	pr_debug("Initializing CPU#%d\n", cpu);
1882 
1883 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
1884 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
1885 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1886 
1887 	/*
1888 	 * Initialize the per-CPU GDT with the boot GDT,
1889 	 * and set up the GDT descriptor:
1890 	 */
1891 	switch_to_new_gdt(cpu);
1892 	load_current_idt();
1893 
1894 	if (IS_ENABLED(CONFIG_X86_64)) {
1895 		loadsegment(fs, 0);
1896 		memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1897 		syscall_init();
1898 
1899 		wrmsrl(MSR_FS_BASE, 0);
1900 		wrmsrl(MSR_KERNEL_GS_BASE, 0);
1901 		barrier();
1902 
1903 		x2apic_setup();
1904 	}
1905 
1906 	mmgrab(&init_mm);
1907 	cur->active_mm = &init_mm;
1908 	BUG_ON(cur->mm);
1909 	initialize_tlbstate_and_flush();
1910 	enter_lazy_tlb(&init_mm, cur);
1911 
1912 	/* Initialize the TSS. */
1913 	tss_setup_ist(tss);
1914 	tss_setup_io_bitmap(tss);
1915 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1916 
1917 	load_TR_desc();
1918 	/*
1919 	 * sp0 points to the entry trampoline stack regardless of what task
1920 	 * is running.
1921 	 */
1922 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1923 
1924 	load_mm_ldt(&init_mm);
1925 
1926 	clear_all_debug_regs();
1927 	dbg_restore_debug_regs();
1928 
1929 	doublefault_init_cpu_tss();
1930 
1931 	fpu__init_cpu();
1932 
1933 	if (is_uv_system())
1934 		uv_cpu_init();
1935 
1936 	load_fixmap_gdt(cpu);
1937 }
1938 
1939 /*
1940  * The microcode loader calls this upon late microcode load to recheck features,
1941  * only when microcode has been updated. Caller holds microcode_mutex and CPU
1942  * hotplug lock.
1943  */
1944 void microcode_check(void)
1945 {
1946 	struct cpuinfo_x86 info;
1947 
1948 	perf_check_microcode();
1949 
1950 	/* Reload CPUID max function as it might've changed. */
1951 	info.cpuid_level = cpuid_eax(0);
1952 
1953 	/*
1954 	 * Copy all capability leafs to pick up the synthetic ones so that
1955 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
1956 	 * get overwritten in get_cpu_cap().
1957 	 */
1958 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
1959 
1960 	get_cpu_cap(&info);
1961 
1962 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
1963 		return;
1964 
1965 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
1966 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
1967 }
1968 
1969 /*
1970  * Invoked from core CPU hotplug code after hotplug operations
1971  */
1972 void arch_smt_update(void)
1973 {
1974 	/* Handle the speculative execution misfeatures */
1975 	cpu_bugs_smt_update();
1976 	/* Check whether IPI broadcasting can be enabled */
1977 	apic_smt_update();
1978 }
1979