xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 715f23b6)
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/init.h>
18 #include <linux/kprobes.h>
19 #include <linux/kgdb.h>
20 #include <linux/smp.h>
21 #include <linux/io.h>
22 #include <linux/syscore_ops.h>
23 
24 #include <asm/stackprotector.h>
25 #include <asm/perf_event.h>
26 #include <asm/mmu_context.h>
27 #include <asm/doublefault.h>
28 #include <asm/archrandom.h>
29 #include <asm/hypervisor.h>
30 #include <asm/processor.h>
31 #include <asm/tlbflush.h>
32 #include <asm/debugreg.h>
33 #include <asm/sections.h>
34 #include <asm/vsyscall.h>
35 #include <linux/topology.h>
36 #include <linux/cpumask.h>
37 #include <asm/pgtable.h>
38 #include <linux/atomic.h>
39 #include <asm/proto.h>
40 #include <asm/setup.h>
41 #include <asm/apic.h>
42 #include <asm/desc.h>
43 #include <asm/fpu/internal.h>
44 #include <asm/mtrr.h>
45 #include <asm/hwcap2.h>
46 #include <linux/numa.h>
47 #include <asm/asm.h>
48 #include <asm/bugs.h>
49 #include <asm/cpu.h>
50 #include <asm/mce.h>
51 #include <asm/msr.h>
52 #include <asm/pat.h>
53 #include <asm/microcode.h>
54 #include <asm/microcode_intel.h>
55 #include <asm/intel-family.h>
56 #include <asm/cpu_device_id.h>
57 #include <asm/uv/uv.h>
58 
59 #include "cpu.h"
60 
61 u32 elf_hwcap2 __read_mostly;
62 
63 /* all of these masks are initialized in setup_cpu_local_masks() */
64 cpumask_var_t cpu_initialized_mask;
65 cpumask_var_t cpu_callout_mask;
66 cpumask_var_t cpu_callin_mask;
67 
68 /* representing cpus for which sibling maps can be computed */
69 cpumask_var_t cpu_sibling_setup_mask;
70 
71 /* Number of siblings per CPU package */
72 int smp_num_siblings = 1;
73 EXPORT_SYMBOL(smp_num_siblings);
74 
75 /* Last level cache ID of each logical CPU */
76 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
77 
78 /* correctly size the local cpu masks */
79 void __init setup_cpu_local_masks(void)
80 {
81 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
82 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
83 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
84 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
85 }
86 
87 static void default_init(struct cpuinfo_x86 *c)
88 {
89 #ifdef CONFIG_X86_64
90 	cpu_detect_cache_sizes(c);
91 #else
92 	/* Not much we can do here... */
93 	/* Check if at least it has cpuid */
94 	if (c->cpuid_level == -1) {
95 		/* No cpuid. It must be an ancient CPU */
96 		if (c->x86 == 4)
97 			strcpy(c->x86_model_id, "486");
98 		else if (c->x86 == 3)
99 			strcpy(c->x86_model_id, "386");
100 	}
101 #endif
102 }
103 
104 static const struct cpu_dev default_cpu = {
105 	.c_init		= default_init,
106 	.c_vendor	= "Unknown",
107 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
108 };
109 
110 static const struct cpu_dev *this_cpu = &default_cpu;
111 
112 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
113 #ifdef CONFIG_X86_64
114 	/*
115 	 * We need valid kernel segments for data and code in long mode too
116 	 * IRET will check the segment types  kkeil 2000/10/28
117 	 * Also sysret mandates a special GDT layout
118 	 *
119 	 * TLS descriptors are currently at a different place compared to i386.
120 	 * Hopefully nobody expects them at a fixed place (Wine?)
121 	 */
122 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
123 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
124 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
125 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
126 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
127 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
128 #else
129 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
130 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
131 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
132 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
133 	/*
134 	 * Segments used for calling PnP BIOS have byte granularity.
135 	 * They code segments and data segments have fixed 64k limits,
136 	 * the transfer segment sizes are set at run time.
137 	 */
138 	/* 32-bit code */
139 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
140 	/* 16-bit code */
141 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
142 	/* 16-bit data */
143 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
144 	/* 16-bit data */
145 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
146 	/* 16-bit data */
147 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
148 	/*
149 	 * The APM segments have byte granularity and their bases
150 	 * are set at run time.  All have 64k limits.
151 	 */
152 	/* 32-bit code */
153 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
154 	/* 16-bit code */
155 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
156 	/* data */
157 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
158 
159 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
160 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
161 	GDT_STACK_CANARY_INIT
162 #endif
163 } };
164 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
165 
166 static int __init x86_mpx_setup(char *s)
167 {
168 	/* require an exact match without trailing characters */
169 	if (strlen(s))
170 		return 0;
171 
172 	/* do not emit a message if the feature is not present */
173 	if (!boot_cpu_has(X86_FEATURE_MPX))
174 		return 1;
175 
176 	setup_clear_cpu_cap(X86_FEATURE_MPX);
177 	pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
178 	return 1;
179 }
180 __setup("nompx", x86_mpx_setup);
181 
182 #ifdef CONFIG_X86_64
183 static int __init x86_nopcid_setup(char *s)
184 {
185 	/* nopcid doesn't accept parameters */
186 	if (s)
187 		return -EINVAL;
188 
189 	/* do not emit a message if the feature is not present */
190 	if (!boot_cpu_has(X86_FEATURE_PCID))
191 		return 0;
192 
193 	setup_clear_cpu_cap(X86_FEATURE_PCID);
194 	pr_info("nopcid: PCID feature disabled\n");
195 	return 0;
196 }
197 early_param("nopcid", x86_nopcid_setup);
198 #endif
199 
200 static int __init x86_noinvpcid_setup(char *s)
201 {
202 	/* noinvpcid doesn't accept parameters */
203 	if (s)
204 		return -EINVAL;
205 
206 	/* do not emit a message if the feature is not present */
207 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
208 		return 0;
209 
210 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
211 	pr_info("noinvpcid: INVPCID feature disabled\n");
212 	return 0;
213 }
214 early_param("noinvpcid", x86_noinvpcid_setup);
215 
216 #ifdef CONFIG_X86_32
217 static int cachesize_override = -1;
218 static int disable_x86_serial_nr = 1;
219 
220 static int __init cachesize_setup(char *str)
221 {
222 	get_option(&str, &cachesize_override);
223 	return 1;
224 }
225 __setup("cachesize=", cachesize_setup);
226 
227 static int __init x86_sep_setup(char *s)
228 {
229 	setup_clear_cpu_cap(X86_FEATURE_SEP);
230 	return 1;
231 }
232 __setup("nosep", x86_sep_setup);
233 
234 /* Standard macro to see if a specific flag is changeable */
235 static inline int flag_is_changeable_p(u32 flag)
236 {
237 	u32 f1, f2;
238 
239 	/*
240 	 * Cyrix and IDT cpus allow disabling of CPUID
241 	 * so the code below may return different results
242 	 * when it is executed before and after enabling
243 	 * the CPUID. Add "volatile" to not allow gcc to
244 	 * optimize the subsequent calls to this function.
245 	 */
246 	asm volatile ("pushfl		\n\t"
247 		      "pushfl		\n\t"
248 		      "popl %0		\n\t"
249 		      "movl %0, %1	\n\t"
250 		      "xorl %2, %0	\n\t"
251 		      "pushl %0		\n\t"
252 		      "popfl		\n\t"
253 		      "pushfl		\n\t"
254 		      "popl %0		\n\t"
255 		      "popfl		\n\t"
256 
257 		      : "=&r" (f1), "=&r" (f2)
258 		      : "ir" (flag));
259 
260 	return ((f1^f2) & flag) != 0;
261 }
262 
263 /* Probe for the CPUID instruction */
264 int have_cpuid_p(void)
265 {
266 	return flag_is_changeable_p(X86_EFLAGS_ID);
267 }
268 
269 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
270 {
271 	unsigned long lo, hi;
272 
273 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
274 		return;
275 
276 	/* Disable processor serial number: */
277 
278 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
279 	lo |= 0x200000;
280 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
281 
282 	pr_notice("CPU serial number disabled.\n");
283 	clear_cpu_cap(c, X86_FEATURE_PN);
284 
285 	/* Disabling the serial number may affect the cpuid level */
286 	c->cpuid_level = cpuid_eax(0);
287 }
288 
289 static int __init x86_serial_nr_setup(char *s)
290 {
291 	disable_x86_serial_nr = 0;
292 	return 1;
293 }
294 __setup("serialnumber", x86_serial_nr_setup);
295 #else
296 static inline int flag_is_changeable_p(u32 flag)
297 {
298 	return 1;
299 }
300 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
301 {
302 }
303 #endif
304 
305 static __init int setup_disable_smep(char *arg)
306 {
307 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
308 	/* Check for things that depend on SMEP being enabled: */
309 	check_mpx_erratum(&boot_cpu_data);
310 	return 1;
311 }
312 __setup("nosmep", setup_disable_smep);
313 
314 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
315 {
316 	if (cpu_has(c, X86_FEATURE_SMEP))
317 		cr4_set_bits(X86_CR4_SMEP);
318 }
319 
320 static __init int setup_disable_smap(char *arg)
321 {
322 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
323 	return 1;
324 }
325 __setup("nosmap", setup_disable_smap);
326 
327 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
328 {
329 	unsigned long eflags = native_save_fl();
330 
331 	/* This should have been cleared long ago */
332 	BUG_ON(eflags & X86_EFLAGS_AC);
333 
334 	if (cpu_has(c, X86_FEATURE_SMAP)) {
335 #ifdef CONFIG_X86_SMAP
336 		cr4_set_bits(X86_CR4_SMAP);
337 #else
338 		cr4_clear_bits(X86_CR4_SMAP);
339 #endif
340 	}
341 }
342 
343 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
344 {
345 	/* Check the boot processor, plus build option for UMIP. */
346 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
347 		goto out;
348 
349 	/* Check the current processor's cpuid bits. */
350 	if (!cpu_has(c, X86_FEATURE_UMIP))
351 		goto out;
352 
353 	cr4_set_bits(X86_CR4_UMIP);
354 
355 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
356 
357 	return;
358 
359 out:
360 	/*
361 	 * Make sure UMIP is disabled in case it was enabled in a
362 	 * previous boot (e.g., via kexec).
363 	 */
364 	cr4_clear_bits(X86_CR4_UMIP);
365 }
366 
367 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
368 static unsigned long cr4_pinned_bits __ro_after_init;
369 
370 void native_write_cr0(unsigned long val)
371 {
372 	unsigned long bits_missing = 0;
373 
374 set_register:
375 	asm volatile("mov %0,%%cr0": "+r" (val), "+m" (__force_order));
376 
377 	if (static_branch_likely(&cr_pinning)) {
378 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
379 			bits_missing = X86_CR0_WP;
380 			val |= bits_missing;
381 			goto set_register;
382 		}
383 		/* Warn after we've set the missing bits. */
384 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
385 	}
386 }
387 EXPORT_SYMBOL(native_write_cr0);
388 
389 void native_write_cr4(unsigned long val)
390 {
391 	unsigned long bits_missing = 0;
392 
393 set_register:
394 	asm volatile("mov %0,%%cr4": "+r" (val), "+m" (cr4_pinned_bits));
395 
396 	if (static_branch_likely(&cr_pinning)) {
397 		if (unlikely((val & cr4_pinned_bits) != cr4_pinned_bits)) {
398 			bits_missing = ~val & cr4_pinned_bits;
399 			val |= bits_missing;
400 			goto set_register;
401 		}
402 		/* Warn after we've set the missing bits. */
403 		WARN_ONCE(bits_missing, "CR4 bits went missing: %lx!?\n",
404 			  bits_missing);
405 	}
406 }
407 EXPORT_SYMBOL(native_write_cr4);
408 
409 void cr4_init(void)
410 {
411 	unsigned long cr4 = __read_cr4();
412 
413 	if (boot_cpu_has(X86_FEATURE_PCID))
414 		cr4 |= X86_CR4_PCIDE;
415 	if (static_branch_likely(&cr_pinning))
416 		cr4 |= cr4_pinned_bits;
417 
418 	__write_cr4(cr4);
419 
420 	/* Initialize cr4 shadow for this CPU. */
421 	this_cpu_write(cpu_tlbstate.cr4, cr4);
422 }
423 
424 /*
425  * Once CPU feature detection is finished (and boot params have been
426  * parsed), record any of the sensitive CR bits that are set, and
427  * enable CR pinning.
428  */
429 static void __init setup_cr_pinning(void)
430 {
431 	unsigned long mask;
432 
433 	mask = (X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP);
434 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & mask;
435 	static_key_enable(&cr_pinning.key);
436 }
437 
438 /*
439  * Protection Keys are not available in 32-bit mode.
440  */
441 static bool pku_disabled;
442 
443 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
444 {
445 	struct pkru_state *pk;
446 
447 	/* check the boot processor, plus compile options for PKU: */
448 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
449 		return;
450 	/* checks the actual processor's cpuid bits: */
451 	if (!cpu_has(c, X86_FEATURE_PKU))
452 		return;
453 	if (pku_disabled)
454 		return;
455 
456 	cr4_set_bits(X86_CR4_PKE);
457 	pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU);
458 	if (pk)
459 		pk->pkru = init_pkru_value;
460 	/*
461 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
462 	 * cpuid bit to be set.  We need to ensure that we
463 	 * update that bit in this CPU's "cpu_info".
464 	 */
465 	get_cpu_cap(c);
466 }
467 
468 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
469 static __init int setup_disable_pku(char *arg)
470 {
471 	/*
472 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
473 	 * runtime checks are against OSPKE so clearing the
474 	 * bit does nothing.
475 	 *
476 	 * This way, we will see "pku" in cpuinfo, but not
477 	 * "ospke", which is exactly what we want.  It shows
478 	 * that the CPU has PKU, but the OS has not enabled it.
479 	 * This happens to be exactly how a system would look
480 	 * if we disabled the config option.
481 	 */
482 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
483 	pku_disabled = true;
484 	return 1;
485 }
486 __setup("nopku", setup_disable_pku);
487 #endif /* CONFIG_X86_64 */
488 
489 /*
490  * Some CPU features depend on higher CPUID levels, which may not always
491  * be available due to CPUID level capping or broken virtualization
492  * software.  Add those features to this table to auto-disable them.
493  */
494 struct cpuid_dependent_feature {
495 	u32 feature;
496 	u32 level;
497 };
498 
499 static const struct cpuid_dependent_feature
500 cpuid_dependent_features[] = {
501 	{ X86_FEATURE_MWAIT,		0x00000005 },
502 	{ X86_FEATURE_DCA,		0x00000009 },
503 	{ X86_FEATURE_XSAVE,		0x0000000d },
504 	{ 0, 0 }
505 };
506 
507 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
508 {
509 	const struct cpuid_dependent_feature *df;
510 
511 	for (df = cpuid_dependent_features; df->feature; df++) {
512 
513 		if (!cpu_has(c, df->feature))
514 			continue;
515 		/*
516 		 * Note: cpuid_level is set to -1 if unavailable, but
517 		 * extended_extended_level is set to 0 if unavailable
518 		 * and the legitimate extended levels are all negative
519 		 * when signed; hence the weird messing around with
520 		 * signs here...
521 		 */
522 		if (!((s32)df->level < 0 ?
523 		     (u32)df->level > (u32)c->extended_cpuid_level :
524 		     (s32)df->level > (s32)c->cpuid_level))
525 			continue;
526 
527 		clear_cpu_cap(c, df->feature);
528 		if (!warn)
529 			continue;
530 
531 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
532 			x86_cap_flag(df->feature), df->level);
533 	}
534 }
535 
536 /*
537  * Naming convention should be: <Name> [(<Codename>)]
538  * This table only is used unless init_<vendor>() below doesn't set it;
539  * in particular, if CPUID levels 0x80000002..4 are supported, this
540  * isn't used
541  */
542 
543 /* Look up CPU names by table lookup. */
544 static const char *table_lookup_model(struct cpuinfo_x86 *c)
545 {
546 #ifdef CONFIG_X86_32
547 	const struct legacy_cpu_model_info *info;
548 
549 	if (c->x86_model >= 16)
550 		return NULL;	/* Range check */
551 
552 	if (!this_cpu)
553 		return NULL;
554 
555 	info = this_cpu->legacy_models;
556 
557 	while (info->family) {
558 		if (info->family == c->x86)
559 			return info->model_names[c->x86_model];
560 		info++;
561 	}
562 #endif
563 	return NULL;		/* Not found */
564 }
565 
566 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
567 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
568 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
569 
570 void load_percpu_segment(int cpu)
571 {
572 #ifdef CONFIG_X86_32
573 	loadsegment(fs, __KERNEL_PERCPU);
574 #else
575 	__loadsegment_simple(gs, 0);
576 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
577 #endif
578 	load_stack_canary_segment();
579 }
580 
581 #ifdef CONFIG_X86_32
582 /* The 32-bit entry code needs to find cpu_entry_area. */
583 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
584 #endif
585 
586 /* Load the original GDT from the per-cpu structure */
587 void load_direct_gdt(int cpu)
588 {
589 	struct desc_ptr gdt_descr;
590 
591 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
592 	gdt_descr.size = GDT_SIZE - 1;
593 	load_gdt(&gdt_descr);
594 }
595 EXPORT_SYMBOL_GPL(load_direct_gdt);
596 
597 /* Load a fixmap remapping of the per-cpu GDT */
598 void load_fixmap_gdt(int cpu)
599 {
600 	struct desc_ptr gdt_descr;
601 
602 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
603 	gdt_descr.size = GDT_SIZE - 1;
604 	load_gdt(&gdt_descr);
605 }
606 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
607 
608 /*
609  * Current gdt points %fs at the "master" per-cpu area: after this,
610  * it's on the real one.
611  */
612 void switch_to_new_gdt(int cpu)
613 {
614 	/* Load the original GDT */
615 	load_direct_gdt(cpu);
616 	/* Reload the per-cpu base */
617 	load_percpu_segment(cpu);
618 }
619 
620 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
621 
622 static void get_model_name(struct cpuinfo_x86 *c)
623 {
624 	unsigned int *v;
625 	char *p, *q, *s;
626 
627 	if (c->extended_cpuid_level < 0x80000004)
628 		return;
629 
630 	v = (unsigned int *)c->x86_model_id;
631 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
632 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
633 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
634 	c->x86_model_id[48] = 0;
635 
636 	/* Trim whitespace */
637 	p = q = s = &c->x86_model_id[0];
638 
639 	while (*p == ' ')
640 		p++;
641 
642 	while (*p) {
643 		/* Note the last non-whitespace index */
644 		if (!isspace(*p))
645 			s = q;
646 
647 		*q++ = *p++;
648 	}
649 
650 	*(s + 1) = '\0';
651 }
652 
653 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
654 {
655 	unsigned int eax, ebx, ecx, edx;
656 
657 	c->x86_max_cores = 1;
658 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
659 		return;
660 
661 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
662 	if (eax & 0x1f)
663 		c->x86_max_cores = (eax >> 26) + 1;
664 }
665 
666 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
667 {
668 	unsigned int n, dummy, ebx, ecx, edx, l2size;
669 
670 	n = c->extended_cpuid_level;
671 
672 	if (n >= 0x80000005) {
673 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
674 		c->x86_cache_size = (ecx>>24) + (edx>>24);
675 #ifdef CONFIG_X86_64
676 		/* On K8 L1 TLB is inclusive, so don't count it */
677 		c->x86_tlbsize = 0;
678 #endif
679 	}
680 
681 	if (n < 0x80000006)	/* Some chips just has a large L1. */
682 		return;
683 
684 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
685 	l2size = ecx >> 16;
686 
687 #ifdef CONFIG_X86_64
688 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
689 #else
690 	/* do processor-specific cache resizing */
691 	if (this_cpu->legacy_cache_size)
692 		l2size = this_cpu->legacy_cache_size(c, l2size);
693 
694 	/* Allow user to override all this if necessary. */
695 	if (cachesize_override != -1)
696 		l2size = cachesize_override;
697 
698 	if (l2size == 0)
699 		return;		/* Again, no L2 cache is possible */
700 #endif
701 
702 	c->x86_cache_size = l2size;
703 }
704 
705 u16 __read_mostly tlb_lli_4k[NR_INFO];
706 u16 __read_mostly tlb_lli_2m[NR_INFO];
707 u16 __read_mostly tlb_lli_4m[NR_INFO];
708 u16 __read_mostly tlb_lld_4k[NR_INFO];
709 u16 __read_mostly tlb_lld_2m[NR_INFO];
710 u16 __read_mostly tlb_lld_4m[NR_INFO];
711 u16 __read_mostly tlb_lld_1g[NR_INFO];
712 
713 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
714 {
715 	if (this_cpu->c_detect_tlb)
716 		this_cpu->c_detect_tlb(c);
717 
718 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
719 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
720 		tlb_lli_4m[ENTRIES]);
721 
722 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
723 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
724 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
725 }
726 
727 int detect_ht_early(struct cpuinfo_x86 *c)
728 {
729 #ifdef CONFIG_SMP
730 	u32 eax, ebx, ecx, edx;
731 
732 	if (!cpu_has(c, X86_FEATURE_HT))
733 		return -1;
734 
735 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
736 		return -1;
737 
738 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
739 		return -1;
740 
741 	cpuid(1, &eax, &ebx, &ecx, &edx);
742 
743 	smp_num_siblings = (ebx & 0xff0000) >> 16;
744 	if (smp_num_siblings == 1)
745 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
746 #endif
747 	return 0;
748 }
749 
750 void detect_ht(struct cpuinfo_x86 *c)
751 {
752 #ifdef CONFIG_SMP
753 	int index_msb, core_bits;
754 
755 	if (detect_ht_early(c) < 0)
756 		return;
757 
758 	index_msb = get_count_order(smp_num_siblings);
759 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
760 
761 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
762 
763 	index_msb = get_count_order(smp_num_siblings);
764 
765 	core_bits = get_count_order(c->x86_max_cores);
766 
767 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
768 				       ((1 << core_bits) - 1);
769 #endif
770 }
771 
772 static void get_cpu_vendor(struct cpuinfo_x86 *c)
773 {
774 	char *v = c->x86_vendor_id;
775 	int i;
776 
777 	for (i = 0; i < X86_VENDOR_NUM; i++) {
778 		if (!cpu_devs[i])
779 			break;
780 
781 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
782 		    (cpu_devs[i]->c_ident[1] &&
783 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
784 
785 			this_cpu = cpu_devs[i];
786 			c->x86_vendor = this_cpu->c_x86_vendor;
787 			return;
788 		}
789 	}
790 
791 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
792 		    "CPU: Your system may be unstable.\n", v);
793 
794 	c->x86_vendor = X86_VENDOR_UNKNOWN;
795 	this_cpu = &default_cpu;
796 }
797 
798 void cpu_detect(struct cpuinfo_x86 *c)
799 {
800 	/* Get vendor name */
801 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
802 	      (unsigned int *)&c->x86_vendor_id[0],
803 	      (unsigned int *)&c->x86_vendor_id[8],
804 	      (unsigned int *)&c->x86_vendor_id[4]);
805 
806 	c->x86 = 4;
807 	/* Intel-defined flags: level 0x00000001 */
808 	if (c->cpuid_level >= 0x00000001) {
809 		u32 junk, tfms, cap0, misc;
810 
811 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
812 		c->x86		= x86_family(tfms);
813 		c->x86_model	= x86_model(tfms);
814 		c->x86_stepping	= x86_stepping(tfms);
815 
816 		if (cap0 & (1<<19)) {
817 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
818 			c->x86_cache_alignment = c->x86_clflush_size;
819 		}
820 	}
821 }
822 
823 static void apply_forced_caps(struct cpuinfo_x86 *c)
824 {
825 	int i;
826 
827 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
828 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
829 		c->x86_capability[i] |= cpu_caps_set[i];
830 	}
831 }
832 
833 static void init_speculation_control(struct cpuinfo_x86 *c)
834 {
835 	/*
836 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
837 	 * and they also have a different bit for STIBP support. Also,
838 	 * a hypervisor might have set the individual AMD bits even on
839 	 * Intel CPUs, for finer-grained selection of what's available.
840 	 */
841 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
842 		set_cpu_cap(c, X86_FEATURE_IBRS);
843 		set_cpu_cap(c, X86_FEATURE_IBPB);
844 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
845 	}
846 
847 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
848 		set_cpu_cap(c, X86_FEATURE_STIBP);
849 
850 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
851 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
852 		set_cpu_cap(c, X86_FEATURE_SSBD);
853 
854 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
855 		set_cpu_cap(c, X86_FEATURE_IBRS);
856 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
857 	}
858 
859 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
860 		set_cpu_cap(c, X86_FEATURE_IBPB);
861 
862 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
863 		set_cpu_cap(c, X86_FEATURE_STIBP);
864 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
865 	}
866 
867 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
868 		set_cpu_cap(c, X86_FEATURE_SSBD);
869 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
870 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
871 	}
872 }
873 
874 static void init_cqm(struct cpuinfo_x86 *c)
875 {
876 	if (!cpu_has(c, X86_FEATURE_CQM_LLC)) {
877 		c->x86_cache_max_rmid  = -1;
878 		c->x86_cache_occ_scale = -1;
879 		return;
880 	}
881 
882 	/* will be overridden if occupancy monitoring exists */
883 	c->x86_cache_max_rmid = cpuid_ebx(0xf);
884 
885 	if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC) ||
886 	    cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL) ||
887 	    cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)) {
888 		u32 eax, ebx, ecx, edx;
889 
890 		/* QoS sub-leaf, EAX=0Fh, ECX=1 */
891 		cpuid_count(0xf, 1, &eax, &ebx, &ecx, &edx);
892 
893 		c->x86_cache_max_rmid  = ecx;
894 		c->x86_cache_occ_scale = ebx;
895 	}
896 }
897 
898 void get_cpu_cap(struct cpuinfo_x86 *c)
899 {
900 	u32 eax, ebx, ecx, edx;
901 
902 	/* Intel-defined flags: level 0x00000001 */
903 	if (c->cpuid_level >= 0x00000001) {
904 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
905 
906 		c->x86_capability[CPUID_1_ECX] = ecx;
907 		c->x86_capability[CPUID_1_EDX] = edx;
908 	}
909 
910 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
911 	if (c->cpuid_level >= 0x00000006)
912 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
913 
914 	/* Additional Intel-defined flags: level 0x00000007 */
915 	if (c->cpuid_level >= 0x00000007) {
916 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
917 		c->x86_capability[CPUID_7_0_EBX] = ebx;
918 		c->x86_capability[CPUID_7_ECX] = ecx;
919 		c->x86_capability[CPUID_7_EDX] = edx;
920 
921 		/* Check valid sub-leaf index before accessing it */
922 		if (eax >= 1) {
923 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
924 			c->x86_capability[CPUID_7_1_EAX] = eax;
925 		}
926 	}
927 
928 	/* Extended state features: level 0x0000000d */
929 	if (c->cpuid_level >= 0x0000000d) {
930 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
931 
932 		c->x86_capability[CPUID_D_1_EAX] = eax;
933 	}
934 
935 	/* AMD-defined flags: level 0x80000001 */
936 	eax = cpuid_eax(0x80000000);
937 	c->extended_cpuid_level = eax;
938 
939 	if ((eax & 0xffff0000) == 0x80000000) {
940 		if (eax >= 0x80000001) {
941 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
942 
943 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
944 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
945 		}
946 	}
947 
948 	if (c->extended_cpuid_level >= 0x80000007) {
949 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
950 
951 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
952 		c->x86_power = edx;
953 	}
954 
955 	if (c->extended_cpuid_level >= 0x80000008) {
956 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
957 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
958 	}
959 
960 	if (c->extended_cpuid_level >= 0x8000000a)
961 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
962 
963 	init_scattered_cpuid_features(c);
964 	init_speculation_control(c);
965 	init_cqm(c);
966 
967 	/*
968 	 * Clear/Set all flags overridden by options, after probe.
969 	 * This needs to happen each time we re-probe, which may happen
970 	 * several times during CPU initialization.
971 	 */
972 	apply_forced_caps(c);
973 }
974 
975 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
976 {
977 	u32 eax, ebx, ecx, edx;
978 
979 	if (c->extended_cpuid_level >= 0x80000008) {
980 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
981 
982 		c->x86_virt_bits = (eax >> 8) & 0xff;
983 		c->x86_phys_bits = eax & 0xff;
984 	}
985 #ifdef CONFIG_X86_32
986 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
987 		c->x86_phys_bits = 36;
988 #endif
989 	c->x86_cache_bits = c->x86_phys_bits;
990 }
991 
992 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
993 {
994 #ifdef CONFIG_X86_32
995 	int i;
996 
997 	/*
998 	 * First of all, decide if this is a 486 or higher
999 	 * It's a 486 if we can modify the AC flag
1000 	 */
1001 	if (flag_is_changeable_p(X86_EFLAGS_AC))
1002 		c->x86 = 4;
1003 	else
1004 		c->x86 = 3;
1005 
1006 	for (i = 0; i < X86_VENDOR_NUM; i++)
1007 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
1008 			c->x86_vendor_id[0] = 0;
1009 			cpu_devs[i]->c_identify(c);
1010 			if (c->x86_vendor_id[0]) {
1011 				get_cpu_vendor(c);
1012 				break;
1013 			}
1014 		}
1015 #endif
1016 }
1017 
1018 #define NO_SPECULATION		BIT(0)
1019 #define NO_MELTDOWN		BIT(1)
1020 #define NO_SSB			BIT(2)
1021 #define NO_L1TF			BIT(3)
1022 #define NO_MDS			BIT(4)
1023 #define MSBDS_ONLY		BIT(5)
1024 #define NO_SWAPGS		BIT(6)
1025 #define NO_ITLB_MULTIHIT	BIT(7)
1026 
1027 #define VULNWL(_vendor, _family, _model, _whitelist)	\
1028 	{ X86_VENDOR_##_vendor, _family, _model, X86_FEATURE_ANY, _whitelist }
1029 
1030 #define VULNWL_INTEL(model, whitelist)		\
1031 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1032 
1033 #define VULNWL_AMD(family, whitelist)		\
1034 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1035 
1036 #define VULNWL_HYGON(family, whitelist)		\
1037 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1038 
1039 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1040 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1041 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1042 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1043 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1044 
1045 	/* Intel Family 6 */
1046 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1047 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION | NO_ITLB_MULTIHIT),
1048 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1049 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1050 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1051 
1052 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1053 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1054 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1055 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1056 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1057 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1058 
1059 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1060 
1061 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1062 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1063 
1064 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1065 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1066 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1067 
1068 	/*
1069 	 * Technically, swapgs isn't serializing on AMD (despite it previously
1070 	 * being documented as such in the APM).  But according to AMD, %gs is
1071 	 * updated non-speculatively, and the issuing of %gs-relative memory
1072 	 * operands will be blocked until the %gs update completes, which is
1073 	 * good enough for our purposes.
1074 	 */
1075 
1076 	VULNWL_INTEL(ATOM_TREMONT_D,		NO_ITLB_MULTIHIT),
1077 
1078 	/* AMD Family 0xf - 0x12 */
1079 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1080 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1081 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1082 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1083 
1084 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1085 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1086 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1087 	{}
1088 };
1089 
1090 static bool __init cpu_matches(unsigned long which)
1091 {
1092 	const struct x86_cpu_id *m = x86_match_cpu(cpu_vuln_whitelist);
1093 
1094 	return m && !!(m->driver_data & which);
1095 }
1096 
1097 u64 x86_read_arch_cap_msr(void)
1098 {
1099 	u64 ia32_cap = 0;
1100 
1101 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1102 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1103 
1104 	return ia32_cap;
1105 }
1106 
1107 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1108 {
1109 	u64 ia32_cap = x86_read_arch_cap_msr();
1110 
1111 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1112 	if (!cpu_matches(NO_ITLB_MULTIHIT) && !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO))
1113 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1114 
1115 	if (cpu_matches(NO_SPECULATION))
1116 		return;
1117 
1118 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1119 	setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1120 
1121 	if (!cpu_matches(NO_SSB) && !(ia32_cap & ARCH_CAP_SSB_NO) &&
1122 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1123 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1124 
1125 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1126 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1127 
1128 	if (!cpu_matches(NO_MDS) && !(ia32_cap & ARCH_CAP_MDS_NO)) {
1129 		setup_force_cpu_bug(X86_BUG_MDS);
1130 		if (cpu_matches(MSBDS_ONLY))
1131 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1132 	}
1133 
1134 	if (!cpu_matches(NO_SWAPGS))
1135 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1136 
1137 	/*
1138 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1139 	 *	- TSX is supported or
1140 	 *	- TSX_CTRL is present
1141 	 *
1142 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1143 	 * the kernel boot e.g. kexec.
1144 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1145 	 * update is not present or running as guest that don't get TSX_CTRL.
1146 	 */
1147 	if (!(ia32_cap & ARCH_CAP_TAA_NO) &&
1148 	    (cpu_has(c, X86_FEATURE_RTM) ||
1149 	     (ia32_cap & ARCH_CAP_TSX_CTRL_MSR)))
1150 		setup_force_cpu_bug(X86_BUG_TAA);
1151 
1152 	if (cpu_matches(NO_MELTDOWN))
1153 		return;
1154 
1155 	/* Rogue Data Cache Load? No! */
1156 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1157 		return;
1158 
1159 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1160 
1161 	if (cpu_matches(NO_L1TF))
1162 		return;
1163 
1164 	setup_force_cpu_bug(X86_BUG_L1TF);
1165 }
1166 
1167 /*
1168  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1169  * unfortunately, that's not true in practice because of early VIA
1170  * chips and (more importantly) broken virtualizers that are not easy
1171  * to detect. In the latter case it doesn't even *fail* reliably, so
1172  * probing for it doesn't even work. Disable it completely on 32-bit
1173  * unless we can find a reliable way to detect all the broken cases.
1174  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1175  */
1176 static void detect_nopl(void)
1177 {
1178 #ifdef CONFIG_X86_32
1179 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1180 #else
1181 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1182 #endif
1183 }
1184 
1185 /*
1186  * Do minimum CPU detection early.
1187  * Fields really needed: vendor, cpuid_level, family, model, mask,
1188  * cache alignment.
1189  * The others are not touched to avoid unwanted side effects.
1190  *
1191  * WARNING: this function is only called on the boot CPU.  Don't add code
1192  * here that is supposed to run on all CPUs.
1193  */
1194 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1195 {
1196 #ifdef CONFIG_X86_64
1197 	c->x86_clflush_size = 64;
1198 	c->x86_phys_bits = 36;
1199 	c->x86_virt_bits = 48;
1200 #else
1201 	c->x86_clflush_size = 32;
1202 	c->x86_phys_bits = 32;
1203 	c->x86_virt_bits = 32;
1204 #endif
1205 	c->x86_cache_alignment = c->x86_clflush_size;
1206 
1207 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1208 	c->extended_cpuid_level = 0;
1209 
1210 	if (!have_cpuid_p())
1211 		identify_cpu_without_cpuid(c);
1212 
1213 	/* cyrix could have cpuid enabled via c_identify()*/
1214 	if (have_cpuid_p()) {
1215 		cpu_detect(c);
1216 		get_cpu_vendor(c);
1217 		get_cpu_cap(c);
1218 		get_cpu_address_sizes(c);
1219 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1220 
1221 		if (this_cpu->c_early_init)
1222 			this_cpu->c_early_init(c);
1223 
1224 		c->cpu_index = 0;
1225 		filter_cpuid_features(c, false);
1226 
1227 		if (this_cpu->c_bsp_init)
1228 			this_cpu->c_bsp_init(c);
1229 	} else {
1230 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1231 	}
1232 
1233 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1234 
1235 	cpu_set_bug_bits(c);
1236 
1237 	fpu__init_system(c);
1238 
1239 #ifdef CONFIG_X86_32
1240 	/*
1241 	 * Regardless of whether PCID is enumerated, the SDM says
1242 	 * that it can't be enabled in 32-bit mode.
1243 	 */
1244 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1245 #endif
1246 
1247 	/*
1248 	 * Later in the boot process pgtable_l5_enabled() relies on
1249 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1250 	 * enabled by this point we need to clear the feature bit to avoid
1251 	 * false-positives at the later stage.
1252 	 *
1253 	 * pgtable_l5_enabled() can be false here for several reasons:
1254 	 *  - 5-level paging is disabled compile-time;
1255 	 *  - it's 32-bit kernel;
1256 	 *  - machine doesn't support 5-level paging;
1257 	 *  - user specified 'no5lvl' in kernel command line.
1258 	 */
1259 	if (!pgtable_l5_enabled())
1260 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1261 
1262 	detect_nopl();
1263 }
1264 
1265 void __init early_cpu_init(void)
1266 {
1267 	const struct cpu_dev *const *cdev;
1268 	int count = 0;
1269 
1270 #ifdef CONFIG_PROCESSOR_SELECT
1271 	pr_info("KERNEL supported cpus:\n");
1272 #endif
1273 
1274 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1275 		const struct cpu_dev *cpudev = *cdev;
1276 
1277 		if (count >= X86_VENDOR_NUM)
1278 			break;
1279 		cpu_devs[count] = cpudev;
1280 		count++;
1281 
1282 #ifdef CONFIG_PROCESSOR_SELECT
1283 		{
1284 			unsigned int j;
1285 
1286 			for (j = 0; j < 2; j++) {
1287 				if (!cpudev->c_ident[j])
1288 					continue;
1289 				pr_info("  %s %s\n", cpudev->c_vendor,
1290 					cpudev->c_ident[j]);
1291 			}
1292 		}
1293 #endif
1294 	}
1295 	early_identify_cpu(&boot_cpu_data);
1296 }
1297 
1298 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1299 {
1300 #ifdef CONFIG_X86_64
1301 	/*
1302 	 * Empirically, writing zero to a segment selector on AMD does
1303 	 * not clear the base, whereas writing zero to a segment
1304 	 * selector on Intel does clear the base.  Intel's behavior
1305 	 * allows slightly faster context switches in the common case
1306 	 * where GS is unused by the prev and next threads.
1307 	 *
1308 	 * Since neither vendor documents this anywhere that I can see,
1309 	 * detect it directly instead of hardcoding the choice by
1310 	 * vendor.
1311 	 *
1312 	 * I've designated AMD's behavior as the "bug" because it's
1313 	 * counterintuitive and less friendly.
1314 	 */
1315 
1316 	unsigned long old_base, tmp;
1317 	rdmsrl(MSR_FS_BASE, old_base);
1318 	wrmsrl(MSR_FS_BASE, 1);
1319 	loadsegment(fs, 0);
1320 	rdmsrl(MSR_FS_BASE, tmp);
1321 	if (tmp != 0)
1322 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1323 	wrmsrl(MSR_FS_BASE, old_base);
1324 #endif
1325 }
1326 
1327 static void generic_identify(struct cpuinfo_x86 *c)
1328 {
1329 	c->extended_cpuid_level = 0;
1330 
1331 	if (!have_cpuid_p())
1332 		identify_cpu_without_cpuid(c);
1333 
1334 	/* cyrix could have cpuid enabled via c_identify()*/
1335 	if (!have_cpuid_p())
1336 		return;
1337 
1338 	cpu_detect(c);
1339 
1340 	get_cpu_vendor(c);
1341 
1342 	get_cpu_cap(c);
1343 
1344 	get_cpu_address_sizes(c);
1345 
1346 	if (c->cpuid_level >= 0x00000001) {
1347 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1348 #ifdef CONFIG_X86_32
1349 # ifdef CONFIG_SMP
1350 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1351 # else
1352 		c->apicid = c->initial_apicid;
1353 # endif
1354 #endif
1355 		c->phys_proc_id = c->initial_apicid;
1356 	}
1357 
1358 	get_model_name(c); /* Default name */
1359 
1360 	detect_null_seg_behavior(c);
1361 
1362 	/*
1363 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1364 	 * systems that run Linux at CPL > 0 may or may not have the
1365 	 * issue, but, even if they have the issue, there's absolutely
1366 	 * nothing we can do about it because we can't use the real IRET
1367 	 * instruction.
1368 	 *
1369 	 * NB: For the time being, only 32-bit kernels support
1370 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1371 	 * whether to apply espfix using paravirt hooks.  If any
1372 	 * non-paravirt system ever shows up that does *not* have the
1373 	 * ESPFIX issue, we can change this.
1374 	 */
1375 #ifdef CONFIG_X86_32
1376 # ifdef CONFIG_PARAVIRT_XXL
1377 	do {
1378 		extern void native_iret(void);
1379 		if (pv_ops.cpu.iret == native_iret)
1380 			set_cpu_bug(c, X86_BUG_ESPFIX);
1381 	} while (0);
1382 # else
1383 	set_cpu_bug(c, X86_BUG_ESPFIX);
1384 # endif
1385 #endif
1386 }
1387 
1388 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
1389 {
1390 	/*
1391 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
1392 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
1393 	 * in case CQM bits really aren't there in this CPU.
1394 	 */
1395 	if (c != &boot_cpu_data) {
1396 		boot_cpu_data.x86_cache_max_rmid =
1397 			min(boot_cpu_data.x86_cache_max_rmid,
1398 			    c->x86_cache_max_rmid);
1399 	}
1400 }
1401 
1402 /*
1403  * Validate that ACPI/mptables have the same information about the
1404  * effective APIC id and update the package map.
1405  */
1406 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1407 {
1408 #ifdef CONFIG_SMP
1409 	unsigned int apicid, cpu = smp_processor_id();
1410 
1411 	apicid = apic->cpu_present_to_apicid(cpu);
1412 
1413 	if (apicid != c->apicid) {
1414 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1415 		       cpu, apicid, c->initial_apicid);
1416 	}
1417 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1418 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1419 #else
1420 	c->logical_proc_id = 0;
1421 #endif
1422 }
1423 
1424 /*
1425  * This does the hard work of actually picking apart the CPU stuff...
1426  */
1427 static void identify_cpu(struct cpuinfo_x86 *c)
1428 {
1429 	int i;
1430 
1431 	c->loops_per_jiffy = loops_per_jiffy;
1432 	c->x86_cache_size = 0;
1433 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1434 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1435 	c->x86_vendor_id[0] = '\0'; /* Unset */
1436 	c->x86_model_id[0] = '\0';  /* Unset */
1437 	c->x86_max_cores = 1;
1438 	c->x86_coreid_bits = 0;
1439 	c->cu_id = 0xff;
1440 #ifdef CONFIG_X86_64
1441 	c->x86_clflush_size = 64;
1442 	c->x86_phys_bits = 36;
1443 	c->x86_virt_bits = 48;
1444 #else
1445 	c->cpuid_level = -1;	/* CPUID not detected */
1446 	c->x86_clflush_size = 32;
1447 	c->x86_phys_bits = 32;
1448 	c->x86_virt_bits = 32;
1449 #endif
1450 	c->x86_cache_alignment = c->x86_clflush_size;
1451 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1452 
1453 	generic_identify(c);
1454 
1455 	if (this_cpu->c_identify)
1456 		this_cpu->c_identify(c);
1457 
1458 	/* Clear/Set all flags overridden by options, after probe */
1459 	apply_forced_caps(c);
1460 
1461 #ifdef CONFIG_X86_64
1462 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1463 #endif
1464 
1465 	/*
1466 	 * Vendor-specific initialization.  In this section we
1467 	 * canonicalize the feature flags, meaning if there are
1468 	 * features a certain CPU supports which CPUID doesn't
1469 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1470 	 * we handle them here.
1471 	 *
1472 	 * At the end of this section, c->x86_capability better
1473 	 * indicate the features this CPU genuinely supports!
1474 	 */
1475 	if (this_cpu->c_init)
1476 		this_cpu->c_init(c);
1477 
1478 	/* Disable the PN if appropriate */
1479 	squash_the_stupid_serial_number(c);
1480 
1481 	/* Set up SMEP/SMAP/UMIP */
1482 	setup_smep(c);
1483 	setup_smap(c);
1484 	setup_umip(c);
1485 
1486 	/*
1487 	 * The vendor-specific functions might have changed features.
1488 	 * Now we do "generic changes."
1489 	 */
1490 
1491 	/* Filter out anything that depends on CPUID levels we don't have */
1492 	filter_cpuid_features(c, true);
1493 
1494 	/* If the model name is still unset, do table lookup. */
1495 	if (!c->x86_model_id[0]) {
1496 		const char *p;
1497 		p = table_lookup_model(c);
1498 		if (p)
1499 			strcpy(c->x86_model_id, p);
1500 		else
1501 			/* Last resort... */
1502 			sprintf(c->x86_model_id, "%02x/%02x",
1503 				c->x86, c->x86_model);
1504 	}
1505 
1506 #ifdef CONFIG_X86_64
1507 	detect_ht(c);
1508 #endif
1509 
1510 	x86_init_rdrand(c);
1511 	x86_init_cache_qos(c);
1512 	setup_pku(c);
1513 
1514 	/*
1515 	 * Clear/Set all flags overridden by options, need do it
1516 	 * before following smp all cpus cap AND.
1517 	 */
1518 	apply_forced_caps(c);
1519 
1520 	/*
1521 	 * On SMP, boot_cpu_data holds the common feature set between
1522 	 * all CPUs; so make sure that we indicate which features are
1523 	 * common between the CPUs.  The first time this routine gets
1524 	 * executed, c == &boot_cpu_data.
1525 	 */
1526 	if (c != &boot_cpu_data) {
1527 		/* AND the already accumulated flags with these */
1528 		for (i = 0; i < NCAPINTS; i++)
1529 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1530 
1531 		/* OR, i.e. replicate the bug flags */
1532 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1533 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1534 	}
1535 
1536 	/* Init Machine Check Exception if available. */
1537 	mcheck_cpu_init(c);
1538 
1539 	select_idle_routine(c);
1540 
1541 #ifdef CONFIG_NUMA
1542 	numa_add_cpu(smp_processor_id());
1543 #endif
1544 }
1545 
1546 /*
1547  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1548  * on 32-bit kernels:
1549  */
1550 #ifdef CONFIG_X86_32
1551 void enable_sep_cpu(void)
1552 {
1553 	struct tss_struct *tss;
1554 	int cpu;
1555 
1556 	if (!boot_cpu_has(X86_FEATURE_SEP))
1557 		return;
1558 
1559 	cpu = get_cpu();
1560 	tss = &per_cpu(cpu_tss_rw, cpu);
1561 
1562 	/*
1563 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1564 	 * see the big comment in struct x86_hw_tss's definition.
1565 	 */
1566 
1567 	tss->x86_tss.ss1 = __KERNEL_CS;
1568 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1569 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1570 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1571 
1572 	put_cpu();
1573 }
1574 #endif
1575 
1576 void __init identify_boot_cpu(void)
1577 {
1578 	identify_cpu(&boot_cpu_data);
1579 #ifdef CONFIG_X86_32
1580 	sysenter_setup();
1581 	enable_sep_cpu();
1582 #endif
1583 	cpu_detect_tlb(&boot_cpu_data);
1584 	setup_cr_pinning();
1585 
1586 	tsx_init();
1587 }
1588 
1589 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1590 {
1591 	BUG_ON(c == &boot_cpu_data);
1592 	identify_cpu(c);
1593 #ifdef CONFIG_X86_32
1594 	enable_sep_cpu();
1595 #endif
1596 	mtrr_ap_init();
1597 	validate_apic_and_package_id(c);
1598 	x86_spec_ctrl_setup_ap();
1599 }
1600 
1601 static __init int setup_noclflush(char *arg)
1602 {
1603 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1604 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1605 	return 1;
1606 }
1607 __setup("noclflush", setup_noclflush);
1608 
1609 void print_cpu_info(struct cpuinfo_x86 *c)
1610 {
1611 	const char *vendor = NULL;
1612 
1613 	if (c->x86_vendor < X86_VENDOR_NUM) {
1614 		vendor = this_cpu->c_vendor;
1615 	} else {
1616 		if (c->cpuid_level >= 0)
1617 			vendor = c->x86_vendor_id;
1618 	}
1619 
1620 	if (vendor && !strstr(c->x86_model_id, vendor))
1621 		pr_cont("%s ", vendor);
1622 
1623 	if (c->x86_model_id[0])
1624 		pr_cont("%s", c->x86_model_id);
1625 	else
1626 		pr_cont("%d86", c->x86);
1627 
1628 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1629 
1630 	if (c->x86_stepping || c->cpuid_level >= 0)
1631 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1632 	else
1633 		pr_cont(")\n");
1634 }
1635 
1636 /*
1637  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1638  * But we need to keep a dummy __setup around otherwise it would
1639  * show up as an environment variable for init.
1640  */
1641 static __init int setup_clearcpuid(char *arg)
1642 {
1643 	return 1;
1644 }
1645 __setup("clearcpuid=", setup_clearcpuid);
1646 
1647 #ifdef CONFIG_X86_64
1648 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1649 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1650 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1651 
1652 /*
1653  * The following percpu variables are hot.  Align current_task to
1654  * cacheline size such that they fall in the same cacheline.
1655  */
1656 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1657 	&init_task;
1658 EXPORT_PER_CPU_SYMBOL(current_task);
1659 
1660 DEFINE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
1661 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1662 
1663 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1664 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1665 
1666 /* May not be marked __init: used by software suspend */
1667 void syscall_init(void)
1668 {
1669 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1670 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1671 
1672 #ifdef CONFIG_IA32_EMULATION
1673 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1674 	/*
1675 	 * This only works on Intel CPUs.
1676 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1677 	 * This does not cause SYSENTER to jump to the wrong location, because
1678 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1679 	 */
1680 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1681 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1682 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1683 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1684 #else
1685 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1686 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1687 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1688 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1689 #endif
1690 
1691 	/* Flags to clear on syscall */
1692 	wrmsrl(MSR_SYSCALL_MASK,
1693 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1694 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1695 }
1696 
1697 DEFINE_PER_CPU(int, debug_stack_usage);
1698 DEFINE_PER_CPU(u32, debug_idt_ctr);
1699 
1700 void debug_stack_set_zero(void)
1701 {
1702 	this_cpu_inc(debug_idt_ctr);
1703 	load_current_idt();
1704 }
1705 NOKPROBE_SYMBOL(debug_stack_set_zero);
1706 
1707 void debug_stack_reset(void)
1708 {
1709 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1710 		return;
1711 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1712 		load_current_idt();
1713 }
1714 NOKPROBE_SYMBOL(debug_stack_reset);
1715 
1716 #else	/* CONFIG_X86_64 */
1717 
1718 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1719 EXPORT_PER_CPU_SYMBOL(current_task);
1720 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1721 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1722 
1723 /*
1724  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1725  * the top of the kernel stack.  Use an extra percpu variable to track the
1726  * top of the kernel stack directly.
1727  */
1728 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1729 	(unsigned long)&init_thread_union + THREAD_SIZE;
1730 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1731 
1732 #ifdef CONFIG_STACKPROTECTOR
1733 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1734 #endif
1735 
1736 #endif	/* CONFIG_X86_64 */
1737 
1738 /*
1739  * Clear all 6 debug registers:
1740  */
1741 static void clear_all_debug_regs(void)
1742 {
1743 	int i;
1744 
1745 	for (i = 0; i < 8; i++) {
1746 		/* Ignore db4, db5 */
1747 		if ((i == 4) || (i == 5))
1748 			continue;
1749 
1750 		set_debugreg(0, i);
1751 	}
1752 }
1753 
1754 #ifdef CONFIG_KGDB
1755 /*
1756  * Restore debug regs if using kgdbwait and you have a kernel debugger
1757  * connection established.
1758  */
1759 static void dbg_restore_debug_regs(void)
1760 {
1761 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1762 		arch_kgdb_ops.correct_hw_break();
1763 }
1764 #else /* ! CONFIG_KGDB */
1765 #define dbg_restore_debug_regs()
1766 #endif /* ! CONFIG_KGDB */
1767 
1768 static void wait_for_master_cpu(int cpu)
1769 {
1770 #ifdef CONFIG_SMP
1771 	/*
1772 	 * wait for ACK from master CPU before continuing
1773 	 * with AP initialization
1774 	 */
1775 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1776 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1777 		cpu_relax();
1778 #endif
1779 }
1780 
1781 #ifdef CONFIG_X86_64
1782 static inline void setup_getcpu(int cpu)
1783 {
1784 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1785 	struct desc_struct d = { };
1786 
1787 	if (boot_cpu_has(X86_FEATURE_RDTSCP))
1788 		write_rdtscp_aux(cpudata);
1789 
1790 	/* Store CPU and node number in limit. */
1791 	d.limit0 = cpudata;
1792 	d.limit1 = cpudata >> 16;
1793 
1794 	d.type = 5;		/* RO data, expand down, accessed */
1795 	d.dpl = 3;		/* Visible to user code */
1796 	d.s = 1;		/* Not a system segment */
1797 	d.p = 1;		/* Present */
1798 	d.d = 1;		/* 32-bit */
1799 
1800 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1801 }
1802 
1803 static inline void ucode_cpu_init(int cpu)
1804 {
1805 	if (cpu)
1806 		load_ucode_ap();
1807 }
1808 
1809 static inline void tss_setup_ist(struct tss_struct *tss)
1810 {
1811 	/* Set up the per-CPU TSS IST stacks */
1812 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1813 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1814 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1815 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1816 }
1817 
1818 #else /* CONFIG_X86_64 */
1819 
1820 static inline void setup_getcpu(int cpu) { }
1821 
1822 static inline void ucode_cpu_init(int cpu)
1823 {
1824 	show_ucode_info_early();
1825 }
1826 
1827 static inline void tss_setup_ist(struct tss_struct *tss) { }
1828 
1829 #endif /* !CONFIG_X86_64 */
1830 
1831 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
1832 {
1833 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
1834 
1835 #ifdef CONFIG_X86_IOPL_IOPERM
1836 	tss->io_bitmap.prev_max = 0;
1837 	tss->io_bitmap.prev_sequence = 0;
1838 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
1839 	/*
1840 	 * Invalidate the extra array entry past the end of the all
1841 	 * permission bitmap as required by the hardware.
1842 	 */
1843 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
1844 #endif
1845 }
1846 
1847 /*
1848  * cpu_init() initializes state that is per-CPU. Some data is already
1849  * initialized (naturally) in the bootstrap process, such as the GDT
1850  * and IDT. We reload them nevertheless, this function acts as a
1851  * 'CPU state barrier', nothing should get across.
1852  */
1853 void cpu_init(void)
1854 {
1855 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
1856 	struct task_struct *cur = current;
1857 	int cpu = raw_smp_processor_id();
1858 
1859 	wait_for_master_cpu(cpu);
1860 
1861 	ucode_cpu_init(cpu);
1862 
1863 #ifdef CONFIG_NUMA
1864 	if (this_cpu_read(numa_node) == 0 &&
1865 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1866 		set_numa_node(early_cpu_to_node(cpu));
1867 #endif
1868 	setup_getcpu(cpu);
1869 
1870 	pr_debug("Initializing CPU#%d\n", cpu);
1871 
1872 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
1873 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
1874 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1875 
1876 	/*
1877 	 * Initialize the per-CPU GDT with the boot GDT,
1878 	 * and set up the GDT descriptor:
1879 	 */
1880 	switch_to_new_gdt(cpu);
1881 	load_current_idt();
1882 
1883 	if (IS_ENABLED(CONFIG_X86_64)) {
1884 		loadsegment(fs, 0);
1885 		memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1886 		syscall_init();
1887 
1888 		wrmsrl(MSR_FS_BASE, 0);
1889 		wrmsrl(MSR_KERNEL_GS_BASE, 0);
1890 		barrier();
1891 
1892 		x2apic_setup();
1893 	}
1894 
1895 	mmgrab(&init_mm);
1896 	cur->active_mm = &init_mm;
1897 	BUG_ON(cur->mm);
1898 	initialize_tlbstate_and_flush();
1899 	enter_lazy_tlb(&init_mm, cur);
1900 
1901 	/* Initialize the TSS. */
1902 	tss_setup_ist(tss);
1903 	tss_setup_io_bitmap(tss);
1904 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1905 
1906 	load_TR_desc();
1907 	/*
1908 	 * sp0 points to the entry trampoline stack regardless of what task
1909 	 * is running.
1910 	 */
1911 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1912 
1913 	load_mm_ldt(&init_mm);
1914 
1915 	clear_all_debug_regs();
1916 	dbg_restore_debug_regs();
1917 
1918 	doublefault_init_cpu_tss();
1919 
1920 	fpu__init_cpu();
1921 
1922 	if (is_uv_system())
1923 		uv_cpu_init();
1924 
1925 	load_fixmap_gdt(cpu);
1926 }
1927 
1928 /*
1929  * The microcode loader calls this upon late microcode load to recheck features,
1930  * only when microcode has been updated. Caller holds microcode_mutex and CPU
1931  * hotplug lock.
1932  */
1933 void microcode_check(void)
1934 {
1935 	struct cpuinfo_x86 info;
1936 
1937 	perf_check_microcode();
1938 
1939 	/* Reload CPUID max function as it might've changed. */
1940 	info.cpuid_level = cpuid_eax(0);
1941 
1942 	/*
1943 	 * Copy all capability leafs to pick up the synthetic ones so that
1944 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
1945 	 * get overwritten in get_cpu_cap().
1946 	 */
1947 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
1948 
1949 	get_cpu_cap(&info);
1950 
1951 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
1952 		return;
1953 
1954 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
1955 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
1956 }
1957 
1958 /*
1959  * Invoked from core CPU hotplug code after hotplug operations
1960  */
1961 void arch_smt_update(void)
1962 {
1963 	/* Handle the speculative execution misfeatures */
1964 	cpu_bugs_smt_update();
1965 	/* Check whether IPI broadcasting can be enabled */
1966 	apic_smt_update();
1967 }
1968