xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 2022ca0a)
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/archrandom.h>
28 #include <asm/hypervisor.h>
29 #include <asm/processor.h>
30 #include <asm/tlbflush.h>
31 #include <asm/debugreg.h>
32 #include <asm/sections.h>
33 #include <asm/vsyscall.h>
34 #include <linux/topology.h>
35 #include <linux/cpumask.h>
36 #include <asm/pgtable.h>
37 #include <linux/atomic.h>
38 #include <asm/proto.h>
39 #include <asm/setup.h>
40 #include <asm/apic.h>
41 #include <asm/desc.h>
42 #include <asm/fpu/internal.h>
43 #include <asm/mtrr.h>
44 #include <asm/hwcap2.h>
45 #include <linux/numa.h>
46 #include <asm/asm.h>
47 #include <asm/bugs.h>
48 #include <asm/cpu.h>
49 #include <asm/mce.h>
50 #include <asm/msr.h>
51 #include <asm/pat.h>
52 #include <asm/microcode.h>
53 #include <asm/microcode_intel.h>
54 #include <asm/intel-family.h>
55 #include <asm/cpu_device_id.h>
56 
57 #ifdef CONFIG_X86_LOCAL_APIC
58 #include <asm/uv/uv.h>
59 #endif
60 
61 #include "cpu.h"
62 
63 u32 elf_hwcap2 __read_mostly;
64 
65 /* all of these masks are initialized in setup_cpu_local_masks() */
66 cpumask_var_t cpu_initialized_mask;
67 cpumask_var_t cpu_callout_mask;
68 cpumask_var_t cpu_callin_mask;
69 
70 /* representing cpus for which sibling maps can be computed */
71 cpumask_var_t cpu_sibling_setup_mask;
72 
73 /* Number of siblings per CPU package */
74 int smp_num_siblings = 1;
75 EXPORT_SYMBOL(smp_num_siblings);
76 
77 /* Last level cache ID of each logical CPU */
78 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
79 
80 /* correctly size the local cpu masks */
81 void __init setup_cpu_local_masks(void)
82 {
83 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
84 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
85 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
86 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
87 }
88 
89 static void default_init(struct cpuinfo_x86 *c)
90 {
91 #ifdef CONFIG_X86_64
92 	cpu_detect_cache_sizes(c);
93 #else
94 	/* Not much we can do here... */
95 	/* Check if at least it has cpuid */
96 	if (c->cpuid_level == -1) {
97 		/* No cpuid. It must be an ancient CPU */
98 		if (c->x86 == 4)
99 			strcpy(c->x86_model_id, "486");
100 		else if (c->x86 == 3)
101 			strcpy(c->x86_model_id, "386");
102 	}
103 #endif
104 }
105 
106 static const struct cpu_dev default_cpu = {
107 	.c_init		= default_init,
108 	.c_vendor	= "Unknown",
109 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
110 };
111 
112 static const struct cpu_dev *this_cpu = &default_cpu;
113 
114 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
115 #ifdef CONFIG_X86_64
116 	/*
117 	 * We need valid kernel segments for data and code in long mode too
118 	 * IRET will check the segment types  kkeil 2000/10/28
119 	 * Also sysret mandates a special GDT layout
120 	 *
121 	 * TLS descriptors are currently at a different place compared to i386.
122 	 * Hopefully nobody expects them at a fixed place (Wine?)
123 	 */
124 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
125 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
126 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
127 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
128 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
129 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
130 #else
131 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
132 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
133 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
134 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
135 	/*
136 	 * Segments used for calling PnP BIOS have byte granularity.
137 	 * They code segments and data segments have fixed 64k limits,
138 	 * the transfer segment sizes are set at run time.
139 	 */
140 	/* 32-bit code */
141 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
142 	/* 16-bit code */
143 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
144 	/* 16-bit data */
145 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
146 	/* 16-bit data */
147 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
148 	/* 16-bit data */
149 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
150 	/*
151 	 * The APM segments have byte granularity and their bases
152 	 * are set at run time.  All have 64k limits.
153 	 */
154 	/* 32-bit code */
155 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
156 	/* 16-bit code */
157 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
158 	/* data */
159 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
160 
161 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
162 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
163 	GDT_STACK_CANARY_INIT
164 #endif
165 } };
166 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
167 
168 static int __init x86_mpx_setup(char *s)
169 {
170 	/* require an exact match without trailing characters */
171 	if (strlen(s))
172 		return 0;
173 
174 	/* do not emit a message if the feature is not present */
175 	if (!boot_cpu_has(X86_FEATURE_MPX))
176 		return 1;
177 
178 	setup_clear_cpu_cap(X86_FEATURE_MPX);
179 	pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
180 	return 1;
181 }
182 __setup("nompx", x86_mpx_setup);
183 
184 #ifdef CONFIG_X86_64
185 static int __init x86_nopcid_setup(char *s)
186 {
187 	/* nopcid doesn't accept parameters */
188 	if (s)
189 		return -EINVAL;
190 
191 	/* do not emit a message if the feature is not present */
192 	if (!boot_cpu_has(X86_FEATURE_PCID))
193 		return 0;
194 
195 	setup_clear_cpu_cap(X86_FEATURE_PCID);
196 	pr_info("nopcid: PCID feature disabled\n");
197 	return 0;
198 }
199 early_param("nopcid", x86_nopcid_setup);
200 #endif
201 
202 static int __init x86_noinvpcid_setup(char *s)
203 {
204 	/* noinvpcid doesn't accept parameters */
205 	if (s)
206 		return -EINVAL;
207 
208 	/* do not emit a message if the feature is not present */
209 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
210 		return 0;
211 
212 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
213 	pr_info("noinvpcid: INVPCID feature disabled\n");
214 	return 0;
215 }
216 early_param("noinvpcid", x86_noinvpcid_setup);
217 
218 #ifdef CONFIG_X86_32
219 static int cachesize_override = -1;
220 static int disable_x86_serial_nr = 1;
221 
222 static int __init cachesize_setup(char *str)
223 {
224 	get_option(&str, &cachesize_override);
225 	return 1;
226 }
227 __setup("cachesize=", cachesize_setup);
228 
229 static int __init x86_sep_setup(char *s)
230 {
231 	setup_clear_cpu_cap(X86_FEATURE_SEP);
232 	return 1;
233 }
234 __setup("nosep", x86_sep_setup);
235 
236 /* Standard macro to see if a specific flag is changeable */
237 static inline int flag_is_changeable_p(u32 flag)
238 {
239 	u32 f1, f2;
240 
241 	/*
242 	 * Cyrix and IDT cpus allow disabling of CPUID
243 	 * so the code below may return different results
244 	 * when it is executed before and after enabling
245 	 * the CPUID. Add "volatile" to not allow gcc to
246 	 * optimize the subsequent calls to this function.
247 	 */
248 	asm volatile ("pushfl		\n\t"
249 		      "pushfl		\n\t"
250 		      "popl %0		\n\t"
251 		      "movl %0, %1	\n\t"
252 		      "xorl %2, %0	\n\t"
253 		      "pushl %0		\n\t"
254 		      "popfl		\n\t"
255 		      "pushfl		\n\t"
256 		      "popl %0		\n\t"
257 		      "popfl		\n\t"
258 
259 		      : "=&r" (f1), "=&r" (f2)
260 		      : "ir" (flag));
261 
262 	return ((f1^f2) & flag) != 0;
263 }
264 
265 /* Probe for the CPUID instruction */
266 int have_cpuid_p(void)
267 {
268 	return flag_is_changeable_p(X86_EFLAGS_ID);
269 }
270 
271 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
272 {
273 	unsigned long lo, hi;
274 
275 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
276 		return;
277 
278 	/* Disable processor serial number: */
279 
280 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
281 	lo |= 0x200000;
282 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
283 
284 	pr_notice("CPU serial number disabled.\n");
285 	clear_cpu_cap(c, X86_FEATURE_PN);
286 
287 	/* Disabling the serial number may affect the cpuid level */
288 	c->cpuid_level = cpuid_eax(0);
289 }
290 
291 static int __init x86_serial_nr_setup(char *s)
292 {
293 	disable_x86_serial_nr = 0;
294 	return 1;
295 }
296 __setup("serialnumber", x86_serial_nr_setup);
297 #else
298 static inline int flag_is_changeable_p(u32 flag)
299 {
300 	return 1;
301 }
302 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
303 {
304 }
305 #endif
306 
307 static __init int setup_disable_smep(char *arg)
308 {
309 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
310 	/* Check for things that depend on SMEP being enabled: */
311 	check_mpx_erratum(&boot_cpu_data);
312 	return 1;
313 }
314 __setup("nosmep", setup_disable_smep);
315 
316 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
317 {
318 	if (cpu_has(c, X86_FEATURE_SMEP))
319 		cr4_set_bits(X86_CR4_SMEP);
320 }
321 
322 static __init int setup_disable_smap(char *arg)
323 {
324 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
325 	return 1;
326 }
327 __setup("nosmap", setup_disable_smap);
328 
329 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
330 {
331 	unsigned long eflags = native_save_fl();
332 
333 	/* This should have been cleared long ago */
334 	BUG_ON(eflags & X86_EFLAGS_AC);
335 
336 	if (cpu_has(c, X86_FEATURE_SMAP)) {
337 #ifdef CONFIG_X86_SMAP
338 		cr4_set_bits(X86_CR4_SMAP);
339 #else
340 		cr4_clear_bits(X86_CR4_SMAP);
341 #endif
342 	}
343 }
344 
345 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
346 {
347 	/* Check the boot processor, plus build option for UMIP. */
348 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
349 		goto out;
350 
351 	/* Check the current processor's cpuid bits. */
352 	if (!cpu_has(c, X86_FEATURE_UMIP))
353 		goto out;
354 
355 	cr4_set_bits(X86_CR4_UMIP);
356 
357 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
358 
359 	return;
360 
361 out:
362 	/*
363 	 * Make sure UMIP is disabled in case it was enabled in a
364 	 * previous boot (e.g., via kexec).
365 	 */
366 	cr4_clear_bits(X86_CR4_UMIP);
367 }
368 
369 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
370 static unsigned long cr4_pinned_bits __ro_after_init;
371 
372 void native_write_cr0(unsigned long val)
373 {
374 	unsigned long bits_missing = 0;
375 
376 set_register:
377 	asm volatile("mov %0,%%cr0": "+r" (val), "+m" (__force_order));
378 
379 	if (static_branch_likely(&cr_pinning)) {
380 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
381 			bits_missing = X86_CR0_WP;
382 			val |= bits_missing;
383 			goto set_register;
384 		}
385 		/* Warn after we've set the missing bits. */
386 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
387 	}
388 }
389 EXPORT_SYMBOL(native_write_cr0);
390 
391 void native_write_cr4(unsigned long val)
392 {
393 	unsigned long bits_missing = 0;
394 
395 set_register:
396 	asm volatile("mov %0,%%cr4": "+r" (val), "+m" (cr4_pinned_bits));
397 
398 	if (static_branch_likely(&cr_pinning)) {
399 		if (unlikely((val & cr4_pinned_bits) != cr4_pinned_bits)) {
400 			bits_missing = ~val & cr4_pinned_bits;
401 			val |= bits_missing;
402 			goto set_register;
403 		}
404 		/* Warn after we've set the missing bits. */
405 		WARN_ONCE(bits_missing, "CR4 bits went missing: %lx!?\n",
406 			  bits_missing);
407 	}
408 }
409 EXPORT_SYMBOL(native_write_cr4);
410 
411 void cr4_init(void)
412 {
413 	unsigned long cr4 = __read_cr4();
414 
415 	if (boot_cpu_has(X86_FEATURE_PCID))
416 		cr4 |= X86_CR4_PCIDE;
417 	if (static_branch_likely(&cr_pinning))
418 		cr4 |= cr4_pinned_bits;
419 
420 	__write_cr4(cr4);
421 
422 	/* Initialize cr4 shadow for this CPU. */
423 	this_cpu_write(cpu_tlbstate.cr4, cr4);
424 }
425 
426 /*
427  * Once CPU feature detection is finished (and boot params have been
428  * parsed), record any of the sensitive CR bits that are set, and
429  * enable CR pinning.
430  */
431 static void __init setup_cr_pinning(void)
432 {
433 	unsigned long mask;
434 
435 	mask = (X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP);
436 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & mask;
437 	static_key_enable(&cr_pinning.key);
438 }
439 
440 /*
441  * Protection Keys are not available in 32-bit mode.
442  */
443 static bool pku_disabled;
444 
445 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
446 {
447 	struct pkru_state *pk;
448 
449 	/* check the boot processor, plus compile options for PKU: */
450 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
451 		return;
452 	/* checks the actual processor's cpuid bits: */
453 	if (!cpu_has(c, X86_FEATURE_PKU))
454 		return;
455 	if (pku_disabled)
456 		return;
457 
458 	cr4_set_bits(X86_CR4_PKE);
459 	pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU);
460 	if (pk)
461 		pk->pkru = init_pkru_value;
462 	/*
463 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
464 	 * cpuid bit to be set.  We need to ensure that we
465 	 * update that bit in this CPU's "cpu_info".
466 	 */
467 	get_cpu_cap(c);
468 }
469 
470 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
471 static __init int setup_disable_pku(char *arg)
472 {
473 	/*
474 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
475 	 * runtime checks are against OSPKE so clearing the
476 	 * bit does nothing.
477 	 *
478 	 * This way, we will see "pku" in cpuinfo, but not
479 	 * "ospke", which is exactly what we want.  It shows
480 	 * that the CPU has PKU, but the OS has not enabled it.
481 	 * This happens to be exactly how a system would look
482 	 * if we disabled the config option.
483 	 */
484 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
485 	pku_disabled = true;
486 	return 1;
487 }
488 __setup("nopku", setup_disable_pku);
489 #endif /* CONFIG_X86_64 */
490 
491 /*
492  * Some CPU features depend on higher CPUID levels, which may not always
493  * be available due to CPUID level capping or broken virtualization
494  * software.  Add those features to this table to auto-disable them.
495  */
496 struct cpuid_dependent_feature {
497 	u32 feature;
498 	u32 level;
499 };
500 
501 static const struct cpuid_dependent_feature
502 cpuid_dependent_features[] = {
503 	{ X86_FEATURE_MWAIT,		0x00000005 },
504 	{ X86_FEATURE_DCA,		0x00000009 },
505 	{ X86_FEATURE_XSAVE,		0x0000000d },
506 	{ 0, 0 }
507 };
508 
509 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
510 {
511 	const struct cpuid_dependent_feature *df;
512 
513 	for (df = cpuid_dependent_features; df->feature; df++) {
514 
515 		if (!cpu_has(c, df->feature))
516 			continue;
517 		/*
518 		 * Note: cpuid_level is set to -1 if unavailable, but
519 		 * extended_extended_level is set to 0 if unavailable
520 		 * and the legitimate extended levels are all negative
521 		 * when signed; hence the weird messing around with
522 		 * signs here...
523 		 */
524 		if (!((s32)df->level < 0 ?
525 		     (u32)df->level > (u32)c->extended_cpuid_level :
526 		     (s32)df->level > (s32)c->cpuid_level))
527 			continue;
528 
529 		clear_cpu_cap(c, df->feature);
530 		if (!warn)
531 			continue;
532 
533 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
534 			x86_cap_flag(df->feature), df->level);
535 	}
536 }
537 
538 /*
539  * Naming convention should be: <Name> [(<Codename>)]
540  * This table only is used unless init_<vendor>() below doesn't set it;
541  * in particular, if CPUID levels 0x80000002..4 are supported, this
542  * isn't used
543  */
544 
545 /* Look up CPU names by table lookup. */
546 static const char *table_lookup_model(struct cpuinfo_x86 *c)
547 {
548 #ifdef CONFIG_X86_32
549 	const struct legacy_cpu_model_info *info;
550 
551 	if (c->x86_model >= 16)
552 		return NULL;	/* Range check */
553 
554 	if (!this_cpu)
555 		return NULL;
556 
557 	info = this_cpu->legacy_models;
558 
559 	while (info->family) {
560 		if (info->family == c->x86)
561 			return info->model_names[c->x86_model];
562 		info++;
563 	}
564 #endif
565 	return NULL;		/* Not found */
566 }
567 
568 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
569 __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
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 
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),
1047 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION),
1048 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION),
1049 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION),
1050 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION),
1051 
1052 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS),
1053 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS),
1054 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS),
1055 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS),
1056 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS),
1057 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS),
1058 
1059 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1060 
1061 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS),
1062 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS),
1063 
1064 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS),
1065 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS),
1066 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS),
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 	/* AMD Family 0xf - 0x12 */
1077 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS),
1078 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS),
1079 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS),
1080 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS),
1081 
1082 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1083 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS),
1084 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS),
1085 	{}
1086 };
1087 
1088 static bool __init cpu_matches(unsigned long which)
1089 {
1090 	const struct x86_cpu_id *m = x86_match_cpu(cpu_vuln_whitelist);
1091 
1092 	return m && !!(m->driver_data & which);
1093 }
1094 
1095 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1096 {
1097 	u64 ia32_cap = 0;
1098 
1099 	if (cpu_matches(NO_SPECULATION))
1100 		return;
1101 
1102 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1103 	setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1104 
1105 	if (cpu_has(c, X86_FEATURE_ARCH_CAPABILITIES))
1106 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1107 
1108 	if (!cpu_matches(NO_SSB) && !(ia32_cap & ARCH_CAP_SSB_NO) &&
1109 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1110 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1111 
1112 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1113 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1114 
1115 	if (!cpu_matches(NO_MDS) && !(ia32_cap & ARCH_CAP_MDS_NO)) {
1116 		setup_force_cpu_bug(X86_BUG_MDS);
1117 		if (cpu_matches(MSBDS_ONLY))
1118 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1119 	}
1120 
1121 	if (!cpu_matches(NO_SWAPGS))
1122 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1123 
1124 	if (cpu_matches(NO_MELTDOWN))
1125 		return;
1126 
1127 	/* Rogue Data Cache Load? No! */
1128 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1129 		return;
1130 
1131 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1132 
1133 	if (cpu_matches(NO_L1TF))
1134 		return;
1135 
1136 	setup_force_cpu_bug(X86_BUG_L1TF);
1137 }
1138 
1139 /*
1140  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1141  * unfortunately, that's not true in practice because of early VIA
1142  * chips and (more importantly) broken virtualizers that are not easy
1143  * to detect. In the latter case it doesn't even *fail* reliably, so
1144  * probing for it doesn't even work. Disable it completely on 32-bit
1145  * unless we can find a reliable way to detect all the broken cases.
1146  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1147  */
1148 static void detect_nopl(void)
1149 {
1150 #ifdef CONFIG_X86_32
1151 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1152 #else
1153 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1154 #endif
1155 }
1156 
1157 /*
1158  * Do minimum CPU detection early.
1159  * Fields really needed: vendor, cpuid_level, family, model, mask,
1160  * cache alignment.
1161  * The others are not touched to avoid unwanted side effects.
1162  *
1163  * WARNING: this function is only called on the boot CPU.  Don't add code
1164  * here that is supposed to run on all CPUs.
1165  */
1166 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1167 {
1168 #ifdef CONFIG_X86_64
1169 	c->x86_clflush_size = 64;
1170 	c->x86_phys_bits = 36;
1171 	c->x86_virt_bits = 48;
1172 #else
1173 	c->x86_clflush_size = 32;
1174 	c->x86_phys_bits = 32;
1175 	c->x86_virt_bits = 32;
1176 #endif
1177 	c->x86_cache_alignment = c->x86_clflush_size;
1178 
1179 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1180 	c->extended_cpuid_level = 0;
1181 
1182 	if (!have_cpuid_p())
1183 		identify_cpu_without_cpuid(c);
1184 
1185 	/* cyrix could have cpuid enabled via c_identify()*/
1186 	if (have_cpuid_p()) {
1187 		cpu_detect(c);
1188 		get_cpu_vendor(c);
1189 		get_cpu_cap(c);
1190 		get_cpu_address_sizes(c);
1191 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1192 
1193 		if (this_cpu->c_early_init)
1194 			this_cpu->c_early_init(c);
1195 
1196 		c->cpu_index = 0;
1197 		filter_cpuid_features(c, false);
1198 
1199 		if (this_cpu->c_bsp_init)
1200 			this_cpu->c_bsp_init(c);
1201 	} else {
1202 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1203 	}
1204 
1205 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1206 
1207 	cpu_set_bug_bits(c);
1208 
1209 	fpu__init_system(c);
1210 
1211 #ifdef CONFIG_X86_32
1212 	/*
1213 	 * Regardless of whether PCID is enumerated, the SDM says
1214 	 * that it can't be enabled in 32-bit mode.
1215 	 */
1216 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1217 #endif
1218 
1219 	/*
1220 	 * Later in the boot process pgtable_l5_enabled() relies on
1221 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1222 	 * enabled by this point we need to clear the feature bit to avoid
1223 	 * false-positives at the later stage.
1224 	 *
1225 	 * pgtable_l5_enabled() can be false here for several reasons:
1226 	 *  - 5-level paging is disabled compile-time;
1227 	 *  - it's 32-bit kernel;
1228 	 *  - machine doesn't support 5-level paging;
1229 	 *  - user specified 'no5lvl' in kernel command line.
1230 	 */
1231 	if (!pgtable_l5_enabled())
1232 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1233 
1234 	detect_nopl();
1235 }
1236 
1237 void __init early_cpu_init(void)
1238 {
1239 	const struct cpu_dev *const *cdev;
1240 	int count = 0;
1241 
1242 #ifdef CONFIG_PROCESSOR_SELECT
1243 	pr_info("KERNEL supported cpus:\n");
1244 #endif
1245 
1246 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1247 		const struct cpu_dev *cpudev = *cdev;
1248 
1249 		if (count >= X86_VENDOR_NUM)
1250 			break;
1251 		cpu_devs[count] = cpudev;
1252 		count++;
1253 
1254 #ifdef CONFIG_PROCESSOR_SELECT
1255 		{
1256 			unsigned int j;
1257 
1258 			for (j = 0; j < 2; j++) {
1259 				if (!cpudev->c_ident[j])
1260 					continue;
1261 				pr_info("  %s %s\n", cpudev->c_vendor,
1262 					cpudev->c_ident[j]);
1263 			}
1264 		}
1265 #endif
1266 	}
1267 	early_identify_cpu(&boot_cpu_data);
1268 }
1269 
1270 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1271 {
1272 #ifdef CONFIG_X86_64
1273 	/*
1274 	 * Empirically, writing zero to a segment selector on AMD does
1275 	 * not clear the base, whereas writing zero to a segment
1276 	 * selector on Intel does clear the base.  Intel's behavior
1277 	 * allows slightly faster context switches in the common case
1278 	 * where GS is unused by the prev and next threads.
1279 	 *
1280 	 * Since neither vendor documents this anywhere that I can see,
1281 	 * detect it directly instead of hardcoding the choice by
1282 	 * vendor.
1283 	 *
1284 	 * I've designated AMD's behavior as the "bug" because it's
1285 	 * counterintuitive and less friendly.
1286 	 */
1287 
1288 	unsigned long old_base, tmp;
1289 	rdmsrl(MSR_FS_BASE, old_base);
1290 	wrmsrl(MSR_FS_BASE, 1);
1291 	loadsegment(fs, 0);
1292 	rdmsrl(MSR_FS_BASE, tmp);
1293 	if (tmp != 0)
1294 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1295 	wrmsrl(MSR_FS_BASE, old_base);
1296 #endif
1297 }
1298 
1299 static void generic_identify(struct cpuinfo_x86 *c)
1300 {
1301 	c->extended_cpuid_level = 0;
1302 
1303 	if (!have_cpuid_p())
1304 		identify_cpu_without_cpuid(c);
1305 
1306 	/* cyrix could have cpuid enabled via c_identify()*/
1307 	if (!have_cpuid_p())
1308 		return;
1309 
1310 	cpu_detect(c);
1311 
1312 	get_cpu_vendor(c);
1313 
1314 	get_cpu_cap(c);
1315 
1316 	get_cpu_address_sizes(c);
1317 
1318 	if (c->cpuid_level >= 0x00000001) {
1319 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1320 #ifdef CONFIG_X86_32
1321 # ifdef CONFIG_SMP
1322 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1323 # else
1324 		c->apicid = c->initial_apicid;
1325 # endif
1326 #endif
1327 		c->phys_proc_id = c->initial_apicid;
1328 	}
1329 
1330 	get_model_name(c); /* Default name */
1331 
1332 	detect_null_seg_behavior(c);
1333 
1334 	/*
1335 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1336 	 * systems that run Linux at CPL > 0 may or may not have the
1337 	 * issue, but, even if they have the issue, there's absolutely
1338 	 * nothing we can do about it because we can't use the real IRET
1339 	 * instruction.
1340 	 *
1341 	 * NB: For the time being, only 32-bit kernels support
1342 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1343 	 * whether to apply espfix using paravirt hooks.  If any
1344 	 * non-paravirt system ever shows up that does *not* have the
1345 	 * ESPFIX issue, we can change this.
1346 	 */
1347 #ifdef CONFIG_X86_32
1348 # ifdef CONFIG_PARAVIRT_XXL
1349 	do {
1350 		extern void native_iret(void);
1351 		if (pv_ops.cpu.iret == native_iret)
1352 			set_cpu_bug(c, X86_BUG_ESPFIX);
1353 	} while (0);
1354 # else
1355 	set_cpu_bug(c, X86_BUG_ESPFIX);
1356 # endif
1357 #endif
1358 }
1359 
1360 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
1361 {
1362 	/*
1363 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
1364 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
1365 	 * in case CQM bits really aren't there in this CPU.
1366 	 */
1367 	if (c != &boot_cpu_data) {
1368 		boot_cpu_data.x86_cache_max_rmid =
1369 			min(boot_cpu_data.x86_cache_max_rmid,
1370 			    c->x86_cache_max_rmid);
1371 	}
1372 }
1373 
1374 /*
1375  * Validate that ACPI/mptables have the same information about the
1376  * effective APIC id and update the package map.
1377  */
1378 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1379 {
1380 #ifdef CONFIG_SMP
1381 	unsigned int apicid, cpu = smp_processor_id();
1382 
1383 	apicid = apic->cpu_present_to_apicid(cpu);
1384 
1385 	if (apicid != c->apicid) {
1386 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1387 		       cpu, apicid, c->initial_apicid);
1388 	}
1389 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1390 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1391 #else
1392 	c->logical_proc_id = 0;
1393 #endif
1394 }
1395 
1396 /*
1397  * This does the hard work of actually picking apart the CPU stuff...
1398  */
1399 static void identify_cpu(struct cpuinfo_x86 *c)
1400 {
1401 	int i;
1402 
1403 	c->loops_per_jiffy = loops_per_jiffy;
1404 	c->x86_cache_size = 0;
1405 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1406 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1407 	c->x86_vendor_id[0] = '\0'; /* Unset */
1408 	c->x86_model_id[0] = '\0';  /* Unset */
1409 	c->x86_max_cores = 1;
1410 	c->x86_coreid_bits = 0;
1411 	c->cu_id = 0xff;
1412 #ifdef CONFIG_X86_64
1413 	c->x86_clflush_size = 64;
1414 	c->x86_phys_bits = 36;
1415 	c->x86_virt_bits = 48;
1416 #else
1417 	c->cpuid_level = -1;	/* CPUID not detected */
1418 	c->x86_clflush_size = 32;
1419 	c->x86_phys_bits = 32;
1420 	c->x86_virt_bits = 32;
1421 #endif
1422 	c->x86_cache_alignment = c->x86_clflush_size;
1423 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1424 
1425 	generic_identify(c);
1426 
1427 	if (this_cpu->c_identify)
1428 		this_cpu->c_identify(c);
1429 
1430 	/* Clear/Set all flags overridden by options, after probe */
1431 	apply_forced_caps(c);
1432 
1433 #ifdef CONFIG_X86_64
1434 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1435 #endif
1436 
1437 	/*
1438 	 * Vendor-specific initialization.  In this section we
1439 	 * canonicalize the feature flags, meaning if there are
1440 	 * features a certain CPU supports which CPUID doesn't
1441 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1442 	 * we handle them here.
1443 	 *
1444 	 * At the end of this section, c->x86_capability better
1445 	 * indicate the features this CPU genuinely supports!
1446 	 */
1447 	if (this_cpu->c_init)
1448 		this_cpu->c_init(c);
1449 
1450 	/* Disable the PN if appropriate */
1451 	squash_the_stupid_serial_number(c);
1452 
1453 	/* Set up SMEP/SMAP/UMIP */
1454 	setup_smep(c);
1455 	setup_smap(c);
1456 	setup_umip(c);
1457 
1458 	/*
1459 	 * The vendor-specific functions might have changed features.
1460 	 * Now we do "generic changes."
1461 	 */
1462 
1463 	/* Filter out anything that depends on CPUID levels we don't have */
1464 	filter_cpuid_features(c, true);
1465 
1466 	/* If the model name is still unset, do table lookup. */
1467 	if (!c->x86_model_id[0]) {
1468 		const char *p;
1469 		p = table_lookup_model(c);
1470 		if (p)
1471 			strcpy(c->x86_model_id, p);
1472 		else
1473 			/* Last resort... */
1474 			sprintf(c->x86_model_id, "%02x/%02x",
1475 				c->x86, c->x86_model);
1476 	}
1477 
1478 #ifdef CONFIG_X86_64
1479 	detect_ht(c);
1480 #endif
1481 
1482 	x86_init_rdrand(c);
1483 	x86_init_cache_qos(c);
1484 	setup_pku(c);
1485 
1486 	/*
1487 	 * Clear/Set all flags overridden by options, need do it
1488 	 * before following smp all cpus cap AND.
1489 	 */
1490 	apply_forced_caps(c);
1491 
1492 	/*
1493 	 * On SMP, boot_cpu_data holds the common feature set between
1494 	 * all CPUs; so make sure that we indicate which features are
1495 	 * common between the CPUs.  The first time this routine gets
1496 	 * executed, c == &boot_cpu_data.
1497 	 */
1498 	if (c != &boot_cpu_data) {
1499 		/* AND the already accumulated flags with these */
1500 		for (i = 0; i < NCAPINTS; i++)
1501 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1502 
1503 		/* OR, i.e. replicate the bug flags */
1504 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1505 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1506 	}
1507 
1508 	/* Init Machine Check Exception if available. */
1509 	mcheck_cpu_init(c);
1510 
1511 	select_idle_routine(c);
1512 
1513 #ifdef CONFIG_NUMA
1514 	numa_add_cpu(smp_processor_id());
1515 #endif
1516 }
1517 
1518 /*
1519  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1520  * on 32-bit kernels:
1521  */
1522 #ifdef CONFIG_X86_32
1523 void enable_sep_cpu(void)
1524 {
1525 	struct tss_struct *tss;
1526 	int cpu;
1527 
1528 	if (!boot_cpu_has(X86_FEATURE_SEP))
1529 		return;
1530 
1531 	cpu = get_cpu();
1532 	tss = &per_cpu(cpu_tss_rw, cpu);
1533 
1534 	/*
1535 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1536 	 * see the big comment in struct x86_hw_tss's definition.
1537 	 */
1538 
1539 	tss->x86_tss.ss1 = __KERNEL_CS;
1540 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1541 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1542 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1543 
1544 	put_cpu();
1545 }
1546 #endif
1547 
1548 void __init identify_boot_cpu(void)
1549 {
1550 	identify_cpu(&boot_cpu_data);
1551 #ifdef CONFIG_X86_32
1552 	sysenter_setup();
1553 	enable_sep_cpu();
1554 #endif
1555 	cpu_detect_tlb(&boot_cpu_data);
1556 	setup_cr_pinning();
1557 }
1558 
1559 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1560 {
1561 	BUG_ON(c == &boot_cpu_data);
1562 	identify_cpu(c);
1563 #ifdef CONFIG_X86_32
1564 	enable_sep_cpu();
1565 #endif
1566 	mtrr_ap_init();
1567 	validate_apic_and_package_id(c);
1568 	x86_spec_ctrl_setup_ap();
1569 }
1570 
1571 static __init int setup_noclflush(char *arg)
1572 {
1573 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1574 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1575 	return 1;
1576 }
1577 __setup("noclflush", setup_noclflush);
1578 
1579 void print_cpu_info(struct cpuinfo_x86 *c)
1580 {
1581 	const char *vendor = NULL;
1582 
1583 	if (c->x86_vendor < X86_VENDOR_NUM) {
1584 		vendor = this_cpu->c_vendor;
1585 	} else {
1586 		if (c->cpuid_level >= 0)
1587 			vendor = c->x86_vendor_id;
1588 	}
1589 
1590 	if (vendor && !strstr(c->x86_model_id, vendor))
1591 		pr_cont("%s ", vendor);
1592 
1593 	if (c->x86_model_id[0])
1594 		pr_cont("%s", c->x86_model_id);
1595 	else
1596 		pr_cont("%d86", c->x86);
1597 
1598 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1599 
1600 	if (c->x86_stepping || c->cpuid_level >= 0)
1601 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1602 	else
1603 		pr_cont(")\n");
1604 }
1605 
1606 /*
1607  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1608  * But we need to keep a dummy __setup around otherwise it would
1609  * show up as an environment variable for init.
1610  */
1611 static __init int setup_clearcpuid(char *arg)
1612 {
1613 	return 1;
1614 }
1615 __setup("clearcpuid=", setup_clearcpuid);
1616 
1617 #ifdef CONFIG_X86_64
1618 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1619 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1620 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1621 
1622 /*
1623  * The following percpu variables are hot.  Align current_task to
1624  * cacheline size such that they fall in the same cacheline.
1625  */
1626 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1627 	&init_task;
1628 EXPORT_PER_CPU_SYMBOL(current_task);
1629 
1630 DEFINE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
1631 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1632 
1633 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1634 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1635 
1636 /* May not be marked __init: used by software suspend */
1637 void syscall_init(void)
1638 {
1639 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1640 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1641 
1642 #ifdef CONFIG_IA32_EMULATION
1643 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1644 	/*
1645 	 * This only works on Intel CPUs.
1646 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1647 	 * This does not cause SYSENTER to jump to the wrong location, because
1648 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1649 	 */
1650 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1651 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1652 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1653 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1654 #else
1655 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1656 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1657 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1658 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1659 #endif
1660 
1661 	/* Flags to clear on syscall */
1662 	wrmsrl(MSR_SYSCALL_MASK,
1663 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1664 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1665 }
1666 
1667 DEFINE_PER_CPU(int, debug_stack_usage);
1668 DEFINE_PER_CPU(u32, debug_idt_ctr);
1669 
1670 void debug_stack_set_zero(void)
1671 {
1672 	this_cpu_inc(debug_idt_ctr);
1673 	load_current_idt();
1674 }
1675 NOKPROBE_SYMBOL(debug_stack_set_zero);
1676 
1677 void debug_stack_reset(void)
1678 {
1679 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1680 		return;
1681 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1682 		load_current_idt();
1683 }
1684 NOKPROBE_SYMBOL(debug_stack_reset);
1685 
1686 #else	/* CONFIG_X86_64 */
1687 
1688 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1689 EXPORT_PER_CPU_SYMBOL(current_task);
1690 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1691 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1692 
1693 /*
1694  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1695  * the top of the kernel stack.  Use an extra percpu variable to track the
1696  * top of the kernel stack directly.
1697  */
1698 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1699 	(unsigned long)&init_thread_union + THREAD_SIZE;
1700 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1701 
1702 #ifdef CONFIG_STACKPROTECTOR
1703 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1704 #endif
1705 
1706 #endif	/* CONFIG_X86_64 */
1707 
1708 /*
1709  * Clear all 6 debug registers:
1710  */
1711 static void clear_all_debug_regs(void)
1712 {
1713 	int i;
1714 
1715 	for (i = 0; i < 8; i++) {
1716 		/* Ignore db4, db5 */
1717 		if ((i == 4) || (i == 5))
1718 			continue;
1719 
1720 		set_debugreg(0, i);
1721 	}
1722 }
1723 
1724 #ifdef CONFIG_KGDB
1725 /*
1726  * Restore debug regs if using kgdbwait and you have a kernel debugger
1727  * connection established.
1728  */
1729 static void dbg_restore_debug_regs(void)
1730 {
1731 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1732 		arch_kgdb_ops.correct_hw_break();
1733 }
1734 #else /* ! CONFIG_KGDB */
1735 #define dbg_restore_debug_regs()
1736 #endif /* ! CONFIG_KGDB */
1737 
1738 static void wait_for_master_cpu(int cpu)
1739 {
1740 #ifdef CONFIG_SMP
1741 	/*
1742 	 * wait for ACK from master CPU before continuing
1743 	 * with AP initialization
1744 	 */
1745 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1746 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1747 		cpu_relax();
1748 #endif
1749 }
1750 
1751 #ifdef CONFIG_X86_64
1752 static void setup_getcpu(int cpu)
1753 {
1754 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1755 	struct desc_struct d = { };
1756 
1757 	if (boot_cpu_has(X86_FEATURE_RDTSCP))
1758 		write_rdtscp_aux(cpudata);
1759 
1760 	/* Store CPU and node number in limit. */
1761 	d.limit0 = cpudata;
1762 	d.limit1 = cpudata >> 16;
1763 
1764 	d.type = 5;		/* RO data, expand down, accessed */
1765 	d.dpl = 3;		/* Visible to user code */
1766 	d.s = 1;		/* Not a system segment */
1767 	d.p = 1;		/* Present */
1768 	d.d = 1;		/* 32-bit */
1769 
1770 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1771 }
1772 #endif
1773 
1774 /*
1775  * cpu_init() initializes state that is per-CPU. Some data is already
1776  * initialized (naturally) in the bootstrap process, such as the GDT
1777  * and IDT. We reload them nevertheless, this function acts as a
1778  * 'CPU state barrier', nothing should get across.
1779  */
1780 #ifdef CONFIG_X86_64
1781 
1782 void cpu_init(void)
1783 {
1784 	int cpu = raw_smp_processor_id();
1785 	struct task_struct *me;
1786 	struct tss_struct *t;
1787 	int i;
1788 
1789 	wait_for_master_cpu(cpu);
1790 
1791 	if (cpu)
1792 		load_ucode_ap();
1793 
1794 	t = &per_cpu(cpu_tss_rw, cpu);
1795 
1796 #ifdef CONFIG_NUMA
1797 	if (this_cpu_read(numa_node) == 0 &&
1798 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1799 		set_numa_node(early_cpu_to_node(cpu));
1800 #endif
1801 	setup_getcpu(cpu);
1802 
1803 	me = current;
1804 
1805 	pr_debug("Initializing CPU#%d\n", cpu);
1806 
1807 	cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1808 
1809 	/*
1810 	 * Initialize the per-CPU GDT with the boot GDT,
1811 	 * and set up the GDT descriptor:
1812 	 */
1813 
1814 	switch_to_new_gdt(cpu);
1815 	loadsegment(fs, 0);
1816 
1817 	load_current_idt();
1818 
1819 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1820 	syscall_init();
1821 
1822 	wrmsrl(MSR_FS_BASE, 0);
1823 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1824 	barrier();
1825 
1826 	x86_configure_nx();
1827 	x2apic_setup();
1828 
1829 	/*
1830 	 * set up and load the per-CPU TSS
1831 	 */
1832 	if (!t->x86_tss.ist[0]) {
1833 		t->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1834 		t->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1835 		t->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1836 		t->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1837 	}
1838 
1839 	t->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
1840 
1841 	/*
1842 	 * <= is required because the CPU will access up to
1843 	 * 8 bits beyond the end of the IO permission bitmap.
1844 	 */
1845 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1846 		t->io_bitmap[i] = ~0UL;
1847 
1848 	mmgrab(&init_mm);
1849 	me->active_mm = &init_mm;
1850 	BUG_ON(me->mm);
1851 	initialize_tlbstate_and_flush();
1852 	enter_lazy_tlb(&init_mm, me);
1853 
1854 	/*
1855 	 * Initialize the TSS.  sp0 points to the entry trampoline stack
1856 	 * regardless of what task is running.
1857 	 */
1858 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1859 	load_TR_desc();
1860 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1861 
1862 	load_mm_ldt(&init_mm);
1863 
1864 	clear_all_debug_regs();
1865 	dbg_restore_debug_regs();
1866 
1867 	fpu__init_cpu();
1868 
1869 	if (is_uv_system())
1870 		uv_cpu_init();
1871 
1872 	load_fixmap_gdt(cpu);
1873 }
1874 
1875 #else
1876 
1877 void cpu_init(void)
1878 {
1879 	int cpu = smp_processor_id();
1880 	struct task_struct *curr = current;
1881 	struct tss_struct *t = &per_cpu(cpu_tss_rw, cpu);
1882 
1883 	wait_for_master_cpu(cpu);
1884 
1885 	show_ucode_info_early();
1886 
1887 	pr_info("Initializing CPU#%d\n", cpu);
1888 
1889 	if (cpu_feature_enabled(X86_FEATURE_VME) ||
1890 	    boot_cpu_has(X86_FEATURE_TSC) ||
1891 	    boot_cpu_has(X86_FEATURE_DE))
1892 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1893 
1894 	load_current_idt();
1895 	switch_to_new_gdt(cpu);
1896 
1897 	/*
1898 	 * Set up and load the per-CPU TSS and LDT
1899 	 */
1900 	mmgrab(&init_mm);
1901 	curr->active_mm = &init_mm;
1902 	BUG_ON(curr->mm);
1903 	initialize_tlbstate_and_flush();
1904 	enter_lazy_tlb(&init_mm, curr);
1905 
1906 	/*
1907 	 * Initialize the TSS.  sp0 points to the entry trampoline stack
1908 	 * regardless of what task is running.
1909 	 */
1910 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1911 	load_TR_desc();
1912 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1913 
1914 	load_mm_ldt(&init_mm);
1915 
1916 	t->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
1917 
1918 #ifdef CONFIG_DOUBLEFAULT
1919 	/* Set up doublefault TSS pointer in the GDT */
1920 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1921 #endif
1922 
1923 	clear_all_debug_regs();
1924 	dbg_restore_debug_regs();
1925 
1926 	fpu__init_cpu();
1927 
1928 	load_fixmap_gdt(cpu);
1929 }
1930 #endif
1931 
1932 /*
1933  * The microcode loader calls this upon late microcode load to recheck features,
1934  * only when microcode has been updated. Caller holds microcode_mutex and CPU
1935  * hotplug lock.
1936  */
1937 void microcode_check(void)
1938 {
1939 	struct cpuinfo_x86 info;
1940 
1941 	perf_check_microcode();
1942 
1943 	/* Reload CPUID max function as it might've changed. */
1944 	info.cpuid_level = cpuid_eax(0);
1945 
1946 	/*
1947 	 * Copy all capability leafs to pick up the synthetic ones so that
1948 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
1949 	 * get overwritten in get_cpu_cap().
1950 	 */
1951 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
1952 
1953 	get_cpu_cap(&info);
1954 
1955 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
1956 		return;
1957 
1958 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
1959 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
1960 }
1961 
1962 /*
1963  * Invoked from core CPU hotplug code after hotplug operations
1964  */
1965 void arch_smt_update(void)
1966 {
1967 	/* Handle the speculative execution misfeatures */
1968 	cpu_bugs_smt_update();
1969 	/* Check whether IPI broadcasting can be enabled */
1970 	apic_smt_update();
1971 }
1972