xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 29c37341)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* cpu_feature_enabled() cannot be used this early */
3 #define USE_EARLY_PGTABLE_L5
4 
5 #include <linux/memblock.h>
6 #include <linux/linkage.h>
7 #include <linux/bitops.h>
8 #include <linux/kernel.h>
9 #include <linux/export.h>
10 #include <linux/percpu.h>
11 #include <linux/string.h>
12 #include <linux/ctype.h>
13 #include <linux/delay.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/clock.h>
16 #include <linux/sched/task.h>
17 #include <linux/sched/smt.h>
18 #include <linux/init.h>
19 #include <linux/kprobes.h>
20 #include <linux/kgdb.h>
21 #include <linux/smp.h>
22 #include <linux/io.h>
23 #include <linux/syscore_ops.h>
24 #include <linux/pgtable.h>
25 
26 #include <asm/stackprotector.h>
27 #include <asm/perf_event.h>
28 #include <asm/mmu_context.h>
29 #include <asm/doublefault.h>
30 #include <asm/archrandom.h>
31 #include <asm/hypervisor.h>
32 #include <asm/processor.h>
33 #include <asm/tlbflush.h>
34 #include <asm/debugreg.h>
35 #include <asm/sections.h>
36 #include <asm/vsyscall.h>
37 #include <linux/topology.h>
38 #include <linux/cpumask.h>
39 #include <linux/atomic.h>
40 #include <asm/proto.h>
41 #include <asm/setup.h>
42 #include <asm/apic.h>
43 #include <asm/desc.h>
44 #include <asm/fpu/internal.h>
45 #include <asm/mtrr.h>
46 #include <asm/hwcap2.h>
47 #include <linux/numa.h>
48 #include <asm/numa.h>
49 #include <asm/asm.h>
50 #include <asm/bugs.h>
51 #include <asm/cpu.h>
52 #include <asm/mce.h>
53 #include <asm/msr.h>
54 #include <asm/memtype.h>
55 #include <asm/microcode.h>
56 #include <asm/microcode_intel.h>
57 #include <asm/intel-family.h>
58 #include <asm/cpu_device_id.h>
59 #include <asm/uv/uv.h>
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 #ifdef CONFIG_X86_64
169 static int __init x86_nopcid_setup(char *s)
170 {
171 	/* nopcid doesn't accept parameters */
172 	if (s)
173 		return -EINVAL;
174 
175 	/* do not emit a message if the feature is not present */
176 	if (!boot_cpu_has(X86_FEATURE_PCID))
177 		return 0;
178 
179 	setup_clear_cpu_cap(X86_FEATURE_PCID);
180 	pr_info("nopcid: PCID feature disabled\n");
181 	return 0;
182 }
183 early_param("nopcid", x86_nopcid_setup);
184 #endif
185 
186 static int __init x86_noinvpcid_setup(char *s)
187 {
188 	/* noinvpcid doesn't accept parameters */
189 	if (s)
190 		return -EINVAL;
191 
192 	/* do not emit a message if the feature is not present */
193 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
194 		return 0;
195 
196 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
197 	pr_info("noinvpcid: INVPCID feature disabled\n");
198 	return 0;
199 }
200 early_param("noinvpcid", x86_noinvpcid_setup);
201 
202 #ifdef CONFIG_X86_32
203 static int cachesize_override = -1;
204 static int disable_x86_serial_nr = 1;
205 
206 static int __init cachesize_setup(char *str)
207 {
208 	get_option(&str, &cachesize_override);
209 	return 1;
210 }
211 __setup("cachesize=", cachesize_setup);
212 
213 static int __init x86_sep_setup(char *s)
214 {
215 	setup_clear_cpu_cap(X86_FEATURE_SEP);
216 	return 1;
217 }
218 __setup("nosep", x86_sep_setup);
219 
220 /* Standard macro to see if a specific flag is changeable */
221 static inline int flag_is_changeable_p(u32 flag)
222 {
223 	u32 f1, f2;
224 
225 	/*
226 	 * Cyrix and IDT cpus allow disabling of CPUID
227 	 * so the code below may return different results
228 	 * when it is executed before and after enabling
229 	 * the CPUID. Add "volatile" to not allow gcc to
230 	 * optimize the subsequent calls to this function.
231 	 */
232 	asm volatile ("pushfl		\n\t"
233 		      "pushfl		\n\t"
234 		      "popl %0		\n\t"
235 		      "movl %0, %1	\n\t"
236 		      "xorl %2, %0	\n\t"
237 		      "pushl %0		\n\t"
238 		      "popfl		\n\t"
239 		      "pushfl		\n\t"
240 		      "popl %0		\n\t"
241 		      "popfl		\n\t"
242 
243 		      : "=&r" (f1), "=&r" (f2)
244 		      : "ir" (flag));
245 
246 	return ((f1^f2) & flag) != 0;
247 }
248 
249 /* Probe for the CPUID instruction */
250 int have_cpuid_p(void)
251 {
252 	return flag_is_changeable_p(X86_EFLAGS_ID);
253 }
254 
255 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
256 {
257 	unsigned long lo, hi;
258 
259 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
260 		return;
261 
262 	/* Disable processor serial number: */
263 
264 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
265 	lo |= 0x200000;
266 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
267 
268 	pr_notice("CPU serial number disabled.\n");
269 	clear_cpu_cap(c, X86_FEATURE_PN);
270 
271 	/* Disabling the serial number may affect the cpuid level */
272 	c->cpuid_level = cpuid_eax(0);
273 }
274 
275 static int __init x86_serial_nr_setup(char *s)
276 {
277 	disable_x86_serial_nr = 0;
278 	return 1;
279 }
280 __setup("serialnumber", x86_serial_nr_setup);
281 #else
282 static inline int flag_is_changeable_p(u32 flag)
283 {
284 	return 1;
285 }
286 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
287 {
288 }
289 #endif
290 
291 static __init int setup_disable_smep(char *arg)
292 {
293 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
294 	return 1;
295 }
296 __setup("nosmep", setup_disable_smep);
297 
298 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
299 {
300 	if (cpu_has(c, X86_FEATURE_SMEP))
301 		cr4_set_bits(X86_CR4_SMEP);
302 }
303 
304 static __init int setup_disable_smap(char *arg)
305 {
306 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
307 	return 1;
308 }
309 __setup("nosmap", setup_disable_smap);
310 
311 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
312 {
313 	unsigned long eflags = native_save_fl();
314 
315 	/* This should have been cleared long ago */
316 	BUG_ON(eflags & X86_EFLAGS_AC);
317 
318 	if (cpu_has(c, X86_FEATURE_SMAP)) {
319 #ifdef CONFIG_X86_SMAP
320 		cr4_set_bits(X86_CR4_SMAP);
321 #else
322 		cr4_clear_bits(X86_CR4_SMAP);
323 #endif
324 	}
325 }
326 
327 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
328 {
329 	/* Check the boot processor, plus build option for UMIP. */
330 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
331 		goto out;
332 
333 	/* Check the current processor's cpuid bits. */
334 	if (!cpu_has(c, X86_FEATURE_UMIP))
335 		goto out;
336 
337 	cr4_set_bits(X86_CR4_UMIP);
338 
339 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
340 
341 	return;
342 
343 out:
344 	/*
345 	 * Make sure UMIP is disabled in case it was enabled in a
346 	 * previous boot (e.g., via kexec).
347 	 */
348 	cr4_clear_bits(X86_CR4_UMIP);
349 }
350 
351 /* These bits should not change their value after CPU init is finished. */
352 static const unsigned long cr4_pinned_mask =
353 	X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP | X86_CR4_FSGSBASE;
354 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
355 static unsigned long cr4_pinned_bits __ro_after_init;
356 
357 void native_write_cr0(unsigned long val)
358 {
359 	unsigned long bits_missing = 0;
360 
361 set_register:
362 	asm volatile("mov %0,%%cr0": "+r" (val), "+m" (__force_order));
363 
364 	if (static_branch_likely(&cr_pinning)) {
365 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
366 			bits_missing = X86_CR0_WP;
367 			val |= bits_missing;
368 			goto set_register;
369 		}
370 		/* Warn after we've set the missing bits. */
371 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
372 	}
373 }
374 EXPORT_SYMBOL(native_write_cr0);
375 
376 void native_write_cr4(unsigned long val)
377 {
378 	unsigned long bits_changed = 0;
379 
380 set_register:
381 	asm volatile("mov %0,%%cr4": "+r" (val), "+m" (cr4_pinned_bits));
382 
383 	if (static_branch_likely(&cr_pinning)) {
384 		if (unlikely((val & cr4_pinned_mask) != cr4_pinned_bits)) {
385 			bits_changed = (val & cr4_pinned_mask) ^ cr4_pinned_bits;
386 			val = (val & ~cr4_pinned_mask) | cr4_pinned_bits;
387 			goto set_register;
388 		}
389 		/* Warn after we've corrected the changed bits. */
390 		WARN_ONCE(bits_changed, "pinned CR4 bits changed: 0x%lx!?\n",
391 			  bits_changed);
392 	}
393 }
394 #if IS_MODULE(CONFIG_LKDTM)
395 EXPORT_SYMBOL_GPL(native_write_cr4);
396 #endif
397 
398 void cr4_update_irqsoff(unsigned long set, unsigned long clear)
399 {
400 	unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
401 
402 	lockdep_assert_irqs_disabled();
403 
404 	newval = (cr4 & ~clear) | set;
405 	if (newval != cr4) {
406 		this_cpu_write(cpu_tlbstate.cr4, newval);
407 		__write_cr4(newval);
408 	}
409 }
410 EXPORT_SYMBOL(cr4_update_irqsoff);
411 
412 /* Read the CR4 shadow. */
413 unsigned long cr4_read_shadow(void)
414 {
415 	return this_cpu_read(cpu_tlbstate.cr4);
416 }
417 EXPORT_SYMBOL_GPL(cr4_read_shadow);
418 
419 void cr4_init(void)
420 {
421 	unsigned long cr4 = __read_cr4();
422 
423 	if (boot_cpu_has(X86_FEATURE_PCID))
424 		cr4 |= X86_CR4_PCIDE;
425 	if (static_branch_likely(&cr_pinning))
426 		cr4 = (cr4 & ~cr4_pinned_mask) | cr4_pinned_bits;
427 
428 	__write_cr4(cr4);
429 
430 	/* Initialize cr4 shadow for this CPU. */
431 	this_cpu_write(cpu_tlbstate.cr4, cr4);
432 }
433 
434 /*
435  * Once CPU feature detection is finished (and boot params have been
436  * parsed), record any of the sensitive CR bits that are set, and
437  * enable CR pinning.
438  */
439 static void __init setup_cr_pinning(void)
440 {
441 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & cr4_pinned_mask;
442 	static_key_enable(&cr_pinning.key);
443 }
444 
445 static __init int x86_nofsgsbase_setup(char *arg)
446 {
447 	/* Require an exact match without trailing characters. */
448 	if (strlen(arg))
449 		return 0;
450 
451 	/* Do not emit a message if the feature is not present. */
452 	if (!boot_cpu_has(X86_FEATURE_FSGSBASE))
453 		return 1;
454 
455 	setup_clear_cpu_cap(X86_FEATURE_FSGSBASE);
456 	pr_info("FSGSBASE disabled via kernel command line\n");
457 	return 1;
458 }
459 __setup("nofsgsbase", x86_nofsgsbase_setup);
460 
461 /*
462  * Protection Keys are not available in 32-bit mode.
463  */
464 static bool pku_disabled;
465 
466 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
467 {
468 	struct pkru_state *pk;
469 
470 	/* check the boot processor, plus compile options for PKU: */
471 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
472 		return;
473 	/* checks the actual processor's cpuid bits: */
474 	if (!cpu_has(c, X86_FEATURE_PKU))
475 		return;
476 	if (pku_disabled)
477 		return;
478 
479 	cr4_set_bits(X86_CR4_PKE);
480 	pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU);
481 	if (pk)
482 		pk->pkru = init_pkru_value;
483 	/*
484 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
485 	 * cpuid bit to be set.  We need to ensure that we
486 	 * update that bit in this CPU's "cpu_info".
487 	 */
488 	set_cpu_cap(c, X86_FEATURE_OSPKE);
489 }
490 
491 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
492 static __init int setup_disable_pku(char *arg)
493 {
494 	/*
495 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
496 	 * runtime checks are against OSPKE so clearing the
497 	 * bit does nothing.
498 	 *
499 	 * This way, we will see "pku" in cpuinfo, but not
500 	 * "ospke", which is exactly what we want.  It shows
501 	 * that the CPU has PKU, but the OS has not enabled it.
502 	 * This happens to be exactly how a system would look
503 	 * if we disabled the config option.
504 	 */
505 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
506 	pku_disabled = true;
507 	return 1;
508 }
509 __setup("nopku", setup_disable_pku);
510 #endif /* CONFIG_X86_64 */
511 
512 /*
513  * Some CPU features depend on higher CPUID levels, which may not always
514  * be available due to CPUID level capping or broken virtualization
515  * software.  Add those features to this table to auto-disable them.
516  */
517 struct cpuid_dependent_feature {
518 	u32 feature;
519 	u32 level;
520 };
521 
522 static const struct cpuid_dependent_feature
523 cpuid_dependent_features[] = {
524 	{ X86_FEATURE_MWAIT,		0x00000005 },
525 	{ X86_FEATURE_DCA,		0x00000009 },
526 	{ X86_FEATURE_XSAVE,		0x0000000d },
527 	{ 0, 0 }
528 };
529 
530 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
531 {
532 	const struct cpuid_dependent_feature *df;
533 
534 	for (df = cpuid_dependent_features; df->feature; df++) {
535 
536 		if (!cpu_has(c, df->feature))
537 			continue;
538 		/*
539 		 * Note: cpuid_level is set to -1 if unavailable, but
540 		 * extended_extended_level is set to 0 if unavailable
541 		 * and the legitimate extended levels are all negative
542 		 * when signed; hence the weird messing around with
543 		 * signs here...
544 		 */
545 		if (!((s32)df->level < 0 ?
546 		     (u32)df->level > (u32)c->extended_cpuid_level :
547 		     (s32)df->level > (s32)c->cpuid_level))
548 			continue;
549 
550 		clear_cpu_cap(c, df->feature);
551 		if (!warn)
552 			continue;
553 
554 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
555 			x86_cap_flag(df->feature), df->level);
556 	}
557 }
558 
559 /*
560  * Naming convention should be: <Name> [(<Codename>)]
561  * This table only is used unless init_<vendor>() below doesn't set it;
562  * in particular, if CPUID levels 0x80000002..4 are supported, this
563  * isn't used
564  */
565 
566 /* Look up CPU names by table lookup. */
567 static const char *table_lookup_model(struct cpuinfo_x86 *c)
568 {
569 #ifdef CONFIG_X86_32
570 	const struct legacy_cpu_model_info *info;
571 
572 	if (c->x86_model >= 16)
573 		return NULL;	/* Range check */
574 
575 	if (!this_cpu)
576 		return NULL;
577 
578 	info = this_cpu->legacy_models;
579 
580 	while (info->family) {
581 		if (info->family == c->x86)
582 			return info->model_names[c->x86_model];
583 		info++;
584 	}
585 #endif
586 	return NULL;		/* Not found */
587 }
588 
589 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
590 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
591 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
592 
593 void load_percpu_segment(int cpu)
594 {
595 #ifdef CONFIG_X86_32
596 	loadsegment(fs, __KERNEL_PERCPU);
597 #else
598 	__loadsegment_simple(gs, 0);
599 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
600 #endif
601 	load_stack_canary_segment();
602 }
603 
604 #ifdef CONFIG_X86_32
605 /* The 32-bit entry code needs to find cpu_entry_area. */
606 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
607 #endif
608 
609 /* Load the original GDT from the per-cpu structure */
610 void load_direct_gdt(int cpu)
611 {
612 	struct desc_ptr gdt_descr;
613 
614 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
615 	gdt_descr.size = GDT_SIZE - 1;
616 	load_gdt(&gdt_descr);
617 }
618 EXPORT_SYMBOL_GPL(load_direct_gdt);
619 
620 /* Load a fixmap remapping of the per-cpu GDT */
621 void load_fixmap_gdt(int cpu)
622 {
623 	struct desc_ptr gdt_descr;
624 
625 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
626 	gdt_descr.size = GDT_SIZE - 1;
627 	load_gdt(&gdt_descr);
628 }
629 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
630 
631 /*
632  * Current gdt points %fs at the "master" per-cpu area: after this,
633  * it's on the real one.
634  */
635 void switch_to_new_gdt(int cpu)
636 {
637 	/* Load the original GDT */
638 	load_direct_gdt(cpu);
639 	/* Reload the per-cpu base */
640 	load_percpu_segment(cpu);
641 }
642 
643 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
644 
645 static void get_model_name(struct cpuinfo_x86 *c)
646 {
647 	unsigned int *v;
648 	char *p, *q, *s;
649 
650 	if (c->extended_cpuid_level < 0x80000004)
651 		return;
652 
653 	v = (unsigned int *)c->x86_model_id;
654 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
655 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
656 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
657 	c->x86_model_id[48] = 0;
658 
659 	/* Trim whitespace */
660 	p = q = s = &c->x86_model_id[0];
661 
662 	while (*p == ' ')
663 		p++;
664 
665 	while (*p) {
666 		/* Note the last non-whitespace index */
667 		if (!isspace(*p))
668 			s = q;
669 
670 		*q++ = *p++;
671 	}
672 
673 	*(s + 1) = '\0';
674 }
675 
676 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
677 {
678 	unsigned int eax, ebx, ecx, edx;
679 
680 	c->x86_max_cores = 1;
681 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
682 		return;
683 
684 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
685 	if (eax & 0x1f)
686 		c->x86_max_cores = (eax >> 26) + 1;
687 }
688 
689 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
690 {
691 	unsigned int n, dummy, ebx, ecx, edx, l2size;
692 
693 	n = c->extended_cpuid_level;
694 
695 	if (n >= 0x80000005) {
696 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
697 		c->x86_cache_size = (ecx>>24) + (edx>>24);
698 #ifdef CONFIG_X86_64
699 		/* On K8 L1 TLB is inclusive, so don't count it */
700 		c->x86_tlbsize = 0;
701 #endif
702 	}
703 
704 	if (n < 0x80000006)	/* Some chips just has a large L1. */
705 		return;
706 
707 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
708 	l2size = ecx >> 16;
709 
710 #ifdef CONFIG_X86_64
711 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
712 #else
713 	/* do processor-specific cache resizing */
714 	if (this_cpu->legacy_cache_size)
715 		l2size = this_cpu->legacy_cache_size(c, l2size);
716 
717 	/* Allow user to override all this if necessary. */
718 	if (cachesize_override != -1)
719 		l2size = cachesize_override;
720 
721 	if (l2size == 0)
722 		return;		/* Again, no L2 cache is possible */
723 #endif
724 
725 	c->x86_cache_size = l2size;
726 }
727 
728 u16 __read_mostly tlb_lli_4k[NR_INFO];
729 u16 __read_mostly tlb_lli_2m[NR_INFO];
730 u16 __read_mostly tlb_lli_4m[NR_INFO];
731 u16 __read_mostly tlb_lld_4k[NR_INFO];
732 u16 __read_mostly tlb_lld_2m[NR_INFO];
733 u16 __read_mostly tlb_lld_4m[NR_INFO];
734 u16 __read_mostly tlb_lld_1g[NR_INFO];
735 
736 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
737 {
738 	if (this_cpu->c_detect_tlb)
739 		this_cpu->c_detect_tlb(c);
740 
741 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
742 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
743 		tlb_lli_4m[ENTRIES]);
744 
745 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
746 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
747 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
748 }
749 
750 int detect_ht_early(struct cpuinfo_x86 *c)
751 {
752 #ifdef CONFIG_SMP
753 	u32 eax, ebx, ecx, edx;
754 
755 	if (!cpu_has(c, X86_FEATURE_HT))
756 		return -1;
757 
758 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
759 		return -1;
760 
761 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
762 		return -1;
763 
764 	cpuid(1, &eax, &ebx, &ecx, &edx);
765 
766 	smp_num_siblings = (ebx & 0xff0000) >> 16;
767 	if (smp_num_siblings == 1)
768 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
769 #endif
770 	return 0;
771 }
772 
773 void detect_ht(struct cpuinfo_x86 *c)
774 {
775 #ifdef CONFIG_SMP
776 	int index_msb, core_bits;
777 
778 	if (detect_ht_early(c) < 0)
779 		return;
780 
781 	index_msb = get_count_order(smp_num_siblings);
782 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
783 
784 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
785 
786 	index_msb = get_count_order(smp_num_siblings);
787 
788 	core_bits = get_count_order(c->x86_max_cores);
789 
790 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
791 				       ((1 << core_bits) - 1);
792 #endif
793 }
794 
795 static void get_cpu_vendor(struct cpuinfo_x86 *c)
796 {
797 	char *v = c->x86_vendor_id;
798 	int i;
799 
800 	for (i = 0; i < X86_VENDOR_NUM; i++) {
801 		if (!cpu_devs[i])
802 			break;
803 
804 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
805 		    (cpu_devs[i]->c_ident[1] &&
806 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
807 
808 			this_cpu = cpu_devs[i];
809 			c->x86_vendor = this_cpu->c_x86_vendor;
810 			return;
811 		}
812 	}
813 
814 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
815 		    "CPU: Your system may be unstable.\n", v);
816 
817 	c->x86_vendor = X86_VENDOR_UNKNOWN;
818 	this_cpu = &default_cpu;
819 }
820 
821 void cpu_detect(struct cpuinfo_x86 *c)
822 {
823 	/* Get vendor name */
824 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
825 	      (unsigned int *)&c->x86_vendor_id[0],
826 	      (unsigned int *)&c->x86_vendor_id[8],
827 	      (unsigned int *)&c->x86_vendor_id[4]);
828 
829 	c->x86 = 4;
830 	/* Intel-defined flags: level 0x00000001 */
831 	if (c->cpuid_level >= 0x00000001) {
832 		u32 junk, tfms, cap0, misc;
833 
834 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
835 		c->x86		= x86_family(tfms);
836 		c->x86_model	= x86_model(tfms);
837 		c->x86_stepping	= x86_stepping(tfms);
838 
839 		if (cap0 & (1<<19)) {
840 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
841 			c->x86_cache_alignment = c->x86_clflush_size;
842 		}
843 	}
844 }
845 
846 static void apply_forced_caps(struct cpuinfo_x86 *c)
847 {
848 	int i;
849 
850 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
851 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
852 		c->x86_capability[i] |= cpu_caps_set[i];
853 	}
854 }
855 
856 static void init_speculation_control(struct cpuinfo_x86 *c)
857 {
858 	/*
859 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
860 	 * and they also have a different bit for STIBP support. Also,
861 	 * a hypervisor might have set the individual AMD bits even on
862 	 * Intel CPUs, for finer-grained selection of what's available.
863 	 */
864 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
865 		set_cpu_cap(c, X86_FEATURE_IBRS);
866 		set_cpu_cap(c, X86_FEATURE_IBPB);
867 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
868 	}
869 
870 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
871 		set_cpu_cap(c, X86_FEATURE_STIBP);
872 
873 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
874 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
875 		set_cpu_cap(c, X86_FEATURE_SSBD);
876 
877 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
878 		set_cpu_cap(c, X86_FEATURE_IBRS);
879 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
880 	}
881 
882 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
883 		set_cpu_cap(c, X86_FEATURE_IBPB);
884 
885 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
886 		set_cpu_cap(c, X86_FEATURE_STIBP);
887 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
888 	}
889 
890 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
891 		set_cpu_cap(c, X86_FEATURE_SSBD);
892 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
893 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
894 	}
895 }
896 
897 void get_cpu_cap(struct cpuinfo_x86 *c)
898 {
899 	u32 eax, ebx, ecx, edx;
900 
901 	/* Intel-defined flags: level 0x00000001 */
902 	if (c->cpuid_level >= 0x00000001) {
903 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
904 
905 		c->x86_capability[CPUID_1_ECX] = ecx;
906 		c->x86_capability[CPUID_1_EDX] = edx;
907 	}
908 
909 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
910 	if (c->cpuid_level >= 0x00000006)
911 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
912 
913 	/* Additional Intel-defined flags: level 0x00000007 */
914 	if (c->cpuid_level >= 0x00000007) {
915 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
916 		c->x86_capability[CPUID_7_0_EBX] = ebx;
917 		c->x86_capability[CPUID_7_ECX] = ecx;
918 		c->x86_capability[CPUID_7_EDX] = edx;
919 
920 		/* Check valid sub-leaf index before accessing it */
921 		if (eax >= 1) {
922 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
923 			c->x86_capability[CPUID_7_1_EAX] = eax;
924 		}
925 	}
926 
927 	/* Extended state features: level 0x0000000d */
928 	if (c->cpuid_level >= 0x0000000d) {
929 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
930 
931 		c->x86_capability[CPUID_D_1_EAX] = eax;
932 	}
933 
934 	/* AMD-defined flags: level 0x80000001 */
935 	eax = cpuid_eax(0x80000000);
936 	c->extended_cpuid_level = eax;
937 
938 	if ((eax & 0xffff0000) == 0x80000000) {
939 		if (eax >= 0x80000001) {
940 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
941 
942 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
943 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
944 		}
945 	}
946 
947 	if (c->extended_cpuid_level >= 0x80000007) {
948 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
949 
950 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
951 		c->x86_power = edx;
952 	}
953 
954 	if (c->extended_cpuid_level >= 0x80000008) {
955 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
956 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
957 	}
958 
959 	if (c->extended_cpuid_level >= 0x8000000a)
960 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
961 
962 	init_scattered_cpuid_features(c);
963 	init_speculation_control(c);
964 
965 	/*
966 	 * Clear/Set all flags overridden by options, after probe.
967 	 * This needs to happen each time we re-probe, which may happen
968 	 * several times during CPU initialization.
969 	 */
970 	apply_forced_caps(c);
971 }
972 
973 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
974 {
975 	u32 eax, ebx, ecx, edx;
976 
977 	if (c->extended_cpuid_level >= 0x80000008) {
978 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
979 
980 		c->x86_virt_bits = (eax >> 8) & 0xff;
981 		c->x86_phys_bits = eax & 0xff;
982 	}
983 #ifdef CONFIG_X86_32
984 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
985 		c->x86_phys_bits = 36;
986 #endif
987 	c->x86_cache_bits = c->x86_phys_bits;
988 }
989 
990 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
991 {
992 #ifdef CONFIG_X86_32
993 	int i;
994 
995 	/*
996 	 * First of all, decide if this is a 486 or higher
997 	 * It's a 486 if we can modify the AC flag
998 	 */
999 	if (flag_is_changeable_p(X86_EFLAGS_AC))
1000 		c->x86 = 4;
1001 	else
1002 		c->x86 = 3;
1003 
1004 	for (i = 0; i < X86_VENDOR_NUM; i++)
1005 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
1006 			c->x86_vendor_id[0] = 0;
1007 			cpu_devs[i]->c_identify(c);
1008 			if (c->x86_vendor_id[0]) {
1009 				get_cpu_vendor(c);
1010 				break;
1011 			}
1012 		}
1013 #endif
1014 }
1015 
1016 #define NO_SPECULATION		BIT(0)
1017 #define NO_MELTDOWN		BIT(1)
1018 #define NO_SSB			BIT(2)
1019 #define NO_L1TF			BIT(3)
1020 #define NO_MDS			BIT(4)
1021 #define MSBDS_ONLY		BIT(5)
1022 #define NO_SWAPGS		BIT(6)
1023 #define NO_ITLB_MULTIHIT	BIT(7)
1024 #define NO_SPECTRE_V2		BIT(8)
1025 
1026 #define VULNWL(vendor, family, model, whitelist)	\
1027 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist)
1028 
1029 #define VULNWL_INTEL(model, whitelist)		\
1030 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1031 
1032 #define VULNWL_AMD(family, whitelist)		\
1033 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1034 
1035 #define VULNWL_HYGON(family, whitelist)		\
1036 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1037 
1038 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1039 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1040 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1041 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1042 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1043 
1044 	/* Intel Family 6 */
1045 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1046 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION | NO_ITLB_MULTIHIT),
1047 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1048 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1049 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1050 
1051 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1052 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1053 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1054 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1055 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1056 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1057 
1058 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1059 
1060 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1061 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1062 
1063 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1064 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1065 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1066 
1067 	/*
1068 	 * Technically, swapgs isn't serializing on AMD (despite it previously
1069 	 * being documented as such in the APM).  But according to AMD, %gs is
1070 	 * updated non-speculatively, and the issuing of %gs-relative memory
1071 	 * operands will be blocked until the %gs update completes, which is
1072 	 * good enough for our purposes.
1073 	 */
1074 
1075 	VULNWL_INTEL(ATOM_TREMONT_D,		NO_ITLB_MULTIHIT),
1076 
1077 	/* AMD Family 0xf - 0x12 */
1078 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1079 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1080 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1081 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1082 
1083 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1084 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1085 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1086 
1087 	/* Zhaoxin Family 7 */
1088 	VULNWL(CENTAUR,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1089 	VULNWL(ZHAOXIN,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1090 	{}
1091 };
1092 
1093 #define VULNBL_INTEL_STEPPINGS(model, steppings, issues)		   \
1094 	X86_MATCH_VENDOR_FAM_MODEL_STEPPINGS_FEATURE(INTEL, 6,		   \
1095 					    INTEL_FAM6_##model, steppings, \
1096 					    X86_FEATURE_ANY, issues)
1097 
1098 #define SRBDS		BIT(0)
1099 
1100 static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = {
1101 	VULNBL_INTEL_STEPPINGS(IVYBRIDGE,	X86_STEPPING_ANY,		SRBDS),
1102 	VULNBL_INTEL_STEPPINGS(HASWELL,		X86_STEPPING_ANY,		SRBDS),
1103 	VULNBL_INTEL_STEPPINGS(HASWELL_L,	X86_STEPPING_ANY,		SRBDS),
1104 	VULNBL_INTEL_STEPPINGS(HASWELL_G,	X86_STEPPING_ANY,		SRBDS),
1105 	VULNBL_INTEL_STEPPINGS(BROADWELL_G,	X86_STEPPING_ANY,		SRBDS),
1106 	VULNBL_INTEL_STEPPINGS(BROADWELL,	X86_STEPPING_ANY,		SRBDS),
1107 	VULNBL_INTEL_STEPPINGS(SKYLAKE_L,	X86_STEPPING_ANY,		SRBDS),
1108 	VULNBL_INTEL_STEPPINGS(SKYLAKE,		X86_STEPPING_ANY,		SRBDS),
1109 	VULNBL_INTEL_STEPPINGS(KABYLAKE_L,	X86_STEPPINGS(0x0, 0xC),	SRBDS),
1110 	VULNBL_INTEL_STEPPINGS(KABYLAKE,	X86_STEPPINGS(0x0, 0xD),	SRBDS),
1111 	{}
1112 };
1113 
1114 static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which)
1115 {
1116 	const struct x86_cpu_id *m = x86_match_cpu(table);
1117 
1118 	return m && !!(m->driver_data & which);
1119 }
1120 
1121 u64 x86_read_arch_cap_msr(void)
1122 {
1123 	u64 ia32_cap = 0;
1124 
1125 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1126 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1127 
1128 	return ia32_cap;
1129 }
1130 
1131 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1132 {
1133 	u64 ia32_cap = x86_read_arch_cap_msr();
1134 
1135 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1136 	if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) &&
1137 	    !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO))
1138 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1139 
1140 	if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION))
1141 		return;
1142 
1143 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1144 
1145 	if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2))
1146 		setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1147 
1148 	if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) &&
1149 	    !(ia32_cap & ARCH_CAP_SSB_NO) &&
1150 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1151 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1152 
1153 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1154 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1155 
1156 	if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) &&
1157 	    !(ia32_cap & ARCH_CAP_MDS_NO)) {
1158 		setup_force_cpu_bug(X86_BUG_MDS);
1159 		if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY))
1160 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1161 	}
1162 
1163 	if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS))
1164 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1165 
1166 	/*
1167 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1168 	 *	- TSX is supported or
1169 	 *	- TSX_CTRL is present
1170 	 *
1171 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1172 	 * the kernel boot e.g. kexec.
1173 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1174 	 * update is not present or running as guest that don't get TSX_CTRL.
1175 	 */
1176 	if (!(ia32_cap & ARCH_CAP_TAA_NO) &&
1177 	    (cpu_has(c, X86_FEATURE_RTM) ||
1178 	     (ia32_cap & ARCH_CAP_TSX_CTRL_MSR)))
1179 		setup_force_cpu_bug(X86_BUG_TAA);
1180 
1181 	/*
1182 	 * SRBDS affects CPUs which support RDRAND or RDSEED and are listed
1183 	 * in the vulnerability blacklist.
1184 	 */
1185 	if ((cpu_has(c, X86_FEATURE_RDRAND) ||
1186 	     cpu_has(c, X86_FEATURE_RDSEED)) &&
1187 	    cpu_matches(cpu_vuln_blacklist, SRBDS))
1188 		    setup_force_cpu_bug(X86_BUG_SRBDS);
1189 
1190 	if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN))
1191 		return;
1192 
1193 	/* Rogue Data Cache Load? No! */
1194 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1195 		return;
1196 
1197 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1198 
1199 	if (cpu_matches(cpu_vuln_whitelist, NO_L1TF))
1200 		return;
1201 
1202 	setup_force_cpu_bug(X86_BUG_L1TF);
1203 }
1204 
1205 /*
1206  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1207  * unfortunately, that's not true in practice because of early VIA
1208  * chips and (more importantly) broken virtualizers that are not easy
1209  * to detect. In the latter case it doesn't even *fail* reliably, so
1210  * probing for it doesn't even work. Disable it completely on 32-bit
1211  * unless we can find a reliable way to detect all the broken cases.
1212  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1213  */
1214 static void detect_nopl(void)
1215 {
1216 #ifdef CONFIG_X86_32
1217 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1218 #else
1219 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1220 #endif
1221 }
1222 
1223 /*
1224  * Do minimum CPU detection early.
1225  * Fields really needed: vendor, cpuid_level, family, model, mask,
1226  * cache alignment.
1227  * The others are not touched to avoid unwanted side effects.
1228  *
1229  * WARNING: this function is only called on the boot CPU.  Don't add code
1230  * here that is supposed to run on all CPUs.
1231  */
1232 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1233 {
1234 #ifdef CONFIG_X86_64
1235 	c->x86_clflush_size = 64;
1236 	c->x86_phys_bits = 36;
1237 	c->x86_virt_bits = 48;
1238 #else
1239 	c->x86_clflush_size = 32;
1240 	c->x86_phys_bits = 32;
1241 	c->x86_virt_bits = 32;
1242 #endif
1243 	c->x86_cache_alignment = c->x86_clflush_size;
1244 
1245 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1246 	c->extended_cpuid_level = 0;
1247 
1248 	if (!have_cpuid_p())
1249 		identify_cpu_without_cpuid(c);
1250 
1251 	/* cyrix could have cpuid enabled via c_identify()*/
1252 	if (have_cpuid_p()) {
1253 		cpu_detect(c);
1254 		get_cpu_vendor(c);
1255 		get_cpu_cap(c);
1256 		get_cpu_address_sizes(c);
1257 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1258 
1259 		if (this_cpu->c_early_init)
1260 			this_cpu->c_early_init(c);
1261 
1262 		c->cpu_index = 0;
1263 		filter_cpuid_features(c, false);
1264 
1265 		if (this_cpu->c_bsp_init)
1266 			this_cpu->c_bsp_init(c);
1267 	} else {
1268 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1269 	}
1270 
1271 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1272 
1273 	cpu_set_bug_bits(c);
1274 
1275 	cpu_set_core_cap_bits(c);
1276 
1277 	fpu__init_system(c);
1278 
1279 #ifdef CONFIG_X86_32
1280 	/*
1281 	 * Regardless of whether PCID is enumerated, the SDM says
1282 	 * that it can't be enabled in 32-bit mode.
1283 	 */
1284 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1285 #endif
1286 
1287 	/*
1288 	 * Later in the boot process pgtable_l5_enabled() relies on
1289 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1290 	 * enabled by this point we need to clear the feature bit to avoid
1291 	 * false-positives at the later stage.
1292 	 *
1293 	 * pgtable_l5_enabled() can be false here for several reasons:
1294 	 *  - 5-level paging is disabled compile-time;
1295 	 *  - it's 32-bit kernel;
1296 	 *  - machine doesn't support 5-level paging;
1297 	 *  - user specified 'no5lvl' in kernel command line.
1298 	 */
1299 	if (!pgtable_l5_enabled())
1300 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1301 
1302 	detect_nopl();
1303 }
1304 
1305 void __init early_cpu_init(void)
1306 {
1307 	const struct cpu_dev *const *cdev;
1308 	int count = 0;
1309 
1310 #ifdef CONFIG_PROCESSOR_SELECT
1311 	pr_info("KERNEL supported cpus:\n");
1312 #endif
1313 
1314 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1315 		const struct cpu_dev *cpudev = *cdev;
1316 
1317 		if (count >= X86_VENDOR_NUM)
1318 			break;
1319 		cpu_devs[count] = cpudev;
1320 		count++;
1321 
1322 #ifdef CONFIG_PROCESSOR_SELECT
1323 		{
1324 			unsigned int j;
1325 
1326 			for (j = 0; j < 2; j++) {
1327 				if (!cpudev->c_ident[j])
1328 					continue;
1329 				pr_info("  %s %s\n", cpudev->c_vendor,
1330 					cpudev->c_ident[j]);
1331 			}
1332 		}
1333 #endif
1334 	}
1335 	early_identify_cpu(&boot_cpu_data);
1336 }
1337 
1338 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1339 {
1340 #ifdef CONFIG_X86_64
1341 	/*
1342 	 * Empirically, writing zero to a segment selector on AMD does
1343 	 * not clear the base, whereas writing zero to a segment
1344 	 * selector on Intel does clear the base.  Intel's behavior
1345 	 * allows slightly faster context switches in the common case
1346 	 * where GS is unused by the prev and next threads.
1347 	 *
1348 	 * Since neither vendor documents this anywhere that I can see,
1349 	 * detect it directly instead of hardcoding the choice by
1350 	 * vendor.
1351 	 *
1352 	 * I've designated AMD's behavior as the "bug" because it's
1353 	 * counterintuitive and less friendly.
1354 	 */
1355 
1356 	unsigned long old_base, tmp;
1357 	rdmsrl(MSR_FS_BASE, old_base);
1358 	wrmsrl(MSR_FS_BASE, 1);
1359 	loadsegment(fs, 0);
1360 	rdmsrl(MSR_FS_BASE, tmp);
1361 	if (tmp != 0)
1362 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1363 	wrmsrl(MSR_FS_BASE, old_base);
1364 #endif
1365 }
1366 
1367 static void generic_identify(struct cpuinfo_x86 *c)
1368 {
1369 	c->extended_cpuid_level = 0;
1370 
1371 	if (!have_cpuid_p())
1372 		identify_cpu_without_cpuid(c);
1373 
1374 	/* cyrix could have cpuid enabled via c_identify()*/
1375 	if (!have_cpuid_p())
1376 		return;
1377 
1378 	cpu_detect(c);
1379 
1380 	get_cpu_vendor(c);
1381 
1382 	get_cpu_cap(c);
1383 
1384 	get_cpu_address_sizes(c);
1385 
1386 	if (c->cpuid_level >= 0x00000001) {
1387 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1388 #ifdef CONFIG_X86_32
1389 # ifdef CONFIG_SMP
1390 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1391 # else
1392 		c->apicid = c->initial_apicid;
1393 # endif
1394 #endif
1395 		c->phys_proc_id = c->initial_apicid;
1396 	}
1397 
1398 	get_model_name(c); /* Default name */
1399 
1400 	detect_null_seg_behavior(c);
1401 
1402 	/*
1403 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1404 	 * systems that run Linux at CPL > 0 may or may not have the
1405 	 * issue, but, even if they have the issue, there's absolutely
1406 	 * nothing we can do about it because we can't use the real IRET
1407 	 * instruction.
1408 	 *
1409 	 * NB: For the time being, only 32-bit kernels support
1410 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1411 	 * whether to apply espfix using paravirt hooks.  If any
1412 	 * non-paravirt system ever shows up that does *not* have the
1413 	 * ESPFIX issue, we can change this.
1414 	 */
1415 #ifdef CONFIG_X86_32
1416 # ifdef CONFIG_PARAVIRT_XXL
1417 	do {
1418 		extern void native_iret(void);
1419 		if (pv_ops.cpu.iret == native_iret)
1420 			set_cpu_bug(c, X86_BUG_ESPFIX);
1421 	} while (0);
1422 # else
1423 	set_cpu_bug(c, X86_BUG_ESPFIX);
1424 # endif
1425 #endif
1426 }
1427 
1428 /*
1429  * Validate that ACPI/mptables have the same information about the
1430  * effective APIC id and update the package map.
1431  */
1432 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1433 {
1434 #ifdef CONFIG_SMP
1435 	unsigned int apicid, cpu = smp_processor_id();
1436 
1437 	apicid = apic->cpu_present_to_apicid(cpu);
1438 
1439 	if (apicid != c->apicid) {
1440 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1441 		       cpu, apicid, c->initial_apicid);
1442 	}
1443 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1444 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1445 #else
1446 	c->logical_proc_id = 0;
1447 #endif
1448 }
1449 
1450 /*
1451  * This does the hard work of actually picking apart the CPU stuff...
1452  */
1453 static void identify_cpu(struct cpuinfo_x86 *c)
1454 {
1455 	int i;
1456 
1457 	c->loops_per_jiffy = loops_per_jiffy;
1458 	c->x86_cache_size = 0;
1459 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1460 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1461 	c->x86_vendor_id[0] = '\0'; /* Unset */
1462 	c->x86_model_id[0] = '\0';  /* Unset */
1463 	c->x86_max_cores = 1;
1464 	c->x86_coreid_bits = 0;
1465 	c->cu_id = 0xff;
1466 #ifdef CONFIG_X86_64
1467 	c->x86_clflush_size = 64;
1468 	c->x86_phys_bits = 36;
1469 	c->x86_virt_bits = 48;
1470 #else
1471 	c->cpuid_level = -1;	/* CPUID not detected */
1472 	c->x86_clflush_size = 32;
1473 	c->x86_phys_bits = 32;
1474 	c->x86_virt_bits = 32;
1475 #endif
1476 	c->x86_cache_alignment = c->x86_clflush_size;
1477 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1478 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
1479 	memset(&c->vmx_capability, 0, sizeof(c->vmx_capability));
1480 #endif
1481 
1482 	generic_identify(c);
1483 
1484 	if (this_cpu->c_identify)
1485 		this_cpu->c_identify(c);
1486 
1487 	/* Clear/Set all flags overridden by options, after probe */
1488 	apply_forced_caps(c);
1489 
1490 #ifdef CONFIG_X86_64
1491 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1492 #endif
1493 
1494 	/*
1495 	 * Vendor-specific initialization.  In this section we
1496 	 * canonicalize the feature flags, meaning if there are
1497 	 * features a certain CPU supports which CPUID doesn't
1498 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1499 	 * we handle them here.
1500 	 *
1501 	 * At the end of this section, c->x86_capability better
1502 	 * indicate the features this CPU genuinely supports!
1503 	 */
1504 	if (this_cpu->c_init)
1505 		this_cpu->c_init(c);
1506 
1507 	/* Disable the PN if appropriate */
1508 	squash_the_stupid_serial_number(c);
1509 
1510 	/* Set up SMEP/SMAP/UMIP */
1511 	setup_smep(c);
1512 	setup_smap(c);
1513 	setup_umip(c);
1514 
1515 	/* Enable FSGSBASE instructions if available. */
1516 	if (cpu_has(c, X86_FEATURE_FSGSBASE)) {
1517 		cr4_set_bits(X86_CR4_FSGSBASE);
1518 		elf_hwcap2 |= HWCAP2_FSGSBASE;
1519 	}
1520 
1521 	/*
1522 	 * The vendor-specific functions might have changed features.
1523 	 * Now we do "generic changes."
1524 	 */
1525 
1526 	/* Filter out anything that depends on CPUID levels we don't have */
1527 	filter_cpuid_features(c, true);
1528 
1529 	/* If the model name is still unset, do table lookup. */
1530 	if (!c->x86_model_id[0]) {
1531 		const char *p;
1532 		p = table_lookup_model(c);
1533 		if (p)
1534 			strcpy(c->x86_model_id, p);
1535 		else
1536 			/* Last resort... */
1537 			sprintf(c->x86_model_id, "%02x/%02x",
1538 				c->x86, c->x86_model);
1539 	}
1540 
1541 #ifdef CONFIG_X86_64
1542 	detect_ht(c);
1543 #endif
1544 
1545 	x86_init_rdrand(c);
1546 	setup_pku(c);
1547 
1548 	/*
1549 	 * Clear/Set all flags overridden by options, need do it
1550 	 * before following smp all cpus cap AND.
1551 	 */
1552 	apply_forced_caps(c);
1553 
1554 	/*
1555 	 * On SMP, boot_cpu_data holds the common feature set between
1556 	 * all CPUs; so make sure that we indicate which features are
1557 	 * common between the CPUs.  The first time this routine gets
1558 	 * executed, c == &boot_cpu_data.
1559 	 */
1560 	if (c != &boot_cpu_data) {
1561 		/* AND the already accumulated flags with these */
1562 		for (i = 0; i < NCAPINTS; i++)
1563 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1564 
1565 		/* OR, i.e. replicate the bug flags */
1566 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1567 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1568 	}
1569 
1570 	/* Init Machine Check Exception if available. */
1571 	mcheck_cpu_init(c);
1572 
1573 	select_idle_routine(c);
1574 
1575 #ifdef CONFIG_NUMA
1576 	numa_add_cpu(smp_processor_id());
1577 #endif
1578 }
1579 
1580 /*
1581  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1582  * on 32-bit kernels:
1583  */
1584 #ifdef CONFIG_X86_32
1585 void enable_sep_cpu(void)
1586 {
1587 	struct tss_struct *tss;
1588 	int cpu;
1589 
1590 	if (!boot_cpu_has(X86_FEATURE_SEP))
1591 		return;
1592 
1593 	cpu = get_cpu();
1594 	tss = &per_cpu(cpu_tss_rw, cpu);
1595 
1596 	/*
1597 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1598 	 * see the big comment in struct x86_hw_tss's definition.
1599 	 */
1600 
1601 	tss->x86_tss.ss1 = __KERNEL_CS;
1602 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1603 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1604 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1605 
1606 	put_cpu();
1607 }
1608 #endif
1609 
1610 void __init identify_boot_cpu(void)
1611 {
1612 	identify_cpu(&boot_cpu_data);
1613 #ifdef CONFIG_X86_32
1614 	sysenter_setup();
1615 	enable_sep_cpu();
1616 #endif
1617 	cpu_detect_tlb(&boot_cpu_data);
1618 	setup_cr_pinning();
1619 
1620 	tsx_init();
1621 }
1622 
1623 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1624 {
1625 	BUG_ON(c == &boot_cpu_data);
1626 	identify_cpu(c);
1627 #ifdef CONFIG_X86_32
1628 	enable_sep_cpu();
1629 #endif
1630 	mtrr_ap_init();
1631 	validate_apic_and_package_id(c);
1632 	x86_spec_ctrl_setup_ap();
1633 	update_srbds_msr();
1634 }
1635 
1636 static __init int setup_noclflush(char *arg)
1637 {
1638 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1639 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1640 	return 1;
1641 }
1642 __setup("noclflush", setup_noclflush);
1643 
1644 void print_cpu_info(struct cpuinfo_x86 *c)
1645 {
1646 	const char *vendor = NULL;
1647 
1648 	if (c->x86_vendor < X86_VENDOR_NUM) {
1649 		vendor = this_cpu->c_vendor;
1650 	} else {
1651 		if (c->cpuid_level >= 0)
1652 			vendor = c->x86_vendor_id;
1653 	}
1654 
1655 	if (vendor && !strstr(c->x86_model_id, vendor))
1656 		pr_cont("%s ", vendor);
1657 
1658 	if (c->x86_model_id[0])
1659 		pr_cont("%s", c->x86_model_id);
1660 	else
1661 		pr_cont("%d86", c->x86);
1662 
1663 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1664 
1665 	if (c->x86_stepping || c->cpuid_level >= 0)
1666 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1667 	else
1668 		pr_cont(")\n");
1669 }
1670 
1671 /*
1672  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1673  * But we need to keep a dummy __setup around otherwise it would
1674  * show up as an environment variable for init.
1675  */
1676 static __init int setup_clearcpuid(char *arg)
1677 {
1678 	return 1;
1679 }
1680 __setup("clearcpuid=", setup_clearcpuid);
1681 
1682 #ifdef CONFIG_X86_64
1683 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1684 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1685 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1686 
1687 /*
1688  * The following percpu variables are hot.  Align current_task to
1689  * cacheline size such that they fall in the same cacheline.
1690  */
1691 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1692 	&init_task;
1693 EXPORT_PER_CPU_SYMBOL(current_task);
1694 
1695 DEFINE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
1696 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1697 
1698 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1699 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1700 
1701 /* May not be marked __init: used by software suspend */
1702 void syscall_init(void)
1703 {
1704 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1705 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1706 
1707 #ifdef CONFIG_IA32_EMULATION
1708 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1709 	/*
1710 	 * This only works on Intel CPUs.
1711 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1712 	 * This does not cause SYSENTER to jump to the wrong location, because
1713 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1714 	 */
1715 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1716 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1717 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1718 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1719 #else
1720 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1721 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1722 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1723 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1724 #endif
1725 
1726 	/* Flags to clear on syscall */
1727 	wrmsrl(MSR_SYSCALL_MASK,
1728 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1729 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1730 }
1731 
1732 #else	/* CONFIG_X86_64 */
1733 
1734 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1735 EXPORT_PER_CPU_SYMBOL(current_task);
1736 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1737 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1738 
1739 /*
1740  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1741  * the top of the kernel stack.  Use an extra percpu variable to track the
1742  * top of the kernel stack directly.
1743  */
1744 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1745 	(unsigned long)&init_thread_union + THREAD_SIZE;
1746 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1747 
1748 #ifdef CONFIG_STACKPROTECTOR
1749 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1750 #endif
1751 
1752 #endif	/* CONFIG_X86_64 */
1753 
1754 /*
1755  * Clear all 6 debug registers:
1756  */
1757 static void clear_all_debug_regs(void)
1758 {
1759 	int i;
1760 
1761 	for (i = 0; i < 8; i++) {
1762 		/* Ignore db4, db5 */
1763 		if ((i == 4) || (i == 5))
1764 			continue;
1765 
1766 		set_debugreg(0, i);
1767 	}
1768 }
1769 
1770 #ifdef CONFIG_KGDB
1771 /*
1772  * Restore debug regs if using kgdbwait and you have a kernel debugger
1773  * connection established.
1774  */
1775 static void dbg_restore_debug_regs(void)
1776 {
1777 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1778 		arch_kgdb_ops.correct_hw_break();
1779 }
1780 #else /* ! CONFIG_KGDB */
1781 #define dbg_restore_debug_regs()
1782 #endif /* ! CONFIG_KGDB */
1783 
1784 static void wait_for_master_cpu(int cpu)
1785 {
1786 #ifdef CONFIG_SMP
1787 	/*
1788 	 * wait for ACK from master CPU before continuing
1789 	 * with AP initialization
1790 	 */
1791 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1792 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1793 		cpu_relax();
1794 #endif
1795 }
1796 
1797 #ifdef CONFIG_X86_64
1798 static inline void setup_getcpu(int cpu)
1799 {
1800 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1801 	struct desc_struct d = { };
1802 
1803 	if (boot_cpu_has(X86_FEATURE_RDTSCP))
1804 		write_rdtscp_aux(cpudata);
1805 
1806 	/* Store CPU and node number in limit. */
1807 	d.limit0 = cpudata;
1808 	d.limit1 = cpudata >> 16;
1809 
1810 	d.type = 5;		/* RO data, expand down, accessed */
1811 	d.dpl = 3;		/* Visible to user code */
1812 	d.s = 1;		/* Not a system segment */
1813 	d.p = 1;		/* Present */
1814 	d.d = 1;		/* 32-bit */
1815 
1816 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1817 }
1818 
1819 static inline void ucode_cpu_init(int cpu)
1820 {
1821 	if (cpu)
1822 		load_ucode_ap();
1823 }
1824 
1825 static inline void tss_setup_ist(struct tss_struct *tss)
1826 {
1827 	/* Set up the per-CPU TSS IST stacks */
1828 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1829 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1830 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1831 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1832 }
1833 
1834 #else /* CONFIG_X86_64 */
1835 
1836 static inline void setup_getcpu(int cpu) { }
1837 
1838 static inline void ucode_cpu_init(int cpu)
1839 {
1840 	show_ucode_info_early();
1841 }
1842 
1843 static inline void tss_setup_ist(struct tss_struct *tss) { }
1844 
1845 #endif /* !CONFIG_X86_64 */
1846 
1847 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
1848 {
1849 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
1850 
1851 #ifdef CONFIG_X86_IOPL_IOPERM
1852 	tss->io_bitmap.prev_max = 0;
1853 	tss->io_bitmap.prev_sequence = 0;
1854 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
1855 	/*
1856 	 * Invalidate the extra array entry past the end of the all
1857 	 * permission bitmap as required by the hardware.
1858 	 */
1859 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
1860 #endif
1861 }
1862 
1863 /*
1864  * cpu_init() initializes state that is per-CPU. Some data is already
1865  * initialized (naturally) in the bootstrap process, such as the GDT
1866  * and IDT. We reload them nevertheless, this function acts as a
1867  * 'CPU state barrier', nothing should get across.
1868  */
1869 void cpu_init(void)
1870 {
1871 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
1872 	struct task_struct *cur = current;
1873 	int cpu = raw_smp_processor_id();
1874 
1875 	wait_for_master_cpu(cpu);
1876 
1877 	ucode_cpu_init(cpu);
1878 
1879 #ifdef CONFIG_NUMA
1880 	if (this_cpu_read(numa_node) == 0 &&
1881 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1882 		set_numa_node(early_cpu_to_node(cpu));
1883 #endif
1884 	setup_getcpu(cpu);
1885 
1886 	pr_debug("Initializing CPU#%d\n", cpu);
1887 
1888 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
1889 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
1890 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1891 
1892 	/*
1893 	 * Initialize the per-CPU GDT with the boot GDT,
1894 	 * and set up the GDT descriptor:
1895 	 */
1896 	switch_to_new_gdt(cpu);
1897 	load_current_idt();
1898 
1899 	if (IS_ENABLED(CONFIG_X86_64)) {
1900 		loadsegment(fs, 0);
1901 		memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1902 		syscall_init();
1903 
1904 		wrmsrl(MSR_FS_BASE, 0);
1905 		wrmsrl(MSR_KERNEL_GS_BASE, 0);
1906 		barrier();
1907 
1908 		x2apic_setup();
1909 	}
1910 
1911 	mmgrab(&init_mm);
1912 	cur->active_mm = &init_mm;
1913 	BUG_ON(cur->mm);
1914 	initialize_tlbstate_and_flush();
1915 	enter_lazy_tlb(&init_mm, cur);
1916 
1917 	/* Initialize the TSS. */
1918 	tss_setup_ist(tss);
1919 	tss_setup_io_bitmap(tss);
1920 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1921 
1922 	load_TR_desc();
1923 	/*
1924 	 * sp0 points to the entry trampoline stack regardless of what task
1925 	 * is running.
1926 	 */
1927 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1928 
1929 	load_mm_ldt(&init_mm);
1930 
1931 	clear_all_debug_regs();
1932 	dbg_restore_debug_regs();
1933 
1934 	doublefault_init_cpu_tss();
1935 
1936 	fpu__init_cpu();
1937 
1938 	if (is_uv_system())
1939 		uv_cpu_init();
1940 
1941 	load_fixmap_gdt(cpu);
1942 }
1943 
1944 /*
1945  * The microcode loader calls this upon late microcode load to recheck features,
1946  * only when microcode has been updated. Caller holds microcode_mutex and CPU
1947  * hotplug lock.
1948  */
1949 void microcode_check(void)
1950 {
1951 	struct cpuinfo_x86 info;
1952 
1953 	perf_check_microcode();
1954 
1955 	/* Reload CPUID max function as it might've changed. */
1956 	info.cpuid_level = cpuid_eax(0);
1957 
1958 	/*
1959 	 * Copy all capability leafs to pick up the synthetic ones so that
1960 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
1961 	 * get overwritten in get_cpu_cap().
1962 	 */
1963 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
1964 
1965 	get_cpu_cap(&info);
1966 
1967 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
1968 		return;
1969 
1970 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
1971 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
1972 }
1973 
1974 /*
1975  * Invoked from core CPU hotplug code after hotplug operations
1976  */
1977 void arch_smt_update(void)
1978 {
1979 	/* Handle the speculative execution misfeatures */
1980 	cpu_bugs_smt_update();
1981 	/* Check whether IPI broadcasting can be enabled */
1982 	apic_smt_update();
1983 }
1984