xref: /linux/arch/x86/kernel/cpu/common.c (revision 1ee54195a305fae3955642af8528bdf67496d353)
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 #include <asm/uv/uv.h>
57 
58 #include "cpu.h"
59 
60 u32 elf_hwcap2 __read_mostly;
61 
62 /* all of these masks are initialized in setup_cpu_local_masks() */
63 cpumask_var_t cpu_initialized_mask;
64 cpumask_var_t cpu_callout_mask;
65 cpumask_var_t cpu_callin_mask;
66 
67 /* representing cpus for which sibling maps can be computed */
68 cpumask_var_t cpu_sibling_setup_mask;
69 
70 /* Number of siblings per CPU package */
71 int smp_num_siblings = 1;
72 EXPORT_SYMBOL(smp_num_siblings);
73 
74 /* Last level cache ID of each logical CPU */
75 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
76 
77 /* correctly size the local cpu masks */
78 void __init setup_cpu_local_masks(void)
79 {
80 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
81 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
82 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
83 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
84 }
85 
86 static void default_init(struct cpuinfo_x86 *c)
87 {
88 #ifdef CONFIG_X86_64
89 	cpu_detect_cache_sizes(c);
90 #else
91 	/* Not much we can do here... */
92 	/* Check if at least it has cpuid */
93 	if (c->cpuid_level == -1) {
94 		/* No cpuid. It must be an ancient CPU */
95 		if (c->x86 == 4)
96 			strcpy(c->x86_model_id, "486");
97 		else if (c->x86 == 3)
98 			strcpy(c->x86_model_id, "386");
99 	}
100 #endif
101 }
102 
103 static const struct cpu_dev default_cpu = {
104 	.c_init		= default_init,
105 	.c_vendor	= "Unknown",
106 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
107 };
108 
109 static const struct cpu_dev *this_cpu = &default_cpu;
110 
111 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
112 #ifdef CONFIG_X86_64
113 	/*
114 	 * We need valid kernel segments for data and code in long mode too
115 	 * IRET will check the segment types  kkeil 2000/10/28
116 	 * Also sysret mandates a special GDT layout
117 	 *
118 	 * TLS descriptors are currently at a different place compared to i386.
119 	 * Hopefully nobody expects them at a fixed place (Wine?)
120 	 */
121 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
122 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
123 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
124 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
125 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
126 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
127 #else
128 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
129 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
130 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
131 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
132 	/*
133 	 * Segments used for calling PnP BIOS have byte granularity.
134 	 * They code segments and data segments have fixed 64k limits,
135 	 * the transfer segment sizes are set at run time.
136 	 */
137 	/* 32-bit code */
138 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
139 	/* 16-bit code */
140 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
141 	/* 16-bit data */
142 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
143 	/* 16-bit data */
144 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
145 	/* 16-bit data */
146 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
147 	/*
148 	 * The APM segments have byte granularity and their bases
149 	 * are set at run time.  All have 64k limits.
150 	 */
151 	/* 32-bit code */
152 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
153 	/* 16-bit code */
154 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
155 	/* data */
156 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
157 
158 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
159 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
160 	GDT_STACK_CANARY_INIT
161 #endif
162 } };
163 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
164 
165 static int __init x86_mpx_setup(char *s)
166 {
167 	/* require an exact match without trailing characters */
168 	if (strlen(s))
169 		return 0;
170 
171 	/* do not emit a message if the feature is not present */
172 	if (!boot_cpu_has(X86_FEATURE_MPX))
173 		return 1;
174 
175 	setup_clear_cpu_cap(X86_FEATURE_MPX);
176 	pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
177 	return 1;
178 }
179 __setup("nompx", x86_mpx_setup);
180 
181 #ifdef CONFIG_X86_64
182 static int __init x86_nopcid_setup(char *s)
183 {
184 	/* nopcid doesn't accept parameters */
185 	if (s)
186 		return -EINVAL;
187 
188 	/* do not emit a message if the feature is not present */
189 	if (!boot_cpu_has(X86_FEATURE_PCID))
190 		return 0;
191 
192 	setup_clear_cpu_cap(X86_FEATURE_PCID);
193 	pr_info("nopcid: PCID feature disabled\n");
194 	return 0;
195 }
196 early_param("nopcid", x86_nopcid_setup);
197 #endif
198 
199 static int __init x86_noinvpcid_setup(char *s)
200 {
201 	/* noinvpcid doesn't accept parameters */
202 	if (s)
203 		return -EINVAL;
204 
205 	/* do not emit a message if the feature is not present */
206 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
207 		return 0;
208 
209 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
210 	pr_info("noinvpcid: INVPCID feature disabled\n");
211 	return 0;
212 }
213 early_param("noinvpcid", x86_noinvpcid_setup);
214 
215 #ifdef CONFIG_X86_32
216 static int cachesize_override = -1;
217 static int disable_x86_serial_nr = 1;
218 
219 static int __init cachesize_setup(char *str)
220 {
221 	get_option(&str, &cachesize_override);
222 	return 1;
223 }
224 __setup("cachesize=", cachesize_setup);
225 
226 static int __init x86_sep_setup(char *s)
227 {
228 	setup_clear_cpu_cap(X86_FEATURE_SEP);
229 	return 1;
230 }
231 __setup("nosep", x86_sep_setup);
232 
233 /* Standard macro to see if a specific flag is changeable */
234 static inline int flag_is_changeable_p(u32 flag)
235 {
236 	u32 f1, f2;
237 
238 	/*
239 	 * Cyrix and IDT cpus allow disabling of CPUID
240 	 * so the code below may return different results
241 	 * when it is executed before and after enabling
242 	 * the CPUID. Add "volatile" to not allow gcc to
243 	 * optimize the subsequent calls to this function.
244 	 */
245 	asm volatile ("pushfl		\n\t"
246 		      "pushfl		\n\t"
247 		      "popl %0		\n\t"
248 		      "movl %0, %1	\n\t"
249 		      "xorl %2, %0	\n\t"
250 		      "pushl %0		\n\t"
251 		      "popfl		\n\t"
252 		      "pushfl		\n\t"
253 		      "popl %0		\n\t"
254 		      "popfl		\n\t"
255 
256 		      : "=&r" (f1), "=&r" (f2)
257 		      : "ir" (flag));
258 
259 	return ((f1^f2) & flag) != 0;
260 }
261 
262 /* Probe for the CPUID instruction */
263 int have_cpuid_p(void)
264 {
265 	return flag_is_changeable_p(X86_EFLAGS_ID);
266 }
267 
268 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
269 {
270 	unsigned long lo, hi;
271 
272 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
273 		return;
274 
275 	/* Disable processor serial number: */
276 
277 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
278 	lo |= 0x200000;
279 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
280 
281 	pr_notice("CPU serial number disabled.\n");
282 	clear_cpu_cap(c, X86_FEATURE_PN);
283 
284 	/* Disabling the serial number may affect the cpuid level */
285 	c->cpuid_level = cpuid_eax(0);
286 }
287 
288 static int __init x86_serial_nr_setup(char *s)
289 {
290 	disable_x86_serial_nr = 0;
291 	return 1;
292 }
293 __setup("serialnumber", x86_serial_nr_setup);
294 #else
295 static inline int flag_is_changeable_p(u32 flag)
296 {
297 	return 1;
298 }
299 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
300 {
301 }
302 #endif
303 
304 static __init int setup_disable_smep(char *arg)
305 {
306 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
307 	/* Check for things that depend on SMEP being enabled: */
308 	check_mpx_erratum(&boot_cpu_data);
309 	return 1;
310 }
311 __setup("nosmep", setup_disable_smep);
312 
313 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
314 {
315 	if (cpu_has(c, X86_FEATURE_SMEP))
316 		cr4_set_bits(X86_CR4_SMEP);
317 }
318 
319 static __init int setup_disable_smap(char *arg)
320 {
321 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
322 	return 1;
323 }
324 __setup("nosmap", setup_disable_smap);
325 
326 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
327 {
328 	unsigned long eflags = native_save_fl();
329 
330 	/* This should have been cleared long ago */
331 	BUG_ON(eflags & X86_EFLAGS_AC);
332 
333 	if (cpu_has(c, X86_FEATURE_SMAP)) {
334 #ifdef CONFIG_X86_SMAP
335 		cr4_set_bits(X86_CR4_SMAP);
336 #else
337 		cr4_clear_bits(X86_CR4_SMAP);
338 #endif
339 	}
340 }
341 
342 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
343 {
344 	/* Check the boot processor, plus build option for UMIP. */
345 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
346 		goto out;
347 
348 	/* Check the current processor's cpuid bits. */
349 	if (!cpu_has(c, X86_FEATURE_UMIP))
350 		goto out;
351 
352 	cr4_set_bits(X86_CR4_UMIP);
353 
354 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
355 
356 	return;
357 
358 out:
359 	/*
360 	 * Make sure UMIP is disabled in case it was enabled in a
361 	 * previous boot (e.g., via kexec).
362 	 */
363 	cr4_clear_bits(X86_CR4_UMIP);
364 }
365 
366 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
367 static unsigned long cr4_pinned_bits __ro_after_init;
368 
369 void native_write_cr0(unsigned long val)
370 {
371 	unsigned long bits_missing = 0;
372 
373 set_register:
374 	asm volatile("mov %0,%%cr0": "+r" (val), "+m" (__force_order));
375 
376 	if (static_branch_likely(&cr_pinning)) {
377 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
378 			bits_missing = X86_CR0_WP;
379 			val |= bits_missing;
380 			goto set_register;
381 		}
382 		/* Warn after we've set the missing bits. */
383 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
384 	}
385 }
386 EXPORT_SYMBOL(native_write_cr0);
387 
388 void native_write_cr4(unsigned long val)
389 {
390 	unsigned long bits_missing = 0;
391 
392 set_register:
393 	asm volatile("mov %0,%%cr4": "+r" (val), "+m" (cr4_pinned_bits));
394 
395 	if (static_branch_likely(&cr_pinning)) {
396 		if (unlikely((val & cr4_pinned_bits) != cr4_pinned_bits)) {
397 			bits_missing = ~val & cr4_pinned_bits;
398 			val |= bits_missing;
399 			goto set_register;
400 		}
401 		/* Warn after we've set the missing bits. */
402 		WARN_ONCE(bits_missing, "CR4 bits went missing: %lx!?\n",
403 			  bits_missing);
404 	}
405 }
406 EXPORT_SYMBOL(native_write_cr4);
407 
408 void cr4_init(void)
409 {
410 	unsigned long cr4 = __read_cr4();
411 
412 	if (boot_cpu_has(X86_FEATURE_PCID))
413 		cr4 |= X86_CR4_PCIDE;
414 	if (static_branch_likely(&cr_pinning))
415 		cr4 |= cr4_pinned_bits;
416 
417 	__write_cr4(cr4);
418 
419 	/* Initialize cr4 shadow for this CPU. */
420 	this_cpu_write(cpu_tlbstate.cr4, cr4);
421 }
422 
423 /*
424  * Once CPU feature detection is finished (and boot params have been
425  * parsed), record any of the sensitive CR bits that are set, and
426  * enable CR pinning.
427  */
428 static void __init setup_cr_pinning(void)
429 {
430 	unsigned long mask;
431 
432 	mask = (X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP);
433 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & mask;
434 	static_key_enable(&cr_pinning.key);
435 }
436 
437 /*
438  * Protection Keys are not available in 32-bit mode.
439  */
440 static bool pku_disabled;
441 
442 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
443 {
444 	struct pkru_state *pk;
445 
446 	/* check the boot processor, plus compile options for PKU: */
447 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
448 		return;
449 	/* checks the actual processor's cpuid bits: */
450 	if (!cpu_has(c, X86_FEATURE_PKU))
451 		return;
452 	if (pku_disabled)
453 		return;
454 
455 	cr4_set_bits(X86_CR4_PKE);
456 	pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU);
457 	if (pk)
458 		pk->pkru = init_pkru_value;
459 	/*
460 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
461 	 * cpuid bit to be set.  We need to ensure that we
462 	 * update that bit in this CPU's "cpu_info".
463 	 */
464 	get_cpu_cap(c);
465 }
466 
467 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
468 static __init int setup_disable_pku(char *arg)
469 {
470 	/*
471 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
472 	 * runtime checks are against OSPKE so clearing the
473 	 * bit does nothing.
474 	 *
475 	 * This way, we will see "pku" in cpuinfo, but not
476 	 * "ospke", which is exactly what we want.  It shows
477 	 * that the CPU has PKU, but the OS has not enabled it.
478 	 * This happens to be exactly how a system would look
479 	 * if we disabled the config option.
480 	 */
481 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
482 	pku_disabled = true;
483 	return 1;
484 }
485 __setup("nopku", setup_disable_pku);
486 #endif /* CONFIG_X86_64 */
487 
488 /*
489  * Some CPU features depend on higher CPUID levels, which may not always
490  * be available due to CPUID level capping or broken virtualization
491  * software.  Add those features to this table to auto-disable them.
492  */
493 struct cpuid_dependent_feature {
494 	u32 feature;
495 	u32 level;
496 };
497 
498 static const struct cpuid_dependent_feature
499 cpuid_dependent_features[] = {
500 	{ X86_FEATURE_MWAIT,		0x00000005 },
501 	{ X86_FEATURE_DCA,		0x00000009 },
502 	{ X86_FEATURE_XSAVE,		0x0000000d },
503 	{ 0, 0 }
504 };
505 
506 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
507 {
508 	const struct cpuid_dependent_feature *df;
509 
510 	for (df = cpuid_dependent_features; df->feature; df++) {
511 
512 		if (!cpu_has(c, df->feature))
513 			continue;
514 		/*
515 		 * Note: cpuid_level is set to -1 if unavailable, but
516 		 * extended_extended_level is set to 0 if unavailable
517 		 * and the legitimate extended levels are all negative
518 		 * when signed; hence the weird messing around with
519 		 * signs here...
520 		 */
521 		if (!((s32)df->level < 0 ?
522 		     (u32)df->level > (u32)c->extended_cpuid_level :
523 		     (s32)df->level > (s32)c->cpuid_level))
524 			continue;
525 
526 		clear_cpu_cap(c, df->feature);
527 		if (!warn)
528 			continue;
529 
530 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
531 			x86_cap_flag(df->feature), df->level);
532 	}
533 }
534 
535 /*
536  * Naming convention should be: <Name> [(<Codename>)]
537  * This table only is used unless init_<vendor>() below doesn't set it;
538  * in particular, if CPUID levels 0x80000002..4 are supported, this
539  * isn't used
540  */
541 
542 /* Look up CPU names by table lookup. */
543 static const char *table_lookup_model(struct cpuinfo_x86 *c)
544 {
545 #ifdef CONFIG_X86_32
546 	const struct legacy_cpu_model_info *info;
547 
548 	if (c->x86_model >= 16)
549 		return NULL;	/* Range check */
550 
551 	if (!this_cpu)
552 		return NULL;
553 
554 	info = this_cpu->legacy_models;
555 
556 	while (info->family) {
557 		if (info->family == c->x86)
558 			return info->model_names[c->x86_model];
559 		info++;
560 	}
561 #endif
562 	return NULL;		/* Not found */
563 }
564 
565 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
566 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
567 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
568 
569 void load_percpu_segment(int cpu)
570 {
571 #ifdef CONFIG_X86_32
572 	loadsegment(fs, __KERNEL_PERCPU);
573 #else
574 	__loadsegment_simple(gs, 0);
575 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
576 #endif
577 	load_stack_canary_segment();
578 }
579 
580 #ifdef CONFIG_X86_32
581 /* The 32-bit entry code needs to find cpu_entry_area. */
582 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
583 #endif
584 
585 /* Load the original GDT from the per-cpu structure */
586 void load_direct_gdt(int cpu)
587 {
588 	struct desc_ptr gdt_descr;
589 
590 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
591 	gdt_descr.size = GDT_SIZE - 1;
592 	load_gdt(&gdt_descr);
593 }
594 EXPORT_SYMBOL_GPL(load_direct_gdt);
595 
596 /* Load a fixmap remapping of the per-cpu GDT */
597 void load_fixmap_gdt(int cpu)
598 {
599 	struct desc_ptr gdt_descr;
600 
601 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
602 	gdt_descr.size = GDT_SIZE - 1;
603 	load_gdt(&gdt_descr);
604 }
605 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
606 
607 /*
608  * Current gdt points %fs at the "master" per-cpu area: after this,
609  * it's on the real one.
610  */
611 void switch_to_new_gdt(int cpu)
612 {
613 	/* Load the original GDT */
614 	load_direct_gdt(cpu);
615 	/* Reload the per-cpu base */
616 	load_percpu_segment(cpu);
617 }
618 
619 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
620 
621 static void get_model_name(struct cpuinfo_x86 *c)
622 {
623 	unsigned int *v;
624 	char *p, *q, *s;
625 
626 	if (c->extended_cpuid_level < 0x80000004)
627 		return;
628 
629 	v = (unsigned int *)c->x86_model_id;
630 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
631 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
632 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
633 	c->x86_model_id[48] = 0;
634 
635 	/* Trim whitespace */
636 	p = q = s = &c->x86_model_id[0];
637 
638 	while (*p == ' ')
639 		p++;
640 
641 	while (*p) {
642 		/* Note the last non-whitespace index */
643 		if (!isspace(*p))
644 			s = q;
645 
646 		*q++ = *p++;
647 	}
648 
649 	*(s + 1) = '\0';
650 }
651 
652 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
653 {
654 	unsigned int eax, ebx, ecx, edx;
655 
656 	c->x86_max_cores = 1;
657 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
658 		return;
659 
660 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
661 	if (eax & 0x1f)
662 		c->x86_max_cores = (eax >> 26) + 1;
663 }
664 
665 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
666 {
667 	unsigned int n, dummy, ebx, ecx, edx, l2size;
668 
669 	n = c->extended_cpuid_level;
670 
671 	if (n >= 0x80000005) {
672 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
673 		c->x86_cache_size = (ecx>>24) + (edx>>24);
674 #ifdef CONFIG_X86_64
675 		/* On K8 L1 TLB is inclusive, so don't count it */
676 		c->x86_tlbsize = 0;
677 #endif
678 	}
679 
680 	if (n < 0x80000006)	/* Some chips just has a large L1. */
681 		return;
682 
683 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
684 	l2size = ecx >> 16;
685 
686 #ifdef CONFIG_X86_64
687 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
688 #else
689 	/* do processor-specific cache resizing */
690 	if (this_cpu->legacy_cache_size)
691 		l2size = this_cpu->legacy_cache_size(c, l2size);
692 
693 	/* Allow user to override all this if necessary. */
694 	if (cachesize_override != -1)
695 		l2size = cachesize_override;
696 
697 	if (l2size == 0)
698 		return;		/* Again, no L2 cache is possible */
699 #endif
700 
701 	c->x86_cache_size = l2size;
702 }
703 
704 u16 __read_mostly tlb_lli_4k[NR_INFO];
705 u16 __read_mostly tlb_lli_2m[NR_INFO];
706 u16 __read_mostly tlb_lli_4m[NR_INFO];
707 u16 __read_mostly tlb_lld_4k[NR_INFO];
708 u16 __read_mostly tlb_lld_2m[NR_INFO];
709 u16 __read_mostly tlb_lld_4m[NR_INFO];
710 u16 __read_mostly tlb_lld_1g[NR_INFO];
711 
712 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
713 {
714 	if (this_cpu->c_detect_tlb)
715 		this_cpu->c_detect_tlb(c);
716 
717 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
718 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
719 		tlb_lli_4m[ENTRIES]);
720 
721 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
722 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
723 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
724 }
725 
726 int detect_ht_early(struct cpuinfo_x86 *c)
727 {
728 #ifdef CONFIG_SMP
729 	u32 eax, ebx, ecx, edx;
730 
731 	if (!cpu_has(c, X86_FEATURE_HT))
732 		return -1;
733 
734 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
735 		return -1;
736 
737 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
738 		return -1;
739 
740 	cpuid(1, &eax, &ebx, &ecx, &edx);
741 
742 	smp_num_siblings = (ebx & 0xff0000) >> 16;
743 	if (smp_num_siblings == 1)
744 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
745 #endif
746 	return 0;
747 }
748 
749 void detect_ht(struct cpuinfo_x86 *c)
750 {
751 #ifdef CONFIG_SMP
752 	int index_msb, core_bits;
753 
754 	if (detect_ht_early(c) < 0)
755 		return;
756 
757 	index_msb = get_count_order(smp_num_siblings);
758 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
759 
760 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
761 
762 	index_msb = get_count_order(smp_num_siblings);
763 
764 	core_bits = get_count_order(c->x86_max_cores);
765 
766 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
767 				       ((1 << core_bits) - 1);
768 #endif
769 }
770 
771 static void get_cpu_vendor(struct cpuinfo_x86 *c)
772 {
773 	char *v = c->x86_vendor_id;
774 	int i;
775 
776 	for (i = 0; i < X86_VENDOR_NUM; i++) {
777 		if (!cpu_devs[i])
778 			break;
779 
780 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
781 		    (cpu_devs[i]->c_ident[1] &&
782 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
783 
784 			this_cpu = cpu_devs[i];
785 			c->x86_vendor = this_cpu->c_x86_vendor;
786 			return;
787 		}
788 	}
789 
790 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
791 		    "CPU: Your system may be unstable.\n", v);
792 
793 	c->x86_vendor = X86_VENDOR_UNKNOWN;
794 	this_cpu = &default_cpu;
795 }
796 
797 void cpu_detect(struct cpuinfo_x86 *c)
798 {
799 	/* Get vendor name */
800 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
801 	      (unsigned int *)&c->x86_vendor_id[0],
802 	      (unsigned int *)&c->x86_vendor_id[8],
803 	      (unsigned int *)&c->x86_vendor_id[4]);
804 
805 	c->x86 = 4;
806 	/* Intel-defined flags: level 0x00000001 */
807 	if (c->cpuid_level >= 0x00000001) {
808 		u32 junk, tfms, cap0, misc;
809 
810 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
811 		c->x86		= x86_family(tfms);
812 		c->x86_model	= x86_model(tfms);
813 		c->x86_stepping	= x86_stepping(tfms);
814 
815 		if (cap0 & (1<<19)) {
816 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
817 			c->x86_cache_alignment = c->x86_clflush_size;
818 		}
819 	}
820 }
821 
822 static void apply_forced_caps(struct cpuinfo_x86 *c)
823 {
824 	int i;
825 
826 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
827 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
828 		c->x86_capability[i] |= cpu_caps_set[i];
829 	}
830 }
831 
832 static void init_speculation_control(struct cpuinfo_x86 *c)
833 {
834 	/*
835 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
836 	 * and they also have a different bit for STIBP support. Also,
837 	 * a hypervisor might have set the individual AMD bits even on
838 	 * Intel CPUs, for finer-grained selection of what's available.
839 	 */
840 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
841 		set_cpu_cap(c, X86_FEATURE_IBRS);
842 		set_cpu_cap(c, X86_FEATURE_IBPB);
843 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
844 	}
845 
846 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
847 		set_cpu_cap(c, X86_FEATURE_STIBP);
848 
849 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
850 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
851 		set_cpu_cap(c, X86_FEATURE_SSBD);
852 
853 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
854 		set_cpu_cap(c, X86_FEATURE_IBRS);
855 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
856 	}
857 
858 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
859 		set_cpu_cap(c, X86_FEATURE_IBPB);
860 
861 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
862 		set_cpu_cap(c, X86_FEATURE_STIBP);
863 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
864 	}
865 
866 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
867 		set_cpu_cap(c, X86_FEATURE_SSBD);
868 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
869 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
870 	}
871 }
872 
873 static void init_cqm(struct cpuinfo_x86 *c)
874 {
875 	if (!cpu_has(c, X86_FEATURE_CQM_LLC)) {
876 		c->x86_cache_max_rmid  = -1;
877 		c->x86_cache_occ_scale = -1;
878 		return;
879 	}
880 
881 	/* will be overridden if occupancy monitoring exists */
882 	c->x86_cache_max_rmid = cpuid_ebx(0xf);
883 
884 	if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC) ||
885 	    cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL) ||
886 	    cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)) {
887 		u32 eax, ebx, ecx, edx;
888 
889 		/* QoS sub-leaf, EAX=0Fh, ECX=1 */
890 		cpuid_count(0xf, 1, &eax, &ebx, &ecx, &edx);
891 
892 		c->x86_cache_max_rmid  = ecx;
893 		c->x86_cache_occ_scale = ebx;
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 	init_cqm(c);
965 
966 	/*
967 	 * Clear/Set all flags overridden by options, after probe.
968 	 * This needs to happen each time we re-probe, which may happen
969 	 * several times during CPU initialization.
970 	 */
971 	apply_forced_caps(c);
972 }
973 
974 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
975 {
976 	u32 eax, ebx, ecx, edx;
977 
978 	if (c->extended_cpuid_level >= 0x80000008) {
979 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
980 
981 		c->x86_virt_bits = (eax >> 8) & 0xff;
982 		c->x86_phys_bits = eax & 0xff;
983 	}
984 #ifdef CONFIG_X86_32
985 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
986 		c->x86_phys_bits = 36;
987 #endif
988 	c->x86_cache_bits = c->x86_phys_bits;
989 }
990 
991 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
992 {
993 #ifdef CONFIG_X86_32
994 	int i;
995 
996 	/*
997 	 * First of all, decide if this is a 486 or higher
998 	 * It's a 486 if we can modify the AC flag
999 	 */
1000 	if (flag_is_changeable_p(X86_EFLAGS_AC))
1001 		c->x86 = 4;
1002 	else
1003 		c->x86 = 3;
1004 
1005 	for (i = 0; i < X86_VENDOR_NUM; i++)
1006 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
1007 			c->x86_vendor_id[0] = 0;
1008 			cpu_devs[i]->c_identify(c);
1009 			if (c->x86_vendor_id[0]) {
1010 				get_cpu_vendor(c);
1011 				break;
1012 			}
1013 		}
1014 #endif
1015 }
1016 
1017 #define NO_SPECULATION		BIT(0)
1018 #define NO_MELTDOWN		BIT(1)
1019 #define NO_SSB			BIT(2)
1020 #define NO_L1TF			BIT(3)
1021 #define NO_MDS			BIT(4)
1022 #define MSBDS_ONLY		BIT(5)
1023 #define NO_SWAPGS		BIT(6)
1024 #define NO_ITLB_MULTIHIT	BIT(7)
1025 
1026 #define VULNWL(_vendor, _family, _model, _whitelist)	\
1027 	{ X86_VENDOR_##_vendor, _family, _model, X86_FEATURE_ANY, _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 };
1088 
1089 static bool __init cpu_matches(unsigned long which)
1090 {
1091 	const struct x86_cpu_id *m = x86_match_cpu(cpu_vuln_whitelist);
1092 
1093 	return m && !!(m->driver_data & which);
1094 }
1095 
1096 u64 x86_read_arch_cap_msr(void)
1097 {
1098 	u64 ia32_cap = 0;
1099 
1100 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1101 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1102 
1103 	return ia32_cap;
1104 }
1105 
1106 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1107 {
1108 	u64 ia32_cap = x86_read_arch_cap_msr();
1109 
1110 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1111 	if (!cpu_matches(NO_ITLB_MULTIHIT) && !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO))
1112 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1113 
1114 	if (cpu_matches(NO_SPECULATION))
1115 		return;
1116 
1117 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1118 	setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1119 
1120 	if (!cpu_matches(NO_SSB) && !(ia32_cap & ARCH_CAP_SSB_NO) &&
1121 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1122 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1123 
1124 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1125 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1126 
1127 	if (!cpu_matches(NO_MDS) && !(ia32_cap & ARCH_CAP_MDS_NO)) {
1128 		setup_force_cpu_bug(X86_BUG_MDS);
1129 		if (cpu_matches(MSBDS_ONLY))
1130 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1131 	}
1132 
1133 	if (!cpu_matches(NO_SWAPGS))
1134 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1135 
1136 	/*
1137 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1138 	 *	- TSX is supported or
1139 	 *	- TSX_CTRL is present
1140 	 *
1141 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1142 	 * the kernel boot e.g. kexec.
1143 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1144 	 * update is not present or running as guest that don't get TSX_CTRL.
1145 	 */
1146 	if (!(ia32_cap & ARCH_CAP_TAA_NO) &&
1147 	    (cpu_has(c, X86_FEATURE_RTM) ||
1148 	     (ia32_cap & ARCH_CAP_TSX_CTRL_MSR)))
1149 		setup_force_cpu_bug(X86_BUG_TAA);
1150 
1151 	if (cpu_matches(NO_MELTDOWN))
1152 		return;
1153 
1154 	/* Rogue Data Cache Load? No! */
1155 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1156 		return;
1157 
1158 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1159 
1160 	if (cpu_matches(NO_L1TF))
1161 		return;
1162 
1163 	setup_force_cpu_bug(X86_BUG_L1TF);
1164 }
1165 
1166 /*
1167  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1168  * unfortunately, that's not true in practice because of early VIA
1169  * chips and (more importantly) broken virtualizers that are not easy
1170  * to detect. In the latter case it doesn't even *fail* reliably, so
1171  * probing for it doesn't even work. Disable it completely on 32-bit
1172  * unless we can find a reliable way to detect all the broken cases.
1173  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1174  */
1175 static void detect_nopl(void)
1176 {
1177 #ifdef CONFIG_X86_32
1178 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1179 #else
1180 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1181 #endif
1182 }
1183 
1184 /*
1185  * Do minimum CPU detection early.
1186  * Fields really needed: vendor, cpuid_level, family, model, mask,
1187  * cache alignment.
1188  * The others are not touched to avoid unwanted side effects.
1189  *
1190  * WARNING: this function is only called on the boot CPU.  Don't add code
1191  * here that is supposed to run on all CPUs.
1192  */
1193 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1194 {
1195 #ifdef CONFIG_X86_64
1196 	c->x86_clflush_size = 64;
1197 	c->x86_phys_bits = 36;
1198 	c->x86_virt_bits = 48;
1199 #else
1200 	c->x86_clflush_size = 32;
1201 	c->x86_phys_bits = 32;
1202 	c->x86_virt_bits = 32;
1203 #endif
1204 	c->x86_cache_alignment = c->x86_clflush_size;
1205 
1206 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1207 	c->extended_cpuid_level = 0;
1208 
1209 	if (!have_cpuid_p())
1210 		identify_cpu_without_cpuid(c);
1211 
1212 	/* cyrix could have cpuid enabled via c_identify()*/
1213 	if (have_cpuid_p()) {
1214 		cpu_detect(c);
1215 		get_cpu_vendor(c);
1216 		get_cpu_cap(c);
1217 		get_cpu_address_sizes(c);
1218 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1219 
1220 		if (this_cpu->c_early_init)
1221 			this_cpu->c_early_init(c);
1222 
1223 		c->cpu_index = 0;
1224 		filter_cpuid_features(c, false);
1225 
1226 		if (this_cpu->c_bsp_init)
1227 			this_cpu->c_bsp_init(c);
1228 	} else {
1229 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1230 	}
1231 
1232 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1233 
1234 	cpu_set_bug_bits(c);
1235 
1236 	fpu__init_system(c);
1237 
1238 #ifdef CONFIG_X86_32
1239 	/*
1240 	 * Regardless of whether PCID is enumerated, the SDM says
1241 	 * that it can't be enabled in 32-bit mode.
1242 	 */
1243 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1244 #endif
1245 
1246 	/*
1247 	 * Later in the boot process pgtable_l5_enabled() relies on
1248 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1249 	 * enabled by this point we need to clear the feature bit to avoid
1250 	 * false-positives at the later stage.
1251 	 *
1252 	 * pgtable_l5_enabled() can be false here for several reasons:
1253 	 *  - 5-level paging is disabled compile-time;
1254 	 *  - it's 32-bit kernel;
1255 	 *  - machine doesn't support 5-level paging;
1256 	 *  - user specified 'no5lvl' in kernel command line.
1257 	 */
1258 	if (!pgtable_l5_enabled())
1259 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1260 
1261 	detect_nopl();
1262 }
1263 
1264 void __init early_cpu_init(void)
1265 {
1266 	const struct cpu_dev *const *cdev;
1267 	int count = 0;
1268 
1269 #ifdef CONFIG_PROCESSOR_SELECT
1270 	pr_info("KERNEL supported cpus:\n");
1271 #endif
1272 
1273 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1274 		const struct cpu_dev *cpudev = *cdev;
1275 
1276 		if (count >= X86_VENDOR_NUM)
1277 			break;
1278 		cpu_devs[count] = cpudev;
1279 		count++;
1280 
1281 #ifdef CONFIG_PROCESSOR_SELECT
1282 		{
1283 			unsigned int j;
1284 
1285 			for (j = 0; j < 2; j++) {
1286 				if (!cpudev->c_ident[j])
1287 					continue;
1288 				pr_info("  %s %s\n", cpudev->c_vendor,
1289 					cpudev->c_ident[j]);
1290 			}
1291 		}
1292 #endif
1293 	}
1294 	early_identify_cpu(&boot_cpu_data);
1295 }
1296 
1297 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1298 {
1299 #ifdef CONFIG_X86_64
1300 	/*
1301 	 * Empirically, writing zero to a segment selector on AMD does
1302 	 * not clear the base, whereas writing zero to a segment
1303 	 * selector on Intel does clear the base.  Intel's behavior
1304 	 * allows slightly faster context switches in the common case
1305 	 * where GS is unused by the prev and next threads.
1306 	 *
1307 	 * Since neither vendor documents this anywhere that I can see,
1308 	 * detect it directly instead of hardcoding the choice by
1309 	 * vendor.
1310 	 *
1311 	 * I've designated AMD's behavior as the "bug" because it's
1312 	 * counterintuitive and less friendly.
1313 	 */
1314 
1315 	unsigned long old_base, tmp;
1316 	rdmsrl(MSR_FS_BASE, old_base);
1317 	wrmsrl(MSR_FS_BASE, 1);
1318 	loadsegment(fs, 0);
1319 	rdmsrl(MSR_FS_BASE, tmp);
1320 	if (tmp != 0)
1321 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1322 	wrmsrl(MSR_FS_BASE, old_base);
1323 #endif
1324 }
1325 
1326 static void generic_identify(struct cpuinfo_x86 *c)
1327 {
1328 	c->extended_cpuid_level = 0;
1329 
1330 	if (!have_cpuid_p())
1331 		identify_cpu_without_cpuid(c);
1332 
1333 	/* cyrix could have cpuid enabled via c_identify()*/
1334 	if (!have_cpuid_p())
1335 		return;
1336 
1337 	cpu_detect(c);
1338 
1339 	get_cpu_vendor(c);
1340 
1341 	get_cpu_cap(c);
1342 
1343 	get_cpu_address_sizes(c);
1344 
1345 	if (c->cpuid_level >= 0x00000001) {
1346 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1347 #ifdef CONFIG_X86_32
1348 # ifdef CONFIG_SMP
1349 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1350 # else
1351 		c->apicid = c->initial_apicid;
1352 # endif
1353 #endif
1354 		c->phys_proc_id = c->initial_apicid;
1355 	}
1356 
1357 	get_model_name(c); /* Default name */
1358 
1359 	detect_null_seg_behavior(c);
1360 
1361 	/*
1362 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1363 	 * systems that run Linux at CPL > 0 may or may not have the
1364 	 * issue, but, even if they have the issue, there's absolutely
1365 	 * nothing we can do about it because we can't use the real IRET
1366 	 * instruction.
1367 	 *
1368 	 * NB: For the time being, only 32-bit kernels support
1369 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1370 	 * whether to apply espfix using paravirt hooks.  If any
1371 	 * non-paravirt system ever shows up that does *not* have the
1372 	 * ESPFIX issue, we can change this.
1373 	 */
1374 #ifdef CONFIG_X86_32
1375 # ifdef CONFIG_PARAVIRT_XXL
1376 	do {
1377 		extern void native_iret(void);
1378 		if (pv_ops.cpu.iret == native_iret)
1379 			set_cpu_bug(c, X86_BUG_ESPFIX);
1380 	} while (0);
1381 # else
1382 	set_cpu_bug(c, X86_BUG_ESPFIX);
1383 # endif
1384 #endif
1385 }
1386 
1387 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
1388 {
1389 	/*
1390 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
1391 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
1392 	 * in case CQM bits really aren't there in this CPU.
1393 	 */
1394 	if (c != &boot_cpu_data) {
1395 		boot_cpu_data.x86_cache_max_rmid =
1396 			min(boot_cpu_data.x86_cache_max_rmid,
1397 			    c->x86_cache_max_rmid);
1398 	}
1399 }
1400 
1401 /*
1402  * Validate that ACPI/mptables have the same information about the
1403  * effective APIC id and update the package map.
1404  */
1405 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1406 {
1407 #ifdef CONFIG_SMP
1408 	unsigned int apicid, cpu = smp_processor_id();
1409 
1410 	apicid = apic->cpu_present_to_apicid(cpu);
1411 
1412 	if (apicid != c->apicid) {
1413 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1414 		       cpu, apicid, c->initial_apicid);
1415 	}
1416 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1417 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1418 #else
1419 	c->logical_proc_id = 0;
1420 #endif
1421 }
1422 
1423 /*
1424  * This does the hard work of actually picking apart the CPU stuff...
1425  */
1426 static void identify_cpu(struct cpuinfo_x86 *c)
1427 {
1428 	int i;
1429 
1430 	c->loops_per_jiffy = loops_per_jiffy;
1431 	c->x86_cache_size = 0;
1432 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1433 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1434 	c->x86_vendor_id[0] = '\0'; /* Unset */
1435 	c->x86_model_id[0] = '\0';  /* Unset */
1436 	c->x86_max_cores = 1;
1437 	c->x86_coreid_bits = 0;
1438 	c->cu_id = 0xff;
1439 #ifdef CONFIG_X86_64
1440 	c->x86_clflush_size = 64;
1441 	c->x86_phys_bits = 36;
1442 	c->x86_virt_bits = 48;
1443 #else
1444 	c->cpuid_level = -1;	/* CPUID not detected */
1445 	c->x86_clflush_size = 32;
1446 	c->x86_phys_bits = 32;
1447 	c->x86_virt_bits = 32;
1448 #endif
1449 	c->x86_cache_alignment = c->x86_clflush_size;
1450 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1451 
1452 	generic_identify(c);
1453 
1454 	if (this_cpu->c_identify)
1455 		this_cpu->c_identify(c);
1456 
1457 	/* Clear/Set all flags overridden by options, after probe */
1458 	apply_forced_caps(c);
1459 
1460 #ifdef CONFIG_X86_64
1461 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1462 #endif
1463 
1464 	/*
1465 	 * Vendor-specific initialization.  In this section we
1466 	 * canonicalize the feature flags, meaning if there are
1467 	 * features a certain CPU supports which CPUID doesn't
1468 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1469 	 * we handle them here.
1470 	 *
1471 	 * At the end of this section, c->x86_capability better
1472 	 * indicate the features this CPU genuinely supports!
1473 	 */
1474 	if (this_cpu->c_init)
1475 		this_cpu->c_init(c);
1476 
1477 	/* Disable the PN if appropriate */
1478 	squash_the_stupid_serial_number(c);
1479 
1480 	/* Set up SMEP/SMAP/UMIP */
1481 	setup_smep(c);
1482 	setup_smap(c);
1483 	setup_umip(c);
1484 
1485 	/*
1486 	 * The vendor-specific functions might have changed features.
1487 	 * Now we do "generic changes."
1488 	 */
1489 
1490 	/* Filter out anything that depends on CPUID levels we don't have */
1491 	filter_cpuid_features(c, true);
1492 
1493 	/* If the model name is still unset, do table lookup. */
1494 	if (!c->x86_model_id[0]) {
1495 		const char *p;
1496 		p = table_lookup_model(c);
1497 		if (p)
1498 			strcpy(c->x86_model_id, p);
1499 		else
1500 			/* Last resort... */
1501 			sprintf(c->x86_model_id, "%02x/%02x",
1502 				c->x86, c->x86_model);
1503 	}
1504 
1505 #ifdef CONFIG_X86_64
1506 	detect_ht(c);
1507 #endif
1508 
1509 	x86_init_rdrand(c);
1510 	x86_init_cache_qos(c);
1511 	setup_pku(c);
1512 
1513 	/*
1514 	 * Clear/Set all flags overridden by options, need do it
1515 	 * before following smp all cpus cap AND.
1516 	 */
1517 	apply_forced_caps(c);
1518 
1519 	/*
1520 	 * On SMP, boot_cpu_data holds the common feature set between
1521 	 * all CPUs; so make sure that we indicate which features are
1522 	 * common between the CPUs.  The first time this routine gets
1523 	 * executed, c == &boot_cpu_data.
1524 	 */
1525 	if (c != &boot_cpu_data) {
1526 		/* AND the already accumulated flags with these */
1527 		for (i = 0; i < NCAPINTS; i++)
1528 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1529 
1530 		/* OR, i.e. replicate the bug flags */
1531 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1532 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1533 	}
1534 
1535 	/* Init Machine Check Exception if available. */
1536 	mcheck_cpu_init(c);
1537 
1538 	select_idle_routine(c);
1539 
1540 #ifdef CONFIG_NUMA
1541 	numa_add_cpu(smp_processor_id());
1542 #endif
1543 }
1544 
1545 /*
1546  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1547  * on 32-bit kernels:
1548  */
1549 #ifdef CONFIG_X86_32
1550 void enable_sep_cpu(void)
1551 {
1552 	struct tss_struct *tss;
1553 	int cpu;
1554 
1555 	if (!boot_cpu_has(X86_FEATURE_SEP))
1556 		return;
1557 
1558 	cpu = get_cpu();
1559 	tss = &per_cpu(cpu_tss_rw, cpu);
1560 
1561 	/*
1562 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1563 	 * see the big comment in struct x86_hw_tss's definition.
1564 	 */
1565 
1566 	tss->x86_tss.ss1 = __KERNEL_CS;
1567 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1568 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1569 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1570 
1571 	put_cpu();
1572 }
1573 #endif
1574 
1575 void __init identify_boot_cpu(void)
1576 {
1577 	identify_cpu(&boot_cpu_data);
1578 #ifdef CONFIG_X86_32
1579 	sysenter_setup();
1580 	enable_sep_cpu();
1581 #endif
1582 	cpu_detect_tlb(&boot_cpu_data);
1583 	setup_cr_pinning();
1584 
1585 	tsx_init();
1586 }
1587 
1588 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1589 {
1590 	BUG_ON(c == &boot_cpu_data);
1591 	identify_cpu(c);
1592 #ifdef CONFIG_X86_32
1593 	enable_sep_cpu();
1594 #endif
1595 	mtrr_ap_init();
1596 	validate_apic_and_package_id(c);
1597 	x86_spec_ctrl_setup_ap();
1598 }
1599 
1600 static __init int setup_noclflush(char *arg)
1601 {
1602 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1603 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1604 	return 1;
1605 }
1606 __setup("noclflush", setup_noclflush);
1607 
1608 void print_cpu_info(struct cpuinfo_x86 *c)
1609 {
1610 	const char *vendor = NULL;
1611 
1612 	if (c->x86_vendor < X86_VENDOR_NUM) {
1613 		vendor = this_cpu->c_vendor;
1614 	} else {
1615 		if (c->cpuid_level >= 0)
1616 			vendor = c->x86_vendor_id;
1617 	}
1618 
1619 	if (vendor && !strstr(c->x86_model_id, vendor))
1620 		pr_cont("%s ", vendor);
1621 
1622 	if (c->x86_model_id[0])
1623 		pr_cont("%s", c->x86_model_id);
1624 	else
1625 		pr_cont("%d86", c->x86);
1626 
1627 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1628 
1629 	if (c->x86_stepping || c->cpuid_level >= 0)
1630 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1631 	else
1632 		pr_cont(")\n");
1633 }
1634 
1635 /*
1636  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1637  * But we need to keep a dummy __setup around otherwise it would
1638  * show up as an environment variable for init.
1639  */
1640 static __init int setup_clearcpuid(char *arg)
1641 {
1642 	return 1;
1643 }
1644 __setup("clearcpuid=", setup_clearcpuid);
1645 
1646 #ifdef CONFIG_X86_64
1647 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1648 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1649 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1650 
1651 /*
1652  * The following percpu variables are hot.  Align current_task to
1653  * cacheline size such that they fall in the same cacheline.
1654  */
1655 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1656 	&init_task;
1657 EXPORT_PER_CPU_SYMBOL(current_task);
1658 
1659 DEFINE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
1660 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1661 
1662 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1663 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1664 
1665 /* May not be marked __init: used by software suspend */
1666 void syscall_init(void)
1667 {
1668 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1669 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1670 
1671 #ifdef CONFIG_IA32_EMULATION
1672 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1673 	/*
1674 	 * This only works on Intel CPUs.
1675 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1676 	 * This does not cause SYSENTER to jump to the wrong location, because
1677 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1678 	 */
1679 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1680 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1681 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1682 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1683 #else
1684 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1685 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1686 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1687 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1688 #endif
1689 
1690 	/* Flags to clear on syscall */
1691 	wrmsrl(MSR_SYSCALL_MASK,
1692 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1693 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1694 }
1695 
1696 DEFINE_PER_CPU(int, debug_stack_usage);
1697 DEFINE_PER_CPU(u32, debug_idt_ctr);
1698 
1699 void debug_stack_set_zero(void)
1700 {
1701 	this_cpu_inc(debug_idt_ctr);
1702 	load_current_idt();
1703 }
1704 NOKPROBE_SYMBOL(debug_stack_set_zero);
1705 
1706 void debug_stack_reset(void)
1707 {
1708 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1709 		return;
1710 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1711 		load_current_idt();
1712 }
1713 NOKPROBE_SYMBOL(debug_stack_reset);
1714 
1715 #else	/* CONFIG_X86_64 */
1716 
1717 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1718 EXPORT_PER_CPU_SYMBOL(current_task);
1719 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1720 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1721 
1722 /*
1723  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1724  * the top of the kernel stack.  Use an extra percpu variable to track the
1725  * top of the kernel stack directly.
1726  */
1727 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1728 	(unsigned long)&init_thread_union + THREAD_SIZE;
1729 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1730 
1731 #ifdef CONFIG_STACKPROTECTOR
1732 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1733 #endif
1734 
1735 #endif	/* CONFIG_X86_64 */
1736 
1737 /*
1738  * Clear all 6 debug registers:
1739  */
1740 static void clear_all_debug_regs(void)
1741 {
1742 	int i;
1743 
1744 	for (i = 0; i < 8; i++) {
1745 		/* Ignore db4, db5 */
1746 		if ((i == 4) || (i == 5))
1747 			continue;
1748 
1749 		set_debugreg(0, i);
1750 	}
1751 }
1752 
1753 #ifdef CONFIG_KGDB
1754 /*
1755  * Restore debug regs if using kgdbwait and you have a kernel debugger
1756  * connection established.
1757  */
1758 static void dbg_restore_debug_regs(void)
1759 {
1760 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1761 		arch_kgdb_ops.correct_hw_break();
1762 }
1763 #else /* ! CONFIG_KGDB */
1764 #define dbg_restore_debug_regs()
1765 #endif /* ! CONFIG_KGDB */
1766 
1767 static void wait_for_master_cpu(int cpu)
1768 {
1769 #ifdef CONFIG_SMP
1770 	/*
1771 	 * wait for ACK from master CPU before continuing
1772 	 * with AP initialization
1773 	 */
1774 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1775 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1776 		cpu_relax();
1777 #endif
1778 }
1779 
1780 #ifdef CONFIG_X86_64
1781 static inline void setup_getcpu(int cpu)
1782 {
1783 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1784 	struct desc_struct d = { };
1785 
1786 	if (boot_cpu_has(X86_FEATURE_RDTSCP))
1787 		write_rdtscp_aux(cpudata);
1788 
1789 	/* Store CPU and node number in limit. */
1790 	d.limit0 = cpudata;
1791 	d.limit1 = cpudata >> 16;
1792 
1793 	d.type = 5;		/* RO data, expand down, accessed */
1794 	d.dpl = 3;		/* Visible to user code */
1795 	d.s = 1;		/* Not a system segment */
1796 	d.p = 1;		/* Present */
1797 	d.d = 1;		/* 32-bit */
1798 
1799 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1800 }
1801 
1802 static inline void ucode_cpu_init(int cpu)
1803 {
1804 	if (cpu)
1805 		load_ucode_ap();
1806 }
1807 
1808 static inline void tss_setup_ist(struct tss_struct *tss)
1809 {
1810 	/* Set up the per-CPU TSS IST stacks */
1811 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1812 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1813 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1814 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1815 }
1816 
1817 static inline void gdt_setup_doublefault_tss(int cpu) { }
1818 
1819 #else /* CONFIG_X86_64 */
1820 
1821 static inline void setup_getcpu(int cpu) { }
1822 
1823 static inline void ucode_cpu_init(int cpu)
1824 {
1825 	show_ucode_info_early();
1826 }
1827 
1828 static inline void tss_setup_ist(struct tss_struct *tss) { }
1829 
1830 static inline void gdt_setup_doublefault_tss(int cpu)
1831 {
1832 #ifdef CONFIG_DOUBLEFAULT
1833 	/* Set up the doublefault TSS pointer in the GDT */
1834 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1835 #endif
1836 }
1837 #endif /* !CONFIG_X86_64 */
1838 
1839 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
1840 {
1841 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
1842 
1843 #ifdef CONFIG_X86_IOPL_IOPERM
1844 	tss->io_bitmap.prev_max = 0;
1845 	tss->io_bitmap.prev_sequence = 0;
1846 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
1847 	/*
1848 	 * Invalidate the extra array entry past the end of the all
1849 	 * permission bitmap as required by the hardware.
1850 	 */
1851 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
1852 #endif
1853 }
1854 
1855 /*
1856  * cpu_init() initializes state that is per-CPU. Some data is already
1857  * initialized (naturally) in the bootstrap process, such as the GDT
1858  * and IDT. We reload them nevertheless, this function acts as a
1859  * 'CPU state barrier', nothing should get across.
1860  */
1861 void cpu_init(void)
1862 {
1863 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
1864 	struct task_struct *cur = current;
1865 	int cpu = raw_smp_processor_id();
1866 
1867 	wait_for_master_cpu(cpu);
1868 
1869 	ucode_cpu_init(cpu);
1870 
1871 #ifdef CONFIG_NUMA
1872 	if (this_cpu_read(numa_node) == 0 &&
1873 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1874 		set_numa_node(early_cpu_to_node(cpu));
1875 #endif
1876 	setup_getcpu(cpu);
1877 
1878 	pr_debug("Initializing CPU#%d\n", cpu);
1879 
1880 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
1881 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
1882 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1883 
1884 	/*
1885 	 * Initialize the per-CPU GDT with the boot GDT,
1886 	 * and set up the GDT descriptor:
1887 	 */
1888 	switch_to_new_gdt(cpu);
1889 	load_current_idt();
1890 
1891 	if (IS_ENABLED(CONFIG_X86_64)) {
1892 		loadsegment(fs, 0);
1893 		memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1894 		syscall_init();
1895 
1896 		wrmsrl(MSR_FS_BASE, 0);
1897 		wrmsrl(MSR_KERNEL_GS_BASE, 0);
1898 		barrier();
1899 
1900 		x2apic_setup();
1901 	}
1902 
1903 	mmgrab(&init_mm);
1904 	cur->active_mm = &init_mm;
1905 	BUG_ON(cur->mm);
1906 	initialize_tlbstate_and_flush();
1907 	enter_lazy_tlb(&init_mm, cur);
1908 
1909 	/* Initialize the TSS. */
1910 	tss_setup_ist(tss);
1911 	tss_setup_io_bitmap(tss);
1912 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1913 
1914 	load_TR_desc();
1915 	/*
1916 	 * sp0 points to the entry trampoline stack regardless of what task
1917 	 * is running.
1918 	 */
1919 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1920 
1921 	load_mm_ldt(&init_mm);
1922 
1923 	clear_all_debug_regs();
1924 	dbg_restore_debug_regs();
1925 
1926 	gdt_setup_doublefault_tss(cpu);
1927 
1928 	fpu__init_cpu();
1929 
1930 	if (is_uv_system())
1931 		uv_cpu_init();
1932 
1933 	load_fixmap_gdt(cpu);
1934 }
1935 
1936 /*
1937  * The microcode loader calls this upon late microcode load to recheck features,
1938  * only when microcode has been updated. Caller holds microcode_mutex and CPU
1939  * hotplug lock.
1940  */
1941 void microcode_check(void)
1942 {
1943 	struct cpuinfo_x86 info;
1944 
1945 	perf_check_microcode();
1946 
1947 	/* Reload CPUID max function as it might've changed. */
1948 	info.cpuid_level = cpuid_eax(0);
1949 
1950 	/*
1951 	 * Copy all capability leafs to pick up the synthetic ones so that
1952 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
1953 	 * get overwritten in get_cpu_cap().
1954 	 */
1955 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
1956 
1957 	get_cpu_cap(&info);
1958 
1959 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
1960 		return;
1961 
1962 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
1963 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
1964 }
1965 
1966 /*
1967  * Invoked from core CPU hotplug code after hotplug operations
1968  */
1969 void arch_smt_update(void)
1970 {
1971 	/* Handle the speculative execution misfeatures */
1972 	cpu_bugs_smt_update();
1973 	/* Check whether IPI broadcasting can be enabled */
1974 	apic_smt_update();
1975 }
1976