xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 9bd5910d)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* cpu_feature_enabled() cannot be used this early */
3 #define USE_EARLY_PGTABLE_L5
4 
5 #include <linux/memblock.h>
6 #include <linux/linkage.h>
7 #include <linux/bitops.h>
8 #include <linux/kernel.h>
9 #include <linux/export.h>
10 #include <linux/percpu.h>
11 #include <linux/string.h>
12 #include <linux/ctype.h>
13 #include <linux/delay.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/clock.h>
16 #include <linux/sched/task.h>
17 #include <linux/sched/smt.h>
18 #include <linux/init.h>
19 #include <linux/kprobes.h>
20 #include <linux/kgdb.h>
21 #include <linux/smp.h>
22 #include <linux/io.h>
23 #include <linux/syscore_ops.h>
24 
25 #include <asm/stackprotector.h>
26 #include <asm/perf_event.h>
27 #include <asm/mmu_context.h>
28 #include <asm/doublefault.h>
29 #include <asm/archrandom.h>
30 #include <asm/hypervisor.h>
31 #include <asm/processor.h>
32 #include <asm/tlbflush.h>
33 #include <asm/debugreg.h>
34 #include <asm/sections.h>
35 #include <asm/vsyscall.h>
36 #include <linux/topology.h>
37 #include <linux/cpumask.h>
38 #include <asm/pgtable.h>
39 #include <linux/atomic.h>
40 #include <asm/proto.h>
41 #include <asm/setup.h>
42 #include <asm/apic.h>
43 #include <asm/desc.h>
44 #include <asm/fpu/internal.h>
45 #include <asm/mtrr.h>
46 #include <asm/hwcap2.h>
47 #include <linux/numa.h>
48 #include <asm/asm.h>
49 #include <asm/bugs.h>
50 #include <asm/cpu.h>
51 #include <asm/mce.h>
52 #include <asm/msr.h>
53 #include <asm/memtype.h>
54 #include <asm/microcode.h>
55 #include <asm/microcode_intel.h>
56 #include <asm/intel-family.h>
57 #include <asm/cpu_device_id.h>
58 #include <asm/uv/uv.h>
59 
60 #include "cpu.h"
61 
62 u32 elf_hwcap2 __read_mostly;
63 
64 /* all of these masks are initialized in setup_cpu_local_masks() */
65 cpumask_var_t cpu_initialized_mask;
66 cpumask_var_t cpu_callout_mask;
67 cpumask_var_t cpu_callin_mask;
68 
69 /* representing cpus for which sibling maps can be computed */
70 cpumask_var_t cpu_sibling_setup_mask;
71 
72 /* Number of siblings per CPU package */
73 int smp_num_siblings = 1;
74 EXPORT_SYMBOL(smp_num_siblings);
75 
76 /* Last level cache ID of each logical CPU */
77 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
78 
79 /* correctly size the local cpu masks */
80 void __init setup_cpu_local_masks(void)
81 {
82 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
83 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
84 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
85 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
86 }
87 
88 static void default_init(struct cpuinfo_x86 *c)
89 {
90 #ifdef CONFIG_X86_64
91 	cpu_detect_cache_sizes(c);
92 #else
93 	/* Not much we can do here... */
94 	/* Check if at least it has cpuid */
95 	if (c->cpuid_level == -1) {
96 		/* No cpuid. It must be an ancient CPU */
97 		if (c->x86 == 4)
98 			strcpy(c->x86_model_id, "486");
99 		else if (c->x86 == 3)
100 			strcpy(c->x86_model_id, "386");
101 	}
102 #endif
103 }
104 
105 static const struct cpu_dev default_cpu = {
106 	.c_init		= default_init,
107 	.c_vendor	= "Unknown",
108 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
109 };
110 
111 static const struct cpu_dev *this_cpu = &default_cpu;
112 
113 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
114 #ifdef CONFIG_X86_64
115 	/*
116 	 * We need valid kernel segments for data and code in long mode too
117 	 * IRET will check the segment types  kkeil 2000/10/28
118 	 * Also sysret mandates a special GDT layout
119 	 *
120 	 * TLS descriptors are currently at a different place compared to i386.
121 	 * Hopefully nobody expects them at a fixed place (Wine?)
122 	 */
123 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
124 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
125 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
126 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
127 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
128 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
129 #else
130 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
131 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
132 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
133 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
134 	/*
135 	 * Segments used for calling PnP BIOS have byte granularity.
136 	 * They code segments and data segments have fixed 64k limits,
137 	 * the transfer segment sizes are set at run time.
138 	 */
139 	/* 32-bit code */
140 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
141 	/* 16-bit code */
142 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
143 	/* 16-bit data */
144 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
145 	/* 16-bit data */
146 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
147 	/* 16-bit data */
148 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
149 	/*
150 	 * The APM segments have byte granularity and their bases
151 	 * are set at run time.  All have 64k limits.
152 	 */
153 	/* 32-bit code */
154 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
155 	/* 16-bit code */
156 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
157 	/* data */
158 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
159 
160 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
161 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
162 	GDT_STACK_CANARY_INIT
163 #endif
164 } };
165 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
166 
167 #ifdef CONFIG_X86_64
168 static int __init x86_nopcid_setup(char *s)
169 {
170 	/* nopcid doesn't accept parameters */
171 	if (s)
172 		return -EINVAL;
173 
174 	/* do not emit a message if the feature is not present */
175 	if (!boot_cpu_has(X86_FEATURE_PCID))
176 		return 0;
177 
178 	setup_clear_cpu_cap(X86_FEATURE_PCID);
179 	pr_info("nopcid: PCID feature disabled\n");
180 	return 0;
181 }
182 early_param("nopcid", x86_nopcid_setup);
183 #endif
184 
185 static int __init x86_noinvpcid_setup(char *s)
186 {
187 	/* noinvpcid doesn't accept parameters */
188 	if (s)
189 		return -EINVAL;
190 
191 	/* do not emit a message if the feature is not present */
192 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
193 		return 0;
194 
195 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
196 	pr_info("noinvpcid: INVPCID feature disabled\n");
197 	return 0;
198 }
199 early_param("noinvpcid", x86_noinvpcid_setup);
200 
201 #ifdef CONFIG_X86_32
202 static int cachesize_override = -1;
203 static int disable_x86_serial_nr = 1;
204 
205 static int __init cachesize_setup(char *str)
206 {
207 	get_option(&str, &cachesize_override);
208 	return 1;
209 }
210 __setup("cachesize=", cachesize_setup);
211 
212 static int __init x86_sep_setup(char *s)
213 {
214 	setup_clear_cpu_cap(X86_FEATURE_SEP);
215 	return 1;
216 }
217 __setup("nosep", x86_sep_setup);
218 
219 /* Standard macro to see if a specific flag is changeable */
220 static inline int flag_is_changeable_p(u32 flag)
221 {
222 	u32 f1, f2;
223 
224 	/*
225 	 * Cyrix and IDT cpus allow disabling of CPUID
226 	 * so the code below may return different results
227 	 * when it is executed before and after enabling
228 	 * the CPUID. Add "volatile" to not allow gcc to
229 	 * optimize the subsequent calls to this function.
230 	 */
231 	asm volatile ("pushfl		\n\t"
232 		      "pushfl		\n\t"
233 		      "popl %0		\n\t"
234 		      "movl %0, %1	\n\t"
235 		      "xorl %2, %0	\n\t"
236 		      "pushl %0		\n\t"
237 		      "popfl		\n\t"
238 		      "pushfl		\n\t"
239 		      "popl %0		\n\t"
240 		      "popfl		\n\t"
241 
242 		      : "=&r" (f1), "=&r" (f2)
243 		      : "ir" (flag));
244 
245 	return ((f1^f2) & flag) != 0;
246 }
247 
248 /* Probe for the CPUID instruction */
249 int have_cpuid_p(void)
250 {
251 	return flag_is_changeable_p(X86_EFLAGS_ID);
252 }
253 
254 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
255 {
256 	unsigned long lo, hi;
257 
258 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
259 		return;
260 
261 	/* Disable processor serial number: */
262 
263 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
264 	lo |= 0x200000;
265 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
266 
267 	pr_notice("CPU serial number disabled.\n");
268 	clear_cpu_cap(c, X86_FEATURE_PN);
269 
270 	/* Disabling the serial number may affect the cpuid level */
271 	c->cpuid_level = cpuid_eax(0);
272 }
273 
274 static int __init x86_serial_nr_setup(char *s)
275 {
276 	disable_x86_serial_nr = 0;
277 	return 1;
278 }
279 __setup("serialnumber", x86_serial_nr_setup);
280 #else
281 static inline int flag_is_changeable_p(u32 flag)
282 {
283 	return 1;
284 }
285 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
286 {
287 }
288 #endif
289 
290 static __init int setup_disable_smep(char *arg)
291 {
292 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
293 	return 1;
294 }
295 __setup("nosmep", setup_disable_smep);
296 
297 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
298 {
299 	if (cpu_has(c, X86_FEATURE_SMEP))
300 		cr4_set_bits(X86_CR4_SMEP);
301 }
302 
303 static __init int setup_disable_smap(char *arg)
304 {
305 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
306 	return 1;
307 }
308 __setup("nosmap", setup_disable_smap);
309 
310 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
311 {
312 	unsigned long eflags = native_save_fl();
313 
314 	/* This should have been cleared long ago */
315 	BUG_ON(eflags & X86_EFLAGS_AC);
316 
317 	if (cpu_has(c, X86_FEATURE_SMAP)) {
318 #ifdef CONFIG_X86_SMAP
319 		cr4_set_bits(X86_CR4_SMAP);
320 #else
321 		cr4_clear_bits(X86_CR4_SMAP);
322 #endif
323 	}
324 }
325 
326 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
327 {
328 	/* Check the boot processor, plus build option for UMIP. */
329 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
330 		goto out;
331 
332 	/* Check the current processor's cpuid bits. */
333 	if (!cpu_has(c, X86_FEATURE_UMIP))
334 		goto out;
335 
336 	cr4_set_bits(X86_CR4_UMIP);
337 
338 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
339 
340 	return;
341 
342 out:
343 	/*
344 	 * Make sure UMIP is disabled in case it was enabled in a
345 	 * previous boot (e.g., via kexec).
346 	 */
347 	cr4_clear_bits(X86_CR4_UMIP);
348 }
349 
350 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
351 static unsigned long cr4_pinned_bits __ro_after_init;
352 
353 void native_write_cr0(unsigned long val)
354 {
355 	unsigned long bits_missing = 0;
356 
357 set_register:
358 	asm volatile("mov %0,%%cr0": "+r" (val), "+m" (__force_order));
359 
360 	if (static_branch_likely(&cr_pinning)) {
361 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
362 			bits_missing = X86_CR0_WP;
363 			val |= bits_missing;
364 			goto set_register;
365 		}
366 		/* Warn after we've set the missing bits. */
367 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
368 	}
369 }
370 EXPORT_SYMBOL(native_write_cr0);
371 
372 void native_write_cr4(unsigned long val)
373 {
374 	unsigned long bits_missing = 0;
375 
376 set_register:
377 	asm volatile("mov %0,%%cr4": "+r" (val), "+m" (cr4_pinned_bits));
378 
379 	if (static_branch_likely(&cr_pinning)) {
380 		if (unlikely((val & cr4_pinned_bits) != cr4_pinned_bits)) {
381 			bits_missing = ~val & cr4_pinned_bits;
382 			val |= bits_missing;
383 			goto set_register;
384 		}
385 		/* Warn after we've set the missing bits. */
386 		WARN_ONCE(bits_missing, "CR4 bits went missing: %lx!?\n",
387 			  bits_missing);
388 	}
389 }
390 EXPORT_SYMBOL(native_write_cr4);
391 
392 void cr4_init(void)
393 {
394 	unsigned long cr4 = __read_cr4();
395 
396 	if (boot_cpu_has(X86_FEATURE_PCID))
397 		cr4 |= X86_CR4_PCIDE;
398 	if (static_branch_likely(&cr_pinning))
399 		cr4 |= cr4_pinned_bits;
400 
401 	__write_cr4(cr4);
402 
403 	/* Initialize cr4 shadow for this CPU. */
404 	this_cpu_write(cpu_tlbstate.cr4, cr4);
405 }
406 
407 /*
408  * Once CPU feature detection is finished (and boot params have been
409  * parsed), record any of the sensitive CR bits that are set, and
410  * enable CR pinning.
411  */
412 static void __init setup_cr_pinning(void)
413 {
414 	unsigned long mask;
415 
416 	mask = (X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP);
417 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & mask;
418 	static_key_enable(&cr_pinning.key);
419 }
420 
421 /*
422  * Protection Keys are not available in 32-bit mode.
423  */
424 static bool pku_disabled;
425 
426 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
427 {
428 	struct pkru_state *pk;
429 
430 	/* check the boot processor, plus compile options for PKU: */
431 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
432 		return;
433 	/* checks the actual processor's cpuid bits: */
434 	if (!cpu_has(c, X86_FEATURE_PKU))
435 		return;
436 	if (pku_disabled)
437 		return;
438 
439 	cr4_set_bits(X86_CR4_PKE);
440 	pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU);
441 	if (pk)
442 		pk->pkru = init_pkru_value;
443 	/*
444 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
445 	 * cpuid bit to be set.  We need to ensure that we
446 	 * update that bit in this CPU's "cpu_info".
447 	 */
448 	get_cpu_cap(c);
449 }
450 
451 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
452 static __init int setup_disable_pku(char *arg)
453 {
454 	/*
455 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
456 	 * runtime checks are against OSPKE so clearing the
457 	 * bit does nothing.
458 	 *
459 	 * This way, we will see "pku" in cpuinfo, but not
460 	 * "ospke", which is exactly what we want.  It shows
461 	 * that the CPU has PKU, but the OS has not enabled it.
462 	 * This happens to be exactly how a system would look
463 	 * if we disabled the config option.
464 	 */
465 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
466 	pku_disabled = true;
467 	return 1;
468 }
469 __setup("nopku", setup_disable_pku);
470 #endif /* CONFIG_X86_64 */
471 
472 /*
473  * Some CPU features depend on higher CPUID levels, which may not always
474  * be available due to CPUID level capping or broken virtualization
475  * software.  Add those features to this table to auto-disable them.
476  */
477 struct cpuid_dependent_feature {
478 	u32 feature;
479 	u32 level;
480 };
481 
482 static const struct cpuid_dependent_feature
483 cpuid_dependent_features[] = {
484 	{ X86_FEATURE_MWAIT,		0x00000005 },
485 	{ X86_FEATURE_DCA,		0x00000009 },
486 	{ X86_FEATURE_XSAVE,		0x0000000d },
487 	{ 0, 0 }
488 };
489 
490 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
491 {
492 	const struct cpuid_dependent_feature *df;
493 
494 	for (df = cpuid_dependent_features; df->feature; df++) {
495 
496 		if (!cpu_has(c, df->feature))
497 			continue;
498 		/*
499 		 * Note: cpuid_level is set to -1 if unavailable, but
500 		 * extended_extended_level is set to 0 if unavailable
501 		 * and the legitimate extended levels are all negative
502 		 * when signed; hence the weird messing around with
503 		 * signs here...
504 		 */
505 		if (!((s32)df->level < 0 ?
506 		     (u32)df->level > (u32)c->extended_cpuid_level :
507 		     (s32)df->level > (s32)c->cpuid_level))
508 			continue;
509 
510 		clear_cpu_cap(c, df->feature);
511 		if (!warn)
512 			continue;
513 
514 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
515 			x86_cap_flag(df->feature), df->level);
516 	}
517 }
518 
519 /*
520  * Naming convention should be: <Name> [(<Codename>)]
521  * This table only is used unless init_<vendor>() below doesn't set it;
522  * in particular, if CPUID levels 0x80000002..4 are supported, this
523  * isn't used
524  */
525 
526 /* Look up CPU names by table lookup. */
527 static const char *table_lookup_model(struct cpuinfo_x86 *c)
528 {
529 #ifdef CONFIG_X86_32
530 	const struct legacy_cpu_model_info *info;
531 
532 	if (c->x86_model >= 16)
533 		return NULL;	/* Range check */
534 
535 	if (!this_cpu)
536 		return NULL;
537 
538 	info = this_cpu->legacy_models;
539 
540 	while (info->family) {
541 		if (info->family == c->x86)
542 			return info->model_names[c->x86_model];
543 		info++;
544 	}
545 #endif
546 	return NULL;		/* Not found */
547 }
548 
549 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
550 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
551 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
552 
553 void load_percpu_segment(int cpu)
554 {
555 #ifdef CONFIG_X86_32
556 	loadsegment(fs, __KERNEL_PERCPU);
557 #else
558 	__loadsegment_simple(gs, 0);
559 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
560 #endif
561 	load_stack_canary_segment();
562 }
563 
564 #ifdef CONFIG_X86_32
565 /* The 32-bit entry code needs to find cpu_entry_area. */
566 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
567 #endif
568 
569 /* Load the original GDT from the per-cpu structure */
570 void load_direct_gdt(int cpu)
571 {
572 	struct desc_ptr gdt_descr;
573 
574 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
575 	gdt_descr.size = GDT_SIZE - 1;
576 	load_gdt(&gdt_descr);
577 }
578 EXPORT_SYMBOL_GPL(load_direct_gdt);
579 
580 /* Load a fixmap remapping of the per-cpu GDT */
581 void load_fixmap_gdt(int cpu)
582 {
583 	struct desc_ptr gdt_descr;
584 
585 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
586 	gdt_descr.size = GDT_SIZE - 1;
587 	load_gdt(&gdt_descr);
588 }
589 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
590 
591 /*
592  * Current gdt points %fs at the "master" per-cpu area: after this,
593  * it's on the real one.
594  */
595 void switch_to_new_gdt(int cpu)
596 {
597 	/* Load the original GDT */
598 	load_direct_gdt(cpu);
599 	/* Reload the per-cpu base */
600 	load_percpu_segment(cpu);
601 }
602 
603 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
604 
605 static void get_model_name(struct cpuinfo_x86 *c)
606 {
607 	unsigned int *v;
608 	char *p, *q, *s;
609 
610 	if (c->extended_cpuid_level < 0x80000004)
611 		return;
612 
613 	v = (unsigned int *)c->x86_model_id;
614 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
615 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
616 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
617 	c->x86_model_id[48] = 0;
618 
619 	/* Trim whitespace */
620 	p = q = s = &c->x86_model_id[0];
621 
622 	while (*p == ' ')
623 		p++;
624 
625 	while (*p) {
626 		/* Note the last non-whitespace index */
627 		if (!isspace(*p))
628 			s = q;
629 
630 		*q++ = *p++;
631 	}
632 
633 	*(s + 1) = '\0';
634 }
635 
636 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
637 {
638 	unsigned int eax, ebx, ecx, edx;
639 
640 	c->x86_max_cores = 1;
641 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
642 		return;
643 
644 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
645 	if (eax & 0x1f)
646 		c->x86_max_cores = (eax >> 26) + 1;
647 }
648 
649 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
650 {
651 	unsigned int n, dummy, ebx, ecx, edx, l2size;
652 
653 	n = c->extended_cpuid_level;
654 
655 	if (n >= 0x80000005) {
656 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
657 		c->x86_cache_size = (ecx>>24) + (edx>>24);
658 #ifdef CONFIG_X86_64
659 		/* On K8 L1 TLB is inclusive, so don't count it */
660 		c->x86_tlbsize = 0;
661 #endif
662 	}
663 
664 	if (n < 0x80000006)	/* Some chips just has a large L1. */
665 		return;
666 
667 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
668 	l2size = ecx >> 16;
669 
670 #ifdef CONFIG_X86_64
671 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
672 #else
673 	/* do processor-specific cache resizing */
674 	if (this_cpu->legacy_cache_size)
675 		l2size = this_cpu->legacy_cache_size(c, l2size);
676 
677 	/* Allow user to override all this if necessary. */
678 	if (cachesize_override != -1)
679 		l2size = cachesize_override;
680 
681 	if (l2size == 0)
682 		return;		/* Again, no L2 cache is possible */
683 #endif
684 
685 	c->x86_cache_size = l2size;
686 }
687 
688 u16 __read_mostly tlb_lli_4k[NR_INFO];
689 u16 __read_mostly tlb_lli_2m[NR_INFO];
690 u16 __read_mostly tlb_lli_4m[NR_INFO];
691 u16 __read_mostly tlb_lld_4k[NR_INFO];
692 u16 __read_mostly tlb_lld_2m[NR_INFO];
693 u16 __read_mostly tlb_lld_4m[NR_INFO];
694 u16 __read_mostly tlb_lld_1g[NR_INFO];
695 
696 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
697 {
698 	if (this_cpu->c_detect_tlb)
699 		this_cpu->c_detect_tlb(c);
700 
701 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
702 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
703 		tlb_lli_4m[ENTRIES]);
704 
705 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
706 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
707 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
708 }
709 
710 int detect_ht_early(struct cpuinfo_x86 *c)
711 {
712 #ifdef CONFIG_SMP
713 	u32 eax, ebx, ecx, edx;
714 
715 	if (!cpu_has(c, X86_FEATURE_HT))
716 		return -1;
717 
718 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
719 		return -1;
720 
721 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
722 		return -1;
723 
724 	cpuid(1, &eax, &ebx, &ecx, &edx);
725 
726 	smp_num_siblings = (ebx & 0xff0000) >> 16;
727 	if (smp_num_siblings == 1)
728 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
729 #endif
730 	return 0;
731 }
732 
733 void detect_ht(struct cpuinfo_x86 *c)
734 {
735 #ifdef CONFIG_SMP
736 	int index_msb, core_bits;
737 
738 	if (detect_ht_early(c) < 0)
739 		return;
740 
741 	index_msb = get_count_order(smp_num_siblings);
742 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
743 
744 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
745 
746 	index_msb = get_count_order(smp_num_siblings);
747 
748 	core_bits = get_count_order(c->x86_max_cores);
749 
750 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
751 				       ((1 << core_bits) - 1);
752 #endif
753 }
754 
755 static void get_cpu_vendor(struct cpuinfo_x86 *c)
756 {
757 	char *v = c->x86_vendor_id;
758 	int i;
759 
760 	for (i = 0; i < X86_VENDOR_NUM; i++) {
761 		if (!cpu_devs[i])
762 			break;
763 
764 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
765 		    (cpu_devs[i]->c_ident[1] &&
766 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
767 
768 			this_cpu = cpu_devs[i];
769 			c->x86_vendor = this_cpu->c_x86_vendor;
770 			return;
771 		}
772 	}
773 
774 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
775 		    "CPU: Your system may be unstable.\n", v);
776 
777 	c->x86_vendor = X86_VENDOR_UNKNOWN;
778 	this_cpu = &default_cpu;
779 }
780 
781 void cpu_detect(struct cpuinfo_x86 *c)
782 {
783 	/* Get vendor name */
784 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
785 	      (unsigned int *)&c->x86_vendor_id[0],
786 	      (unsigned int *)&c->x86_vendor_id[8],
787 	      (unsigned int *)&c->x86_vendor_id[4]);
788 
789 	c->x86 = 4;
790 	/* Intel-defined flags: level 0x00000001 */
791 	if (c->cpuid_level >= 0x00000001) {
792 		u32 junk, tfms, cap0, misc;
793 
794 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
795 		c->x86		= x86_family(tfms);
796 		c->x86_model	= x86_model(tfms);
797 		c->x86_stepping	= x86_stepping(tfms);
798 
799 		if (cap0 & (1<<19)) {
800 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
801 			c->x86_cache_alignment = c->x86_clflush_size;
802 		}
803 	}
804 }
805 
806 static void apply_forced_caps(struct cpuinfo_x86 *c)
807 {
808 	int i;
809 
810 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
811 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
812 		c->x86_capability[i] |= cpu_caps_set[i];
813 	}
814 }
815 
816 static void init_speculation_control(struct cpuinfo_x86 *c)
817 {
818 	/*
819 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
820 	 * and they also have a different bit for STIBP support. Also,
821 	 * a hypervisor might have set the individual AMD bits even on
822 	 * Intel CPUs, for finer-grained selection of what's available.
823 	 */
824 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
825 		set_cpu_cap(c, X86_FEATURE_IBRS);
826 		set_cpu_cap(c, X86_FEATURE_IBPB);
827 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
828 	}
829 
830 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
831 		set_cpu_cap(c, X86_FEATURE_STIBP);
832 
833 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
834 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
835 		set_cpu_cap(c, X86_FEATURE_SSBD);
836 
837 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
838 		set_cpu_cap(c, X86_FEATURE_IBRS);
839 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
840 	}
841 
842 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
843 		set_cpu_cap(c, X86_FEATURE_IBPB);
844 
845 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
846 		set_cpu_cap(c, X86_FEATURE_STIBP);
847 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
848 	}
849 
850 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
851 		set_cpu_cap(c, X86_FEATURE_SSBD);
852 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
853 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
854 	}
855 }
856 
857 static void init_cqm(struct cpuinfo_x86 *c)
858 {
859 	if (!cpu_has(c, X86_FEATURE_CQM_LLC)) {
860 		c->x86_cache_max_rmid  = -1;
861 		c->x86_cache_occ_scale = -1;
862 		return;
863 	}
864 
865 	/* will be overridden if occupancy monitoring exists */
866 	c->x86_cache_max_rmid = cpuid_ebx(0xf);
867 
868 	if (cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC) ||
869 	    cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL) ||
870 	    cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)) {
871 		u32 eax, ebx, ecx, edx;
872 
873 		/* QoS sub-leaf, EAX=0Fh, ECX=1 */
874 		cpuid_count(0xf, 1, &eax, &ebx, &ecx, &edx);
875 
876 		c->x86_cache_max_rmid  = ecx;
877 		c->x86_cache_occ_scale = ebx;
878 	}
879 }
880 
881 void get_cpu_cap(struct cpuinfo_x86 *c)
882 {
883 	u32 eax, ebx, ecx, edx;
884 
885 	/* Intel-defined flags: level 0x00000001 */
886 	if (c->cpuid_level >= 0x00000001) {
887 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
888 
889 		c->x86_capability[CPUID_1_ECX] = ecx;
890 		c->x86_capability[CPUID_1_EDX] = edx;
891 	}
892 
893 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
894 	if (c->cpuid_level >= 0x00000006)
895 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
896 
897 	/* Additional Intel-defined flags: level 0x00000007 */
898 	if (c->cpuid_level >= 0x00000007) {
899 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
900 		c->x86_capability[CPUID_7_0_EBX] = ebx;
901 		c->x86_capability[CPUID_7_ECX] = ecx;
902 		c->x86_capability[CPUID_7_EDX] = edx;
903 
904 		/* Check valid sub-leaf index before accessing it */
905 		if (eax >= 1) {
906 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
907 			c->x86_capability[CPUID_7_1_EAX] = eax;
908 		}
909 	}
910 
911 	/* Extended state features: level 0x0000000d */
912 	if (c->cpuid_level >= 0x0000000d) {
913 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
914 
915 		c->x86_capability[CPUID_D_1_EAX] = eax;
916 	}
917 
918 	/* AMD-defined flags: level 0x80000001 */
919 	eax = cpuid_eax(0x80000000);
920 	c->extended_cpuid_level = eax;
921 
922 	if ((eax & 0xffff0000) == 0x80000000) {
923 		if (eax >= 0x80000001) {
924 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
925 
926 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
927 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
928 		}
929 	}
930 
931 	if (c->extended_cpuid_level >= 0x80000007) {
932 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
933 
934 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
935 		c->x86_power = edx;
936 	}
937 
938 	if (c->extended_cpuid_level >= 0x80000008) {
939 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
940 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
941 	}
942 
943 	if (c->extended_cpuid_level >= 0x8000000a)
944 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
945 
946 	init_scattered_cpuid_features(c);
947 	init_speculation_control(c);
948 	init_cqm(c);
949 
950 	/*
951 	 * Clear/Set all flags overridden by options, after probe.
952 	 * This needs to happen each time we re-probe, which may happen
953 	 * several times during CPU initialization.
954 	 */
955 	apply_forced_caps(c);
956 }
957 
958 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
959 {
960 	u32 eax, ebx, ecx, edx;
961 
962 	if (c->extended_cpuid_level >= 0x80000008) {
963 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
964 
965 		c->x86_virt_bits = (eax >> 8) & 0xff;
966 		c->x86_phys_bits = eax & 0xff;
967 	}
968 #ifdef CONFIG_X86_32
969 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
970 		c->x86_phys_bits = 36;
971 #endif
972 	c->x86_cache_bits = c->x86_phys_bits;
973 }
974 
975 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
976 {
977 #ifdef CONFIG_X86_32
978 	int i;
979 
980 	/*
981 	 * First of all, decide if this is a 486 or higher
982 	 * It's a 486 if we can modify the AC flag
983 	 */
984 	if (flag_is_changeable_p(X86_EFLAGS_AC))
985 		c->x86 = 4;
986 	else
987 		c->x86 = 3;
988 
989 	for (i = 0; i < X86_VENDOR_NUM; i++)
990 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
991 			c->x86_vendor_id[0] = 0;
992 			cpu_devs[i]->c_identify(c);
993 			if (c->x86_vendor_id[0]) {
994 				get_cpu_vendor(c);
995 				break;
996 			}
997 		}
998 #endif
999 }
1000 
1001 #define NO_SPECULATION		BIT(0)
1002 #define NO_MELTDOWN		BIT(1)
1003 #define NO_SSB			BIT(2)
1004 #define NO_L1TF			BIT(3)
1005 #define NO_MDS			BIT(4)
1006 #define MSBDS_ONLY		BIT(5)
1007 #define NO_SWAPGS		BIT(6)
1008 #define NO_ITLB_MULTIHIT	BIT(7)
1009 #define NO_SPECTRE_V2		BIT(8)
1010 
1011 #define VULNWL(_vendor, _family, _model, _whitelist)	\
1012 	{ X86_VENDOR_##_vendor, _family, _model, X86_FEATURE_ANY, _whitelist }
1013 
1014 #define VULNWL_INTEL(model, whitelist)		\
1015 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1016 
1017 #define VULNWL_AMD(family, whitelist)		\
1018 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1019 
1020 #define VULNWL_HYGON(family, whitelist)		\
1021 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1022 
1023 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1024 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1025 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1026 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1027 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1028 
1029 	/* Intel Family 6 */
1030 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1031 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION | NO_ITLB_MULTIHIT),
1032 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1033 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1034 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1035 
1036 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1037 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1038 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1039 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1040 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1041 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1042 
1043 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1044 
1045 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1046 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1047 
1048 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1049 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1050 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1051 
1052 	/*
1053 	 * Technically, swapgs isn't serializing on AMD (despite it previously
1054 	 * being documented as such in the APM).  But according to AMD, %gs is
1055 	 * updated non-speculatively, and the issuing of %gs-relative memory
1056 	 * operands will be blocked until the %gs update completes, which is
1057 	 * good enough for our purposes.
1058 	 */
1059 
1060 	VULNWL_INTEL(ATOM_TREMONT_D,		NO_ITLB_MULTIHIT),
1061 
1062 	/* AMD Family 0xf - 0x12 */
1063 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1064 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1065 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1066 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1067 
1068 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1069 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1070 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1071 
1072 	/* Zhaoxin Family 7 */
1073 	VULNWL(CENTAUR,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1074 	VULNWL(ZHAOXIN,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1075 	{}
1076 };
1077 
1078 static bool __init cpu_matches(unsigned long which)
1079 {
1080 	const struct x86_cpu_id *m = x86_match_cpu(cpu_vuln_whitelist);
1081 
1082 	return m && !!(m->driver_data & which);
1083 }
1084 
1085 u64 x86_read_arch_cap_msr(void)
1086 {
1087 	u64 ia32_cap = 0;
1088 
1089 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1090 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1091 
1092 	return ia32_cap;
1093 }
1094 
1095 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1096 {
1097 	u64 ia32_cap = x86_read_arch_cap_msr();
1098 
1099 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1100 	if (!cpu_matches(NO_ITLB_MULTIHIT) && !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO))
1101 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1102 
1103 	if (cpu_matches(NO_SPECULATION))
1104 		return;
1105 
1106 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1107 
1108 	if (!cpu_matches(NO_SPECTRE_V2))
1109 		setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1110 
1111 	if (!cpu_matches(NO_SSB) && !(ia32_cap & ARCH_CAP_SSB_NO) &&
1112 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1113 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1114 
1115 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1116 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1117 
1118 	if (!cpu_matches(NO_MDS) && !(ia32_cap & ARCH_CAP_MDS_NO)) {
1119 		setup_force_cpu_bug(X86_BUG_MDS);
1120 		if (cpu_matches(MSBDS_ONLY))
1121 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1122 	}
1123 
1124 	if (!cpu_matches(NO_SWAPGS))
1125 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1126 
1127 	/*
1128 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1129 	 *	- TSX is supported or
1130 	 *	- TSX_CTRL is present
1131 	 *
1132 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1133 	 * the kernel boot e.g. kexec.
1134 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1135 	 * update is not present or running as guest that don't get TSX_CTRL.
1136 	 */
1137 	if (!(ia32_cap & ARCH_CAP_TAA_NO) &&
1138 	    (cpu_has(c, X86_FEATURE_RTM) ||
1139 	     (ia32_cap & ARCH_CAP_TSX_CTRL_MSR)))
1140 		setup_force_cpu_bug(X86_BUG_TAA);
1141 
1142 	if (cpu_matches(NO_MELTDOWN))
1143 		return;
1144 
1145 	/* Rogue Data Cache Load? No! */
1146 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1147 		return;
1148 
1149 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1150 
1151 	if (cpu_matches(NO_L1TF))
1152 		return;
1153 
1154 	setup_force_cpu_bug(X86_BUG_L1TF);
1155 }
1156 
1157 /*
1158  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1159  * unfortunately, that's not true in practice because of early VIA
1160  * chips and (more importantly) broken virtualizers that are not easy
1161  * to detect. In the latter case it doesn't even *fail* reliably, so
1162  * probing for it doesn't even work. Disable it completely on 32-bit
1163  * unless we can find a reliable way to detect all the broken cases.
1164  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1165  */
1166 static void detect_nopl(void)
1167 {
1168 #ifdef CONFIG_X86_32
1169 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1170 #else
1171 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1172 #endif
1173 }
1174 
1175 /*
1176  * Do minimum CPU detection early.
1177  * Fields really needed: vendor, cpuid_level, family, model, mask,
1178  * cache alignment.
1179  * The others are not touched to avoid unwanted side effects.
1180  *
1181  * WARNING: this function is only called on the boot CPU.  Don't add code
1182  * here that is supposed to run on all CPUs.
1183  */
1184 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1185 {
1186 #ifdef CONFIG_X86_64
1187 	c->x86_clflush_size = 64;
1188 	c->x86_phys_bits = 36;
1189 	c->x86_virt_bits = 48;
1190 #else
1191 	c->x86_clflush_size = 32;
1192 	c->x86_phys_bits = 32;
1193 	c->x86_virt_bits = 32;
1194 #endif
1195 	c->x86_cache_alignment = c->x86_clflush_size;
1196 
1197 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1198 	c->extended_cpuid_level = 0;
1199 
1200 	if (!have_cpuid_p())
1201 		identify_cpu_without_cpuid(c);
1202 
1203 	/* cyrix could have cpuid enabled via c_identify()*/
1204 	if (have_cpuid_p()) {
1205 		cpu_detect(c);
1206 		get_cpu_vendor(c);
1207 		get_cpu_cap(c);
1208 		get_cpu_address_sizes(c);
1209 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1210 
1211 		if (this_cpu->c_early_init)
1212 			this_cpu->c_early_init(c);
1213 
1214 		c->cpu_index = 0;
1215 		filter_cpuid_features(c, false);
1216 
1217 		if (this_cpu->c_bsp_init)
1218 			this_cpu->c_bsp_init(c);
1219 	} else {
1220 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1221 	}
1222 
1223 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1224 
1225 	cpu_set_bug_bits(c);
1226 
1227 	fpu__init_system(c);
1228 
1229 #ifdef CONFIG_X86_32
1230 	/*
1231 	 * Regardless of whether PCID is enumerated, the SDM says
1232 	 * that it can't be enabled in 32-bit mode.
1233 	 */
1234 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1235 #endif
1236 
1237 	/*
1238 	 * Later in the boot process pgtable_l5_enabled() relies on
1239 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1240 	 * enabled by this point we need to clear the feature bit to avoid
1241 	 * false-positives at the later stage.
1242 	 *
1243 	 * pgtable_l5_enabled() can be false here for several reasons:
1244 	 *  - 5-level paging is disabled compile-time;
1245 	 *  - it's 32-bit kernel;
1246 	 *  - machine doesn't support 5-level paging;
1247 	 *  - user specified 'no5lvl' in kernel command line.
1248 	 */
1249 	if (!pgtable_l5_enabled())
1250 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1251 
1252 	detect_nopl();
1253 }
1254 
1255 void __init early_cpu_init(void)
1256 {
1257 	const struct cpu_dev *const *cdev;
1258 	int count = 0;
1259 
1260 #ifdef CONFIG_PROCESSOR_SELECT
1261 	pr_info("KERNEL supported cpus:\n");
1262 #endif
1263 
1264 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1265 		const struct cpu_dev *cpudev = *cdev;
1266 
1267 		if (count >= X86_VENDOR_NUM)
1268 			break;
1269 		cpu_devs[count] = cpudev;
1270 		count++;
1271 
1272 #ifdef CONFIG_PROCESSOR_SELECT
1273 		{
1274 			unsigned int j;
1275 
1276 			for (j = 0; j < 2; j++) {
1277 				if (!cpudev->c_ident[j])
1278 					continue;
1279 				pr_info("  %s %s\n", cpudev->c_vendor,
1280 					cpudev->c_ident[j]);
1281 			}
1282 		}
1283 #endif
1284 	}
1285 	early_identify_cpu(&boot_cpu_data);
1286 }
1287 
1288 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1289 {
1290 #ifdef CONFIG_X86_64
1291 	/*
1292 	 * Empirically, writing zero to a segment selector on AMD does
1293 	 * not clear the base, whereas writing zero to a segment
1294 	 * selector on Intel does clear the base.  Intel's behavior
1295 	 * allows slightly faster context switches in the common case
1296 	 * where GS is unused by the prev and next threads.
1297 	 *
1298 	 * Since neither vendor documents this anywhere that I can see,
1299 	 * detect it directly instead of hardcoding the choice by
1300 	 * vendor.
1301 	 *
1302 	 * I've designated AMD's behavior as the "bug" because it's
1303 	 * counterintuitive and less friendly.
1304 	 */
1305 
1306 	unsigned long old_base, tmp;
1307 	rdmsrl(MSR_FS_BASE, old_base);
1308 	wrmsrl(MSR_FS_BASE, 1);
1309 	loadsegment(fs, 0);
1310 	rdmsrl(MSR_FS_BASE, tmp);
1311 	if (tmp != 0)
1312 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1313 	wrmsrl(MSR_FS_BASE, old_base);
1314 #endif
1315 }
1316 
1317 static void generic_identify(struct cpuinfo_x86 *c)
1318 {
1319 	c->extended_cpuid_level = 0;
1320 
1321 	if (!have_cpuid_p())
1322 		identify_cpu_without_cpuid(c);
1323 
1324 	/* cyrix could have cpuid enabled via c_identify()*/
1325 	if (!have_cpuid_p())
1326 		return;
1327 
1328 	cpu_detect(c);
1329 
1330 	get_cpu_vendor(c);
1331 
1332 	get_cpu_cap(c);
1333 
1334 	get_cpu_address_sizes(c);
1335 
1336 	if (c->cpuid_level >= 0x00000001) {
1337 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1338 #ifdef CONFIG_X86_32
1339 # ifdef CONFIG_SMP
1340 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1341 # else
1342 		c->apicid = c->initial_apicid;
1343 # endif
1344 #endif
1345 		c->phys_proc_id = c->initial_apicid;
1346 	}
1347 
1348 	get_model_name(c); /* Default name */
1349 
1350 	detect_null_seg_behavior(c);
1351 
1352 	/*
1353 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1354 	 * systems that run Linux at CPL > 0 may or may not have the
1355 	 * issue, but, even if they have the issue, there's absolutely
1356 	 * nothing we can do about it because we can't use the real IRET
1357 	 * instruction.
1358 	 *
1359 	 * NB: For the time being, only 32-bit kernels support
1360 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1361 	 * whether to apply espfix using paravirt hooks.  If any
1362 	 * non-paravirt system ever shows up that does *not* have the
1363 	 * ESPFIX issue, we can change this.
1364 	 */
1365 #ifdef CONFIG_X86_32
1366 # ifdef CONFIG_PARAVIRT_XXL
1367 	do {
1368 		extern void native_iret(void);
1369 		if (pv_ops.cpu.iret == native_iret)
1370 			set_cpu_bug(c, X86_BUG_ESPFIX);
1371 	} while (0);
1372 # else
1373 	set_cpu_bug(c, X86_BUG_ESPFIX);
1374 # endif
1375 #endif
1376 }
1377 
1378 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
1379 {
1380 	/*
1381 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
1382 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
1383 	 * in case CQM bits really aren't there in this CPU.
1384 	 */
1385 	if (c != &boot_cpu_data) {
1386 		boot_cpu_data.x86_cache_max_rmid =
1387 			min(boot_cpu_data.x86_cache_max_rmid,
1388 			    c->x86_cache_max_rmid);
1389 	}
1390 }
1391 
1392 /*
1393  * Validate that ACPI/mptables have the same information about the
1394  * effective APIC id and update the package map.
1395  */
1396 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1397 {
1398 #ifdef CONFIG_SMP
1399 	unsigned int apicid, cpu = smp_processor_id();
1400 
1401 	apicid = apic->cpu_present_to_apicid(cpu);
1402 
1403 	if (apicid != c->apicid) {
1404 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1405 		       cpu, apicid, c->initial_apicid);
1406 	}
1407 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1408 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1409 #else
1410 	c->logical_proc_id = 0;
1411 #endif
1412 }
1413 
1414 /*
1415  * This does the hard work of actually picking apart the CPU stuff...
1416  */
1417 static void identify_cpu(struct cpuinfo_x86 *c)
1418 {
1419 	int i;
1420 
1421 	c->loops_per_jiffy = loops_per_jiffy;
1422 	c->x86_cache_size = 0;
1423 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1424 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1425 	c->x86_vendor_id[0] = '\0'; /* Unset */
1426 	c->x86_model_id[0] = '\0';  /* Unset */
1427 	c->x86_max_cores = 1;
1428 	c->x86_coreid_bits = 0;
1429 	c->cu_id = 0xff;
1430 #ifdef CONFIG_X86_64
1431 	c->x86_clflush_size = 64;
1432 	c->x86_phys_bits = 36;
1433 	c->x86_virt_bits = 48;
1434 #else
1435 	c->cpuid_level = -1;	/* CPUID not detected */
1436 	c->x86_clflush_size = 32;
1437 	c->x86_phys_bits = 32;
1438 	c->x86_virt_bits = 32;
1439 #endif
1440 	c->x86_cache_alignment = c->x86_clflush_size;
1441 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1442 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
1443 	memset(&c->vmx_capability, 0, sizeof(c->vmx_capability));
1444 #endif
1445 
1446 	generic_identify(c);
1447 
1448 	if (this_cpu->c_identify)
1449 		this_cpu->c_identify(c);
1450 
1451 	/* Clear/Set all flags overridden by options, after probe */
1452 	apply_forced_caps(c);
1453 
1454 #ifdef CONFIG_X86_64
1455 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1456 #endif
1457 
1458 	/*
1459 	 * Vendor-specific initialization.  In this section we
1460 	 * canonicalize the feature flags, meaning if there are
1461 	 * features a certain CPU supports which CPUID doesn't
1462 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1463 	 * we handle them here.
1464 	 *
1465 	 * At the end of this section, c->x86_capability better
1466 	 * indicate the features this CPU genuinely supports!
1467 	 */
1468 	if (this_cpu->c_init)
1469 		this_cpu->c_init(c);
1470 
1471 	/* Disable the PN if appropriate */
1472 	squash_the_stupid_serial_number(c);
1473 
1474 	/* Set up SMEP/SMAP/UMIP */
1475 	setup_smep(c);
1476 	setup_smap(c);
1477 	setup_umip(c);
1478 
1479 	/*
1480 	 * The vendor-specific functions might have changed features.
1481 	 * Now we do "generic changes."
1482 	 */
1483 
1484 	/* Filter out anything that depends on CPUID levels we don't have */
1485 	filter_cpuid_features(c, true);
1486 
1487 	/* If the model name is still unset, do table lookup. */
1488 	if (!c->x86_model_id[0]) {
1489 		const char *p;
1490 		p = table_lookup_model(c);
1491 		if (p)
1492 			strcpy(c->x86_model_id, p);
1493 		else
1494 			/* Last resort... */
1495 			sprintf(c->x86_model_id, "%02x/%02x",
1496 				c->x86, c->x86_model);
1497 	}
1498 
1499 #ifdef CONFIG_X86_64
1500 	detect_ht(c);
1501 #endif
1502 
1503 	x86_init_rdrand(c);
1504 	x86_init_cache_qos(c);
1505 	setup_pku(c);
1506 
1507 	/*
1508 	 * Clear/Set all flags overridden by options, need do it
1509 	 * before following smp all cpus cap AND.
1510 	 */
1511 	apply_forced_caps(c);
1512 
1513 	/*
1514 	 * On SMP, boot_cpu_data holds the common feature set between
1515 	 * all CPUs; so make sure that we indicate which features are
1516 	 * common between the CPUs.  The first time this routine gets
1517 	 * executed, c == &boot_cpu_data.
1518 	 */
1519 	if (c != &boot_cpu_data) {
1520 		/* AND the already accumulated flags with these */
1521 		for (i = 0; i < NCAPINTS; i++)
1522 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1523 
1524 		/* OR, i.e. replicate the bug flags */
1525 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1526 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1527 	}
1528 
1529 	/* Init Machine Check Exception if available. */
1530 	mcheck_cpu_init(c);
1531 
1532 	select_idle_routine(c);
1533 
1534 #ifdef CONFIG_NUMA
1535 	numa_add_cpu(smp_processor_id());
1536 #endif
1537 }
1538 
1539 /*
1540  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1541  * on 32-bit kernels:
1542  */
1543 #ifdef CONFIG_X86_32
1544 void enable_sep_cpu(void)
1545 {
1546 	struct tss_struct *tss;
1547 	int cpu;
1548 
1549 	if (!boot_cpu_has(X86_FEATURE_SEP))
1550 		return;
1551 
1552 	cpu = get_cpu();
1553 	tss = &per_cpu(cpu_tss_rw, cpu);
1554 
1555 	/*
1556 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1557 	 * see the big comment in struct x86_hw_tss's definition.
1558 	 */
1559 
1560 	tss->x86_tss.ss1 = __KERNEL_CS;
1561 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1562 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1563 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1564 
1565 	put_cpu();
1566 }
1567 #endif
1568 
1569 void __init identify_boot_cpu(void)
1570 {
1571 	identify_cpu(&boot_cpu_data);
1572 #ifdef CONFIG_X86_32
1573 	sysenter_setup();
1574 	enable_sep_cpu();
1575 #endif
1576 	cpu_detect_tlb(&boot_cpu_data);
1577 	setup_cr_pinning();
1578 
1579 	tsx_init();
1580 }
1581 
1582 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1583 {
1584 	BUG_ON(c == &boot_cpu_data);
1585 	identify_cpu(c);
1586 #ifdef CONFIG_X86_32
1587 	enable_sep_cpu();
1588 #endif
1589 	mtrr_ap_init();
1590 	validate_apic_and_package_id(c);
1591 	x86_spec_ctrl_setup_ap();
1592 }
1593 
1594 static __init int setup_noclflush(char *arg)
1595 {
1596 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1597 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1598 	return 1;
1599 }
1600 __setup("noclflush", setup_noclflush);
1601 
1602 void print_cpu_info(struct cpuinfo_x86 *c)
1603 {
1604 	const char *vendor = NULL;
1605 
1606 	if (c->x86_vendor < X86_VENDOR_NUM) {
1607 		vendor = this_cpu->c_vendor;
1608 	} else {
1609 		if (c->cpuid_level >= 0)
1610 			vendor = c->x86_vendor_id;
1611 	}
1612 
1613 	if (vendor && !strstr(c->x86_model_id, vendor))
1614 		pr_cont("%s ", vendor);
1615 
1616 	if (c->x86_model_id[0])
1617 		pr_cont("%s", c->x86_model_id);
1618 	else
1619 		pr_cont("%d86", c->x86);
1620 
1621 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1622 
1623 	if (c->x86_stepping || c->cpuid_level >= 0)
1624 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1625 	else
1626 		pr_cont(")\n");
1627 }
1628 
1629 /*
1630  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1631  * But we need to keep a dummy __setup around otherwise it would
1632  * show up as an environment variable for init.
1633  */
1634 static __init int setup_clearcpuid(char *arg)
1635 {
1636 	return 1;
1637 }
1638 __setup("clearcpuid=", setup_clearcpuid);
1639 
1640 #ifdef CONFIG_X86_64
1641 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1642 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1643 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1644 
1645 /*
1646  * The following percpu variables are hot.  Align current_task to
1647  * cacheline size such that they fall in the same cacheline.
1648  */
1649 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1650 	&init_task;
1651 EXPORT_PER_CPU_SYMBOL(current_task);
1652 
1653 DEFINE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
1654 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1655 
1656 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1657 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1658 
1659 /* May not be marked __init: used by software suspend */
1660 void syscall_init(void)
1661 {
1662 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1663 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1664 
1665 #ifdef CONFIG_IA32_EMULATION
1666 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1667 	/*
1668 	 * This only works on Intel CPUs.
1669 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1670 	 * This does not cause SYSENTER to jump to the wrong location, because
1671 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1672 	 */
1673 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1674 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1675 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1676 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1677 #else
1678 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1679 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1680 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1681 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1682 #endif
1683 
1684 	/* Flags to clear on syscall */
1685 	wrmsrl(MSR_SYSCALL_MASK,
1686 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1687 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1688 }
1689 
1690 DEFINE_PER_CPU(int, debug_stack_usage);
1691 DEFINE_PER_CPU(u32, debug_idt_ctr);
1692 
1693 void debug_stack_set_zero(void)
1694 {
1695 	this_cpu_inc(debug_idt_ctr);
1696 	load_current_idt();
1697 }
1698 NOKPROBE_SYMBOL(debug_stack_set_zero);
1699 
1700 void debug_stack_reset(void)
1701 {
1702 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1703 		return;
1704 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1705 		load_current_idt();
1706 }
1707 NOKPROBE_SYMBOL(debug_stack_reset);
1708 
1709 #else	/* CONFIG_X86_64 */
1710 
1711 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1712 EXPORT_PER_CPU_SYMBOL(current_task);
1713 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1714 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1715 
1716 /*
1717  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1718  * the top of the kernel stack.  Use an extra percpu variable to track the
1719  * top of the kernel stack directly.
1720  */
1721 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1722 	(unsigned long)&init_thread_union + THREAD_SIZE;
1723 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1724 
1725 #ifdef CONFIG_STACKPROTECTOR
1726 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1727 #endif
1728 
1729 #endif	/* CONFIG_X86_64 */
1730 
1731 /*
1732  * Clear all 6 debug registers:
1733  */
1734 static void clear_all_debug_regs(void)
1735 {
1736 	int i;
1737 
1738 	for (i = 0; i < 8; i++) {
1739 		/* Ignore db4, db5 */
1740 		if ((i == 4) || (i == 5))
1741 			continue;
1742 
1743 		set_debugreg(0, i);
1744 	}
1745 }
1746 
1747 #ifdef CONFIG_KGDB
1748 /*
1749  * Restore debug regs if using kgdbwait and you have a kernel debugger
1750  * connection established.
1751  */
1752 static void dbg_restore_debug_regs(void)
1753 {
1754 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1755 		arch_kgdb_ops.correct_hw_break();
1756 }
1757 #else /* ! CONFIG_KGDB */
1758 #define dbg_restore_debug_regs()
1759 #endif /* ! CONFIG_KGDB */
1760 
1761 static void wait_for_master_cpu(int cpu)
1762 {
1763 #ifdef CONFIG_SMP
1764 	/*
1765 	 * wait for ACK from master CPU before continuing
1766 	 * with AP initialization
1767 	 */
1768 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1769 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1770 		cpu_relax();
1771 #endif
1772 }
1773 
1774 #ifdef CONFIG_X86_64
1775 static inline void setup_getcpu(int cpu)
1776 {
1777 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1778 	struct desc_struct d = { };
1779 
1780 	if (boot_cpu_has(X86_FEATURE_RDTSCP))
1781 		write_rdtscp_aux(cpudata);
1782 
1783 	/* Store CPU and node number in limit. */
1784 	d.limit0 = cpudata;
1785 	d.limit1 = cpudata >> 16;
1786 
1787 	d.type = 5;		/* RO data, expand down, accessed */
1788 	d.dpl = 3;		/* Visible to user code */
1789 	d.s = 1;		/* Not a system segment */
1790 	d.p = 1;		/* Present */
1791 	d.d = 1;		/* 32-bit */
1792 
1793 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1794 }
1795 
1796 static inline void ucode_cpu_init(int cpu)
1797 {
1798 	if (cpu)
1799 		load_ucode_ap();
1800 }
1801 
1802 static inline void tss_setup_ist(struct tss_struct *tss)
1803 {
1804 	/* Set up the per-CPU TSS IST stacks */
1805 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1806 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1807 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1808 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1809 }
1810 
1811 #else /* CONFIG_X86_64 */
1812 
1813 static inline void setup_getcpu(int cpu) { }
1814 
1815 static inline void ucode_cpu_init(int cpu)
1816 {
1817 	show_ucode_info_early();
1818 }
1819 
1820 static inline void tss_setup_ist(struct tss_struct *tss) { }
1821 
1822 #endif /* !CONFIG_X86_64 */
1823 
1824 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
1825 {
1826 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
1827 
1828 #ifdef CONFIG_X86_IOPL_IOPERM
1829 	tss->io_bitmap.prev_max = 0;
1830 	tss->io_bitmap.prev_sequence = 0;
1831 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
1832 	/*
1833 	 * Invalidate the extra array entry past the end of the all
1834 	 * permission bitmap as required by the hardware.
1835 	 */
1836 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
1837 #endif
1838 }
1839 
1840 /*
1841  * cpu_init() initializes state that is per-CPU. Some data is already
1842  * initialized (naturally) in the bootstrap process, such as the GDT
1843  * and IDT. We reload them nevertheless, this function acts as a
1844  * 'CPU state barrier', nothing should get across.
1845  */
1846 void cpu_init(void)
1847 {
1848 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
1849 	struct task_struct *cur = current;
1850 	int cpu = raw_smp_processor_id();
1851 
1852 	wait_for_master_cpu(cpu);
1853 
1854 	ucode_cpu_init(cpu);
1855 
1856 #ifdef CONFIG_NUMA
1857 	if (this_cpu_read(numa_node) == 0 &&
1858 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1859 		set_numa_node(early_cpu_to_node(cpu));
1860 #endif
1861 	setup_getcpu(cpu);
1862 
1863 	pr_debug("Initializing CPU#%d\n", cpu);
1864 
1865 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
1866 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
1867 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1868 
1869 	/*
1870 	 * Initialize the per-CPU GDT with the boot GDT,
1871 	 * and set up the GDT descriptor:
1872 	 */
1873 	switch_to_new_gdt(cpu);
1874 	load_current_idt();
1875 
1876 	if (IS_ENABLED(CONFIG_X86_64)) {
1877 		loadsegment(fs, 0);
1878 		memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1879 		syscall_init();
1880 
1881 		wrmsrl(MSR_FS_BASE, 0);
1882 		wrmsrl(MSR_KERNEL_GS_BASE, 0);
1883 		barrier();
1884 
1885 		x2apic_setup();
1886 	}
1887 
1888 	mmgrab(&init_mm);
1889 	cur->active_mm = &init_mm;
1890 	BUG_ON(cur->mm);
1891 	initialize_tlbstate_and_flush();
1892 	enter_lazy_tlb(&init_mm, cur);
1893 
1894 	/* Initialize the TSS. */
1895 	tss_setup_ist(tss);
1896 	tss_setup_io_bitmap(tss);
1897 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1898 
1899 	load_TR_desc();
1900 	/*
1901 	 * sp0 points to the entry trampoline stack regardless of what task
1902 	 * is running.
1903 	 */
1904 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1905 
1906 	load_mm_ldt(&init_mm);
1907 
1908 	clear_all_debug_regs();
1909 	dbg_restore_debug_regs();
1910 
1911 	doublefault_init_cpu_tss();
1912 
1913 	fpu__init_cpu();
1914 
1915 	if (is_uv_system())
1916 		uv_cpu_init();
1917 
1918 	load_fixmap_gdt(cpu);
1919 }
1920 
1921 /*
1922  * The microcode loader calls this upon late microcode load to recheck features,
1923  * only when microcode has been updated. Caller holds microcode_mutex and CPU
1924  * hotplug lock.
1925  */
1926 void microcode_check(void)
1927 {
1928 	struct cpuinfo_x86 info;
1929 
1930 	perf_check_microcode();
1931 
1932 	/* Reload CPUID max function as it might've changed. */
1933 	info.cpuid_level = cpuid_eax(0);
1934 
1935 	/*
1936 	 * Copy all capability leafs to pick up the synthetic ones so that
1937 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
1938 	 * get overwritten in get_cpu_cap().
1939 	 */
1940 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
1941 
1942 	get_cpu_cap(&info);
1943 
1944 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
1945 		return;
1946 
1947 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
1948 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
1949 }
1950 
1951 /*
1952  * Invoked from core CPU hotplug code after hotplug operations
1953  */
1954 void arch_smt_update(void)
1955 {
1956 	/* Handle the speculative execution misfeatures */
1957 	cpu_bugs_smt_update();
1958 	/* Check whether IPI broadcasting can be enabled */
1959 	apic_smt_update();
1960 }
1961