xref: /linux/arch/x86/kernel/cpu/common.c (revision e21f9e2e862e9eb3dd64eaddb6256b3e5098660f)
1 #include <linux/bootmem.h>
2 #include <linux/linkage.h>
3 #include <linux/bitops.h>
4 #include <linux/kernel.h>
5 #include <linux/export.h>
6 #include <linux/percpu.h>
7 #include <linux/string.h>
8 #include <linux/ctype.h>
9 #include <linux/delay.h>
10 #include <linux/sched/mm.h>
11 #include <linux/sched/clock.h>
12 #include <linux/sched/task.h>
13 #include <linux/init.h>
14 #include <linux/kprobes.h>
15 #include <linux/kgdb.h>
16 #include <linux/smp.h>
17 #include <linux/io.h>
18 #include <linux/syscore_ops.h>
19 
20 #include <asm/stackprotector.h>
21 #include <asm/perf_event.h>
22 #include <asm/mmu_context.h>
23 #include <asm/archrandom.h>
24 #include <asm/hypervisor.h>
25 #include <asm/processor.h>
26 #include <asm/tlbflush.h>
27 #include <asm/debugreg.h>
28 #include <asm/sections.h>
29 #include <asm/vsyscall.h>
30 #include <linux/topology.h>
31 #include <linux/cpumask.h>
32 #include <asm/pgtable.h>
33 #include <linux/atomic.h>
34 #include <asm/proto.h>
35 #include <asm/setup.h>
36 #include <asm/apic.h>
37 #include <asm/desc.h>
38 #include <asm/fpu/internal.h>
39 #include <asm/mtrr.h>
40 #include <asm/hwcap2.h>
41 #include <linux/numa.h>
42 #include <asm/asm.h>
43 #include <asm/bugs.h>
44 #include <asm/cpu.h>
45 #include <asm/mce.h>
46 #include <asm/msr.h>
47 #include <asm/pat.h>
48 #include <asm/microcode.h>
49 #include <asm/microcode_intel.h>
50 #include <asm/intel-family.h>
51 #include <asm/cpu_device_id.h>
52 
53 #ifdef CONFIG_X86_LOCAL_APIC
54 #include <asm/uv/uv.h>
55 #endif
56 
57 #include "cpu.h"
58 
59 u32 elf_hwcap2 __read_mostly;
60 
61 /* all of these masks are initialized in setup_cpu_local_masks() */
62 cpumask_var_t cpu_initialized_mask;
63 cpumask_var_t cpu_callout_mask;
64 cpumask_var_t cpu_callin_mask;
65 
66 /* representing cpus for which sibling maps can be computed */
67 cpumask_var_t cpu_sibling_setup_mask;
68 
69 /* correctly size the local cpu masks */
70 void __init setup_cpu_local_masks(void)
71 {
72 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
73 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
74 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
75 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
76 }
77 
78 static void default_init(struct cpuinfo_x86 *c)
79 {
80 #ifdef CONFIG_X86_64
81 	cpu_detect_cache_sizes(c);
82 #else
83 	/* Not much we can do here... */
84 	/* Check if at least it has cpuid */
85 	if (c->cpuid_level == -1) {
86 		/* No cpuid. It must be an ancient CPU */
87 		if (c->x86 == 4)
88 			strcpy(c->x86_model_id, "486");
89 		else if (c->x86 == 3)
90 			strcpy(c->x86_model_id, "386");
91 	}
92 #endif
93 }
94 
95 static const struct cpu_dev default_cpu = {
96 	.c_init		= default_init,
97 	.c_vendor	= "Unknown",
98 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
99 };
100 
101 static const struct cpu_dev *this_cpu = &default_cpu;
102 
103 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
104 #ifdef CONFIG_X86_64
105 	/*
106 	 * We need valid kernel segments for data and code in long mode too
107 	 * IRET will check the segment types  kkeil 2000/10/28
108 	 * Also sysret mandates a special GDT layout
109 	 *
110 	 * TLS descriptors are currently at a different place compared to i386.
111 	 * Hopefully nobody expects them at a fixed place (Wine?)
112 	 */
113 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
114 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
115 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
116 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
117 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
118 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
119 #else
120 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
121 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
122 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
123 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
124 	/*
125 	 * Segments used for calling PnP BIOS have byte granularity.
126 	 * They code segments and data segments have fixed 64k limits,
127 	 * the transfer segment sizes are set at run time.
128 	 */
129 	/* 32-bit code */
130 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
131 	/* 16-bit code */
132 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
133 	/* 16-bit data */
134 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
135 	/* 16-bit data */
136 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
137 	/* 16-bit data */
138 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
139 	/*
140 	 * The APM segments have byte granularity and their bases
141 	 * are set at run time.  All have 64k limits.
142 	 */
143 	/* 32-bit code */
144 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
145 	/* 16-bit code */
146 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
147 	/* data */
148 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
149 
150 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
151 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
152 	GDT_STACK_CANARY_INIT
153 #endif
154 } };
155 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
156 
157 static int __init x86_mpx_setup(char *s)
158 {
159 	/* require an exact match without trailing characters */
160 	if (strlen(s))
161 		return 0;
162 
163 	/* do not emit a message if the feature is not present */
164 	if (!boot_cpu_has(X86_FEATURE_MPX))
165 		return 1;
166 
167 	setup_clear_cpu_cap(X86_FEATURE_MPX);
168 	pr_info("nompx: Intel Memory Protection Extensions (MPX) disabled\n");
169 	return 1;
170 }
171 __setup("nompx", x86_mpx_setup);
172 
173 #ifdef CONFIG_X86_64
174 static int __init x86_nopcid_setup(char *s)
175 {
176 	/* nopcid doesn't accept parameters */
177 	if (s)
178 		return -EINVAL;
179 
180 	/* do not emit a message if the feature is not present */
181 	if (!boot_cpu_has(X86_FEATURE_PCID))
182 		return 0;
183 
184 	setup_clear_cpu_cap(X86_FEATURE_PCID);
185 	pr_info("nopcid: PCID feature disabled\n");
186 	return 0;
187 }
188 early_param("nopcid", x86_nopcid_setup);
189 #endif
190 
191 static int __init x86_noinvpcid_setup(char *s)
192 {
193 	/* noinvpcid doesn't accept parameters */
194 	if (s)
195 		return -EINVAL;
196 
197 	/* do not emit a message if the feature is not present */
198 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
199 		return 0;
200 
201 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
202 	pr_info("noinvpcid: INVPCID feature disabled\n");
203 	return 0;
204 }
205 early_param("noinvpcid", x86_noinvpcid_setup);
206 
207 #ifdef CONFIG_X86_32
208 static int cachesize_override = -1;
209 static int disable_x86_serial_nr = 1;
210 
211 static int __init cachesize_setup(char *str)
212 {
213 	get_option(&str, &cachesize_override);
214 	return 1;
215 }
216 __setup("cachesize=", cachesize_setup);
217 
218 static int __init x86_sep_setup(char *s)
219 {
220 	setup_clear_cpu_cap(X86_FEATURE_SEP);
221 	return 1;
222 }
223 __setup("nosep", x86_sep_setup);
224 
225 /* Standard macro to see if a specific flag is changeable */
226 static inline int flag_is_changeable_p(u32 flag)
227 {
228 	u32 f1, f2;
229 
230 	/*
231 	 * Cyrix and IDT cpus allow disabling of CPUID
232 	 * so the code below may return different results
233 	 * when it is executed before and after enabling
234 	 * the CPUID. Add "volatile" to not allow gcc to
235 	 * optimize the subsequent calls to this function.
236 	 */
237 	asm volatile ("pushfl		\n\t"
238 		      "pushfl		\n\t"
239 		      "popl %0		\n\t"
240 		      "movl %0, %1	\n\t"
241 		      "xorl %2, %0	\n\t"
242 		      "pushl %0		\n\t"
243 		      "popfl		\n\t"
244 		      "pushfl		\n\t"
245 		      "popl %0		\n\t"
246 		      "popfl		\n\t"
247 
248 		      : "=&r" (f1), "=&r" (f2)
249 		      : "ir" (flag));
250 
251 	return ((f1^f2) & flag) != 0;
252 }
253 
254 /* Probe for the CPUID instruction */
255 int have_cpuid_p(void)
256 {
257 	return flag_is_changeable_p(X86_EFLAGS_ID);
258 }
259 
260 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
261 {
262 	unsigned long lo, hi;
263 
264 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
265 		return;
266 
267 	/* Disable processor serial number: */
268 
269 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
270 	lo |= 0x200000;
271 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
272 
273 	pr_notice("CPU serial number disabled.\n");
274 	clear_cpu_cap(c, X86_FEATURE_PN);
275 
276 	/* Disabling the serial number may affect the cpuid level */
277 	c->cpuid_level = cpuid_eax(0);
278 }
279 
280 static int __init x86_serial_nr_setup(char *s)
281 {
282 	disable_x86_serial_nr = 0;
283 	return 1;
284 }
285 __setup("serialnumber", x86_serial_nr_setup);
286 #else
287 static inline int flag_is_changeable_p(u32 flag)
288 {
289 	return 1;
290 }
291 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
292 {
293 }
294 #endif
295 
296 static __init int setup_disable_smep(char *arg)
297 {
298 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
299 	/* Check for things that depend on SMEP being enabled: */
300 	check_mpx_erratum(&boot_cpu_data);
301 	return 1;
302 }
303 __setup("nosmep", setup_disable_smep);
304 
305 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
306 {
307 	if (cpu_has(c, X86_FEATURE_SMEP))
308 		cr4_set_bits(X86_CR4_SMEP);
309 }
310 
311 static __init int setup_disable_smap(char *arg)
312 {
313 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
314 	return 1;
315 }
316 __setup("nosmap", setup_disable_smap);
317 
318 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
319 {
320 	unsigned long eflags = native_save_fl();
321 
322 	/* This should have been cleared long ago */
323 	BUG_ON(eflags & X86_EFLAGS_AC);
324 
325 	if (cpu_has(c, X86_FEATURE_SMAP)) {
326 #ifdef CONFIG_X86_SMAP
327 		cr4_set_bits(X86_CR4_SMAP);
328 #else
329 		cr4_clear_bits(X86_CR4_SMAP);
330 #endif
331 	}
332 }
333 
334 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
335 {
336 	/* Check the boot processor, plus build option for UMIP. */
337 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
338 		goto out;
339 
340 	/* Check the current processor's cpuid bits. */
341 	if (!cpu_has(c, X86_FEATURE_UMIP))
342 		goto out;
343 
344 	cr4_set_bits(X86_CR4_UMIP);
345 
346 	pr_info("x86/cpu: Activated the Intel User Mode Instruction Prevention (UMIP) CPU feature\n");
347 
348 	return;
349 
350 out:
351 	/*
352 	 * Make sure UMIP is disabled in case it was enabled in a
353 	 * previous boot (e.g., via kexec).
354 	 */
355 	cr4_clear_bits(X86_CR4_UMIP);
356 }
357 
358 /*
359  * Protection Keys are not available in 32-bit mode.
360  */
361 static bool pku_disabled;
362 
363 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
364 {
365 	/* check the boot processor, plus compile options for PKU: */
366 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
367 		return;
368 	/* checks the actual processor's cpuid bits: */
369 	if (!cpu_has(c, X86_FEATURE_PKU))
370 		return;
371 	if (pku_disabled)
372 		return;
373 
374 	cr4_set_bits(X86_CR4_PKE);
375 	/*
376 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
377 	 * cpuid bit to be set.  We need to ensure that we
378 	 * update that bit in this CPU's "cpu_info".
379 	 */
380 	get_cpu_cap(c);
381 }
382 
383 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
384 static __init int setup_disable_pku(char *arg)
385 {
386 	/*
387 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
388 	 * runtime checks are against OSPKE so clearing the
389 	 * bit does nothing.
390 	 *
391 	 * This way, we will see "pku" in cpuinfo, but not
392 	 * "ospke", which is exactly what we want.  It shows
393 	 * that the CPU has PKU, but the OS has not enabled it.
394 	 * This happens to be exactly how a system would look
395 	 * if we disabled the config option.
396 	 */
397 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
398 	pku_disabled = true;
399 	return 1;
400 }
401 __setup("nopku", setup_disable_pku);
402 #endif /* CONFIG_X86_64 */
403 
404 /*
405  * Some CPU features depend on higher CPUID levels, which may not always
406  * be available due to CPUID level capping or broken virtualization
407  * software.  Add those features to this table to auto-disable them.
408  */
409 struct cpuid_dependent_feature {
410 	u32 feature;
411 	u32 level;
412 };
413 
414 static const struct cpuid_dependent_feature
415 cpuid_dependent_features[] = {
416 	{ X86_FEATURE_MWAIT,		0x00000005 },
417 	{ X86_FEATURE_DCA,		0x00000009 },
418 	{ X86_FEATURE_XSAVE,		0x0000000d },
419 	{ 0, 0 }
420 };
421 
422 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
423 {
424 	const struct cpuid_dependent_feature *df;
425 
426 	for (df = cpuid_dependent_features; df->feature; df++) {
427 
428 		if (!cpu_has(c, df->feature))
429 			continue;
430 		/*
431 		 * Note: cpuid_level is set to -1 if unavailable, but
432 		 * extended_extended_level is set to 0 if unavailable
433 		 * and the legitimate extended levels are all negative
434 		 * when signed; hence the weird messing around with
435 		 * signs here...
436 		 */
437 		if (!((s32)df->level < 0 ?
438 		     (u32)df->level > (u32)c->extended_cpuid_level :
439 		     (s32)df->level > (s32)c->cpuid_level))
440 			continue;
441 
442 		clear_cpu_cap(c, df->feature);
443 		if (!warn)
444 			continue;
445 
446 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
447 			x86_cap_flag(df->feature), df->level);
448 	}
449 }
450 
451 /*
452  * Naming convention should be: <Name> [(<Codename>)]
453  * This table only is used unless init_<vendor>() below doesn't set it;
454  * in particular, if CPUID levels 0x80000002..4 are supported, this
455  * isn't used
456  */
457 
458 /* Look up CPU names by table lookup. */
459 static const char *table_lookup_model(struct cpuinfo_x86 *c)
460 {
461 #ifdef CONFIG_X86_32
462 	const struct legacy_cpu_model_info *info;
463 
464 	if (c->x86_model >= 16)
465 		return NULL;	/* Range check */
466 
467 	if (!this_cpu)
468 		return NULL;
469 
470 	info = this_cpu->legacy_models;
471 
472 	while (info->family) {
473 		if (info->family == c->x86)
474 			return info->model_names[c->x86_model];
475 		info++;
476 	}
477 #endif
478 	return NULL;		/* Not found */
479 }
480 
481 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS];
482 __u32 cpu_caps_set[NCAPINTS + NBUGINTS];
483 
484 void load_percpu_segment(int cpu)
485 {
486 #ifdef CONFIG_X86_32
487 	loadsegment(fs, __KERNEL_PERCPU);
488 #else
489 	__loadsegment_simple(gs, 0);
490 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
491 #endif
492 	load_stack_canary_segment();
493 }
494 
495 #ifdef CONFIG_X86_32
496 /* The 32-bit entry code needs to find cpu_entry_area. */
497 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
498 #endif
499 
500 #ifdef CONFIG_X86_64
501 /*
502  * Special IST stacks which the CPU switches to when it calls
503  * an IST-marked descriptor entry. Up to 7 stacks (hardware
504  * limit), all of them are 4K, except the debug stack which
505  * is 8K.
506  */
507 static const unsigned int exception_stack_sizes[N_EXCEPTION_STACKS] = {
508 	  [0 ... N_EXCEPTION_STACKS - 1]	= EXCEPTION_STKSZ,
509 	  [DEBUG_STACK - 1]			= DEBUG_STKSZ
510 };
511 #endif
512 
513 /* Load the original GDT from the per-cpu structure */
514 void load_direct_gdt(int cpu)
515 {
516 	struct desc_ptr gdt_descr;
517 
518 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
519 	gdt_descr.size = GDT_SIZE - 1;
520 	load_gdt(&gdt_descr);
521 }
522 EXPORT_SYMBOL_GPL(load_direct_gdt);
523 
524 /* Load a fixmap remapping of the per-cpu GDT */
525 void load_fixmap_gdt(int cpu)
526 {
527 	struct desc_ptr gdt_descr;
528 
529 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
530 	gdt_descr.size = GDT_SIZE - 1;
531 	load_gdt(&gdt_descr);
532 }
533 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
534 
535 /*
536  * Current gdt points %fs at the "master" per-cpu area: after this,
537  * it's on the real one.
538  */
539 void switch_to_new_gdt(int cpu)
540 {
541 	/* Load the original GDT */
542 	load_direct_gdt(cpu);
543 	/* Reload the per-cpu base */
544 	load_percpu_segment(cpu);
545 }
546 
547 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
548 
549 static void get_model_name(struct cpuinfo_x86 *c)
550 {
551 	unsigned int *v;
552 	char *p, *q, *s;
553 
554 	if (c->extended_cpuid_level < 0x80000004)
555 		return;
556 
557 	v = (unsigned int *)c->x86_model_id;
558 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
559 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
560 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
561 	c->x86_model_id[48] = 0;
562 
563 	/* Trim whitespace */
564 	p = q = s = &c->x86_model_id[0];
565 
566 	while (*p == ' ')
567 		p++;
568 
569 	while (*p) {
570 		/* Note the last non-whitespace index */
571 		if (!isspace(*p))
572 			s = q;
573 
574 		*q++ = *p++;
575 	}
576 
577 	*(s + 1) = '\0';
578 }
579 
580 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
581 {
582 	unsigned int n, dummy, ebx, ecx, edx, l2size;
583 
584 	n = c->extended_cpuid_level;
585 
586 	if (n >= 0x80000005) {
587 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
588 		c->x86_cache_size = (ecx>>24) + (edx>>24);
589 #ifdef CONFIG_X86_64
590 		/* On K8 L1 TLB is inclusive, so don't count it */
591 		c->x86_tlbsize = 0;
592 #endif
593 	}
594 
595 	if (n < 0x80000006)	/* Some chips just has a large L1. */
596 		return;
597 
598 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
599 	l2size = ecx >> 16;
600 
601 #ifdef CONFIG_X86_64
602 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
603 #else
604 	/* do processor-specific cache resizing */
605 	if (this_cpu->legacy_cache_size)
606 		l2size = this_cpu->legacy_cache_size(c, l2size);
607 
608 	/* Allow user to override all this if necessary. */
609 	if (cachesize_override != -1)
610 		l2size = cachesize_override;
611 
612 	if (l2size == 0)
613 		return;		/* Again, no L2 cache is possible */
614 #endif
615 
616 	c->x86_cache_size = l2size;
617 }
618 
619 u16 __read_mostly tlb_lli_4k[NR_INFO];
620 u16 __read_mostly tlb_lli_2m[NR_INFO];
621 u16 __read_mostly tlb_lli_4m[NR_INFO];
622 u16 __read_mostly tlb_lld_4k[NR_INFO];
623 u16 __read_mostly tlb_lld_2m[NR_INFO];
624 u16 __read_mostly tlb_lld_4m[NR_INFO];
625 u16 __read_mostly tlb_lld_1g[NR_INFO];
626 
627 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
628 {
629 	if (this_cpu->c_detect_tlb)
630 		this_cpu->c_detect_tlb(c);
631 
632 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
633 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
634 		tlb_lli_4m[ENTRIES]);
635 
636 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
637 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
638 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
639 }
640 
641 void detect_ht(struct cpuinfo_x86 *c)
642 {
643 #ifdef CONFIG_SMP
644 	u32 eax, ebx, ecx, edx;
645 	int index_msb, core_bits;
646 	static bool printed;
647 
648 	if (!cpu_has(c, X86_FEATURE_HT))
649 		return;
650 
651 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
652 		goto out;
653 
654 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
655 		return;
656 
657 	cpuid(1, &eax, &ebx, &ecx, &edx);
658 
659 	smp_num_siblings = (ebx & 0xff0000) >> 16;
660 
661 	if (smp_num_siblings == 1) {
662 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
663 		goto out;
664 	}
665 
666 	if (smp_num_siblings <= 1)
667 		goto out;
668 
669 	index_msb = get_count_order(smp_num_siblings);
670 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
671 
672 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
673 
674 	index_msb = get_count_order(smp_num_siblings);
675 
676 	core_bits = get_count_order(c->x86_max_cores);
677 
678 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
679 				       ((1 << core_bits) - 1);
680 
681 out:
682 	if (!printed && (c->x86_max_cores * smp_num_siblings) > 1) {
683 		pr_info("CPU: Physical Processor ID: %d\n",
684 			c->phys_proc_id);
685 		pr_info("CPU: Processor Core ID: %d\n",
686 			c->cpu_core_id);
687 		printed = 1;
688 	}
689 #endif
690 }
691 
692 static void get_cpu_vendor(struct cpuinfo_x86 *c)
693 {
694 	char *v = c->x86_vendor_id;
695 	int i;
696 
697 	for (i = 0; i < X86_VENDOR_NUM; i++) {
698 		if (!cpu_devs[i])
699 			break;
700 
701 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
702 		    (cpu_devs[i]->c_ident[1] &&
703 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
704 
705 			this_cpu = cpu_devs[i];
706 			c->x86_vendor = this_cpu->c_x86_vendor;
707 			return;
708 		}
709 	}
710 
711 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
712 		    "CPU: Your system may be unstable.\n", v);
713 
714 	c->x86_vendor = X86_VENDOR_UNKNOWN;
715 	this_cpu = &default_cpu;
716 }
717 
718 void cpu_detect(struct cpuinfo_x86 *c)
719 {
720 	/* Get vendor name */
721 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
722 	      (unsigned int *)&c->x86_vendor_id[0],
723 	      (unsigned int *)&c->x86_vendor_id[8],
724 	      (unsigned int *)&c->x86_vendor_id[4]);
725 
726 	c->x86 = 4;
727 	/* Intel-defined flags: level 0x00000001 */
728 	if (c->cpuid_level >= 0x00000001) {
729 		u32 junk, tfms, cap0, misc;
730 
731 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
732 		c->x86		= x86_family(tfms);
733 		c->x86_model	= x86_model(tfms);
734 		c->x86_stepping	= x86_stepping(tfms);
735 
736 		if (cap0 & (1<<19)) {
737 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
738 			c->x86_cache_alignment = c->x86_clflush_size;
739 		}
740 	}
741 }
742 
743 static void apply_forced_caps(struct cpuinfo_x86 *c)
744 {
745 	int i;
746 
747 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
748 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
749 		c->x86_capability[i] |= cpu_caps_set[i];
750 	}
751 }
752 
753 static void init_speculation_control(struct cpuinfo_x86 *c)
754 {
755 	/*
756 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
757 	 * and they also have a different bit for STIBP support. Also,
758 	 * a hypervisor might have set the individual AMD bits even on
759 	 * Intel CPUs, for finer-grained selection of what's available.
760 	 *
761 	 * We use the AMD bits in 0x8000_0008 EBX as the generic hardware
762 	 * features, which are visible in /proc/cpuinfo and used by the
763 	 * kernel. So set those accordingly from the Intel bits.
764 	 */
765 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
766 		set_cpu_cap(c, X86_FEATURE_IBRS);
767 		set_cpu_cap(c, X86_FEATURE_IBPB);
768 	}
769 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
770 		set_cpu_cap(c, X86_FEATURE_STIBP);
771 }
772 
773 void get_cpu_cap(struct cpuinfo_x86 *c)
774 {
775 	u32 eax, ebx, ecx, edx;
776 
777 	/* Intel-defined flags: level 0x00000001 */
778 	if (c->cpuid_level >= 0x00000001) {
779 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
780 
781 		c->x86_capability[CPUID_1_ECX] = ecx;
782 		c->x86_capability[CPUID_1_EDX] = edx;
783 	}
784 
785 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
786 	if (c->cpuid_level >= 0x00000006)
787 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
788 
789 	/* Additional Intel-defined flags: level 0x00000007 */
790 	if (c->cpuid_level >= 0x00000007) {
791 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
792 		c->x86_capability[CPUID_7_0_EBX] = ebx;
793 		c->x86_capability[CPUID_7_ECX] = ecx;
794 		c->x86_capability[CPUID_7_EDX] = edx;
795 	}
796 
797 	/* Extended state features: level 0x0000000d */
798 	if (c->cpuid_level >= 0x0000000d) {
799 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
800 
801 		c->x86_capability[CPUID_D_1_EAX] = eax;
802 	}
803 
804 	/* Additional Intel-defined flags: level 0x0000000F */
805 	if (c->cpuid_level >= 0x0000000F) {
806 
807 		/* QoS sub-leaf, EAX=0Fh, ECX=0 */
808 		cpuid_count(0x0000000F, 0, &eax, &ebx, &ecx, &edx);
809 		c->x86_capability[CPUID_F_0_EDX] = edx;
810 
811 		if (cpu_has(c, X86_FEATURE_CQM_LLC)) {
812 			/* will be overridden if occupancy monitoring exists */
813 			c->x86_cache_max_rmid = ebx;
814 
815 			/* QoS sub-leaf, EAX=0Fh, ECX=1 */
816 			cpuid_count(0x0000000F, 1, &eax, &ebx, &ecx, &edx);
817 			c->x86_capability[CPUID_F_1_EDX] = edx;
818 
819 			if ((cpu_has(c, X86_FEATURE_CQM_OCCUP_LLC)) ||
820 			      ((cpu_has(c, X86_FEATURE_CQM_MBM_TOTAL)) ||
821 			       (cpu_has(c, X86_FEATURE_CQM_MBM_LOCAL)))) {
822 				c->x86_cache_max_rmid = ecx;
823 				c->x86_cache_occ_scale = ebx;
824 			}
825 		} else {
826 			c->x86_cache_max_rmid = -1;
827 			c->x86_cache_occ_scale = -1;
828 		}
829 	}
830 
831 	/* AMD-defined flags: level 0x80000001 */
832 	eax = cpuid_eax(0x80000000);
833 	c->extended_cpuid_level = eax;
834 
835 	if ((eax & 0xffff0000) == 0x80000000) {
836 		if (eax >= 0x80000001) {
837 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
838 
839 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
840 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
841 		}
842 	}
843 
844 	if (c->extended_cpuid_level >= 0x80000007) {
845 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
846 
847 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
848 		c->x86_power = edx;
849 	}
850 
851 	if (c->extended_cpuid_level >= 0x8000000a)
852 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
853 
854 	init_scattered_cpuid_features(c);
855 	init_speculation_control(c);
856 
857 	/*
858 	 * Clear/Set all flags overridden by options, after probe.
859 	 * This needs to happen each time we re-probe, which may happen
860 	 * several times during CPU initialization.
861 	 */
862 	apply_forced_caps(c);
863 }
864 
865 static void get_cpu_address_sizes(struct cpuinfo_x86 *c)
866 {
867 	u32 eax, ebx, ecx, edx;
868 
869 	if (c->extended_cpuid_level >= 0x80000008) {
870 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
871 
872 		c->x86_virt_bits = (eax >> 8) & 0xff;
873 		c->x86_phys_bits = eax & 0xff;
874 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
875 	}
876 #ifdef CONFIG_X86_32
877 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
878 		c->x86_phys_bits = 36;
879 #endif
880 }
881 
882 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
883 {
884 #ifdef CONFIG_X86_32
885 	int i;
886 
887 	/*
888 	 * First of all, decide if this is a 486 or higher
889 	 * It's a 486 if we can modify the AC flag
890 	 */
891 	if (flag_is_changeable_p(X86_EFLAGS_AC))
892 		c->x86 = 4;
893 	else
894 		c->x86 = 3;
895 
896 	for (i = 0; i < X86_VENDOR_NUM; i++)
897 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
898 			c->x86_vendor_id[0] = 0;
899 			cpu_devs[i]->c_identify(c);
900 			if (c->x86_vendor_id[0]) {
901 				get_cpu_vendor(c);
902 				break;
903 			}
904 		}
905 #endif
906 }
907 
908 static const __initconst struct x86_cpu_id cpu_no_speculation[] = {
909 	{ X86_VENDOR_INTEL,	6, INTEL_FAM6_ATOM_CEDARVIEW,	X86_FEATURE_ANY },
910 	{ X86_VENDOR_INTEL,	6, INTEL_FAM6_ATOM_CLOVERVIEW,	X86_FEATURE_ANY },
911 	{ X86_VENDOR_INTEL,	6, INTEL_FAM6_ATOM_LINCROFT,	X86_FEATURE_ANY },
912 	{ X86_VENDOR_INTEL,	6, INTEL_FAM6_ATOM_PENWELL,	X86_FEATURE_ANY },
913 	{ X86_VENDOR_INTEL,	6, INTEL_FAM6_ATOM_PINEVIEW,	X86_FEATURE_ANY },
914 	{ X86_VENDOR_CENTAUR,	5 },
915 	{ X86_VENDOR_INTEL,	5 },
916 	{ X86_VENDOR_NSC,	5 },
917 	{ X86_VENDOR_ANY,	4 },
918 	{}
919 };
920 
921 static const __initconst struct x86_cpu_id cpu_no_meltdown[] = {
922 	{ X86_VENDOR_AMD },
923 	{}
924 };
925 
926 static bool __init cpu_vulnerable_to_meltdown(struct cpuinfo_x86 *c)
927 {
928 	u64 ia32_cap = 0;
929 
930 	if (x86_match_cpu(cpu_no_meltdown))
931 		return false;
932 
933 	if (cpu_has(c, X86_FEATURE_ARCH_CAPABILITIES))
934 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
935 
936 	/* Rogue Data Cache Load? No! */
937 	if (ia32_cap & ARCH_CAP_RDCL_NO)
938 		return false;
939 
940 	return true;
941 }
942 
943 /*
944  * Do minimum CPU detection early.
945  * Fields really needed: vendor, cpuid_level, family, model, mask,
946  * cache alignment.
947  * The others are not touched to avoid unwanted side effects.
948  *
949  * WARNING: this function is only called on the boot CPU.  Don't add code
950  * here that is supposed to run on all CPUs.
951  */
952 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
953 {
954 #ifdef CONFIG_X86_64
955 	c->x86_clflush_size = 64;
956 	c->x86_phys_bits = 36;
957 	c->x86_virt_bits = 48;
958 #else
959 	c->x86_clflush_size = 32;
960 	c->x86_phys_bits = 32;
961 	c->x86_virt_bits = 32;
962 #endif
963 	c->x86_cache_alignment = c->x86_clflush_size;
964 
965 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
966 	c->extended_cpuid_level = 0;
967 
968 	/* cyrix could have cpuid enabled via c_identify()*/
969 	if (have_cpuid_p()) {
970 		cpu_detect(c);
971 		get_cpu_vendor(c);
972 		get_cpu_cap(c);
973 		get_cpu_address_sizes(c);
974 		setup_force_cpu_cap(X86_FEATURE_CPUID);
975 
976 		if (this_cpu->c_early_init)
977 			this_cpu->c_early_init(c);
978 
979 		c->cpu_index = 0;
980 		filter_cpuid_features(c, false);
981 
982 		if (this_cpu->c_bsp_init)
983 			this_cpu->c_bsp_init(c);
984 	} else {
985 		identify_cpu_without_cpuid(c);
986 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
987 	}
988 
989 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
990 
991 	if (!x86_match_cpu(cpu_no_speculation)) {
992 		if (cpu_vulnerable_to_meltdown(c))
993 			setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
994 		setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
995 		setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
996 	}
997 
998 	fpu__init_system(c);
999 
1000 #ifdef CONFIG_X86_32
1001 	/*
1002 	 * Regardless of whether PCID is enumerated, the SDM says
1003 	 * that it can't be enabled in 32-bit mode.
1004 	 */
1005 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1006 #endif
1007 }
1008 
1009 void __init early_cpu_init(void)
1010 {
1011 	const struct cpu_dev *const *cdev;
1012 	int count = 0;
1013 
1014 #ifdef CONFIG_PROCESSOR_SELECT
1015 	pr_info("KERNEL supported cpus:\n");
1016 #endif
1017 
1018 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1019 		const struct cpu_dev *cpudev = *cdev;
1020 
1021 		if (count >= X86_VENDOR_NUM)
1022 			break;
1023 		cpu_devs[count] = cpudev;
1024 		count++;
1025 
1026 #ifdef CONFIG_PROCESSOR_SELECT
1027 		{
1028 			unsigned int j;
1029 
1030 			for (j = 0; j < 2; j++) {
1031 				if (!cpudev->c_ident[j])
1032 					continue;
1033 				pr_info("  %s %s\n", cpudev->c_vendor,
1034 					cpudev->c_ident[j]);
1035 			}
1036 		}
1037 #endif
1038 	}
1039 	early_identify_cpu(&boot_cpu_data);
1040 }
1041 
1042 /*
1043  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1044  * unfortunately, that's not true in practice because of early VIA
1045  * chips and (more importantly) broken virtualizers that are not easy
1046  * to detect. In the latter case it doesn't even *fail* reliably, so
1047  * probing for it doesn't even work. Disable it completely on 32-bit
1048  * unless we can find a reliable way to detect all the broken cases.
1049  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1050  */
1051 static void detect_nopl(struct cpuinfo_x86 *c)
1052 {
1053 #ifdef CONFIG_X86_32
1054 	clear_cpu_cap(c, X86_FEATURE_NOPL);
1055 #else
1056 	set_cpu_cap(c, X86_FEATURE_NOPL);
1057 #endif
1058 }
1059 
1060 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1061 {
1062 #ifdef CONFIG_X86_64
1063 	/*
1064 	 * Empirically, writing zero to a segment selector on AMD does
1065 	 * not clear the base, whereas writing zero to a segment
1066 	 * selector on Intel does clear the base.  Intel's behavior
1067 	 * allows slightly faster context switches in the common case
1068 	 * where GS is unused by the prev and next threads.
1069 	 *
1070 	 * Since neither vendor documents this anywhere that I can see,
1071 	 * detect it directly instead of hardcoding the choice by
1072 	 * vendor.
1073 	 *
1074 	 * I've designated AMD's behavior as the "bug" because it's
1075 	 * counterintuitive and less friendly.
1076 	 */
1077 
1078 	unsigned long old_base, tmp;
1079 	rdmsrl(MSR_FS_BASE, old_base);
1080 	wrmsrl(MSR_FS_BASE, 1);
1081 	loadsegment(fs, 0);
1082 	rdmsrl(MSR_FS_BASE, tmp);
1083 	if (tmp != 0)
1084 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1085 	wrmsrl(MSR_FS_BASE, old_base);
1086 #endif
1087 }
1088 
1089 static void generic_identify(struct cpuinfo_x86 *c)
1090 {
1091 	c->extended_cpuid_level = 0;
1092 
1093 	if (!have_cpuid_p())
1094 		identify_cpu_without_cpuid(c);
1095 
1096 	/* cyrix could have cpuid enabled via c_identify()*/
1097 	if (!have_cpuid_p())
1098 		return;
1099 
1100 	cpu_detect(c);
1101 
1102 	get_cpu_vendor(c);
1103 
1104 	get_cpu_cap(c);
1105 
1106 	get_cpu_address_sizes(c);
1107 
1108 	if (c->cpuid_level >= 0x00000001) {
1109 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1110 #ifdef CONFIG_X86_32
1111 # ifdef CONFIG_SMP
1112 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1113 # else
1114 		c->apicid = c->initial_apicid;
1115 # endif
1116 #endif
1117 		c->phys_proc_id = c->initial_apicid;
1118 	}
1119 
1120 	get_model_name(c); /* Default name */
1121 
1122 	detect_nopl(c);
1123 
1124 	detect_null_seg_behavior(c);
1125 
1126 	/*
1127 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1128 	 * systems that run Linux at CPL > 0 may or may not have the
1129 	 * issue, but, even if they have the issue, there's absolutely
1130 	 * nothing we can do about it because we can't use the real IRET
1131 	 * instruction.
1132 	 *
1133 	 * NB: For the time being, only 32-bit kernels support
1134 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1135 	 * whether to apply espfix using paravirt hooks.  If any
1136 	 * non-paravirt system ever shows up that does *not* have the
1137 	 * ESPFIX issue, we can change this.
1138 	 */
1139 #ifdef CONFIG_X86_32
1140 # ifdef CONFIG_PARAVIRT
1141 	do {
1142 		extern void native_iret(void);
1143 		if (pv_cpu_ops.iret == native_iret)
1144 			set_cpu_bug(c, X86_BUG_ESPFIX);
1145 	} while (0);
1146 # else
1147 	set_cpu_bug(c, X86_BUG_ESPFIX);
1148 # endif
1149 #endif
1150 }
1151 
1152 static void x86_init_cache_qos(struct cpuinfo_x86 *c)
1153 {
1154 	/*
1155 	 * The heavy lifting of max_rmid and cache_occ_scale are handled
1156 	 * in get_cpu_cap().  Here we just set the max_rmid for the boot_cpu
1157 	 * in case CQM bits really aren't there in this CPU.
1158 	 */
1159 	if (c != &boot_cpu_data) {
1160 		boot_cpu_data.x86_cache_max_rmid =
1161 			min(boot_cpu_data.x86_cache_max_rmid,
1162 			    c->x86_cache_max_rmid);
1163 	}
1164 }
1165 
1166 /*
1167  * Validate that ACPI/mptables have the same information about the
1168  * effective APIC id and update the package map.
1169  */
1170 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1171 {
1172 #ifdef CONFIG_SMP
1173 	unsigned int apicid, cpu = smp_processor_id();
1174 
1175 	apicid = apic->cpu_present_to_apicid(cpu);
1176 
1177 	if (apicid != c->apicid) {
1178 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1179 		       cpu, apicid, c->initial_apicid);
1180 	}
1181 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1182 #else
1183 	c->logical_proc_id = 0;
1184 #endif
1185 }
1186 
1187 /*
1188  * This does the hard work of actually picking apart the CPU stuff...
1189  */
1190 static void identify_cpu(struct cpuinfo_x86 *c)
1191 {
1192 	int i;
1193 
1194 	c->loops_per_jiffy = loops_per_jiffy;
1195 	c->x86_cache_size = 0;
1196 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1197 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1198 	c->x86_vendor_id[0] = '\0'; /* Unset */
1199 	c->x86_model_id[0] = '\0';  /* Unset */
1200 	c->x86_max_cores = 1;
1201 	c->x86_coreid_bits = 0;
1202 	c->cu_id = 0xff;
1203 #ifdef CONFIG_X86_64
1204 	c->x86_clflush_size = 64;
1205 	c->x86_phys_bits = 36;
1206 	c->x86_virt_bits = 48;
1207 #else
1208 	c->cpuid_level = -1;	/* CPUID not detected */
1209 	c->x86_clflush_size = 32;
1210 	c->x86_phys_bits = 32;
1211 	c->x86_virt_bits = 32;
1212 #endif
1213 	c->x86_cache_alignment = c->x86_clflush_size;
1214 	memset(&c->x86_capability, 0, sizeof c->x86_capability);
1215 
1216 	generic_identify(c);
1217 
1218 	if (this_cpu->c_identify)
1219 		this_cpu->c_identify(c);
1220 
1221 	/* Clear/Set all flags overridden by options, after probe */
1222 	apply_forced_caps(c);
1223 
1224 #ifdef CONFIG_X86_64
1225 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1226 #endif
1227 
1228 	/*
1229 	 * Vendor-specific initialization.  In this section we
1230 	 * canonicalize the feature flags, meaning if there are
1231 	 * features a certain CPU supports which CPUID doesn't
1232 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1233 	 * we handle them here.
1234 	 *
1235 	 * At the end of this section, c->x86_capability better
1236 	 * indicate the features this CPU genuinely supports!
1237 	 */
1238 	if (this_cpu->c_init)
1239 		this_cpu->c_init(c);
1240 
1241 	/* Disable the PN if appropriate */
1242 	squash_the_stupid_serial_number(c);
1243 
1244 	/* Set up SMEP/SMAP/UMIP */
1245 	setup_smep(c);
1246 	setup_smap(c);
1247 	setup_umip(c);
1248 
1249 	/*
1250 	 * The vendor-specific functions might have changed features.
1251 	 * Now we do "generic changes."
1252 	 */
1253 
1254 	/* Filter out anything that depends on CPUID levels we don't have */
1255 	filter_cpuid_features(c, true);
1256 
1257 	/* If the model name is still unset, do table lookup. */
1258 	if (!c->x86_model_id[0]) {
1259 		const char *p;
1260 		p = table_lookup_model(c);
1261 		if (p)
1262 			strcpy(c->x86_model_id, p);
1263 		else
1264 			/* Last resort... */
1265 			sprintf(c->x86_model_id, "%02x/%02x",
1266 				c->x86, c->x86_model);
1267 	}
1268 
1269 #ifdef CONFIG_X86_64
1270 	detect_ht(c);
1271 #endif
1272 
1273 	x86_init_rdrand(c);
1274 	x86_init_cache_qos(c);
1275 	setup_pku(c);
1276 
1277 	/*
1278 	 * Clear/Set all flags overridden by options, need do it
1279 	 * before following smp all cpus cap AND.
1280 	 */
1281 	apply_forced_caps(c);
1282 
1283 	/*
1284 	 * On SMP, boot_cpu_data holds the common feature set between
1285 	 * all CPUs; so make sure that we indicate which features are
1286 	 * common between the CPUs.  The first time this routine gets
1287 	 * executed, c == &boot_cpu_data.
1288 	 */
1289 	if (c != &boot_cpu_data) {
1290 		/* AND the already accumulated flags with these */
1291 		for (i = 0; i < NCAPINTS; i++)
1292 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1293 
1294 		/* OR, i.e. replicate the bug flags */
1295 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1296 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1297 	}
1298 
1299 	/* Init Machine Check Exception if available. */
1300 	mcheck_cpu_init(c);
1301 
1302 	select_idle_routine(c);
1303 
1304 #ifdef CONFIG_NUMA
1305 	numa_add_cpu(smp_processor_id());
1306 #endif
1307 }
1308 
1309 /*
1310  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1311  * on 32-bit kernels:
1312  */
1313 #ifdef CONFIG_X86_32
1314 void enable_sep_cpu(void)
1315 {
1316 	struct tss_struct *tss;
1317 	int cpu;
1318 
1319 	if (!boot_cpu_has(X86_FEATURE_SEP))
1320 		return;
1321 
1322 	cpu = get_cpu();
1323 	tss = &per_cpu(cpu_tss_rw, cpu);
1324 
1325 	/*
1326 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1327 	 * see the big comment in struct x86_hw_tss's definition.
1328 	 */
1329 
1330 	tss->x86_tss.ss1 = __KERNEL_CS;
1331 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1332 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1333 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1334 
1335 	put_cpu();
1336 }
1337 #endif
1338 
1339 void __init identify_boot_cpu(void)
1340 {
1341 	identify_cpu(&boot_cpu_data);
1342 #ifdef CONFIG_X86_32
1343 	sysenter_setup();
1344 	enable_sep_cpu();
1345 #endif
1346 	cpu_detect_tlb(&boot_cpu_data);
1347 }
1348 
1349 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1350 {
1351 	BUG_ON(c == &boot_cpu_data);
1352 	identify_cpu(c);
1353 #ifdef CONFIG_X86_32
1354 	enable_sep_cpu();
1355 #endif
1356 	mtrr_ap_init();
1357 	validate_apic_and_package_id(c);
1358 }
1359 
1360 static __init int setup_noclflush(char *arg)
1361 {
1362 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1363 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1364 	return 1;
1365 }
1366 __setup("noclflush", setup_noclflush);
1367 
1368 void print_cpu_info(struct cpuinfo_x86 *c)
1369 {
1370 	const char *vendor = NULL;
1371 
1372 	if (c->x86_vendor < X86_VENDOR_NUM) {
1373 		vendor = this_cpu->c_vendor;
1374 	} else {
1375 		if (c->cpuid_level >= 0)
1376 			vendor = c->x86_vendor_id;
1377 	}
1378 
1379 	if (vendor && !strstr(c->x86_model_id, vendor))
1380 		pr_cont("%s ", vendor);
1381 
1382 	if (c->x86_model_id[0])
1383 		pr_cont("%s", c->x86_model_id);
1384 	else
1385 		pr_cont("%d86", c->x86);
1386 
1387 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1388 
1389 	if (c->x86_stepping || c->cpuid_level >= 0)
1390 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1391 	else
1392 		pr_cont(")\n");
1393 }
1394 
1395 /*
1396  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1397  * But we need to keep a dummy __setup around otherwise it would
1398  * show up as an environment variable for init.
1399  */
1400 static __init int setup_clearcpuid(char *arg)
1401 {
1402 	return 1;
1403 }
1404 __setup("clearcpuid=", setup_clearcpuid);
1405 
1406 #ifdef CONFIG_X86_64
1407 DEFINE_PER_CPU_FIRST(union irq_stack_union,
1408 		     irq_stack_union) __aligned(PAGE_SIZE) __visible;
1409 EXPORT_PER_CPU_SYMBOL_GPL(irq_stack_union);
1410 
1411 /*
1412  * The following percpu variables are hot.  Align current_task to
1413  * cacheline size such that they fall in the same cacheline.
1414  */
1415 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1416 	&init_task;
1417 EXPORT_PER_CPU_SYMBOL(current_task);
1418 
1419 DEFINE_PER_CPU(char *, irq_stack_ptr) =
1420 	init_per_cpu_var(irq_stack_union.irq_stack) + IRQ_STACK_SIZE;
1421 
1422 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1423 
1424 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1425 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1426 
1427 /* May not be marked __init: used by software suspend */
1428 void syscall_init(void)
1429 {
1430 	extern char _entry_trampoline[];
1431 	extern char entry_SYSCALL_64_trampoline[];
1432 
1433 	int cpu = smp_processor_id();
1434 	unsigned long SYSCALL64_entry_trampoline =
1435 		(unsigned long)get_cpu_entry_area(cpu)->entry_trampoline +
1436 		(entry_SYSCALL_64_trampoline - _entry_trampoline);
1437 
1438 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1439 	if (static_cpu_has(X86_FEATURE_PTI))
1440 		wrmsrl(MSR_LSTAR, SYSCALL64_entry_trampoline);
1441 	else
1442 		wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1443 
1444 #ifdef CONFIG_IA32_EMULATION
1445 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1446 	/*
1447 	 * This only works on Intel CPUs.
1448 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1449 	 * This does not cause SYSENTER to jump to the wrong location, because
1450 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1451 	 */
1452 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1453 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1));
1454 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1455 #else
1456 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1457 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1458 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1459 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1460 #endif
1461 
1462 	/* Flags to clear on syscall */
1463 	wrmsrl(MSR_SYSCALL_MASK,
1464 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1465 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1466 }
1467 
1468 /*
1469  * Copies of the original ist values from the tss are only accessed during
1470  * debugging, no special alignment required.
1471  */
1472 DEFINE_PER_CPU(struct orig_ist, orig_ist);
1473 
1474 static DEFINE_PER_CPU(unsigned long, debug_stack_addr);
1475 DEFINE_PER_CPU(int, debug_stack_usage);
1476 
1477 int is_debug_stack(unsigned long addr)
1478 {
1479 	return __this_cpu_read(debug_stack_usage) ||
1480 		(addr <= __this_cpu_read(debug_stack_addr) &&
1481 		 addr > (__this_cpu_read(debug_stack_addr) - DEBUG_STKSZ));
1482 }
1483 NOKPROBE_SYMBOL(is_debug_stack);
1484 
1485 DEFINE_PER_CPU(u32, debug_idt_ctr);
1486 
1487 void debug_stack_set_zero(void)
1488 {
1489 	this_cpu_inc(debug_idt_ctr);
1490 	load_current_idt();
1491 }
1492 NOKPROBE_SYMBOL(debug_stack_set_zero);
1493 
1494 void debug_stack_reset(void)
1495 {
1496 	if (WARN_ON(!this_cpu_read(debug_idt_ctr)))
1497 		return;
1498 	if (this_cpu_dec_return(debug_idt_ctr) == 0)
1499 		load_current_idt();
1500 }
1501 NOKPROBE_SYMBOL(debug_stack_reset);
1502 
1503 #else	/* CONFIG_X86_64 */
1504 
1505 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1506 EXPORT_PER_CPU_SYMBOL(current_task);
1507 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1508 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1509 
1510 /*
1511  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1512  * the top of the kernel stack.  Use an extra percpu variable to track the
1513  * top of the kernel stack directly.
1514  */
1515 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1516 	(unsigned long)&init_thread_union + THREAD_SIZE;
1517 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1518 
1519 #ifdef CONFIG_CC_STACKPROTECTOR
1520 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1521 #endif
1522 
1523 #endif	/* CONFIG_X86_64 */
1524 
1525 /*
1526  * Clear all 6 debug registers:
1527  */
1528 static void clear_all_debug_regs(void)
1529 {
1530 	int i;
1531 
1532 	for (i = 0; i < 8; i++) {
1533 		/* Ignore db4, db5 */
1534 		if ((i == 4) || (i == 5))
1535 			continue;
1536 
1537 		set_debugreg(0, i);
1538 	}
1539 }
1540 
1541 #ifdef CONFIG_KGDB
1542 /*
1543  * Restore debug regs if using kgdbwait and you have a kernel debugger
1544  * connection established.
1545  */
1546 static void dbg_restore_debug_regs(void)
1547 {
1548 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1549 		arch_kgdb_ops.correct_hw_break();
1550 }
1551 #else /* ! CONFIG_KGDB */
1552 #define dbg_restore_debug_regs()
1553 #endif /* ! CONFIG_KGDB */
1554 
1555 static void wait_for_master_cpu(int cpu)
1556 {
1557 #ifdef CONFIG_SMP
1558 	/*
1559 	 * wait for ACK from master CPU before continuing
1560 	 * with AP initialization
1561 	 */
1562 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1563 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1564 		cpu_relax();
1565 #endif
1566 }
1567 
1568 /*
1569  * cpu_init() initializes state that is per-CPU. Some data is already
1570  * initialized (naturally) in the bootstrap process, such as the GDT
1571  * and IDT. We reload them nevertheless, this function acts as a
1572  * 'CPU state barrier', nothing should get across.
1573  * A lot of state is already set up in PDA init for 64 bit
1574  */
1575 #ifdef CONFIG_X86_64
1576 
1577 void cpu_init(void)
1578 {
1579 	struct orig_ist *oist;
1580 	struct task_struct *me;
1581 	struct tss_struct *t;
1582 	unsigned long v;
1583 	int cpu = raw_smp_processor_id();
1584 	int i;
1585 
1586 	wait_for_master_cpu(cpu);
1587 
1588 	/*
1589 	 * Initialize the CR4 shadow before doing anything that could
1590 	 * try to read it.
1591 	 */
1592 	cr4_init_shadow();
1593 
1594 	if (cpu)
1595 		load_ucode_ap();
1596 
1597 	t = &per_cpu(cpu_tss_rw, cpu);
1598 	oist = &per_cpu(orig_ist, cpu);
1599 
1600 #ifdef CONFIG_NUMA
1601 	if (this_cpu_read(numa_node) == 0 &&
1602 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1603 		set_numa_node(early_cpu_to_node(cpu));
1604 #endif
1605 
1606 	me = current;
1607 
1608 	pr_debug("Initializing CPU#%d\n", cpu);
1609 
1610 	cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1611 
1612 	/*
1613 	 * Initialize the per-CPU GDT with the boot GDT,
1614 	 * and set up the GDT descriptor:
1615 	 */
1616 
1617 	switch_to_new_gdt(cpu);
1618 	loadsegment(fs, 0);
1619 
1620 	load_current_idt();
1621 
1622 	memset(me->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1623 	syscall_init();
1624 
1625 	wrmsrl(MSR_FS_BASE, 0);
1626 	wrmsrl(MSR_KERNEL_GS_BASE, 0);
1627 	barrier();
1628 
1629 	x86_configure_nx();
1630 	x2apic_setup();
1631 
1632 	/*
1633 	 * set up and load the per-CPU TSS
1634 	 */
1635 	if (!oist->ist[0]) {
1636 		char *estacks = get_cpu_entry_area(cpu)->exception_stacks;
1637 
1638 		for (v = 0; v < N_EXCEPTION_STACKS; v++) {
1639 			estacks += exception_stack_sizes[v];
1640 			oist->ist[v] = t->x86_tss.ist[v] =
1641 					(unsigned long)estacks;
1642 			if (v == DEBUG_STACK-1)
1643 				per_cpu(debug_stack_addr, cpu) = (unsigned long)estacks;
1644 		}
1645 	}
1646 
1647 	t->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
1648 
1649 	/*
1650 	 * <= is required because the CPU will access up to
1651 	 * 8 bits beyond the end of the IO permission bitmap.
1652 	 */
1653 	for (i = 0; i <= IO_BITMAP_LONGS; i++)
1654 		t->io_bitmap[i] = ~0UL;
1655 
1656 	mmgrab(&init_mm);
1657 	me->active_mm = &init_mm;
1658 	BUG_ON(me->mm);
1659 	initialize_tlbstate_and_flush();
1660 	enter_lazy_tlb(&init_mm, me);
1661 
1662 	/*
1663 	 * Initialize the TSS.  sp0 points to the entry trampoline stack
1664 	 * regardless of what task is running.
1665 	 */
1666 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1667 	load_TR_desc();
1668 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1669 
1670 	load_mm_ldt(&init_mm);
1671 
1672 	clear_all_debug_regs();
1673 	dbg_restore_debug_regs();
1674 
1675 	fpu__init_cpu();
1676 
1677 	if (is_uv_system())
1678 		uv_cpu_init();
1679 
1680 	load_fixmap_gdt(cpu);
1681 }
1682 
1683 #else
1684 
1685 void cpu_init(void)
1686 {
1687 	int cpu = smp_processor_id();
1688 	struct task_struct *curr = current;
1689 	struct tss_struct *t = &per_cpu(cpu_tss_rw, cpu);
1690 
1691 	wait_for_master_cpu(cpu);
1692 
1693 	/*
1694 	 * Initialize the CR4 shadow before doing anything that could
1695 	 * try to read it.
1696 	 */
1697 	cr4_init_shadow();
1698 
1699 	show_ucode_info_early();
1700 
1701 	pr_info("Initializing CPU#%d\n", cpu);
1702 
1703 	if (cpu_feature_enabled(X86_FEATURE_VME) ||
1704 	    boot_cpu_has(X86_FEATURE_TSC) ||
1705 	    boot_cpu_has(X86_FEATURE_DE))
1706 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1707 
1708 	load_current_idt();
1709 	switch_to_new_gdt(cpu);
1710 
1711 	/*
1712 	 * Set up and load the per-CPU TSS and LDT
1713 	 */
1714 	mmgrab(&init_mm);
1715 	curr->active_mm = &init_mm;
1716 	BUG_ON(curr->mm);
1717 	initialize_tlbstate_and_flush();
1718 	enter_lazy_tlb(&init_mm, curr);
1719 
1720 	/*
1721 	 * Initialize the TSS.  Don't bother initializing sp0, as the initial
1722 	 * task never enters user mode.
1723 	 */
1724 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1725 	load_TR_desc();
1726 
1727 	load_mm_ldt(&init_mm);
1728 
1729 	t->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET;
1730 
1731 #ifdef CONFIG_DOUBLEFAULT
1732 	/* Set up doublefault TSS pointer in the GDT */
1733 	__set_tss_desc(cpu, GDT_ENTRY_DOUBLEFAULT_TSS, &doublefault_tss);
1734 #endif
1735 
1736 	clear_all_debug_regs();
1737 	dbg_restore_debug_regs();
1738 
1739 	fpu__init_cpu();
1740 
1741 	load_fixmap_gdt(cpu);
1742 }
1743 #endif
1744 
1745 static void bsp_resume(void)
1746 {
1747 	if (this_cpu->c_bsp_resume)
1748 		this_cpu->c_bsp_resume(&boot_cpu_data);
1749 }
1750 
1751 static struct syscore_ops cpu_syscore_ops = {
1752 	.resume		= bsp_resume,
1753 };
1754 
1755 static int __init init_cpu_syscore(void)
1756 {
1757 	register_syscore_ops(&cpu_syscore_ops);
1758 	return 0;
1759 }
1760 core_initcall(init_cpu_syscore);
1761 
1762 /*
1763  * The microcode loader calls this upon late microcode load to recheck features,
1764  * only when microcode has been updated. Caller holds microcode_mutex and CPU
1765  * hotplug lock.
1766  */
1767 void microcode_check(void)
1768 {
1769 	struct cpuinfo_x86 info;
1770 
1771 	perf_check_microcode();
1772 
1773 	/* Reload CPUID max function as it might've changed. */
1774 	info.cpuid_level = cpuid_eax(0);
1775 
1776 	/*
1777 	 * Copy all capability leafs to pick up the synthetic ones so that
1778 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
1779 	 * get overwritten in get_cpu_cap().
1780 	 */
1781 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
1782 
1783 	get_cpu_cap(&info);
1784 
1785 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
1786 		return;
1787 
1788 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
1789 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
1790 }
1791