xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 276e552e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* cpu_feature_enabled() cannot be used this early */
3 #define USE_EARLY_PGTABLE_L5
4 
5 #include <linux/memblock.h>
6 #include <linux/linkage.h>
7 #include <linux/bitops.h>
8 #include <linux/kernel.h>
9 #include <linux/export.h>
10 #include <linux/percpu.h>
11 #include <linux/string.h>
12 #include <linux/ctype.h>
13 #include <linux/delay.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/clock.h>
16 #include <linux/sched/task.h>
17 #include <linux/sched/smt.h>
18 #include <linux/init.h>
19 #include <linux/kprobes.h>
20 #include <linux/kgdb.h>
21 #include <linux/smp.h>
22 #include <linux/io.h>
23 #include <linux/syscore_ops.h>
24 #include <linux/pgtable.h>
25 
26 #include <asm/cmdline.h>
27 #include <asm/stackprotector.h>
28 #include <asm/perf_event.h>
29 #include <asm/mmu_context.h>
30 #include <asm/doublefault.h>
31 #include <asm/archrandom.h>
32 #include <asm/hypervisor.h>
33 #include <asm/processor.h>
34 #include <asm/tlbflush.h>
35 #include <asm/debugreg.h>
36 #include <asm/sections.h>
37 #include <asm/vsyscall.h>
38 #include <linux/topology.h>
39 #include <linux/cpumask.h>
40 #include <linux/atomic.h>
41 #include <asm/proto.h>
42 #include <asm/setup.h>
43 #include <asm/apic.h>
44 #include <asm/desc.h>
45 #include <asm/fpu/internal.h>
46 #include <asm/mtrr.h>
47 #include <asm/hwcap2.h>
48 #include <linux/numa.h>
49 #include <asm/numa.h>
50 #include <asm/asm.h>
51 #include <asm/bugs.h>
52 #include <asm/cpu.h>
53 #include <asm/mce.h>
54 #include <asm/msr.h>
55 #include <asm/memtype.h>
56 #include <asm/microcode.h>
57 #include <asm/microcode_intel.h>
58 #include <asm/intel-family.h>
59 #include <asm/cpu_device_id.h>
60 #include <asm/uv/uv.h>
61 
62 #include "cpu.h"
63 
64 u32 elf_hwcap2 __read_mostly;
65 
66 /* all of these masks are initialized in setup_cpu_local_masks() */
67 cpumask_var_t cpu_initialized_mask;
68 cpumask_var_t cpu_callout_mask;
69 cpumask_var_t cpu_callin_mask;
70 
71 /* representing cpus for which sibling maps can be computed */
72 cpumask_var_t cpu_sibling_setup_mask;
73 
74 /* Number of siblings per CPU package */
75 int smp_num_siblings = 1;
76 EXPORT_SYMBOL(smp_num_siblings);
77 
78 /* Last level cache ID of each logical CPU */
79 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
80 
81 /* correctly size the local cpu masks */
82 void __init setup_cpu_local_masks(void)
83 {
84 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
85 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
86 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
87 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
88 }
89 
90 static void default_init(struct cpuinfo_x86 *c)
91 {
92 #ifdef CONFIG_X86_64
93 	cpu_detect_cache_sizes(c);
94 #else
95 	/* Not much we can do here... */
96 	/* Check if at least it has cpuid */
97 	if (c->cpuid_level == -1) {
98 		/* No cpuid. It must be an ancient CPU */
99 		if (c->x86 == 4)
100 			strcpy(c->x86_model_id, "486");
101 		else if (c->x86 == 3)
102 			strcpy(c->x86_model_id, "386");
103 	}
104 #endif
105 }
106 
107 static const struct cpu_dev default_cpu = {
108 	.c_init		= default_init,
109 	.c_vendor	= "Unknown",
110 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
111 };
112 
113 static const struct cpu_dev *this_cpu = &default_cpu;
114 
115 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
116 #ifdef CONFIG_X86_64
117 	/*
118 	 * We need valid kernel segments for data and code in long mode too
119 	 * IRET will check the segment types  kkeil 2000/10/28
120 	 * Also sysret mandates a special GDT layout
121 	 *
122 	 * TLS descriptors are currently at a different place compared to i386.
123 	 * Hopefully nobody expects them at a fixed place (Wine?)
124 	 */
125 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
126 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
127 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
128 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
129 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
130 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
131 #else
132 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
133 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
134 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
135 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
136 	/*
137 	 * Segments used for calling PnP BIOS have byte granularity.
138 	 * They code segments and data segments have fixed 64k limits,
139 	 * the transfer segment sizes are set at run time.
140 	 */
141 	/* 32-bit code */
142 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
143 	/* 16-bit code */
144 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
145 	/* 16-bit data */
146 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
147 	/* 16-bit data */
148 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
149 	/* 16-bit data */
150 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
151 	/*
152 	 * The APM segments have byte granularity and their bases
153 	 * are set at run time.  All have 64k limits.
154 	 */
155 	/* 32-bit code */
156 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
157 	/* 16-bit code */
158 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
159 	/* data */
160 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
161 
162 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
163 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
164 #endif
165 } };
166 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
167 
168 #ifdef CONFIG_X86_64
169 static int __init x86_nopcid_setup(char *s)
170 {
171 	/* nopcid doesn't accept parameters */
172 	if (s)
173 		return -EINVAL;
174 
175 	/* do not emit a message if the feature is not present */
176 	if (!boot_cpu_has(X86_FEATURE_PCID))
177 		return 0;
178 
179 	setup_clear_cpu_cap(X86_FEATURE_PCID);
180 	pr_info("nopcid: PCID feature disabled\n");
181 	return 0;
182 }
183 early_param("nopcid", x86_nopcid_setup);
184 #endif
185 
186 static int __init x86_noinvpcid_setup(char *s)
187 {
188 	/* noinvpcid doesn't accept parameters */
189 	if (s)
190 		return -EINVAL;
191 
192 	/* do not emit a message if the feature is not present */
193 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
194 		return 0;
195 
196 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
197 	pr_info("noinvpcid: INVPCID feature disabled\n");
198 	return 0;
199 }
200 early_param("noinvpcid", x86_noinvpcid_setup);
201 
202 #ifdef CONFIG_X86_32
203 static int cachesize_override = -1;
204 static int disable_x86_serial_nr = 1;
205 
206 static int __init cachesize_setup(char *str)
207 {
208 	get_option(&str, &cachesize_override);
209 	return 1;
210 }
211 __setup("cachesize=", cachesize_setup);
212 
213 static int __init x86_sep_setup(char *s)
214 {
215 	setup_clear_cpu_cap(X86_FEATURE_SEP);
216 	return 1;
217 }
218 __setup("nosep", x86_sep_setup);
219 
220 /* Standard macro to see if a specific flag is changeable */
221 static inline int flag_is_changeable_p(u32 flag)
222 {
223 	u32 f1, f2;
224 
225 	/*
226 	 * Cyrix and IDT cpus allow disabling of CPUID
227 	 * so the code below may return different results
228 	 * when it is executed before and after enabling
229 	 * the CPUID. Add "volatile" to not allow gcc to
230 	 * optimize the subsequent calls to this function.
231 	 */
232 	asm volatile ("pushfl		\n\t"
233 		      "pushfl		\n\t"
234 		      "popl %0		\n\t"
235 		      "movl %0, %1	\n\t"
236 		      "xorl %2, %0	\n\t"
237 		      "pushl %0		\n\t"
238 		      "popfl		\n\t"
239 		      "pushfl		\n\t"
240 		      "popl %0		\n\t"
241 		      "popfl		\n\t"
242 
243 		      : "=&r" (f1), "=&r" (f2)
244 		      : "ir" (flag));
245 
246 	return ((f1^f2) & flag) != 0;
247 }
248 
249 /* Probe for the CPUID instruction */
250 int have_cpuid_p(void)
251 {
252 	return flag_is_changeable_p(X86_EFLAGS_ID);
253 }
254 
255 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
256 {
257 	unsigned long lo, hi;
258 
259 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
260 		return;
261 
262 	/* Disable processor serial number: */
263 
264 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
265 	lo |= 0x200000;
266 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
267 
268 	pr_notice("CPU serial number disabled.\n");
269 	clear_cpu_cap(c, X86_FEATURE_PN);
270 
271 	/* Disabling the serial number may affect the cpuid level */
272 	c->cpuid_level = cpuid_eax(0);
273 }
274 
275 static int __init x86_serial_nr_setup(char *s)
276 {
277 	disable_x86_serial_nr = 0;
278 	return 1;
279 }
280 __setup("serialnumber", x86_serial_nr_setup);
281 #else
282 static inline int flag_is_changeable_p(u32 flag)
283 {
284 	return 1;
285 }
286 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
287 {
288 }
289 #endif
290 
291 static __init int setup_disable_smep(char *arg)
292 {
293 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
294 	return 1;
295 }
296 __setup("nosmep", setup_disable_smep);
297 
298 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
299 {
300 	if (cpu_has(c, X86_FEATURE_SMEP))
301 		cr4_set_bits(X86_CR4_SMEP);
302 }
303 
304 static __init int setup_disable_smap(char *arg)
305 {
306 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
307 	return 1;
308 }
309 __setup("nosmap", setup_disable_smap);
310 
311 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
312 {
313 	unsigned long eflags = native_save_fl();
314 
315 	/* This should have been cleared long ago */
316 	BUG_ON(eflags & X86_EFLAGS_AC);
317 
318 	if (cpu_has(c, X86_FEATURE_SMAP)) {
319 #ifdef CONFIG_X86_SMAP
320 		cr4_set_bits(X86_CR4_SMAP);
321 #else
322 		cr4_clear_bits(X86_CR4_SMAP);
323 #endif
324 	}
325 }
326 
327 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
328 {
329 	/* Check the boot processor, plus build option for UMIP. */
330 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
331 		goto out;
332 
333 	/* Check the current processor's cpuid bits. */
334 	if (!cpu_has(c, X86_FEATURE_UMIP))
335 		goto out;
336 
337 	cr4_set_bits(X86_CR4_UMIP);
338 
339 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
340 
341 	return;
342 
343 out:
344 	/*
345 	 * Make sure UMIP is disabled in case it was enabled in a
346 	 * previous boot (e.g., via kexec).
347 	 */
348 	cr4_clear_bits(X86_CR4_UMIP);
349 }
350 
351 /* These bits should not change their value after CPU init is finished. */
352 static const unsigned long cr4_pinned_mask =
353 	X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP | X86_CR4_FSGSBASE;
354 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
355 static unsigned long cr4_pinned_bits __ro_after_init;
356 
357 void native_write_cr0(unsigned long val)
358 {
359 	unsigned long bits_missing = 0;
360 
361 set_register:
362 	asm volatile("mov %0,%%cr0": "+r" (val) : : "memory");
363 
364 	if (static_branch_likely(&cr_pinning)) {
365 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
366 			bits_missing = X86_CR0_WP;
367 			val |= bits_missing;
368 			goto set_register;
369 		}
370 		/* Warn after we've set the missing bits. */
371 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
372 	}
373 }
374 EXPORT_SYMBOL(native_write_cr0);
375 
376 void native_write_cr4(unsigned long val)
377 {
378 	unsigned long bits_changed = 0;
379 
380 set_register:
381 	asm volatile("mov %0,%%cr4": "+r" (val) : : "memory");
382 
383 	if (static_branch_likely(&cr_pinning)) {
384 		if (unlikely((val & cr4_pinned_mask) != cr4_pinned_bits)) {
385 			bits_changed = (val & cr4_pinned_mask) ^ cr4_pinned_bits;
386 			val = (val & ~cr4_pinned_mask) | cr4_pinned_bits;
387 			goto set_register;
388 		}
389 		/* Warn after we've corrected the changed bits. */
390 		WARN_ONCE(bits_changed, "pinned CR4 bits changed: 0x%lx!?\n",
391 			  bits_changed);
392 	}
393 }
394 #if IS_MODULE(CONFIG_LKDTM)
395 EXPORT_SYMBOL_GPL(native_write_cr4);
396 #endif
397 
398 void cr4_update_irqsoff(unsigned long set, unsigned long clear)
399 {
400 	unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
401 
402 	lockdep_assert_irqs_disabled();
403 
404 	newval = (cr4 & ~clear) | set;
405 	if (newval != cr4) {
406 		this_cpu_write(cpu_tlbstate.cr4, newval);
407 		__write_cr4(newval);
408 	}
409 }
410 EXPORT_SYMBOL(cr4_update_irqsoff);
411 
412 /* Read the CR4 shadow. */
413 unsigned long cr4_read_shadow(void)
414 {
415 	return this_cpu_read(cpu_tlbstate.cr4);
416 }
417 EXPORT_SYMBOL_GPL(cr4_read_shadow);
418 
419 void cr4_init(void)
420 {
421 	unsigned long cr4 = __read_cr4();
422 
423 	if (boot_cpu_has(X86_FEATURE_PCID))
424 		cr4 |= X86_CR4_PCIDE;
425 	if (static_branch_likely(&cr_pinning))
426 		cr4 = (cr4 & ~cr4_pinned_mask) | cr4_pinned_bits;
427 
428 	__write_cr4(cr4);
429 
430 	/* Initialize cr4 shadow for this CPU. */
431 	this_cpu_write(cpu_tlbstate.cr4, cr4);
432 }
433 
434 /*
435  * Once CPU feature detection is finished (and boot params have been
436  * parsed), record any of the sensitive CR bits that are set, and
437  * enable CR pinning.
438  */
439 static void __init setup_cr_pinning(void)
440 {
441 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & cr4_pinned_mask;
442 	static_key_enable(&cr_pinning.key);
443 }
444 
445 static __init int x86_nofsgsbase_setup(char *arg)
446 {
447 	/* Require an exact match without trailing characters. */
448 	if (strlen(arg))
449 		return 0;
450 
451 	/* Do not emit a message if the feature is not present. */
452 	if (!boot_cpu_has(X86_FEATURE_FSGSBASE))
453 		return 1;
454 
455 	setup_clear_cpu_cap(X86_FEATURE_FSGSBASE);
456 	pr_info("FSGSBASE disabled via kernel command line\n");
457 	return 1;
458 }
459 __setup("nofsgsbase", x86_nofsgsbase_setup);
460 
461 /*
462  * Protection Keys are not available in 32-bit mode.
463  */
464 static bool pku_disabled;
465 
466 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
467 {
468 	struct pkru_state *pk;
469 
470 	/* check the boot processor, plus compile options for PKU: */
471 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
472 		return;
473 	/* checks the actual processor's cpuid bits: */
474 	if (!cpu_has(c, X86_FEATURE_PKU))
475 		return;
476 	if (pku_disabled)
477 		return;
478 
479 	cr4_set_bits(X86_CR4_PKE);
480 	pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU);
481 	if (pk)
482 		pk->pkru = init_pkru_value;
483 	/*
484 	 * Setting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
485 	 * cpuid bit to be set.  We need to ensure that we
486 	 * update that bit in this CPU's "cpu_info".
487 	 */
488 	set_cpu_cap(c, X86_FEATURE_OSPKE);
489 }
490 
491 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
492 static __init int setup_disable_pku(char *arg)
493 {
494 	/*
495 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
496 	 * runtime checks are against OSPKE so clearing the
497 	 * bit does nothing.
498 	 *
499 	 * This way, we will see "pku" in cpuinfo, but not
500 	 * "ospke", which is exactly what we want.  It shows
501 	 * that the CPU has PKU, but the OS has not enabled it.
502 	 * This happens to be exactly how a system would look
503 	 * if we disabled the config option.
504 	 */
505 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
506 	pku_disabled = true;
507 	return 1;
508 }
509 __setup("nopku", setup_disable_pku);
510 #endif /* CONFIG_X86_64 */
511 
512 /*
513  * Some CPU features depend on higher CPUID levels, which may not always
514  * be available due to CPUID level capping or broken virtualization
515  * software.  Add those features to this table to auto-disable them.
516  */
517 struct cpuid_dependent_feature {
518 	u32 feature;
519 	u32 level;
520 };
521 
522 static const struct cpuid_dependent_feature
523 cpuid_dependent_features[] = {
524 	{ X86_FEATURE_MWAIT,		0x00000005 },
525 	{ X86_FEATURE_DCA,		0x00000009 },
526 	{ X86_FEATURE_XSAVE,		0x0000000d },
527 	{ 0, 0 }
528 };
529 
530 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
531 {
532 	const struct cpuid_dependent_feature *df;
533 
534 	for (df = cpuid_dependent_features; df->feature; df++) {
535 
536 		if (!cpu_has(c, df->feature))
537 			continue;
538 		/*
539 		 * Note: cpuid_level is set to -1 if unavailable, but
540 		 * extended_extended_level is set to 0 if unavailable
541 		 * and the legitimate extended levels are all negative
542 		 * when signed; hence the weird messing around with
543 		 * signs here...
544 		 */
545 		if (!((s32)df->level < 0 ?
546 		     (u32)df->level > (u32)c->extended_cpuid_level :
547 		     (s32)df->level > (s32)c->cpuid_level))
548 			continue;
549 
550 		clear_cpu_cap(c, df->feature);
551 		if (!warn)
552 			continue;
553 
554 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
555 			x86_cap_flag(df->feature), df->level);
556 	}
557 }
558 
559 /*
560  * Naming convention should be: <Name> [(<Codename>)]
561  * This table only is used unless init_<vendor>() below doesn't set it;
562  * in particular, if CPUID levels 0x80000002..4 are supported, this
563  * isn't used
564  */
565 
566 /* Look up CPU names by table lookup. */
567 static const char *table_lookup_model(struct cpuinfo_x86 *c)
568 {
569 #ifdef CONFIG_X86_32
570 	const struct legacy_cpu_model_info *info;
571 
572 	if (c->x86_model >= 16)
573 		return NULL;	/* Range check */
574 
575 	if (!this_cpu)
576 		return NULL;
577 
578 	info = this_cpu->legacy_models;
579 
580 	while (info->family) {
581 		if (info->family == c->x86)
582 			return info->model_names[c->x86_model];
583 		info++;
584 	}
585 #endif
586 	return NULL;		/* Not found */
587 }
588 
589 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
590 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
591 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
592 
593 void load_percpu_segment(int cpu)
594 {
595 #ifdef CONFIG_X86_32
596 	loadsegment(fs, __KERNEL_PERCPU);
597 #else
598 	__loadsegment_simple(gs, 0);
599 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
600 #endif
601 }
602 
603 #ifdef CONFIG_X86_32
604 /* The 32-bit entry code needs to find cpu_entry_area. */
605 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
606 #endif
607 
608 /* Load the original GDT from the per-cpu structure */
609 void load_direct_gdt(int cpu)
610 {
611 	struct desc_ptr gdt_descr;
612 
613 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
614 	gdt_descr.size = GDT_SIZE - 1;
615 	load_gdt(&gdt_descr);
616 }
617 EXPORT_SYMBOL_GPL(load_direct_gdt);
618 
619 /* Load a fixmap remapping of the per-cpu GDT */
620 void load_fixmap_gdt(int cpu)
621 {
622 	struct desc_ptr gdt_descr;
623 
624 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
625 	gdt_descr.size = GDT_SIZE - 1;
626 	load_gdt(&gdt_descr);
627 }
628 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
629 
630 /*
631  * Current gdt points %fs at the "master" per-cpu area: after this,
632  * it's on the real one.
633  */
634 void switch_to_new_gdt(int cpu)
635 {
636 	/* Load the original GDT */
637 	load_direct_gdt(cpu);
638 	/* Reload the per-cpu base */
639 	load_percpu_segment(cpu);
640 }
641 
642 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
643 
644 static void get_model_name(struct cpuinfo_x86 *c)
645 {
646 	unsigned int *v;
647 	char *p, *q, *s;
648 
649 	if (c->extended_cpuid_level < 0x80000004)
650 		return;
651 
652 	v = (unsigned int *)c->x86_model_id;
653 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
654 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
655 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
656 	c->x86_model_id[48] = 0;
657 
658 	/* Trim whitespace */
659 	p = q = s = &c->x86_model_id[0];
660 
661 	while (*p == ' ')
662 		p++;
663 
664 	while (*p) {
665 		/* Note the last non-whitespace index */
666 		if (!isspace(*p))
667 			s = q;
668 
669 		*q++ = *p++;
670 	}
671 
672 	*(s + 1) = '\0';
673 }
674 
675 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
676 {
677 	unsigned int eax, ebx, ecx, edx;
678 
679 	c->x86_max_cores = 1;
680 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
681 		return;
682 
683 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
684 	if (eax & 0x1f)
685 		c->x86_max_cores = (eax >> 26) + 1;
686 }
687 
688 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
689 {
690 	unsigned int n, dummy, ebx, ecx, edx, l2size;
691 
692 	n = c->extended_cpuid_level;
693 
694 	if (n >= 0x80000005) {
695 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
696 		c->x86_cache_size = (ecx>>24) + (edx>>24);
697 #ifdef CONFIG_X86_64
698 		/* On K8 L1 TLB is inclusive, so don't count it */
699 		c->x86_tlbsize = 0;
700 #endif
701 	}
702 
703 	if (n < 0x80000006)	/* Some chips just has a large L1. */
704 		return;
705 
706 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
707 	l2size = ecx >> 16;
708 
709 #ifdef CONFIG_X86_64
710 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
711 #else
712 	/* do processor-specific cache resizing */
713 	if (this_cpu->legacy_cache_size)
714 		l2size = this_cpu->legacy_cache_size(c, l2size);
715 
716 	/* Allow user to override all this if necessary. */
717 	if (cachesize_override != -1)
718 		l2size = cachesize_override;
719 
720 	if (l2size == 0)
721 		return;		/* Again, no L2 cache is possible */
722 #endif
723 
724 	c->x86_cache_size = l2size;
725 }
726 
727 u16 __read_mostly tlb_lli_4k[NR_INFO];
728 u16 __read_mostly tlb_lli_2m[NR_INFO];
729 u16 __read_mostly tlb_lli_4m[NR_INFO];
730 u16 __read_mostly tlb_lld_4k[NR_INFO];
731 u16 __read_mostly tlb_lld_2m[NR_INFO];
732 u16 __read_mostly tlb_lld_4m[NR_INFO];
733 u16 __read_mostly tlb_lld_1g[NR_INFO];
734 
735 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
736 {
737 	if (this_cpu->c_detect_tlb)
738 		this_cpu->c_detect_tlb(c);
739 
740 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
741 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
742 		tlb_lli_4m[ENTRIES]);
743 
744 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
745 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
746 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
747 }
748 
749 int detect_ht_early(struct cpuinfo_x86 *c)
750 {
751 #ifdef CONFIG_SMP
752 	u32 eax, ebx, ecx, edx;
753 
754 	if (!cpu_has(c, X86_FEATURE_HT))
755 		return -1;
756 
757 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
758 		return -1;
759 
760 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
761 		return -1;
762 
763 	cpuid(1, &eax, &ebx, &ecx, &edx);
764 
765 	smp_num_siblings = (ebx & 0xff0000) >> 16;
766 	if (smp_num_siblings == 1)
767 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
768 #endif
769 	return 0;
770 }
771 
772 void detect_ht(struct cpuinfo_x86 *c)
773 {
774 #ifdef CONFIG_SMP
775 	int index_msb, core_bits;
776 
777 	if (detect_ht_early(c) < 0)
778 		return;
779 
780 	index_msb = get_count_order(smp_num_siblings);
781 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
782 
783 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
784 
785 	index_msb = get_count_order(smp_num_siblings);
786 
787 	core_bits = get_count_order(c->x86_max_cores);
788 
789 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
790 				       ((1 << core_bits) - 1);
791 #endif
792 }
793 
794 static void get_cpu_vendor(struct cpuinfo_x86 *c)
795 {
796 	char *v = c->x86_vendor_id;
797 	int i;
798 
799 	for (i = 0; i < X86_VENDOR_NUM; i++) {
800 		if (!cpu_devs[i])
801 			break;
802 
803 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
804 		    (cpu_devs[i]->c_ident[1] &&
805 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
806 
807 			this_cpu = cpu_devs[i];
808 			c->x86_vendor = this_cpu->c_x86_vendor;
809 			return;
810 		}
811 	}
812 
813 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
814 		    "CPU: Your system may be unstable.\n", v);
815 
816 	c->x86_vendor = X86_VENDOR_UNKNOWN;
817 	this_cpu = &default_cpu;
818 }
819 
820 void cpu_detect(struct cpuinfo_x86 *c)
821 {
822 	/* Get vendor name */
823 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
824 	      (unsigned int *)&c->x86_vendor_id[0],
825 	      (unsigned int *)&c->x86_vendor_id[8],
826 	      (unsigned int *)&c->x86_vendor_id[4]);
827 
828 	c->x86 = 4;
829 	/* Intel-defined flags: level 0x00000001 */
830 	if (c->cpuid_level >= 0x00000001) {
831 		u32 junk, tfms, cap0, misc;
832 
833 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
834 		c->x86		= x86_family(tfms);
835 		c->x86_model	= x86_model(tfms);
836 		c->x86_stepping	= x86_stepping(tfms);
837 
838 		if (cap0 & (1<<19)) {
839 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
840 			c->x86_cache_alignment = c->x86_clflush_size;
841 		}
842 	}
843 }
844 
845 static void apply_forced_caps(struct cpuinfo_x86 *c)
846 {
847 	int i;
848 
849 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
850 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
851 		c->x86_capability[i] |= cpu_caps_set[i];
852 	}
853 }
854 
855 static void init_speculation_control(struct cpuinfo_x86 *c)
856 {
857 	/*
858 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
859 	 * and they also have a different bit for STIBP support. Also,
860 	 * a hypervisor might have set the individual AMD bits even on
861 	 * Intel CPUs, for finer-grained selection of what's available.
862 	 */
863 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
864 		set_cpu_cap(c, X86_FEATURE_IBRS);
865 		set_cpu_cap(c, X86_FEATURE_IBPB);
866 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
867 	}
868 
869 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
870 		set_cpu_cap(c, X86_FEATURE_STIBP);
871 
872 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
873 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
874 		set_cpu_cap(c, X86_FEATURE_SSBD);
875 
876 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
877 		set_cpu_cap(c, X86_FEATURE_IBRS);
878 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
879 	}
880 
881 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
882 		set_cpu_cap(c, X86_FEATURE_IBPB);
883 
884 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
885 		set_cpu_cap(c, X86_FEATURE_STIBP);
886 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
887 	}
888 
889 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
890 		set_cpu_cap(c, X86_FEATURE_SSBD);
891 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
892 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
893 	}
894 }
895 
896 void get_cpu_cap(struct cpuinfo_x86 *c)
897 {
898 	u32 eax, ebx, ecx, edx;
899 
900 	/* Intel-defined flags: level 0x00000001 */
901 	if (c->cpuid_level >= 0x00000001) {
902 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
903 
904 		c->x86_capability[CPUID_1_ECX] = ecx;
905 		c->x86_capability[CPUID_1_EDX] = edx;
906 	}
907 
908 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
909 	if (c->cpuid_level >= 0x00000006)
910 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
911 
912 	/* Additional Intel-defined flags: level 0x00000007 */
913 	if (c->cpuid_level >= 0x00000007) {
914 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
915 		c->x86_capability[CPUID_7_0_EBX] = ebx;
916 		c->x86_capability[CPUID_7_ECX] = ecx;
917 		c->x86_capability[CPUID_7_EDX] = edx;
918 
919 		/* Check valid sub-leaf index before accessing it */
920 		if (eax >= 1) {
921 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
922 			c->x86_capability[CPUID_7_1_EAX] = eax;
923 		}
924 	}
925 
926 	/* Extended state features: level 0x0000000d */
927 	if (c->cpuid_level >= 0x0000000d) {
928 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
929 
930 		c->x86_capability[CPUID_D_1_EAX] = eax;
931 	}
932 
933 	/* AMD-defined flags: level 0x80000001 */
934 	eax = cpuid_eax(0x80000000);
935 	c->extended_cpuid_level = eax;
936 
937 	if ((eax & 0xffff0000) == 0x80000000) {
938 		if (eax >= 0x80000001) {
939 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
940 
941 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
942 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
943 		}
944 	}
945 
946 	if (c->extended_cpuid_level >= 0x80000007) {
947 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
948 
949 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
950 		c->x86_power = edx;
951 	}
952 
953 	if (c->extended_cpuid_level >= 0x80000008) {
954 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
955 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
956 	}
957 
958 	if (c->extended_cpuid_level >= 0x8000000a)
959 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
960 
961 	if (c->extended_cpuid_level >= 0x8000001f)
962 		c->x86_capability[CPUID_8000_001F_EAX] = cpuid_eax(0x8000001f);
963 
964 	init_scattered_cpuid_features(c);
965 	init_speculation_control(c);
966 
967 	/*
968 	 * Clear/Set all flags overridden by options, after probe.
969 	 * This needs to happen each time we re-probe, which may happen
970 	 * several times during CPU initialization.
971 	 */
972 	apply_forced_caps(c);
973 }
974 
975 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
976 {
977 	u32 eax, ebx, ecx, edx;
978 
979 	if (c->extended_cpuid_level >= 0x80000008) {
980 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
981 
982 		c->x86_virt_bits = (eax >> 8) & 0xff;
983 		c->x86_phys_bits = eax & 0xff;
984 	}
985 #ifdef CONFIG_X86_32
986 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
987 		c->x86_phys_bits = 36;
988 #endif
989 	c->x86_cache_bits = c->x86_phys_bits;
990 }
991 
992 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
993 {
994 #ifdef CONFIG_X86_32
995 	int i;
996 
997 	/*
998 	 * First of all, decide if this is a 486 or higher
999 	 * It's a 486 if we can modify the AC flag
1000 	 */
1001 	if (flag_is_changeable_p(X86_EFLAGS_AC))
1002 		c->x86 = 4;
1003 	else
1004 		c->x86 = 3;
1005 
1006 	for (i = 0; i < X86_VENDOR_NUM; i++)
1007 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
1008 			c->x86_vendor_id[0] = 0;
1009 			cpu_devs[i]->c_identify(c);
1010 			if (c->x86_vendor_id[0]) {
1011 				get_cpu_vendor(c);
1012 				break;
1013 			}
1014 		}
1015 #endif
1016 }
1017 
1018 #define NO_SPECULATION		BIT(0)
1019 #define NO_MELTDOWN		BIT(1)
1020 #define NO_SSB			BIT(2)
1021 #define NO_L1TF			BIT(3)
1022 #define NO_MDS			BIT(4)
1023 #define MSBDS_ONLY		BIT(5)
1024 #define NO_SWAPGS		BIT(6)
1025 #define NO_ITLB_MULTIHIT	BIT(7)
1026 #define NO_SPECTRE_V2		BIT(8)
1027 
1028 #define VULNWL(vendor, family, model, whitelist)	\
1029 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist)
1030 
1031 #define VULNWL_INTEL(model, whitelist)		\
1032 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1033 
1034 #define VULNWL_AMD(family, whitelist)		\
1035 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1036 
1037 #define VULNWL_HYGON(family, whitelist)		\
1038 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1039 
1040 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1041 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1042 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1043 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1044 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1045 
1046 	/* Intel Family 6 */
1047 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1048 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION | NO_ITLB_MULTIHIT),
1049 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1050 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1051 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1052 
1053 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1054 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1055 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1056 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1057 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1058 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1059 
1060 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1061 
1062 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1063 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1064 
1065 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1066 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1067 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1068 
1069 	/*
1070 	 * Technically, swapgs isn't serializing on AMD (despite it previously
1071 	 * being documented as such in the APM).  But according to AMD, %gs is
1072 	 * updated non-speculatively, and the issuing of %gs-relative memory
1073 	 * operands will be blocked until the %gs update completes, which is
1074 	 * good enough for our purposes.
1075 	 */
1076 
1077 	VULNWL_INTEL(ATOM_TREMONT_D,		NO_ITLB_MULTIHIT),
1078 
1079 	/* AMD Family 0xf - 0x12 */
1080 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1081 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1082 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1083 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1084 
1085 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1086 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1087 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1088 
1089 	/* Zhaoxin Family 7 */
1090 	VULNWL(CENTAUR,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1091 	VULNWL(ZHAOXIN,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1092 	{}
1093 };
1094 
1095 #define VULNBL_INTEL_STEPPINGS(model, steppings, issues)		   \
1096 	X86_MATCH_VENDOR_FAM_MODEL_STEPPINGS_FEATURE(INTEL, 6,		   \
1097 					    INTEL_FAM6_##model, steppings, \
1098 					    X86_FEATURE_ANY, issues)
1099 
1100 #define SRBDS		BIT(0)
1101 
1102 static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = {
1103 	VULNBL_INTEL_STEPPINGS(IVYBRIDGE,	X86_STEPPING_ANY,		SRBDS),
1104 	VULNBL_INTEL_STEPPINGS(HASWELL,		X86_STEPPING_ANY,		SRBDS),
1105 	VULNBL_INTEL_STEPPINGS(HASWELL_L,	X86_STEPPING_ANY,		SRBDS),
1106 	VULNBL_INTEL_STEPPINGS(HASWELL_G,	X86_STEPPING_ANY,		SRBDS),
1107 	VULNBL_INTEL_STEPPINGS(BROADWELL_G,	X86_STEPPING_ANY,		SRBDS),
1108 	VULNBL_INTEL_STEPPINGS(BROADWELL,	X86_STEPPING_ANY,		SRBDS),
1109 	VULNBL_INTEL_STEPPINGS(SKYLAKE_L,	X86_STEPPING_ANY,		SRBDS),
1110 	VULNBL_INTEL_STEPPINGS(SKYLAKE,		X86_STEPPING_ANY,		SRBDS),
1111 	VULNBL_INTEL_STEPPINGS(KABYLAKE_L,	X86_STEPPINGS(0x0, 0xC),	SRBDS),
1112 	VULNBL_INTEL_STEPPINGS(KABYLAKE,	X86_STEPPINGS(0x0, 0xD),	SRBDS),
1113 	{}
1114 };
1115 
1116 static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which)
1117 {
1118 	const struct x86_cpu_id *m = x86_match_cpu(table);
1119 
1120 	return m && !!(m->driver_data & which);
1121 }
1122 
1123 u64 x86_read_arch_cap_msr(void)
1124 {
1125 	u64 ia32_cap = 0;
1126 
1127 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1128 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1129 
1130 	return ia32_cap;
1131 }
1132 
1133 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1134 {
1135 	u64 ia32_cap = x86_read_arch_cap_msr();
1136 
1137 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1138 	if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) &&
1139 	    !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO))
1140 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1141 
1142 	if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION))
1143 		return;
1144 
1145 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1146 
1147 	if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2))
1148 		setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1149 
1150 	if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) &&
1151 	    !(ia32_cap & ARCH_CAP_SSB_NO) &&
1152 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1153 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1154 
1155 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1156 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1157 
1158 	if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) &&
1159 	    !(ia32_cap & ARCH_CAP_MDS_NO)) {
1160 		setup_force_cpu_bug(X86_BUG_MDS);
1161 		if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY))
1162 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1163 	}
1164 
1165 	if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS))
1166 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1167 
1168 	/*
1169 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1170 	 *	- TSX is supported or
1171 	 *	- TSX_CTRL is present
1172 	 *
1173 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1174 	 * the kernel boot e.g. kexec.
1175 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1176 	 * update is not present or running as guest that don't get TSX_CTRL.
1177 	 */
1178 	if (!(ia32_cap & ARCH_CAP_TAA_NO) &&
1179 	    (cpu_has(c, X86_FEATURE_RTM) ||
1180 	     (ia32_cap & ARCH_CAP_TSX_CTRL_MSR)))
1181 		setup_force_cpu_bug(X86_BUG_TAA);
1182 
1183 	/*
1184 	 * SRBDS affects CPUs which support RDRAND or RDSEED and are listed
1185 	 * in the vulnerability blacklist.
1186 	 */
1187 	if ((cpu_has(c, X86_FEATURE_RDRAND) ||
1188 	     cpu_has(c, X86_FEATURE_RDSEED)) &&
1189 	    cpu_matches(cpu_vuln_blacklist, SRBDS))
1190 		    setup_force_cpu_bug(X86_BUG_SRBDS);
1191 
1192 	if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN))
1193 		return;
1194 
1195 	/* Rogue Data Cache Load? No! */
1196 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1197 		return;
1198 
1199 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1200 
1201 	if (cpu_matches(cpu_vuln_whitelist, NO_L1TF))
1202 		return;
1203 
1204 	setup_force_cpu_bug(X86_BUG_L1TF);
1205 }
1206 
1207 /*
1208  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1209  * unfortunately, that's not true in practice because of early VIA
1210  * chips and (more importantly) broken virtualizers that are not easy
1211  * to detect. In the latter case it doesn't even *fail* reliably, so
1212  * probing for it doesn't even work. Disable it completely on 32-bit
1213  * unless we can find a reliable way to detect all the broken cases.
1214  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1215  */
1216 static void detect_nopl(void)
1217 {
1218 #ifdef CONFIG_X86_32
1219 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1220 #else
1221 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1222 #endif
1223 }
1224 
1225 /*
1226  * We parse cpu parameters early because fpu__init_system() is executed
1227  * before parse_early_param().
1228  */
1229 static void __init cpu_parse_early_param(void)
1230 {
1231 	char arg[128];
1232 	char *argptr = arg;
1233 	int arglen, res, bit;
1234 
1235 #ifdef CONFIG_X86_32
1236 	if (cmdline_find_option_bool(boot_command_line, "no387"))
1237 #ifdef CONFIG_MATH_EMULATION
1238 		setup_clear_cpu_cap(X86_FEATURE_FPU);
1239 #else
1240 		pr_err("Option 'no387' required CONFIG_MATH_EMULATION enabled.\n");
1241 #endif
1242 
1243 	if (cmdline_find_option_bool(boot_command_line, "nofxsr"))
1244 		setup_clear_cpu_cap(X86_FEATURE_FXSR);
1245 #endif
1246 
1247 	if (cmdline_find_option_bool(boot_command_line, "noxsave"))
1248 		setup_clear_cpu_cap(X86_FEATURE_XSAVE);
1249 
1250 	if (cmdline_find_option_bool(boot_command_line, "noxsaveopt"))
1251 		setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
1252 
1253 	if (cmdline_find_option_bool(boot_command_line, "noxsaves"))
1254 		setup_clear_cpu_cap(X86_FEATURE_XSAVES);
1255 
1256 	arglen = cmdline_find_option(boot_command_line, "clearcpuid", arg, sizeof(arg));
1257 	if (arglen <= 0)
1258 		return;
1259 
1260 	pr_info("Clearing CPUID bits:");
1261 	do {
1262 		res = get_option(&argptr, &bit);
1263 		if (res == 0 || res == 3)
1264 			break;
1265 
1266 		/* If the argument was too long, the last bit may be cut off */
1267 		if (res == 1 && arglen >= sizeof(arg))
1268 			break;
1269 
1270 		if (bit >= 0 && bit < NCAPINTS * 32) {
1271 			pr_cont(" " X86_CAP_FMT, x86_cap_flag(bit));
1272 			setup_clear_cpu_cap(bit);
1273 		}
1274 	} while (res == 2);
1275 	pr_cont("\n");
1276 }
1277 
1278 /*
1279  * Do minimum CPU detection early.
1280  * Fields really needed: vendor, cpuid_level, family, model, mask,
1281  * cache alignment.
1282  * The others are not touched to avoid unwanted side effects.
1283  *
1284  * WARNING: this function is only called on the boot CPU.  Don't add code
1285  * here that is supposed to run on all CPUs.
1286  */
1287 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1288 {
1289 #ifdef CONFIG_X86_64
1290 	c->x86_clflush_size = 64;
1291 	c->x86_phys_bits = 36;
1292 	c->x86_virt_bits = 48;
1293 #else
1294 	c->x86_clflush_size = 32;
1295 	c->x86_phys_bits = 32;
1296 	c->x86_virt_bits = 32;
1297 #endif
1298 	c->x86_cache_alignment = c->x86_clflush_size;
1299 
1300 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1301 	c->extended_cpuid_level = 0;
1302 
1303 	if (!have_cpuid_p())
1304 		identify_cpu_without_cpuid(c);
1305 
1306 	/* cyrix could have cpuid enabled via c_identify()*/
1307 	if (have_cpuid_p()) {
1308 		cpu_detect(c);
1309 		get_cpu_vendor(c);
1310 		get_cpu_cap(c);
1311 		get_cpu_address_sizes(c);
1312 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1313 		cpu_parse_early_param();
1314 
1315 		if (this_cpu->c_early_init)
1316 			this_cpu->c_early_init(c);
1317 
1318 		c->cpu_index = 0;
1319 		filter_cpuid_features(c, false);
1320 
1321 		if (this_cpu->c_bsp_init)
1322 			this_cpu->c_bsp_init(c);
1323 	} else {
1324 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1325 	}
1326 
1327 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1328 
1329 	cpu_set_bug_bits(c);
1330 
1331 	sld_setup(c);
1332 
1333 	fpu__init_system(c);
1334 
1335 #ifdef CONFIG_X86_32
1336 	/*
1337 	 * Regardless of whether PCID is enumerated, the SDM says
1338 	 * that it can't be enabled in 32-bit mode.
1339 	 */
1340 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1341 #endif
1342 
1343 	/*
1344 	 * Later in the boot process pgtable_l5_enabled() relies on
1345 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1346 	 * enabled by this point we need to clear the feature bit to avoid
1347 	 * false-positives at the later stage.
1348 	 *
1349 	 * pgtable_l5_enabled() can be false here for several reasons:
1350 	 *  - 5-level paging is disabled compile-time;
1351 	 *  - it's 32-bit kernel;
1352 	 *  - machine doesn't support 5-level paging;
1353 	 *  - user specified 'no5lvl' in kernel command line.
1354 	 */
1355 	if (!pgtable_l5_enabled())
1356 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1357 
1358 	detect_nopl();
1359 }
1360 
1361 void __init early_cpu_init(void)
1362 {
1363 	const struct cpu_dev *const *cdev;
1364 	int count = 0;
1365 
1366 #ifdef CONFIG_PROCESSOR_SELECT
1367 	pr_info("KERNEL supported cpus:\n");
1368 #endif
1369 
1370 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1371 		const struct cpu_dev *cpudev = *cdev;
1372 
1373 		if (count >= X86_VENDOR_NUM)
1374 			break;
1375 		cpu_devs[count] = cpudev;
1376 		count++;
1377 
1378 #ifdef CONFIG_PROCESSOR_SELECT
1379 		{
1380 			unsigned int j;
1381 
1382 			for (j = 0; j < 2; j++) {
1383 				if (!cpudev->c_ident[j])
1384 					continue;
1385 				pr_info("  %s %s\n", cpudev->c_vendor,
1386 					cpudev->c_ident[j]);
1387 			}
1388 		}
1389 #endif
1390 	}
1391 	early_identify_cpu(&boot_cpu_data);
1392 }
1393 
1394 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1395 {
1396 #ifdef CONFIG_X86_64
1397 	/*
1398 	 * Empirically, writing zero to a segment selector on AMD does
1399 	 * not clear the base, whereas writing zero to a segment
1400 	 * selector on Intel does clear the base.  Intel's behavior
1401 	 * allows slightly faster context switches in the common case
1402 	 * where GS is unused by the prev and next threads.
1403 	 *
1404 	 * Since neither vendor documents this anywhere that I can see,
1405 	 * detect it directly instead of hard-coding the choice by
1406 	 * vendor.
1407 	 *
1408 	 * I've designated AMD's behavior as the "bug" because it's
1409 	 * counterintuitive and less friendly.
1410 	 */
1411 
1412 	unsigned long old_base, tmp;
1413 	rdmsrl(MSR_FS_BASE, old_base);
1414 	wrmsrl(MSR_FS_BASE, 1);
1415 	loadsegment(fs, 0);
1416 	rdmsrl(MSR_FS_BASE, tmp);
1417 	if (tmp != 0)
1418 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1419 	wrmsrl(MSR_FS_BASE, old_base);
1420 #endif
1421 }
1422 
1423 static void generic_identify(struct cpuinfo_x86 *c)
1424 {
1425 	c->extended_cpuid_level = 0;
1426 
1427 	if (!have_cpuid_p())
1428 		identify_cpu_without_cpuid(c);
1429 
1430 	/* cyrix could have cpuid enabled via c_identify()*/
1431 	if (!have_cpuid_p())
1432 		return;
1433 
1434 	cpu_detect(c);
1435 
1436 	get_cpu_vendor(c);
1437 
1438 	get_cpu_cap(c);
1439 
1440 	get_cpu_address_sizes(c);
1441 
1442 	if (c->cpuid_level >= 0x00000001) {
1443 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1444 #ifdef CONFIG_X86_32
1445 # ifdef CONFIG_SMP
1446 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1447 # else
1448 		c->apicid = c->initial_apicid;
1449 # endif
1450 #endif
1451 		c->phys_proc_id = c->initial_apicid;
1452 	}
1453 
1454 	get_model_name(c); /* Default name */
1455 
1456 	detect_null_seg_behavior(c);
1457 
1458 	/*
1459 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1460 	 * systems that run Linux at CPL > 0 may or may not have the
1461 	 * issue, but, even if they have the issue, there's absolutely
1462 	 * nothing we can do about it because we can't use the real IRET
1463 	 * instruction.
1464 	 *
1465 	 * NB: For the time being, only 32-bit kernels support
1466 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1467 	 * whether to apply espfix using paravirt hooks.  If any
1468 	 * non-paravirt system ever shows up that does *not* have the
1469 	 * ESPFIX issue, we can change this.
1470 	 */
1471 #ifdef CONFIG_X86_32
1472 	set_cpu_bug(c, X86_BUG_ESPFIX);
1473 #endif
1474 }
1475 
1476 /*
1477  * Validate that ACPI/mptables have the same information about the
1478  * effective APIC id and update the package map.
1479  */
1480 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1481 {
1482 #ifdef CONFIG_SMP
1483 	unsigned int apicid, cpu = smp_processor_id();
1484 
1485 	apicid = apic->cpu_present_to_apicid(cpu);
1486 
1487 	if (apicid != c->apicid) {
1488 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1489 		       cpu, apicid, c->initial_apicid);
1490 	}
1491 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1492 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1493 #else
1494 	c->logical_proc_id = 0;
1495 #endif
1496 }
1497 
1498 /*
1499  * This does the hard work of actually picking apart the CPU stuff...
1500  */
1501 static void identify_cpu(struct cpuinfo_x86 *c)
1502 {
1503 	int i;
1504 
1505 	c->loops_per_jiffy = loops_per_jiffy;
1506 	c->x86_cache_size = 0;
1507 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1508 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1509 	c->x86_vendor_id[0] = '\0'; /* Unset */
1510 	c->x86_model_id[0] = '\0';  /* Unset */
1511 	c->x86_max_cores = 1;
1512 	c->x86_coreid_bits = 0;
1513 	c->cu_id = 0xff;
1514 #ifdef CONFIG_X86_64
1515 	c->x86_clflush_size = 64;
1516 	c->x86_phys_bits = 36;
1517 	c->x86_virt_bits = 48;
1518 #else
1519 	c->cpuid_level = -1;	/* CPUID not detected */
1520 	c->x86_clflush_size = 32;
1521 	c->x86_phys_bits = 32;
1522 	c->x86_virt_bits = 32;
1523 #endif
1524 	c->x86_cache_alignment = c->x86_clflush_size;
1525 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1526 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
1527 	memset(&c->vmx_capability, 0, sizeof(c->vmx_capability));
1528 #endif
1529 
1530 	generic_identify(c);
1531 
1532 	if (this_cpu->c_identify)
1533 		this_cpu->c_identify(c);
1534 
1535 	/* Clear/Set all flags overridden by options, after probe */
1536 	apply_forced_caps(c);
1537 
1538 #ifdef CONFIG_X86_64
1539 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1540 #endif
1541 
1542 	/*
1543 	 * Vendor-specific initialization.  In this section we
1544 	 * canonicalize the feature flags, meaning if there are
1545 	 * features a certain CPU supports which CPUID doesn't
1546 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1547 	 * we handle them here.
1548 	 *
1549 	 * At the end of this section, c->x86_capability better
1550 	 * indicate the features this CPU genuinely supports!
1551 	 */
1552 	if (this_cpu->c_init)
1553 		this_cpu->c_init(c);
1554 
1555 	/* Disable the PN if appropriate */
1556 	squash_the_stupid_serial_number(c);
1557 
1558 	/* Set up SMEP/SMAP/UMIP */
1559 	setup_smep(c);
1560 	setup_smap(c);
1561 	setup_umip(c);
1562 
1563 	/* Enable FSGSBASE instructions if available. */
1564 	if (cpu_has(c, X86_FEATURE_FSGSBASE)) {
1565 		cr4_set_bits(X86_CR4_FSGSBASE);
1566 		elf_hwcap2 |= HWCAP2_FSGSBASE;
1567 	}
1568 
1569 	/*
1570 	 * The vendor-specific functions might have changed features.
1571 	 * Now we do "generic changes."
1572 	 */
1573 
1574 	/* Filter out anything that depends on CPUID levels we don't have */
1575 	filter_cpuid_features(c, true);
1576 
1577 	/* If the model name is still unset, do table lookup. */
1578 	if (!c->x86_model_id[0]) {
1579 		const char *p;
1580 		p = table_lookup_model(c);
1581 		if (p)
1582 			strcpy(c->x86_model_id, p);
1583 		else
1584 			/* Last resort... */
1585 			sprintf(c->x86_model_id, "%02x/%02x",
1586 				c->x86, c->x86_model);
1587 	}
1588 
1589 #ifdef CONFIG_X86_64
1590 	detect_ht(c);
1591 #endif
1592 
1593 	x86_init_rdrand(c);
1594 	setup_pku(c);
1595 
1596 	/*
1597 	 * Clear/Set all flags overridden by options, need do it
1598 	 * before following smp all cpus cap AND.
1599 	 */
1600 	apply_forced_caps(c);
1601 
1602 	/*
1603 	 * On SMP, boot_cpu_data holds the common feature set between
1604 	 * all CPUs; so make sure that we indicate which features are
1605 	 * common between the CPUs.  The first time this routine gets
1606 	 * executed, c == &boot_cpu_data.
1607 	 */
1608 	if (c != &boot_cpu_data) {
1609 		/* AND the already accumulated flags with these */
1610 		for (i = 0; i < NCAPINTS; i++)
1611 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1612 
1613 		/* OR, i.e. replicate the bug flags */
1614 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1615 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1616 	}
1617 
1618 	/* Init Machine Check Exception if available. */
1619 	mcheck_cpu_init(c);
1620 
1621 	select_idle_routine(c);
1622 
1623 #ifdef CONFIG_NUMA
1624 	numa_add_cpu(smp_processor_id());
1625 #endif
1626 }
1627 
1628 /*
1629  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1630  * on 32-bit kernels:
1631  */
1632 #ifdef CONFIG_X86_32
1633 void enable_sep_cpu(void)
1634 {
1635 	struct tss_struct *tss;
1636 	int cpu;
1637 
1638 	if (!boot_cpu_has(X86_FEATURE_SEP))
1639 		return;
1640 
1641 	cpu = get_cpu();
1642 	tss = &per_cpu(cpu_tss_rw, cpu);
1643 
1644 	/*
1645 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1646 	 * see the big comment in struct x86_hw_tss's definition.
1647 	 */
1648 
1649 	tss->x86_tss.ss1 = __KERNEL_CS;
1650 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1651 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1652 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1653 
1654 	put_cpu();
1655 }
1656 #endif
1657 
1658 void __init identify_boot_cpu(void)
1659 {
1660 	identify_cpu(&boot_cpu_data);
1661 #ifdef CONFIG_X86_32
1662 	sysenter_setup();
1663 	enable_sep_cpu();
1664 #endif
1665 	cpu_detect_tlb(&boot_cpu_data);
1666 	setup_cr_pinning();
1667 
1668 	tsx_init();
1669 }
1670 
1671 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1672 {
1673 	BUG_ON(c == &boot_cpu_data);
1674 	identify_cpu(c);
1675 #ifdef CONFIG_X86_32
1676 	enable_sep_cpu();
1677 #endif
1678 	mtrr_ap_init();
1679 	validate_apic_and_package_id(c);
1680 	x86_spec_ctrl_setup_ap();
1681 	update_srbds_msr();
1682 }
1683 
1684 static __init int setup_noclflush(char *arg)
1685 {
1686 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1687 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1688 	return 1;
1689 }
1690 __setup("noclflush", setup_noclflush);
1691 
1692 void print_cpu_info(struct cpuinfo_x86 *c)
1693 {
1694 	const char *vendor = NULL;
1695 
1696 	if (c->x86_vendor < X86_VENDOR_NUM) {
1697 		vendor = this_cpu->c_vendor;
1698 	} else {
1699 		if (c->cpuid_level >= 0)
1700 			vendor = c->x86_vendor_id;
1701 	}
1702 
1703 	if (vendor && !strstr(c->x86_model_id, vendor))
1704 		pr_cont("%s ", vendor);
1705 
1706 	if (c->x86_model_id[0])
1707 		pr_cont("%s", c->x86_model_id);
1708 	else
1709 		pr_cont("%d86", c->x86);
1710 
1711 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1712 
1713 	if (c->x86_stepping || c->cpuid_level >= 0)
1714 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1715 	else
1716 		pr_cont(")\n");
1717 }
1718 
1719 /*
1720  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1721  * But we need to keep a dummy __setup around otherwise it would
1722  * show up as an environment variable for init.
1723  */
1724 static __init int setup_clearcpuid(char *arg)
1725 {
1726 	return 1;
1727 }
1728 __setup("clearcpuid=", setup_clearcpuid);
1729 
1730 #ifdef CONFIG_X86_64
1731 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1732 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1733 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1734 
1735 /*
1736  * The following percpu variables are hot.  Align current_task to
1737  * cacheline size such that they fall in the same cacheline.
1738  */
1739 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1740 	&init_task;
1741 EXPORT_PER_CPU_SYMBOL(current_task);
1742 
1743 DEFINE_PER_CPU(void *, hardirq_stack_ptr);
1744 DEFINE_PER_CPU(bool, hardirq_stack_inuse);
1745 
1746 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1747 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1748 
1749 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) = TOP_OF_INIT_STACK;
1750 
1751 /* May not be marked __init: used by software suspend */
1752 void syscall_init(void)
1753 {
1754 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1755 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1756 
1757 #ifdef CONFIG_IA32_EMULATION
1758 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1759 	/*
1760 	 * This only works on Intel CPUs.
1761 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1762 	 * This does not cause SYSENTER to jump to the wrong location, because
1763 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1764 	 */
1765 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1766 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1767 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1768 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1769 #else
1770 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1771 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1772 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1773 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1774 #endif
1775 
1776 	/* Flags to clear on syscall */
1777 	wrmsrl(MSR_SYSCALL_MASK,
1778 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1779 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1780 }
1781 
1782 #else	/* CONFIG_X86_64 */
1783 
1784 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1785 EXPORT_PER_CPU_SYMBOL(current_task);
1786 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1787 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1788 
1789 /*
1790  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1791  * the top of the kernel stack.  Use an extra percpu variable to track the
1792  * top of the kernel stack directly.
1793  */
1794 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1795 	(unsigned long)&init_thread_union + THREAD_SIZE;
1796 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1797 
1798 #ifdef CONFIG_STACKPROTECTOR
1799 DEFINE_PER_CPU(unsigned long, __stack_chk_guard);
1800 EXPORT_PER_CPU_SYMBOL(__stack_chk_guard);
1801 #endif
1802 
1803 #endif	/* CONFIG_X86_64 */
1804 
1805 /*
1806  * Clear all 6 debug registers:
1807  */
1808 static void clear_all_debug_regs(void)
1809 {
1810 	int i;
1811 
1812 	for (i = 0; i < 8; i++) {
1813 		/* Ignore db4, db5 */
1814 		if ((i == 4) || (i == 5))
1815 			continue;
1816 
1817 		set_debugreg(0, i);
1818 	}
1819 }
1820 
1821 #ifdef CONFIG_KGDB
1822 /*
1823  * Restore debug regs if using kgdbwait and you have a kernel debugger
1824  * connection established.
1825  */
1826 static void dbg_restore_debug_regs(void)
1827 {
1828 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1829 		arch_kgdb_ops.correct_hw_break();
1830 }
1831 #else /* ! CONFIG_KGDB */
1832 #define dbg_restore_debug_regs()
1833 #endif /* ! CONFIG_KGDB */
1834 
1835 static void wait_for_master_cpu(int cpu)
1836 {
1837 #ifdef CONFIG_SMP
1838 	/*
1839 	 * wait for ACK from master CPU before continuing
1840 	 * with AP initialization
1841 	 */
1842 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1843 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1844 		cpu_relax();
1845 #endif
1846 }
1847 
1848 #ifdef CONFIG_X86_64
1849 static inline void setup_getcpu(int cpu)
1850 {
1851 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1852 	struct desc_struct d = { };
1853 
1854 	if (boot_cpu_has(X86_FEATURE_RDTSCP) || boot_cpu_has(X86_FEATURE_RDPID))
1855 		wrmsr(MSR_TSC_AUX, cpudata, 0);
1856 
1857 	/* Store CPU and node number in limit. */
1858 	d.limit0 = cpudata;
1859 	d.limit1 = cpudata >> 16;
1860 
1861 	d.type = 5;		/* RO data, expand down, accessed */
1862 	d.dpl = 3;		/* Visible to user code */
1863 	d.s = 1;		/* Not a system segment */
1864 	d.p = 1;		/* Present */
1865 	d.d = 1;		/* 32-bit */
1866 
1867 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1868 }
1869 
1870 static inline void ucode_cpu_init(int cpu)
1871 {
1872 	if (cpu)
1873 		load_ucode_ap();
1874 }
1875 
1876 static inline void tss_setup_ist(struct tss_struct *tss)
1877 {
1878 	/* Set up the per-CPU TSS IST stacks */
1879 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1880 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1881 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1882 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1883 	/* Only mapped when SEV-ES is active */
1884 	tss->x86_tss.ist[IST_INDEX_VC] = __this_cpu_ist_top_va(VC);
1885 }
1886 
1887 #else /* CONFIG_X86_64 */
1888 
1889 static inline void setup_getcpu(int cpu) { }
1890 
1891 static inline void ucode_cpu_init(int cpu)
1892 {
1893 	show_ucode_info_early();
1894 }
1895 
1896 static inline void tss_setup_ist(struct tss_struct *tss) { }
1897 
1898 #endif /* !CONFIG_X86_64 */
1899 
1900 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
1901 {
1902 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
1903 
1904 #ifdef CONFIG_X86_IOPL_IOPERM
1905 	tss->io_bitmap.prev_max = 0;
1906 	tss->io_bitmap.prev_sequence = 0;
1907 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
1908 	/*
1909 	 * Invalidate the extra array entry past the end of the all
1910 	 * permission bitmap as required by the hardware.
1911 	 */
1912 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
1913 #endif
1914 }
1915 
1916 /*
1917  * Setup everything needed to handle exceptions from the IDT, including the IST
1918  * exceptions which use paranoid_entry().
1919  */
1920 void cpu_init_exception_handling(void)
1921 {
1922 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
1923 	int cpu = raw_smp_processor_id();
1924 
1925 	/* paranoid_entry() gets the CPU number from the GDT */
1926 	setup_getcpu(cpu);
1927 
1928 	/* IST vectors need TSS to be set up. */
1929 	tss_setup_ist(tss);
1930 	tss_setup_io_bitmap(tss);
1931 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1932 
1933 	load_TR_desc();
1934 
1935 	/* Finally load the IDT */
1936 	load_current_idt();
1937 }
1938 
1939 /*
1940  * cpu_init() initializes state that is per-CPU. Some data is already
1941  * initialized (naturally) in the bootstrap process, such as the GDT
1942  * and IDT. We reload them nevertheless, this function acts as a
1943  * 'CPU state barrier', nothing should get across.
1944  */
1945 void cpu_init(void)
1946 {
1947 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
1948 	struct task_struct *cur = current;
1949 	int cpu = raw_smp_processor_id();
1950 
1951 	wait_for_master_cpu(cpu);
1952 
1953 	ucode_cpu_init(cpu);
1954 
1955 #ifdef CONFIG_NUMA
1956 	if (this_cpu_read(numa_node) == 0 &&
1957 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1958 		set_numa_node(early_cpu_to_node(cpu));
1959 #endif
1960 	setup_getcpu(cpu);
1961 
1962 	pr_debug("Initializing CPU#%d\n", cpu);
1963 
1964 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
1965 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
1966 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1967 
1968 	/*
1969 	 * Initialize the per-CPU GDT with the boot GDT,
1970 	 * and set up the GDT descriptor:
1971 	 */
1972 	switch_to_new_gdt(cpu);
1973 	load_current_idt();
1974 
1975 	if (IS_ENABLED(CONFIG_X86_64)) {
1976 		loadsegment(fs, 0);
1977 		memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1978 		syscall_init();
1979 
1980 		wrmsrl(MSR_FS_BASE, 0);
1981 		wrmsrl(MSR_KERNEL_GS_BASE, 0);
1982 		barrier();
1983 
1984 		x2apic_setup();
1985 	}
1986 
1987 	mmgrab(&init_mm);
1988 	cur->active_mm = &init_mm;
1989 	BUG_ON(cur->mm);
1990 	initialize_tlbstate_and_flush();
1991 	enter_lazy_tlb(&init_mm, cur);
1992 
1993 	/* Initialize the TSS. */
1994 	tss_setup_ist(tss);
1995 	tss_setup_io_bitmap(tss);
1996 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1997 
1998 	load_TR_desc();
1999 	/*
2000 	 * sp0 points to the entry trampoline stack regardless of what task
2001 	 * is running.
2002 	 */
2003 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
2004 
2005 	load_mm_ldt(&init_mm);
2006 
2007 	clear_all_debug_regs();
2008 	dbg_restore_debug_regs();
2009 
2010 	doublefault_init_cpu_tss();
2011 
2012 	fpu__init_cpu();
2013 
2014 	if (is_uv_system())
2015 		uv_cpu_init();
2016 
2017 	load_fixmap_gdt(cpu);
2018 }
2019 
2020 /*
2021  * The microcode loader calls this upon late microcode load to recheck features,
2022  * only when microcode has been updated. Caller holds microcode_mutex and CPU
2023  * hotplug lock.
2024  */
2025 void microcode_check(void)
2026 {
2027 	struct cpuinfo_x86 info;
2028 
2029 	perf_check_microcode();
2030 
2031 	/* Reload CPUID max function as it might've changed. */
2032 	info.cpuid_level = cpuid_eax(0);
2033 
2034 	/*
2035 	 * Copy all capability leafs to pick up the synthetic ones so that
2036 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
2037 	 * get overwritten in get_cpu_cap().
2038 	 */
2039 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
2040 
2041 	get_cpu_cap(&info);
2042 
2043 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
2044 		return;
2045 
2046 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
2047 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
2048 }
2049 
2050 /*
2051  * Invoked from core CPU hotplug code after hotplug operations
2052  */
2053 void arch_smt_update(void)
2054 {
2055 	/* Handle the speculative execution misfeatures */
2056 	cpu_bugs_smt_update();
2057 	/* Check whether IPI broadcasting can be enabled */
2058 	apic_smt_update();
2059 }
2060