xref: /openbmc/linux/arch/x86/kernel/cpu/common.c (revision 55fd7e02)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /* cpu_feature_enabled() cannot be used this early */
3 #define USE_EARLY_PGTABLE_L5
4 
5 #include <linux/memblock.h>
6 #include <linux/linkage.h>
7 #include <linux/bitops.h>
8 #include <linux/kernel.h>
9 #include <linux/export.h>
10 #include <linux/percpu.h>
11 #include <linux/string.h>
12 #include <linux/ctype.h>
13 #include <linux/delay.h>
14 #include <linux/sched/mm.h>
15 #include <linux/sched/clock.h>
16 #include <linux/sched/task.h>
17 #include <linux/sched/smt.h>
18 #include <linux/init.h>
19 #include <linux/kprobes.h>
20 #include <linux/kgdb.h>
21 #include <linux/smp.h>
22 #include <linux/io.h>
23 #include <linux/syscore_ops.h>
24 #include <linux/pgtable.h>
25 
26 #include <asm/stackprotector.h>
27 #include <asm/perf_event.h>
28 #include <asm/mmu_context.h>
29 #include <asm/doublefault.h>
30 #include <asm/archrandom.h>
31 #include <asm/hypervisor.h>
32 #include <asm/processor.h>
33 #include <asm/tlbflush.h>
34 #include <asm/debugreg.h>
35 #include <asm/sections.h>
36 #include <asm/vsyscall.h>
37 #include <linux/topology.h>
38 #include <linux/cpumask.h>
39 #include <linux/atomic.h>
40 #include <asm/proto.h>
41 #include <asm/setup.h>
42 #include <asm/apic.h>
43 #include <asm/desc.h>
44 #include <asm/fpu/internal.h>
45 #include <asm/mtrr.h>
46 #include <asm/hwcap2.h>
47 #include <linux/numa.h>
48 #include <asm/asm.h>
49 #include <asm/bugs.h>
50 #include <asm/cpu.h>
51 #include <asm/mce.h>
52 #include <asm/msr.h>
53 #include <asm/memtype.h>
54 #include <asm/microcode.h>
55 #include <asm/microcode_intel.h>
56 #include <asm/intel-family.h>
57 #include <asm/cpu_device_id.h>
58 #include <asm/uv/uv.h>
59 
60 #include "cpu.h"
61 
62 u32 elf_hwcap2 __read_mostly;
63 
64 /* all of these masks are initialized in setup_cpu_local_masks() */
65 cpumask_var_t cpu_initialized_mask;
66 cpumask_var_t cpu_callout_mask;
67 cpumask_var_t cpu_callin_mask;
68 
69 /* representing cpus for which sibling maps can be computed */
70 cpumask_var_t cpu_sibling_setup_mask;
71 
72 /* Number of siblings per CPU package */
73 int smp_num_siblings = 1;
74 EXPORT_SYMBOL(smp_num_siblings);
75 
76 /* Last level cache ID of each logical CPU */
77 DEFINE_PER_CPU_READ_MOSTLY(u16, cpu_llc_id) = BAD_APICID;
78 
79 /* correctly size the local cpu masks */
80 void __init setup_cpu_local_masks(void)
81 {
82 	alloc_bootmem_cpumask_var(&cpu_initialized_mask);
83 	alloc_bootmem_cpumask_var(&cpu_callin_mask);
84 	alloc_bootmem_cpumask_var(&cpu_callout_mask);
85 	alloc_bootmem_cpumask_var(&cpu_sibling_setup_mask);
86 }
87 
88 static void default_init(struct cpuinfo_x86 *c)
89 {
90 #ifdef CONFIG_X86_64
91 	cpu_detect_cache_sizes(c);
92 #else
93 	/* Not much we can do here... */
94 	/* Check if at least it has cpuid */
95 	if (c->cpuid_level == -1) {
96 		/* No cpuid. It must be an ancient CPU */
97 		if (c->x86 == 4)
98 			strcpy(c->x86_model_id, "486");
99 		else if (c->x86 == 3)
100 			strcpy(c->x86_model_id, "386");
101 	}
102 #endif
103 }
104 
105 static const struct cpu_dev default_cpu = {
106 	.c_init		= default_init,
107 	.c_vendor	= "Unknown",
108 	.c_x86_vendor	= X86_VENDOR_UNKNOWN,
109 };
110 
111 static const struct cpu_dev *this_cpu = &default_cpu;
112 
113 DEFINE_PER_CPU_PAGE_ALIGNED(struct gdt_page, gdt_page) = { .gdt = {
114 #ifdef CONFIG_X86_64
115 	/*
116 	 * We need valid kernel segments for data and code in long mode too
117 	 * IRET will check the segment types  kkeil 2000/10/28
118 	 * Also sysret mandates a special GDT layout
119 	 *
120 	 * TLS descriptors are currently at a different place compared to i386.
121 	 * Hopefully nobody expects them at a fixed place (Wine?)
122 	 */
123 	[GDT_ENTRY_KERNEL32_CS]		= GDT_ENTRY_INIT(0xc09b, 0, 0xfffff),
124 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xa09b, 0, 0xfffff),
125 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc093, 0, 0xfffff),
126 	[GDT_ENTRY_DEFAULT_USER32_CS]	= GDT_ENTRY_INIT(0xc0fb, 0, 0xfffff),
127 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f3, 0, 0xfffff),
128 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xa0fb, 0, 0xfffff),
129 #else
130 	[GDT_ENTRY_KERNEL_CS]		= GDT_ENTRY_INIT(0xc09a, 0, 0xfffff),
131 	[GDT_ENTRY_KERNEL_DS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
132 	[GDT_ENTRY_DEFAULT_USER_CS]	= GDT_ENTRY_INIT(0xc0fa, 0, 0xfffff),
133 	[GDT_ENTRY_DEFAULT_USER_DS]	= GDT_ENTRY_INIT(0xc0f2, 0, 0xfffff),
134 	/*
135 	 * Segments used for calling PnP BIOS have byte granularity.
136 	 * They code segments and data segments have fixed 64k limits,
137 	 * the transfer segment sizes are set at run time.
138 	 */
139 	/* 32-bit code */
140 	[GDT_ENTRY_PNPBIOS_CS32]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
141 	/* 16-bit code */
142 	[GDT_ENTRY_PNPBIOS_CS16]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
143 	/* 16-bit data */
144 	[GDT_ENTRY_PNPBIOS_DS]		= GDT_ENTRY_INIT(0x0092, 0, 0xffff),
145 	/* 16-bit data */
146 	[GDT_ENTRY_PNPBIOS_TS1]		= GDT_ENTRY_INIT(0x0092, 0, 0),
147 	/* 16-bit data */
148 	[GDT_ENTRY_PNPBIOS_TS2]		= GDT_ENTRY_INIT(0x0092, 0, 0),
149 	/*
150 	 * The APM segments have byte granularity and their bases
151 	 * are set at run time.  All have 64k limits.
152 	 */
153 	/* 32-bit code */
154 	[GDT_ENTRY_APMBIOS_BASE]	= GDT_ENTRY_INIT(0x409a, 0, 0xffff),
155 	/* 16-bit code */
156 	[GDT_ENTRY_APMBIOS_BASE+1]	= GDT_ENTRY_INIT(0x009a, 0, 0xffff),
157 	/* data */
158 	[GDT_ENTRY_APMBIOS_BASE+2]	= GDT_ENTRY_INIT(0x4092, 0, 0xffff),
159 
160 	[GDT_ENTRY_ESPFIX_SS]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
161 	[GDT_ENTRY_PERCPU]		= GDT_ENTRY_INIT(0xc092, 0, 0xfffff),
162 	GDT_STACK_CANARY_INIT
163 #endif
164 } };
165 EXPORT_PER_CPU_SYMBOL_GPL(gdt_page);
166 
167 #ifdef CONFIG_X86_64
168 static int __init x86_nopcid_setup(char *s)
169 {
170 	/* nopcid doesn't accept parameters */
171 	if (s)
172 		return -EINVAL;
173 
174 	/* do not emit a message if the feature is not present */
175 	if (!boot_cpu_has(X86_FEATURE_PCID))
176 		return 0;
177 
178 	setup_clear_cpu_cap(X86_FEATURE_PCID);
179 	pr_info("nopcid: PCID feature disabled\n");
180 	return 0;
181 }
182 early_param("nopcid", x86_nopcid_setup);
183 #endif
184 
185 static int __init x86_noinvpcid_setup(char *s)
186 {
187 	/* noinvpcid doesn't accept parameters */
188 	if (s)
189 		return -EINVAL;
190 
191 	/* do not emit a message if the feature is not present */
192 	if (!boot_cpu_has(X86_FEATURE_INVPCID))
193 		return 0;
194 
195 	setup_clear_cpu_cap(X86_FEATURE_INVPCID);
196 	pr_info("noinvpcid: INVPCID feature disabled\n");
197 	return 0;
198 }
199 early_param("noinvpcid", x86_noinvpcid_setup);
200 
201 #ifdef CONFIG_X86_32
202 static int cachesize_override = -1;
203 static int disable_x86_serial_nr = 1;
204 
205 static int __init cachesize_setup(char *str)
206 {
207 	get_option(&str, &cachesize_override);
208 	return 1;
209 }
210 __setup("cachesize=", cachesize_setup);
211 
212 static int __init x86_sep_setup(char *s)
213 {
214 	setup_clear_cpu_cap(X86_FEATURE_SEP);
215 	return 1;
216 }
217 __setup("nosep", x86_sep_setup);
218 
219 /* Standard macro to see if a specific flag is changeable */
220 static inline int flag_is_changeable_p(u32 flag)
221 {
222 	u32 f1, f2;
223 
224 	/*
225 	 * Cyrix and IDT cpus allow disabling of CPUID
226 	 * so the code below may return different results
227 	 * when it is executed before and after enabling
228 	 * the CPUID. Add "volatile" to not allow gcc to
229 	 * optimize the subsequent calls to this function.
230 	 */
231 	asm volatile ("pushfl		\n\t"
232 		      "pushfl		\n\t"
233 		      "popl %0		\n\t"
234 		      "movl %0, %1	\n\t"
235 		      "xorl %2, %0	\n\t"
236 		      "pushl %0		\n\t"
237 		      "popfl		\n\t"
238 		      "pushfl		\n\t"
239 		      "popl %0		\n\t"
240 		      "popfl		\n\t"
241 
242 		      : "=&r" (f1), "=&r" (f2)
243 		      : "ir" (flag));
244 
245 	return ((f1^f2) & flag) != 0;
246 }
247 
248 /* Probe for the CPUID instruction */
249 int have_cpuid_p(void)
250 {
251 	return flag_is_changeable_p(X86_EFLAGS_ID);
252 }
253 
254 static void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
255 {
256 	unsigned long lo, hi;
257 
258 	if (!cpu_has(c, X86_FEATURE_PN) || !disable_x86_serial_nr)
259 		return;
260 
261 	/* Disable processor serial number: */
262 
263 	rdmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
264 	lo |= 0x200000;
265 	wrmsr(MSR_IA32_BBL_CR_CTL, lo, hi);
266 
267 	pr_notice("CPU serial number disabled.\n");
268 	clear_cpu_cap(c, X86_FEATURE_PN);
269 
270 	/* Disabling the serial number may affect the cpuid level */
271 	c->cpuid_level = cpuid_eax(0);
272 }
273 
274 static int __init x86_serial_nr_setup(char *s)
275 {
276 	disable_x86_serial_nr = 0;
277 	return 1;
278 }
279 __setup("serialnumber", x86_serial_nr_setup);
280 #else
281 static inline int flag_is_changeable_p(u32 flag)
282 {
283 	return 1;
284 }
285 static inline void squash_the_stupid_serial_number(struct cpuinfo_x86 *c)
286 {
287 }
288 #endif
289 
290 static __init int setup_disable_smep(char *arg)
291 {
292 	setup_clear_cpu_cap(X86_FEATURE_SMEP);
293 	return 1;
294 }
295 __setup("nosmep", setup_disable_smep);
296 
297 static __always_inline void setup_smep(struct cpuinfo_x86 *c)
298 {
299 	if (cpu_has(c, X86_FEATURE_SMEP))
300 		cr4_set_bits(X86_CR4_SMEP);
301 }
302 
303 static __init int setup_disable_smap(char *arg)
304 {
305 	setup_clear_cpu_cap(X86_FEATURE_SMAP);
306 	return 1;
307 }
308 __setup("nosmap", setup_disable_smap);
309 
310 static __always_inline void setup_smap(struct cpuinfo_x86 *c)
311 {
312 	unsigned long eflags = native_save_fl();
313 
314 	/* This should have been cleared long ago */
315 	BUG_ON(eflags & X86_EFLAGS_AC);
316 
317 	if (cpu_has(c, X86_FEATURE_SMAP)) {
318 #ifdef CONFIG_X86_SMAP
319 		cr4_set_bits(X86_CR4_SMAP);
320 #else
321 		cr4_clear_bits(X86_CR4_SMAP);
322 #endif
323 	}
324 }
325 
326 static __always_inline void setup_umip(struct cpuinfo_x86 *c)
327 {
328 	/* Check the boot processor, plus build option for UMIP. */
329 	if (!cpu_feature_enabled(X86_FEATURE_UMIP))
330 		goto out;
331 
332 	/* Check the current processor's cpuid bits. */
333 	if (!cpu_has(c, X86_FEATURE_UMIP))
334 		goto out;
335 
336 	cr4_set_bits(X86_CR4_UMIP);
337 
338 	pr_info_once("x86/cpu: User Mode Instruction Prevention (UMIP) activated\n");
339 
340 	return;
341 
342 out:
343 	/*
344 	 * Make sure UMIP is disabled in case it was enabled in a
345 	 * previous boot (e.g., via kexec).
346 	 */
347 	cr4_clear_bits(X86_CR4_UMIP);
348 }
349 
350 /* These bits should not change their value after CPU init is finished. */
351 static const unsigned long cr4_pinned_mask =
352 	X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_UMIP | X86_CR4_FSGSBASE;
353 static DEFINE_STATIC_KEY_FALSE_RO(cr_pinning);
354 static unsigned long cr4_pinned_bits __ro_after_init;
355 
356 void native_write_cr0(unsigned long val)
357 {
358 	unsigned long bits_missing = 0;
359 
360 set_register:
361 	asm volatile("mov %0,%%cr0": "+r" (val), "+m" (__force_order));
362 
363 	if (static_branch_likely(&cr_pinning)) {
364 		if (unlikely((val & X86_CR0_WP) != X86_CR0_WP)) {
365 			bits_missing = X86_CR0_WP;
366 			val |= bits_missing;
367 			goto set_register;
368 		}
369 		/* Warn after we've set the missing bits. */
370 		WARN_ONCE(bits_missing, "CR0 WP bit went missing!?\n");
371 	}
372 }
373 EXPORT_SYMBOL(native_write_cr0);
374 
375 void native_write_cr4(unsigned long val)
376 {
377 	unsigned long bits_changed = 0;
378 
379 set_register:
380 	asm volatile("mov %0,%%cr4": "+r" (val), "+m" (cr4_pinned_bits));
381 
382 	if (static_branch_likely(&cr_pinning)) {
383 		if (unlikely((val & cr4_pinned_mask) != cr4_pinned_bits)) {
384 			bits_changed = (val & cr4_pinned_mask) ^ cr4_pinned_bits;
385 			val = (val & ~cr4_pinned_mask) | cr4_pinned_bits;
386 			goto set_register;
387 		}
388 		/* Warn after we've corrected the changed bits. */
389 		WARN_ONCE(bits_changed, "pinned CR4 bits changed: 0x%lx!?\n",
390 			  bits_changed);
391 	}
392 }
393 #if IS_MODULE(CONFIG_LKDTM)
394 EXPORT_SYMBOL_GPL(native_write_cr4);
395 #endif
396 
397 void cr4_update_irqsoff(unsigned long set, unsigned long clear)
398 {
399 	unsigned long newval, cr4 = this_cpu_read(cpu_tlbstate.cr4);
400 
401 	lockdep_assert_irqs_disabled();
402 
403 	newval = (cr4 & ~clear) | set;
404 	if (newval != cr4) {
405 		this_cpu_write(cpu_tlbstate.cr4, newval);
406 		__write_cr4(newval);
407 	}
408 }
409 EXPORT_SYMBOL(cr4_update_irqsoff);
410 
411 /* Read the CR4 shadow. */
412 unsigned long cr4_read_shadow(void)
413 {
414 	return this_cpu_read(cpu_tlbstate.cr4);
415 }
416 EXPORT_SYMBOL_GPL(cr4_read_shadow);
417 
418 void cr4_init(void)
419 {
420 	unsigned long cr4 = __read_cr4();
421 
422 	if (boot_cpu_has(X86_FEATURE_PCID))
423 		cr4 |= X86_CR4_PCIDE;
424 	if (static_branch_likely(&cr_pinning))
425 		cr4 = (cr4 & ~cr4_pinned_mask) | cr4_pinned_bits;
426 
427 	__write_cr4(cr4);
428 
429 	/* Initialize cr4 shadow for this CPU. */
430 	this_cpu_write(cpu_tlbstate.cr4, cr4);
431 }
432 
433 /*
434  * Once CPU feature detection is finished (and boot params have been
435  * parsed), record any of the sensitive CR bits that are set, and
436  * enable CR pinning.
437  */
438 static void __init setup_cr_pinning(void)
439 {
440 	cr4_pinned_bits = this_cpu_read(cpu_tlbstate.cr4) & cr4_pinned_mask;
441 	static_key_enable(&cr_pinning.key);
442 }
443 
444 /*
445  * Protection Keys are not available in 32-bit mode.
446  */
447 static bool pku_disabled;
448 
449 static __always_inline void setup_pku(struct cpuinfo_x86 *c)
450 {
451 	struct pkru_state *pk;
452 
453 	/* check the boot processor, plus compile options for PKU: */
454 	if (!cpu_feature_enabled(X86_FEATURE_PKU))
455 		return;
456 	/* checks the actual processor's cpuid bits: */
457 	if (!cpu_has(c, X86_FEATURE_PKU))
458 		return;
459 	if (pku_disabled)
460 		return;
461 
462 	cr4_set_bits(X86_CR4_PKE);
463 	pk = get_xsave_addr(&init_fpstate.xsave, XFEATURE_PKRU);
464 	if (pk)
465 		pk->pkru = init_pkru_value;
466 	/*
467 	 * Seting X86_CR4_PKE will cause the X86_FEATURE_OSPKE
468 	 * cpuid bit to be set.  We need to ensure that we
469 	 * update that bit in this CPU's "cpu_info".
470 	 */
471 	set_cpu_cap(c, X86_FEATURE_OSPKE);
472 }
473 
474 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
475 static __init int setup_disable_pku(char *arg)
476 {
477 	/*
478 	 * Do not clear the X86_FEATURE_PKU bit.  All of the
479 	 * runtime checks are against OSPKE so clearing the
480 	 * bit does nothing.
481 	 *
482 	 * This way, we will see "pku" in cpuinfo, but not
483 	 * "ospke", which is exactly what we want.  It shows
484 	 * that the CPU has PKU, but the OS has not enabled it.
485 	 * This happens to be exactly how a system would look
486 	 * if we disabled the config option.
487 	 */
488 	pr_info("x86: 'nopku' specified, disabling Memory Protection Keys\n");
489 	pku_disabled = true;
490 	return 1;
491 }
492 __setup("nopku", setup_disable_pku);
493 #endif /* CONFIG_X86_64 */
494 
495 /*
496  * Some CPU features depend on higher CPUID levels, which may not always
497  * be available due to CPUID level capping or broken virtualization
498  * software.  Add those features to this table to auto-disable them.
499  */
500 struct cpuid_dependent_feature {
501 	u32 feature;
502 	u32 level;
503 };
504 
505 static const struct cpuid_dependent_feature
506 cpuid_dependent_features[] = {
507 	{ X86_FEATURE_MWAIT,		0x00000005 },
508 	{ X86_FEATURE_DCA,		0x00000009 },
509 	{ X86_FEATURE_XSAVE,		0x0000000d },
510 	{ 0, 0 }
511 };
512 
513 static void filter_cpuid_features(struct cpuinfo_x86 *c, bool warn)
514 {
515 	const struct cpuid_dependent_feature *df;
516 
517 	for (df = cpuid_dependent_features; df->feature; df++) {
518 
519 		if (!cpu_has(c, df->feature))
520 			continue;
521 		/*
522 		 * Note: cpuid_level is set to -1 if unavailable, but
523 		 * extended_extended_level is set to 0 if unavailable
524 		 * and the legitimate extended levels are all negative
525 		 * when signed; hence the weird messing around with
526 		 * signs here...
527 		 */
528 		if (!((s32)df->level < 0 ?
529 		     (u32)df->level > (u32)c->extended_cpuid_level :
530 		     (s32)df->level > (s32)c->cpuid_level))
531 			continue;
532 
533 		clear_cpu_cap(c, df->feature);
534 		if (!warn)
535 			continue;
536 
537 		pr_warn("CPU: CPU feature " X86_CAP_FMT " disabled, no CPUID level 0x%x\n",
538 			x86_cap_flag(df->feature), df->level);
539 	}
540 }
541 
542 /*
543  * Naming convention should be: <Name> [(<Codename>)]
544  * This table only is used unless init_<vendor>() below doesn't set it;
545  * in particular, if CPUID levels 0x80000002..4 are supported, this
546  * isn't used
547  */
548 
549 /* Look up CPU names by table lookup. */
550 static const char *table_lookup_model(struct cpuinfo_x86 *c)
551 {
552 #ifdef CONFIG_X86_32
553 	const struct legacy_cpu_model_info *info;
554 
555 	if (c->x86_model >= 16)
556 		return NULL;	/* Range check */
557 
558 	if (!this_cpu)
559 		return NULL;
560 
561 	info = this_cpu->legacy_models;
562 
563 	while (info->family) {
564 		if (info->family == c->x86)
565 			return info->model_names[c->x86_model];
566 		info++;
567 	}
568 #endif
569 	return NULL;		/* Not found */
570 }
571 
572 /* Aligned to unsigned long to avoid split lock in atomic bitmap ops */
573 __u32 cpu_caps_cleared[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
574 __u32 cpu_caps_set[NCAPINTS + NBUGINTS] __aligned(sizeof(unsigned long));
575 
576 void load_percpu_segment(int cpu)
577 {
578 #ifdef CONFIG_X86_32
579 	loadsegment(fs, __KERNEL_PERCPU);
580 #else
581 	__loadsegment_simple(gs, 0);
582 	wrmsrl(MSR_GS_BASE, cpu_kernelmode_gs_base(cpu));
583 #endif
584 	load_stack_canary_segment();
585 }
586 
587 #ifdef CONFIG_X86_32
588 /* The 32-bit entry code needs to find cpu_entry_area. */
589 DEFINE_PER_CPU(struct cpu_entry_area *, cpu_entry_area);
590 #endif
591 
592 /* Load the original GDT from the per-cpu structure */
593 void load_direct_gdt(int cpu)
594 {
595 	struct desc_ptr gdt_descr;
596 
597 	gdt_descr.address = (long)get_cpu_gdt_rw(cpu);
598 	gdt_descr.size = GDT_SIZE - 1;
599 	load_gdt(&gdt_descr);
600 }
601 EXPORT_SYMBOL_GPL(load_direct_gdt);
602 
603 /* Load a fixmap remapping of the per-cpu GDT */
604 void load_fixmap_gdt(int cpu)
605 {
606 	struct desc_ptr gdt_descr;
607 
608 	gdt_descr.address = (long)get_cpu_gdt_ro(cpu);
609 	gdt_descr.size = GDT_SIZE - 1;
610 	load_gdt(&gdt_descr);
611 }
612 EXPORT_SYMBOL_GPL(load_fixmap_gdt);
613 
614 /*
615  * Current gdt points %fs at the "master" per-cpu area: after this,
616  * it's on the real one.
617  */
618 void switch_to_new_gdt(int cpu)
619 {
620 	/* Load the original GDT */
621 	load_direct_gdt(cpu);
622 	/* Reload the per-cpu base */
623 	load_percpu_segment(cpu);
624 }
625 
626 static const struct cpu_dev *cpu_devs[X86_VENDOR_NUM] = {};
627 
628 static void get_model_name(struct cpuinfo_x86 *c)
629 {
630 	unsigned int *v;
631 	char *p, *q, *s;
632 
633 	if (c->extended_cpuid_level < 0x80000004)
634 		return;
635 
636 	v = (unsigned int *)c->x86_model_id;
637 	cpuid(0x80000002, &v[0], &v[1], &v[2], &v[3]);
638 	cpuid(0x80000003, &v[4], &v[5], &v[6], &v[7]);
639 	cpuid(0x80000004, &v[8], &v[9], &v[10], &v[11]);
640 	c->x86_model_id[48] = 0;
641 
642 	/* Trim whitespace */
643 	p = q = s = &c->x86_model_id[0];
644 
645 	while (*p == ' ')
646 		p++;
647 
648 	while (*p) {
649 		/* Note the last non-whitespace index */
650 		if (!isspace(*p))
651 			s = q;
652 
653 		*q++ = *p++;
654 	}
655 
656 	*(s + 1) = '\0';
657 }
658 
659 void detect_num_cpu_cores(struct cpuinfo_x86 *c)
660 {
661 	unsigned int eax, ebx, ecx, edx;
662 
663 	c->x86_max_cores = 1;
664 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
665 		return;
666 
667 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
668 	if (eax & 0x1f)
669 		c->x86_max_cores = (eax >> 26) + 1;
670 }
671 
672 void cpu_detect_cache_sizes(struct cpuinfo_x86 *c)
673 {
674 	unsigned int n, dummy, ebx, ecx, edx, l2size;
675 
676 	n = c->extended_cpuid_level;
677 
678 	if (n >= 0x80000005) {
679 		cpuid(0x80000005, &dummy, &ebx, &ecx, &edx);
680 		c->x86_cache_size = (ecx>>24) + (edx>>24);
681 #ifdef CONFIG_X86_64
682 		/* On K8 L1 TLB is inclusive, so don't count it */
683 		c->x86_tlbsize = 0;
684 #endif
685 	}
686 
687 	if (n < 0x80000006)	/* Some chips just has a large L1. */
688 		return;
689 
690 	cpuid(0x80000006, &dummy, &ebx, &ecx, &edx);
691 	l2size = ecx >> 16;
692 
693 #ifdef CONFIG_X86_64
694 	c->x86_tlbsize += ((ebx >> 16) & 0xfff) + (ebx & 0xfff);
695 #else
696 	/* do processor-specific cache resizing */
697 	if (this_cpu->legacy_cache_size)
698 		l2size = this_cpu->legacy_cache_size(c, l2size);
699 
700 	/* Allow user to override all this if necessary. */
701 	if (cachesize_override != -1)
702 		l2size = cachesize_override;
703 
704 	if (l2size == 0)
705 		return;		/* Again, no L2 cache is possible */
706 #endif
707 
708 	c->x86_cache_size = l2size;
709 }
710 
711 u16 __read_mostly tlb_lli_4k[NR_INFO];
712 u16 __read_mostly tlb_lli_2m[NR_INFO];
713 u16 __read_mostly tlb_lli_4m[NR_INFO];
714 u16 __read_mostly tlb_lld_4k[NR_INFO];
715 u16 __read_mostly tlb_lld_2m[NR_INFO];
716 u16 __read_mostly tlb_lld_4m[NR_INFO];
717 u16 __read_mostly tlb_lld_1g[NR_INFO];
718 
719 static void cpu_detect_tlb(struct cpuinfo_x86 *c)
720 {
721 	if (this_cpu->c_detect_tlb)
722 		this_cpu->c_detect_tlb(c);
723 
724 	pr_info("Last level iTLB entries: 4KB %d, 2MB %d, 4MB %d\n",
725 		tlb_lli_4k[ENTRIES], tlb_lli_2m[ENTRIES],
726 		tlb_lli_4m[ENTRIES]);
727 
728 	pr_info("Last level dTLB entries: 4KB %d, 2MB %d, 4MB %d, 1GB %d\n",
729 		tlb_lld_4k[ENTRIES], tlb_lld_2m[ENTRIES],
730 		tlb_lld_4m[ENTRIES], tlb_lld_1g[ENTRIES]);
731 }
732 
733 int detect_ht_early(struct cpuinfo_x86 *c)
734 {
735 #ifdef CONFIG_SMP
736 	u32 eax, ebx, ecx, edx;
737 
738 	if (!cpu_has(c, X86_FEATURE_HT))
739 		return -1;
740 
741 	if (cpu_has(c, X86_FEATURE_CMP_LEGACY))
742 		return -1;
743 
744 	if (cpu_has(c, X86_FEATURE_XTOPOLOGY))
745 		return -1;
746 
747 	cpuid(1, &eax, &ebx, &ecx, &edx);
748 
749 	smp_num_siblings = (ebx & 0xff0000) >> 16;
750 	if (smp_num_siblings == 1)
751 		pr_info_once("CPU0: Hyper-Threading is disabled\n");
752 #endif
753 	return 0;
754 }
755 
756 void detect_ht(struct cpuinfo_x86 *c)
757 {
758 #ifdef CONFIG_SMP
759 	int index_msb, core_bits;
760 
761 	if (detect_ht_early(c) < 0)
762 		return;
763 
764 	index_msb = get_count_order(smp_num_siblings);
765 	c->phys_proc_id = apic->phys_pkg_id(c->initial_apicid, index_msb);
766 
767 	smp_num_siblings = smp_num_siblings / c->x86_max_cores;
768 
769 	index_msb = get_count_order(smp_num_siblings);
770 
771 	core_bits = get_count_order(c->x86_max_cores);
772 
773 	c->cpu_core_id = apic->phys_pkg_id(c->initial_apicid, index_msb) &
774 				       ((1 << core_bits) - 1);
775 #endif
776 }
777 
778 static void get_cpu_vendor(struct cpuinfo_x86 *c)
779 {
780 	char *v = c->x86_vendor_id;
781 	int i;
782 
783 	for (i = 0; i < X86_VENDOR_NUM; i++) {
784 		if (!cpu_devs[i])
785 			break;
786 
787 		if (!strcmp(v, cpu_devs[i]->c_ident[0]) ||
788 		    (cpu_devs[i]->c_ident[1] &&
789 		     !strcmp(v, cpu_devs[i]->c_ident[1]))) {
790 
791 			this_cpu = cpu_devs[i];
792 			c->x86_vendor = this_cpu->c_x86_vendor;
793 			return;
794 		}
795 	}
796 
797 	pr_err_once("CPU: vendor_id '%s' unknown, using generic init.\n" \
798 		    "CPU: Your system may be unstable.\n", v);
799 
800 	c->x86_vendor = X86_VENDOR_UNKNOWN;
801 	this_cpu = &default_cpu;
802 }
803 
804 void cpu_detect(struct cpuinfo_x86 *c)
805 {
806 	/* Get vendor name */
807 	cpuid(0x00000000, (unsigned int *)&c->cpuid_level,
808 	      (unsigned int *)&c->x86_vendor_id[0],
809 	      (unsigned int *)&c->x86_vendor_id[8],
810 	      (unsigned int *)&c->x86_vendor_id[4]);
811 
812 	c->x86 = 4;
813 	/* Intel-defined flags: level 0x00000001 */
814 	if (c->cpuid_level >= 0x00000001) {
815 		u32 junk, tfms, cap0, misc;
816 
817 		cpuid(0x00000001, &tfms, &misc, &junk, &cap0);
818 		c->x86		= x86_family(tfms);
819 		c->x86_model	= x86_model(tfms);
820 		c->x86_stepping	= x86_stepping(tfms);
821 
822 		if (cap0 & (1<<19)) {
823 			c->x86_clflush_size = ((misc >> 8) & 0xff) * 8;
824 			c->x86_cache_alignment = c->x86_clflush_size;
825 		}
826 	}
827 }
828 
829 static void apply_forced_caps(struct cpuinfo_x86 *c)
830 {
831 	int i;
832 
833 	for (i = 0; i < NCAPINTS + NBUGINTS; i++) {
834 		c->x86_capability[i] &= ~cpu_caps_cleared[i];
835 		c->x86_capability[i] |= cpu_caps_set[i];
836 	}
837 }
838 
839 static void init_speculation_control(struct cpuinfo_x86 *c)
840 {
841 	/*
842 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
843 	 * and they also have a different bit for STIBP support. Also,
844 	 * a hypervisor might have set the individual AMD bits even on
845 	 * Intel CPUs, for finer-grained selection of what's available.
846 	 */
847 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
848 		set_cpu_cap(c, X86_FEATURE_IBRS);
849 		set_cpu_cap(c, X86_FEATURE_IBPB);
850 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
851 	}
852 
853 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
854 		set_cpu_cap(c, X86_FEATURE_STIBP);
855 
856 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL_SSBD) ||
857 	    cpu_has(c, X86_FEATURE_VIRT_SSBD))
858 		set_cpu_cap(c, X86_FEATURE_SSBD);
859 
860 	if (cpu_has(c, X86_FEATURE_AMD_IBRS)) {
861 		set_cpu_cap(c, X86_FEATURE_IBRS);
862 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
863 	}
864 
865 	if (cpu_has(c, X86_FEATURE_AMD_IBPB))
866 		set_cpu_cap(c, X86_FEATURE_IBPB);
867 
868 	if (cpu_has(c, X86_FEATURE_AMD_STIBP)) {
869 		set_cpu_cap(c, X86_FEATURE_STIBP);
870 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
871 	}
872 
873 	if (cpu_has(c, X86_FEATURE_AMD_SSBD)) {
874 		set_cpu_cap(c, X86_FEATURE_SSBD);
875 		set_cpu_cap(c, X86_FEATURE_MSR_SPEC_CTRL);
876 		clear_cpu_cap(c, X86_FEATURE_VIRT_SSBD);
877 	}
878 }
879 
880 void get_cpu_cap(struct cpuinfo_x86 *c)
881 {
882 	u32 eax, ebx, ecx, edx;
883 
884 	/* Intel-defined flags: level 0x00000001 */
885 	if (c->cpuid_level >= 0x00000001) {
886 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
887 
888 		c->x86_capability[CPUID_1_ECX] = ecx;
889 		c->x86_capability[CPUID_1_EDX] = edx;
890 	}
891 
892 	/* Thermal and Power Management Leaf: level 0x00000006 (eax) */
893 	if (c->cpuid_level >= 0x00000006)
894 		c->x86_capability[CPUID_6_EAX] = cpuid_eax(0x00000006);
895 
896 	/* Additional Intel-defined flags: level 0x00000007 */
897 	if (c->cpuid_level >= 0x00000007) {
898 		cpuid_count(0x00000007, 0, &eax, &ebx, &ecx, &edx);
899 		c->x86_capability[CPUID_7_0_EBX] = ebx;
900 		c->x86_capability[CPUID_7_ECX] = ecx;
901 		c->x86_capability[CPUID_7_EDX] = edx;
902 
903 		/* Check valid sub-leaf index before accessing it */
904 		if (eax >= 1) {
905 			cpuid_count(0x00000007, 1, &eax, &ebx, &ecx, &edx);
906 			c->x86_capability[CPUID_7_1_EAX] = eax;
907 		}
908 	}
909 
910 	/* Extended state features: level 0x0000000d */
911 	if (c->cpuid_level >= 0x0000000d) {
912 		cpuid_count(0x0000000d, 1, &eax, &ebx, &ecx, &edx);
913 
914 		c->x86_capability[CPUID_D_1_EAX] = eax;
915 	}
916 
917 	/* AMD-defined flags: level 0x80000001 */
918 	eax = cpuid_eax(0x80000000);
919 	c->extended_cpuid_level = eax;
920 
921 	if ((eax & 0xffff0000) == 0x80000000) {
922 		if (eax >= 0x80000001) {
923 			cpuid(0x80000001, &eax, &ebx, &ecx, &edx);
924 
925 			c->x86_capability[CPUID_8000_0001_ECX] = ecx;
926 			c->x86_capability[CPUID_8000_0001_EDX] = edx;
927 		}
928 	}
929 
930 	if (c->extended_cpuid_level >= 0x80000007) {
931 		cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
932 
933 		c->x86_capability[CPUID_8000_0007_EBX] = ebx;
934 		c->x86_power = edx;
935 	}
936 
937 	if (c->extended_cpuid_level >= 0x80000008) {
938 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
939 		c->x86_capability[CPUID_8000_0008_EBX] = ebx;
940 	}
941 
942 	if (c->extended_cpuid_level >= 0x8000000a)
943 		c->x86_capability[CPUID_8000_000A_EDX] = cpuid_edx(0x8000000a);
944 
945 	init_scattered_cpuid_features(c);
946 	init_speculation_control(c);
947 
948 	/*
949 	 * Clear/Set all flags overridden by options, after probe.
950 	 * This needs to happen each time we re-probe, which may happen
951 	 * several times during CPU initialization.
952 	 */
953 	apply_forced_caps(c);
954 }
955 
956 void get_cpu_address_sizes(struct cpuinfo_x86 *c)
957 {
958 	u32 eax, ebx, ecx, edx;
959 
960 	if (c->extended_cpuid_level >= 0x80000008) {
961 		cpuid(0x80000008, &eax, &ebx, &ecx, &edx);
962 
963 		c->x86_virt_bits = (eax >> 8) & 0xff;
964 		c->x86_phys_bits = eax & 0xff;
965 	}
966 #ifdef CONFIG_X86_32
967 	else if (cpu_has(c, X86_FEATURE_PAE) || cpu_has(c, X86_FEATURE_PSE36))
968 		c->x86_phys_bits = 36;
969 #endif
970 	c->x86_cache_bits = c->x86_phys_bits;
971 }
972 
973 static void identify_cpu_without_cpuid(struct cpuinfo_x86 *c)
974 {
975 #ifdef CONFIG_X86_32
976 	int i;
977 
978 	/*
979 	 * First of all, decide if this is a 486 or higher
980 	 * It's a 486 if we can modify the AC flag
981 	 */
982 	if (flag_is_changeable_p(X86_EFLAGS_AC))
983 		c->x86 = 4;
984 	else
985 		c->x86 = 3;
986 
987 	for (i = 0; i < X86_VENDOR_NUM; i++)
988 		if (cpu_devs[i] && cpu_devs[i]->c_identify) {
989 			c->x86_vendor_id[0] = 0;
990 			cpu_devs[i]->c_identify(c);
991 			if (c->x86_vendor_id[0]) {
992 				get_cpu_vendor(c);
993 				break;
994 			}
995 		}
996 #endif
997 }
998 
999 #define NO_SPECULATION		BIT(0)
1000 #define NO_MELTDOWN		BIT(1)
1001 #define NO_SSB			BIT(2)
1002 #define NO_L1TF			BIT(3)
1003 #define NO_MDS			BIT(4)
1004 #define MSBDS_ONLY		BIT(5)
1005 #define NO_SWAPGS		BIT(6)
1006 #define NO_ITLB_MULTIHIT	BIT(7)
1007 #define NO_SPECTRE_V2		BIT(8)
1008 
1009 #define VULNWL(vendor, family, model, whitelist)	\
1010 	X86_MATCH_VENDOR_FAM_MODEL(vendor, family, model, whitelist)
1011 
1012 #define VULNWL_INTEL(model, whitelist)		\
1013 	VULNWL(INTEL, 6, INTEL_FAM6_##model, whitelist)
1014 
1015 #define VULNWL_AMD(family, whitelist)		\
1016 	VULNWL(AMD, family, X86_MODEL_ANY, whitelist)
1017 
1018 #define VULNWL_HYGON(family, whitelist)		\
1019 	VULNWL(HYGON, family, X86_MODEL_ANY, whitelist)
1020 
1021 static const __initconst struct x86_cpu_id cpu_vuln_whitelist[] = {
1022 	VULNWL(ANY,	4, X86_MODEL_ANY,	NO_SPECULATION),
1023 	VULNWL(CENTAUR,	5, X86_MODEL_ANY,	NO_SPECULATION),
1024 	VULNWL(INTEL,	5, X86_MODEL_ANY,	NO_SPECULATION),
1025 	VULNWL(NSC,	5, X86_MODEL_ANY,	NO_SPECULATION),
1026 
1027 	/* Intel Family 6 */
1028 	VULNWL_INTEL(ATOM_SALTWELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1029 	VULNWL_INTEL(ATOM_SALTWELL_TABLET,	NO_SPECULATION | NO_ITLB_MULTIHIT),
1030 	VULNWL_INTEL(ATOM_SALTWELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1031 	VULNWL_INTEL(ATOM_BONNELL,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1032 	VULNWL_INTEL(ATOM_BONNELL_MID,		NO_SPECULATION | NO_ITLB_MULTIHIT),
1033 
1034 	VULNWL_INTEL(ATOM_SILVERMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1035 	VULNWL_INTEL(ATOM_SILVERMONT_D,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1036 	VULNWL_INTEL(ATOM_SILVERMONT_MID,	NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1037 	VULNWL_INTEL(ATOM_AIRMONT,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1038 	VULNWL_INTEL(XEON_PHI_KNL,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1039 	VULNWL_INTEL(XEON_PHI_KNM,		NO_SSB | NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1040 
1041 	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
1042 
1043 	VULNWL_INTEL(ATOM_AIRMONT_MID,		NO_L1TF | MSBDS_ONLY | NO_SWAPGS | NO_ITLB_MULTIHIT),
1044 	VULNWL_INTEL(ATOM_AIRMONT_NP,		NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1045 
1046 	VULNWL_INTEL(ATOM_GOLDMONT,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1047 	VULNWL_INTEL(ATOM_GOLDMONT_D,		NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1048 	VULNWL_INTEL(ATOM_GOLDMONT_PLUS,	NO_MDS | NO_L1TF | NO_SWAPGS | NO_ITLB_MULTIHIT),
1049 
1050 	/*
1051 	 * Technically, swapgs isn't serializing on AMD (despite it previously
1052 	 * being documented as such in the APM).  But according to AMD, %gs is
1053 	 * updated non-speculatively, and the issuing of %gs-relative memory
1054 	 * operands will be blocked until the %gs update completes, which is
1055 	 * good enough for our purposes.
1056 	 */
1057 
1058 	VULNWL_INTEL(ATOM_TREMONT_D,		NO_ITLB_MULTIHIT),
1059 
1060 	/* AMD Family 0xf - 0x12 */
1061 	VULNWL_AMD(0x0f,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1062 	VULNWL_AMD(0x10,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1063 	VULNWL_AMD(0x11,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1064 	VULNWL_AMD(0x12,	NO_MELTDOWN | NO_SSB | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1065 
1066 	/* FAMILY_ANY must be last, otherwise 0x0f - 0x12 matches won't work */
1067 	VULNWL_AMD(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1068 	VULNWL_HYGON(X86_FAMILY_ANY,	NO_MELTDOWN | NO_L1TF | NO_MDS | NO_SWAPGS | NO_ITLB_MULTIHIT),
1069 
1070 	/* Zhaoxin Family 7 */
1071 	VULNWL(CENTAUR,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1072 	VULNWL(ZHAOXIN,	7, X86_MODEL_ANY,	NO_SPECTRE_V2 | NO_SWAPGS),
1073 	{}
1074 };
1075 
1076 #define VULNBL_INTEL_STEPPINGS(model, steppings, issues)		   \
1077 	X86_MATCH_VENDOR_FAM_MODEL_STEPPINGS_FEATURE(INTEL, 6,		   \
1078 					    INTEL_FAM6_##model, steppings, \
1079 					    X86_FEATURE_ANY, issues)
1080 
1081 #define SRBDS		BIT(0)
1082 
1083 static const struct x86_cpu_id cpu_vuln_blacklist[] __initconst = {
1084 	VULNBL_INTEL_STEPPINGS(IVYBRIDGE,	X86_STEPPING_ANY,		SRBDS),
1085 	VULNBL_INTEL_STEPPINGS(HASWELL,		X86_STEPPING_ANY,		SRBDS),
1086 	VULNBL_INTEL_STEPPINGS(HASWELL_L,	X86_STEPPING_ANY,		SRBDS),
1087 	VULNBL_INTEL_STEPPINGS(HASWELL_G,	X86_STEPPING_ANY,		SRBDS),
1088 	VULNBL_INTEL_STEPPINGS(BROADWELL_G,	X86_STEPPING_ANY,		SRBDS),
1089 	VULNBL_INTEL_STEPPINGS(BROADWELL,	X86_STEPPING_ANY,		SRBDS),
1090 	VULNBL_INTEL_STEPPINGS(SKYLAKE_L,	X86_STEPPING_ANY,		SRBDS),
1091 	VULNBL_INTEL_STEPPINGS(SKYLAKE,		X86_STEPPING_ANY,		SRBDS),
1092 	VULNBL_INTEL_STEPPINGS(KABYLAKE_L,	X86_STEPPINGS(0x0, 0xC),	SRBDS),
1093 	VULNBL_INTEL_STEPPINGS(KABYLAKE,	X86_STEPPINGS(0x0, 0xD),	SRBDS),
1094 	{}
1095 };
1096 
1097 static bool __init cpu_matches(const struct x86_cpu_id *table, unsigned long which)
1098 {
1099 	const struct x86_cpu_id *m = x86_match_cpu(table);
1100 
1101 	return m && !!(m->driver_data & which);
1102 }
1103 
1104 u64 x86_read_arch_cap_msr(void)
1105 {
1106 	u64 ia32_cap = 0;
1107 
1108 	if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1109 		rdmsrl(MSR_IA32_ARCH_CAPABILITIES, ia32_cap);
1110 
1111 	return ia32_cap;
1112 }
1113 
1114 static void __init cpu_set_bug_bits(struct cpuinfo_x86 *c)
1115 {
1116 	u64 ia32_cap = x86_read_arch_cap_msr();
1117 
1118 	/* Set ITLB_MULTIHIT bug if cpu is not in the whitelist and not mitigated */
1119 	if (!cpu_matches(cpu_vuln_whitelist, NO_ITLB_MULTIHIT) &&
1120 	    !(ia32_cap & ARCH_CAP_PSCHANGE_MC_NO))
1121 		setup_force_cpu_bug(X86_BUG_ITLB_MULTIHIT);
1122 
1123 	if (cpu_matches(cpu_vuln_whitelist, NO_SPECULATION))
1124 		return;
1125 
1126 	setup_force_cpu_bug(X86_BUG_SPECTRE_V1);
1127 
1128 	if (!cpu_matches(cpu_vuln_whitelist, NO_SPECTRE_V2))
1129 		setup_force_cpu_bug(X86_BUG_SPECTRE_V2);
1130 
1131 	if (!cpu_matches(cpu_vuln_whitelist, NO_SSB) &&
1132 	    !(ia32_cap & ARCH_CAP_SSB_NO) &&
1133 	   !cpu_has(c, X86_FEATURE_AMD_SSB_NO))
1134 		setup_force_cpu_bug(X86_BUG_SPEC_STORE_BYPASS);
1135 
1136 	if (ia32_cap & ARCH_CAP_IBRS_ALL)
1137 		setup_force_cpu_cap(X86_FEATURE_IBRS_ENHANCED);
1138 
1139 	if (!cpu_matches(cpu_vuln_whitelist, NO_MDS) &&
1140 	    !(ia32_cap & ARCH_CAP_MDS_NO)) {
1141 		setup_force_cpu_bug(X86_BUG_MDS);
1142 		if (cpu_matches(cpu_vuln_whitelist, MSBDS_ONLY))
1143 			setup_force_cpu_bug(X86_BUG_MSBDS_ONLY);
1144 	}
1145 
1146 	if (!cpu_matches(cpu_vuln_whitelist, NO_SWAPGS))
1147 		setup_force_cpu_bug(X86_BUG_SWAPGS);
1148 
1149 	/*
1150 	 * When the CPU is not mitigated for TAA (TAA_NO=0) set TAA bug when:
1151 	 *	- TSX is supported or
1152 	 *	- TSX_CTRL is present
1153 	 *
1154 	 * TSX_CTRL check is needed for cases when TSX could be disabled before
1155 	 * the kernel boot e.g. kexec.
1156 	 * TSX_CTRL check alone is not sufficient for cases when the microcode
1157 	 * update is not present or running as guest that don't get TSX_CTRL.
1158 	 */
1159 	if (!(ia32_cap & ARCH_CAP_TAA_NO) &&
1160 	    (cpu_has(c, X86_FEATURE_RTM) ||
1161 	     (ia32_cap & ARCH_CAP_TSX_CTRL_MSR)))
1162 		setup_force_cpu_bug(X86_BUG_TAA);
1163 
1164 	/*
1165 	 * SRBDS affects CPUs which support RDRAND or RDSEED and are listed
1166 	 * in the vulnerability blacklist.
1167 	 */
1168 	if ((cpu_has(c, X86_FEATURE_RDRAND) ||
1169 	     cpu_has(c, X86_FEATURE_RDSEED)) &&
1170 	    cpu_matches(cpu_vuln_blacklist, SRBDS))
1171 		    setup_force_cpu_bug(X86_BUG_SRBDS);
1172 
1173 	if (cpu_matches(cpu_vuln_whitelist, NO_MELTDOWN))
1174 		return;
1175 
1176 	/* Rogue Data Cache Load? No! */
1177 	if (ia32_cap & ARCH_CAP_RDCL_NO)
1178 		return;
1179 
1180 	setup_force_cpu_bug(X86_BUG_CPU_MELTDOWN);
1181 
1182 	if (cpu_matches(cpu_vuln_whitelist, NO_L1TF))
1183 		return;
1184 
1185 	setup_force_cpu_bug(X86_BUG_L1TF);
1186 }
1187 
1188 /*
1189  * The NOPL instruction is supposed to exist on all CPUs of family >= 6;
1190  * unfortunately, that's not true in practice because of early VIA
1191  * chips and (more importantly) broken virtualizers that are not easy
1192  * to detect. In the latter case it doesn't even *fail* reliably, so
1193  * probing for it doesn't even work. Disable it completely on 32-bit
1194  * unless we can find a reliable way to detect all the broken cases.
1195  * Enable it explicitly on 64-bit for non-constant inputs of cpu_has().
1196  */
1197 static void detect_nopl(void)
1198 {
1199 #ifdef CONFIG_X86_32
1200 	setup_clear_cpu_cap(X86_FEATURE_NOPL);
1201 #else
1202 	setup_force_cpu_cap(X86_FEATURE_NOPL);
1203 #endif
1204 }
1205 
1206 /*
1207  * Do minimum CPU detection early.
1208  * Fields really needed: vendor, cpuid_level, family, model, mask,
1209  * cache alignment.
1210  * The others are not touched to avoid unwanted side effects.
1211  *
1212  * WARNING: this function is only called on the boot CPU.  Don't add code
1213  * here that is supposed to run on all CPUs.
1214  */
1215 static void __init early_identify_cpu(struct cpuinfo_x86 *c)
1216 {
1217 #ifdef CONFIG_X86_64
1218 	c->x86_clflush_size = 64;
1219 	c->x86_phys_bits = 36;
1220 	c->x86_virt_bits = 48;
1221 #else
1222 	c->x86_clflush_size = 32;
1223 	c->x86_phys_bits = 32;
1224 	c->x86_virt_bits = 32;
1225 #endif
1226 	c->x86_cache_alignment = c->x86_clflush_size;
1227 
1228 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1229 	c->extended_cpuid_level = 0;
1230 
1231 	if (!have_cpuid_p())
1232 		identify_cpu_without_cpuid(c);
1233 
1234 	/* cyrix could have cpuid enabled via c_identify()*/
1235 	if (have_cpuid_p()) {
1236 		cpu_detect(c);
1237 		get_cpu_vendor(c);
1238 		get_cpu_cap(c);
1239 		get_cpu_address_sizes(c);
1240 		setup_force_cpu_cap(X86_FEATURE_CPUID);
1241 
1242 		if (this_cpu->c_early_init)
1243 			this_cpu->c_early_init(c);
1244 
1245 		c->cpu_index = 0;
1246 		filter_cpuid_features(c, false);
1247 
1248 		if (this_cpu->c_bsp_init)
1249 			this_cpu->c_bsp_init(c);
1250 	} else {
1251 		setup_clear_cpu_cap(X86_FEATURE_CPUID);
1252 	}
1253 
1254 	setup_force_cpu_cap(X86_FEATURE_ALWAYS);
1255 
1256 	cpu_set_bug_bits(c);
1257 
1258 	cpu_set_core_cap_bits(c);
1259 
1260 	fpu__init_system(c);
1261 
1262 #ifdef CONFIG_X86_32
1263 	/*
1264 	 * Regardless of whether PCID is enumerated, the SDM says
1265 	 * that it can't be enabled in 32-bit mode.
1266 	 */
1267 	setup_clear_cpu_cap(X86_FEATURE_PCID);
1268 #endif
1269 
1270 	/*
1271 	 * Later in the boot process pgtable_l5_enabled() relies on
1272 	 * cpu_feature_enabled(X86_FEATURE_LA57). If 5-level paging is not
1273 	 * enabled by this point we need to clear the feature bit to avoid
1274 	 * false-positives at the later stage.
1275 	 *
1276 	 * pgtable_l5_enabled() can be false here for several reasons:
1277 	 *  - 5-level paging is disabled compile-time;
1278 	 *  - it's 32-bit kernel;
1279 	 *  - machine doesn't support 5-level paging;
1280 	 *  - user specified 'no5lvl' in kernel command line.
1281 	 */
1282 	if (!pgtable_l5_enabled())
1283 		setup_clear_cpu_cap(X86_FEATURE_LA57);
1284 
1285 	detect_nopl();
1286 }
1287 
1288 void __init early_cpu_init(void)
1289 {
1290 	const struct cpu_dev *const *cdev;
1291 	int count = 0;
1292 
1293 #ifdef CONFIG_PROCESSOR_SELECT
1294 	pr_info("KERNEL supported cpus:\n");
1295 #endif
1296 
1297 	for (cdev = __x86_cpu_dev_start; cdev < __x86_cpu_dev_end; cdev++) {
1298 		const struct cpu_dev *cpudev = *cdev;
1299 
1300 		if (count >= X86_VENDOR_NUM)
1301 			break;
1302 		cpu_devs[count] = cpudev;
1303 		count++;
1304 
1305 #ifdef CONFIG_PROCESSOR_SELECT
1306 		{
1307 			unsigned int j;
1308 
1309 			for (j = 0; j < 2; j++) {
1310 				if (!cpudev->c_ident[j])
1311 					continue;
1312 				pr_info("  %s %s\n", cpudev->c_vendor,
1313 					cpudev->c_ident[j]);
1314 			}
1315 		}
1316 #endif
1317 	}
1318 	early_identify_cpu(&boot_cpu_data);
1319 }
1320 
1321 static void detect_null_seg_behavior(struct cpuinfo_x86 *c)
1322 {
1323 #ifdef CONFIG_X86_64
1324 	/*
1325 	 * Empirically, writing zero to a segment selector on AMD does
1326 	 * not clear the base, whereas writing zero to a segment
1327 	 * selector on Intel does clear the base.  Intel's behavior
1328 	 * allows slightly faster context switches in the common case
1329 	 * where GS is unused by the prev and next threads.
1330 	 *
1331 	 * Since neither vendor documents this anywhere that I can see,
1332 	 * detect it directly instead of hardcoding the choice by
1333 	 * vendor.
1334 	 *
1335 	 * I've designated AMD's behavior as the "bug" because it's
1336 	 * counterintuitive and less friendly.
1337 	 */
1338 
1339 	unsigned long old_base, tmp;
1340 	rdmsrl(MSR_FS_BASE, old_base);
1341 	wrmsrl(MSR_FS_BASE, 1);
1342 	loadsegment(fs, 0);
1343 	rdmsrl(MSR_FS_BASE, tmp);
1344 	if (tmp != 0)
1345 		set_cpu_bug(c, X86_BUG_NULL_SEG);
1346 	wrmsrl(MSR_FS_BASE, old_base);
1347 #endif
1348 }
1349 
1350 static void generic_identify(struct cpuinfo_x86 *c)
1351 {
1352 	c->extended_cpuid_level = 0;
1353 
1354 	if (!have_cpuid_p())
1355 		identify_cpu_without_cpuid(c);
1356 
1357 	/* cyrix could have cpuid enabled via c_identify()*/
1358 	if (!have_cpuid_p())
1359 		return;
1360 
1361 	cpu_detect(c);
1362 
1363 	get_cpu_vendor(c);
1364 
1365 	get_cpu_cap(c);
1366 
1367 	get_cpu_address_sizes(c);
1368 
1369 	if (c->cpuid_level >= 0x00000001) {
1370 		c->initial_apicid = (cpuid_ebx(1) >> 24) & 0xFF;
1371 #ifdef CONFIG_X86_32
1372 # ifdef CONFIG_SMP
1373 		c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1374 # else
1375 		c->apicid = c->initial_apicid;
1376 # endif
1377 #endif
1378 		c->phys_proc_id = c->initial_apicid;
1379 	}
1380 
1381 	get_model_name(c); /* Default name */
1382 
1383 	detect_null_seg_behavior(c);
1384 
1385 	/*
1386 	 * ESPFIX is a strange bug.  All real CPUs have it.  Paravirt
1387 	 * systems that run Linux at CPL > 0 may or may not have the
1388 	 * issue, but, even if they have the issue, there's absolutely
1389 	 * nothing we can do about it because we can't use the real IRET
1390 	 * instruction.
1391 	 *
1392 	 * NB: For the time being, only 32-bit kernels support
1393 	 * X86_BUG_ESPFIX as such.  64-bit kernels directly choose
1394 	 * whether to apply espfix using paravirt hooks.  If any
1395 	 * non-paravirt system ever shows up that does *not* have the
1396 	 * ESPFIX issue, we can change this.
1397 	 */
1398 #ifdef CONFIG_X86_32
1399 # ifdef CONFIG_PARAVIRT_XXL
1400 	do {
1401 		extern void native_iret(void);
1402 		if (pv_ops.cpu.iret == native_iret)
1403 			set_cpu_bug(c, X86_BUG_ESPFIX);
1404 	} while (0);
1405 # else
1406 	set_cpu_bug(c, X86_BUG_ESPFIX);
1407 # endif
1408 #endif
1409 }
1410 
1411 /*
1412  * Validate that ACPI/mptables have the same information about the
1413  * effective APIC id and update the package map.
1414  */
1415 static void validate_apic_and_package_id(struct cpuinfo_x86 *c)
1416 {
1417 #ifdef CONFIG_SMP
1418 	unsigned int apicid, cpu = smp_processor_id();
1419 
1420 	apicid = apic->cpu_present_to_apicid(cpu);
1421 
1422 	if (apicid != c->apicid) {
1423 		pr_err(FW_BUG "CPU%u: APIC id mismatch. Firmware: %x APIC: %x\n",
1424 		       cpu, apicid, c->initial_apicid);
1425 	}
1426 	BUG_ON(topology_update_package_map(c->phys_proc_id, cpu));
1427 	BUG_ON(topology_update_die_map(c->cpu_die_id, cpu));
1428 #else
1429 	c->logical_proc_id = 0;
1430 #endif
1431 }
1432 
1433 /*
1434  * This does the hard work of actually picking apart the CPU stuff...
1435  */
1436 static void identify_cpu(struct cpuinfo_x86 *c)
1437 {
1438 	int i;
1439 
1440 	c->loops_per_jiffy = loops_per_jiffy;
1441 	c->x86_cache_size = 0;
1442 	c->x86_vendor = X86_VENDOR_UNKNOWN;
1443 	c->x86_model = c->x86_stepping = 0;	/* So far unknown... */
1444 	c->x86_vendor_id[0] = '\0'; /* Unset */
1445 	c->x86_model_id[0] = '\0';  /* Unset */
1446 	c->x86_max_cores = 1;
1447 	c->x86_coreid_bits = 0;
1448 	c->cu_id = 0xff;
1449 #ifdef CONFIG_X86_64
1450 	c->x86_clflush_size = 64;
1451 	c->x86_phys_bits = 36;
1452 	c->x86_virt_bits = 48;
1453 #else
1454 	c->cpuid_level = -1;	/* CPUID not detected */
1455 	c->x86_clflush_size = 32;
1456 	c->x86_phys_bits = 32;
1457 	c->x86_virt_bits = 32;
1458 #endif
1459 	c->x86_cache_alignment = c->x86_clflush_size;
1460 	memset(&c->x86_capability, 0, sizeof(c->x86_capability));
1461 #ifdef CONFIG_X86_VMX_FEATURE_NAMES
1462 	memset(&c->vmx_capability, 0, sizeof(c->vmx_capability));
1463 #endif
1464 
1465 	generic_identify(c);
1466 
1467 	if (this_cpu->c_identify)
1468 		this_cpu->c_identify(c);
1469 
1470 	/* Clear/Set all flags overridden by options, after probe */
1471 	apply_forced_caps(c);
1472 
1473 #ifdef CONFIG_X86_64
1474 	c->apicid = apic->phys_pkg_id(c->initial_apicid, 0);
1475 #endif
1476 
1477 	/*
1478 	 * Vendor-specific initialization.  In this section we
1479 	 * canonicalize the feature flags, meaning if there are
1480 	 * features a certain CPU supports which CPUID doesn't
1481 	 * tell us, CPUID claiming incorrect flags, or other bugs,
1482 	 * we handle them here.
1483 	 *
1484 	 * At the end of this section, c->x86_capability better
1485 	 * indicate the features this CPU genuinely supports!
1486 	 */
1487 	if (this_cpu->c_init)
1488 		this_cpu->c_init(c);
1489 
1490 	/* Disable the PN if appropriate */
1491 	squash_the_stupid_serial_number(c);
1492 
1493 	/* Set up SMEP/SMAP/UMIP */
1494 	setup_smep(c);
1495 	setup_smap(c);
1496 	setup_umip(c);
1497 
1498 	/*
1499 	 * The vendor-specific functions might have changed features.
1500 	 * Now we do "generic changes."
1501 	 */
1502 
1503 	/* Filter out anything that depends on CPUID levels we don't have */
1504 	filter_cpuid_features(c, true);
1505 
1506 	/* If the model name is still unset, do table lookup. */
1507 	if (!c->x86_model_id[0]) {
1508 		const char *p;
1509 		p = table_lookup_model(c);
1510 		if (p)
1511 			strcpy(c->x86_model_id, p);
1512 		else
1513 			/* Last resort... */
1514 			sprintf(c->x86_model_id, "%02x/%02x",
1515 				c->x86, c->x86_model);
1516 	}
1517 
1518 #ifdef CONFIG_X86_64
1519 	detect_ht(c);
1520 #endif
1521 
1522 	x86_init_rdrand(c);
1523 	setup_pku(c);
1524 
1525 	/*
1526 	 * Clear/Set all flags overridden by options, need do it
1527 	 * before following smp all cpus cap AND.
1528 	 */
1529 	apply_forced_caps(c);
1530 
1531 	/*
1532 	 * On SMP, boot_cpu_data holds the common feature set between
1533 	 * all CPUs; so make sure that we indicate which features are
1534 	 * common between the CPUs.  The first time this routine gets
1535 	 * executed, c == &boot_cpu_data.
1536 	 */
1537 	if (c != &boot_cpu_data) {
1538 		/* AND the already accumulated flags with these */
1539 		for (i = 0; i < NCAPINTS; i++)
1540 			boot_cpu_data.x86_capability[i] &= c->x86_capability[i];
1541 
1542 		/* OR, i.e. replicate the bug flags */
1543 		for (i = NCAPINTS; i < NCAPINTS + NBUGINTS; i++)
1544 			c->x86_capability[i] |= boot_cpu_data.x86_capability[i];
1545 	}
1546 
1547 	/* Init Machine Check Exception if available. */
1548 	mcheck_cpu_init(c);
1549 
1550 	select_idle_routine(c);
1551 
1552 #ifdef CONFIG_NUMA
1553 	numa_add_cpu(smp_processor_id());
1554 #endif
1555 }
1556 
1557 /*
1558  * Set up the CPU state needed to execute SYSENTER/SYSEXIT instructions
1559  * on 32-bit kernels:
1560  */
1561 #ifdef CONFIG_X86_32
1562 void enable_sep_cpu(void)
1563 {
1564 	struct tss_struct *tss;
1565 	int cpu;
1566 
1567 	if (!boot_cpu_has(X86_FEATURE_SEP))
1568 		return;
1569 
1570 	cpu = get_cpu();
1571 	tss = &per_cpu(cpu_tss_rw, cpu);
1572 
1573 	/*
1574 	 * We cache MSR_IA32_SYSENTER_CS's value in the TSS's ss1 field --
1575 	 * see the big comment in struct x86_hw_tss's definition.
1576 	 */
1577 
1578 	tss->x86_tss.ss1 = __KERNEL_CS;
1579 	wrmsr(MSR_IA32_SYSENTER_CS, tss->x86_tss.ss1, 0);
1580 	wrmsr(MSR_IA32_SYSENTER_ESP, (unsigned long)(cpu_entry_stack(cpu) + 1), 0);
1581 	wrmsr(MSR_IA32_SYSENTER_EIP, (unsigned long)entry_SYSENTER_32, 0);
1582 
1583 	put_cpu();
1584 }
1585 #endif
1586 
1587 void __init identify_boot_cpu(void)
1588 {
1589 	identify_cpu(&boot_cpu_data);
1590 #ifdef CONFIG_X86_32
1591 	sysenter_setup();
1592 	enable_sep_cpu();
1593 #endif
1594 	cpu_detect_tlb(&boot_cpu_data);
1595 	setup_cr_pinning();
1596 
1597 	tsx_init();
1598 }
1599 
1600 void identify_secondary_cpu(struct cpuinfo_x86 *c)
1601 {
1602 	BUG_ON(c == &boot_cpu_data);
1603 	identify_cpu(c);
1604 #ifdef CONFIG_X86_32
1605 	enable_sep_cpu();
1606 #endif
1607 	mtrr_ap_init();
1608 	validate_apic_and_package_id(c);
1609 	x86_spec_ctrl_setup_ap();
1610 	update_srbds_msr();
1611 }
1612 
1613 static __init int setup_noclflush(char *arg)
1614 {
1615 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSH);
1616 	setup_clear_cpu_cap(X86_FEATURE_CLFLUSHOPT);
1617 	return 1;
1618 }
1619 __setup("noclflush", setup_noclflush);
1620 
1621 void print_cpu_info(struct cpuinfo_x86 *c)
1622 {
1623 	const char *vendor = NULL;
1624 
1625 	if (c->x86_vendor < X86_VENDOR_NUM) {
1626 		vendor = this_cpu->c_vendor;
1627 	} else {
1628 		if (c->cpuid_level >= 0)
1629 			vendor = c->x86_vendor_id;
1630 	}
1631 
1632 	if (vendor && !strstr(c->x86_model_id, vendor))
1633 		pr_cont("%s ", vendor);
1634 
1635 	if (c->x86_model_id[0])
1636 		pr_cont("%s", c->x86_model_id);
1637 	else
1638 		pr_cont("%d86", c->x86);
1639 
1640 	pr_cont(" (family: 0x%x, model: 0x%x", c->x86, c->x86_model);
1641 
1642 	if (c->x86_stepping || c->cpuid_level >= 0)
1643 		pr_cont(", stepping: 0x%x)\n", c->x86_stepping);
1644 	else
1645 		pr_cont(")\n");
1646 }
1647 
1648 /*
1649  * clearcpuid= was already parsed in fpu__init_parse_early_param.
1650  * But we need to keep a dummy __setup around otherwise it would
1651  * show up as an environment variable for init.
1652  */
1653 static __init int setup_clearcpuid(char *arg)
1654 {
1655 	return 1;
1656 }
1657 __setup("clearcpuid=", setup_clearcpuid);
1658 
1659 #ifdef CONFIG_X86_64
1660 DEFINE_PER_CPU_FIRST(struct fixed_percpu_data,
1661 		     fixed_percpu_data) __aligned(PAGE_SIZE) __visible;
1662 EXPORT_PER_CPU_SYMBOL_GPL(fixed_percpu_data);
1663 
1664 /*
1665  * The following percpu variables are hot.  Align current_task to
1666  * cacheline size such that they fall in the same cacheline.
1667  */
1668 DEFINE_PER_CPU(struct task_struct *, current_task) ____cacheline_aligned =
1669 	&init_task;
1670 EXPORT_PER_CPU_SYMBOL(current_task);
1671 
1672 DEFINE_PER_CPU(struct irq_stack *, hardirq_stack_ptr);
1673 DEFINE_PER_CPU(unsigned int, irq_count) __visible = -1;
1674 
1675 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1676 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1677 
1678 /* May not be marked __init: used by software suspend */
1679 void syscall_init(void)
1680 {
1681 	wrmsr(MSR_STAR, 0, (__USER32_CS << 16) | __KERNEL_CS);
1682 	wrmsrl(MSR_LSTAR, (unsigned long)entry_SYSCALL_64);
1683 
1684 #ifdef CONFIG_IA32_EMULATION
1685 	wrmsrl(MSR_CSTAR, (unsigned long)entry_SYSCALL_compat);
1686 	/*
1687 	 * This only works on Intel CPUs.
1688 	 * On AMD CPUs these MSRs are 32-bit, CPU truncates MSR_IA32_SYSENTER_EIP.
1689 	 * This does not cause SYSENTER to jump to the wrong location, because
1690 	 * AMD doesn't allow SYSENTER in long mode (either 32- or 64-bit).
1691 	 */
1692 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS);
1693 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP,
1694 		    (unsigned long)(cpu_entry_stack(smp_processor_id()) + 1));
1695 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, (u64)entry_SYSENTER_compat);
1696 #else
1697 	wrmsrl(MSR_CSTAR, (unsigned long)ignore_sysret);
1698 	wrmsrl_safe(MSR_IA32_SYSENTER_CS, (u64)GDT_ENTRY_INVALID_SEG);
1699 	wrmsrl_safe(MSR_IA32_SYSENTER_ESP, 0ULL);
1700 	wrmsrl_safe(MSR_IA32_SYSENTER_EIP, 0ULL);
1701 #endif
1702 
1703 	/* Flags to clear on syscall */
1704 	wrmsrl(MSR_SYSCALL_MASK,
1705 	       X86_EFLAGS_TF|X86_EFLAGS_DF|X86_EFLAGS_IF|
1706 	       X86_EFLAGS_IOPL|X86_EFLAGS_AC|X86_EFLAGS_NT);
1707 }
1708 
1709 #else	/* CONFIG_X86_64 */
1710 
1711 DEFINE_PER_CPU(struct task_struct *, current_task) = &init_task;
1712 EXPORT_PER_CPU_SYMBOL(current_task);
1713 DEFINE_PER_CPU(int, __preempt_count) = INIT_PREEMPT_COUNT;
1714 EXPORT_PER_CPU_SYMBOL(__preempt_count);
1715 
1716 /*
1717  * On x86_32, vm86 modifies tss.sp0, so sp0 isn't a reliable way to find
1718  * the top of the kernel stack.  Use an extra percpu variable to track the
1719  * top of the kernel stack directly.
1720  */
1721 DEFINE_PER_CPU(unsigned long, cpu_current_top_of_stack) =
1722 	(unsigned long)&init_thread_union + THREAD_SIZE;
1723 EXPORT_PER_CPU_SYMBOL(cpu_current_top_of_stack);
1724 
1725 #ifdef CONFIG_STACKPROTECTOR
1726 DEFINE_PER_CPU_ALIGNED(struct stack_canary, stack_canary);
1727 #endif
1728 
1729 #endif	/* CONFIG_X86_64 */
1730 
1731 /*
1732  * Clear all 6 debug registers:
1733  */
1734 static void clear_all_debug_regs(void)
1735 {
1736 	int i;
1737 
1738 	for (i = 0; i < 8; i++) {
1739 		/* Ignore db4, db5 */
1740 		if ((i == 4) || (i == 5))
1741 			continue;
1742 
1743 		set_debugreg(0, i);
1744 	}
1745 }
1746 
1747 #ifdef CONFIG_KGDB
1748 /*
1749  * Restore debug regs if using kgdbwait and you have a kernel debugger
1750  * connection established.
1751  */
1752 static void dbg_restore_debug_regs(void)
1753 {
1754 	if (unlikely(kgdb_connected && arch_kgdb_ops.correct_hw_break))
1755 		arch_kgdb_ops.correct_hw_break();
1756 }
1757 #else /* ! CONFIG_KGDB */
1758 #define dbg_restore_debug_regs()
1759 #endif /* ! CONFIG_KGDB */
1760 
1761 static void wait_for_master_cpu(int cpu)
1762 {
1763 #ifdef CONFIG_SMP
1764 	/*
1765 	 * wait for ACK from master CPU before continuing
1766 	 * with AP initialization
1767 	 */
1768 	WARN_ON(cpumask_test_and_set_cpu(cpu, cpu_initialized_mask));
1769 	while (!cpumask_test_cpu(cpu, cpu_callout_mask))
1770 		cpu_relax();
1771 #endif
1772 }
1773 
1774 #ifdef CONFIG_X86_64
1775 static inline void setup_getcpu(int cpu)
1776 {
1777 	unsigned long cpudata = vdso_encode_cpunode(cpu, early_cpu_to_node(cpu));
1778 	struct desc_struct d = { };
1779 
1780 	if (boot_cpu_has(X86_FEATURE_RDTSCP))
1781 		write_rdtscp_aux(cpudata);
1782 
1783 	/* Store CPU and node number in limit. */
1784 	d.limit0 = cpudata;
1785 	d.limit1 = cpudata >> 16;
1786 
1787 	d.type = 5;		/* RO data, expand down, accessed */
1788 	d.dpl = 3;		/* Visible to user code */
1789 	d.s = 1;		/* Not a system segment */
1790 	d.p = 1;		/* Present */
1791 	d.d = 1;		/* 32-bit */
1792 
1793 	write_gdt_entry(get_cpu_gdt_rw(cpu), GDT_ENTRY_CPUNODE, &d, DESCTYPE_S);
1794 }
1795 
1796 static inline void ucode_cpu_init(int cpu)
1797 {
1798 	if (cpu)
1799 		load_ucode_ap();
1800 }
1801 
1802 static inline void tss_setup_ist(struct tss_struct *tss)
1803 {
1804 	/* Set up the per-CPU TSS IST stacks */
1805 	tss->x86_tss.ist[IST_INDEX_DF] = __this_cpu_ist_top_va(DF);
1806 	tss->x86_tss.ist[IST_INDEX_NMI] = __this_cpu_ist_top_va(NMI);
1807 	tss->x86_tss.ist[IST_INDEX_DB] = __this_cpu_ist_top_va(DB);
1808 	tss->x86_tss.ist[IST_INDEX_MCE] = __this_cpu_ist_top_va(MCE);
1809 }
1810 
1811 #else /* CONFIG_X86_64 */
1812 
1813 static inline void setup_getcpu(int cpu) { }
1814 
1815 static inline void ucode_cpu_init(int cpu)
1816 {
1817 	show_ucode_info_early();
1818 }
1819 
1820 static inline void tss_setup_ist(struct tss_struct *tss) { }
1821 
1822 #endif /* !CONFIG_X86_64 */
1823 
1824 static inline void tss_setup_io_bitmap(struct tss_struct *tss)
1825 {
1826 	tss->x86_tss.io_bitmap_base = IO_BITMAP_OFFSET_INVALID;
1827 
1828 #ifdef CONFIG_X86_IOPL_IOPERM
1829 	tss->io_bitmap.prev_max = 0;
1830 	tss->io_bitmap.prev_sequence = 0;
1831 	memset(tss->io_bitmap.bitmap, 0xff, sizeof(tss->io_bitmap.bitmap));
1832 	/*
1833 	 * Invalidate the extra array entry past the end of the all
1834 	 * permission bitmap as required by the hardware.
1835 	 */
1836 	tss->io_bitmap.mapall[IO_BITMAP_LONGS] = ~0UL;
1837 #endif
1838 }
1839 
1840 /*
1841  * cpu_init() initializes state that is per-CPU. Some data is already
1842  * initialized (naturally) in the bootstrap process, such as the GDT
1843  * and IDT. We reload them nevertheless, this function acts as a
1844  * 'CPU state barrier', nothing should get across.
1845  */
1846 void cpu_init(void)
1847 {
1848 	struct tss_struct *tss = this_cpu_ptr(&cpu_tss_rw);
1849 	struct task_struct *cur = current;
1850 	int cpu = raw_smp_processor_id();
1851 
1852 	wait_for_master_cpu(cpu);
1853 
1854 	ucode_cpu_init(cpu);
1855 
1856 #ifdef CONFIG_NUMA
1857 	if (this_cpu_read(numa_node) == 0 &&
1858 	    early_cpu_to_node(cpu) != NUMA_NO_NODE)
1859 		set_numa_node(early_cpu_to_node(cpu));
1860 #endif
1861 	setup_getcpu(cpu);
1862 
1863 	pr_debug("Initializing CPU#%d\n", cpu);
1864 
1865 	if (IS_ENABLED(CONFIG_X86_64) || cpu_feature_enabled(X86_FEATURE_VME) ||
1866 	    boot_cpu_has(X86_FEATURE_TSC) || boot_cpu_has(X86_FEATURE_DE))
1867 		cr4_clear_bits(X86_CR4_VME|X86_CR4_PVI|X86_CR4_TSD|X86_CR4_DE);
1868 
1869 	/*
1870 	 * Initialize the per-CPU GDT with the boot GDT,
1871 	 * and set up the GDT descriptor:
1872 	 */
1873 	switch_to_new_gdt(cpu);
1874 	load_current_idt();
1875 
1876 	if (IS_ENABLED(CONFIG_X86_64)) {
1877 		loadsegment(fs, 0);
1878 		memset(cur->thread.tls_array, 0, GDT_ENTRY_TLS_ENTRIES * 8);
1879 		syscall_init();
1880 
1881 		wrmsrl(MSR_FS_BASE, 0);
1882 		wrmsrl(MSR_KERNEL_GS_BASE, 0);
1883 		barrier();
1884 
1885 		x2apic_setup();
1886 	}
1887 
1888 	mmgrab(&init_mm);
1889 	cur->active_mm = &init_mm;
1890 	BUG_ON(cur->mm);
1891 	initialize_tlbstate_and_flush();
1892 	enter_lazy_tlb(&init_mm, cur);
1893 
1894 	/* Initialize the TSS. */
1895 	tss_setup_ist(tss);
1896 	tss_setup_io_bitmap(tss);
1897 	set_tss_desc(cpu, &get_cpu_entry_area(cpu)->tss.x86_tss);
1898 
1899 	load_TR_desc();
1900 	/*
1901 	 * sp0 points to the entry trampoline stack regardless of what task
1902 	 * is running.
1903 	 */
1904 	load_sp0((unsigned long)(cpu_entry_stack(cpu) + 1));
1905 
1906 	load_mm_ldt(&init_mm);
1907 
1908 	clear_all_debug_regs();
1909 	dbg_restore_debug_regs();
1910 
1911 	doublefault_init_cpu_tss();
1912 
1913 	fpu__init_cpu();
1914 
1915 	if (is_uv_system())
1916 		uv_cpu_init();
1917 
1918 	load_fixmap_gdt(cpu);
1919 }
1920 
1921 /*
1922  * The microcode loader calls this upon late microcode load to recheck features,
1923  * only when microcode has been updated. Caller holds microcode_mutex and CPU
1924  * hotplug lock.
1925  */
1926 void microcode_check(void)
1927 {
1928 	struct cpuinfo_x86 info;
1929 
1930 	perf_check_microcode();
1931 
1932 	/* Reload CPUID max function as it might've changed. */
1933 	info.cpuid_level = cpuid_eax(0);
1934 
1935 	/*
1936 	 * Copy all capability leafs to pick up the synthetic ones so that
1937 	 * memcmp() below doesn't fail on that. The ones coming from CPUID will
1938 	 * get overwritten in get_cpu_cap().
1939 	 */
1940 	memcpy(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability));
1941 
1942 	get_cpu_cap(&info);
1943 
1944 	if (!memcmp(&info.x86_capability, &boot_cpu_data.x86_capability, sizeof(info.x86_capability)))
1945 		return;
1946 
1947 	pr_warn("x86/CPU: CPU features have changed after loading microcode, but might not take effect.\n");
1948 	pr_warn("x86/CPU: Please consider either early loading through initrd/built-in or a potential BIOS update.\n");
1949 }
1950 
1951 /*
1952  * Invoked from core CPU hotplug code after hotplug operations
1953  */
1954 void arch_smt_update(void)
1955 {
1956 	/* Handle the speculative execution misfeatures */
1957 	cpu_bugs_smt_update();
1958 	/* Check whether IPI broadcasting can be enabled */
1959 	apic_smt_update();
1960 }
1961