xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision 3381df09)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/kernel.h>
3 
4 #include <linux/string.h>
5 #include <linux/bitops.h>
6 #include <linux/smp.h>
7 #include <linux/sched.h>
8 #include <linux/sched/clock.h>
9 #include <linux/thread_info.h>
10 #include <linux/init.h>
11 #include <linux/uaccess.h>
12 
13 #include <asm/cpufeature.h>
14 #include <asm/pgtable.h>
15 #include <asm/msr.h>
16 #include <asm/bugs.h>
17 #include <asm/cpu.h>
18 #include <asm/intel-family.h>
19 #include <asm/microcode_intel.h>
20 #include <asm/hwcap2.h>
21 #include <asm/elf.h>
22 #include <asm/cpu_device_id.h>
23 #include <asm/cmdline.h>
24 #include <asm/traps.h>
25 
26 #ifdef CONFIG_X86_64
27 #include <linux/topology.h>
28 #endif
29 
30 #include "cpu.h"
31 
32 #ifdef CONFIG_X86_LOCAL_APIC
33 #include <asm/mpspec.h>
34 #include <asm/apic.h>
35 #endif
36 
37 enum split_lock_detect_state {
38 	sld_off = 0,
39 	sld_warn,
40 	sld_fatal,
41 };
42 
43 /*
44  * Default to sld_off because most systems do not support split lock detection
45  * split_lock_setup() will switch this to sld_warn on systems that support
46  * split lock detect, unless there is a command line override.
47  */
48 static enum split_lock_detect_state sld_state __ro_after_init = sld_off;
49 static u64 msr_test_ctrl_cache __ro_after_init;
50 
51 /*
52  * Processors which have self-snooping capability can handle conflicting
53  * memory type across CPUs by snooping its own cache. However, there exists
54  * CPU models in which having conflicting memory types still leads to
55  * unpredictable behavior, machine check errors, or hangs. Clear this
56  * feature to prevent its use on machines with known erratas.
57  */
58 static void check_memory_type_self_snoop_errata(struct cpuinfo_x86 *c)
59 {
60 	switch (c->x86_model) {
61 	case INTEL_FAM6_CORE_YONAH:
62 	case INTEL_FAM6_CORE2_MEROM:
63 	case INTEL_FAM6_CORE2_MEROM_L:
64 	case INTEL_FAM6_CORE2_PENRYN:
65 	case INTEL_FAM6_CORE2_DUNNINGTON:
66 	case INTEL_FAM6_NEHALEM:
67 	case INTEL_FAM6_NEHALEM_G:
68 	case INTEL_FAM6_NEHALEM_EP:
69 	case INTEL_FAM6_NEHALEM_EX:
70 	case INTEL_FAM6_WESTMERE:
71 	case INTEL_FAM6_WESTMERE_EP:
72 	case INTEL_FAM6_SANDYBRIDGE:
73 		setup_clear_cpu_cap(X86_FEATURE_SELFSNOOP);
74 	}
75 }
76 
77 static bool ring3mwait_disabled __read_mostly;
78 
79 static int __init ring3mwait_disable(char *__unused)
80 {
81 	ring3mwait_disabled = true;
82 	return 0;
83 }
84 __setup("ring3mwait=disable", ring3mwait_disable);
85 
86 static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c)
87 {
88 	/*
89 	 * Ring 3 MONITOR/MWAIT feature cannot be detected without
90 	 * cpu model and family comparison.
91 	 */
92 	if (c->x86 != 6)
93 		return;
94 	switch (c->x86_model) {
95 	case INTEL_FAM6_XEON_PHI_KNL:
96 	case INTEL_FAM6_XEON_PHI_KNM:
97 		break;
98 	default:
99 		return;
100 	}
101 
102 	if (ring3mwait_disabled)
103 		return;
104 
105 	set_cpu_cap(c, X86_FEATURE_RING3MWAIT);
106 	this_cpu_or(msr_misc_features_shadow,
107 		    1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT);
108 
109 	if (c == &boot_cpu_data)
110 		ELF_HWCAP2 |= HWCAP2_RING3MWAIT;
111 }
112 
113 /*
114  * Early microcode releases for the Spectre v2 mitigation were broken.
115  * Information taken from;
116  * - https://newsroom.intel.com/wp-content/uploads/sites/11/2018/03/microcode-update-guidance.pdf
117  * - https://kb.vmware.com/s/article/52345
118  * - Microcode revisions observed in the wild
119  * - Release note from 20180108 microcode release
120  */
121 struct sku_microcode {
122 	u8 model;
123 	u8 stepping;
124 	u32 microcode;
125 };
126 static const struct sku_microcode spectre_bad_microcodes[] = {
127 	{ INTEL_FAM6_KABYLAKE,		0x0B,	0x80 },
128 	{ INTEL_FAM6_KABYLAKE,		0x0A,	0x80 },
129 	{ INTEL_FAM6_KABYLAKE,		0x09,	0x80 },
130 	{ INTEL_FAM6_KABYLAKE_L,	0x0A,	0x80 },
131 	{ INTEL_FAM6_KABYLAKE_L,	0x09,	0x80 },
132 	{ INTEL_FAM6_SKYLAKE_X,		0x03,	0x0100013e },
133 	{ INTEL_FAM6_SKYLAKE_X,		0x04,	0x0200003c },
134 	{ INTEL_FAM6_BROADWELL,		0x04,	0x28 },
135 	{ INTEL_FAM6_BROADWELL_G,	0x01,	0x1b },
136 	{ INTEL_FAM6_BROADWELL_D,	0x02,	0x14 },
137 	{ INTEL_FAM6_BROADWELL_D,	0x03,	0x07000011 },
138 	{ INTEL_FAM6_BROADWELL_X,	0x01,	0x0b000025 },
139 	{ INTEL_FAM6_HASWELL_L,		0x01,	0x21 },
140 	{ INTEL_FAM6_HASWELL_G,		0x01,	0x18 },
141 	{ INTEL_FAM6_HASWELL,		0x03,	0x23 },
142 	{ INTEL_FAM6_HASWELL_X,		0x02,	0x3b },
143 	{ INTEL_FAM6_HASWELL_X,		0x04,	0x10 },
144 	{ INTEL_FAM6_IVYBRIDGE_X,	0x04,	0x42a },
145 	/* Observed in the wild */
146 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x06,	0x61b },
147 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x07,	0x712 },
148 };
149 
150 static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
151 {
152 	int i;
153 
154 	/*
155 	 * We know that the hypervisor lie to us on the microcode version so
156 	 * we may as well hope that it is running the correct version.
157 	 */
158 	if (cpu_has(c, X86_FEATURE_HYPERVISOR))
159 		return false;
160 
161 	if (c->x86 != 6)
162 		return false;
163 
164 	for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
165 		if (c->x86_model == spectre_bad_microcodes[i].model &&
166 		    c->x86_stepping == spectre_bad_microcodes[i].stepping)
167 			return (c->microcode <= spectre_bad_microcodes[i].microcode);
168 	}
169 	return false;
170 }
171 
172 static void early_init_intel(struct cpuinfo_x86 *c)
173 {
174 	u64 misc_enable;
175 
176 	/* Unmask CPUID levels if masked: */
177 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
178 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
179 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
180 			c->cpuid_level = cpuid_eax(0);
181 			get_cpu_cap(c);
182 		}
183 	}
184 
185 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
186 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
187 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
188 
189 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
190 		c->microcode = intel_get_microcode_revision();
191 
192 	/* Now if any of them are set, check the blacklist and clear the lot */
193 	if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
194 	     cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
195 	     cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
196 	     cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
197 		pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
198 		setup_clear_cpu_cap(X86_FEATURE_IBRS);
199 		setup_clear_cpu_cap(X86_FEATURE_IBPB);
200 		setup_clear_cpu_cap(X86_FEATURE_STIBP);
201 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
202 		setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL);
203 		setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
204 		setup_clear_cpu_cap(X86_FEATURE_SSBD);
205 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD);
206 	}
207 
208 	/*
209 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
210 	 *
211 	 * A race condition between speculative fetches and invalidating
212 	 * a large page.  This is worked around in microcode, but we
213 	 * need the microcode to have already been loaded... so if it is
214 	 * not, recommend a BIOS update and disable large pages.
215 	 */
216 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_stepping <= 2 &&
217 	    c->microcode < 0x20e) {
218 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
219 		clear_cpu_cap(c, X86_FEATURE_PSE);
220 	}
221 
222 #ifdef CONFIG_X86_64
223 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
224 #else
225 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
226 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
227 		c->x86_cache_alignment = 128;
228 #endif
229 
230 	/* CPUID workaround for 0F33/0F34 CPU */
231 	if (c->x86 == 0xF && c->x86_model == 0x3
232 	    && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
233 		c->x86_phys_bits = 36;
234 
235 	/*
236 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
237 	 * with P/T states and does not stop in deep C-states.
238 	 *
239 	 * It is also reliable across cores and sockets. (but not across
240 	 * cabinets - we turn it off in that case explicitly.)
241 	 */
242 	if (c->x86_power & (1 << 8)) {
243 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
244 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
245 	}
246 
247 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
248 	if (c->x86 == 6) {
249 		switch (c->x86_model) {
250 		case INTEL_FAM6_ATOM_SALTWELL_MID:
251 		case INTEL_FAM6_ATOM_SALTWELL_TABLET:
252 		case INTEL_FAM6_ATOM_SILVERMONT_MID:
253 		case INTEL_FAM6_ATOM_AIRMONT_NP:
254 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
255 			break;
256 		default:
257 			break;
258 		}
259 	}
260 
261 	/*
262 	 * There is a known erratum on Pentium III and Core Solo
263 	 * and Core Duo CPUs.
264 	 * " Page with PAT set to WC while associated MTRR is UC
265 	 *   may consolidate to UC "
266 	 * Because of this erratum, it is better to stick with
267 	 * setting WC in MTRR rather than using PAT on these CPUs.
268 	 *
269 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
270 	 */
271 	if (c->x86 == 6 && c->x86_model < 15)
272 		clear_cpu_cap(c, X86_FEATURE_PAT);
273 
274 	/*
275 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
276 	 * clear the fast string and enhanced fast string CPU capabilities.
277 	 */
278 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
279 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
280 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
281 			pr_info("Disabled fast string operations\n");
282 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
283 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
284 		}
285 	}
286 
287 	/*
288 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
289 	 * "The operating system also is required to invalidate (i.e., flush)
290 	 *  the TLB when any changes are made to any of the page table entries.
291 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
292 	 *
293 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
294 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
295 	 * to be modified.
296 	 */
297 	if (c->x86 == 5 && c->x86_model == 9) {
298 		pr_info("Disabling PGE capability bit\n");
299 		setup_clear_cpu_cap(X86_FEATURE_PGE);
300 	}
301 
302 	if (c->cpuid_level >= 0x00000001) {
303 		u32 eax, ebx, ecx, edx;
304 
305 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
306 		/*
307 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
308 		 * apicids which are reserved per package. Store the resulting
309 		 * shift value for the package management code.
310 		 */
311 		if (edx & (1U << 28))
312 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
313 	}
314 
315 	check_memory_type_self_snoop_errata(c);
316 
317 	/*
318 	 * Get the number of SMT siblings early from the extended topology
319 	 * leaf, if available. Otherwise try the legacy SMT detection.
320 	 */
321 	if (detect_extended_topology_early(c) < 0)
322 		detect_ht_early(c);
323 }
324 
325 #ifdef CONFIG_X86_32
326 /*
327  *	Early probe support logic for ppro memory erratum #50
328  *
329  *	This is called before we do cpu ident work
330  */
331 
332 int ppro_with_ram_bug(void)
333 {
334 	/* Uses data from early_cpu_detect now */
335 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
336 	    boot_cpu_data.x86 == 6 &&
337 	    boot_cpu_data.x86_model == 1 &&
338 	    boot_cpu_data.x86_stepping < 8) {
339 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
340 		return 1;
341 	}
342 	return 0;
343 }
344 
345 static void intel_smp_check(struct cpuinfo_x86 *c)
346 {
347 	/* calling is from identify_secondary_cpu() ? */
348 	if (!c->cpu_index)
349 		return;
350 
351 	/*
352 	 * Mask B, Pentium, but not Pentium MMX
353 	 */
354 	if (c->x86 == 5 &&
355 	    c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
356 	    c->x86_model <= 3) {
357 		/*
358 		 * Remember we have B step Pentia with bugs
359 		 */
360 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
361 				    "with B stepping processors.\n");
362 	}
363 }
364 
365 static int forcepae;
366 static int __init forcepae_setup(char *__unused)
367 {
368 	forcepae = 1;
369 	return 1;
370 }
371 __setup("forcepae", forcepae_setup);
372 
373 static void intel_workarounds(struct cpuinfo_x86 *c)
374 {
375 #ifdef CONFIG_X86_F00F_BUG
376 	/*
377 	 * All models of Pentium and Pentium with MMX technology CPUs
378 	 * have the F0 0F bug, which lets nonprivileged users lock up the
379 	 * system. Announce that the fault handler will be checking for it.
380 	 * The Quark is also family 5, but does not have the same bug.
381 	 */
382 	clear_cpu_bug(c, X86_BUG_F00F);
383 	if (c->x86 == 5 && c->x86_model < 9) {
384 		static int f00f_workaround_enabled;
385 
386 		set_cpu_bug(c, X86_BUG_F00F);
387 		if (!f00f_workaround_enabled) {
388 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
389 			f00f_workaround_enabled = 1;
390 		}
391 	}
392 #endif
393 
394 	/*
395 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
396 	 * model 3 mask 3
397 	 */
398 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
399 		clear_cpu_cap(c, X86_FEATURE_SEP);
400 
401 	/*
402 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
403 	 * functionally usable PAE implementation.
404 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
405 	 */
406 	if (forcepae) {
407 		pr_warn("PAE forced!\n");
408 		set_cpu_cap(c, X86_FEATURE_PAE);
409 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
410 	}
411 
412 	/*
413 	 * P4 Xeon erratum 037 workaround.
414 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
415 	 */
416 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
417 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
418 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
419 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
420 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
421 		}
422 	}
423 
424 	/*
425 	 * See if we have a good local APIC by checking for buggy Pentia,
426 	 * i.e. all B steppings and the C2 stepping of P54C when using their
427 	 * integrated APIC (see 11AP erratum in "Pentium Processor
428 	 * Specification Update").
429 	 */
430 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
431 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
432 		set_cpu_bug(c, X86_BUG_11AP);
433 
434 
435 #ifdef CONFIG_X86_INTEL_USERCOPY
436 	/*
437 	 * Set up the preferred alignment for movsl bulk memory moves
438 	 */
439 	switch (c->x86) {
440 	case 4:		/* 486: untested */
441 		break;
442 	case 5:		/* Old Pentia: untested */
443 		break;
444 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
445 		movsl_mask.mask = 7;
446 		break;
447 	case 15:	/* P4 is OK down to 8-byte alignment */
448 		movsl_mask.mask = 7;
449 		break;
450 	}
451 #endif
452 
453 	intel_smp_check(c);
454 }
455 #else
456 static void intel_workarounds(struct cpuinfo_x86 *c)
457 {
458 }
459 #endif
460 
461 static void srat_detect_node(struct cpuinfo_x86 *c)
462 {
463 #ifdef CONFIG_NUMA
464 	unsigned node;
465 	int cpu = smp_processor_id();
466 
467 	/* Don't do the funky fallback heuristics the AMD version employs
468 	   for now. */
469 	node = numa_cpu_node(cpu);
470 	if (node == NUMA_NO_NODE || !node_online(node)) {
471 		/* reuse the value from init_cpu_to_node() */
472 		node = cpu_to_node(cpu);
473 	}
474 	numa_set_node(cpu, node);
475 #endif
476 }
477 
478 #define MSR_IA32_TME_ACTIVATE		0x982
479 
480 /* Helpers to access TME_ACTIVATE MSR */
481 #define TME_ACTIVATE_LOCKED(x)		(x & 0x1)
482 #define TME_ACTIVATE_ENABLED(x)		(x & 0x2)
483 
484 #define TME_ACTIVATE_POLICY(x)		((x >> 4) & 0xf)	/* Bits 7:4 */
485 #define TME_ACTIVATE_POLICY_AES_XTS_128	0
486 
487 #define TME_ACTIVATE_KEYID_BITS(x)	((x >> 32) & 0xf)	/* Bits 35:32 */
488 
489 #define TME_ACTIVATE_CRYPTO_ALGS(x)	((x >> 48) & 0xffff)	/* Bits 63:48 */
490 #define TME_ACTIVATE_CRYPTO_AES_XTS_128	1
491 
492 /* Values for mktme_status (SW only construct) */
493 #define MKTME_ENABLED			0
494 #define MKTME_DISABLED			1
495 #define MKTME_UNINITIALIZED		2
496 static int mktme_status = MKTME_UNINITIALIZED;
497 
498 static void detect_tme(struct cpuinfo_x86 *c)
499 {
500 	u64 tme_activate, tme_policy, tme_crypto_algs;
501 	int keyid_bits = 0, nr_keyids = 0;
502 	static u64 tme_activate_cpu0 = 0;
503 
504 	rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
505 
506 	if (mktme_status != MKTME_UNINITIALIZED) {
507 		if (tme_activate != tme_activate_cpu0) {
508 			/* Broken BIOS? */
509 			pr_err_once("x86/tme: configuration is inconsistent between CPUs\n");
510 			pr_err_once("x86/tme: MKTME is not usable\n");
511 			mktme_status = MKTME_DISABLED;
512 
513 			/* Proceed. We may need to exclude bits from x86_phys_bits. */
514 		}
515 	} else {
516 		tme_activate_cpu0 = tme_activate;
517 	}
518 
519 	if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
520 		pr_info_once("x86/tme: not enabled by BIOS\n");
521 		mktme_status = MKTME_DISABLED;
522 		return;
523 	}
524 
525 	if (mktme_status != MKTME_UNINITIALIZED)
526 		goto detect_keyid_bits;
527 
528 	pr_info("x86/tme: enabled by BIOS\n");
529 
530 	tme_policy = TME_ACTIVATE_POLICY(tme_activate);
531 	if (tme_policy != TME_ACTIVATE_POLICY_AES_XTS_128)
532 		pr_warn("x86/tme: Unknown policy is active: %#llx\n", tme_policy);
533 
534 	tme_crypto_algs = TME_ACTIVATE_CRYPTO_ALGS(tme_activate);
535 	if (!(tme_crypto_algs & TME_ACTIVATE_CRYPTO_AES_XTS_128)) {
536 		pr_err("x86/mktme: No known encryption algorithm is supported: %#llx\n",
537 				tme_crypto_algs);
538 		mktme_status = MKTME_DISABLED;
539 	}
540 detect_keyid_bits:
541 	keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
542 	nr_keyids = (1UL << keyid_bits) - 1;
543 	if (nr_keyids) {
544 		pr_info_once("x86/mktme: enabled by BIOS\n");
545 		pr_info_once("x86/mktme: %d KeyIDs available\n", nr_keyids);
546 	} else {
547 		pr_info_once("x86/mktme: disabled by BIOS\n");
548 	}
549 
550 	if (mktme_status == MKTME_UNINITIALIZED) {
551 		/* MKTME is usable */
552 		mktme_status = MKTME_ENABLED;
553 	}
554 
555 	/*
556 	 * KeyID bits effectively lower the number of physical address
557 	 * bits.  Update cpuinfo_x86::x86_phys_bits accordingly.
558 	 */
559 	c->x86_phys_bits -= keyid_bits;
560 }
561 
562 static void init_cpuid_fault(struct cpuinfo_x86 *c)
563 {
564 	u64 msr;
565 
566 	if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
567 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
568 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
569 	}
570 }
571 
572 static void init_intel_misc_features(struct cpuinfo_x86 *c)
573 {
574 	u64 msr;
575 
576 	if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
577 		return;
578 
579 	/* Clear all MISC features */
580 	this_cpu_write(msr_misc_features_shadow, 0);
581 
582 	/* Check features and update capabilities and shadow control bits */
583 	init_cpuid_fault(c);
584 	probe_xeon_phi_r3mwait(c);
585 
586 	msr = this_cpu_read(msr_misc_features_shadow);
587 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
588 }
589 
590 static void split_lock_init(void);
591 
592 static void init_intel(struct cpuinfo_x86 *c)
593 {
594 	early_init_intel(c);
595 
596 	intel_workarounds(c);
597 
598 	/*
599 	 * Detect the extended topology information if available. This
600 	 * will reinitialise the initial_apicid which will be used
601 	 * in init_intel_cacheinfo()
602 	 */
603 	detect_extended_topology(c);
604 
605 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
606 		/*
607 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
608 		 * detection.
609 		 */
610 		detect_num_cpu_cores(c);
611 #ifdef CONFIG_X86_32
612 		detect_ht(c);
613 #endif
614 	}
615 
616 	init_intel_cacheinfo(c);
617 
618 	if (c->cpuid_level > 9) {
619 		unsigned eax = cpuid_eax(10);
620 		/* Check for version and the number of counters */
621 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
622 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
623 	}
624 
625 	if (cpu_has(c, X86_FEATURE_XMM2))
626 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
627 
628 	if (boot_cpu_has(X86_FEATURE_DS)) {
629 		unsigned int l1, l2;
630 
631 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
632 		if (!(l1 & (1<<11)))
633 			set_cpu_cap(c, X86_FEATURE_BTS);
634 		if (!(l1 & (1<<12)))
635 			set_cpu_cap(c, X86_FEATURE_PEBS);
636 	}
637 
638 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
639 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
640 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
641 
642 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
643 		((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
644 		set_cpu_bug(c, X86_BUG_MONITOR);
645 
646 #ifdef CONFIG_X86_64
647 	if (c->x86 == 15)
648 		c->x86_cache_alignment = c->x86_clflush_size * 2;
649 	if (c->x86 == 6)
650 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
651 #else
652 	/*
653 	 * Names for the Pentium II/Celeron processors
654 	 * detectable only by also checking the cache size.
655 	 * Dixon is NOT a Celeron.
656 	 */
657 	if (c->x86 == 6) {
658 		unsigned int l2 = c->x86_cache_size;
659 		char *p = NULL;
660 
661 		switch (c->x86_model) {
662 		case 5:
663 			if (l2 == 0)
664 				p = "Celeron (Covington)";
665 			else if (l2 == 256)
666 				p = "Mobile Pentium II (Dixon)";
667 			break;
668 
669 		case 6:
670 			if (l2 == 128)
671 				p = "Celeron (Mendocino)";
672 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
673 				p = "Celeron-A";
674 			break;
675 
676 		case 8:
677 			if (l2 == 128)
678 				p = "Celeron (Coppermine)";
679 			break;
680 		}
681 
682 		if (p)
683 			strcpy(c->x86_model_id, p);
684 	}
685 
686 	if (c->x86 == 15)
687 		set_cpu_cap(c, X86_FEATURE_P4);
688 	if (c->x86 == 6)
689 		set_cpu_cap(c, X86_FEATURE_P3);
690 #endif
691 
692 	/* Work around errata */
693 	srat_detect_node(c);
694 
695 	init_ia32_feat_ctl(c);
696 
697 	if (cpu_has(c, X86_FEATURE_TME))
698 		detect_tme(c);
699 
700 	init_intel_misc_features(c);
701 
702 	if (tsx_ctrl_state == TSX_CTRL_ENABLE)
703 		tsx_enable();
704 	if (tsx_ctrl_state == TSX_CTRL_DISABLE)
705 		tsx_disable();
706 
707 	split_lock_init();
708 }
709 
710 #ifdef CONFIG_X86_32
711 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
712 {
713 	/*
714 	 * Intel PIII Tualatin. This comes in two flavours.
715 	 * One has 256kb of cache, the other 512. We have no way
716 	 * to determine which, so we use a boottime override
717 	 * for the 512kb model, and assume 256 otherwise.
718 	 */
719 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
720 		size = 256;
721 
722 	/*
723 	 * Intel Quark SoC X1000 contains a 4-way set associative
724 	 * 16K cache with a 16 byte cache line and 256 lines per tag
725 	 */
726 	if ((c->x86 == 5) && (c->x86_model == 9))
727 		size = 16;
728 	return size;
729 }
730 #endif
731 
732 #define TLB_INST_4K	0x01
733 #define TLB_INST_4M	0x02
734 #define TLB_INST_2M_4M	0x03
735 
736 #define TLB_INST_ALL	0x05
737 #define TLB_INST_1G	0x06
738 
739 #define TLB_DATA_4K	0x11
740 #define TLB_DATA_4M	0x12
741 #define TLB_DATA_2M_4M	0x13
742 #define TLB_DATA_4K_4M	0x14
743 
744 #define TLB_DATA_1G	0x16
745 
746 #define TLB_DATA0_4K	0x21
747 #define TLB_DATA0_4M	0x22
748 #define TLB_DATA0_2M_4M	0x23
749 
750 #define STLB_4K		0x41
751 #define STLB_4K_2M	0x42
752 
753 static const struct _tlb_table intel_tlb_table[] = {
754 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
755 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
756 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
757 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
758 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
759 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
760 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages" },
761 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
762 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
763 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
764 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
765 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
766 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
767 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
768 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
769 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
770 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
771 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
772 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
773 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
774 	{ 0x6b, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 8-way associative" },
775 	{ 0x6c, TLB_DATA_2M_4M,		128,	" TLB_DATA 2 MByte or 4 MByte pages, 8-way associative" },
776 	{ 0x6d, TLB_DATA_1G,		16,	" TLB_DATA 1 GByte pages, fully associative" },
777 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
778 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
779 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
780 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
781 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
782 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
783 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
784 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
785 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
786 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
787 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
788 	{ 0xc2, TLB_DATA_2M_4M,		16,	" TLB_DATA 2 MByte/4MByte pages, 4-way associative" },
789 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
790 	{ 0x00, 0, 0 }
791 };
792 
793 static void intel_tlb_lookup(const unsigned char desc)
794 {
795 	unsigned char k;
796 	if (desc == 0)
797 		return;
798 
799 	/* look up this descriptor in the table */
800 	for (k = 0; intel_tlb_table[k].descriptor != desc &&
801 	     intel_tlb_table[k].descriptor != 0; k++)
802 		;
803 
804 	if (intel_tlb_table[k].tlb_type == 0)
805 		return;
806 
807 	switch (intel_tlb_table[k].tlb_type) {
808 	case STLB_4K:
809 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
810 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
811 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
812 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
813 		break;
814 	case STLB_4K_2M:
815 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
816 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
817 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
818 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
819 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
820 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
821 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
822 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
823 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
824 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
825 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
826 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
827 		break;
828 	case TLB_INST_ALL:
829 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
830 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
831 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
832 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
833 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
834 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
835 		break;
836 	case TLB_INST_4K:
837 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
838 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
839 		break;
840 	case TLB_INST_4M:
841 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
842 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
843 		break;
844 	case TLB_INST_2M_4M:
845 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
846 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
847 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
848 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
849 		break;
850 	case TLB_DATA_4K:
851 	case TLB_DATA0_4K:
852 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
853 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
854 		break;
855 	case TLB_DATA_4M:
856 	case TLB_DATA0_4M:
857 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
858 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
859 		break;
860 	case TLB_DATA_2M_4M:
861 	case TLB_DATA0_2M_4M:
862 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
863 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
864 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
865 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
866 		break;
867 	case TLB_DATA_4K_4M:
868 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
869 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
870 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
871 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
872 		break;
873 	case TLB_DATA_1G:
874 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
875 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
876 		break;
877 	}
878 }
879 
880 static void intel_detect_tlb(struct cpuinfo_x86 *c)
881 {
882 	int i, j, n;
883 	unsigned int regs[4];
884 	unsigned char *desc = (unsigned char *)regs;
885 
886 	if (c->cpuid_level < 2)
887 		return;
888 
889 	/* Number of times to iterate */
890 	n = cpuid_eax(2) & 0xFF;
891 
892 	for (i = 0 ; i < n ; i++) {
893 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
894 
895 		/* If bit 31 is set, this is an unknown format */
896 		for (j = 0 ; j < 3 ; j++)
897 			if (regs[j] & (1 << 31))
898 				regs[j] = 0;
899 
900 		/* Byte 0 is level count, not a descriptor */
901 		for (j = 1 ; j < 16 ; j++)
902 			intel_tlb_lookup(desc[j]);
903 	}
904 }
905 
906 static const struct cpu_dev intel_cpu_dev = {
907 	.c_vendor	= "Intel",
908 	.c_ident	= { "GenuineIntel" },
909 #ifdef CONFIG_X86_32
910 	.legacy_models = {
911 		{ .family = 4, .model_names =
912 		  {
913 			  [0] = "486 DX-25/33",
914 			  [1] = "486 DX-50",
915 			  [2] = "486 SX",
916 			  [3] = "486 DX/2",
917 			  [4] = "486 SL",
918 			  [5] = "486 SX/2",
919 			  [7] = "486 DX/2-WB",
920 			  [8] = "486 DX/4",
921 			  [9] = "486 DX/4-WB"
922 		  }
923 		},
924 		{ .family = 5, .model_names =
925 		  {
926 			  [0] = "Pentium 60/66 A-step",
927 			  [1] = "Pentium 60/66",
928 			  [2] = "Pentium 75 - 200",
929 			  [3] = "OverDrive PODP5V83",
930 			  [4] = "Pentium MMX",
931 			  [7] = "Mobile Pentium 75 - 200",
932 			  [8] = "Mobile Pentium MMX",
933 			  [9] = "Quark SoC X1000",
934 		  }
935 		},
936 		{ .family = 6, .model_names =
937 		  {
938 			  [0] = "Pentium Pro A-step",
939 			  [1] = "Pentium Pro",
940 			  [3] = "Pentium II (Klamath)",
941 			  [4] = "Pentium II (Deschutes)",
942 			  [5] = "Pentium II (Deschutes)",
943 			  [6] = "Mobile Pentium II",
944 			  [7] = "Pentium III (Katmai)",
945 			  [8] = "Pentium III (Coppermine)",
946 			  [10] = "Pentium III (Cascades)",
947 			  [11] = "Pentium III (Tualatin)",
948 		  }
949 		},
950 		{ .family = 15, .model_names =
951 		  {
952 			  [0] = "Pentium 4 (Unknown)",
953 			  [1] = "Pentium 4 (Willamette)",
954 			  [2] = "Pentium 4 (Northwood)",
955 			  [4] = "Pentium 4 (Foster)",
956 			  [5] = "Pentium 4 (Foster)",
957 		  }
958 		},
959 	},
960 	.legacy_cache_size = intel_size_cache,
961 #endif
962 	.c_detect_tlb	= intel_detect_tlb,
963 	.c_early_init   = early_init_intel,
964 	.c_init		= init_intel,
965 	.c_x86_vendor	= X86_VENDOR_INTEL,
966 };
967 
968 cpu_dev_register(intel_cpu_dev);
969 
970 #undef pr_fmt
971 #define pr_fmt(fmt) "x86/split lock detection: " fmt
972 
973 static const struct {
974 	const char			*option;
975 	enum split_lock_detect_state	state;
976 } sld_options[] __initconst = {
977 	{ "off",	sld_off   },
978 	{ "warn",	sld_warn  },
979 	{ "fatal",	sld_fatal },
980 };
981 
982 static inline bool match_option(const char *arg, int arglen, const char *opt)
983 {
984 	int len = strlen(opt);
985 
986 	return len == arglen && !strncmp(arg, opt, len);
987 }
988 
989 static bool split_lock_verify_msr(bool on)
990 {
991 	u64 ctrl, tmp;
992 
993 	if (rdmsrl_safe(MSR_TEST_CTRL, &ctrl))
994 		return false;
995 	if (on)
996 		ctrl |= MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
997 	else
998 		ctrl &= ~MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
999 	if (wrmsrl_safe(MSR_TEST_CTRL, ctrl))
1000 		return false;
1001 	rdmsrl(MSR_TEST_CTRL, tmp);
1002 	return ctrl == tmp;
1003 }
1004 
1005 static void __init split_lock_setup(void)
1006 {
1007 	enum split_lock_detect_state state = sld_warn;
1008 	char arg[20];
1009 	int i, ret;
1010 
1011 	if (!split_lock_verify_msr(false)) {
1012 		pr_info("MSR access failed: Disabled\n");
1013 		return;
1014 	}
1015 
1016 	ret = cmdline_find_option(boot_command_line, "split_lock_detect",
1017 				  arg, sizeof(arg));
1018 	if (ret >= 0) {
1019 		for (i = 0; i < ARRAY_SIZE(sld_options); i++) {
1020 			if (match_option(arg, ret, sld_options[i].option)) {
1021 				state = sld_options[i].state;
1022 				break;
1023 			}
1024 		}
1025 	}
1026 
1027 	switch (state) {
1028 	case sld_off:
1029 		pr_info("disabled\n");
1030 		return;
1031 	case sld_warn:
1032 		pr_info("warning about user-space split_locks\n");
1033 		break;
1034 	case sld_fatal:
1035 		pr_info("sending SIGBUS on user-space split_locks\n");
1036 		break;
1037 	}
1038 
1039 	rdmsrl(MSR_TEST_CTRL, msr_test_ctrl_cache);
1040 
1041 	if (!split_lock_verify_msr(true)) {
1042 		pr_info("MSR access failed: Disabled\n");
1043 		return;
1044 	}
1045 
1046 	sld_state = state;
1047 	setup_force_cpu_cap(X86_FEATURE_SPLIT_LOCK_DETECT);
1048 }
1049 
1050 /*
1051  * MSR_TEST_CTRL is per core, but we treat it like a per CPU MSR. Locking
1052  * is not implemented as one thread could undo the setting of the other
1053  * thread immediately after dropping the lock anyway.
1054  */
1055 static void sld_update_msr(bool on)
1056 {
1057 	u64 test_ctrl_val = msr_test_ctrl_cache;
1058 
1059 	if (on)
1060 		test_ctrl_val |= MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
1061 
1062 	wrmsrl(MSR_TEST_CTRL, test_ctrl_val);
1063 }
1064 
1065 static void split_lock_init(void)
1066 {
1067 	split_lock_verify_msr(sld_state != sld_off);
1068 }
1069 
1070 static void split_lock_warn(unsigned long ip)
1071 {
1072 	pr_warn_ratelimited("#AC: %s/%d took a split_lock trap at address: 0x%lx\n",
1073 			    current->comm, current->pid, ip);
1074 
1075 	/*
1076 	 * Disable the split lock detection for this task so it can make
1077 	 * progress and set TIF_SLD so the detection is re-enabled via
1078 	 * switch_to_sld() when the task is scheduled out.
1079 	 */
1080 	sld_update_msr(false);
1081 	set_tsk_thread_flag(current, TIF_SLD);
1082 }
1083 
1084 bool handle_guest_split_lock(unsigned long ip)
1085 {
1086 	if (sld_state == sld_warn) {
1087 		split_lock_warn(ip);
1088 		return true;
1089 	}
1090 
1091 	pr_warn_once("#AC: %s/%d %s split_lock trap at address: 0x%lx\n",
1092 		     current->comm, current->pid,
1093 		     sld_state == sld_fatal ? "fatal" : "bogus", ip);
1094 
1095 	current->thread.error_code = 0;
1096 	current->thread.trap_nr = X86_TRAP_AC;
1097 	force_sig_fault(SIGBUS, BUS_ADRALN, NULL);
1098 	return false;
1099 }
1100 EXPORT_SYMBOL_GPL(handle_guest_split_lock);
1101 
1102 bool handle_user_split_lock(struct pt_regs *regs, long error_code)
1103 {
1104 	if ((regs->flags & X86_EFLAGS_AC) || sld_state == sld_fatal)
1105 		return false;
1106 	split_lock_warn(regs->ip);
1107 	return true;
1108 }
1109 
1110 /*
1111  * This function is called only when switching between tasks with
1112  * different split-lock detection modes. It sets the MSR for the
1113  * mode of the new task. This is right most of the time, but since
1114  * the MSR is shared by hyperthreads on a physical core there can
1115  * be glitches when the two threads need different modes.
1116  */
1117 void switch_to_sld(unsigned long tifn)
1118 {
1119 	sld_update_msr(!(tifn & _TIF_SLD));
1120 }
1121 
1122 #define SPLIT_LOCK_CPU(model) {X86_VENDOR_INTEL, 6, model, X86_FEATURE_ANY}
1123 
1124 /*
1125  * The following processors have the split lock detection feature. But
1126  * since they don't have the IA32_CORE_CAPABILITIES MSR, the feature cannot
1127  * be enumerated. Enable it by family and model matching on these
1128  * processors.
1129  */
1130 static const struct x86_cpu_id split_lock_cpu_ids[] __initconst = {
1131 	SPLIT_LOCK_CPU(INTEL_FAM6_ICELAKE_X),
1132 	SPLIT_LOCK_CPU(INTEL_FAM6_ICELAKE_L),
1133 	{}
1134 };
1135 
1136 void __init cpu_set_core_cap_bits(struct cpuinfo_x86 *c)
1137 {
1138 	u64 ia32_core_caps = 0;
1139 
1140 	if (c->x86_vendor != X86_VENDOR_INTEL)
1141 		return;
1142 	if (cpu_has(c, X86_FEATURE_CORE_CAPABILITIES)) {
1143 		/* Enumerate features reported in IA32_CORE_CAPABILITIES MSR. */
1144 		rdmsrl(MSR_IA32_CORE_CAPS, ia32_core_caps);
1145 	} else if (!boot_cpu_has(X86_FEATURE_HYPERVISOR)) {
1146 		/* Enumerate split lock detection by family and model. */
1147 		if (x86_match_cpu(split_lock_cpu_ids))
1148 			ia32_core_caps |= MSR_IA32_CORE_CAPS_SPLIT_LOCK_DETECT;
1149 	}
1150 
1151 	if (ia32_core_caps & MSR_IA32_CORE_CAPS_SPLIT_LOCK_DETECT)
1152 		split_lock_setup();
1153 }
1154