xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision 2634682f)
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/kernel.h>
3 #include <linux/pgtable.h>
4 
5 #include <linux/string.h>
6 #include <linux/bitops.h>
7 #include <linux/smp.h>
8 #include <linux/sched.h>
9 #include <linux/sched/clock.h>
10 #include <linux/thread_info.h>
11 #include <linux/init.h>
12 #include <linux/uaccess.h>
13 
14 #include <asm/cpufeature.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 #include <asm/resctrl.h>
26 #include <asm/numa.h>
27 
28 #ifdef CONFIG_X86_64
29 #include <linux/topology.h>
30 #endif
31 
32 #include "cpu.h"
33 
34 #ifdef CONFIG_X86_LOCAL_APIC
35 #include <asm/mpspec.h>
36 #include <asm/apic.h>
37 #endif
38 
39 enum split_lock_detect_state {
40 	sld_off = 0,
41 	sld_warn,
42 	sld_fatal,
43 };
44 
45 /*
46  * Default to sld_off because most systems do not support split lock detection
47  * split_lock_setup() will switch this to sld_warn on systems that support
48  * split lock detect, unless there is a command line override.
49  */
50 static enum split_lock_detect_state sld_state __ro_after_init = sld_off;
51 static u64 msr_test_ctrl_cache __ro_after_init;
52 
53 /*
54  * With a name like MSR_TEST_CTL it should go without saying, but don't touch
55  * MSR_TEST_CTL unless the CPU is one of the whitelisted models.  Writing it
56  * on CPUs that do not support SLD can cause fireworks, even when writing '0'.
57  */
58 static bool cpu_model_supports_sld __ro_after_init;
59 
60 /*
61  * Processors which have self-snooping capability can handle conflicting
62  * memory type across CPUs by snooping its own cache. However, there exists
63  * CPU models in which having conflicting memory types still leads to
64  * unpredictable behavior, machine check errors, or hangs. Clear this
65  * feature to prevent its use on machines with known erratas.
66  */
67 static void check_memory_type_self_snoop_errata(struct cpuinfo_x86 *c)
68 {
69 	switch (c->x86_model) {
70 	case INTEL_FAM6_CORE_YONAH:
71 	case INTEL_FAM6_CORE2_MEROM:
72 	case INTEL_FAM6_CORE2_MEROM_L:
73 	case INTEL_FAM6_CORE2_PENRYN:
74 	case INTEL_FAM6_CORE2_DUNNINGTON:
75 	case INTEL_FAM6_NEHALEM:
76 	case INTEL_FAM6_NEHALEM_G:
77 	case INTEL_FAM6_NEHALEM_EP:
78 	case INTEL_FAM6_NEHALEM_EX:
79 	case INTEL_FAM6_WESTMERE:
80 	case INTEL_FAM6_WESTMERE_EP:
81 	case INTEL_FAM6_SANDYBRIDGE:
82 		setup_clear_cpu_cap(X86_FEATURE_SELFSNOOP);
83 	}
84 }
85 
86 static bool ring3mwait_disabled __read_mostly;
87 
88 static int __init ring3mwait_disable(char *__unused)
89 {
90 	ring3mwait_disabled = true;
91 	return 0;
92 }
93 __setup("ring3mwait=disable", ring3mwait_disable);
94 
95 static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c)
96 {
97 	/*
98 	 * Ring 3 MONITOR/MWAIT feature cannot be detected without
99 	 * cpu model and family comparison.
100 	 */
101 	if (c->x86 != 6)
102 		return;
103 	switch (c->x86_model) {
104 	case INTEL_FAM6_XEON_PHI_KNL:
105 	case INTEL_FAM6_XEON_PHI_KNM:
106 		break;
107 	default:
108 		return;
109 	}
110 
111 	if (ring3mwait_disabled)
112 		return;
113 
114 	set_cpu_cap(c, X86_FEATURE_RING3MWAIT);
115 	this_cpu_or(msr_misc_features_shadow,
116 		    1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT);
117 
118 	if (c == &boot_cpu_data)
119 		ELF_HWCAP2 |= HWCAP2_RING3MWAIT;
120 }
121 
122 /*
123  * Early microcode releases for the Spectre v2 mitigation were broken.
124  * Information taken from;
125  * - https://newsroom.intel.com/wp-content/uploads/sites/11/2018/03/microcode-update-guidance.pdf
126  * - https://kb.vmware.com/s/article/52345
127  * - Microcode revisions observed in the wild
128  * - Release note from 20180108 microcode release
129  */
130 struct sku_microcode {
131 	u8 model;
132 	u8 stepping;
133 	u32 microcode;
134 };
135 static const struct sku_microcode spectre_bad_microcodes[] = {
136 	{ INTEL_FAM6_KABYLAKE,		0x0B,	0x80 },
137 	{ INTEL_FAM6_KABYLAKE,		0x0A,	0x80 },
138 	{ INTEL_FAM6_KABYLAKE,		0x09,	0x80 },
139 	{ INTEL_FAM6_KABYLAKE_L,	0x0A,	0x80 },
140 	{ INTEL_FAM6_KABYLAKE_L,	0x09,	0x80 },
141 	{ INTEL_FAM6_SKYLAKE_X,		0x03,	0x0100013e },
142 	{ INTEL_FAM6_SKYLAKE_X,		0x04,	0x0200003c },
143 	{ INTEL_FAM6_BROADWELL,		0x04,	0x28 },
144 	{ INTEL_FAM6_BROADWELL_G,	0x01,	0x1b },
145 	{ INTEL_FAM6_BROADWELL_D,	0x02,	0x14 },
146 	{ INTEL_FAM6_BROADWELL_D,	0x03,	0x07000011 },
147 	{ INTEL_FAM6_BROADWELL_X,	0x01,	0x0b000025 },
148 	{ INTEL_FAM6_HASWELL_L,		0x01,	0x21 },
149 	{ INTEL_FAM6_HASWELL_G,		0x01,	0x18 },
150 	{ INTEL_FAM6_HASWELL,		0x03,	0x23 },
151 	{ INTEL_FAM6_HASWELL_X,		0x02,	0x3b },
152 	{ INTEL_FAM6_HASWELL_X,		0x04,	0x10 },
153 	{ INTEL_FAM6_IVYBRIDGE_X,	0x04,	0x42a },
154 	/* Observed in the wild */
155 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x06,	0x61b },
156 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x07,	0x712 },
157 };
158 
159 static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
160 {
161 	int i;
162 
163 	/*
164 	 * We know that the hypervisor lie to us on the microcode version so
165 	 * we may as well hope that it is running the correct version.
166 	 */
167 	if (cpu_has(c, X86_FEATURE_HYPERVISOR))
168 		return false;
169 
170 	if (c->x86 != 6)
171 		return false;
172 
173 	for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
174 		if (c->x86_model == spectre_bad_microcodes[i].model &&
175 		    c->x86_stepping == spectre_bad_microcodes[i].stepping)
176 			return (c->microcode <= spectre_bad_microcodes[i].microcode);
177 	}
178 	return false;
179 }
180 
181 static void early_init_intel(struct cpuinfo_x86 *c)
182 {
183 	u64 misc_enable;
184 
185 	/* Unmask CPUID levels if masked: */
186 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
187 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
188 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
189 			c->cpuid_level = cpuid_eax(0);
190 			get_cpu_cap(c);
191 		}
192 	}
193 
194 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
195 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
196 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
197 
198 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
199 		c->microcode = intel_get_microcode_revision();
200 
201 	/* Now if any of them are set, check the blacklist and clear the lot */
202 	if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
203 	     cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
204 	     cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
205 	     cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
206 		pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
207 		setup_clear_cpu_cap(X86_FEATURE_IBRS);
208 		setup_clear_cpu_cap(X86_FEATURE_IBPB);
209 		setup_clear_cpu_cap(X86_FEATURE_STIBP);
210 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
211 		setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL);
212 		setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
213 		setup_clear_cpu_cap(X86_FEATURE_SSBD);
214 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD);
215 	}
216 
217 	/*
218 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
219 	 *
220 	 * A race condition between speculative fetches and invalidating
221 	 * a large page.  This is worked around in microcode, but we
222 	 * need the microcode to have already been loaded... so if it is
223 	 * not, recommend a BIOS update and disable large pages.
224 	 */
225 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_stepping <= 2 &&
226 	    c->microcode < 0x20e) {
227 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
228 		clear_cpu_cap(c, X86_FEATURE_PSE);
229 	}
230 
231 #ifdef CONFIG_X86_64
232 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
233 #else
234 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
235 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
236 		c->x86_cache_alignment = 128;
237 #endif
238 
239 	/* CPUID workaround for 0F33/0F34 CPU */
240 	if (c->x86 == 0xF && c->x86_model == 0x3
241 	    && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
242 		c->x86_phys_bits = 36;
243 
244 	/*
245 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
246 	 * with P/T states and does not stop in deep C-states.
247 	 *
248 	 * It is also reliable across cores and sockets. (but not across
249 	 * cabinets - we turn it off in that case explicitly.)
250 	 */
251 	if (c->x86_power & (1 << 8)) {
252 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
253 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
254 	}
255 
256 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
257 	if (c->x86 == 6) {
258 		switch (c->x86_model) {
259 		case INTEL_FAM6_ATOM_SALTWELL_MID:
260 		case INTEL_FAM6_ATOM_SALTWELL_TABLET:
261 		case INTEL_FAM6_ATOM_SILVERMONT_MID:
262 		case INTEL_FAM6_ATOM_AIRMONT_NP:
263 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
264 			break;
265 		default:
266 			break;
267 		}
268 	}
269 
270 	/*
271 	 * There is a known erratum on Pentium III and Core Solo
272 	 * and Core Duo CPUs.
273 	 * " Page with PAT set to WC while associated MTRR is UC
274 	 *   may consolidate to UC "
275 	 * Because of this erratum, it is better to stick with
276 	 * setting WC in MTRR rather than using PAT on these CPUs.
277 	 *
278 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
279 	 */
280 	if (c->x86 == 6 && c->x86_model < 15)
281 		clear_cpu_cap(c, X86_FEATURE_PAT);
282 
283 	/*
284 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
285 	 * clear the fast string and enhanced fast string CPU capabilities.
286 	 */
287 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
288 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
289 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
290 			pr_info("Disabled fast string operations\n");
291 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
292 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
293 		}
294 	}
295 
296 	/*
297 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
298 	 * "The operating system also is required to invalidate (i.e., flush)
299 	 *  the TLB when any changes are made to any of the page table entries.
300 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
301 	 *
302 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
303 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
304 	 * to be modified.
305 	 */
306 	if (c->x86 == 5 && c->x86_model == 9) {
307 		pr_info("Disabling PGE capability bit\n");
308 		setup_clear_cpu_cap(X86_FEATURE_PGE);
309 	}
310 
311 	if (c->cpuid_level >= 0x00000001) {
312 		u32 eax, ebx, ecx, edx;
313 
314 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
315 		/*
316 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
317 		 * apicids which are reserved per package. Store the resulting
318 		 * shift value for the package management code.
319 		 */
320 		if (edx & (1U << 28))
321 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
322 	}
323 
324 	check_memory_type_self_snoop_errata(c);
325 
326 	/*
327 	 * Get the number of SMT siblings early from the extended topology
328 	 * leaf, if available. Otherwise try the legacy SMT detection.
329 	 */
330 	if (detect_extended_topology_early(c) < 0)
331 		detect_ht_early(c);
332 }
333 
334 static void bsp_init_intel(struct cpuinfo_x86 *c)
335 {
336 	resctrl_cpu_detect(c);
337 }
338 
339 #ifdef CONFIG_X86_32
340 /*
341  *	Early probe support logic for ppro memory erratum #50
342  *
343  *	This is called before we do cpu ident work
344  */
345 
346 int ppro_with_ram_bug(void)
347 {
348 	/* Uses data from early_cpu_detect now */
349 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
350 	    boot_cpu_data.x86 == 6 &&
351 	    boot_cpu_data.x86_model == 1 &&
352 	    boot_cpu_data.x86_stepping < 8) {
353 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
354 		return 1;
355 	}
356 	return 0;
357 }
358 
359 static void intel_smp_check(struct cpuinfo_x86 *c)
360 {
361 	/* calling is from identify_secondary_cpu() ? */
362 	if (!c->cpu_index)
363 		return;
364 
365 	/*
366 	 * Mask B, Pentium, but not Pentium MMX
367 	 */
368 	if (c->x86 == 5 &&
369 	    c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
370 	    c->x86_model <= 3) {
371 		/*
372 		 * Remember we have B step Pentia with bugs
373 		 */
374 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
375 				    "with B stepping processors.\n");
376 	}
377 }
378 
379 static int forcepae;
380 static int __init forcepae_setup(char *__unused)
381 {
382 	forcepae = 1;
383 	return 1;
384 }
385 __setup("forcepae", forcepae_setup);
386 
387 static void intel_workarounds(struct cpuinfo_x86 *c)
388 {
389 #ifdef CONFIG_X86_F00F_BUG
390 	/*
391 	 * All models of Pentium and Pentium with MMX technology CPUs
392 	 * have the F0 0F bug, which lets nonprivileged users lock up the
393 	 * system. Announce that the fault handler will be checking for it.
394 	 * The Quark is also family 5, but does not have the same bug.
395 	 */
396 	clear_cpu_bug(c, X86_BUG_F00F);
397 	if (c->x86 == 5 && c->x86_model < 9) {
398 		static int f00f_workaround_enabled;
399 
400 		set_cpu_bug(c, X86_BUG_F00F);
401 		if (!f00f_workaround_enabled) {
402 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
403 			f00f_workaround_enabled = 1;
404 		}
405 	}
406 #endif
407 
408 	/*
409 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
410 	 * model 3 mask 3
411 	 */
412 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
413 		clear_cpu_cap(c, X86_FEATURE_SEP);
414 
415 	/*
416 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
417 	 * functionally usable PAE implementation.
418 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
419 	 */
420 	if (forcepae) {
421 		pr_warn("PAE forced!\n");
422 		set_cpu_cap(c, X86_FEATURE_PAE);
423 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
424 	}
425 
426 	/*
427 	 * P4 Xeon erratum 037 workaround.
428 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
429 	 */
430 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
431 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
432 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
433 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
434 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
435 		}
436 	}
437 
438 	/*
439 	 * See if we have a good local APIC by checking for buggy Pentia,
440 	 * i.e. all B steppings and the C2 stepping of P54C when using their
441 	 * integrated APIC (see 11AP erratum in "Pentium Processor
442 	 * Specification Update").
443 	 */
444 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
445 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
446 		set_cpu_bug(c, X86_BUG_11AP);
447 
448 
449 #ifdef CONFIG_X86_INTEL_USERCOPY
450 	/*
451 	 * Set up the preferred alignment for movsl bulk memory moves
452 	 */
453 	switch (c->x86) {
454 	case 4:		/* 486: untested */
455 		break;
456 	case 5:		/* Old Pentia: untested */
457 		break;
458 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
459 		movsl_mask.mask = 7;
460 		break;
461 	case 15:	/* P4 is OK down to 8-byte alignment */
462 		movsl_mask.mask = 7;
463 		break;
464 	}
465 #endif
466 
467 	intel_smp_check(c);
468 }
469 #else
470 static void intel_workarounds(struct cpuinfo_x86 *c)
471 {
472 }
473 #endif
474 
475 static void srat_detect_node(struct cpuinfo_x86 *c)
476 {
477 #ifdef CONFIG_NUMA
478 	unsigned node;
479 	int cpu = smp_processor_id();
480 
481 	/* Don't do the funky fallback heuristics the AMD version employs
482 	   for now. */
483 	node = numa_cpu_node(cpu);
484 	if (node == NUMA_NO_NODE || !node_online(node)) {
485 		/* reuse the value from init_cpu_to_node() */
486 		node = cpu_to_node(cpu);
487 	}
488 	numa_set_node(cpu, node);
489 #endif
490 }
491 
492 #define MSR_IA32_TME_ACTIVATE		0x982
493 
494 /* Helpers to access TME_ACTIVATE MSR */
495 #define TME_ACTIVATE_LOCKED(x)		(x & 0x1)
496 #define TME_ACTIVATE_ENABLED(x)		(x & 0x2)
497 
498 #define TME_ACTIVATE_POLICY(x)		((x >> 4) & 0xf)	/* Bits 7:4 */
499 #define TME_ACTIVATE_POLICY_AES_XTS_128	0
500 
501 #define TME_ACTIVATE_KEYID_BITS(x)	((x >> 32) & 0xf)	/* Bits 35:32 */
502 
503 #define TME_ACTIVATE_CRYPTO_ALGS(x)	((x >> 48) & 0xffff)	/* Bits 63:48 */
504 #define TME_ACTIVATE_CRYPTO_AES_XTS_128	1
505 
506 /* Values for mktme_status (SW only construct) */
507 #define MKTME_ENABLED			0
508 #define MKTME_DISABLED			1
509 #define MKTME_UNINITIALIZED		2
510 static int mktme_status = MKTME_UNINITIALIZED;
511 
512 static void detect_tme(struct cpuinfo_x86 *c)
513 {
514 	u64 tme_activate, tme_policy, tme_crypto_algs;
515 	int keyid_bits = 0, nr_keyids = 0;
516 	static u64 tme_activate_cpu0 = 0;
517 
518 	rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
519 
520 	if (mktme_status != MKTME_UNINITIALIZED) {
521 		if (tme_activate != tme_activate_cpu0) {
522 			/* Broken BIOS? */
523 			pr_err_once("x86/tme: configuration is inconsistent between CPUs\n");
524 			pr_err_once("x86/tme: MKTME is not usable\n");
525 			mktme_status = MKTME_DISABLED;
526 
527 			/* Proceed. We may need to exclude bits from x86_phys_bits. */
528 		}
529 	} else {
530 		tme_activate_cpu0 = tme_activate;
531 	}
532 
533 	if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
534 		pr_info_once("x86/tme: not enabled by BIOS\n");
535 		mktme_status = MKTME_DISABLED;
536 		return;
537 	}
538 
539 	if (mktme_status != MKTME_UNINITIALIZED)
540 		goto detect_keyid_bits;
541 
542 	pr_info("x86/tme: enabled by BIOS\n");
543 
544 	tme_policy = TME_ACTIVATE_POLICY(tme_activate);
545 	if (tme_policy != TME_ACTIVATE_POLICY_AES_XTS_128)
546 		pr_warn("x86/tme: Unknown policy is active: %#llx\n", tme_policy);
547 
548 	tme_crypto_algs = TME_ACTIVATE_CRYPTO_ALGS(tme_activate);
549 	if (!(tme_crypto_algs & TME_ACTIVATE_CRYPTO_AES_XTS_128)) {
550 		pr_err("x86/mktme: No known encryption algorithm is supported: %#llx\n",
551 				tme_crypto_algs);
552 		mktme_status = MKTME_DISABLED;
553 	}
554 detect_keyid_bits:
555 	keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
556 	nr_keyids = (1UL << keyid_bits) - 1;
557 	if (nr_keyids) {
558 		pr_info_once("x86/mktme: enabled by BIOS\n");
559 		pr_info_once("x86/mktme: %d KeyIDs available\n", nr_keyids);
560 	} else {
561 		pr_info_once("x86/mktme: disabled by BIOS\n");
562 	}
563 
564 	if (mktme_status == MKTME_UNINITIALIZED) {
565 		/* MKTME is usable */
566 		mktme_status = MKTME_ENABLED;
567 	}
568 
569 	/*
570 	 * KeyID bits effectively lower the number of physical address
571 	 * bits.  Update cpuinfo_x86::x86_phys_bits accordingly.
572 	 */
573 	c->x86_phys_bits -= keyid_bits;
574 }
575 
576 static void init_cpuid_fault(struct cpuinfo_x86 *c)
577 {
578 	u64 msr;
579 
580 	if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
581 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
582 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
583 	}
584 }
585 
586 static void init_intel_misc_features(struct cpuinfo_x86 *c)
587 {
588 	u64 msr;
589 
590 	if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
591 		return;
592 
593 	/* Clear all MISC features */
594 	this_cpu_write(msr_misc_features_shadow, 0);
595 
596 	/* Check features and update capabilities and shadow control bits */
597 	init_cpuid_fault(c);
598 	probe_xeon_phi_r3mwait(c);
599 
600 	msr = this_cpu_read(msr_misc_features_shadow);
601 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
602 }
603 
604 static void split_lock_init(void);
605 
606 static void init_intel(struct cpuinfo_x86 *c)
607 {
608 	early_init_intel(c);
609 
610 	intel_workarounds(c);
611 
612 	/*
613 	 * Detect the extended topology information if available. This
614 	 * will reinitialise the initial_apicid which will be used
615 	 * in init_intel_cacheinfo()
616 	 */
617 	detect_extended_topology(c);
618 
619 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
620 		/*
621 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
622 		 * detection.
623 		 */
624 		detect_num_cpu_cores(c);
625 #ifdef CONFIG_X86_32
626 		detect_ht(c);
627 #endif
628 	}
629 
630 	init_intel_cacheinfo(c);
631 
632 	if (c->cpuid_level > 9) {
633 		unsigned eax = cpuid_eax(10);
634 		/* Check for version and the number of counters */
635 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
636 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
637 	}
638 
639 	if (cpu_has(c, X86_FEATURE_XMM2))
640 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
641 
642 	if (boot_cpu_has(X86_FEATURE_DS)) {
643 		unsigned int l1, l2;
644 
645 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
646 		if (!(l1 & (1<<11)))
647 			set_cpu_cap(c, X86_FEATURE_BTS);
648 		if (!(l1 & (1<<12)))
649 			set_cpu_cap(c, X86_FEATURE_PEBS);
650 	}
651 
652 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
653 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
654 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
655 
656 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
657 		((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
658 		set_cpu_bug(c, X86_BUG_MONITOR);
659 
660 #ifdef CONFIG_X86_64
661 	if (c->x86 == 15)
662 		c->x86_cache_alignment = c->x86_clflush_size * 2;
663 	if (c->x86 == 6)
664 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
665 #else
666 	/*
667 	 * Names for the Pentium II/Celeron processors
668 	 * detectable only by also checking the cache size.
669 	 * Dixon is NOT a Celeron.
670 	 */
671 	if (c->x86 == 6) {
672 		unsigned int l2 = c->x86_cache_size;
673 		char *p = NULL;
674 
675 		switch (c->x86_model) {
676 		case 5:
677 			if (l2 == 0)
678 				p = "Celeron (Covington)";
679 			else if (l2 == 256)
680 				p = "Mobile Pentium II (Dixon)";
681 			break;
682 
683 		case 6:
684 			if (l2 == 128)
685 				p = "Celeron (Mendocino)";
686 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
687 				p = "Celeron-A";
688 			break;
689 
690 		case 8:
691 			if (l2 == 128)
692 				p = "Celeron (Coppermine)";
693 			break;
694 		}
695 
696 		if (p)
697 			strcpy(c->x86_model_id, p);
698 	}
699 
700 	if (c->x86 == 15)
701 		set_cpu_cap(c, X86_FEATURE_P4);
702 	if (c->x86 == 6)
703 		set_cpu_cap(c, X86_FEATURE_P3);
704 #endif
705 
706 	/* Work around errata */
707 	srat_detect_node(c);
708 
709 	init_ia32_feat_ctl(c);
710 
711 	if (cpu_has(c, X86_FEATURE_TME))
712 		detect_tme(c);
713 
714 	init_intel_misc_features(c);
715 
716 	if (tsx_ctrl_state == TSX_CTRL_ENABLE)
717 		tsx_enable();
718 	if (tsx_ctrl_state == TSX_CTRL_DISABLE)
719 		tsx_disable();
720 
721 	split_lock_init();
722 }
723 
724 #ifdef CONFIG_X86_32
725 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
726 {
727 	/*
728 	 * Intel PIII Tualatin. This comes in two flavours.
729 	 * One has 256kb of cache, the other 512. We have no way
730 	 * to determine which, so we use a boottime override
731 	 * for the 512kb model, and assume 256 otherwise.
732 	 */
733 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
734 		size = 256;
735 
736 	/*
737 	 * Intel Quark SoC X1000 contains a 4-way set associative
738 	 * 16K cache with a 16 byte cache line and 256 lines per tag
739 	 */
740 	if ((c->x86 == 5) && (c->x86_model == 9))
741 		size = 16;
742 	return size;
743 }
744 #endif
745 
746 #define TLB_INST_4K	0x01
747 #define TLB_INST_4M	0x02
748 #define TLB_INST_2M_4M	0x03
749 
750 #define TLB_INST_ALL	0x05
751 #define TLB_INST_1G	0x06
752 
753 #define TLB_DATA_4K	0x11
754 #define TLB_DATA_4M	0x12
755 #define TLB_DATA_2M_4M	0x13
756 #define TLB_DATA_4K_4M	0x14
757 
758 #define TLB_DATA_1G	0x16
759 
760 #define TLB_DATA0_4K	0x21
761 #define TLB_DATA0_4M	0x22
762 #define TLB_DATA0_2M_4M	0x23
763 
764 #define STLB_4K		0x41
765 #define STLB_4K_2M	0x42
766 
767 static const struct _tlb_table intel_tlb_table[] = {
768 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
769 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
770 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
771 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
772 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
773 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
774 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages" },
775 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
776 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
777 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
778 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
779 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
780 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
781 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
782 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
783 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
784 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
785 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
786 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
787 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
788 	{ 0x6b, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 8-way associative" },
789 	{ 0x6c, TLB_DATA_2M_4M,		128,	" TLB_DATA 2 MByte or 4 MByte pages, 8-way associative" },
790 	{ 0x6d, TLB_DATA_1G,		16,	" TLB_DATA 1 GByte pages, fully associative" },
791 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
792 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
793 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
794 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
795 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
796 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
797 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
798 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
799 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
800 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
801 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
802 	{ 0xc2, TLB_DATA_2M_4M,		16,	" TLB_DATA 2 MByte/4MByte pages, 4-way associative" },
803 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
804 	{ 0x00, 0, 0 }
805 };
806 
807 static void intel_tlb_lookup(const unsigned char desc)
808 {
809 	unsigned char k;
810 	if (desc == 0)
811 		return;
812 
813 	/* look up this descriptor in the table */
814 	for (k = 0; intel_tlb_table[k].descriptor != desc &&
815 	     intel_tlb_table[k].descriptor != 0; k++)
816 		;
817 
818 	if (intel_tlb_table[k].tlb_type == 0)
819 		return;
820 
821 	switch (intel_tlb_table[k].tlb_type) {
822 	case STLB_4K:
823 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
824 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
825 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
826 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
827 		break;
828 	case STLB_4K_2M:
829 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
830 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
831 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
832 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
833 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
834 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
835 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
836 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
837 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
838 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
839 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
840 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
841 		break;
842 	case TLB_INST_ALL:
843 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
844 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
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_INST_4K:
851 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
852 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
853 		break;
854 	case TLB_INST_4M:
855 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
856 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
857 		break;
858 	case TLB_INST_2M_4M:
859 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
860 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
861 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
862 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
863 		break;
864 	case TLB_DATA_4K:
865 	case TLB_DATA0_4K:
866 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
867 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
868 		break;
869 	case TLB_DATA_4M:
870 	case TLB_DATA0_4M:
871 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
872 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
873 		break;
874 	case TLB_DATA_2M_4M:
875 	case TLB_DATA0_2M_4M:
876 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
877 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
878 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
879 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
880 		break;
881 	case TLB_DATA_4K_4M:
882 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
883 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
884 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
885 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
886 		break;
887 	case TLB_DATA_1G:
888 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
889 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
890 		break;
891 	}
892 }
893 
894 static void intel_detect_tlb(struct cpuinfo_x86 *c)
895 {
896 	int i, j, n;
897 	unsigned int regs[4];
898 	unsigned char *desc = (unsigned char *)regs;
899 
900 	if (c->cpuid_level < 2)
901 		return;
902 
903 	/* Number of times to iterate */
904 	n = cpuid_eax(2) & 0xFF;
905 
906 	for (i = 0 ; i < n ; i++) {
907 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
908 
909 		/* If bit 31 is set, this is an unknown format */
910 		for (j = 0 ; j < 3 ; j++)
911 			if (regs[j] & (1 << 31))
912 				regs[j] = 0;
913 
914 		/* Byte 0 is level count, not a descriptor */
915 		for (j = 1 ; j < 16 ; j++)
916 			intel_tlb_lookup(desc[j]);
917 	}
918 }
919 
920 static const struct cpu_dev intel_cpu_dev = {
921 	.c_vendor	= "Intel",
922 	.c_ident	= { "GenuineIntel" },
923 #ifdef CONFIG_X86_32
924 	.legacy_models = {
925 		{ .family = 4, .model_names =
926 		  {
927 			  [0] = "486 DX-25/33",
928 			  [1] = "486 DX-50",
929 			  [2] = "486 SX",
930 			  [3] = "486 DX/2",
931 			  [4] = "486 SL",
932 			  [5] = "486 SX/2",
933 			  [7] = "486 DX/2-WB",
934 			  [8] = "486 DX/4",
935 			  [9] = "486 DX/4-WB"
936 		  }
937 		},
938 		{ .family = 5, .model_names =
939 		  {
940 			  [0] = "Pentium 60/66 A-step",
941 			  [1] = "Pentium 60/66",
942 			  [2] = "Pentium 75 - 200",
943 			  [3] = "OverDrive PODP5V83",
944 			  [4] = "Pentium MMX",
945 			  [7] = "Mobile Pentium 75 - 200",
946 			  [8] = "Mobile Pentium MMX",
947 			  [9] = "Quark SoC X1000",
948 		  }
949 		},
950 		{ .family = 6, .model_names =
951 		  {
952 			  [0] = "Pentium Pro A-step",
953 			  [1] = "Pentium Pro",
954 			  [3] = "Pentium II (Klamath)",
955 			  [4] = "Pentium II (Deschutes)",
956 			  [5] = "Pentium II (Deschutes)",
957 			  [6] = "Mobile Pentium II",
958 			  [7] = "Pentium III (Katmai)",
959 			  [8] = "Pentium III (Coppermine)",
960 			  [10] = "Pentium III (Cascades)",
961 			  [11] = "Pentium III (Tualatin)",
962 		  }
963 		},
964 		{ .family = 15, .model_names =
965 		  {
966 			  [0] = "Pentium 4 (Unknown)",
967 			  [1] = "Pentium 4 (Willamette)",
968 			  [2] = "Pentium 4 (Northwood)",
969 			  [4] = "Pentium 4 (Foster)",
970 			  [5] = "Pentium 4 (Foster)",
971 		  }
972 		},
973 	},
974 	.legacy_cache_size = intel_size_cache,
975 #endif
976 	.c_detect_tlb	= intel_detect_tlb,
977 	.c_early_init   = early_init_intel,
978 	.c_bsp_init	= bsp_init_intel,
979 	.c_init		= init_intel,
980 	.c_x86_vendor	= X86_VENDOR_INTEL,
981 };
982 
983 cpu_dev_register(intel_cpu_dev);
984 
985 #undef pr_fmt
986 #define pr_fmt(fmt) "x86/split lock detection: " fmt
987 
988 static const struct {
989 	const char			*option;
990 	enum split_lock_detect_state	state;
991 } sld_options[] __initconst = {
992 	{ "off",	sld_off   },
993 	{ "warn",	sld_warn  },
994 	{ "fatal",	sld_fatal },
995 };
996 
997 static inline bool match_option(const char *arg, int arglen, const char *opt)
998 {
999 	int len = strlen(opt);
1000 
1001 	return len == arglen && !strncmp(arg, opt, len);
1002 }
1003 
1004 static bool split_lock_verify_msr(bool on)
1005 {
1006 	u64 ctrl, tmp;
1007 
1008 	if (rdmsrl_safe(MSR_TEST_CTRL, &ctrl))
1009 		return false;
1010 	if (on)
1011 		ctrl |= MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
1012 	else
1013 		ctrl &= ~MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
1014 	if (wrmsrl_safe(MSR_TEST_CTRL, ctrl))
1015 		return false;
1016 	rdmsrl(MSR_TEST_CTRL, tmp);
1017 	return ctrl == tmp;
1018 }
1019 
1020 static void __init split_lock_setup(void)
1021 {
1022 	enum split_lock_detect_state state = sld_warn;
1023 	char arg[20];
1024 	int i, ret;
1025 
1026 	if (!split_lock_verify_msr(false)) {
1027 		pr_info("MSR access failed: Disabled\n");
1028 		return;
1029 	}
1030 
1031 	ret = cmdline_find_option(boot_command_line, "split_lock_detect",
1032 				  arg, sizeof(arg));
1033 	if (ret >= 0) {
1034 		for (i = 0; i < ARRAY_SIZE(sld_options); i++) {
1035 			if (match_option(arg, ret, sld_options[i].option)) {
1036 				state = sld_options[i].state;
1037 				break;
1038 			}
1039 		}
1040 	}
1041 
1042 	switch (state) {
1043 	case sld_off:
1044 		pr_info("disabled\n");
1045 		return;
1046 	case sld_warn:
1047 		pr_info("warning about user-space split_locks\n");
1048 		break;
1049 	case sld_fatal:
1050 		pr_info("sending SIGBUS on user-space split_locks\n");
1051 		break;
1052 	}
1053 
1054 	rdmsrl(MSR_TEST_CTRL, msr_test_ctrl_cache);
1055 
1056 	if (!split_lock_verify_msr(true)) {
1057 		pr_info("MSR access failed: Disabled\n");
1058 		return;
1059 	}
1060 
1061 	sld_state = state;
1062 	setup_force_cpu_cap(X86_FEATURE_SPLIT_LOCK_DETECT);
1063 }
1064 
1065 /*
1066  * MSR_TEST_CTRL is per core, but we treat it like a per CPU MSR. Locking
1067  * is not implemented as one thread could undo the setting of the other
1068  * thread immediately after dropping the lock anyway.
1069  */
1070 static void sld_update_msr(bool on)
1071 {
1072 	u64 test_ctrl_val = msr_test_ctrl_cache;
1073 
1074 	if (on)
1075 		test_ctrl_val |= MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
1076 
1077 	wrmsrl(MSR_TEST_CTRL, test_ctrl_val);
1078 }
1079 
1080 static void split_lock_init(void)
1081 {
1082 	if (cpu_model_supports_sld)
1083 		split_lock_verify_msr(sld_state != sld_off);
1084 }
1085 
1086 static void split_lock_warn(unsigned long ip)
1087 {
1088 	pr_warn_ratelimited("#AC: %s/%d took a split_lock trap at address: 0x%lx\n",
1089 			    current->comm, current->pid, ip);
1090 
1091 	/*
1092 	 * Disable the split lock detection for this task so it can make
1093 	 * progress and set TIF_SLD so the detection is re-enabled via
1094 	 * switch_to_sld() when the task is scheduled out.
1095 	 */
1096 	sld_update_msr(false);
1097 	set_tsk_thread_flag(current, TIF_SLD);
1098 }
1099 
1100 bool handle_guest_split_lock(unsigned long ip)
1101 {
1102 	if (sld_state == sld_warn) {
1103 		split_lock_warn(ip);
1104 		return true;
1105 	}
1106 
1107 	pr_warn_once("#AC: %s/%d %s split_lock trap at address: 0x%lx\n",
1108 		     current->comm, current->pid,
1109 		     sld_state == sld_fatal ? "fatal" : "bogus", ip);
1110 
1111 	current->thread.error_code = 0;
1112 	current->thread.trap_nr = X86_TRAP_AC;
1113 	force_sig_fault(SIGBUS, BUS_ADRALN, NULL);
1114 	return false;
1115 }
1116 EXPORT_SYMBOL_GPL(handle_guest_split_lock);
1117 
1118 bool handle_user_split_lock(struct pt_regs *regs, long error_code)
1119 {
1120 	if ((regs->flags & X86_EFLAGS_AC) || sld_state == sld_fatal)
1121 		return false;
1122 	split_lock_warn(regs->ip);
1123 	return true;
1124 }
1125 
1126 /*
1127  * This function is called only when switching between tasks with
1128  * different split-lock detection modes. It sets the MSR for the
1129  * mode of the new task. This is right most of the time, but since
1130  * the MSR is shared by hyperthreads on a physical core there can
1131  * be glitches when the two threads need different modes.
1132  */
1133 void switch_to_sld(unsigned long tifn)
1134 {
1135 	sld_update_msr(!(tifn & _TIF_SLD));
1136 }
1137 
1138 /*
1139  * Bits in the IA32_CORE_CAPABILITIES are not architectural, so they should
1140  * only be trusted if it is confirmed that a CPU model implements a
1141  * specific feature at a particular bit position.
1142  *
1143  * The possible driver data field values:
1144  *
1145  * - 0: CPU models that are known to have the per-core split-lock detection
1146  *	feature even though they do not enumerate IA32_CORE_CAPABILITIES.
1147  *
1148  * - 1: CPU models which may enumerate IA32_CORE_CAPABILITIES and if so use
1149  *      bit 5 to enumerate the per-core split-lock detection feature.
1150  */
1151 static const struct x86_cpu_id split_lock_cpu_ids[] __initconst = {
1152 	X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_X,		0),
1153 	X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_L,		0),
1154 	X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_D,		0),
1155 	X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT,	1),
1156 	X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT_D,	1),
1157 	X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT_L,	1),
1158 	X86_MATCH_INTEL_FAM6_MODEL(TIGERLAKE_L,		1),
1159 	X86_MATCH_INTEL_FAM6_MODEL(TIGERLAKE,		1),
1160 	X86_MATCH_INTEL_FAM6_MODEL(SAPPHIRERAPIDS_X,	1),
1161 	X86_MATCH_INTEL_FAM6_MODEL(ALDERLAKE,		1),
1162 	{}
1163 };
1164 
1165 void __init cpu_set_core_cap_bits(struct cpuinfo_x86 *c)
1166 {
1167 	const struct x86_cpu_id *m;
1168 	u64 ia32_core_caps;
1169 
1170 	if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
1171 		return;
1172 
1173 	m = x86_match_cpu(split_lock_cpu_ids);
1174 	if (!m)
1175 		return;
1176 
1177 	switch (m->driver_data) {
1178 	case 0:
1179 		break;
1180 	case 1:
1181 		if (!cpu_has(c, X86_FEATURE_CORE_CAPABILITIES))
1182 			return;
1183 		rdmsrl(MSR_IA32_CORE_CAPS, ia32_core_caps);
1184 		if (!(ia32_core_caps & MSR_IA32_CORE_CAPS_SPLIT_LOCK_DETECT))
1185 			return;
1186 		break;
1187 	default:
1188 		return;
1189 	}
1190 
1191 	cpu_model_supports_sld = true;
1192 	split_lock_setup();
1193 }
1194