xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision 0e17c50f)
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 
23 #ifdef CONFIG_X86_64
24 #include <linux/topology.h>
25 #endif
26 
27 #include "cpu.h"
28 
29 #ifdef CONFIG_X86_LOCAL_APIC
30 #include <asm/mpspec.h>
31 #include <asm/apic.h>
32 #endif
33 
34 /*
35  * Processors which have self-snooping capability can handle conflicting
36  * memory type across CPUs by snooping its own cache. However, there exists
37  * CPU models in which having conflicting memory types still leads to
38  * unpredictable behavior, machine check errors, or hangs. Clear this
39  * feature to prevent its use on machines with known erratas.
40  */
41 static void check_memory_type_self_snoop_errata(struct cpuinfo_x86 *c)
42 {
43 	switch (c->x86_model) {
44 	case INTEL_FAM6_CORE_YONAH:
45 	case INTEL_FAM6_CORE2_MEROM:
46 	case INTEL_FAM6_CORE2_MEROM_L:
47 	case INTEL_FAM6_CORE2_PENRYN:
48 	case INTEL_FAM6_CORE2_DUNNINGTON:
49 	case INTEL_FAM6_NEHALEM:
50 	case INTEL_FAM6_NEHALEM_G:
51 	case INTEL_FAM6_NEHALEM_EP:
52 	case INTEL_FAM6_NEHALEM_EX:
53 	case INTEL_FAM6_WESTMERE:
54 	case INTEL_FAM6_WESTMERE_EP:
55 	case INTEL_FAM6_SANDYBRIDGE:
56 		setup_clear_cpu_cap(X86_FEATURE_SELFSNOOP);
57 	}
58 }
59 
60 static bool ring3mwait_disabled __read_mostly;
61 
62 static int __init ring3mwait_disable(char *__unused)
63 {
64 	ring3mwait_disabled = true;
65 	return 0;
66 }
67 __setup("ring3mwait=disable", ring3mwait_disable);
68 
69 static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c)
70 {
71 	/*
72 	 * Ring 3 MONITOR/MWAIT feature cannot be detected without
73 	 * cpu model and family comparison.
74 	 */
75 	if (c->x86 != 6)
76 		return;
77 	switch (c->x86_model) {
78 	case INTEL_FAM6_XEON_PHI_KNL:
79 	case INTEL_FAM6_XEON_PHI_KNM:
80 		break;
81 	default:
82 		return;
83 	}
84 
85 	if (ring3mwait_disabled)
86 		return;
87 
88 	set_cpu_cap(c, X86_FEATURE_RING3MWAIT);
89 	this_cpu_or(msr_misc_features_shadow,
90 		    1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT);
91 
92 	if (c == &boot_cpu_data)
93 		ELF_HWCAP2 |= HWCAP2_RING3MWAIT;
94 }
95 
96 /*
97  * Early microcode releases for the Spectre v2 mitigation were broken.
98  * Information taken from;
99  * - https://newsroom.intel.com/wp-content/uploads/sites/11/2018/03/microcode-update-guidance.pdf
100  * - https://kb.vmware.com/s/article/52345
101  * - Microcode revisions observed in the wild
102  * - Release note from 20180108 microcode release
103  */
104 struct sku_microcode {
105 	u8 model;
106 	u8 stepping;
107 	u32 microcode;
108 };
109 static const struct sku_microcode spectre_bad_microcodes[] = {
110 	{ INTEL_FAM6_KABYLAKE,		0x0B,	0x80 },
111 	{ INTEL_FAM6_KABYLAKE,		0x0A,	0x80 },
112 	{ INTEL_FAM6_KABYLAKE,		0x09,	0x80 },
113 	{ INTEL_FAM6_KABYLAKE_L,	0x0A,	0x80 },
114 	{ INTEL_FAM6_KABYLAKE_L,	0x09,	0x80 },
115 	{ INTEL_FAM6_SKYLAKE_X,		0x03,	0x0100013e },
116 	{ INTEL_FAM6_SKYLAKE_X,		0x04,	0x0200003c },
117 	{ INTEL_FAM6_BROADWELL,		0x04,	0x28 },
118 	{ INTEL_FAM6_BROADWELL_G,	0x01,	0x1b },
119 	{ INTEL_FAM6_BROADWELL_D,	0x02,	0x14 },
120 	{ INTEL_FAM6_BROADWELL_D,	0x03,	0x07000011 },
121 	{ INTEL_FAM6_BROADWELL_X,	0x01,	0x0b000025 },
122 	{ INTEL_FAM6_HASWELL_L,		0x01,	0x21 },
123 	{ INTEL_FAM6_HASWELL_G,		0x01,	0x18 },
124 	{ INTEL_FAM6_HASWELL,		0x03,	0x23 },
125 	{ INTEL_FAM6_HASWELL_X,		0x02,	0x3b },
126 	{ INTEL_FAM6_HASWELL_X,		0x04,	0x10 },
127 	{ INTEL_FAM6_IVYBRIDGE_X,	0x04,	0x42a },
128 	/* Observed in the wild */
129 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x06,	0x61b },
130 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x07,	0x712 },
131 };
132 
133 static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
134 {
135 	int i;
136 
137 	/*
138 	 * We know that the hypervisor lie to us on the microcode version so
139 	 * we may as well hope that it is running the correct version.
140 	 */
141 	if (cpu_has(c, X86_FEATURE_HYPERVISOR))
142 		return false;
143 
144 	if (c->x86 != 6)
145 		return false;
146 
147 	for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
148 		if (c->x86_model == spectre_bad_microcodes[i].model &&
149 		    c->x86_stepping == spectre_bad_microcodes[i].stepping)
150 			return (c->microcode <= spectre_bad_microcodes[i].microcode);
151 	}
152 	return false;
153 }
154 
155 static void early_init_intel(struct cpuinfo_x86 *c)
156 {
157 	u64 misc_enable;
158 
159 	/* Unmask CPUID levels if masked: */
160 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
161 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
162 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
163 			c->cpuid_level = cpuid_eax(0);
164 			get_cpu_cap(c);
165 		}
166 	}
167 
168 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
169 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
170 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
171 
172 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
173 		c->microcode = intel_get_microcode_revision();
174 
175 	/* Now if any of them are set, check the blacklist and clear the lot */
176 	if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
177 	     cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
178 	     cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
179 	     cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
180 		pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
181 		setup_clear_cpu_cap(X86_FEATURE_IBRS);
182 		setup_clear_cpu_cap(X86_FEATURE_IBPB);
183 		setup_clear_cpu_cap(X86_FEATURE_STIBP);
184 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
185 		setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL);
186 		setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
187 		setup_clear_cpu_cap(X86_FEATURE_SSBD);
188 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD);
189 	}
190 
191 	/*
192 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
193 	 *
194 	 * A race condition between speculative fetches and invalidating
195 	 * a large page.  This is worked around in microcode, but we
196 	 * need the microcode to have already been loaded... so if it is
197 	 * not, recommend a BIOS update and disable large pages.
198 	 */
199 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_stepping <= 2 &&
200 	    c->microcode < 0x20e) {
201 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
202 		clear_cpu_cap(c, X86_FEATURE_PSE);
203 	}
204 
205 #ifdef CONFIG_X86_64
206 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
207 #else
208 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
209 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
210 		c->x86_cache_alignment = 128;
211 #endif
212 
213 	/* CPUID workaround for 0F33/0F34 CPU */
214 	if (c->x86 == 0xF && c->x86_model == 0x3
215 	    && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
216 		c->x86_phys_bits = 36;
217 
218 	/*
219 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
220 	 * with P/T states and does not stop in deep C-states.
221 	 *
222 	 * It is also reliable across cores and sockets. (but not across
223 	 * cabinets - we turn it off in that case explicitly.)
224 	 */
225 	if (c->x86_power & (1 << 8)) {
226 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
227 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
228 	}
229 
230 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
231 	if (c->x86 == 6) {
232 		switch (c->x86_model) {
233 		case INTEL_FAM6_ATOM_SALTWELL_MID:
234 		case INTEL_FAM6_ATOM_SALTWELL_TABLET:
235 		case INTEL_FAM6_ATOM_SILVERMONT_MID:
236 		case INTEL_FAM6_ATOM_AIRMONT_NP:
237 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
238 			break;
239 		default:
240 			break;
241 		}
242 	}
243 
244 	/*
245 	 * There is a known erratum on Pentium III and Core Solo
246 	 * and Core Duo CPUs.
247 	 * " Page with PAT set to WC while associated MTRR is UC
248 	 *   may consolidate to UC "
249 	 * Because of this erratum, it is better to stick with
250 	 * setting WC in MTRR rather than using PAT on these CPUs.
251 	 *
252 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
253 	 */
254 	if (c->x86 == 6 && c->x86_model < 15)
255 		clear_cpu_cap(c, X86_FEATURE_PAT);
256 
257 	/*
258 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
259 	 * clear the fast string and enhanced fast string CPU capabilities.
260 	 */
261 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
262 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
263 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
264 			pr_info("Disabled fast string operations\n");
265 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
266 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
267 		}
268 	}
269 
270 	/*
271 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
272 	 * "The operating system also is required to invalidate (i.e., flush)
273 	 *  the TLB when any changes are made to any of the page table entries.
274 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
275 	 *
276 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
277 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
278 	 * to be modified.
279 	 */
280 	if (c->x86 == 5 && c->x86_model == 9) {
281 		pr_info("Disabling PGE capability bit\n");
282 		setup_clear_cpu_cap(X86_FEATURE_PGE);
283 	}
284 
285 	if (c->cpuid_level >= 0x00000001) {
286 		u32 eax, ebx, ecx, edx;
287 
288 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
289 		/*
290 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
291 		 * apicids which are reserved per package. Store the resulting
292 		 * shift value for the package management code.
293 		 */
294 		if (edx & (1U << 28))
295 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
296 	}
297 
298 	check_memory_type_self_snoop_errata(c);
299 
300 	/*
301 	 * Get the number of SMT siblings early from the extended topology
302 	 * leaf, if available. Otherwise try the legacy SMT detection.
303 	 */
304 	if (detect_extended_topology_early(c) < 0)
305 		detect_ht_early(c);
306 }
307 
308 #ifdef CONFIG_X86_32
309 /*
310  *	Early probe support logic for ppro memory erratum #50
311  *
312  *	This is called before we do cpu ident work
313  */
314 
315 int ppro_with_ram_bug(void)
316 {
317 	/* Uses data from early_cpu_detect now */
318 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
319 	    boot_cpu_data.x86 == 6 &&
320 	    boot_cpu_data.x86_model == 1 &&
321 	    boot_cpu_data.x86_stepping < 8) {
322 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
323 		return 1;
324 	}
325 	return 0;
326 }
327 
328 static void intel_smp_check(struct cpuinfo_x86 *c)
329 {
330 	/* calling is from identify_secondary_cpu() ? */
331 	if (!c->cpu_index)
332 		return;
333 
334 	/*
335 	 * Mask B, Pentium, but not Pentium MMX
336 	 */
337 	if (c->x86 == 5 &&
338 	    c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
339 	    c->x86_model <= 3) {
340 		/*
341 		 * Remember we have B step Pentia with bugs
342 		 */
343 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
344 				    "with B stepping processors.\n");
345 	}
346 }
347 
348 static int forcepae;
349 static int __init forcepae_setup(char *__unused)
350 {
351 	forcepae = 1;
352 	return 1;
353 }
354 __setup("forcepae", forcepae_setup);
355 
356 static void intel_workarounds(struct cpuinfo_x86 *c)
357 {
358 #ifdef CONFIG_X86_F00F_BUG
359 	/*
360 	 * All models of Pentium and Pentium with MMX technology CPUs
361 	 * have the F0 0F bug, which lets nonprivileged users lock up the
362 	 * system. Announce that the fault handler will be checking for it.
363 	 * The Quark is also family 5, but does not have the same bug.
364 	 */
365 	clear_cpu_bug(c, X86_BUG_F00F);
366 	if (c->x86 == 5 && c->x86_model < 9) {
367 		static int f00f_workaround_enabled;
368 
369 		set_cpu_bug(c, X86_BUG_F00F);
370 		if (!f00f_workaround_enabled) {
371 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
372 			f00f_workaround_enabled = 1;
373 		}
374 	}
375 #endif
376 
377 	/*
378 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
379 	 * model 3 mask 3
380 	 */
381 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
382 		clear_cpu_cap(c, X86_FEATURE_SEP);
383 
384 	/*
385 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
386 	 * functionally usable PAE implementation.
387 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
388 	 */
389 	if (forcepae) {
390 		pr_warn("PAE forced!\n");
391 		set_cpu_cap(c, X86_FEATURE_PAE);
392 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
393 	}
394 
395 	/*
396 	 * P4 Xeon erratum 037 workaround.
397 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
398 	 */
399 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
400 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
401 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
402 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
403 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
404 		}
405 	}
406 
407 	/*
408 	 * See if we have a good local APIC by checking for buggy Pentia,
409 	 * i.e. all B steppings and the C2 stepping of P54C when using their
410 	 * integrated APIC (see 11AP erratum in "Pentium Processor
411 	 * Specification Update").
412 	 */
413 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
414 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
415 		set_cpu_bug(c, X86_BUG_11AP);
416 
417 
418 #ifdef CONFIG_X86_INTEL_USERCOPY
419 	/*
420 	 * Set up the preferred alignment for movsl bulk memory moves
421 	 */
422 	switch (c->x86) {
423 	case 4:		/* 486: untested */
424 		break;
425 	case 5:		/* Old Pentia: untested */
426 		break;
427 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
428 		movsl_mask.mask = 7;
429 		break;
430 	case 15:	/* P4 is OK down to 8-byte alignment */
431 		movsl_mask.mask = 7;
432 		break;
433 	}
434 #endif
435 
436 	intel_smp_check(c);
437 }
438 #else
439 static void intel_workarounds(struct cpuinfo_x86 *c)
440 {
441 }
442 #endif
443 
444 static void srat_detect_node(struct cpuinfo_x86 *c)
445 {
446 #ifdef CONFIG_NUMA
447 	unsigned node;
448 	int cpu = smp_processor_id();
449 
450 	/* Don't do the funky fallback heuristics the AMD version employs
451 	   for now. */
452 	node = numa_cpu_node(cpu);
453 	if (node == NUMA_NO_NODE || !node_online(node)) {
454 		/* reuse the value from init_cpu_to_node() */
455 		node = cpu_to_node(cpu);
456 	}
457 	numa_set_node(cpu, node);
458 #endif
459 }
460 
461 #define MSR_IA32_TME_ACTIVATE		0x982
462 
463 /* Helpers to access TME_ACTIVATE MSR */
464 #define TME_ACTIVATE_LOCKED(x)		(x & 0x1)
465 #define TME_ACTIVATE_ENABLED(x)		(x & 0x2)
466 
467 #define TME_ACTIVATE_POLICY(x)		((x >> 4) & 0xf)	/* Bits 7:4 */
468 #define TME_ACTIVATE_POLICY_AES_XTS_128	0
469 
470 #define TME_ACTIVATE_KEYID_BITS(x)	((x >> 32) & 0xf)	/* Bits 35:32 */
471 
472 #define TME_ACTIVATE_CRYPTO_ALGS(x)	((x >> 48) & 0xffff)	/* Bits 63:48 */
473 #define TME_ACTIVATE_CRYPTO_AES_XTS_128	1
474 
475 /* Values for mktme_status (SW only construct) */
476 #define MKTME_ENABLED			0
477 #define MKTME_DISABLED			1
478 #define MKTME_UNINITIALIZED		2
479 static int mktme_status = MKTME_UNINITIALIZED;
480 
481 static void detect_tme(struct cpuinfo_x86 *c)
482 {
483 	u64 tme_activate, tme_policy, tme_crypto_algs;
484 	int keyid_bits = 0, nr_keyids = 0;
485 	static u64 tme_activate_cpu0 = 0;
486 
487 	rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
488 
489 	if (mktme_status != MKTME_UNINITIALIZED) {
490 		if (tme_activate != tme_activate_cpu0) {
491 			/* Broken BIOS? */
492 			pr_err_once("x86/tme: configuration is inconsistent between CPUs\n");
493 			pr_err_once("x86/tme: MKTME is not usable\n");
494 			mktme_status = MKTME_DISABLED;
495 
496 			/* Proceed. We may need to exclude bits from x86_phys_bits. */
497 		}
498 	} else {
499 		tme_activate_cpu0 = tme_activate;
500 	}
501 
502 	if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
503 		pr_info_once("x86/tme: not enabled by BIOS\n");
504 		mktme_status = MKTME_DISABLED;
505 		return;
506 	}
507 
508 	if (mktme_status != MKTME_UNINITIALIZED)
509 		goto detect_keyid_bits;
510 
511 	pr_info("x86/tme: enabled by BIOS\n");
512 
513 	tme_policy = TME_ACTIVATE_POLICY(tme_activate);
514 	if (tme_policy != TME_ACTIVATE_POLICY_AES_XTS_128)
515 		pr_warn("x86/tme: Unknown policy is active: %#llx\n", tme_policy);
516 
517 	tme_crypto_algs = TME_ACTIVATE_CRYPTO_ALGS(tme_activate);
518 	if (!(tme_crypto_algs & TME_ACTIVATE_CRYPTO_AES_XTS_128)) {
519 		pr_err("x86/mktme: No known encryption algorithm is supported: %#llx\n",
520 				tme_crypto_algs);
521 		mktme_status = MKTME_DISABLED;
522 	}
523 detect_keyid_bits:
524 	keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
525 	nr_keyids = (1UL << keyid_bits) - 1;
526 	if (nr_keyids) {
527 		pr_info_once("x86/mktme: enabled by BIOS\n");
528 		pr_info_once("x86/mktme: %d KeyIDs available\n", nr_keyids);
529 	} else {
530 		pr_info_once("x86/mktme: disabled by BIOS\n");
531 	}
532 
533 	if (mktme_status == MKTME_UNINITIALIZED) {
534 		/* MKTME is usable */
535 		mktme_status = MKTME_ENABLED;
536 	}
537 
538 	/*
539 	 * KeyID bits effectively lower the number of physical address
540 	 * bits.  Update cpuinfo_x86::x86_phys_bits accordingly.
541 	 */
542 	c->x86_phys_bits -= keyid_bits;
543 }
544 
545 static void init_cpuid_fault(struct cpuinfo_x86 *c)
546 {
547 	u64 msr;
548 
549 	if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
550 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
551 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
552 	}
553 }
554 
555 static void init_intel_misc_features(struct cpuinfo_x86 *c)
556 {
557 	u64 msr;
558 
559 	if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
560 		return;
561 
562 	/* Clear all MISC features */
563 	this_cpu_write(msr_misc_features_shadow, 0);
564 
565 	/* Check features and update capabilities and shadow control bits */
566 	init_cpuid_fault(c);
567 	probe_xeon_phi_r3mwait(c);
568 
569 	msr = this_cpu_read(msr_misc_features_shadow);
570 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
571 }
572 
573 static void init_intel(struct cpuinfo_x86 *c)
574 {
575 	early_init_intel(c);
576 
577 	intel_workarounds(c);
578 
579 	/*
580 	 * Detect the extended topology information if available. This
581 	 * will reinitialise the initial_apicid which will be used
582 	 * in init_intel_cacheinfo()
583 	 */
584 	detect_extended_topology(c);
585 
586 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
587 		/*
588 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
589 		 * detection.
590 		 */
591 		detect_num_cpu_cores(c);
592 #ifdef CONFIG_X86_32
593 		detect_ht(c);
594 #endif
595 	}
596 
597 	init_intel_cacheinfo(c);
598 
599 	if (c->cpuid_level > 9) {
600 		unsigned eax = cpuid_eax(10);
601 		/* Check for version and the number of counters */
602 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
603 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
604 	}
605 
606 	if (cpu_has(c, X86_FEATURE_XMM2))
607 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
608 
609 	if (boot_cpu_has(X86_FEATURE_DS)) {
610 		unsigned int l1, l2;
611 
612 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
613 		if (!(l1 & (1<<11)))
614 			set_cpu_cap(c, X86_FEATURE_BTS);
615 		if (!(l1 & (1<<12)))
616 			set_cpu_cap(c, X86_FEATURE_PEBS);
617 	}
618 
619 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
620 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
621 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
622 
623 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
624 		((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
625 		set_cpu_bug(c, X86_BUG_MONITOR);
626 
627 #ifdef CONFIG_X86_64
628 	if (c->x86 == 15)
629 		c->x86_cache_alignment = c->x86_clflush_size * 2;
630 	if (c->x86 == 6)
631 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
632 #else
633 	/*
634 	 * Names for the Pentium II/Celeron processors
635 	 * detectable only by also checking the cache size.
636 	 * Dixon is NOT a Celeron.
637 	 */
638 	if (c->x86 == 6) {
639 		unsigned int l2 = c->x86_cache_size;
640 		char *p = NULL;
641 
642 		switch (c->x86_model) {
643 		case 5:
644 			if (l2 == 0)
645 				p = "Celeron (Covington)";
646 			else if (l2 == 256)
647 				p = "Mobile Pentium II (Dixon)";
648 			break;
649 
650 		case 6:
651 			if (l2 == 128)
652 				p = "Celeron (Mendocino)";
653 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
654 				p = "Celeron-A";
655 			break;
656 
657 		case 8:
658 			if (l2 == 128)
659 				p = "Celeron (Coppermine)";
660 			break;
661 		}
662 
663 		if (p)
664 			strcpy(c->x86_model_id, p);
665 	}
666 
667 	if (c->x86 == 15)
668 		set_cpu_cap(c, X86_FEATURE_P4);
669 	if (c->x86 == 6)
670 		set_cpu_cap(c, X86_FEATURE_P3);
671 #endif
672 
673 	/* Work around errata */
674 	srat_detect_node(c);
675 
676 	init_ia32_feat_ctl(c);
677 
678 	if (cpu_has(c, X86_FEATURE_TME))
679 		detect_tme(c);
680 
681 	init_intel_misc_features(c);
682 
683 	if (tsx_ctrl_state == TSX_CTRL_ENABLE)
684 		tsx_enable();
685 	if (tsx_ctrl_state == TSX_CTRL_DISABLE)
686 		tsx_disable();
687 }
688 
689 #ifdef CONFIG_X86_32
690 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
691 {
692 	/*
693 	 * Intel PIII Tualatin. This comes in two flavours.
694 	 * One has 256kb of cache, the other 512. We have no way
695 	 * to determine which, so we use a boottime override
696 	 * for the 512kb model, and assume 256 otherwise.
697 	 */
698 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
699 		size = 256;
700 
701 	/*
702 	 * Intel Quark SoC X1000 contains a 4-way set associative
703 	 * 16K cache with a 16 byte cache line and 256 lines per tag
704 	 */
705 	if ((c->x86 == 5) && (c->x86_model == 9))
706 		size = 16;
707 	return size;
708 }
709 #endif
710 
711 #define TLB_INST_4K	0x01
712 #define TLB_INST_4M	0x02
713 #define TLB_INST_2M_4M	0x03
714 
715 #define TLB_INST_ALL	0x05
716 #define TLB_INST_1G	0x06
717 
718 #define TLB_DATA_4K	0x11
719 #define TLB_DATA_4M	0x12
720 #define TLB_DATA_2M_4M	0x13
721 #define TLB_DATA_4K_4M	0x14
722 
723 #define TLB_DATA_1G	0x16
724 
725 #define TLB_DATA0_4K	0x21
726 #define TLB_DATA0_4M	0x22
727 #define TLB_DATA0_2M_4M	0x23
728 
729 #define STLB_4K		0x41
730 #define STLB_4K_2M	0x42
731 
732 static const struct _tlb_table intel_tlb_table[] = {
733 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
734 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
735 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
736 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
737 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
738 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
739 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages" },
740 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
741 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
742 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
743 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
744 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
745 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
746 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
747 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
748 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
749 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
750 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
751 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
752 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
753 	{ 0x6b, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 8-way associative" },
754 	{ 0x6c, TLB_DATA_2M_4M,		128,	" TLB_DATA 2 MByte or 4 MByte pages, 8-way associative" },
755 	{ 0x6d, TLB_DATA_1G,		16,	" TLB_DATA 1 GByte pages, fully associative" },
756 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
757 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
758 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
759 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
760 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
761 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
762 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
763 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
764 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
765 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
766 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
767 	{ 0xc2, TLB_DATA_2M_4M,		16,	" TLB_DATA 2 MByte/4MByte pages, 4-way associative" },
768 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
769 	{ 0x00, 0, 0 }
770 };
771 
772 static void intel_tlb_lookup(const unsigned char desc)
773 {
774 	unsigned char k;
775 	if (desc == 0)
776 		return;
777 
778 	/* look up this descriptor in the table */
779 	for (k = 0; intel_tlb_table[k].descriptor != desc &&
780 	     intel_tlb_table[k].descriptor != 0; k++)
781 		;
782 
783 	if (intel_tlb_table[k].tlb_type == 0)
784 		return;
785 
786 	switch (intel_tlb_table[k].tlb_type) {
787 	case STLB_4K:
788 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
789 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
790 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
791 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
792 		break;
793 	case STLB_4K_2M:
794 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
795 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
796 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
797 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
798 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
799 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
800 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
801 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
802 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
803 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
804 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
805 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
806 		break;
807 	case TLB_INST_ALL:
808 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
809 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
810 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
811 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
812 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
813 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
814 		break;
815 	case TLB_INST_4K:
816 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
817 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
818 		break;
819 	case TLB_INST_4M:
820 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
821 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
822 		break;
823 	case TLB_INST_2M_4M:
824 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
825 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
826 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
827 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
828 		break;
829 	case TLB_DATA_4K:
830 	case TLB_DATA0_4K:
831 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
832 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
833 		break;
834 	case TLB_DATA_4M:
835 	case TLB_DATA0_4M:
836 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
837 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
838 		break;
839 	case TLB_DATA_2M_4M:
840 	case TLB_DATA0_2M_4M:
841 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
842 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
843 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
844 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
845 		break;
846 	case TLB_DATA_4K_4M:
847 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
848 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
849 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
850 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
851 		break;
852 	case TLB_DATA_1G:
853 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
854 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
855 		break;
856 	}
857 }
858 
859 static void intel_detect_tlb(struct cpuinfo_x86 *c)
860 {
861 	int i, j, n;
862 	unsigned int regs[4];
863 	unsigned char *desc = (unsigned char *)regs;
864 
865 	if (c->cpuid_level < 2)
866 		return;
867 
868 	/* Number of times to iterate */
869 	n = cpuid_eax(2) & 0xFF;
870 
871 	for (i = 0 ; i < n ; i++) {
872 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
873 
874 		/* If bit 31 is set, this is an unknown format */
875 		for (j = 0 ; j < 3 ; j++)
876 			if (regs[j] & (1 << 31))
877 				regs[j] = 0;
878 
879 		/* Byte 0 is level count, not a descriptor */
880 		for (j = 1 ; j < 16 ; j++)
881 			intel_tlb_lookup(desc[j]);
882 	}
883 }
884 
885 static const struct cpu_dev intel_cpu_dev = {
886 	.c_vendor	= "Intel",
887 	.c_ident	= { "GenuineIntel" },
888 #ifdef CONFIG_X86_32
889 	.legacy_models = {
890 		{ .family = 4, .model_names =
891 		  {
892 			  [0] = "486 DX-25/33",
893 			  [1] = "486 DX-50",
894 			  [2] = "486 SX",
895 			  [3] = "486 DX/2",
896 			  [4] = "486 SL",
897 			  [5] = "486 SX/2",
898 			  [7] = "486 DX/2-WB",
899 			  [8] = "486 DX/4",
900 			  [9] = "486 DX/4-WB"
901 		  }
902 		},
903 		{ .family = 5, .model_names =
904 		  {
905 			  [0] = "Pentium 60/66 A-step",
906 			  [1] = "Pentium 60/66",
907 			  [2] = "Pentium 75 - 200",
908 			  [3] = "OverDrive PODP5V83",
909 			  [4] = "Pentium MMX",
910 			  [7] = "Mobile Pentium 75 - 200",
911 			  [8] = "Mobile Pentium MMX",
912 			  [9] = "Quark SoC X1000",
913 		  }
914 		},
915 		{ .family = 6, .model_names =
916 		  {
917 			  [0] = "Pentium Pro A-step",
918 			  [1] = "Pentium Pro",
919 			  [3] = "Pentium II (Klamath)",
920 			  [4] = "Pentium II (Deschutes)",
921 			  [5] = "Pentium II (Deschutes)",
922 			  [6] = "Mobile Pentium II",
923 			  [7] = "Pentium III (Katmai)",
924 			  [8] = "Pentium III (Coppermine)",
925 			  [10] = "Pentium III (Cascades)",
926 			  [11] = "Pentium III (Tualatin)",
927 		  }
928 		},
929 		{ .family = 15, .model_names =
930 		  {
931 			  [0] = "Pentium 4 (Unknown)",
932 			  [1] = "Pentium 4 (Willamette)",
933 			  [2] = "Pentium 4 (Northwood)",
934 			  [4] = "Pentium 4 (Foster)",
935 			  [5] = "Pentium 4 (Foster)",
936 		  }
937 		},
938 	},
939 	.legacy_cache_size = intel_size_cache,
940 #endif
941 	.c_detect_tlb	= intel_detect_tlb,
942 	.c_early_init   = early_init_intel,
943 	.c_init		= init_intel,
944 	.c_x86_vendor	= X86_VENDOR_INTEL,
945 };
946 
947 cpu_dev_register(intel_cpu_dev);
948