xref: /linux/arch/x86/kernel/cpu/intel.c (revision f4cdf7ca9a1fdcca413157df19753f388a5a224e)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/bitops.h>
4 #include <linux/init.h>
5 #include <linux/kernel.h>
6 #include <linux/minmax.h>
7 #include <linux/smp.h>
8 #include <linux/string.h>
9 #include <linux/types.h>
10 
11 #ifdef CONFIG_X86_64
12 #include <linux/topology.h>
13 #endif
14 
15 #include <asm/bugs.h>
16 #include <asm/cpu_device_id.h>
17 #include <asm/cpufeature.h>
18 #include <asm/cpu.h>
19 #include <asm/cpuid/api.h>
20 #include <asm/hwcap2.h>
21 #include <asm/intel-family.h>
22 #include <asm/microcode.h>
23 #include <asm/msr.h>
24 #include <asm/numa.h>
25 #include <asm/resctrl.h>
26 #include <asm/thermal.h>
27 #include <asm/uaccess.h>
28 
29 #include "cpu.h"
30 
31 /*
32  * Processors which have self-snooping capability can handle conflicting
33  * memory type across CPUs by snooping its own cache. However, there exists
34  * CPU models in which having conflicting memory types still leads to
35  * unpredictable behavior, machine check errors, or hangs. Clear this
36  * feature to prevent its use on machines with known erratas.
37  */
38 static void check_memory_type_self_snoop_errata(struct cpuinfo_x86 *c)
39 {
40 	switch (c->x86_vfm) {
41 	case INTEL_CORE_YONAH:
42 	case INTEL_CORE2_MEROM:
43 	case INTEL_CORE2_MEROM_L:
44 	case INTEL_CORE2_PENRYN:
45 	case INTEL_CORE2_DUNNINGTON:
46 	case INTEL_NEHALEM:
47 	case INTEL_NEHALEM_G:
48 	case INTEL_NEHALEM_EP:
49 	case INTEL_NEHALEM_EX:
50 	case INTEL_WESTMERE:
51 	case INTEL_WESTMERE_EP:
52 	case INTEL_SANDYBRIDGE:
53 		setup_clear_cpu_cap(X86_FEATURE_SELFSNOOP);
54 	}
55 }
56 
57 static bool ring3mwait_disabled __read_mostly;
58 
59 static int __init ring3mwait_disable(char *__unused)
60 {
61 	ring3mwait_disabled = true;
62 	return 1;
63 }
64 __setup("ring3mwait=disable", ring3mwait_disable);
65 
66 static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c)
67 {
68 	/*
69 	 * Ring 3 MONITOR/MWAIT feature cannot be detected without
70 	 * cpu model and family comparison.
71 	 */
72 	if (c->x86 != 6)
73 		return;
74 	switch (c->x86_vfm) {
75 	case INTEL_XEON_PHI_KNL:
76 	case INTEL_XEON_PHI_KNM:
77 		break;
78 	default:
79 		return;
80 	}
81 
82 	if (ring3mwait_disabled)
83 		return;
84 
85 	set_cpu_cap(c, X86_FEATURE_RING3MWAIT);
86 	this_cpu_or(msr_misc_features_shadow,
87 		    1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT);
88 
89 	if (c == &boot_cpu_data)
90 		ELF_HWCAP2 |= HWCAP2_RING3MWAIT;
91 }
92 
93 /*
94  * Early microcode releases for the Spectre v2 mitigation were broken.
95  * Information taken from;
96  * - https://newsroom.intel.com/wp-content/uploads/sites/11/2018/03/microcode-update-guidance.pdf
97  * - https://kb.vmware.com/s/article/52345
98  * - Microcode revisions observed in the wild
99  * - Release note from 20180108 microcode release
100  */
101 struct sku_microcode {
102 	u32 vfm;
103 	u8 stepping;
104 	u32 microcode;
105 };
106 static const struct sku_microcode spectre_bad_microcodes[] = {
107 	{ INTEL_KABYLAKE,	0x0B,	0x80 },
108 	{ INTEL_KABYLAKE,	0x0A,	0x80 },
109 	{ INTEL_KABYLAKE,	0x09,	0x80 },
110 	{ INTEL_KABYLAKE_L,	0x0A,	0x80 },
111 	{ INTEL_KABYLAKE_L,	0x09,	0x80 },
112 	{ INTEL_SKYLAKE_X,	0x03,	0x0100013e },
113 	{ INTEL_SKYLAKE_X,	0x04,	0x0200003c },
114 	{ INTEL_BROADWELL,	0x04,	0x28 },
115 	{ INTEL_BROADWELL_G,	0x01,	0x1b },
116 	{ INTEL_BROADWELL_D,	0x02,	0x14 },
117 	{ INTEL_BROADWELL_D,	0x03,	0x07000011 },
118 	{ INTEL_BROADWELL_X,	0x01,	0x0b000025 },
119 	{ INTEL_HASWELL_L,	0x01,	0x21 },
120 	{ INTEL_HASWELL_G,	0x01,	0x18 },
121 	{ INTEL_HASWELL,	0x03,	0x23 },
122 	{ INTEL_HASWELL_X,	0x02,	0x3b },
123 	{ INTEL_HASWELL_X,	0x04,	0x10 },
124 	{ INTEL_IVYBRIDGE_X,	0x04,	0x42a },
125 	/* Observed in the wild */
126 	{ INTEL_SANDYBRIDGE_X,	0x06,	0x61b },
127 	{ INTEL_SANDYBRIDGE_X,	0x07,	0x712 },
128 };
129 
130 static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
131 {
132 	int i;
133 
134 	/*
135 	 * We know that the hypervisor lie to us on the microcode version so
136 	 * we may as well hope that it is running the correct version.
137 	 */
138 	if (cpu_has(c, X86_FEATURE_HYPERVISOR))
139 		return false;
140 
141 	for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
142 		if (c->x86_vfm == spectre_bad_microcodes[i].vfm &&
143 		    c->x86_stepping == spectre_bad_microcodes[i].stepping)
144 			return (c->microcode <= spectre_bad_microcodes[i].microcode);
145 	}
146 	return false;
147 }
148 
149 #define MSR_IA32_TME_ACTIVATE		0x982
150 
151 /* Helpers to access TME_ACTIVATE MSR */
152 #define TME_ACTIVATE_LOCKED(x)		(x & 0x1)
153 #define TME_ACTIVATE_ENABLED(x)		(x & 0x2)
154 
155 #define TME_ACTIVATE_KEYID_BITS(x)	((x >> 32) & 0xf)	/* Bits 35:32 */
156 
157 static void detect_tme_early(struct cpuinfo_x86 *c)
158 {
159 	u64 tme_activate;
160 	int keyid_bits;
161 
162 	rdmsrq(MSR_IA32_TME_ACTIVATE, tme_activate);
163 
164 	if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
165 		pr_info_once("x86/tme: not enabled by BIOS\n");
166 		clear_cpu_cap(c, X86_FEATURE_TME);
167 		return;
168 	}
169 	pr_info_once("x86/tme: enabled by BIOS\n");
170 	keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
171 	if (!keyid_bits)
172 		return;
173 
174 	/*
175 	 * KeyID bits are set by BIOS and can be present regardless
176 	 * of whether the kernel is using them. They effectively lower
177 	 * the number of physical address bits.
178 	 *
179 	 * Update cpuinfo_x86::x86_phys_bits accordingly.
180 	 */
181 	c->x86_phys_bits -= keyid_bits;
182 	pr_info_once("x86/mktme: BIOS enabled: x86_phys_bits reduced by %d\n",
183 		     keyid_bits);
184 }
185 
186 void intel_unlock_cpuid_leafs(struct cpuinfo_x86 *c)
187 {
188 	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
189 		return;
190 
191 	if (c->x86_vfm < INTEL_PENTIUM_M_DOTHAN)
192 		return;
193 
194 	/*
195 	 * The BIOS can have limited CPUID to leaf 2, which breaks feature
196 	 * enumeration. Unlock it and update the maximum leaf info.
197 	 */
198 	if (msr_clear_bit(MSR_IA32_MISC_ENABLE, MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0)
199 		c->cpuid_level = cpuid_eax(0);
200 }
201 
202 /*
203  * Use CPUID to generate a "vfm" value. Useful before cpuinfo_x86
204  * structures are populated.
205  */
206 static u32 intel_cpuid_vfm(void)
207 {
208 	u32 eax   = cpuid_eax(1);
209 	u32 fam   = x86_family(eax);
210 	u32 model = x86_model(eax);
211 
212 	return IFM(fam, model);
213 }
214 
215 u32 intel_get_platform_id(void)
216 {
217 	unsigned int val[2];
218 
219 	if (x86_hypervisor_present)
220 		return 0;
221 
222 	/*
223 	 * This can be called early. Use CPUID directly instead of
224 	 * relying on cpuinfo_x86 which may not be fully initialized.
225 	 * The PII does not have MSR_IA32_PLATFORM_ID. Everything
226 	 * before _it_ has no microcode (for Linux at least).
227 	 */
228 	if (intel_cpuid_vfm() <= INTEL_PENTIUM_II_KLAMATH)
229 		return 0;
230 
231 	/* get processor flags from MSR 0x17 */
232 	native_rdmsr(MSR_IA32_PLATFORM_ID, val[0], val[1]);
233 
234 	return (val[1] >> 18) & 7;
235 }
236 
237 static void early_init_intel(struct cpuinfo_x86 *c)
238 {
239 	u64 misc_enable;
240 
241 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
242 		c->microcode = intel_get_microcode_revision();
243 	c->intel_platform_id = intel_get_platform_id();
244 
245 	/* Now if any of them are set, check the blacklist and clear the lot */
246 	if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
247 	     cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
248 	     cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
249 	     cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
250 		pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
251 		setup_clear_cpu_cap(X86_FEATURE_IBRS);
252 		setup_clear_cpu_cap(X86_FEATURE_IBPB);
253 		setup_clear_cpu_cap(X86_FEATURE_STIBP);
254 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
255 		setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL);
256 		setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
257 		setup_clear_cpu_cap(X86_FEATURE_SSBD);
258 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD);
259 	}
260 
261 	/*
262 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
263 	 *
264 	 * A race condition between speculative fetches and invalidating
265 	 * a large page.  This is worked around in microcode, but we
266 	 * need the microcode to have already been loaded... so if it is
267 	 * not, recommend a BIOS update and disable large pages.
268 	 */
269 	if (c->x86_vfm == INTEL_ATOM_BONNELL && c->x86_stepping <= 2 &&
270 	    c->microcode < 0x20e) {
271 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
272 		clear_cpu_cap(c, X86_FEATURE_PSE);
273 	}
274 
275 #ifndef CONFIG_X86_64
276 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
277 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
278 		c->x86_cache_alignment = 128;
279 #endif
280 
281 	/* CPUID workaround for 0F33/0F34 CPU */
282 	if (c->x86_vfm == INTEL_P4_PRESCOTT &&
283 	    (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
284 		c->x86_phys_bits = 36;
285 
286 	/*
287 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
288 	 * with P/T states and does not stop in deep C-states.
289 	 *
290 	 * It is also reliable across cores and sockets. (but not across
291 	 * cabinets - we turn it off in that case explicitly.)
292 	 *
293 	 * Use a model-specific check for some older CPUs that have invariant
294 	 * TSC but may not report it architecturally via 8000_0007.
295 	 */
296 	if (c->x86_power & (1 << 8)) {
297 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
298 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
299 	} else if ((c->x86_vfm >= INTEL_P4_PRESCOTT && c->x86_vfm <= INTEL_P4_CEDARMILL) ||
300 		   (c->x86_vfm >= INTEL_CORE_YONAH  && c->x86_vfm <= INTEL_IVYBRIDGE)) {
301 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
302 	}
303 
304 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
305 	switch (c->x86_vfm) {
306 	case INTEL_ATOM_SALTWELL_MID:
307 	case INTEL_ATOM_SALTWELL_TABLET:
308 	case INTEL_ATOM_SILVERMONT_MID:
309 	case INTEL_ATOM_AIRMONT_NP:
310 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
311 		break;
312 	}
313 
314 	/*
315 	 * PAT is broken on early family 6 CPUs, the last of which
316 	 * is "Yonah" where the erratum is named "AN7":
317 	 *
318 	 * 	Page with PAT (Page Attribute Table) Set to USWC
319 	 * 	(Uncacheable Speculative Write Combine) While
320 	 * 	Associated MTRR (Memory Type Range Register) Is UC
321 	 * 	(Uncacheable) May Consolidate to UC
322 	 *
323 	 * Disable PAT and fall back to MTRR on these CPUs.
324 	 */
325 	if (c->x86_vfm >= INTEL_PENTIUM_PRO &&
326 	    c->x86_vfm <= INTEL_CORE_YONAH)
327 		clear_cpu_cap(c, X86_FEATURE_PAT);
328 
329 	/*
330 	 * Modern CPUs are generally expected to have a sane fast string
331 	 * implementation. However, BIOSes typically have a knob to tweak
332 	 * the architectural MISC_ENABLE.FAST_STRING enable bit.
333 	 *
334 	 * Adhere to the preference and program the Linux-defined fast
335 	 * string flag and enhanced fast string capabilities accordingly.
336 	 */
337 	if (c->x86_vfm >= INTEL_PENTIUM_M_DOTHAN) {
338 		rdmsrq(MSR_IA32_MISC_ENABLE, misc_enable);
339 		if (misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING) {
340 			/* X86_FEATURE_ERMS is set based on CPUID */
341 			set_cpu_cap(c, X86_FEATURE_REP_GOOD);
342 		} else {
343 			pr_info("Disabled fast string operations\n");
344 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
345 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
346 		}
347 	}
348 
349 	/*
350 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
351 	 * "The operating system also is required to invalidate (i.e., flush)
352 	 *  the TLB when any changes are made to any of the page table entries.
353 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
354 	 *
355 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
356 	 * should be false so that __flush_tlb_all() causes CR3 instead of CR4.PGE
357 	 * to be modified.
358 	 */
359 	if (c->x86_vfm == INTEL_QUARK_X1000) {
360 		pr_info("Disabling PGE capability bit\n");
361 		setup_clear_cpu_cap(X86_FEATURE_PGE);
362 	}
363 
364 	check_memory_type_self_snoop_errata(c);
365 
366 	/*
367 	 * Adjust the number of physical bits early because it affects the
368 	 * valid bits of the MTRR mask registers.
369 	 */
370 	if (cpu_has(c, X86_FEATURE_TME))
371 		detect_tme_early(c);
372 }
373 
374 static void bsp_init_intel(struct cpuinfo_x86 *c)
375 {
376 	resctrl_cpu_detect(c);
377 }
378 
379 #ifdef CONFIG_X86_32
380 /*
381  *	Early probe support logic for ppro memory erratum #50
382  *
383  *	This is called before we do cpu ident work
384  */
385 
386 int ppro_with_ram_bug(void)
387 {
388 	/* Uses data from early_cpu_detect now */
389 	if (boot_cpu_data.x86_vfm == INTEL_PENTIUM_PRO &&
390 	    boot_cpu_data.x86_stepping < 8) {
391 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
392 		return 1;
393 	}
394 	return 0;
395 }
396 
397 static void intel_smp_check(struct cpuinfo_x86 *c)
398 {
399 	/* calling is from identify_secondary_cpu() ? */
400 	if (!c->cpu_index)
401 		return;
402 
403 	/*
404 	 * Mask B, Pentium, but not Pentium MMX
405 	 */
406 	if (c->x86_vfm >= INTEL_FAM5_START && c->x86_vfm < INTEL_PENTIUM_MMX &&
407 	    c->x86_stepping >= 1 && c->x86_stepping <= 4) {
408 		/*
409 		 * Remember we have B step Pentia with bugs
410 		 */
411 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
412 				    "with B stepping processors.\n");
413 	}
414 }
415 
416 static int forcepae;
417 static int __init forcepae_setup(char *__unused)
418 {
419 	forcepae = 1;
420 	return 1;
421 }
422 __setup("forcepae", forcepae_setup);
423 
424 static void intel_workarounds(struct cpuinfo_x86 *c)
425 {
426 	/*
427 	 * All models of Pentium and Pentium with MMX technology CPUs
428 	 * have the F0 0F bug, which lets nonprivileged users lock up the
429 	 * system. The fault handler always checks for it.
430 	 * The Quark is also family 5, but does not have the same bug.
431 	 */
432 	if (IS_ENABLED(CONFIG_X86_F00F_BUG) &&
433 	    (c->x86_vfm >= INTEL_FAM5_START && c->x86_vfm < INTEL_QUARK_X1000))
434 		set_cpu_bug(c, X86_BUG_F00F);
435 
436 	/*
437 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
438 	 * model 3 mask 3
439 	 */
440 	if ((c->x86_vfm == INTEL_PENTIUM_II_KLAMATH && c->x86_stepping < 3) ||
441 	    c->x86_vfm < INTEL_PENTIUM_II_KLAMATH)
442 		clear_cpu_cap(c, X86_FEATURE_SEP);
443 
444 	/*
445 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
446 	 * functionally usable PAE implementation.
447 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
448 	 */
449 	if (forcepae) {
450 		pr_warn("PAE forced!\n");
451 		set_cpu_cap(c, X86_FEATURE_PAE);
452 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
453 	}
454 
455 	/*
456 	 * P4 Xeon erratum 037 workaround.
457 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
458 	 */
459 	if (c->x86_vfm == INTEL_P4_WILLAMETTE && c->x86_stepping == 1) {
460 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
461 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
462 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
463 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
464 		}
465 	}
466 
467 	/*
468 	 * See if we have a good local APIC by checking for buggy Pentia,
469 	 * i.e. all B steppings and the C2 stepping of P54C when using their
470 	 * integrated APIC (see 11AP erratum in "Pentium Processor
471 	 * Specification Update").
472 	 */
473 	if (boot_cpu_has(X86_FEATURE_APIC) && c->x86_vfm == INTEL_PENTIUM_75 &&
474 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
475 		set_cpu_bug(c, X86_BUG_11AP);
476 
477 #ifdef CONFIG_X86_INTEL_USERCOPY
478 	/*
479 	 * MOVSL bulk memory moves can be slow when source and dest are not
480 	 * both 8-byte aligned. PII/PIII only like MOVSL with 8-byte alignment.
481 	 *
482 	 * Set the preferred alignment for Pentium Pro and newer processors, as
483 	 * it has only been tested on these.
484 	 */
485 	if (c->x86_vfm >= INTEL_PENTIUM_PRO)
486 		movsl_mask.mask = 7;
487 #endif
488 
489 	intel_smp_check(c);
490 }
491 #else
492 static void intel_workarounds(struct cpuinfo_x86 *c)
493 {
494 }
495 #endif
496 
497 static void srat_detect_node(struct cpuinfo_x86 *c)
498 {
499 #ifdef CONFIG_NUMA
500 	unsigned node;
501 	int cpu = smp_processor_id();
502 
503 	/* Don't do the funky fallback heuristics the AMD version employs
504 	   for now. */
505 	node = numa_cpu_node(cpu);
506 	if (node == NUMA_NO_NODE || !node_online(node)) {
507 		/* reuse the value from init_cpu_to_node() */
508 		node = cpu_to_node(cpu);
509 	}
510 	numa_set_node(cpu, node);
511 #endif
512 }
513 
514 static void init_cpuid_fault(struct cpuinfo_x86 *c)
515 {
516 	u64 msr;
517 
518 	if (!rdmsrq_safe(MSR_PLATFORM_INFO, &msr)) {
519 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
520 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
521 	}
522 }
523 
524 static void init_intel_misc_features(struct cpuinfo_x86 *c)
525 {
526 	u64 msr;
527 
528 	if (rdmsrq_safe(MSR_MISC_FEATURES_ENABLES, &msr))
529 		return;
530 
531 	/* Clear all MISC features */
532 	this_cpu_write(msr_misc_features_shadow, 0);
533 
534 	/* Check features and update capabilities and shadow control bits */
535 	init_cpuid_fault(c);
536 	probe_xeon_phi_r3mwait(c);
537 
538 	msr = this_cpu_read(msr_misc_features_shadow);
539 	wrmsrq(MSR_MISC_FEATURES_ENABLES, msr);
540 }
541 
542 /*
543  * This is a list of Intel CPUs that are known to suffer from downclocking when
544  * ZMM registers (512-bit vectors) are used.  On these CPUs, when the kernel
545  * executes SIMD-optimized code such as cryptography functions or CRCs, it
546  * should prefer 256-bit (YMM) code to 512-bit (ZMM) code.
547  */
548 static const struct x86_cpu_id zmm_exclusion_list[] = {
549 	X86_MATCH_VFM(INTEL_SKYLAKE_X,		0),
550 	X86_MATCH_VFM(INTEL_ICELAKE_X,		0),
551 	X86_MATCH_VFM(INTEL_ICELAKE_D,		0),
552 	X86_MATCH_VFM(INTEL_ICELAKE,		0),
553 	X86_MATCH_VFM(INTEL_ICELAKE_L,		0),
554 	X86_MATCH_VFM(INTEL_ICELAKE_NNPI,	0),
555 	X86_MATCH_VFM(INTEL_TIGERLAKE_L,	0),
556 	X86_MATCH_VFM(INTEL_TIGERLAKE,		0),
557 	/* Allow Rocket Lake and later, and Sapphire Rapids and later. */
558 	{},
559 };
560 
561 static void init_intel(struct cpuinfo_x86 *c)
562 {
563 	early_init_intel(c);
564 
565 	intel_workarounds(c);
566 
567 	init_intel_cacheinfo(c);
568 
569 	if (c->cpuid_level > 9) {
570 		unsigned eax = cpuid_eax(10);
571 		/* Check for version and the number of counters */
572 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
573 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
574 	}
575 
576 	if (cpu_has(c, X86_FEATURE_XMM2))
577 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
578 
579 	if (boot_cpu_has(X86_FEATURE_DS)) {
580 		u64 l;
581 
582 		rdmsrq(MSR_IA32_MISC_ENABLE, l);
583 		if (!(l & MSR_IA32_MISC_ENABLE_BTS_UNAVAIL))
584 			set_cpu_cap(c, X86_FEATURE_BTS);
585 		if (!(l & MSR_IA32_MISC_ENABLE_PEBS_UNAVAIL))
586 			set_cpu_cap(c, X86_FEATURE_PEBS);
587 	}
588 
589 	if (boot_cpu_has(X86_FEATURE_CLFLUSH) &&
590 	    (c->x86_vfm == INTEL_CORE2_DUNNINGTON ||
591 	     c->x86_vfm == INTEL_NEHALEM_EX ||
592 	     c->x86_vfm == INTEL_WESTMERE_EX))
593 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
594 
595 	if (boot_cpu_has(X86_FEATURE_MWAIT) &&
596 	    (c->x86_vfm == INTEL_ATOM_GOLDMONT ||
597 	     c->x86_vfm == INTEL_LUNARLAKE_M))
598 		set_cpu_bug(c, X86_BUG_MONITOR);
599 
600 #ifdef CONFIG_X86_64
601 	if (c->x86 == 15)
602 		c->x86_cache_alignment = c->x86_clflush_size * 2;
603 #else
604 	/*
605 	 * Names for the Pentium II/Celeron processors
606 	 * detectable only by also checking the cache size.
607 	 * Dixon is NOT a Celeron.
608 	 */
609 	if (c->x86 == 6) {
610 		unsigned int l2 = c->x86_cache_size;
611 		char *p = NULL;
612 
613 		switch (c->x86_model) {
614 		case 5:
615 			if (l2 == 0)
616 				p = "Celeron (Covington)";
617 			else if (l2 == 256)
618 				p = "Mobile Pentium II (Dixon)";
619 			break;
620 
621 		case 6:
622 			if (l2 == 128)
623 				p = "Celeron (Mendocino)";
624 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
625 				p = "Celeron-A";
626 			break;
627 
628 		case 8:
629 			if (l2 == 128)
630 				p = "Celeron (Coppermine)";
631 			break;
632 		}
633 
634 		if (p)
635 			strcpy(c->x86_model_id, p);
636 	}
637 #endif
638 
639 	if (x86_match_cpu(zmm_exclusion_list))
640 		set_cpu_cap(c, X86_FEATURE_PREFER_YMM);
641 
642 	/* Work around errata */
643 	srat_detect_node(c);
644 
645 	init_ia32_feat_ctl(c);
646 
647 	init_intel_misc_features(c);
648 
649 	split_lock_init();
650 
651 	intel_init_thermal(c);
652 }
653 
654 #ifdef CONFIG_X86_32
655 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
656 {
657 	/*
658 	 * Intel PIII Tualatin. This comes in two flavours.
659 	 * One has 256kb of cache, the other 512. We have no way
660 	 * to determine which, so we use a boottime override
661 	 * for the 512kb model, and assume 256 otherwise.
662 	 */
663 	if (c->x86_vfm == INTEL_PENTIUM_III_TUALATIN && size == 0)
664 		size = 256;
665 
666 	/*
667 	 * Intel Quark SoC X1000 contains a 4-way set associative
668 	 * 16K cache with a 16 byte cache line and 256 lines per tag
669 	 */
670 	if (c->x86_vfm == INTEL_QUARK_X1000)
671 		size = 16;
672 	return size;
673 }
674 #endif
675 
676 static void intel_tlb_lookup(const struct leaf_0x2_table *desc)
677 {
678 	short entries = desc->entries;
679 
680 	switch (desc->t_type) {
681 	case STLB_4K:
682 		tlb_lli_4k = max(tlb_lli_4k, entries);
683 		tlb_lld_4k = max(tlb_lld_4k, entries);
684 		break;
685 	case STLB_4K_2M:
686 		tlb_lli_4k = max(tlb_lli_4k, entries);
687 		tlb_lld_4k = max(tlb_lld_4k, entries);
688 		tlb_lli_2m = max(tlb_lli_2m, entries);
689 		tlb_lld_2m = max(tlb_lld_2m, entries);
690 		tlb_lli_4m = max(tlb_lli_4m, entries);
691 		tlb_lld_4m = max(tlb_lld_4m, entries);
692 		break;
693 	case TLB_INST_ALL:
694 		tlb_lli_4k = max(tlb_lli_4k, entries);
695 		tlb_lli_2m = max(tlb_lli_2m, entries);
696 		tlb_lli_4m = max(tlb_lli_4m, entries);
697 		break;
698 	case TLB_INST_4K:
699 		tlb_lli_4k = max(tlb_lli_4k, entries);
700 		break;
701 	case TLB_INST_4M:
702 		tlb_lli_4m = max(tlb_lli_4m, entries);
703 		break;
704 	case TLB_INST_2M_4M:
705 		tlb_lli_2m = max(tlb_lli_2m, entries);
706 		tlb_lli_4m = max(tlb_lli_4m, entries);
707 		break;
708 	case TLB_DATA_4K:
709 	case TLB_DATA0_4K:
710 		tlb_lld_4k = max(tlb_lld_4k, entries);
711 		break;
712 	case TLB_DATA_4M:
713 	case TLB_DATA0_4M:
714 		tlb_lld_4m = max(tlb_lld_4m, entries);
715 		break;
716 	case TLB_DATA_2M_4M:
717 	case TLB_DATA0_2M_4M:
718 		tlb_lld_2m = max(tlb_lld_2m, entries);
719 		tlb_lld_4m = max(tlb_lld_4m, entries);
720 		break;
721 	case TLB_DATA_4K_4M:
722 		tlb_lld_4k = max(tlb_lld_4k, entries);
723 		tlb_lld_4m = max(tlb_lld_4m, entries);
724 		break;
725 	case TLB_DATA_1G_2M_4M:
726 		tlb_lld_2m = max(tlb_lld_2m, TLB_0x63_2M_4M_ENTRIES);
727 		tlb_lld_4m = max(tlb_lld_4m, TLB_0x63_2M_4M_ENTRIES);
728 		fallthrough;
729 	case TLB_DATA_1G:
730 		tlb_lld_1g = max(tlb_lld_1g, entries);
731 		break;
732 	}
733 }
734 
735 static void intel_detect_tlb(struct cpuinfo_x86 *c)
736 {
737 	const struct leaf_0x2_table *desc;
738 	union leaf_0x2_regs regs;
739 	u8 *ptr;
740 
741 	if (c->cpuid_level < 2)
742 		return;
743 
744 	cpuid_leaf_0x2(&regs);
745 	for_each_cpuid_0x2_desc(regs, ptr, desc)
746 		intel_tlb_lookup(desc);
747 }
748 
749 static const struct cpu_dev intel_cpu_dev = {
750 	.c_vendor	= "Intel",
751 	.c_ident	= { "GenuineIntel" },
752 #ifdef CONFIG_X86_32
753 	.legacy_models = {
754 		{ .family = 4, .model_names =
755 		  {
756 			  [0] = "486 DX-25/33",
757 			  [1] = "486 DX-50",
758 			  [2] = "486 SX",
759 			  [3] = "486 DX/2",
760 			  [4] = "486 SL",
761 			  [5] = "486 SX/2",
762 			  [7] = "486 DX/2-WB",
763 			  [8] = "486 DX/4",
764 			  [9] = "486 DX/4-WB"
765 		  }
766 		},
767 		{ .family = 5, .model_names =
768 		  {
769 			  [0] = "Pentium 60/66 A-step",
770 			  [1] = "Pentium 60/66",
771 			  [2] = "Pentium 75 - 200",
772 			  [3] = "OverDrive PODP5V83",
773 			  [4] = "Pentium MMX",
774 			  [7] = "Mobile Pentium 75 - 200",
775 			  [8] = "Mobile Pentium MMX",
776 			  [9] = "Quark SoC X1000",
777 		  }
778 		},
779 		{ .family = 6, .model_names =
780 		  {
781 			  [0] = "Pentium Pro A-step",
782 			  [1] = "Pentium Pro",
783 			  [3] = "Pentium II (Klamath)",
784 			  [4] = "Pentium II (Deschutes)",
785 			  [5] = "Pentium II (Deschutes)",
786 			  [6] = "Mobile Pentium II",
787 			  [7] = "Pentium III (Katmai)",
788 			  [8] = "Pentium III (Coppermine)",
789 			  [10] = "Pentium III (Cascades)",
790 			  [11] = "Pentium III (Tualatin)",
791 		  }
792 		},
793 		{ .family = 15, .model_names =
794 		  {
795 			  [0] = "Pentium 4 (Unknown)",
796 			  [1] = "Pentium 4 (Willamette)",
797 			  [2] = "Pentium 4 (Northwood)",
798 			  [4] = "Pentium 4 (Foster)",
799 			  [5] = "Pentium 4 (Foster)",
800 		  }
801 		},
802 	},
803 	.legacy_cache_size = intel_size_cache,
804 #endif
805 	.c_detect_tlb	= intel_detect_tlb,
806 	.c_early_init   = early_init_intel,
807 	.c_bsp_init	= bsp_init_intel,
808 	.c_init		= init_intel,
809 	.c_x86_vendor	= X86_VENDOR_INTEL,
810 };
811 
812 cpu_dev_register(intel_cpu_dev);
813