xref: /linux/arch/x86/kernel/cpu/intel.c (revision cd4cdc53cc5e7009ea6ef305336682d7800becf7)
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  */
check_memory_type_self_snoop_errata(struct cpuinfo_x86 * c)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 
ring3mwait_disable(char * __unused)59 static int __init ring3mwait_disable(char *__unused)
60 {
61 	ring3mwait_disabled = true;
62 	return 1;
63 }
64 __setup("ring3mwait=disable", ring3mwait_disable);
65 
probe_xeon_phi_r3mwait(struct cpuinfo_x86 * c)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 
bad_spectre_microcode(struct cpuinfo_x86 * c)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 
detect_tme_early(struct cpuinfo_x86 * c)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 
intel_unlock_cpuid_leafs(struct cpuinfo_x86 * c)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 
early_init_intel(struct cpuinfo_x86 * c)202 static void early_init_intel(struct cpuinfo_x86 *c)
203 {
204 	u64 misc_enable;
205 
206 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
207 		c->microcode = intel_get_microcode_revision();
208 	c->intel_platform_id = intel_get_platform_id();
209 
210 	/* Now if any of them are set, check the blacklist and clear the lot */
211 	if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
212 	     cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
213 	     cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
214 	     cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
215 		pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
216 		setup_clear_cpu_cap(X86_FEATURE_IBRS);
217 		setup_clear_cpu_cap(X86_FEATURE_IBPB);
218 		setup_clear_cpu_cap(X86_FEATURE_STIBP);
219 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
220 		setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL);
221 		setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
222 		setup_clear_cpu_cap(X86_FEATURE_SSBD);
223 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD);
224 	}
225 
226 	/*
227 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
228 	 *
229 	 * A race condition between speculative fetches and invalidating
230 	 * a large page.  This is worked around in microcode, but we
231 	 * need the microcode to have already been loaded... so if it is
232 	 * not, recommend a BIOS update and disable large pages.
233 	 */
234 	if (c->x86_vfm == INTEL_ATOM_BONNELL && c->x86_stepping <= 2 &&
235 	    c->microcode < 0x20e) {
236 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
237 		clear_cpu_cap(c, X86_FEATURE_PSE);
238 	}
239 
240 #ifndef CONFIG_X86_64
241 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
242 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
243 		c->x86_cache_alignment = 128;
244 #endif
245 
246 	/* CPUID workaround for 0F33/0F34 CPU */
247 	if (c->x86_vfm == INTEL_P4_PRESCOTT &&
248 	    (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
249 		c->x86_phys_bits = 36;
250 
251 	/*
252 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
253 	 * with P/T states and does not stop in deep C-states.
254 	 *
255 	 * It is also reliable across cores and sockets. (but not across
256 	 * cabinets - we turn it off in that case explicitly.)
257 	 *
258 	 * Use a model-specific check for some older CPUs that have invariant
259 	 * TSC but may not report it architecturally via 8000_0007.
260 	 */
261 	if (c->x86_power & (1 << 8)) {
262 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
263 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
264 	} else if ((c->x86_vfm >= INTEL_P4_PRESCOTT && c->x86_vfm <= INTEL_P4_CEDARMILL) ||
265 		   (c->x86_vfm >= INTEL_CORE_YONAH  && c->x86_vfm <= INTEL_IVYBRIDGE)) {
266 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
267 	}
268 
269 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
270 	switch (c->x86_vfm) {
271 	case INTEL_ATOM_SALTWELL_MID:
272 	case INTEL_ATOM_SALTWELL_TABLET:
273 	case INTEL_ATOM_SILVERMONT_MID:
274 	case INTEL_ATOM_AIRMONT_NP:
275 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
276 		break;
277 	}
278 
279 	/*
280 	 * PAT is broken on early family 6 CPUs, the last of which
281 	 * is "Yonah" where the erratum is named "AN7":
282 	 *
283 	 * 	Page with PAT (Page Attribute Table) Set to USWC
284 	 * 	(Uncacheable Speculative Write Combine) While
285 	 * 	Associated MTRR (Memory Type Range Register) Is UC
286 	 * 	(Uncacheable) May Consolidate to UC
287 	 *
288 	 * Disable PAT and fall back to MTRR on these CPUs.
289 	 */
290 	if (c->x86_vfm >= INTEL_PENTIUM_PRO &&
291 	    c->x86_vfm <= INTEL_CORE_YONAH)
292 		clear_cpu_cap(c, X86_FEATURE_PAT);
293 
294 	/*
295 	 * Modern CPUs are generally expected to have a sane fast string
296 	 * implementation. However, BIOSes typically have a knob to tweak
297 	 * the architectural MISC_ENABLE.FAST_STRING enable bit.
298 	 *
299 	 * Adhere to the preference and program the Linux-defined fast
300 	 * string flag and enhanced fast string capabilities accordingly.
301 	 */
302 	if (c->x86_vfm >= INTEL_PENTIUM_M_DOTHAN) {
303 		rdmsrq(MSR_IA32_MISC_ENABLE, misc_enable);
304 		if (misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING) {
305 			/* X86_FEATURE_ERMS is set based on CPUID */
306 			set_cpu_cap(c, X86_FEATURE_REP_GOOD);
307 		} else {
308 			pr_info("Disabled fast string operations\n");
309 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
310 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
311 		}
312 	}
313 
314 	/*
315 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
316 	 * "The operating system also is required to invalidate (i.e., flush)
317 	 *  the TLB when any changes are made to any of the page table entries.
318 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
319 	 *
320 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
321 	 * should be false so that __flush_tlb_all() causes CR3 instead of CR4.PGE
322 	 * to be modified.
323 	 */
324 	if (c->x86_vfm == INTEL_QUARK_X1000) {
325 		pr_info("Disabling PGE capability bit\n");
326 		setup_clear_cpu_cap(X86_FEATURE_PGE);
327 	}
328 
329 	check_memory_type_self_snoop_errata(c);
330 
331 	/*
332 	 * Adjust the number of physical bits early because it affects the
333 	 * valid bits of the MTRR mask registers.
334 	 */
335 	if (cpu_has(c, X86_FEATURE_TME))
336 		detect_tme_early(c);
337 }
338 
bsp_init_intel(struct cpuinfo_x86 * c)339 static void bsp_init_intel(struct cpuinfo_x86 *c)
340 {
341 	resctrl_cpu_detect(c);
342 }
343 
344 #ifdef CONFIG_X86_32
345 /*
346  *	Early probe support logic for ppro memory erratum #50
347  *
348  *	This is called before we do cpu ident work
349  */
350 
ppro_with_ram_bug(void)351 int ppro_with_ram_bug(void)
352 {
353 	/* Uses data from early_cpu_detect now */
354 	if (boot_cpu_data.x86_vfm == INTEL_PENTIUM_PRO &&
355 	    boot_cpu_data.x86_stepping < 8) {
356 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
357 		return 1;
358 	}
359 	return 0;
360 }
361 
intel_smp_check(struct cpuinfo_x86 * c)362 static void intel_smp_check(struct cpuinfo_x86 *c)
363 {
364 	/* calling is from identify_secondary_cpu() ? */
365 	if (!c->cpu_index)
366 		return;
367 
368 	/*
369 	 * Mask B, Pentium, but not Pentium MMX
370 	 */
371 	if (c->x86_vfm >= INTEL_FAM5_START && c->x86_vfm < INTEL_PENTIUM_MMX &&
372 	    c->x86_stepping >= 1 && c->x86_stepping <= 4) {
373 		/*
374 		 * Remember we have B step Pentia with bugs
375 		 */
376 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
377 				    "with B stepping processors.\n");
378 	}
379 }
380 
381 static int forcepae;
forcepae_setup(char * __unused)382 static int __init forcepae_setup(char *__unused)
383 {
384 	forcepae = 1;
385 	return 1;
386 }
387 __setup("forcepae", forcepae_setup);
388 
intel_workarounds(struct cpuinfo_x86 * c)389 static void intel_workarounds(struct cpuinfo_x86 *c)
390 {
391 #ifdef CONFIG_X86_F00F_BUG
392 	/*
393 	 * All models of Pentium and Pentium with MMX technology CPUs
394 	 * have the F0 0F bug, which lets nonprivileged users lock up the
395 	 * system. Announce that the fault handler will be checking for it.
396 	 * The Quark is also family 5, but does not have the same bug.
397 	 */
398 	clear_cpu_bug(c, X86_BUG_F00F);
399 	if (c->x86_vfm >= INTEL_FAM5_START && c->x86_vfm < INTEL_QUARK_X1000) {
400 		static int f00f_workaround_enabled;
401 
402 		set_cpu_bug(c, X86_BUG_F00F);
403 		if (!f00f_workaround_enabled) {
404 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
405 			f00f_workaround_enabled = 1;
406 		}
407 	}
408 #endif
409 
410 	/*
411 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
412 	 * model 3 mask 3
413 	 */
414 	if ((c->x86_vfm == INTEL_PENTIUM_II_KLAMATH && c->x86_stepping < 3) ||
415 	    c->x86_vfm < INTEL_PENTIUM_II_KLAMATH)
416 		clear_cpu_cap(c, X86_FEATURE_SEP);
417 
418 	/*
419 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
420 	 * functionally usable PAE implementation.
421 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
422 	 */
423 	if (forcepae) {
424 		pr_warn("PAE forced!\n");
425 		set_cpu_cap(c, X86_FEATURE_PAE);
426 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
427 	}
428 
429 	/*
430 	 * P4 Xeon erratum 037 workaround.
431 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
432 	 */
433 	if (c->x86_vfm == INTEL_P4_WILLAMETTE && c->x86_stepping == 1) {
434 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
435 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
436 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
437 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
438 		}
439 	}
440 
441 	/*
442 	 * See if we have a good local APIC by checking for buggy Pentia,
443 	 * i.e. all B steppings and the C2 stepping of P54C when using their
444 	 * integrated APIC (see 11AP erratum in "Pentium Processor
445 	 * Specification Update").
446 	 */
447 	if (boot_cpu_has(X86_FEATURE_APIC) && c->x86_vfm == INTEL_PENTIUM_75 &&
448 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
449 		set_cpu_bug(c, X86_BUG_11AP);
450 
451 #ifdef CONFIG_X86_INTEL_USERCOPY
452 	/*
453 	 * MOVSL bulk memory moves can be slow when source and dest are not
454 	 * both 8-byte aligned. PII/PIII only like MOVSL with 8-byte alignment.
455 	 *
456 	 * Set the preferred alignment for Pentium Pro and newer processors, as
457 	 * it has only been tested on these.
458 	 */
459 	if (c->x86_vfm >= INTEL_PENTIUM_PRO)
460 		movsl_mask.mask = 7;
461 #endif
462 
463 	intel_smp_check(c);
464 }
465 #else
intel_workarounds(struct cpuinfo_x86 * c)466 static void intel_workarounds(struct cpuinfo_x86 *c)
467 {
468 }
469 #endif
470 
srat_detect_node(struct cpuinfo_x86 * c)471 static void srat_detect_node(struct cpuinfo_x86 *c)
472 {
473 #ifdef CONFIG_NUMA
474 	unsigned node;
475 	int cpu = smp_processor_id();
476 
477 	/* Don't do the funky fallback heuristics the AMD version employs
478 	   for now. */
479 	node = numa_cpu_node(cpu);
480 	if (node == NUMA_NO_NODE || !node_online(node)) {
481 		/* reuse the value from init_cpu_to_node() */
482 		node = cpu_to_node(cpu);
483 	}
484 	numa_set_node(cpu, node);
485 #endif
486 }
487 
init_cpuid_fault(struct cpuinfo_x86 * c)488 static void init_cpuid_fault(struct cpuinfo_x86 *c)
489 {
490 	u64 msr;
491 
492 	if (!rdmsrq_safe(MSR_PLATFORM_INFO, &msr)) {
493 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
494 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
495 	}
496 }
497 
init_intel_misc_features(struct cpuinfo_x86 * c)498 static void init_intel_misc_features(struct cpuinfo_x86 *c)
499 {
500 	u64 msr;
501 
502 	if (rdmsrq_safe(MSR_MISC_FEATURES_ENABLES, &msr))
503 		return;
504 
505 	/* Clear all MISC features */
506 	this_cpu_write(msr_misc_features_shadow, 0);
507 
508 	/* Check features and update capabilities and shadow control bits */
509 	init_cpuid_fault(c);
510 	probe_xeon_phi_r3mwait(c);
511 
512 	msr = this_cpu_read(msr_misc_features_shadow);
513 	wrmsrq(MSR_MISC_FEATURES_ENABLES, msr);
514 }
515 
516 /*
517  * This is a list of Intel CPUs that are known to suffer from downclocking when
518  * ZMM registers (512-bit vectors) are used.  On these CPUs, when the kernel
519  * executes SIMD-optimized code such as cryptography functions or CRCs, it
520  * should prefer 256-bit (YMM) code to 512-bit (ZMM) code.
521  */
522 static const struct x86_cpu_id zmm_exclusion_list[] = {
523 	X86_MATCH_VFM(INTEL_SKYLAKE_X,		0),
524 	X86_MATCH_VFM(INTEL_ICELAKE_X,		0),
525 	X86_MATCH_VFM(INTEL_ICELAKE_D,		0),
526 	X86_MATCH_VFM(INTEL_ICELAKE,		0),
527 	X86_MATCH_VFM(INTEL_ICELAKE_L,		0),
528 	X86_MATCH_VFM(INTEL_ICELAKE_NNPI,	0),
529 	X86_MATCH_VFM(INTEL_TIGERLAKE_L,	0),
530 	X86_MATCH_VFM(INTEL_TIGERLAKE,		0),
531 	/* Allow Rocket Lake and later, and Sapphire Rapids and later. */
532 	{},
533 };
534 
init_intel(struct cpuinfo_x86 * c)535 static void init_intel(struct cpuinfo_x86 *c)
536 {
537 	early_init_intel(c);
538 
539 	intel_workarounds(c);
540 
541 	init_intel_cacheinfo(c);
542 
543 	if (c->cpuid_level > 9) {
544 		unsigned eax = cpuid_eax(10);
545 		/* Check for version and the number of counters */
546 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
547 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
548 	}
549 
550 	if (cpu_has(c, X86_FEATURE_XMM2))
551 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
552 
553 	if (boot_cpu_has(X86_FEATURE_DS)) {
554 		unsigned int l1, l2;
555 
556 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
557 		if (!(l1 & MSR_IA32_MISC_ENABLE_BTS_UNAVAIL))
558 			set_cpu_cap(c, X86_FEATURE_BTS);
559 		if (!(l1 & MSR_IA32_MISC_ENABLE_PEBS_UNAVAIL))
560 			set_cpu_cap(c, X86_FEATURE_PEBS);
561 	}
562 
563 	if (boot_cpu_has(X86_FEATURE_CLFLUSH) &&
564 	    (c->x86_vfm == INTEL_CORE2_DUNNINGTON ||
565 	     c->x86_vfm == INTEL_NEHALEM_EX ||
566 	     c->x86_vfm == INTEL_WESTMERE_EX))
567 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
568 
569 	if (boot_cpu_has(X86_FEATURE_MWAIT) &&
570 	    (c->x86_vfm == INTEL_ATOM_GOLDMONT ||
571 	     c->x86_vfm == INTEL_LUNARLAKE_M))
572 		set_cpu_bug(c, X86_BUG_MONITOR);
573 
574 #ifdef CONFIG_X86_64
575 	if (c->x86 == 15)
576 		c->x86_cache_alignment = c->x86_clflush_size * 2;
577 #else
578 	/*
579 	 * Names for the Pentium II/Celeron processors
580 	 * detectable only by also checking the cache size.
581 	 * Dixon is NOT a Celeron.
582 	 */
583 	if (c->x86 == 6) {
584 		unsigned int l2 = c->x86_cache_size;
585 		char *p = NULL;
586 
587 		switch (c->x86_model) {
588 		case 5:
589 			if (l2 == 0)
590 				p = "Celeron (Covington)";
591 			else if (l2 == 256)
592 				p = "Mobile Pentium II (Dixon)";
593 			break;
594 
595 		case 6:
596 			if (l2 == 128)
597 				p = "Celeron (Mendocino)";
598 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
599 				p = "Celeron-A";
600 			break;
601 
602 		case 8:
603 			if (l2 == 128)
604 				p = "Celeron (Coppermine)";
605 			break;
606 		}
607 
608 		if (p)
609 			strcpy(c->x86_model_id, p);
610 	}
611 #endif
612 
613 	if (x86_match_cpu(zmm_exclusion_list))
614 		set_cpu_cap(c, X86_FEATURE_PREFER_YMM);
615 
616 	/* Work around errata */
617 	srat_detect_node(c);
618 
619 	init_ia32_feat_ctl(c);
620 
621 	init_intel_misc_features(c);
622 
623 	split_lock_init();
624 
625 	intel_init_thermal(c);
626 }
627 
628 #ifdef CONFIG_X86_32
intel_size_cache(struct cpuinfo_x86 * c,unsigned int size)629 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
630 {
631 	/*
632 	 * Intel PIII Tualatin. This comes in two flavours.
633 	 * One has 256kb of cache, the other 512. We have no way
634 	 * to determine which, so we use a boottime override
635 	 * for the 512kb model, and assume 256 otherwise.
636 	 */
637 	if (c->x86_vfm == INTEL_PENTIUM_III_TUALATIN && size == 0)
638 		size = 256;
639 
640 	/*
641 	 * Intel Quark SoC X1000 contains a 4-way set associative
642 	 * 16K cache with a 16 byte cache line and 256 lines per tag
643 	 */
644 	if (c->x86_vfm == INTEL_QUARK_X1000)
645 		size = 16;
646 	return size;
647 }
648 #endif
649 
intel_tlb_lookup(const struct leaf_0x2_table * desc)650 static void intel_tlb_lookup(const struct leaf_0x2_table *desc)
651 {
652 	short entries = desc->entries;
653 
654 	switch (desc->t_type) {
655 	case STLB_4K:
656 		tlb_lli_4k = max(tlb_lli_4k, entries);
657 		tlb_lld_4k = max(tlb_lld_4k, entries);
658 		break;
659 	case STLB_4K_2M:
660 		tlb_lli_4k = max(tlb_lli_4k, entries);
661 		tlb_lld_4k = max(tlb_lld_4k, entries);
662 		tlb_lli_2m = max(tlb_lli_2m, entries);
663 		tlb_lld_2m = max(tlb_lld_2m, entries);
664 		tlb_lli_4m = max(tlb_lli_4m, entries);
665 		tlb_lld_4m = max(tlb_lld_4m, entries);
666 		break;
667 	case TLB_INST_ALL:
668 		tlb_lli_4k = max(tlb_lli_4k, entries);
669 		tlb_lli_2m = max(tlb_lli_2m, entries);
670 		tlb_lli_4m = max(tlb_lli_4m, entries);
671 		break;
672 	case TLB_INST_4K:
673 		tlb_lli_4k = max(tlb_lli_4k, entries);
674 		break;
675 	case TLB_INST_4M:
676 		tlb_lli_4m = max(tlb_lli_4m, entries);
677 		break;
678 	case TLB_INST_2M_4M:
679 		tlb_lli_2m = max(tlb_lli_2m, entries);
680 		tlb_lli_4m = max(tlb_lli_4m, entries);
681 		break;
682 	case TLB_DATA_4K:
683 	case TLB_DATA0_4K:
684 		tlb_lld_4k = max(tlb_lld_4k, entries);
685 		break;
686 	case TLB_DATA_4M:
687 	case TLB_DATA0_4M:
688 		tlb_lld_4m = max(tlb_lld_4m, entries);
689 		break;
690 	case TLB_DATA_2M_4M:
691 	case TLB_DATA0_2M_4M:
692 		tlb_lld_2m = max(tlb_lld_2m, entries);
693 		tlb_lld_4m = max(tlb_lld_4m, entries);
694 		break;
695 	case TLB_DATA_4K_4M:
696 		tlb_lld_4k = max(tlb_lld_4k, entries);
697 		tlb_lld_4m = max(tlb_lld_4m, entries);
698 		break;
699 	case TLB_DATA_1G_2M_4M:
700 		tlb_lld_2m = max(tlb_lld_2m, TLB_0x63_2M_4M_ENTRIES);
701 		tlb_lld_4m = max(tlb_lld_4m, TLB_0x63_2M_4M_ENTRIES);
702 		fallthrough;
703 	case TLB_DATA_1G:
704 		tlb_lld_1g = max(tlb_lld_1g, entries);
705 		break;
706 	}
707 }
708 
intel_detect_tlb(struct cpuinfo_x86 * c)709 static void intel_detect_tlb(struct cpuinfo_x86 *c)
710 {
711 	const struct leaf_0x2_table *desc;
712 	union leaf_0x2_regs regs;
713 	u8 *ptr;
714 
715 	if (c->cpuid_level < 2)
716 		return;
717 
718 	cpuid_leaf_0x2(&regs);
719 	for_each_cpuid_0x2_desc(regs, ptr, desc)
720 		intel_tlb_lookup(desc);
721 }
722 
723 static const struct cpu_dev intel_cpu_dev = {
724 	.c_vendor	= "Intel",
725 	.c_ident	= { "GenuineIntel" },
726 #ifdef CONFIG_X86_32
727 	.legacy_models = {
728 		{ .family = 4, .model_names =
729 		  {
730 			  [0] = "486 DX-25/33",
731 			  [1] = "486 DX-50",
732 			  [2] = "486 SX",
733 			  [3] = "486 DX/2",
734 			  [4] = "486 SL",
735 			  [5] = "486 SX/2",
736 			  [7] = "486 DX/2-WB",
737 			  [8] = "486 DX/4",
738 			  [9] = "486 DX/4-WB"
739 		  }
740 		},
741 		{ .family = 5, .model_names =
742 		  {
743 			  [0] = "Pentium 60/66 A-step",
744 			  [1] = "Pentium 60/66",
745 			  [2] = "Pentium 75 - 200",
746 			  [3] = "OverDrive PODP5V83",
747 			  [4] = "Pentium MMX",
748 			  [7] = "Mobile Pentium 75 - 200",
749 			  [8] = "Mobile Pentium MMX",
750 			  [9] = "Quark SoC X1000",
751 		  }
752 		},
753 		{ .family = 6, .model_names =
754 		  {
755 			  [0] = "Pentium Pro A-step",
756 			  [1] = "Pentium Pro",
757 			  [3] = "Pentium II (Klamath)",
758 			  [4] = "Pentium II (Deschutes)",
759 			  [5] = "Pentium II (Deschutes)",
760 			  [6] = "Mobile Pentium II",
761 			  [7] = "Pentium III (Katmai)",
762 			  [8] = "Pentium III (Coppermine)",
763 			  [10] = "Pentium III (Cascades)",
764 			  [11] = "Pentium III (Tualatin)",
765 		  }
766 		},
767 		{ .family = 15, .model_names =
768 		  {
769 			  [0] = "Pentium 4 (Unknown)",
770 			  [1] = "Pentium 4 (Willamette)",
771 			  [2] = "Pentium 4 (Northwood)",
772 			  [4] = "Pentium 4 (Foster)",
773 			  [5] = "Pentium 4 (Foster)",
774 		  }
775 		},
776 	},
777 	.legacy_cache_size = intel_size_cache,
778 #endif
779 	.c_detect_tlb	= intel_detect_tlb,
780 	.c_early_init   = early_init_intel,
781 	.c_bsp_init	= bsp_init_intel,
782 	.c_init		= init_intel,
783 	.c_x86_vendor	= X86_VENDOR_INTEL,
784 };
785 
786 cpu_dev_register(intel_cpu_dev);
787