xref: /linux/arch/x86/kernel/cpu/intel.c (revision 8a3dc0f7c4ccf13098dba804be06799b4bd46c7a)
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 
10 #ifdef CONFIG_X86_64
11 #include <linux/topology.h>
12 #endif
13 
14 #include <asm/bugs.h>
15 #include <asm/cpu_device_id.h>
16 #include <asm/cpufeature.h>
17 #include <asm/cpu.h>
18 #include <asm/hwcap2.h>
19 #include <asm/intel-family.h>
20 #include <asm/microcode.h>
21 #include <asm/msr.h>
22 #include <asm/numa.h>
23 #include <asm/resctrl.h>
24 #include <asm/thermal.h>
25 #include <asm/uaccess.h>
26 
27 #include "cpu.h"
28 
29 /*
30  * Processors which have self-snooping capability can handle conflicting
31  * memory type across CPUs by snooping its own cache. However, there exists
32  * CPU models in which having conflicting memory types still leads to
33  * unpredictable behavior, machine check errors, or hangs. Clear this
34  * feature to prevent its use on machines with known erratas.
35  */
36 static void check_memory_type_self_snoop_errata(struct cpuinfo_x86 *c)
37 {
38 	switch (c->x86_vfm) {
39 	case INTEL_CORE_YONAH:
40 	case INTEL_CORE2_MEROM:
41 	case INTEL_CORE2_MEROM_L:
42 	case INTEL_CORE2_PENRYN:
43 	case INTEL_CORE2_DUNNINGTON:
44 	case INTEL_NEHALEM:
45 	case INTEL_NEHALEM_G:
46 	case INTEL_NEHALEM_EP:
47 	case INTEL_NEHALEM_EX:
48 	case INTEL_WESTMERE:
49 	case INTEL_WESTMERE_EP:
50 	case INTEL_SANDYBRIDGE:
51 		setup_clear_cpu_cap(X86_FEATURE_SELFSNOOP);
52 	}
53 }
54 
55 static bool ring3mwait_disabled __read_mostly;
56 
57 static int __init ring3mwait_disable(char *__unused)
58 {
59 	ring3mwait_disabled = true;
60 	return 1;
61 }
62 __setup("ring3mwait=disable", ring3mwait_disable);
63 
64 static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c)
65 {
66 	/*
67 	 * Ring 3 MONITOR/MWAIT feature cannot be detected without
68 	 * cpu model and family comparison.
69 	 */
70 	if (c->x86 != 6)
71 		return;
72 	switch (c->x86_vfm) {
73 	case INTEL_XEON_PHI_KNL:
74 	case INTEL_XEON_PHI_KNM:
75 		break;
76 	default:
77 		return;
78 	}
79 
80 	if (ring3mwait_disabled)
81 		return;
82 
83 	set_cpu_cap(c, X86_FEATURE_RING3MWAIT);
84 	this_cpu_or(msr_misc_features_shadow,
85 		    1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT);
86 
87 	if (c == &boot_cpu_data)
88 		ELF_HWCAP2 |= HWCAP2_RING3MWAIT;
89 }
90 
91 /*
92  * Early microcode releases for the Spectre v2 mitigation were broken.
93  * Information taken from;
94  * - https://newsroom.intel.com/wp-content/uploads/sites/11/2018/03/microcode-update-guidance.pdf
95  * - https://kb.vmware.com/s/article/52345
96  * - Microcode revisions observed in the wild
97  * - Release note from 20180108 microcode release
98  */
99 struct sku_microcode {
100 	u32 vfm;
101 	u8 stepping;
102 	u32 microcode;
103 };
104 static const struct sku_microcode spectre_bad_microcodes[] = {
105 	{ INTEL_KABYLAKE,	0x0B,	0x80 },
106 	{ INTEL_KABYLAKE,	0x0A,	0x80 },
107 	{ INTEL_KABYLAKE,	0x09,	0x80 },
108 	{ INTEL_KABYLAKE_L,	0x0A,	0x80 },
109 	{ INTEL_KABYLAKE_L,	0x09,	0x80 },
110 	{ INTEL_SKYLAKE_X,	0x03,	0x0100013e },
111 	{ INTEL_SKYLAKE_X,	0x04,	0x0200003c },
112 	{ INTEL_BROADWELL,	0x04,	0x28 },
113 	{ INTEL_BROADWELL_G,	0x01,	0x1b },
114 	{ INTEL_BROADWELL_D,	0x02,	0x14 },
115 	{ INTEL_BROADWELL_D,	0x03,	0x07000011 },
116 	{ INTEL_BROADWELL_X,	0x01,	0x0b000025 },
117 	{ INTEL_HASWELL_L,	0x01,	0x21 },
118 	{ INTEL_HASWELL_G,	0x01,	0x18 },
119 	{ INTEL_HASWELL,	0x03,	0x23 },
120 	{ INTEL_HASWELL_X,	0x02,	0x3b },
121 	{ INTEL_HASWELL_X,	0x04,	0x10 },
122 	{ INTEL_IVYBRIDGE_X,	0x04,	0x42a },
123 	/* Observed in the wild */
124 	{ INTEL_SANDYBRIDGE_X,	0x06,	0x61b },
125 	{ INTEL_SANDYBRIDGE_X,	0x07,	0x712 },
126 };
127 
128 static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
129 {
130 	int i;
131 
132 	/*
133 	 * We know that the hypervisor lie to us on the microcode version so
134 	 * we may as well hope that it is running the correct version.
135 	 */
136 	if (cpu_has(c, X86_FEATURE_HYPERVISOR))
137 		return false;
138 
139 	for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
140 		if (c->x86_vfm == spectre_bad_microcodes[i].vfm &&
141 		    c->x86_stepping == spectre_bad_microcodes[i].stepping)
142 			return (c->microcode <= spectre_bad_microcodes[i].microcode);
143 	}
144 	return false;
145 }
146 
147 #define MSR_IA32_TME_ACTIVATE		0x982
148 
149 /* Helpers to access TME_ACTIVATE MSR */
150 #define TME_ACTIVATE_LOCKED(x)		(x & 0x1)
151 #define TME_ACTIVATE_ENABLED(x)		(x & 0x2)
152 
153 #define TME_ACTIVATE_KEYID_BITS(x)	((x >> 32) & 0xf)	/* Bits 35:32 */
154 
155 static void detect_tme_early(struct cpuinfo_x86 *c)
156 {
157 	u64 tme_activate;
158 	int keyid_bits;
159 
160 	rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
161 
162 	if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
163 		pr_info_once("x86/tme: not enabled by BIOS\n");
164 		clear_cpu_cap(c, X86_FEATURE_TME);
165 		return;
166 	}
167 	pr_info_once("x86/tme: enabled by BIOS\n");
168 	keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
169 	if (!keyid_bits)
170 		return;
171 
172 	/*
173 	 * KeyID bits are set by BIOS and can be present regardless
174 	 * of whether the kernel is using them. They effectively lower
175 	 * the number of physical address bits.
176 	 *
177 	 * Update cpuinfo_x86::x86_phys_bits accordingly.
178 	 */
179 	c->x86_phys_bits -= keyid_bits;
180 	pr_info_once("x86/mktme: BIOS enabled: x86_phys_bits reduced by %d\n",
181 		     keyid_bits);
182 }
183 
184 void intel_unlock_cpuid_leafs(struct cpuinfo_x86 *c)
185 {
186 	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
187 		return;
188 
189 	if (c->x86 < 6 || (c->x86 == 6 && c->x86_model < 0xd))
190 		return;
191 
192 	/*
193 	 * The BIOS can have limited CPUID to leaf 2, which breaks feature
194 	 * enumeration. Unlock it and update the maximum leaf info.
195 	 */
196 	if (msr_clear_bit(MSR_IA32_MISC_ENABLE, MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0)
197 		c->cpuid_level = cpuid_eax(0);
198 }
199 
200 static void early_init_intel(struct cpuinfo_x86 *c)
201 {
202 	u64 misc_enable;
203 
204 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
205 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
206 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
207 
208 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
209 		c->microcode = intel_get_microcode_revision();
210 
211 	/* Now if any of them are set, check the blacklist and clear the lot */
212 	if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
213 	     cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
214 	     cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
215 	     cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
216 		pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
217 		setup_clear_cpu_cap(X86_FEATURE_IBRS);
218 		setup_clear_cpu_cap(X86_FEATURE_IBPB);
219 		setup_clear_cpu_cap(X86_FEATURE_STIBP);
220 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
221 		setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL);
222 		setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
223 		setup_clear_cpu_cap(X86_FEATURE_SSBD);
224 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD);
225 	}
226 
227 	/*
228 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
229 	 *
230 	 * A race condition between speculative fetches and invalidating
231 	 * a large page.  This is worked around in microcode, but we
232 	 * need the microcode to have already been loaded... so if it is
233 	 * not, recommend a BIOS update and disable large pages.
234 	 */
235 	if (c->x86_vfm == INTEL_ATOM_BONNELL && c->x86_stepping <= 2 &&
236 	    c->microcode < 0x20e) {
237 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
238 		clear_cpu_cap(c, X86_FEATURE_PSE);
239 	}
240 
241 #ifdef CONFIG_X86_64
242 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
243 #else
244 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
245 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
246 		c->x86_cache_alignment = 128;
247 #endif
248 
249 	/* CPUID workaround for 0F33/0F34 CPU */
250 	if (c->x86 == 0xF && c->x86_model == 0x3
251 	    && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
252 		c->x86_phys_bits = 36;
253 
254 	/*
255 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
256 	 * with P/T states and does not stop in deep C-states.
257 	 *
258 	 * It is also reliable across cores and sockets. (but not across
259 	 * cabinets - we turn it off in that case explicitly.)
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 	}
265 
266 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
267 	switch (c->x86_vfm) {
268 	case INTEL_ATOM_SALTWELL_MID:
269 	case INTEL_ATOM_SALTWELL_TABLET:
270 	case INTEL_ATOM_SILVERMONT_MID:
271 	case INTEL_ATOM_AIRMONT_NP:
272 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
273 		break;
274 	}
275 
276 	/*
277 	 * PAT is broken on early family 6 CPUs, the last of which
278 	 * is "Yonah" where the erratum is named "AN7":
279 	 *
280 	 * 	Page with PAT (Page Attribute Table) Set to USWC
281 	 * 	(Uncacheable Speculative Write Combine) While
282 	 * 	Associated MTRR (Memory Type Range Register) Is UC
283 	 * 	(Uncacheable) May Consolidate to UC
284 	 *
285 	 * Disable PAT and fall back to MTRR on these CPUs.
286 	 */
287 	if (c->x86_vfm >= INTEL_PENTIUM_PRO &&
288 	    c->x86_vfm <= INTEL_CORE_YONAH)
289 		clear_cpu_cap(c, X86_FEATURE_PAT);
290 
291 	/*
292 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
293 	 * clear the fast string and enhanced fast string CPU capabilities.
294 	 */
295 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
296 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
297 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
298 			pr_info("Disabled fast string operations\n");
299 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
300 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
301 		}
302 	}
303 
304 	/*
305 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
306 	 * "The operating system also is required to invalidate (i.e., flush)
307 	 *  the TLB when any changes are made to any of the page table entries.
308 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
309 	 *
310 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
311 	 * should be false so that __flush_tlb_all() causes CR3 instead of CR4.PGE
312 	 * to be modified.
313 	 */
314 	if (c->x86_vfm == INTEL_QUARK_X1000) {
315 		pr_info("Disabling PGE capability bit\n");
316 		setup_clear_cpu_cap(X86_FEATURE_PGE);
317 	}
318 
319 	check_memory_type_self_snoop_errata(c);
320 
321 	/*
322 	 * Adjust the number of physical bits early because it affects the
323 	 * valid bits of the MTRR mask registers.
324 	 */
325 	if (cpu_has(c, X86_FEATURE_TME))
326 		detect_tme_early(c);
327 }
328 
329 static void bsp_init_intel(struct cpuinfo_x86 *c)
330 {
331 	resctrl_cpu_detect(c);
332 }
333 
334 #ifdef CONFIG_X86_32
335 /*
336  *	Early probe support logic for ppro memory erratum #50
337  *
338  *	This is called before we do cpu ident work
339  */
340 
341 int ppro_with_ram_bug(void)
342 {
343 	/* Uses data from early_cpu_detect now */
344 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
345 	    boot_cpu_data.x86 == 6 &&
346 	    boot_cpu_data.x86_model == 1 &&
347 	    boot_cpu_data.x86_stepping < 8) {
348 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
349 		return 1;
350 	}
351 	return 0;
352 }
353 
354 static void intel_smp_check(struct cpuinfo_x86 *c)
355 {
356 	/* calling is from identify_secondary_cpu() ? */
357 	if (!c->cpu_index)
358 		return;
359 
360 	/*
361 	 * Mask B, Pentium, but not Pentium MMX
362 	 */
363 	if (c->x86 == 5 &&
364 	    c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
365 	    c->x86_model <= 3) {
366 		/*
367 		 * Remember we have B step Pentia with bugs
368 		 */
369 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
370 				    "with B stepping processors.\n");
371 	}
372 }
373 
374 static int forcepae;
375 static int __init forcepae_setup(char *__unused)
376 {
377 	forcepae = 1;
378 	return 1;
379 }
380 __setup("forcepae", forcepae_setup);
381 
382 static void intel_workarounds(struct cpuinfo_x86 *c)
383 {
384 #ifdef CONFIG_X86_F00F_BUG
385 	/*
386 	 * All models of Pentium and Pentium with MMX technology CPUs
387 	 * have the F0 0F bug, which lets nonprivileged users lock up the
388 	 * system. Announce that the fault handler will be checking for it.
389 	 * The Quark is also family 5, but does not have the same bug.
390 	 */
391 	clear_cpu_bug(c, X86_BUG_F00F);
392 	if (c->x86 == 5 && c->x86_model < 9) {
393 		static int f00f_workaround_enabled;
394 
395 		set_cpu_bug(c, X86_BUG_F00F);
396 		if (!f00f_workaround_enabled) {
397 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
398 			f00f_workaround_enabled = 1;
399 		}
400 	}
401 #endif
402 
403 	/*
404 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
405 	 * model 3 mask 3
406 	 */
407 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
408 		clear_cpu_cap(c, X86_FEATURE_SEP);
409 
410 	/*
411 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
412 	 * functionally usable PAE implementation.
413 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
414 	 */
415 	if (forcepae) {
416 		pr_warn("PAE forced!\n");
417 		set_cpu_cap(c, X86_FEATURE_PAE);
418 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
419 	}
420 
421 	/*
422 	 * P4 Xeon erratum 037 workaround.
423 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
424 	 */
425 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
426 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
427 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
428 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
429 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
430 		}
431 	}
432 
433 	/*
434 	 * See if we have a good local APIC by checking for buggy Pentia,
435 	 * i.e. all B steppings and the C2 stepping of P54C when using their
436 	 * integrated APIC (see 11AP erratum in "Pentium Processor
437 	 * Specification Update").
438 	 */
439 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
440 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
441 		set_cpu_bug(c, X86_BUG_11AP);
442 
443 
444 #ifdef CONFIG_X86_INTEL_USERCOPY
445 	/*
446 	 * Set up the preferred alignment for movsl bulk memory moves
447 	 */
448 	switch (c->x86) {
449 	case 4:		/* 486: untested */
450 		break;
451 	case 5:		/* Old Pentia: untested */
452 		break;
453 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
454 		movsl_mask.mask = 7;
455 		break;
456 	case 15:	/* P4 is OK down to 8-byte alignment */
457 		movsl_mask.mask = 7;
458 		break;
459 	}
460 #endif
461 
462 	intel_smp_check(c);
463 }
464 #else
465 static void intel_workarounds(struct cpuinfo_x86 *c)
466 {
467 }
468 #endif
469 
470 static void srat_detect_node(struct cpuinfo_x86 *c)
471 {
472 #ifdef CONFIG_NUMA
473 	unsigned node;
474 	int cpu = smp_processor_id();
475 
476 	/* Don't do the funky fallback heuristics the AMD version employs
477 	   for now. */
478 	node = numa_cpu_node(cpu);
479 	if (node == NUMA_NO_NODE || !node_online(node)) {
480 		/* reuse the value from init_cpu_to_node() */
481 		node = cpu_to_node(cpu);
482 	}
483 	numa_set_node(cpu, node);
484 #endif
485 }
486 
487 static void init_cpuid_fault(struct cpuinfo_x86 *c)
488 {
489 	u64 msr;
490 
491 	if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
492 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
493 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
494 	}
495 }
496 
497 static void init_intel_misc_features(struct cpuinfo_x86 *c)
498 {
499 	u64 msr;
500 
501 	if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
502 		return;
503 
504 	/* Clear all MISC features */
505 	this_cpu_write(msr_misc_features_shadow, 0);
506 
507 	/* Check features and update capabilities and shadow control bits */
508 	init_cpuid_fault(c);
509 	probe_xeon_phi_r3mwait(c);
510 
511 	msr = this_cpu_read(msr_misc_features_shadow);
512 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
513 }
514 
515 static void init_intel(struct cpuinfo_x86 *c)
516 {
517 	early_init_intel(c);
518 
519 	intel_workarounds(c);
520 
521 	init_intel_cacheinfo(c);
522 
523 	if (c->cpuid_level > 9) {
524 		unsigned eax = cpuid_eax(10);
525 		/* Check for version and the number of counters */
526 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
527 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
528 	}
529 
530 	if (cpu_has(c, X86_FEATURE_XMM2))
531 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
532 
533 	if (boot_cpu_has(X86_FEATURE_DS)) {
534 		unsigned int l1, l2;
535 
536 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
537 		if (!(l1 & MSR_IA32_MISC_ENABLE_BTS_UNAVAIL))
538 			set_cpu_cap(c, X86_FEATURE_BTS);
539 		if (!(l1 & MSR_IA32_MISC_ENABLE_PEBS_UNAVAIL))
540 			set_cpu_cap(c, X86_FEATURE_PEBS);
541 	}
542 
543 	if (boot_cpu_has(X86_FEATURE_CLFLUSH) &&
544 	    (c->x86_vfm == INTEL_CORE2_DUNNINGTON ||
545 	     c->x86_vfm == INTEL_NEHALEM_EX ||
546 	     c->x86_vfm == INTEL_WESTMERE_EX))
547 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
548 
549 	if (boot_cpu_has(X86_FEATURE_MWAIT) &&
550 	    (c->x86_vfm == INTEL_ATOM_GOLDMONT ||
551 	     c->x86_vfm == INTEL_LUNARLAKE_M))
552 		set_cpu_bug(c, X86_BUG_MONITOR);
553 
554 #ifdef CONFIG_X86_64
555 	if (c->x86 == 15)
556 		c->x86_cache_alignment = c->x86_clflush_size * 2;
557 	if (c->x86 == 6)
558 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
559 #else
560 	/*
561 	 * Names for the Pentium II/Celeron processors
562 	 * detectable only by also checking the cache size.
563 	 * Dixon is NOT a Celeron.
564 	 */
565 	if (c->x86 == 6) {
566 		unsigned int l2 = c->x86_cache_size;
567 		char *p = NULL;
568 
569 		switch (c->x86_model) {
570 		case 5:
571 			if (l2 == 0)
572 				p = "Celeron (Covington)";
573 			else if (l2 == 256)
574 				p = "Mobile Pentium II (Dixon)";
575 			break;
576 
577 		case 6:
578 			if (l2 == 128)
579 				p = "Celeron (Mendocino)";
580 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
581 				p = "Celeron-A";
582 			break;
583 
584 		case 8:
585 			if (l2 == 128)
586 				p = "Celeron (Coppermine)";
587 			break;
588 		}
589 
590 		if (p)
591 			strcpy(c->x86_model_id, p);
592 	}
593 #endif
594 
595 	/* Work around errata */
596 	srat_detect_node(c);
597 
598 	init_ia32_feat_ctl(c);
599 
600 	init_intel_misc_features(c);
601 
602 	split_lock_init();
603 
604 	intel_init_thermal(c);
605 }
606 
607 #ifdef CONFIG_X86_32
608 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
609 {
610 	/*
611 	 * Intel PIII Tualatin. This comes in two flavours.
612 	 * One has 256kb of cache, the other 512. We have no way
613 	 * to determine which, so we use a boottime override
614 	 * for the 512kb model, and assume 256 otherwise.
615 	 */
616 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
617 		size = 256;
618 
619 	/*
620 	 * Intel Quark SoC X1000 contains a 4-way set associative
621 	 * 16K cache with a 16 byte cache line and 256 lines per tag
622 	 */
623 	if ((c->x86 == 5) && (c->x86_model == 9))
624 		size = 16;
625 	return size;
626 }
627 #endif
628 
629 #define TLB_INST_4K		0x01
630 #define TLB_INST_4M		0x02
631 #define TLB_INST_2M_4M		0x03
632 
633 #define TLB_INST_ALL		0x05
634 #define TLB_INST_1G		0x06
635 
636 #define TLB_DATA_4K		0x11
637 #define TLB_DATA_4M		0x12
638 #define TLB_DATA_2M_4M		0x13
639 #define TLB_DATA_4K_4M		0x14
640 
641 #define TLB_DATA_1G		0x16
642 #define TLB_DATA_1G_2M_4M	0x17
643 
644 #define TLB_DATA0_4K		0x21
645 #define TLB_DATA0_4M		0x22
646 #define TLB_DATA0_2M_4M		0x23
647 
648 #define STLB_4K			0x41
649 #define STLB_4K_2M		0x42
650 
651 /*
652  * All of leaf 0x2's one-byte TLB descriptors implies the same number of
653  * entries for their respective TLB types.  The 0x63 descriptor is an
654  * exception: it implies 4 dTLB entries for 1GB pages 32 dTLB entries
655  * for 2MB or 4MB pages.  Encode descriptor 0x63 dTLB entry count for
656  * 2MB/4MB pages here, as its count for dTLB 1GB pages is already at the
657  * intel_tlb_table[] mapping.
658  */
659 #define TLB_0x63_2M_4M_ENTRIES	32
660 
661 struct _tlb_table {
662 	unsigned char descriptor;
663 	char tlb_type;
664 	unsigned int entries;
665 };
666 
667 static const struct _tlb_table intel_tlb_table[] = {
668 	{ 0x01, TLB_INST_4K,		32},	/* TLB_INST 4 KByte pages, 4-way set associative */
669 	{ 0x02, TLB_INST_4M,		2},	/* TLB_INST 4 MByte pages, full associative */
670 	{ 0x03, TLB_DATA_4K,		64},	/* TLB_DATA 4 KByte pages, 4-way set associative */
671 	{ 0x04, TLB_DATA_4M,		8},	/* TLB_DATA 4 MByte pages, 4-way set associative */
672 	{ 0x05, TLB_DATA_4M,		32},	/* TLB_DATA 4 MByte pages, 4-way set associative */
673 	{ 0x0b, TLB_INST_4M,		4},	/* TLB_INST 4 MByte pages, 4-way set associative */
674 	{ 0x4f, TLB_INST_4K,		32},	/* TLB_INST 4 KByte pages */
675 	{ 0x50, TLB_INST_ALL,		64},	/* TLB_INST 4 KByte and 2-MByte or 4-MByte pages */
676 	{ 0x51, TLB_INST_ALL,		128},	/* TLB_INST 4 KByte and 2-MByte or 4-MByte pages */
677 	{ 0x52, TLB_INST_ALL,		256},	/* TLB_INST 4 KByte and 2-MByte or 4-MByte pages */
678 	{ 0x55, TLB_INST_2M_4M,		7},	/* TLB_INST 2-MByte or 4-MByte pages, fully associative */
679 	{ 0x56, TLB_DATA0_4M,		16},	/* TLB_DATA0 4 MByte pages, 4-way set associative */
680 	{ 0x57, TLB_DATA0_4K,		16},	/* TLB_DATA0 4 KByte pages, 4-way associative */
681 	{ 0x59, TLB_DATA0_4K,		16},	/* TLB_DATA0 4 KByte pages, fully associative */
682 	{ 0x5a, TLB_DATA0_2M_4M,	32},	/* TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative */
683 	{ 0x5b, TLB_DATA_4K_4M,		64},	/* TLB_DATA 4 KByte and 4 MByte pages */
684 	{ 0x5c, TLB_DATA_4K_4M,		128},	/* TLB_DATA 4 KByte and 4 MByte pages */
685 	{ 0x5d, TLB_DATA_4K_4M,		256},	/* TLB_DATA 4 KByte and 4 MByte pages */
686 	{ 0x61, TLB_INST_4K,		48},	/* TLB_INST 4 KByte pages, full associative */
687 	{ 0x63, TLB_DATA_1G_2M_4M,	4},	/* TLB_DATA 1 GByte pages, 4-way set associative
688 						 * (plus 32 entries TLB_DATA 2 MByte or 4 MByte pages, not encoded here) */
689 	{ 0x6b, TLB_DATA_4K,		256},	/* TLB_DATA 4 KByte pages, 8-way associative */
690 	{ 0x6c, TLB_DATA_2M_4M,		128},	/* TLB_DATA 2 MByte or 4 MByte pages, 8-way associative */
691 	{ 0x6d, TLB_DATA_1G,		16},	/* TLB_DATA 1 GByte pages, fully associative */
692 	{ 0x76, TLB_INST_2M_4M,		8},	/* TLB_INST 2-MByte or 4-MByte pages, fully associative */
693 	{ 0xb0, TLB_INST_4K,		128},	/* TLB_INST 4 KByte pages, 4-way set associative */
694 	{ 0xb1, TLB_INST_2M_4M,		4},	/* TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries */
695 	{ 0xb2, TLB_INST_4K,		64},	/* TLB_INST 4KByte pages, 4-way set associative */
696 	{ 0xb3, TLB_DATA_4K,		128},	/* TLB_DATA 4 KByte pages, 4-way set associative */
697 	{ 0xb4, TLB_DATA_4K,		256},	/* TLB_DATA 4 KByte pages, 4-way associative */
698 	{ 0xb5, TLB_INST_4K,		64},	/* TLB_INST 4 KByte pages, 8-way set associative */
699 	{ 0xb6, TLB_INST_4K,		128},	/* TLB_INST 4 KByte pages, 8-way set associative */
700 	{ 0xba, TLB_DATA_4K,		64},	/* TLB_DATA 4 KByte pages, 4-way associative */
701 	{ 0xc0, TLB_DATA_4K_4M,		8},	/* TLB_DATA 4 KByte and 4 MByte pages, 4-way associative */
702 	{ 0xc1, STLB_4K_2M,		1024},	/* STLB 4 KByte and 2 MByte pages, 8-way associative */
703 	{ 0xc2, TLB_DATA_2M_4M,		16},	/* TLB_DATA 2 MByte/4MByte pages, 4-way associative */
704 	{ 0xca, STLB_4K,		512},	/* STLB 4 KByte pages, 4-way associative */
705 	{ 0x00, 0, 0 }
706 };
707 
708 static void intel_tlb_lookup(const unsigned char desc)
709 {
710 	unsigned int entries;
711 	unsigned char k;
712 
713 	if (desc == 0)
714 		return;
715 
716 	/* look up this descriptor in the table */
717 	for (k = 0; intel_tlb_table[k].descriptor != desc &&
718 	     intel_tlb_table[k].descriptor != 0; k++)
719 		;
720 
721 	if (intel_tlb_table[k].tlb_type == 0)
722 		return;
723 
724 	entries = intel_tlb_table[k].entries;
725 	switch (intel_tlb_table[k].tlb_type) {
726 	case STLB_4K:
727 		tlb_lli_4k = max(tlb_lli_4k, entries);
728 		tlb_lld_4k = max(tlb_lld_4k, entries);
729 		break;
730 	case STLB_4K_2M:
731 		tlb_lli_4k = max(tlb_lli_4k, entries);
732 		tlb_lld_4k = max(tlb_lld_4k, entries);
733 		tlb_lli_2m = max(tlb_lli_2m, entries);
734 		tlb_lld_2m = max(tlb_lld_2m, entries);
735 		tlb_lli_4m = max(tlb_lli_4m, entries);
736 		tlb_lld_4m = max(tlb_lld_4m, entries);
737 		break;
738 	case TLB_INST_ALL:
739 		tlb_lli_4k = max(tlb_lli_4k, entries);
740 		tlb_lli_2m = max(tlb_lli_2m, entries);
741 		tlb_lli_4m = max(tlb_lli_4m, entries);
742 		break;
743 	case TLB_INST_4K:
744 		tlb_lli_4k = max(tlb_lli_4k, entries);
745 		break;
746 	case TLB_INST_4M:
747 		tlb_lli_4m = max(tlb_lli_4m, entries);
748 		break;
749 	case TLB_INST_2M_4M:
750 		tlb_lli_2m = max(tlb_lli_2m, entries);
751 		tlb_lli_4m = max(tlb_lli_4m, entries);
752 		break;
753 	case TLB_DATA_4K:
754 	case TLB_DATA0_4K:
755 		tlb_lld_4k = max(tlb_lld_4k, entries);
756 		break;
757 	case TLB_DATA_4M:
758 	case TLB_DATA0_4M:
759 		tlb_lld_4m = max(tlb_lld_4m, entries);
760 		break;
761 	case TLB_DATA_2M_4M:
762 	case TLB_DATA0_2M_4M:
763 		tlb_lld_2m = max(tlb_lld_2m, entries);
764 		tlb_lld_4m = max(tlb_lld_4m, entries);
765 		break;
766 	case TLB_DATA_4K_4M:
767 		tlb_lld_4k = max(tlb_lld_4k, entries);
768 		tlb_lld_4m = max(tlb_lld_4m, entries);
769 		break;
770 	case TLB_DATA_1G_2M_4M:
771 		tlb_lld_2m = max(tlb_lld_2m, TLB_0x63_2M_4M_ENTRIES);
772 		tlb_lld_4m = max(tlb_lld_4m, TLB_0x63_2M_4M_ENTRIES);
773 		fallthrough;
774 	case TLB_DATA_1G:
775 		tlb_lld_1g = max(tlb_lld_1g, entries);
776 		break;
777 	}
778 }
779 
780 static void intel_detect_tlb(struct cpuinfo_x86 *c)
781 {
782 	int i, j, n;
783 	unsigned int regs[4];
784 	unsigned char *desc = (unsigned char *)regs;
785 
786 	if (c->cpuid_level < 2)
787 		return;
788 
789 	/* Number of times to iterate */
790 	n = cpuid_eax(2) & 0xFF;
791 
792 	for (i = 0 ; i < n ; i++) {
793 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
794 
795 		/* If bit 31 is set, this is an unknown format */
796 		for (j = 0 ; j < 4 ; j++)
797 			if (regs[j] & (1 << 31))
798 				regs[j] = 0;
799 
800 		/* Byte 0 is level count, not a descriptor */
801 		for (j = 1 ; j < 16 ; j++)
802 			intel_tlb_lookup(desc[j]);
803 	}
804 }
805 
806 static const struct cpu_dev intel_cpu_dev = {
807 	.c_vendor	= "Intel",
808 	.c_ident	= { "GenuineIntel" },
809 #ifdef CONFIG_X86_32
810 	.legacy_models = {
811 		{ .family = 4, .model_names =
812 		  {
813 			  [0] = "486 DX-25/33",
814 			  [1] = "486 DX-50",
815 			  [2] = "486 SX",
816 			  [3] = "486 DX/2",
817 			  [4] = "486 SL",
818 			  [5] = "486 SX/2",
819 			  [7] = "486 DX/2-WB",
820 			  [8] = "486 DX/4",
821 			  [9] = "486 DX/4-WB"
822 		  }
823 		},
824 		{ .family = 5, .model_names =
825 		  {
826 			  [0] = "Pentium 60/66 A-step",
827 			  [1] = "Pentium 60/66",
828 			  [2] = "Pentium 75 - 200",
829 			  [3] = "OverDrive PODP5V83",
830 			  [4] = "Pentium MMX",
831 			  [7] = "Mobile Pentium 75 - 200",
832 			  [8] = "Mobile Pentium MMX",
833 			  [9] = "Quark SoC X1000",
834 		  }
835 		},
836 		{ .family = 6, .model_names =
837 		  {
838 			  [0] = "Pentium Pro A-step",
839 			  [1] = "Pentium Pro",
840 			  [3] = "Pentium II (Klamath)",
841 			  [4] = "Pentium II (Deschutes)",
842 			  [5] = "Pentium II (Deschutes)",
843 			  [6] = "Mobile Pentium II",
844 			  [7] = "Pentium III (Katmai)",
845 			  [8] = "Pentium III (Coppermine)",
846 			  [10] = "Pentium III (Cascades)",
847 			  [11] = "Pentium III (Tualatin)",
848 		  }
849 		},
850 		{ .family = 15, .model_names =
851 		  {
852 			  [0] = "Pentium 4 (Unknown)",
853 			  [1] = "Pentium 4 (Willamette)",
854 			  [2] = "Pentium 4 (Northwood)",
855 			  [4] = "Pentium 4 (Foster)",
856 			  [5] = "Pentium 4 (Foster)",
857 		  }
858 		},
859 	},
860 	.legacy_cache_size = intel_size_cache,
861 #endif
862 	.c_detect_tlb	= intel_detect_tlb,
863 	.c_early_init   = early_init_intel,
864 	.c_bsp_init	= bsp_init_intel,
865 	.c_init		= init_intel,
866 	.c_x86_vendor	= X86_VENDOR_INTEL,
867 };
868 
869 cpu_dev_register(intel_cpu_dev);
870