xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision 6f69e2a3)
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  * Just in case our CPU detection goes bad, or you have a weird system,
36  * allow a way to override the automatic disabling of MPX.
37  */
38 static int forcempx;
39 
40 static int __init forcempx_setup(char *__unused)
41 {
42 	forcempx = 1;
43 
44 	return 1;
45 }
46 __setup("intel-skd-046-workaround=disable", forcempx_setup);
47 
48 void check_mpx_erratum(struct cpuinfo_x86 *c)
49 {
50 	if (forcempx)
51 		return;
52 	/*
53 	 * Turn off the MPX feature on CPUs where SMEP is not
54 	 * available or disabled.
55 	 *
56 	 * Works around Intel Erratum SKD046: "Branch Instructions
57 	 * May Initialize MPX Bound Registers Incorrectly".
58 	 *
59 	 * This might falsely disable MPX on systems without
60 	 * SMEP, like Atom processors without SMEP.  But there
61 	 * is no such hardware known at the moment.
62 	 */
63 	if (cpu_has(c, X86_FEATURE_MPX) && !cpu_has(c, X86_FEATURE_SMEP)) {
64 		setup_clear_cpu_cap(X86_FEATURE_MPX);
65 		pr_warn("x86/mpx: Disabling MPX since SMEP not present\n");
66 	}
67 }
68 
69 /*
70  * Processors which have self-snooping capability can handle conflicting
71  * memory type across CPUs by snooping its own cache. However, there exists
72  * CPU models in which having conflicting memory types still leads to
73  * unpredictable behavior, machine check errors, or hangs. Clear this
74  * feature to prevent its use on machines with known erratas.
75  */
76 static void check_memory_type_self_snoop_errata(struct cpuinfo_x86 *c)
77 {
78 	switch (c->x86_model) {
79 	case INTEL_FAM6_CORE_YONAH:
80 	case INTEL_FAM6_CORE2_MEROM:
81 	case INTEL_FAM6_CORE2_MEROM_L:
82 	case INTEL_FAM6_CORE2_PENRYN:
83 	case INTEL_FAM6_CORE2_DUNNINGTON:
84 	case INTEL_FAM6_NEHALEM:
85 	case INTEL_FAM6_NEHALEM_G:
86 	case INTEL_FAM6_NEHALEM_EP:
87 	case INTEL_FAM6_NEHALEM_EX:
88 	case INTEL_FAM6_WESTMERE:
89 	case INTEL_FAM6_WESTMERE_EP:
90 	case INTEL_FAM6_SANDYBRIDGE:
91 		setup_clear_cpu_cap(X86_FEATURE_SELFSNOOP);
92 	}
93 }
94 
95 static bool ring3mwait_disabled __read_mostly;
96 
97 static int __init ring3mwait_disable(char *__unused)
98 {
99 	ring3mwait_disabled = true;
100 	return 0;
101 }
102 __setup("ring3mwait=disable", ring3mwait_disable);
103 
104 static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c)
105 {
106 	/*
107 	 * Ring 3 MONITOR/MWAIT feature cannot be detected without
108 	 * cpu model and family comparison.
109 	 */
110 	if (c->x86 != 6)
111 		return;
112 	switch (c->x86_model) {
113 	case INTEL_FAM6_XEON_PHI_KNL:
114 	case INTEL_FAM6_XEON_PHI_KNM:
115 		break;
116 	default:
117 		return;
118 	}
119 
120 	if (ring3mwait_disabled)
121 		return;
122 
123 	set_cpu_cap(c, X86_FEATURE_RING3MWAIT);
124 	this_cpu_or(msr_misc_features_shadow,
125 		    1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT);
126 
127 	if (c == &boot_cpu_data)
128 		ELF_HWCAP2 |= HWCAP2_RING3MWAIT;
129 }
130 
131 /*
132  * Early microcode releases for the Spectre v2 mitigation were broken.
133  * Information taken from;
134  * - https://newsroom.intel.com/wp-content/uploads/sites/11/2018/03/microcode-update-guidance.pdf
135  * - https://kb.vmware.com/s/article/52345
136  * - Microcode revisions observed in the wild
137  * - Release note from 20180108 microcode release
138  */
139 struct sku_microcode {
140 	u8 model;
141 	u8 stepping;
142 	u32 microcode;
143 };
144 static const struct sku_microcode spectre_bad_microcodes[] = {
145 	{ INTEL_FAM6_KABYLAKE,		0x0B,	0x80 },
146 	{ INTEL_FAM6_KABYLAKE,		0x0A,	0x80 },
147 	{ INTEL_FAM6_KABYLAKE,		0x09,	0x80 },
148 	{ INTEL_FAM6_KABYLAKE_L,	0x0A,	0x80 },
149 	{ INTEL_FAM6_KABYLAKE_L,	0x09,	0x80 },
150 	{ INTEL_FAM6_SKYLAKE_X,		0x03,	0x0100013e },
151 	{ INTEL_FAM6_SKYLAKE_X,		0x04,	0x0200003c },
152 	{ INTEL_FAM6_BROADWELL,		0x04,	0x28 },
153 	{ INTEL_FAM6_BROADWELL_G,	0x01,	0x1b },
154 	{ INTEL_FAM6_BROADWELL_D,	0x02,	0x14 },
155 	{ INTEL_FAM6_BROADWELL_D,	0x03,	0x07000011 },
156 	{ INTEL_FAM6_BROADWELL_X,	0x01,	0x0b000025 },
157 	{ INTEL_FAM6_HASWELL_L,		0x01,	0x21 },
158 	{ INTEL_FAM6_HASWELL_G,		0x01,	0x18 },
159 	{ INTEL_FAM6_HASWELL,		0x03,	0x23 },
160 	{ INTEL_FAM6_HASWELL_X,		0x02,	0x3b },
161 	{ INTEL_FAM6_HASWELL_X,		0x04,	0x10 },
162 	{ INTEL_FAM6_IVYBRIDGE_X,	0x04,	0x42a },
163 	/* Observed in the wild */
164 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x06,	0x61b },
165 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x07,	0x712 },
166 };
167 
168 static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
169 {
170 	int i;
171 
172 	/*
173 	 * We know that the hypervisor lie to us on the microcode version so
174 	 * we may as well hope that it is running the correct version.
175 	 */
176 	if (cpu_has(c, X86_FEATURE_HYPERVISOR))
177 		return false;
178 
179 	if (c->x86 != 6)
180 		return false;
181 
182 	for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
183 		if (c->x86_model == spectre_bad_microcodes[i].model &&
184 		    c->x86_stepping == spectre_bad_microcodes[i].stepping)
185 			return (c->microcode <= spectre_bad_microcodes[i].microcode);
186 	}
187 	return false;
188 }
189 
190 static void early_init_intel(struct cpuinfo_x86 *c)
191 {
192 	u64 misc_enable;
193 
194 	/* Unmask CPUID levels if masked: */
195 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
196 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
197 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
198 			c->cpuid_level = cpuid_eax(0);
199 			get_cpu_cap(c);
200 		}
201 	}
202 
203 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
204 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
205 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
206 
207 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
208 		c->microcode = intel_get_microcode_revision();
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 == 6 && c->x86_model == 0x1c && 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 #ifdef CONFIG_X86_64
241 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
242 #else
243 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
244 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
245 		c->x86_cache_alignment = 128;
246 #endif
247 
248 	/* CPUID workaround for 0F33/0F34 CPU */
249 	if (c->x86 == 0xF && c->x86_model == 0x3
250 	    && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
251 		c->x86_phys_bits = 36;
252 
253 	/*
254 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
255 	 * with P/T states and does not stop in deep C-states.
256 	 *
257 	 * It is also reliable across cores and sockets. (but not across
258 	 * cabinets - we turn it off in that case explicitly.)
259 	 */
260 	if (c->x86_power & (1 << 8)) {
261 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
262 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
263 	}
264 
265 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
266 	if (c->x86 == 6) {
267 		switch (c->x86_model) {
268 		case INTEL_FAM6_ATOM_SALTWELL_MID:
269 		case INTEL_FAM6_ATOM_SALTWELL_TABLET:
270 		case INTEL_FAM6_ATOM_SILVERMONT_MID:
271 		case INTEL_FAM6_ATOM_AIRMONT_NP:
272 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
273 			break;
274 		default:
275 			break;
276 		}
277 	}
278 
279 	/*
280 	 * There is a known erratum on Pentium III and Core Solo
281 	 * and Core Duo CPUs.
282 	 * " Page with PAT set to WC while associated MTRR is UC
283 	 *   may consolidate to UC "
284 	 * Because of this erratum, it is better to stick with
285 	 * setting WC in MTRR rather than using PAT on these CPUs.
286 	 *
287 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
288 	 */
289 	if (c->x86 == 6 && c->x86_model < 15)
290 		clear_cpu_cap(c, X86_FEATURE_PAT);
291 
292 	/*
293 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
294 	 * clear the fast string and enhanced fast string CPU capabilities.
295 	 */
296 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
297 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
298 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
299 			pr_info("Disabled fast string operations\n");
300 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
301 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
302 		}
303 	}
304 
305 	/*
306 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
307 	 * "The operating system also is required to invalidate (i.e., flush)
308 	 *  the TLB when any changes are made to any of the page table entries.
309 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
310 	 *
311 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
312 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
313 	 * to be modified.
314 	 */
315 	if (c->x86 == 5 && c->x86_model == 9) {
316 		pr_info("Disabling PGE capability bit\n");
317 		setup_clear_cpu_cap(X86_FEATURE_PGE);
318 	}
319 
320 	if (c->cpuid_level >= 0x00000001) {
321 		u32 eax, ebx, ecx, edx;
322 
323 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
324 		/*
325 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
326 		 * apicids which are reserved per package. Store the resulting
327 		 * shift value for the package management code.
328 		 */
329 		if (edx & (1U << 28))
330 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
331 	}
332 
333 	check_mpx_erratum(c);
334 	check_memory_type_self_snoop_errata(c);
335 
336 	/*
337 	 * Get the number of SMT siblings early from the extended topology
338 	 * leaf, if available. Otherwise try the legacy SMT detection.
339 	 */
340 	if (detect_extended_topology_early(c) < 0)
341 		detect_ht_early(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 
351 int ppro_with_ram_bug(void)
352 {
353 	/* Uses data from early_cpu_detect now */
354 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
355 	    boot_cpu_data.x86 == 6 &&
356 	    boot_cpu_data.x86_model == 1 &&
357 	    boot_cpu_data.x86_stepping < 8) {
358 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
359 		return 1;
360 	}
361 	return 0;
362 }
363 
364 static void intel_smp_check(struct cpuinfo_x86 *c)
365 {
366 	/* calling is from identify_secondary_cpu() ? */
367 	if (!c->cpu_index)
368 		return;
369 
370 	/*
371 	 * Mask B, Pentium, but not Pentium MMX
372 	 */
373 	if (c->x86 == 5 &&
374 	    c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
375 	    c->x86_model <= 3) {
376 		/*
377 		 * Remember we have B step Pentia with bugs
378 		 */
379 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
380 				    "with B stepping processors.\n");
381 	}
382 }
383 
384 static int forcepae;
385 static int __init forcepae_setup(char *__unused)
386 {
387 	forcepae = 1;
388 	return 1;
389 }
390 __setup("forcepae", forcepae_setup);
391 
392 static void intel_workarounds(struct cpuinfo_x86 *c)
393 {
394 #ifdef CONFIG_X86_F00F_BUG
395 	/*
396 	 * All models of Pentium and Pentium with MMX technology CPUs
397 	 * have the F0 0F bug, which lets nonprivileged users lock up the
398 	 * system. Announce that the fault handler will be checking for it.
399 	 * The Quark is also family 5, but does not have the same bug.
400 	 */
401 	clear_cpu_bug(c, X86_BUG_F00F);
402 	if (c->x86 == 5 && c->x86_model < 9) {
403 		static int f00f_workaround_enabled;
404 
405 		set_cpu_bug(c, X86_BUG_F00F);
406 		if (!f00f_workaround_enabled) {
407 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
408 			f00f_workaround_enabled = 1;
409 		}
410 	}
411 #endif
412 
413 	/*
414 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
415 	 * model 3 mask 3
416 	 */
417 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
418 		clear_cpu_cap(c, X86_FEATURE_SEP);
419 
420 	/*
421 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
422 	 * functionally usable PAE implementation.
423 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
424 	 */
425 	if (forcepae) {
426 		pr_warn("PAE forced!\n");
427 		set_cpu_cap(c, X86_FEATURE_PAE);
428 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
429 	}
430 
431 	/*
432 	 * P4 Xeon erratum 037 workaround.
433 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
434 	 */
435 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
436 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
437 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
438 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
439 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
440 		}
441 	}
442 
443 	/*
444 	 * See if we have a good local APIC by checking for buggy Pentia,
445 	 * i.e. all B steppings and the C2 stepping of P54C when using their
446 	 * integrated APIC (see 11AP erratum in "Pentium Processor
447 	 * Specification Update").
448 	 */
449 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
450 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
451 		set_cpu_bug(c, X86_BUG_11AP);
452 
453 
454 #ifdef CONFIG_X86_INTEL_USERCOPY
455 	/*
456 	 * Set up the preferred alignment for movsl bulk memory moves
457 	 */
458 	switch (c->x86) {
459 	case 4:		/* 486: untested */
460 		break;
461 	case 5:		/* Old Pentia: untested */
462 		break;
463 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
464 		movsl_mask.mask = 7;
465 		break;
466 	case 15:	/* P4 is OK down to 8-byte alignment */
467 		movsl_mask.mask = 7;
468 		break;
469 	}
470 #endif
471 
472 	intel_smp_check(c);
473 }
474 #else
475 static void intel_workarounds(struct cpuinfo_x86 *c)
476 {
477 }
478 #endif
479 
480 static void srat_detect_node(struct cpuinfo_x86 *c)
481 {
482 #ifdef CONFIG_NUMA
483 	unsigned node;
484 	int cpu = smp_processor_id();
485 
486 	/* Don't do the funky fallback heuristics the AMD version employs
487 	   for now. */
488 	node = numa_cpu_node(cpu);
489 	if (node == NUMA_NO_NODE || !node_online(node)) {
490 		/* reuse the value from init_cpu_to_node() */
491 		node = cpu_to_node(cpu);
492 	}
493 	numa_set_node(cpu, node);
494 #endif
495 }
496 
497 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
498 {
499 	/* Intel VMX MSR indicated features */
500 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
501 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
502 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
503 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
504 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
505 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
506 #define x86_VMX_FEATURE_EPT_CAP_AD		0x00200000
507 
508 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
509 	u32 msr_vpid_cap, msr_ept_cap;
510 
511 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
512 	clear_cpu_cap(c, X86_FEATURE_VNMI);
513 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
514 	clear_cpu_cap(c, X86_FEATURE_EPT);
515 	clear_cpu_cap(c, X86_FEATURE_VPID);
516 	clear_cpu_cap(c, X86_FEATURE_EPT_AD);
517 
518 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
519 	msr_ctl = vmx_msr_high | vmx_msr_low;
520 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
521 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
522 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
523 		set_cpu_cap(c, X86_FEATURE_VNMI);
524 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
525 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
526 		      vmx_msr_low, vmx_msr_high);
527 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
528 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
529 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
530 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
531 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT) {
532 			set_cpu_cap(c, X86_FEATURE_EPT);
533 			rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
534 			      msr_ept_cap, msr_vpid_cap);
535 			if (msr_ept_cap & x86_VMX_FEATURE_EPT_CAP_AD)
536 				set_cpu_cap(c, X86_FEATURE_EPT_AD);
537 		}
538 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
539 			set_cpu_cap(c, X86_FEATURE_VPID);
540 	}
541 }
542 
543 #define MSR_IA32_TME_ACTIVATE		0x982
544 
545 /* Helpers to access TME_ACTIVATE MSR */
546 #define TME_ACTIVATE_LOCKED(x)		(x & 0x1)
547 #define TME_ACTIVATE_ENABLED(x)		(x & 0x2)
548 
549 #define TME_ACTIVATE_POLICY(x)		((x >> 4) & 0xf)	/* Bits 7:4 */
550 #define TME_ACTIVATE_POLICY_AES_XTS_128	0
551 
552 #define TME_ACTIVATE_KEYID_BITS(x)	((x >> 32) & 0xf)	/* Bits 35:32 */
553 
554 #define TME_ACTIVATE_CRYPTO_ALGS(x)	((x >> 48) & 0xffff)	/* Bits 63:48 */
555 #define TME_ACTIVATE_CRYPTO_AES_XTS_128	1
556 
557 /* Values for mktme_status (SW only construct) */
558 #define MKTME_ENABLED			0
559 #define MKTME_DISABLED			1
560 #define MKTME_UNINITIALIZED		2
561 static int mktme_status = MKTME_UNINITIALIZED;
562 
563 static void detect_tme(struct cpuinfo_x86 *c)
564 {
565 	u64 tme_activate, tme_policy, tme_crypto_algs;
566 	int keyid_bits = 0, nr_keyids = 0;
567 	static u64 tme_activate_cpu0 = 0;
568 
569 	rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
570 
571 	if (mktme_status != MKTME_UNINITIALIZED) {
572 		if (tme_activate != tme_activate_cpu0) {
573 			/* Broken BIOS? */
574 			pr_err_once("x86/tme: configuration is inconsistent between CPUs\n");
575 			pr_err_once("x86/tme: MKTME is not usable\n");
576 			mktme_status = MKTME_DISABLED;
577 
578 			/* Proceed. We may need to exclude bits from x86_phys_bits. */
579 		}
580 	} else {
581 		tme_activate_cpu0 = tme_activate;
582 	}
583 
584 	if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
585 		pr_info_once("x86/tme: not enabled by BIOS\n");
586 		mktme_status = MKTME_DISABLED;
587 		return;
588 	}
589 
590 	if (mktme_status != MKTME_UNINITIALIZED)
591 		goto detect_keyid_bits;
592 
593 	pr_info("x86/tme: enabled by BIOS\n");
594 
595 	tme_policy = TME_ACTIVATE_POLICY(tme_activate);
596 	if (tme_policy != TME_ACTIVATE_POLICY_AES_XTS_128)
597 		pr_warn("x86/tme: Unknown policy is active: %#llx\n", tme_policy);
598 
599 	tme_crypto_algs = TME_ACTIVATE_CRYPTO_ALGS(tme_activate);
600 	if (!(tme_crypto_algs & TME_ACTIVATE_CRYPTO_AES_XTS_128)) {
601 		pr_err("x86/mktme: No known encryption algorithm is supported: %#llx\n",
602 				tme_crypto_algs);
603 		mktme_status = MKTME_DISABLED;
604 	}
605 detect_keyid_bits:
606 	keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
607 	nr_keyids = (1UL << keyid_bits) - 1;
608 	if (nr_keyids) {
609 		pr_info_once("x86/mktme: enabled by BIOS\n");
610 		pr_info_once("x86/mktme: %d KeyIDs available\n", nr_keyids);
611 	} else {
612 		pr_info_once("x86/mktme: disabled by BIOS\n");
613 	}
614 
615 	if (mktme_status == MKTME_UNINITIALIZED) {
616 		/* MKTME is usable */
617 		mktme_status = MKTME_ENABLED;
618 	}
619 
620 	/*
621 	 * KeyID bits effectively lower the number of physical address
622 	 * bits.  Update cpuinfo_x86::x86_phys_bits accordingly.
623 	 */
624 	c->x86_phys_bits -= keyid_bits;
625 }
626 
627 static void init_cpuid_fault(struct cpuinfo_x86 *c)
628 {
629 	u64 msr;
630 
631 	if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
632 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
633 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
634 	}
635 }
636 
637 static void init_intel_misc_features(struct cpuinfo_x86 *c)
638 {
639 	u64 msr;
640 
641 	if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
642 		return;
643 
644 	/* Clear all MISC features */
645 	this_cpu_write(msr_misc_features_shadow, 0);
646 
647 	/* Check features and update capabilities and shadow control bits */
648 	init_cpuid_fault(c);
649 	probe_xeon_phi_r3mwait(c);
650 
651 	msr = this_cpu_read(msr_misc_features_shadow);
652 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
653 }
654 
655 static void init_intel(struct cpuinfo_x86 *c)
656 {
657 	early_init_intel(c);
658 
659 	intel_workarounds(c);
660 
661 	/*
662 	 * Detect the extended topology information if available. This
663 	 * will reinitialise the initial_apicid which will be used
664 	 * in init_intel_cacheinfo()
665 	 */
666 	detect_extended_topology(c);
667 
668 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
669 		/*
670 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
671 		 * detection.
672 		 */
673 		detect_num_cpu_cores(c);
674 #ifdef CONFIG_X86_32
675 		detect_ht(c);
676 #endif
677 	}
678 
679 	init_intel_cacheinfo(c);
680 
681 	if (c->cpuid_level > 9) {
682 		unsigned eax = cpuid_eax(10);
683 		/* Check for version and the number of counters */
684 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
685 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
686 	}
687 
688 	if (cpu_has(c, X86_FEATURE_XMM2))
689 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
690 
691 	if (boot_cpu_has(X86_FEATURE_DS)) {
692 		unsigned int l1, l2;
693 
694 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
695 		if (!(l1 & (1<<11)))
696 			set_cpu_cap(c, X86_FEATURE_BTS);
697 		if (!(l1 & (1<<12)))
698 			set_cpu_cap(c, X86_FEATURE_PEBS);
699 	}
700 
701 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
702 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
703 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
704 
705 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
706 		((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
707 		set_cpu_bug(c, X86_BUG_MONITOR);
708 
709 #ifdef CONFIG_X86_64
710 	if (c->x86 == 15)
711 		c->x86_cache_alignment = c->x86_clflush_size * 2;
712 	if (c->x86 == 6)
713 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
714 #else
715 	/*
716 	 * Names for the Pentium II/Celeron processors
717 	 * detectable only by also checking the cache size.
718 	 * Dixon is NOT a Celeron.
719 	 */
720 	if (c->x86 == 6) {
721 		unsigned int l2 = c->x86_cache_size;
722 		char *p = NULL;
723 
724 		switch (c->x86_model) {
725 		case 5:
726 			if (l2 == 0)
727 				p = "Celeron (Covington)";
728 			else if (l2 == 256)
729 				p = "Mobile Pentium II (Dixon)";
730 			break;
731 
732 		case 6:
733 			if (l2 == 128)
734 				p = "Celeron (Mendocino)";
735 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
736 				p = "Celeron-A";
737 			break;
738 
739 		case 8:
740 			if (l2 == 128)
741 				p = "Celeron (Coppermine)";
742 			break;
743 		}
744 
745 		if (p)
746 			strcpy(c->x86_model_id, p);
747 	}
748 
749 	if (c->x86 == 15)
750 		set_cpu_cap(c, X86_FEATURE_P4);
751 	if (c->x86 == 6)
752 		set_cpu_cap(c, X86_FEATURE_P3);
753 #endif
754 
755 	/* Work around errata */
756 	srat_detect_node(c);
757 
758 	if (cpu_has(c, X86_FEATURE_VMX))
759 		detect_vmx_virtcap(c);
760 
761 	if (cpu_has(c, X86_FEATURE_TME))
762 		detect_tme(c);
763 
764 	init_intel_misc_features(c);
765 
766 	if (tsx_ctrl_state == TSX_CTRL_ENABLE)
767 		tsx_enable();
768 	if (tsx_ctrl_state == TSX_CTRL_DISABLE)
769 		tsx_disable();
770 }
771 
772 #ifdef CONFIG_X86_32
773 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
774 {
775 	/*
776 	 * Intel PIII Tualatin. This comes in two flavours.
777 	 * One has 256kb of cache, the other 512. We have no way
778 	 * to determine which, so we use a boottime override
779 	 * for the 512kb model, and assume 256 otherwise.
780 	 */
781 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
782 		size = 256;
783 
784 	/*
785 	 * Intel Quark SoC X1000 contains a 4-way set associative
786 	 * 16K cache with a 16 byte cache line and 256 lines per tag
787 	 */
788 	if ((c->x86 == 5) && (c->x86_model == 9))
789 		size = 16;
790 	return size;
791 }
792 #endif
793 
794 #define TLB_INST_4K	0x01
795 #define TLB_INST_4M	0x02
796 #define TLB_INST_2M_4M	0x03
797 
798 #define TLB_INST_ALL	0x05
799 #define TLB_INST_1G	0x06
800 
801 #define TLB_DATA_4K	0x11
802 #define TLB_DATA_4M	0x12
803 #define TLB_DATA_2M_4M	0x13
804 #define TLB_DATA_4K_4M	0x14
805 
806 #define TLB_DATA_1G	0x16
807 
808 #define TLB_DATA0_4K	0x21
809 #define TLB_DATA0_4M	0x22
810 #define TLB_DATA0_2M_4M	0x23
811 
812 #define STLB_4K		0x41
813 #define STLB_4K_2M	0x42
814 
815 static const struct _tlb_table intel_tlb_table[] = {
816 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
817 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
818 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
819 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
820 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
821 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
822 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages" },
823 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
824 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
825 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
826 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
827 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
828 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
829 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
830 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
831 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
832 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
833 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
834 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
835 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
836 	{ 0x6b, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 8-way associative" },
837 	{ 0x6c, TLB_DATA_2M_4M,		128,	" TLB_DATA 2 MByte or 4 MByte pages, 8-way associative" },
838 	{ 0x6d, TLB_DATA_1G,		16,	" TLB_DATA 1 GByte pages, fully associative" },
839 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
840 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
841 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
842 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
843 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
844 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
845 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
846 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
847 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
848 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
849 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
850 	{ 0xc2, TLB_DATA_2M_4M,		16,	" TLB_DATA 2 MByte/4MByte pages, 4-way associative" },
851 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
852 	{ 0x00, 0, 0 }
853 };
854 
855 static void intel_tlb_lookup(const unsigned char desc)
856 {
857 	unsigned char k;
858 	if (desc == 0)
859 		return;
860 
861 	/* look up this descriptor in the table */
862 	for (k = 0; intel_tlb_table[k].descriptor != desc &&
863 	     intel_tlb_table[k].descriptor != 0; k++)
864 		;
865 
866 	if (intel_tlb_table[k].tlb_type == 0)
867 		return;
868 
869 	switch (intel_tlb_table[k].tlb_type) {
870 	case STLB_4K:
871 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
872 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
873 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
874 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
875 		break;
876 	case STLB_4K_2M:
877 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
878 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
879 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
880 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
881 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
882 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
883 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
884 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
885 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
886 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
887 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
888 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
889 		break;
890 	case TLB_INST_ALL:
891 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
892 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
893 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
894 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
895 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
896 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
897 		break;
898 	case TLB_INST_4K:
899 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
900 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
901 		break;
902 	case TLB_INST_4M:
903 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
904 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
905 		break;
906 	case TLB_INST_2M_4M:
907 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
908 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
909 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
910 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
911 		break;
912 	case TLB_DATA_4K:
913 	case TLB_DATA0_4K:
914 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
915 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
916 		break;
917 	case TLB_DATA_4M:
918 	case TLB_DATA0_4M:
919 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
920 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
921 		break;
922 	case TLB_DATA_2M_4M:
923 	case TLB_DATA0_2M_4M:
924 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
925 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
926 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
927 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
928 		break;
929 	case TLB_DATA_4K_4M:
930 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
931 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
932 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
933 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
934 		break;
935 	case TLB_DATA_1G:
936 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
937 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
938 		break;
939 	}
940 }
941 
942 static void intel_detect_tlb(struct cpuinfo_x86 *c)
943 {
944 	int i, j, n;
945 	unsigned int regs[4];
946 	unsigned char *desc = (unsigned char *)regs;
947 
948 	if (c->cpuid_level < 2)
949 		return;
950 
951 	/* Number of times to iterate */
952 	n = cpuid_eax(2) & 0xFF;
953 
954 	for (i = 0 ; i < n ; i++) {
955 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
956 
957 		/* If bit 31 is set, this is an unknown format */
958 		for (j = 0 ; j < 3 ; j++)
959 			if (regs[j] & (1 << 31))
960 				regs[j] = 0;
961 
962 		/* Byte 0 is level count, not a descriptor */
963 		for (j = 1 ; j < 16 ; j++)
964 			intel_tlb_lookup(desc[j]);
965 	}
966 }
967 
968 static const struct cpu_dev intel_cpu_dev = {
969 	.c_vendor	= "Intel",
970 	.c_ident	= { "GenuineIntel" },
971 #ifdef CONFIG_X86_32
972 	.legacy_models = {
973 		{ .family = 4, .model_names =
974 		  {
975 			  [0] = "486 DX-25/33",
976 			  [1] = "486 DX-50",
977 			  [2] = "486 SX",
978 			  [3] = "486 DX/2",
979 			  [4] = "486 SL",
980 			  [5] = "486 SX/2",
981 			  [7] = "486 DX/2-WB",
982 			  [8] = "486 DX/4",
983 			  [9] = "486 DX/4-WB"
984 		  }
985 		},
986 		{ .family = 5, .model_names =
987 		  {
988 			  [0] = "Pentium 60/66 A-step",
989 			  [1] = "Pentium 60/66",
990 			  [2] = "Pentium 75 - 200",
991 			  [3] = "OverDrive PODP5V83",
992 			  [4] = "Pentium MMX",
993 			  [7] = "Mobile Pentium 75 - 200",
994 			  [8] = "Mobile Pentium MMX",
995 			  [9] = "Quark SoC X1000",
996 		  }
997 		},
998 		{ .family = 6, .model_names =
999 		  {
1000 			  [0] = "Pentium Pro A-step",
1001 			  [1] = "Pentium Pro",
1002 			  [3] = "Pentium II (Klamath)",
1003 			  [4] = "Pentium II (Deschutes)",
1004 			  [5] = "Pentium II (Deschutes)",
1005 			  [6] = "Mobile Pentium II",
1006 			  [7] = "Pentium III (Katmai)",
1007 			  [8] = "Pentium III (Coppermine)",
1008 			  [10] = "Pentium III (Cascades)",
1009 			  [11] = "Pentium III (Tualatin)",
1010 		  }
1011 		},
1012 		{ .family = 15, .model_names =
1013 		  {
1014 			  [0] = "Pentium 4 (Unknown)",
1015 			  [1] = "Pentium 4 (Willamette)",
1016 			  [2] = "Pentium 4 (Northwood)",
1017 			  [4] = "Pentium 4 (Foster)",
1018 			  [5] = "Pentium 4 (Foster)",
1019 		  }
1020 		},
1021 	},
1022 	.legacy_cache_size = intel_size_cache,
1023 #endif
1024 	.c_detect_tlb	= intel_detect_tlb,
1025 	.c_early_init   = early_init_intel,
1026 	.c_init		= init_intel,
1027 	.c_x86_vendor	= X86_VENDOR_INTEL,
1028 };
1029 
1030 cpu_dev_register(intel_cpu_dev);
1031