xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision d2ba09c1)
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 static bool ring3mwait_disabled __read_mostly;
70 
71 static int __init ring3mwait_disable(char *__unused)
72 {
73 	ring3mwait_disabled = true;
74 	return 0;
75 }
76 __setup("ring3mwait=disable", ring3mwait_disable);
77 
78 static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c)
79 {
80 	/*
81 	 * Ring 3 MONITOR/MWAIT feature cannot be detected without
82 	 * cpu model and family comparison.
83 	 */
84 	if (c->x86 != 6)
85 		return;
86 	switch (c->x86_model) {
87 	case INTEL_FAM6_XEON_PHI_KNL:
88 	case INTEL_FAM6_XEON_PHI_KNM:
89 		break;
90 	default:
91 		return;
92 	}
93 
94 	if (ring3mwait_disabled)
95 		return;
96 
97 	set_cpu_cap(c, X86_FEATURE_RING3MWAIT);
98 	this_cpu_or(msr_misc_features_shadow,
99 		    1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT);
100 
101 	if (c == &boot_cpu_data)
102 		ELF_HWCAP2 |= HWCAP2_RING3MWAIT;
103 }
104 
105 /*
106  * Early microcode releases for the Spectre v2 mitigation were broken.
107  * Information taken from;
108  * - https://newsroom.intel.com/wp-content/uploads/sites/11/2018/03/microcode-update-guidance.pdf
109  * - https://kb.vmware.com/s/article/52345
110  * - Microcode revisions observed in the wild
111  * - Release note from 20180108 microcode release
112  */
113 struct sku_microcode {
114 	u8 model;
115 	u8 stepping;
116 	u32 microcode;
117 };
118 static const struct sku_microcode spectre_bad_microcodes[] = {
119 	{ INTEL_FAM6_KABYLAKE_DESKTOP,	0x0B,	0x80 },
120 	{ INTEL_FAM6_KABYLAKE_DESKTOP,	0x0A,	0x80 },
121 	{ INTEL_FAM6_KABYLAKE_DESKTOP,	0x09,	0x80 },
122 	{ INTEL_FAM6_KABYLAKE_MOBILE,	0x0A,	0x80 },
123 	{ INTEL_FAM6_KABYLAKE_MOBILE,	0x09,	0x80 },
124 	{ INTEL_FAM6_SKYLAKE_X,		0x03,	0x0100013e },
125 	{ INTEL_FAM6_SKYLAKE_X,		0x04,	0x0200003c },
126 	{ INTEL_FAM6_BROADWELL_CORE,	0x04,	0x28 },
127 	{ INTEL_FAM6_BROADWELL_GT3E,	0x01,	0x1b },
128 	{ INTEL_FAM6_BROADWELL_XEON_D,	0x02,	0x14 },
129 	{ INTEL_FAM6_BROADWELL_XEON_D,	0x03,	0x07000011 },
130 	{ INTEL_FAM6_BROADWELL_X,	0x01,	0x0b000025 },
131 	{ INTEL_FAM6_HASWELL_ULT,	0x01,	0x21 },
132 	{ INTEL_FAM6_HASWELL_GT3E,	0x01,	0x18 },
133 	{ INTEL_FAM6_HASWELL_CORE,	0x03,	0x23 },
134 	{ INTEL_FAM6_HASWELL_X,		0x02,	0x3b },
135 	{ INTEL_FAM6_HASWELL_X,		0x04,	0x10 },
136 	{ INTEL_FAM6_IVYBRIDGE_X,	0x04,	0x42a },
137 	/* Observed in the wild */
138 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x06,	0x61b },
139 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x07,	0x712 },
140 };
141 
142 static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
143 {
144 	int i;
145 
146 	/*
147 	 * We know that the hypervisor lie to us on the microcode version so
148 	 * we may as well hope that it is running the correct version.
149 	 */
150 	if (cpu_has(c, X86_FEATURE_HYPERVISOR))
151 		return false;
152 
153 	for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
154 		if (c->x86_model == spectre_bad_microcodes[i].model &&
155 		    c->x86_stepping == spectre_bad_microcodes[i].stepping)
156 			return (c->microcode <= spectre_bad_microcodes[i].microcode);
157 	}
158 	return false;
159 }
160 
161 static void early_init_intel(struct cpuinfo_x86 *c)
162 {
163 	u64 misc_enable;
164 
165 	/* Unmask CPUID levels if masked: */
166 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
167 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
168 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
169 			c->cpuid_level = cpuid_eax(0);
170 			get_cpu_cap(c);
171 		}
172 	}
173 
174 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
175 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
176 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
177 
178 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
179 		c->microcode = intel_get_microcode_revision();
180 
181 	/* Now if any of them are set, check the blacklist and clear the lot */
182 	if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
183 	     cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
184 	     cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
185 	     cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
186 		pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
187 		setup_clear_cpu_cap(X86_FEATURE_IBRS);
188 		setup_clear_cpu_cap(X86_FEATURE_IBPB);
189 		setup_clear_cpu_cap(X86_FEATURE_STIBP);
190 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
191 		setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
192 	}
193 
194 	/*
195 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
196 	 *
197 	 * A race condition between speculative fetches and invalidating
198 	 * a large page.  This is worked around in microcode, but we
199 	 * need the microcode to have already been loaded... so if it is
200 	 * not, recommend a BIOS update and disable large pages.
201 	 */
202 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_stepping <= 2 &&
203 	    c->microcode < 0x20e) {
204 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
205 		clear_cpu_cap(c, X86_FEATURE_PSE);
206 	}
207 
208 #ifdef CONFIG_X86_64
209 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
210 #else
211 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
212 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
213 		c->x86_cache_alignment = 128;
214 #endif
215 
216 	/* CPUID workaround for 0F33/0F34 CPU */
217 	if (c->x86 == 0xF && c->x86_model == 0x3
218 	    && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
219 		c->x86_phys_bits = 36;
220 
221 	/*
222 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
223 	 * with P/T states and does not stop in deep C-states.
224 	 *
225 	 * It is also reliable across cores and sockets. (but not across
226 	 * cabinets - we turn it off in that case explicitly.)
227 	 */
228 	if (c->x86_power & (1 << 8)) {
229 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
230 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
231 	}
232 
233 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
234 	if (c->x86 == 6) {
235 		switch (c->x86_model) {
236 		case 0x27:	/* Penwell */
237 		case 0x35:	/* Cloverview */
238 		case 0x4a:	/* Merrifield */
239 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
240 			break;
241 		default:
242 			break;
243 		}
244 	}
245 
246 	/*
247 	 * There is a known erratum on Pentium III and Core Solo
248 	 * and Core Duo CPUs.
249 	 * " Page with PAT set to WC while associated MTRR is UC
250 	 *   may consolidate to UC "
251 	 * Because of this erratum, it is better to stick with
252 	 * setting WC in MTRR rather than using PAT on these CPUs.
253 	 *
254 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
255 	 */
256 	if (c->x86 == 6 && c->x86_model < 15)
257 		clear_cpu_cap(c, X86_FEATURE_PAT);
258 
259 	/*
260 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
261 	 * clear the fast string and enhanced fast string CPU capabilities.
262 	 */
263 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
264 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
265 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
266 			pr_info("Disabled fast string operations\n");
267 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
268 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
269 		}
270 	}
271 
272 	/*
273 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
274 	 * "The operating system also is required to invalidate (i.e., flush)
275 	 *  the TLB when any changes are made to any of the page table entries.
276 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
277 	 *
278 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
279 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
280 	 * to be modified.
281 	 */
282 	if (c->x86 == 5 && c->x86_model == 9) {
283 		pr_info("Disabling PGE capability bit\n");
284 		setup_clear_cpu_cap(X86_FEATURE_PGE);
285 	}
286 
287 	if (c->cpuid_level >= 0x00000001) {
288 		u32 eax, ebx, ecx, edx;
289 
290 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
291 		/*
292 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
293 		 * apicids which are reserved per package. Store the resulting
294 		 * shift value for the package management code.
295 		 */
296 		if (edx & (1U << 28))
297 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
298 	}
299 
300 	check_mpx_erratum(c);
301 }
302 
303 #ifdef CONFIG_X86_32
304 /*
305  *	Early probe support logic for ppro memory erratum #50
306  *
307  *	This is called before we do cpu ident work
308  */
309 
310 int ppro_with_ram_bug(void)
311 {
312 	/* Uses data from early_cpu_detect now */
313 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
314 	    boot_cpu_data.x86 == 6 &&
315 	    boot_cpu_data.x86_model == 1 &&
316 	    boot_cpu_data.x86_stepping < 8) {
317 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
318 		return 1;
319 	}
320 	return 0;
321 }
322 
323 static void intel_smp_check(struct cpuinfo_x86 *c)
324 {
325 	/* calling is from identify_secondary_cpu() ? */
326 	if (!c->cpu_index)
327 		return;
328 
329 	/*
330 	 * Mask B, Pentium, but not Pentium MMX
331 	 */
332 	if (c->x86 == 5 &&
333 	    c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
334 	    c->x86_model <= 3) {
335 		/*
336 		 * Remember we have B step Pentia with bugs
337 		 */
338 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
339 				    "with B stepping processors.\n");
340 	}
341 }
342 
343 static int forcepae;
344 static int __init forcepae_setup(char *__unused)
345 {
346 	forcepae = 1;
347 	return 1;
348 }
349 __setup("forcepae", forcepae_setup);
350 
351 static void intel_workarounds(struct cpuinfo_x86 *c)
352 {
353 #ifdef CONFIG_X86_F00F_BUG
354 	/*
355 	 * All models of Pentium and Pentium with MMX technology CPUs
356 	 * have the F0 0F bug, which lets nonprivileged users lock up the
357 	 * system. Announce that the fault handler will be checking for it.
358 	 * The Quark is also family 5, but does not have the same bug.
359 	 */
360 	clear_cpu_bug(c, X86_BUG_F00F);
361 	if (c->x86 == 5 && c->x86_model < 9) {
362 		static int f00f_workaround_enabled;
363 
364 		set_cpu_bug(c, X86_BUG_F00F);
365 		if (!f00f_workaround_enabled) {
366 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
367 			f00f_workaround_enabled = 1;
368 		}
369 	}
370 #endif
371 
372 	/*
373 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
374 	 * model 3 mask 3
375 	 */
376 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
377 		clear_cpu_cap(c, X86_FEATURE_SEP);
378 
379 	/*
380 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
381 	 * functionally usable PAE implementation.
382 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
383 	 */
384 	if (forcepae) {
385 		pr_warn("PAE forced!\n");
386 		set_cpu_cap(c, X86_FEATURE_PAE);
387 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
388 	}
389 
390 	/*
391 	 * P4 Xeon erratum 037 workaround.
392 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
393 	 */
394 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
395 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
396 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
397 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
398 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
399 		}
400 	}
401 
402 	/*
403 	 * See if we have a good local APIC by checking for buggy Pentia,
404 	 * i.e. all B steppings and the C2 stepping of P54C when using their
405 	 * integrated APIC (see 11AP erratum in "Pentium Processor
406 	 * Specification Update").
407 	 */
408 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
409 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
410 		set_cpu_bug(c, X86_BUG_11AP);
411 
412 
413 #ifdef CONFIG_X86_INTEL_USERCOPY
414 	/*
415 	 * Set up the preferred alignment for movsl bulk memory moves
416 	 */
417 	switch (c->x86) {
418 	case 4:		/* 486: untested */
419 		break;
420 	case 5:		/* Old Pentia: untested */
421 		break;
422 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
423 		movsl_mask.mask = 7;
424 		break;
425 	case 15:	/* P4 is OK down to 8-byte alignment */
426 		movsl_mask.mask = 7;
427 		break;
428 	}
429 #endif
430 
431 	intel_smp_check(c);
432 }
433 #else
434 static void intel_workarounds(struct cpuinfo_x86 *c)
435 {
436 }
437 #endif
438 
439 static void srat_detect_node(struct cpuinfo_x86 *c)
440 {
441 #ifdef CONFIG_NUMA
442 	unsigned node;
443 	int cpu = smp_processor_id();
444 
445 	/* Don't do the funky fallback heuristics the AMD version employs
446 	   for now. */
447 	node = numa_cpu_node(cpu);
448 	if (node == NUMA_NO_NODE || !node_online(node)) {
449 		/* reuse the value from init_cpu_to_node() */
450 		node = cpu_to_node(cpu);
451 	}
452 	numa_set_node(cpu, node);
453 #endif
454 }
455 
456 /*
457  * find out the number of processor cores on the die
458  */
459 static int intel_num_cpu_cores(struct cpuinfo_x86 *c)
460 {
461 	unsigned int eax, ebx, ecx, edx;
462 
463 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
464 		return 1;
465 
466 	/* Intel has a non-standard dependency on %ecx for this CPUID level. */
467 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
468 	if (eax & 0x1f)
469 		return (eax >> 26) + 1;
470 	else
471 		return 1;
472 }
473 
474 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
475 {
476 	/* Intel VMX MSR indicated features */
477 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
478 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
479 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
480 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
481 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
482 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
483 
484 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
485 
486 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
487 	clear_cpu_cap(c, X86_FEATURE_VNMI);
488 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
489 	clear_cpu_cap(c, X86_FEATURE_EPT);
490 	clear_cpu_cap(c, X86_FEATURE_VPID);
491 
492 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
493 	msr_ctl = vmx_msr_high | vmx_msr_low;
494 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
495 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
496 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
497 		set_cpu_cap(c, X86_FEATURE_VNMI);
498 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
499 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
500 		      vmx_msr_low, vmx_msr_high);
501 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
502 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
503 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
504 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
505 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
506 			set_cpu_cap(c, X86_FEATURE_EPT);
507 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
508 			set_cpu_cap(c, X86_FEATURE_VPID);
509 	}
510 }
511 
512 #define MSR_IA32_TME_ACTIVATE		0x982
513 
514 /* Helpers to access TME_ACTIVATE MSR */
515 #define TME_ACTIVATE_LOCKED(x)		(x & 0x1)
516 #define TME_ACTIVATE_ENABLED(x)		(x & 0x2)
517 
518 #define TME_ACTIVATE_POLICY(x)		((x >> 4) & 0xf)	/* Bits 7:4 */
519 #define TME_ACTIVATE_POLICY_AES_XTS_128	0
520 
521 #define TME_ACTIVATE_KEYID_BITS(x)	((x >> 32) & 0xf)	/* Bits 35:32 */
522 
523 #define TME_ACTIVATE_CRYPTO_ALGS(x)	((x >> 48) & 0xffff)	/* Bits 63:48 */
524 #define TME_ACTIVATE_CRYPTO_AES_XTS_128	1
525 
526 /* Values for mktme_status (SW only construct) */
527 #define MKTME_ENABLED			0
528 #define MKTME_DISABLED			1
529 #define MKTME_UNINITIALIZED		2
530 static int mktme_status = MKTME_UNINITIALIZED;
531 
532 static void detect_tme(struct cpuinfo_x86 *c)
533 {
534 	u64 tme_activate, tme_policy, tme_crypto_algs;
535 	int keyid_bits = 0, nr_keyids = 0;
536 	static u64 tme_activate_cpu0 = 0;
537 
538 	rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
539 
540 	if (mktme_status != MKTME_UNINITIALIZED) {
541 		if (tme_activate != tme_activate_cpu0) {
542 			/* Broken BIOS? */
543 			pr_err_once("x86/tme: configuration is inconsistent between CPUs\n");
544 			pr_err_once("x86/tme: MKTME is not usable\n");
545 			mktme_status = MKTME_DISABLED;
546 
547 			/* Proceed. We may need to exclude bits from x86_phys_bits. */
548 		}
549 	} else {
550 		tme_activate_cpu0 = tme_activate;
551 	}
552 
553 	if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
554 		pr_info_once("x86/tme: not enabled by BIOS\n");
555 		mktme_status = MKTME_DISABLED;
556 		return;
557 	}
558 
559 	if (mktme_status != MKTME_UNINITIALIZED)
560 		goto detect_keyid_bits;
561 
562 	pr_info("x86/tme: enabled by BIOS\n");
563 
564 	tme_policy = TME_ACTIVATE_POLICY(tme_activate);
565 	if (tme_policy != TME_ACTIVATE_POLICY_AES_XTS_128)
566 		pr_warn("x86/tme: Unknown policy is active: %#llx\n", tme_policy);
567 
568 	tme_crypto_algs = TME_ACTIVATE_CRYPTO_ALGS(tme_activate);
569 	if (!(tme_crypto_algs & TME_ACTIVATE_CRYPTO_AES_XTS_128)) {
570 		pr_err("x86/mktme: No known encryption algorithm is supported: %#llx\n",
571 				tme_crypto_algs);
572 		mktme_status = MKTME_DISABLED;
573 	}
574 detect_keyid_bits:
575 	keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
576 	nr_keyids = (1UL << keyid_bits) - 1;
577 	if (nr_keyids) {
578 		pr_info_once("x86/mktme: enabled by BIOS\n");
579 		pr_info_once("x86/mktme: %d KeyIDs available\n", nr_keyids);
580 	} else {
581 		pr_info_once("x86/mktme: disabled by BIOS\n");
582 	}
583 
584 	if (mktme_status == MKTME_UNINITIALIZED) {
585 		/* MKTME is usable */
586 		mktme_status = MKTME_ENABLED;
587 	}
588 
589 	/*
590 	 * KeyID bits effectively lower the number of physical address
591 	 * bits.  Update cpuinfo_x86::x86_phys_bits accordingly.
592 	 */
593 	c->x86_phys_bits -= keyid_bits;
594 }
595 
596 static void init_intel_energy_perf(struct cpuinfo_x86 *c)
597 {
598 	u64 epb;
599 
600 	/*
601 	 * Initialize MSR_IA32_ENERGY_PERF_BIAS if not already initialized.
602 	 * (x86_energy_perf_policy(8) is available to change it at run-time.)
603 	 */
604 	if (!cpu_has(c, X86_FEATURE_EPB))
605 		return;
606 
607 	rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
608 	if ((epb & 0xF) != ENERGY_PERF_BIAS_PERFORMANCE)
609 		return;
610 
611 	pr_warn_once("ENERGY_PERF_BIAS: Set to 'normal', was 'performance'\n");
612 	pr_warn_once("ENERGY_PERF_BIAS: View and update with x86_energy_perf_policy(8)\n");
613 	epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
614 	wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
615 }
616 
617 static void intel_bsp_resume(struct cpuinfo_x86 *c)
618 {
619 	/*
620 	 * MSR_IA32_ENERGY_PERF_BIAS is lost across suspend/resume,
621 	 * so reinitialize it properly like during bootup:
622 	 */
623 	init_intel_energy_perf(c);
624 }
625 
626 static void init_cpuid_fault(struct cpuinfo_x86 *c)
627 {
628 	u64 msr;
629 
630 	if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
631 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
632 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
633 	}
634 }
635 
636 static void init_intel_misc_features(struct cpuinfo_x86 *c)
637 {
638 	u64 msr;
639 
640 	if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
641 		return;
642 
643 	/* Clear all MISC features */
644 	this_cpu_write(msr_misc_features_shadow, 0);
645 
646 	/* Check features and update capabilities and shadow control bits */
647 	init_cpuid_fault(c);
648 	probe_xeon_phi_r3mwait(c);
649 
650 	msr = this_cpu_read(msr_misc_features_shadow);
651 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
652 }
653 
654 static void init_intel(struct cpuinfo_x86 *c)
655 {
656 	unsigned int l2 = 0;
657 
658 	early_init_intel(c);
659 
660 	intel_workarounds(c);
661 
662 	/*
663 	 * Detect the extended topology information if available. This
664 	 * will reinitialise the initial_apicid which will be used
665 	 * in init_intel_cacheinfo()
666 	 */
667 	detect_extended_topology(c);
668 
669 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
670 		/*
671 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
672 		 * detection.
673 		 */
674 		c->x86_max_cores = intel_num_cpu_cores(c);
675 #ifdef CONFIG_X86_32
676 		detect_ht(c);
677 #endif
678 	}
679 
680 	l2 = init_intel_cacheinfo(c);
681 
682 	/* Detect legacy cache sizes if init_intel_cacheinfo did not */
683 	if (l2 == 0) {
684 		cpu_detect_cache_sizes(c);
685 		l2 = c->x86_cache_size;
686 	}
687 
688 	if (c->cpuid_level > 9) {
689 		unsigned eax = cpuid_eax(10);
690 		/* Check for version and the number of counters */
691 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
692 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
693 	}
694 
695 	if (cpu_has(c, X86_FEATURE_XMM2))
696 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
697 
698 	if (boot_cpu_has(X86_FEATURE_DS)) {
699 		unsigned int l1;
700 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
701 		if (!(l1 & (1<<11)))
702 			set_cpu_cap(c, X86_FEATURE_BTS);
703 		if (!(l1 & (1<<12)))
704 			set_cpu_cap(c, X86_FEATURE_PEBS);
705 	}
706 
707 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
708 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
709 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
710 
711 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
712 		((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
713 		set_cpu_bug(c, X86_BUG_MONITOR);
714 
715 #ifdef CONFIG_X86_64
716 	if (c->x86 == 15)
717 		c->x86_cache_alignment = c->x86_clflush_size * 2;
718 	if (c->x86 == 6)
719 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
720 #else
721 	/*
722 	 * Names for the Pentium II/Celeron processors
723 	 * detectable only by also checking the cache size.
724 	 * Dixon is NOT a Celeron.
725 	 */
726 	if (c->x86 == 6) {
727 		char *p = NULL;
728 
729 		switch (c->x86_model) {
730 		case 5:
731 			if (l2 == 0)
732 				p = "Celeron (Covington)";
733 			else if (l2 == 256)
734 				p = "Mobile Pentium II (Dixon)";
735 			break;
736 
737 		case 6:
738 			if (l2 == 128)
739 				p = "Celeron (Mendocino)";
740 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
741 				p = "Celeron-A";
742 			break;
743 
744 		case 8:
745 			if (l2 == 128)
746 				p = "Celeron (Coppermine)";
747 			break;
748 		}
749 
750 		if (p)
751 			strcpy(c->x86_model_id, p);
752 	}
753 
754 	if (c->x86 == 15)
755 		set_cpu_cap(c, X86_FEATURE_P4);
756 	if (c->x86 == 6)
757 		set_cpu_cap(c, X86_FEATURE_P3);
758 #endif
759 
760 	/* Work around errata */
761 	srat_detect_node(c);
762 
763 	if (cpu_has(c, X86_FEATURE_VMX))
764 		detect_vmx_virtcap(c);
765 
766 	if (cpu_has(c, X86_FEATURE_TME))
767 		detect_tme(c);
768 
769 	init_intel_energy_perf(c);
770 
771 	init_intel_misc_features(c);
772 }
773 
774 #ifdef CONFIG_X86_32
775 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
776 {
777 	/*
778 	 * Intel PIII Tualatin. This comes in two flavours.
779 	 * One has 256kb of cache, the other 512. We have no way
780 	 * to determine which, so we use a boottime override
781 	 * for the 512kb model, and assume 256 otherwise.
782 	 */
783 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
784 		size = 256;
785 
786 	/*
787 	 * Intel Quark SoC X1000 contains a 4-way set associative
788 	 * 16K cache with a 16 byte cache line and 256 lines per tag
789 	 */
790 	if ((c->x86 == 5) && (c->x86_model == 9))
791 		size = 16;
792 	return size;
793 }
794 #endif
795 
796 #define TLB_INST_4K	0x01
797 #define TLB_INST_4M	0x02
798 #define TLB_INST_2M_4M	0x03
799 
800 #define TLB_INST_ALL	0x05
801 #define TLB_INST_1G	0x06
802 
803 #define TLB_DATA_4K	0x11
804 #define TLB_DATA_4M	0x12
805 #define TLB_DATA_2M_4M	0x13
806 #define TLB_DATA_4K_4M	0x14
807 
808 #define TLB_DATA_1G	0x16
809 
810 #define TLB_DATA0_4K	0x21
811 #define TLB_DATA0_4M	0x22
812 #define TLB_DATA0_2M_4M	0x23
813 
814 #define STLB_4K		0x41
815 #define STLB_4K_2M	0x42
816 
817 static const struct _tlb_table intel_tlb_table[] = {
818 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
819 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
820 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
821 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
822 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
823 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
824 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages */" },
825 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
826 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
827 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
828 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
829 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
830 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
831 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
832 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
833 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
834 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
835 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
836 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
837 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
838 	{ 0x6b, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 8-way associative" },
839 	{ 0x6c, TLB_DATA_2M_4M,		128,	" TLB_DATA 2 MByte or 4 MByte pages, 8-way associative" },
840 	{ 0x6d, TLB_DATA_1G,		16,	" TLB_DATA 1 GByte pages, fully associative" },
841 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
842 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
843 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
844 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
845 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
846 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
847 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
848 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
849 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
850 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
851 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
852 	{ 0xc2, TLB_DATA_2M_4M,		16,	" DTLB 2 MByte/4MByte pages, 4-way associative" },
853 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
854 	{ 0x00, 0, 0 }
855 };
856 
857 static void intel_tlb_lookup(const unsigned char desc)
858 {
859 	unsigned char k;
860 	if (desc == 0)
861 		return;
862 
863 	/* look up this descriptor in the table */
864 	for (k = 0; intel_tlb_table[k].descriptor != desc && \
865 			intel_tlb_table[k].descriptor != 0; k++)
866 		;
867 
868 	if (intel_tlb_table[k].tlb_type == 0)
869 		return;
870 
871 	switch (intel_tlb_table[k].tlb_type) {
872 	case STLB_4K:
873 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
874 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
875 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
876 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
877 		break;
878 	case STLB_4K_2M:
879 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
880 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
881 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
882 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
883 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
884 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
885 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
886 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
887 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
888 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
889 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
890 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
891 		break;
892 	case TLB_INST_ALL:
893 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
894 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
895 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
896 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
897 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
898 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
899 		break;
900 	case TLB_INST_4K:
901 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
902 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
903 		break;
904 	case TLB_INST_4M:
905 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
906 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
907 		break;
908 	case TLB_INST_2M_4M:
909 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
910 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
911 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
912 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
913 		break;
914 	case TLB_DATA_4K:
915 	case TLB_DATA0_4K:
916 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
917 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
918 		break;
919 	case TLB_DATA_4M:
920 	case TLB_DATA0_4M:
921 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
922 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
923 		break;
924 	case TLB_DATA_2M_4M:
925 	case TLB_DATA0_2M_4M:
926 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
927 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
928 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
929 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
930 		break;
931 	case TLB_DATA_4K_4M:
932 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
933 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
934 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
935 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
936 		break;
937 	case TLB_DATA_1G:
938 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
939 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
940 		break;
941 	}
942 }
943 
944 static void intel_detect_tlb(struct cpuinfo_x86 *c)
945 {
946 	int i, j, n;
947 	unsigned int regs[4];
948 	unsigned char *desc = (unsigned char *)regs;
949 
950 	if (c->cpuid_level < 2)
951 		return;
952 
953 	/* Number of times to iterate */
954 	n = cpuid_eax(2) & 0xFF;
955 
956 	for (i = 0 ; i < n ; i++) {
957 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
958 
959 		/* If bit 31 is set, this is an unknown format */
960 		for (j = 0 ; j < 3 ; j++)
961 			if (regs[j] & (1 << 31))
962 				regs[j] = 0;
963 
964 		/* Byte 0 is level count, not a descriptor */
965 		for (j = 1 ; j < 16 ; j++)
966 			intel_tlb_lookup(desc[j]);
967 	}
968 }
969 
970 static const struct cpu_dev intel_cpu_dev = {
971 	.c_vendor	= "Intel",
972 	.c_ident	= { "GenuineIntel" },
973 #ifdef CONFIG_X86_32
974 	.legacy_models = {
975 		{ .family = 4, .model_names =
976 		  {
977 			  [0] = "486 DX-25/33",
978 			  [1] = "486 DX-50",
979 			  [2] = "486 SX",
980 			  [3] = "486 DX/2",
981 			  [4] = "486 SL",
982 			  [5] = "486 SX/2",
983 			  [7] = "486 DX/2-WB",
984 			  [8] = "486 DX/4",
985 			  [9] = "486 DX/4-WB"
986 		  }
987 		},
988 		{ .family = 5, .model_names =
989 		  {
990 			  [0] = "Pentium 60/66 A-step",
991 			  [1] = "Pentium 60/66",
992 			  [2] = "Pentium 75 - 200",
993 			  [3] = "OverDrive PODP5V83",
994 			  [4] = "Pentium MMX",
995 			  [7] = "Mobile Pentium 75 - 200",
996 			  [8] = "Mobile Pentium MMX",
997 			  [9] = "Quark SoC X1000",
998 		  }
999 		},
1000 		{ .family = 6, .model_names =
1001 		  {
1002 			  [0] = "Pentium Pro A-step",
1003 			  [1] = "Pentium Pro",
1004 			  [3] = "Pentium II (Klamath)",
1005 			  [4] = "Pentium II (Deschutes)",
1006 			  [5] = "Pentium II (Deschutes)",
1007 			  [6] = "Mobile Pentium II",
1008 			  [7] = "Pentium III (Katmai)",
1009 			  [8] = "Pentium III (Coppermine)",
1010 			  [10] = "Pentium III (Cascades)",
1011 			  [11] = "Pentium III (Tualatin)",
1012 		  }
1013 		},
1014 		{ .family = 15, .model_names =
1015 		  {
1016 			  [0] = "Pentium 4 (Unknown)",
1017 			  [1] = "Pentium 4 (Willamette)",
1018 			  [2] = "Pentium 4 (Northwood)",
1019 			  [4] = "Pentium 4 (Foster)",
1020 			  [5] = "Pentium 4 (Foster)",
1021 		  }
1022 		},
1023 	},
1024 	.legacy_cache_size = intel_size_cache,
1025 #endif
1026 	.c_detect_tlb	= intel_detect_tlb,
1027 	.c_early_init   = early_init_intel,
1028 	.c_init		= init_intel,
1029 	.c_bsp_resume	= intel_bsp_resume,
1030 	.c_x86_vendor	= X86_VENDOR_INTEL,
1031 };
1032 
1033 cpu_dev_register(intel_cpu_dev);
1034 
1035