xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision 6d99a79c)
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 	if (c->x86 != 6)
154 		return false;
155 
156 	for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
157 		if (c->x86_model == spectre_bad_microcodes[i].model &&
158 		    c->x86_stepping == spectre_bad_microcodes[i].stepping)
159 			return (c->microcode <= spectre_bad_microcodes[i].microcode);
160 	}
161 	return false;
162 }
163 
164 static void early_init_intel(struct cpuinfo_x86 *c)
165 {
166 	u64 misc_enable;
167 
168 	/* Unmask CPUID levels if masked: */
169 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
170 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
171 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
172 			c->cpuid_level = cpuid_eax(0);
173 			get_cpu_cap(c);
174 		}
175 	}
176 
177 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
178 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
179 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
180 
181 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
182 		c->microcode = intel_get_microcode_revision();
183 
184 	/* Now if any of them are set, check the blacklist and clear the lot */
185 	if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
186 	     cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
187 	     cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
188 	     cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
189 		pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
190 		setup_clear_cpu_cap(X86_FEATURE_IBRS);
191 		setup_clear_cpu_cap(X86_FEATURE_IBPB);
192 		setup_clear_cpu_cap(X86_FEATURE_STIBP);
193 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
194 		setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL);
195 		setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
196 		setup_clear_cpu_cap(X86_FEATURE_SSBD);
197 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD);
198 	}
199 
200 	/*
201 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
202 	 *
203 	 * A race condition between speculative fetches and invalidating
204 	 * a large page.  This is worked around in microcode, but we
205 	 * need the microcode to have already been loaded... so if it is
206 	 * not, recommend a BIOS update and disable large pages.
207 	 */
208 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_stepping <= 2 &&
209 	    c->microcode < 0x20e) {
210 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
211 		clear_cpu_cap(c, X86_FEATURE_PSE);
212 	}
213 
214 #ifdef CONFIG_X86_64
215 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
216 #else
217 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
218 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
219 		c->x86_cache_alignment = 128;
220 #endif
221 
222 	/* CPUID workaround for 0F33/0F34 CPU */
223 	if (c->x86 == 0xF && c->x86_model == 0x3
224 	    && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
225 		c->x86_phys_bits = 36;
226 
227 	/*
228 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
229 	 * with P/T states and does not stop in deep C-states.
230 	 *
231 	 * It is also reliable across cores and sockets. (but not across
232 	 * cabinets - we turn it off in that case explicitly.)
233 	 */
234 	if (c->x86_power & (1 << 8)) {
235 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
236 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
237 	}
238 
239 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
240 	if (c->x86 == 6) {
241 		switch (c->x86_model) {
242 		case 0x27:	/* Penwell */
243 		case 0x35:	/* Cloverview */
244 		case 0x4a:	/* Merrifield */
245 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
246 			break;
247 		default:
248 			break;
249 		}
250 	}
251 
252 	/*
253 	 * There is a known erratum on Pentium III and Core Solo
254 	 * and Core Duo CPUs.
255 	 * " Page with PAT set to WC while associated MTRR is UC
256 	 *   may consolidate to UC "
257 	 * Because of this erratum, it is better to stick with
258 	 * setting WC in MTRR rather than using PAT on these CPUs.
259 	 *
260 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
261 	 */
262 	if (c->x86 == 6 && c->x86_model < 15)
263 		clear_cpu_cap(c, X86_FEATURE_PAT);
264 
265 	/*
266 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
267 	 * clear the fast string and enhanced fast string CPU capabilities.
268 	 */
269 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
270 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
271 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
272 			pr_info("Disabled fast string operations\n");
273 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
274 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
275 		}
276 	}
277 
278 	/*
279 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
280 	 * "The operating system also is required to invalidate (i.e., flush)
281 	 *  the TLB when any changes are made to any of the page table entries.
282 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
283 	 *
284 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
285 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
286 	 * to be modified.
287 	 */
288 	if (c->x86 == 5 && c->x86_model == 9) {
289 		pr_info("Disabling PGE capability bit\n");
290 		setup_clear_cpu_cap(X86_FEATURE_PGE);
291 	}
292 
293 	if (c->cpuid_level >= 0x00000001) {
294 		u32 eax, ebx, ecx, edx;
295 
296 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
297 		/*
298 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
299 		 * apicids which are reserved per package. Store the resulting
300 		 * shift value for the package management code.
301 		 */
302 		if (edx & (1U << 28))
303 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
304 	}
305 
306 	check_mpx_erratum(c);
307 
308 	/*
309 	 * Get the number of SMT siblings early from the extended topology
310 	 * leaf, if available. Otherwise try the legacy SMT detection.
311 	 */
312 	if (detect_extended_topology_early(c) < 0)
313 		detect_ht_early(c);
314 }
315 
316 #ifdef CONFIG_X86_32
317 /*
318  *	Early probe support logic for ppro memory erratum #50
319  *
320  *	This is called before we do cpu ident work
321  */
322 
323 int ppro_with_ram_bug(void)
324 {
325 	/* Uses data from early_cpu_detect now */
326 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
327 	    boot_cpu_data.x86 == 6 &&
328 	    boot_cpu_data.x86_model == 1 &&
329 	    boot_cpu_data.x86_stepping < 8) {
330 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
331 		return 1;
332 	}
333 	return 0;
334 }
335 
336 static void intel_smp_check(struct cpuinfo_x86 *c)
337 {
338 	/* calling is from identify_secondary_cpu() ? */
339 	if (!c->cpu_index)
340 		return;
341 
342 	/*
343 	 * Mask B, Pentium, but not Pentium MMX
344 	 */
345 	if (c->x86 == 5 &&
346 	    c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
347 	    c->x86_model <= 3) {
348 		/*
349 		 * Remember we have B step Pentia with bugs
350 		 */
351 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
352 				    "with B stepping processors.\n");
353 	}
354 }
355 
356 static int forcepae;
357 static int __init forcepae_setup(char *__unused)
358 {
359 	forcepae = 1;
360 	return 1;
361 }
362 __setup("forcepae", forcepae_setup);
363 
364 static void intel_workarounds(struct cpuinfo_x86 *c)
365 {
366 #ifdef CONFIG_X86_F00F_BUG
367 	/*
368 	 * All models of Pentium and Pentium with MMX technology CPUs
369 	 * have the F0 0F bug, which lets nonprivileged users lock up the
370 	 * system. Announce that the fault handler will be checking for it.
371 	 * The Quark is also family 5, but does not have the same bug.
372 	 */
373 	clear_cpu_bug(c, X86_BUG_F00F);
374 	if (c->x86 == 5 && c->x86_model < 9) {
375 		static int f00f_workaround_enabled;
376 
377 		set_cpu_bug(c, X86_BUG_F00F);
378 		if (!f00f_workaround_enabled) {
379 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
380 			f00f_workaround_enabled = 1;
381 		}
382 	}
383 #endif
384 
385 	/*
386 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
387 	 * model 3 mask 3
388 	 */
389 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
390 		clear_cpu_cap(c, X86_FEATURE_SEP);
391 
392 	/*
393 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
394 	 * functionally usable PAE implementation.
395 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
396 	 */
397 	if (forcepae) {
398 		pr_warn("PAE forced!\n");
399 		set_cpu_cap(c, X86_FEATURE_PAE);
400 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
401 	}
402 
403 	/*
404 	 * P4 Xeon erratum 037 workaround.
405 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
406 	 */
407 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
408 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
409 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
410 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
411 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
412 		}
413 	}
414 
415 	/*
416 	 * See if we have a good local APIC by checking for buggy Pentia,
417 	 * i.e. all B steppings and the C2 stepping of P54C when using their
418 	 * integrated APIC (see 11AP erratum in "Pentium Processor
419 	 * Specification Update").
420 	 */
421 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
422 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
423 		set_cpu_bug(c, X86_BUG_11AP);
424 
425 
426 #ifdef CONFIG_X86_INTEL_USERCOPY
427 	/*
428 	 * Set up the preferred alignment for movsl bulk memory moves
429 	 */
430 	switch (c->x86) {
431 	case 4:		/* 486: untested */
432 		break;
433 	case 5:		/* Old Pentia: untested */
434 		break;
435 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
436 		movsl_mask.mask = 7;
437 		break;
438 	case 15:	/* P4 is OK down to 8-byte alignment */
439 		movsl_mask.mask = 7;
440 		break;
441 	}
442 #endif
443 
444 	intel_smp_check(c);
445 }
446 #else
447 static void intel_workarounds(struct cpuinfo_x86 *c)
448 {
449 }
450 #endif
451 
452 static void srat_detect_node(struct cpuinfo_x86 *c)
453 {
454 #ifdef CONFIG_NUMA
455 	unsigned node;
456 	int cpu = smp_processor_id();
457 
458 	/* Don't do the funky fallback heuristics the AMD version employs
459 	   for now. */
460 	node = numa_cpu_node(cpu);
461 	if (node == NUMA_NO_NODE || !node_online(node)) {
462 		/* reuse the value from init_cpu_to_node() */
463 		node = cpu_to_node(cpu);
464 	}
465 	numa_set_node(cpu, node);
466 #endif
467 }
468 
469 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
470 {
471 	/* Intel VMX MSR indicated features */
472 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
473 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
474 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
475 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
476 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
477 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
478 #define x86_VMX_FEATURE_EPT_CAP_AD		0x00200000
479 
480 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
481 	u32 msr_vpid_cap, msr_ept_cap;
482 
483 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
484 	clear_cpu_cap(c, X86_FEATURE_VNMI);
485 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
486 	clear_cpu_cap(c, X86_FEATURE_EPT);
487 	clear_cpu_cap(c, X86_FEATURE_VPID);
488 	clear_cpu_cap(c, X86_FEATURE_EPT_AD);
489 
490 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
491 	msr_ctl = vmx_msr_high | vmx_msr_low;
492 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
493 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
494 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
495 		set_cpu_cap(c, X86_FEATURE_VNMI);
496 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
497 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
498 		      vmx_msr_low, vmx_msr_high);
499 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
500 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
501 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
502 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
503 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT) {
504 			set_cpu_cap(c, X86_FEATURE_EPT);
505 			rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
506 			      msr_ept_cap, msr_vpid_cap);
507 			if (msr_ept_cap & x86_VMX_FEATURE_EPT_CAP_AD)
508 				set_cpu_cap(c, X86_FEATURE_EPT_AD);
509 		}
510 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
511 			set_cpu_cap(c, X86_FEATURE_VPID);
512 	}
513 }
514 
515 #define MSR_IA32_TME_ACTIVATE		0x982
516 
517 /* Helpers to access TME_ACTIVATE MSR */
518 #define TME_ACTIVATE_LOCKED(x)		(x & 0x1)
519 #define TME_ACTIVATE_ENABLED(x)		(x & 0x2)
520 
521 #define TME_ACTIVATE_POLICY(x)		((x >> 4) & 0xf)	/* Bits 7:4 */
522 #define TME_ACTIVATE_POLICY_AES_XTS_128	0
523 
524 #define TME_ACTIVATE_KEYID_BITS(x)	((x >> 32) & 0xf)	/* Bits 35:32 */
525 
526 #define TME_ACTIVATE_CRYPTO_ALGS(x)	((x >> 48) & 0xffff)	/* Bits 63:48 */
527 #define TME_ACTIVATE_CRYPTO_AES_XTS_128	1
528 
529 /* Values for mktme_status (SW only construct) */
530 #define MKTME_ENABLED			0
531 #define MKTME_DISABLED			1
532 #define MKTME_UNINITIALIZED		2
533 static int mktme_status = MKTME_UNINITIALIZED;
534 
535 static void detect_tme(struct cpuinfo_x86 *c)
536 {
537 	u64 tme_activate, tme_policy, tme_crypto_algs;
538 	int keyid_bits = 0, nr_keyids = 0;
539 	static u64 tme_activate_cpu0 = 0;
540 
541 	rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
542 
543 	if (mktme_status != MKTME_UNINITIALIZED) {
544 		if (tme_activate != tme_activate_cpu0) {
545 			/* Broken BIOS? */
546 			pr_err_once("x86/tme: configuration is inconsistent between CPUs\n");
547 			pr_err_once("x86/tme: MKTME is not usable\n");
548 			mktme_status = MKTME_DISABLED;
549 
550 			/* Proceed. We may need to exclude bits from x86_phys_bits. */
551 		}
552 	} else {
553 		tme_activate_cpu0 = tme_activate;
554 	}
555 
556 	if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
557 		pr_info_once("x86/tme: not enabled by BIOS\n");
558 		mktme_status = MKTME_DISABLED;
559 		return;
560 	}
561 
562 	if (mktme_status != MKTME_UNINITIALIZED)
563 		goto detect_keyid_bits;
564 
565 	pr_info("x86/tme: enabled by BIOS\n");
566 
567 	tme_policy = TME_ACTIVATE_POLICY(tme_activate);
568 	if (tme_policy != TME_ACTIVATE_POLICY_AES_XTS_128)
569 		pr_warn("x86/tme: Unknown policy is active: %#llx\n", tme_policy);
570 
571 	tme_crypto_algs = TME_ACTIVATE_CRYPTO_ALGS(tme_activate);
572 	if (!(tme_crypto_algs & TME_ACTIVATE_CRYPTO_AES_XTS_128)) {
573 		pr_err("x86/mktme: No known encryption algorithm is supported: %#llx\n",
574 				tme_crypto_algs);
575 		mktme_status = MKTME_DISABLED;
576 	}
577 detect_keyid_bits:
578 	keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
579 	nr_keyids = (1UL << keyid_bits) - 1;
580 	if (nr_keyids) {
581 		pr_info_once("x86/mktme: enabled by BIOS\n");
582 		pr_info_once("x86/mktme: %d KeyIDs available\n", nr_keyids);
583 	} else {
584 		pr_info_once("x86/mktme: disabled by BIOS\n");
585 	}
586 
587 	if (mktme_status == MKTME_UNINITIALIZED) {
588 		/* MKTME is usable */
589 		mktme_status = MKTME_ENABLED;
590 	}
591 
592 	/*
593 	 * KeyID bits effectively lower the number of physical address
594 	 * bits.  Update cpuinfo_x86::x86_phys_bits accordingly.
595 	 */
596 	c->x86_phys_bits -= keyid_bits;
597 }
598 
599 static void init_intel_energy_perf(struct cpuinfo_x86 *c)
600 {
601 	u64 epb;
602 
603 	/*
604 	 * Initialize MSR_IA32_ENERGY_PERF_BIAS if not already initialized.
605 	 * (x86_energy_perf_policy(8) is available to change it at run-time.)
606 	 */
607 	if (!cpu_has(c, X86_FEATURE_EPB))
608 		return;
609 
610 	rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
611 	if ((epb & 0xF) != ENERGY_PERF_BIAS_PERFORMANCE)
612 		return;
613 
614 	pr_warn_once("ENERGY_PERF_BIAS: Set to 'normal', was 'performance'\n");
615 	pr_warn_once("ENERGY_PERF_BIAS: View and update with x86_energy_perf_policy(8)\n");
616 	epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
617 	wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
618 }
619 
620 static void intel_bsp_resume(struct cpuinfo_x86 *c)
621 {
622 	/*
623 	 * MSR_IA32_ENERGY_PERF_BIAS is lost across suspend/resume,
624 	 * so reinitialize it properly like during bootup:
625 	 */
626 	init_intel_energy_perf(c);
627 }
628 
629 static void init_cpuid_fault(struct cpuinfo_x86 *c)
630 {
631 	u64 msr;
632 
633 	if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
634 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
635 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
636 	}
637 }
638 
639 static void init_intel_misc_features(struct cpuinfo_x86 *c)
640 {
641 	u64 msr;
642 
643 	if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
644 		return;
645 
646 	/* Clear all MISC features */
647 	this_cpu_write(msr_misc_features_shadow, 0);
648 
649 	/* Check features and update capabilities and shadow control bits */
650 	init_cpuid_fault(c);
651 	probe_xeon_phi_r3mwait(c);
652 
653 	msr = this_cpu_read(msr_misc_features_shadow);
654 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
655 }
656 
657 static void init_intel(struct cpuinfo_x86 *c)
658 {
659 	early_init_intel(c);
660 
661 	intel_workarounds(c);
662 
663 	/*
664 	 * Detect the extended topology information if available. This
665 	 * will reinitialise the initial_apicid which will be used
666 	 * in init_intel_cacheinfo()
667 	 */
668 	detect_extended_topology(c);
669 
670 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
671 		/*
672 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
673 		 * detection.
674 		 */
675 		detect_num_cpu_cores(c);
676 #ifdef CONFIG_X86_32
677 		detect_ht(c);
678 #endif
679 	}
680 
681 	init_intel_cacheinfo(c);
682 
683 	if (c->cpuid_level > 9) {
684 		unsigned eax = cpuid_eax(10);
685 		/* Check for version and the number of counters */
686 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
687 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
688 	}
689 
690 	if (cpu_has(c, X86_FEATURE_XMM2))
691 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
692 
693 	if (boot_cpu_has(X86_FEATURE_DS)) {
694 		unsigned int l1, l2;
695 
696 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
697 		if (!(l1 & (1<<11)))
698 			set_cpu_cap(c, X86_FEATURE_BTS);
699 		if (!(l1 & (1<<12)))
700 			set_cpu_cap(c, X86_FEATURE_PEBS);
701 	}
702 
703 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
704 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
705 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
706 
707 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
708 		((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
709 		set_cpu_bug(c, X86_BUG_MONITOR);
710 
711 #ifdef CONFIG_X86_64
712 	if (c->x86 == 15)
713 		c->x86_cache_alignment = c->x86_clflush_size * 2;
714 	if (c->x86 == 6)
715 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
716 #else
717 	/*
718 	 * Names for the Pentium II/Celeron processors
719 	 * detectable only by also checking the cache size.
720 	 * Dixon is NOT a Celeron.
721 	 */
722 	if (c->x86 == 6) {
723 		unsigned int l2 = c->x86_cache_size;
724 		char *p = NULL;
725 
726 		switch (c->x86_model) {
727 		case 5:
728 			if (l2 == 0)
729 				p = "Celeron (Covington)";
730 			else if (l2 == 256)
731 				p = "Mobile Pentium II (Dixon)";
732 			break;
733 
734 		case 6:
735 			if (l2 == 128)
736 				p = "Celeron (Mendocino)";
737 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
738 				p = "Celeron-A";
739 			break;
740 
741 		case 8:
742 			if (l2 == 128)
743 				p = "Celeron (Coppermine)";
744 			break;
745 		}
746 
747 		if (p)
748 			strcpy(c->x86_model_id, p);
749 	}
750 
751 	if (c->x86 == 15)
752 		set_cpu_cap(c, X86_FEATURE_P4);
753 	if (c->x86 == 6)
754 		set_cpu_cap(c, X86_FEATURE_P3);
755 #endif
756 
757 	/* Work around errata */
758 	srat_detect_node(c);
759 
760 	if (cpu_has(c, X86_FEATURE_VMX))
761 		detect_vmx_virtcap(c);
762 
763 	if (cpu_has(c, X86_FEATURE_TME))
764 		detect_tme(c);
765 
766 	init_intel_energy_perf(c);
767 
768 	init_intel_misc_features(c);
769 }
770 
771 #ifdef CONFIG_X86_32
772 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
773 {
774 	/*
775 	 * Intel PIII Tualatin. This comes in two flavours.
776 	 * One has 256kb of cache, the other 512. We have no way
777 	 * to determine which, so we use a boottime override
778 	 * for the 512kb model, and assume 256 otherwise.
779 	 */
780 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
781 		size = 256;
782 
783 	/*
784 	 * Intel Quark SoC X1000 contains a 4-way set associative
785 	 * 16K cache with a 16 byte cache line and 256 lines per tag
786 	 */
787 	if ((c->x86 == 5) && (c->x86_model == 9))
788 		size = 16;
789 	return size;
790 }
791 #endif
792 
793 #define TLB_INST_4K	0x01
794 #define TLB_INST_4M	0x02
795 #define TLB_INST_2M_4M	0x03
796 
797 #define TLB_INST_ALL	0x05
798 #define TLB_INST_1G	0x06
799 
800 #define TLB_DATA_4K	0x11
801 #define TLB_DATA_4M	0x12
802 #define TLB_DATA_2M_4M	0x13
803 #define TLB_DATA_4K_4M	0x14
804 
805 #define TLB_DATA_1G	0x16
806 
807 #define TLB_DATA0_4K	0x21
808 #define TLB_DATA0_4M	0x22
809 #define TLB_DATA0_2M_4M	0x23
810 
811 #define STLB_4K		0x41
812 #define STLB_4K_2M	0x42
813 
814 static const struct _tlb_table intel_tlb_table[] = {
815 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
816 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
817 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
818 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
819 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
820 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
821 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages */" },
822 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
823 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
824 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
825 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
826 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
827 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
828 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
829 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
830 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
831 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
832 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
833 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
834 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
835 	{ 0x6b, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 8-way associative" },
836 	{ 0x6c, TLB_DATA_2M_4M,		128,	" TLB_DATA 2 MByte or 4 MByte pages, 8-way associative" },
837 	{ 0x6d, TLB_DATA_1G,		16,	" TLB_DATA 1 GByte pages, fully associative" },
838 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
839 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
840 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
841 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
842 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
843 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
844 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
845 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
846 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
847 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
848 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
849 	{ 0xc2, TLB_DATA_2M_4M,		16,	" DTLB 2 MByte/4MByte pages, 4-way associative" },
850 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
851 	{ 0x00, 0, 0 }
852 };
853 
854 static void intel_tlb_lookup(const unsigned char desc)
855 {
856 	unsigned char k;
857 	if (desc == 0)
858 		return;
859 
860 	/* look up this descriptor in the table */
861 	for (k = 0; intel_tlb_table[k].descriptor != desc && \
862 			intel_tlb_table[k].descriptor != 0; k++)
863 		;
864 
865 	if (intel_tlb_table[k].tlb_type == 0)
866 		return;
867 
868 	switch (intel_tlb_table[k].tlb_type) {
869 	case STLB_4K:
870 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
871 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
872 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
873 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
874 		break;
875 	case STLB_4K_2M:
876 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
877 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
878 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
879 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
880 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
881 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
882 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
883 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
884 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
885 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
886 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
887 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
888 		break;
889 	case TLB_INST_ALL:
890 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
891 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
892 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
893 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
894 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
895 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
896 		break;
897 	case TLB_INST_4K:
898 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
899 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
900 		break;
901 	case TLB_INST_4M:
902 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
903 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
904 		break;
905 	case TLB_INST_2M_4M:
906 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
907 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
908 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
909 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
910 		break;
911 	case TLB_DATA_4K:
912 	case TLB_DATA0_4K:
913 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
914 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
915 		break;
916 	case TLB_DATA_4M:
917 	case TLB_DATA0_4M:
918 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
919 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
920 		break;
921 	case TLB_DATA_2M_4M:
922 	case TLB_DATA0_2M_4M:
923 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
924 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
925 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
926 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
927 		break;
928 	case TLB_DATA_4K_4M:
929 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
930 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
931 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
932 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
933 		break;
934 	case TLB_DATA_1G:
935 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
936 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
937 		break;
938 	}
939 }
940 
941 static void intel_detect_tlb(struct cpuinfo_x86 *c)
942 {
943 	int i, j, n;
944 	unsigned int regs[4];
945 	unsigned char *desc = (unsigned char *)regs;
946 
947 	if (c->cpuid_level < 2)
948 		return;
949 
950 	/* Number of times to iterate */
951 	n = cpuid_eax(2) & 0xFF;
952 
953 	for (i = 0 ; i < n ; i++) {
954 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
955 
956 		/* If bit 31 is set, this is an unknown format */
957 		for (j = 0 ; j < 3 ; j++)
958 			if (regs[j] & (1 << 31))
959 				regs[j] = 0;
960 
961 		/* Byte 0 is level count, not a descriptor */
962 		for (j = 1 ; j < 16 ; j++)
963 			intel_tlb_lookup(desc[j]);
964 	}
965 }
966 
967 static const struct cpu_dev intel_cpu_dev = {
968 	.c_vendor	= "Intel",
969 	.c_ident	= { "GenuineIntel" },
970 #ifdef CONFIG_X86_32
971 	.legacy_models = {
972 		{ .family = 4, .model_names =
973 		  {
974 			  [0] = "486 DX-25/33",
975 			  [1] = "486 DX-50",
976 			  [2] = "486 SX",
977 			  [3] = "486 DX/2",
978 			  [4] = "486 SL",
979 			  [5] = "486 SX/2",
980 			  [7] = "486 DX/2-WB",
981 			  [8] = "486 DX/4",
982 			  [9] = "486 DX/4-WB"
983 		  }
984 		},
985 		{ .family = 5, .model_names =
986 		  {
987 			  [0] = "Pentium 60/66 A-step",
988 			  [1] = "Pentium 60/66",
989 			  [2] = "Pentium 75 - 200",
990 			  [3] = "OverDrive PODP5V83",
991 			  [4] = "Pentium MMX",
992 			  [7] = "Mobile Pentium 75 - 200",
993 			  [8] = "Mobile Pentium MMX",
994 			  [9] = "Quark SoC X1000",
995 		  }
996 		},
997 		{ .family = 6, .model_names =
998 		  {
999 			  [0] = "Pentium Pro A-step",
1000 			  [1] = "Pentium Pro",
1001 			  [3] = "Pentium II (Klamath)",
1002 			  [4] = "Pentium II (Deschutes)",
1003 			  [5] = "Pentium II (Deschutes)",
1004 			  [6] = "Mobile Pentium II",
1005 			  [7] = "Pentium III (Katmai)",
1006 			  [8] = "Pentium III (Coppermine)",
1007 			  [10] = "Pentium III (Cascades)",
1008 			  [11] = "Pentium III (Tualatin)",
1009 		  }
1010 		},
1011 		{ .family = 15, .model_names =
1012 		  {
1013 			  [0] = "Pentium 4 (Unknown)",
1014 			  [1] = "Pentium 4 (Willamette)",
1015 			  [2] = "Pentium 4 (Northwood)",
1016 			  [4] = "Pentium 4 (Foster)",
1017 			  [5] = "Pentium 4 (Foster)",
1018 		  }
1019 		},
1020 	},
1021 	.legacy_cache_size = intel_size_cache,
1022 #endif
1023 	.c_detect_tlb	= intel_detect_tlb,
1024 	.c_early_init   = early_init_intel,
1025 	.c_init		= init_intel,
1026 	.c_bsp_resume	= intel_bsp_resume,
1027 	.c_x86_vendor	= X86_VENDOR_INTEL,
1028 };
1029 
1030 cpu_dev_register(intel_cpu_dev);
1031 
1032