xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision a20eefae)
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_DESKTOP,	0x0B,	0x80 },
146 	{ INTEL_FAM6_KABYLAKE_DESKTOP,	0x0A,	0x80 },
147 	{ INTEL_FAM6_KABYLAKE_DESKTOP,	0x09,	0x80 },
148 	{ INTEL_FAM6_KABYLAKE_MOBILE,	0x0A,	0x80 },
149 	{ INTEL_FAM6_KABYLAKE_MOBILE,	0x09,	0x80 },
150 	{ INTEL_FAM6_SKYLAKE_X,		0x03,	0x0100013e },
151 	{ INTEL_FAM6_SKYLAKE_X,		0x04,	0x0200003c },
152 	{ INTEL_FAM6_BROADWELL_CORE,	0x04,	0x28 },
153 	{ INTEL_FAM6_BROADWELL_GT3E,	0x01,	0x1b },
154 	{ INTEL_FAM6_BROADWELL_XEON_D,	0x02,	0x14 },
155 	{ INTEL_FAM6_BROADWELL_XEON_D,	0x03,	0x07000011 },
156 	{ INTEL_FAM6_BROADWELL_X,	0x01,	0x0b000025 },
157 	{ INTEL_FAM6_HASWELL_ULT,	0x01,	0x21 },
158 	{ INTEL_FAM6_HASWELL_GT3E,	0x01,	0x18 },
159 	{ INTEL_FAM6_HASWELL_CORE,	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 0x27:	/* Penwell */
269 		case 0x35:	/* Cloverview */
270 		case 0x4a:	/* Merrifield */
271 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
272 			break;
273 		default:
274 			break;
275 		}
276 	}
277 
278 	/*
279 	 * There is a known erratum on Pentium III and Core Solo
280 	 * and Core Duo CPUs.
281 	 * " Page with PAT set to WC while associated MTRR is UC
282 	 *   may consolidate to UC "
283 	 * Because of this erratum, it is better to stick with
284 	 * setting WC in MTRR rather than using PAT on these CPUs.
285 	 *
286 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
287 	 */
288 	if (c->x86 == 6 && c->x86_model < 15)
289 		clear_cpu_cap(c, X86_FEATURE_PAT);
290 
291 	/*
292 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
293 	 * clear the fast string and enhanced fast string CPU capabilities.
294 	 */
295 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
296 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
297 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
298 			pr_info("Disabled fast string operations\n");
299 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
300 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
301 		}
302 	}
303 
304 	/*
305 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
306 	 * "The operating system also is required to invalidate (i.e., flush)
307 	 *  the TLB when any changes are made to any of the page table entries.
308 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
309 	 *
310 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
311 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
312 	 * to be modified.
313 	 */
314 	if (c->x86 == 5 && c->x86_model == 9) {
315 		pr_info("Disabling PGE capability bit\n");
316 		setup_clear_cpu_cap(X86_FEATURE_PGE);
317 	}
318 
319 	if (c->cpuid_level >= 0x00000001) {
320 		u32 eax, ebx, ecx, edx;
321 
322 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
323 		/*
324 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
325 		 * apicids which are reserved per package. Store the resulting
326 		 * shift value for the package management code.
327 		 */
328 		if (edx & (1U << 28))
329 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
330 	}
331 
332 	check_mpx_erratum(c);
333 	check_memory_type_self_snoop_errata(c);
334 
335 	/*
336 	 * Get the number of SMT siblings early from the extended topology
337 	 * leaf, if available. Otherwise try the legacy SMT detection.
338 	 */
339 	if (detect_extended_topology_early(c) < 0)
340 		detect_ht_early(c);
341 }
342 
343 #ifdef CONFIG_X86_32
344 /*
345  *	Early probe support logic for ppro memory erratum #50
346  *
347  *	This is called before we do cpu ident work
348  */
349 
350 int ppro_with_ram_bug(void)
351 {
352 	/* Uses data from early_cpu_detect now */
353 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
354 	    boot_cpu_data.x86 == 6 &&
355 	    boot_cpu_data.x86_model == 1 &&
356 	    boot_cpu_data.x86_stepping < 8) {
357 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
358 		return 1;
359 	}
360 	return 0;
361 }
362 
363 static void intel_smp_check(struct cpuinfo_x86 *c)
364 {
365 	/* calling is from identify_secondary_cpu() ? */
366 	if (!c->cpu_index)
367 		return;
368 
369 	/*
370 	 * Mask B, Pentium, but not Pentium MMX
371 	 */
372 	if (c->x86 == 5 &&
373 	    c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
374 	    c->x86_model <= 3) {
375 		/*
376 		 * Remember we have B step Pentia with bugs
377 		 */
378 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
379 				    "with B stepping processors.\n");
380 	}
381 }
382 
383 static int forcepae;
384 static int __init forcepae_setup(char *__unused)
385 {
386 	forcepae = 1;
387 	return 1;
388 }
389 __setup("forcepae", forcepae_setup);
390 
391 static void intel_workarounds(struct cpuinfo_x86 *c)
392 {
393 #ifdef CONFIG_X86_F00F_BUG
394 	/*
395 	 * All models of Pentium and Pentium with MMX technology CPUs
396 	 * have the F0 0F bug, which lets nonprivileged users lock up the
397 	 * system. Announce that the fault handler will be checking for it.
398 	 * The Quark is also family 5, but does not have the same bug.
399 	 */
400 	clear_cpu_bug(c, X86_BUG_F00F);
401 	if (c->x86 == 5 && c->x86_model < 9) {
402 		static int f00f_workaround_enabled;
403 
404 		set_cpu_bug(c, X86_BUG_F00F);
405 		if (!f00f_workaround_enabled) {
406 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
407 			f00f_workaround_enabled = 1;
408 		}
409 	}
410 #endif
411 
412 	/*
413 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
414 	 * model 3 mask 3
415 	 */
416 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
417 		clear_cpu_cap(c, X86_FEATURE_SEP);
418 
419 	/*
420 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
421 	 * functionally usable PAE implementation.
422 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
423 	 */
424 	if (forcepae) {
425 		pr_warn("PAE forced!\n");
426 		set_cpu_cap(c, X86_FEATURE_PAE);
427 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
428 	}
429 
430 	/*
431 	 * P4 Xeon erratum 037 workaround.
432 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
433 	 */
434 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
435 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
436 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
437 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
438 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
439 		}
440 	}
441 
442 	/*
443 	 * See if we have a good local APIC by checking for buggy Pentia,
444 	 * i.e. all B steppings and the C2 stepping of P54C when using their
445 	 * integrated APIC (see 11AP erratum in "Pentium Processor
446 	 * Specification Update").
447 	 */
448 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
449 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
450 		set_cpu_bug(c, X86_BUG_11AP);
451 
452 
453 #ifdef CONFIG_X86_INTEL_USERCOPY
454 	/*
455 	 * Set up the preferred alignment for movsl bulk memory moves
456 	 */
457 	switch (c->x86) {
458 	case 4:		/* 486: untested */
459 		break;
460 	case 5:		/* Old Pentia: untested */
461 		break;
462 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
463 		movsl_mask.mask = 7;
464 		break;
465 	case 15:	/* P4 is OK down to 8-byte alignment */
466 		movsl_mask.mask = 7;
467 		break;
468 	}
469 #endif
470 
471 	intel_smp_check(c);
472 }
473 #else
474 static void intel_workarounds(struct cpuinfo_x86 *c)
475 {
476 }
477 #endif
478 
479 static void srat_detect_node(struct cpuinfo_x86 *c)
480 {
481 #ifdef CONFIG_NUMA
482 	unsigned node;
483 	int cpu = smp_processor_id();
484 
485 	/* Don't do the funky fallback heuristics the AMD version employs
486 	   for now. */
487 	node = numa_cpu_node(cpu);
488 	if (node == NUMA_NO_NODE || !node_online(node)) {
489 		/* reuse the value from init_cpu_to_node() */
490 		node = cpu_to_node(cpu);
491 	}
492 	numa_set_node(cpu, node);
493 #endif
494 }
495 
496 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
497 {
498 	/* Intel VMX MSR indicated features */
499 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
500 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
501 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
502 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
503 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
504 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
505 #define x86_VMX_FEATURE_EPT_CAP_AD		0x00200000
506 
507 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
508 	u32 msr_vpid_cap, msr_ept_cap;
509 
510 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
511 	clear_cpu_cap(c, X86_FEATURE_VNMI);
512 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
513 	clear_cpu_cap(c, X86_FEATURE_EPT);
514 	clear_cpu_cap(c, X86_FEATURE_VPID);
515 	clear_cpu_cap(c, X86_FEATURE_EPT_AD);
516 
517 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
518 	msr_ctl = vmx_msr_high | vmx_msr_low;
519 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
520 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
521 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
522 		set_cpu_cap(c, X86_FEATURE_VNMI);
523 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
524 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
525 		      vmx_msr_low, vmx_msr_high);
526 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
527 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
528 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
529 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
530 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT) {
531 			set_cpu_cap(c, X86_FEATURE_EPT);
532 			rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
533 			      msr_ept_cap, msr_vpid_cap);
534 			if (msr_ept_cap & x86_VMX_FEATURE_EPT_CAP_AD)
535 				set_cpu_cap(c, X86_FEATURE_EPT_AD);
536 		}
537 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
538 			set_cpu_cap(c, X86_FEATURE_VPID);
539 	}
540 }
541 
542 #define MSR_IA32_TME_ACTIVATE		0x982
543 
544 /* Helpers to access TME_ACTIVATE MSR */
545 #define TME_ACTIVATE_LOCKED(x)		(x & 0x1)
546 #define TME_ACTIVATE_ENABLED(x)		(x & 0x2)
547 
548 #define TME_ACTIVATE_POLICY(x)		((x >> 4) & 0xf)	/* Bits 7:4 */
549 #define TME_ACTIVATE_POLICY_AES_XTS_128	0
550 
551 #define TME_ACTIVATE_KEYID_BITS(x)	((x >> 32) & 0xf)	/* Bits 35:32 */
552 
553 #define TME_ACTIVATE_CRYPTO_ALGS(x)	((x >> 48) & 0xffff)	/* Bits 63:48 */
554 #define TME_ACTIVATE_CRYPTO_AES_XTS_128	1
555 
556 /* Values for mktme_status (SW only construct) */
557 #define MKTME_ENABLED			0
558 #define MKTME_DISABLED			1
559 #define MKTME_UNINITIALIZED		2
560 static int mktme_status = MKTME_UNINITIALIZED;
561 
562 static void detect_tme(struct cpuinfo_x86 *c)
563 {
564 	u64 tme_activate, tme_policy, tme_crypto_algs;
565 	int keyid_bits = 0, nr_keyids = 0;
566 	static u64 tme_activate_cpu0 = 0;
567 
568 	rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
569 
570 	if (mktme_status != MKTME_UNINITIALIZED) {
571 		if (tme_activate != tme_activate_cpu0) {
572 			/* Broken BIOS? */
573 			pr_err_once("x86/tme: configuration is inconsistent between CPUs\n");
574 			pr_err_once("x86/tme: MKTME is not usable\n");
575 			mktme_status = MKTME_DISABLED;
576 
577 			/* Proceed. We may need to exclude bits from x86_phys_bits. */
578 		}
579 	} else {
580 		tme_activate_cpu0 = tme_activate;
581 	}
582 
583 	if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
584 		pr_info_once("x86/tme: not enabled by BIOS\n");
585 		mktme_status = MKTME_DISABLED;
586 		return;
587 	}
588 
589 	if (mktme_status != MKTME_UNINITIALIZED)
590 		goto detect_keyid_bits;
591 
592 	pr_info("x86/tme: enabled by BIOS\n");
593 
594 	tme_policy = TME_ACTIVATE_POLICY(tme_activate);
595 	if (tme_policy != TME_ACTIVATE_POLICY_AES_XTS_128)
596 		pr_warn("x86/tme: Unknown policy is active: %#llx\n", tme_policy);
597 
598 	tme_crypto_algs = TME_ACTIVATE_CRYPTO_ALGS(tme_activate);
599 	if (!(tme_crypto_algs & TME_ACTIVATE_CRYPTO_AES_XTS_128)) {
600 		pr_err("x86/mktme: No known encryption algorithm is supported: %#llx\n",
601 				tme_crypto_algs);
602 		mktme_status = MKTME_DISABLED;
603 	}
604 detect_keyid_bits:
605 	keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
606 	nr_keyids = (1UL << keyid_bits) - 1;
607 	if (nr_keyids) {
608 		pr_info_once("x86/mktme: enabled by BIOS\n");
609 		pr_info_once("x86/mktme: %d KeyIDs available\n", nr_keyids);
610 	} else {
611 		pr_info_once("x86/mktme: disabled by BIOS\n");
612 	}
613 
614 	if (mktme_status == MKTME_UNINITIALIZED) {
615 		/* MKTME is usable */
616 		mktme_status = MKTME_ENABLED;
617 	}
618 
619 	/*
620 	 * KeyID bits effectively lower the number of physical address
621 	 * bits.  Update cpuinfo_x86::x86_phys_bits accordingly.
622 	 */
623 	c->x86_phys_bits -= keyid_bits;
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 	early_init_intel(c);
657 
658 	intel_workarounds(c);
659 
660 	/*
661 	 * Detect the extended topology information if available. This
662 	 * will reinitialise the initial_apicid which will be used
663 	 * in init_intel_cacheinfo()
664 	 */
665 	detect_extended_topology(c);
666 
667 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
668 		/*
669 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
670 		 * detection.
671 		 */
672 		detect_num_cpu_cores(c);
673 #ifdef CONFIG_X86_32
674 		detect_ht(c);
675 #endif
676 	}
677 
678 	init_intel_cacheinfo(c);
679 
680 	if (c->cpuid_level > 9) {
681 		unsigned eax = cpuid_eax(10);
682 		/* Check for version and the number of counters */
683 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
684 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
685 	}
686 
687 	if (cpu_has(c, X86_FEATURE_XMM2))
688 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
689 
690 	if (boot_cpu_has(X86_FEATURE_DS)) {
691 		unsigned int l1, l2;
692 
693 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
694 		if (!(l1 & (1<<11)))
695 			set_cpu_cap(c, X86_FEATURE_BTS);
696 		if (!(l1 & (1<<12)))
697 			set_cpu_cap(c, X86_FEATURE_PEBS);
698 	}
699 
700 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
701 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
702 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
703 
704 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
705 		((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
706 		set_cpu_bug(c, X86_BUG_MONITOR);
707 
708 #ifdef CONFIG_X86_64
709 	if (c->x86 == 15)
710 		c->x86_cache_alignment = c->x86_clflush_size * 2;
711 	if (c->x86 == 6)
712 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
713 #else
714 	/*
715 	 * Names for the Pentium II/Celeron processors
716 	 * detectable only by also checking the cache size.
717 	 * Dixon is NOT a Celeron.
718 	 */
719 	if (c->x86 == 6) {
720 		unsigned int l2 = c->x86_cache_size;
721 		char *p = NULL;
722 
723 		switch (c->x86_model) {
724 		case 5:
725 			if (l2 == 0)
726 				p = "Celeron (Covington)";
727 			else if (l2 == 256)
728 				p = "Mobile Pentium II (Dixon)";
729 			break;
730 
731 		case 6:
732 			if (l2 == 128)
733 				p = "Celeron (Mendocino)";
734 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
735 				p = "Celeron-A";
736 			break;
737 
738 		case 8:
739 			if (l2 == 128)
740 				p = "Celeron (Coppermine)";
741 			break;
742 		}
743 
744 		if (p)
745 			strcpy(c->x86_model_id, p);
746 	}
747 
748 	if (c->x86 == 15)
749 		set_cpu_cap(c, X86_FEATURE_P4);
750 	if (c->x86 == 6)
751 		set_cpu_cap(c, X86_FEATURE_P3);
752 #endif
753 
754 	/* Work around errata */
755 	srat_detect_node(c);
756 
757 	if (cpu_has(c, X86_FEATURE_VMX))
758 		detect_vmx_virtcap(c);
759 
760 	if (cpu_has(c, X86_FEATURE_TME))
761 		detect_tme(c);
762 
763 	init_intel_misc_features(c);
764 }
765 
766 #ifdef CONFIG_X86_32
767 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
768 {
769 	/*
770 	 * Intel PIII Tualatin. This comes in two flavours.
771 	 * One has 256kb of cache, the other 512. We have no way
772 	 * to determine which, so we use a boottime override
773 	 * for the 512kb model, and assume 256 otherwise.
774 	 */
775 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
776 		size = 256;
777 
778 	/*
779 	 * Intel Quark SoC X1000 contains a 4-way set associative
780 	 * 16K cache with a 16 byte cache line and 256 lines per tag
781 	 */
782 	if ((c->x86 == 5) && (c->x86_model == 9))
783 		size = 16;
784 	return size;
785 }
786 #endif
787 
788 #define TLB_INST_4K	0x01
789 #define TLB_INST_4M	0x02
790 #define TLB_INST_2M_4M	0x03
791 
792 #define TLB_INST_ALL	0x05
793 #define TLB_INST_1G	0x06
794 
795 #define TLB_DATA_4K	0x11
796 #define TLB_DATA_4M	0x12
797 #define TLB_DATA_2M_4M	0x13
798 #define TLB_DATA_4K_4M	0x14
799 
800 #define TLB_DATA_1G	0x16
801 
802 #define TLB_DATA0_4K	0x21
803 #define TLB_DATA0_4M	0x22
804 #define TLB_DATA0_2M_4M	0x23
805 
806 #define STLB_4K		0x41
807 #define STLB_4K_2M	0x42
808 
809 static const struct _tlb_table intel_tlb_table[] = {
810 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
811 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
812 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
813 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
814 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
815 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
816 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages */" },
817 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
818 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
819 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
820 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
821 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
822 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
823 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
824 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
825 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
826 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
827 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
828 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
829 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
830 	{ 0x6b, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 8-way associative" },
831 	{ 0x6c, TLB_DATA_2M_4M,		128,	" TLB_DATA 2 MByte or 4 MByte pages, 8-way associative" },
832 	{ 0x6d, TLB_DATA_1G,		16,	" TLB_DATA 1 GByte pages, fully associative" },
833 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
834 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
835 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
836 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
837 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
838 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
839 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
840 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
841 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
842 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
843 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
844 	{ 0xc2, TLB_DATA_2M_4M,		16,	" DTLB 2 MByte/4MByte pages, 4-way associative" },
845 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
846 	{ 0x00, 0, 0 }
847 };
848 
849 static void intel_tlb_lookup(const unsigned char desc)
850 {
851 	unsigned char k;
852 	if (desc == 0)
853 		return;
854 
855 	/* look up this descriptor in the table */
856 	for (k = 0; intel_tlb_table[k].descriptor != desc && \
857 			intel_tlb_table[k].descriptor != 0; k++)
858 		;
859 
860 	if (intel_tlb_table[k].tlb_type == 0)
861 		return;
862 
863 	switch (intel_tlb_table[k].tlb_type) {
864 	case STLB_4K:
865 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
866 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
867 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
868 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
869 		break;
870 	case STLB_4K_2M:
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 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
876 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
877 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
878 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
879 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
880 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
881 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
882 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
883 		break;
884 	case TLB_INST_ALL:
885 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
886 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
887 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
888 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
889 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
890 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
891 		break;
892 	case TLB_INST_4K:
893 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
894 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
895 		break;
896 	case TLB_INST_4M:
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_2M_4M:
901 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
902 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
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_DATA_4K:
907 	case TLB_DATA0_4K:
908 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
909 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
910 		break;
911 	case TLB_DATA_4M:
912 	case TLB_DATA0_4M:
913 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
914 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
915 		break;
916 	case TLB_DATA_2M_4M:
917 	case TLB_DATA0_2M_4M:
918 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
919 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
920 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
921 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
922 		break;
923 	case TLB_DATA_4K_4M:
924 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
925 			tlb_lld_4k[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_1G:
930 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
931 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
932 		break;
933 	}
934 }
935 
936 static void intel_detect_tlb(struct cpuinfo_x86 *c)
937 {
938 	int i, j, n;
939 	unsigned int regs[4];
940 	unsigned char *desc = (unsigned char *)regs;
941 
942 	if (c->cpuid_level < 2)
943 		return;
944 
945 	/* Number of times to iterate */
946 	n = cpuid_eax(2) & 0xFF;
947 
948 	for (i = 0 ; i < n ; i++) {
949 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
950 
951 		/* If bit 31 is set, this is an unknown format */
952 		for (j = 0 ; j < 3 ; j++)
953 			if (regs[j] & (1 << 31))
954 				regs[j] = 0;
955 
956 		/* Byte 0 is level count, not a descriptor */
957 		for (j = 1 ; j < 16 ; j++)
958 			intel_tlb_lookup(desc[j]);
959 	}
960 }
961 
962 static const struct cpu_dev intel_cpu_dev = {
963 	.c_vendor	= "Intel",
964 	.c_ident	= { "GenuineIntel" },
965 #ifdef CONFIG_X86_32
966 	.legacy_models = {
967 		{ .family = 4, .model_names =
968 		  {
969 			  [0] = "486 DX-25/33",
970 			  [1] = "486 DX-50",
971 			  [2] = "486 SX",
972 			  [3] = "486 DX/2",
973 			  [4] = "486 SL",
974 			  [5] = "486 SX/2",
975 			  [7] = "486 DX/2-WB",
976 			  [8] = "486 DX/4",
977 			  [9] = "486 DX/4-WB"
978 		  }
979 		},
980 		{ .family = 5, .model_names =
981 		  {
982 			  [0] = "Pentium 60/66 A-step",
983 			  [1] = "Pentium 60/66",
984 			  [2] = "Pentium 75 - 200",
985 			  [3] = "OverDrive PODP5V83",
986 			  [4] = "Pentium MMX",
987 			  [7] = "Mobile Pentium 75 - 200",
988 			  [8] = "Mobile Pentium MMX",
989 			  [9] = "Quark SoC X1000",
990 		  }
991 		},
992 		{ .family = 6, .model_names =
993 		  {
994 			  [0] = "Pentium Pro A-step",
995 			  [1] = "Pentium Pro",
996 			  [3] = "Pentium II (Klamath)",
997 			  [4] = "Pentium II (Deschutes)",
998 			  [5] = "Pentium II (Deschutes)",
999 			  [6] = "Mobile Pentium II",
1000 			  [7] = "Pentium III (Katmai)",
1001 			  [8] = "Pentium III (Coppermine)",
1002 			  [10] = "Pentium III (Cascades)",
1003 			  [11] = "Pentium III (Tualatin)",
1004 		  }
1005 		},
1006 		{ .family = 15, .model_names =
1007 		  {
1008 			  [0] = "Pentium 4 (Unknown)",
1009 			  [1] = "Pentium 4 (Willamette)",
1010 			  [2] = "Pentium 4 (Northwood)",
1011 			  [4] = "Pentium 4 (Foster)",
1012 			  [5] = "Pentium 4 (Foster)",
1013 		  }
1014 		},
1015 	},
1016 	.legacy_cache_size = intel_size_cache,
1017 #endif
1018 	.c_detect_tlb	= intel_detect_tlb,
1019 	.c_early_init   = early_init_intel,
1020 	.c_init		= init_intel,
1021 	.c_x86_vendor	= X86_VENDOR_INTEL,
1022 };
1023 
1024 cpu_dev_register(intel_cpu_dev);
1025