xref: /openbmc/linux/arch/x86/kernel/cpu/intel.c (revision 01a6e126)
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/11/2018/01/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_SKYLAKE_DESKTOP,	0x03,	0xc2 },
127 	{ INTEL_FAM6_BROADWELL_CORE,	0x04,	0x28 },
128 	{ INTEL_FAM6_BROADWELL_GT3E,	0x01,	0x1b },
129 	{ INTEL_FAM6_BROADWELL_XEON_D,	0x02,	0x14 },
130 	{ INTEL_FAM6_BROADWELL_XEON_D,	0x03,	0x07000011 },
131 	{ INTEL_FAM6_BROADWELL_X,	0x01,	0x0b000025 },
132 	{ INTEL_FAM6_HASWELL_ULT,	0x01,	0x21 },
133 	{ INTEL_FAM6_HASWELL_GT3E,	0x01,	0x18 },
134 	{ INTEL_FAM6_HASWELL_CORE,	0x03,	0x23 },
135 	{ INTEL_FAM6_HASWELL_X,		0x02,	0x3b },
136 	{ INTEL_FAM6_HASWELL_X,		0x04,	0x10 },
137 	{ INTEL_FAM6_IVYBRIDGE_X,	0x04,	0x42a },
138 	/* Observed in the wild */
139 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x06,	0x61b },
140 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x07,	0x712 },
141 };
142 
143 static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
144 {
145 	int i;
146 
147 	for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
148 		if (c->x86_model == spectre_bad_microcodes[i].model &&
149 		    c->x86_stepping == spectre_bad_microcodes[i].stepping)
150 			return (c->microcode <= spectre_bad_microcodes[i].microcode);
151 	}
152 	return false;
153 }
154 
155 static void early_init_intel(struct cpuinfo_x86 *c)
156 {
157 	u64 misc_enable;
158 
159 	/* Unmask CPUID levels if masked: */
160 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
161 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
162 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
163 			c->cpuid_level = cpuid_eax(0);
164 			get_cpu_cap(c);
165 		}
166 	}
167 
168 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
169 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
170 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
171 
172 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
173 		c->microcode = intel_get_microcode_revision();
174 
175 	/* Now if any of them are set, check the blacklist and clear the lot */
176 	if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
177 	     cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
178 	     cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
179 	     cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
180 		pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
181 		setup_clear_cpu_cap(X86_FEATURE_IBRS);
182 		setup_clear_cpu_cap(X86_FEATURE_IBPB);
183 		setup_clear_cpu_cap(X86_FEATURE_STIBP);
184 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
185 		setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
186 	}
187 
188 	/*
189 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
190 	 *
191 	 * A race condition between speculative fetches and invalidating
192 	 * a large page.  This is worked around in microcode, but we
193 	 * need the microcode to have already been loaded... so if it is
194 	 * not, recommend a BIOS update and disable large pages.
195 	 */
196 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_stepping <= 2 &&
197 	    c->microcode < 0x20e) {
198 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
199 		clear_cpu_cap(c, X86_FEATURE_PSE);
200 	}
201 
202 #ifdef CONFIG_X86_64
203 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
204 #else
205 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
206 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
207 		c->x86_cache_alignment = 128;
208 #endif
209 
210 	/* CPUID workaround for 0F33/0F34 CPU */
211 	if (c->x86 == 0xF && c->x86_model == 0x3
212 	    && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
213 		c->x86_phys_bits = 36;
214 
215 	/*
216 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
217 	 * with P/T states and does not stop in deep C-states.
218 	 *
219 	 * It is also reliable across cores and sockets. (but not across
220 	 * cabinets - we turn it off in that case explicitly.)
221 	 */
222 	if (c->x86_power & (1 << 8)) {
223 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
224 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
225 	}
226 
227 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
228 	if (c->x86 == 6) {
229 		switch (c->x86_model) {
230 		case 0x27:	/* Penwell */
231 		case 0x35:	/* Cloverview */
232 		case 0x4a:	/* Merrifield */
233 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
234 			break;
235 		default:
236 			break;
237 		}
238 	}
239 
240 	/*
241 	 * There is a known erratum on Pentium III and Core Solo
242 	 * and Core Duo CPUs.
243 	 * " Page with PAT set to WC while associated MTRR is UC
244 	 *   may consolidate to UC "
245 	 * Because of this erratum, it is better to stick with
246 	 * setting WC in MTRR rather than using PAT on these CPUs.
247 	 *
248 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
249 	 */
250 	if (c->x86 == 6 && c->x86_model < 15)
251 		clear_cpu_cap(c, X86_FEATURE_PAT);
252 
253 	/*
254 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
255 	 * clear the fast string and enhanced fast string CPU capabilities.
256 	 */
257 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
258 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
259 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
260 			pr_info("Disabled fast string operations\n");
261 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
262 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
263 		}
264 	}
265 
266 	/*
267 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
268 	 * "The operating system also is required to invalidate (i.e., flush)
269 	 *  the TLB when any changes are made to any of the page table entries.
270 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
271 	 *
272 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
273 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
274 	 * to be modified.
275 	 */
276 	if (c->x86 == 5 && c->x86_model == 9) {
277 		pr_info("Disabling PGE capability bit\n");
278 		setup_clear_cpu_cap(X86_FEATURE_PGE);
279 	}
280 
281 	if (c->cpuid_level >= 0x00000001) {
282 		u32 eax, ebx, ecx, edx;
283 
284 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
285 		/*
286 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
287 		 * apicids which are reserved per package. Store the resulting
288 		 * shift value for the package management code.
289 		 */
290 		if (edx & (1U << 28))
291 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
292 	}
293 
294 	check_mpx_erratum(c);
295 }
296 
297 #ifdef CONFIG_X86_32
298 /*
299  *	Early probe support logic for ppro memory erratum #50
300  *
301  *	This is called before we do cpu ident work
302  */
303 
304 int ppro_with_ram_bug(void)
305 {
306 	/* Uses data from early_cpu_detect now */
307 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
308 	    boot_cpu_data.x86 == 6 &&
309 	    boot_cpu_data.x86_model == 1 &&
310 	    boot_cpu_data.x86_stepping < 8) {
311 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
312 		return 1;
313 	}
314 	return 0;
315 }
316 
317 static void intel_smp_check(struct cpuinfo_x86 *c)
318 {
319 	/* calling is from identify_secondary_cpu() ? */
320 	if (!c->cpu_index)
321 		return;
322 
323 	/*
324 	 * Mask B, Pentium, but not Pentium MMX
325 	 */
326 	if (c->x86 == 5 &&
327 	    c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
328 	    c->x86_model <= 3) {
329 		/*
330 		 * Remember we have B step Pentia with bugs
331 		 */
332 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
333 				    "with B stepping processors.\n");
334 	}
335 }
336 
337 static int forcepae;
338 static int __init forcepae_setup(char *__unused)
339 {
340 	forcepae = 1;
341 	return 1;
342 }
343 __setup("forcepae", forcepae_setup);
344 
345 static void intel_workarounds(struct cpuinfo_x86 *c)
346 {
347 #ifdef CONFIG_X86_F00F_BUG
348 	/*
349 	 * All models of Pentium and Pentium with MMX technology CPUs
350 	 * have the F0 0F bug, which lets nonprivileged users lock up the
351 	 * system. Announce that the fault handler will be checking for it.
352 	 * The Quark is also family 5, but does not have the same bug.
353 	 */
354 	clear_cpu_bug(c, X86_BUG_F00F);
355 	if (c->x86 == 5 && c->x86_model < 9) {
356 		static int f00f_workaround_enabled;
357 
358 		set_cpu_bug(c, X86_BUG_F00F);
359 		if (!f00f_workaround_enabled) {
360 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
361 			f00f_workaround_enabled = 1;
362 		}
363 	}
364 #endif
365 
366 	/*
367 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
368 	 * model 3 mask 3
369 	 */
370 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
371 		clear_cpu_cap(c, X86_FEATURE_SEP);
372 
373 	/*
374 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
375 	 * functionally usable PAE implementation.
376 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
377 	 */
378 	if (forcepae) {
379 		pr_warn("PAE forced!\n");
380 		set_cpu_cap(c, X86_FEATURE_PAE);
381 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
382 	}
383 
384 	/*
385 	 * P4 Xeon erratum 037 workaround.
386 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
387 	 */
388 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
389 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
390 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
391 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
392 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
393 		}
394 	}
395 
396 	/*
397 	 * See if we have a good local APIC by checking for buggy Pentia,
398 	 * i.e. all B steppings and the C2 stepping of P54C when using their
399 	 * integrated APIC (see 11AP erratum in "Pentium Processor
400 	 * Specification Update").
401 	 */
402 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
403 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
404 		set_cpu_bug(c, X86_BUG_11AP);
405 
406 
407 #ifdef CONFIG_X86_INTEL_USERCOPY
408 	/*
409 	 * Set up the preferred alignment for movsl bulk memory moves
410 	 */
411 	switch (c->x86) {
412 	case 4:		/* 486: untested */
413 		break;
414 	case 5:		/* Old Pentia: untested */
415 		break;
416 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
417 		movsl_mask.mask = 7;
418 		break;
419 	case 15:	/* P4 is OK down to 8-byte alignment */
420 		movsl_mask.mask = 7;
421 		break;
422 	}
423 #endif
424 
425 	intel_smp_check(c);
426 }
427 #else
428 static void intel_workarounds(struct cpuinfo_x86 *c)
429 {
430 }
431 #endif
432 
433 static void srat_detect_node(struct cpuinfo_x86 *c)
434 {
435 #ifdef CONFIG_NUMA
436 	unsigned node;
437 	int cpu = smp_processor_id();
438 
439 	/* Don't do the funky fallback heuristics the AMD version employs
440 	   for now. */
441 	node = numa_cpu_node(cpu);
442 	if (node == NUMA_NO_NODE || !node_online(node)) {
443 		/* reuse the value from init_cpu_to_node() */
444 		node = cpu_to_node(cpu);
445 	}
446 	numa_set_node(cpu, node);
447 #endif
448 }
449 
450 /*
451  * find out the number of processor cores on the die
452  */
453 static int intel_num_cpu_cores(struct cpuinfo_x86 *c)
454 {
455 	unsigned int eax, ebx, ecx, edx;
456 
457 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
458 		return 1;
459 
460 	/* Intel has a non-standard dependency on %ecx for this CPUID level. */
461 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
462 	if (eax & 0x1f)
463 		return (eax >> 26) + 1;
464 	else
465 		return 1;
466 }
467 
468 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
469 {
470 	/* Intel VMX MSR indicated features */
471 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
472 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
473 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
474 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
475 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
476 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
477 
478 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
479 
480 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
481 	clear_cpu_cap(c, X86_FEATURE_VNMI);
482 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
483 	clear_cpu_cap(c, X86_FEATURE_EPT);
484 	clear_cpu_cap(c, X86_FEATURE_VPID);
485 
486 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
487 	msr_ctl = vmx_msr_high | vmx_msr_low;
488 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
489 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
490 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
491 		set_cpu_cap(c, X86_FEATURE_VNMI);
492 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
493 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
494 		      vmx_msr_low, vmx_msr_high);
495 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
496 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
497 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
498 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
499 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
500 			set_cpu_cap(c, X86_FEATURE_EPT);
501 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
502 			set_cpu_cap(c, X86_FEATURE_VPID);
503 	}
504 }
505 
506 static void init_intel_energy_perf(struct cpuinfo_x86 *c)
507 {
508 	u64 epb;
509 
510 	/*
511 	 * Initialize MSR_IA32_ENERGY_PERF_BIAS if not already initialized.
512 	 * (x86_energy_perf_policy(8) is available to change it at run-time.)
513 	 */
514 	if (!cpu_has(c, X86_FEATURE_EPB))
515 		return;
516 
517 	rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
518 	if ((epb & 0xF) != ENERGY_PERF_BIAS_PERFORMANCE)
519 		return;
520 
521 	pr_warn_once("ENERGY_PERF_BIAS: Set to 'normal', was 'performance'\n");
522 	pr_warn_once("ENERGY_PERF_BIAS: View and update with x86_energy_perf_policy(8)\n");
523 	epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
524 	wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
525 }
526 
527 static void intel_bsp_resume(struct cpuinfo_x86 *c)
528 {
529 	/*
530 	 * MSR_IA32_ENERGY_PERF_BIAS is lost across suspend/resume,
531 	 * so reinitialize it properly like during bootup:
532 	 */
533 	init_intel_energy_perf(c);
534 }
535 
536 static void init_cpuid_fault(struct cpuinfo_x86 *c)
537 {
538 	u64 msr;
539 
540 	if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
541 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
542 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
543 	}
544 }
545 
546 static void init_intel_misc_features(struct cpuinfo_x86 *c)
547 {
548 	u64 msr;
549 
550 	if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
551 		return;
552 
553 	/* Clear all MISC features */
554 	this_cpu_write(msr_misc_features_shadow, 0);
555 
556 	/* Check features and update capabilities and shadow control bits */
557 	init_cpuid_fault(c);
558 	probe_xeon_phi_r3mwait(c);
559 
560 	msr = this_cpu_read(msr_misc_features_shadow);
561 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
562 }
563 
564 static void init_intel(struct cpuinfo_x86 *c)
565 {
566 	unsigned int l2 = 0;
567 
568 	early_init_intel(c);
569 
570 	intel_workarounds(c);
571 
572 	/*
573 	 * Detect the extended topology information if available. This
574 	 * will reinitialise the initial_apicid which will be used
575 	 * in init_intel_cacheinfo()
576 	 */
577 	detect_extended_topology(c);
578 
579 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
580 		/*
581 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
582 		 * detection.
583 		 */
584 		c->x86_max_cores = intel_num_cpu_cores(c);
585 #ifdef CONFIG_X86_32
586 		detect_ht(c);
587 #endif
588 	}
589 
590 	l2 = init_intel_cacheinfo(c);
591 
592 	/* Detect legacy cache sizes if init_intel_cacheinfo did not */
593 	if (l2 == 0) {
594 		cpu_detect_cache_sizes(c);
595 		l2 = c->x86_cache_size;
596 	}
597 
598 	if (c->cpuid_level > 9) {
599 		unsigned eax = cpuid_eax(10);
600 		/* Check for version and the number of counters */
601 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
602 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
603 	}
604 
605 	if (cpu_has(c, X86_FEATURE_XMM2))
606 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
607 
608 	if (boot_cpu_has(X86_FEATURE_DS)) {
609 		unsigned int l1;
610 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
611 		if (!(l1 & (1<<11)))
612 			set_cpu_cap(c, X86_FEATURE_BTS);
613 		if (!(l1 & (1<<12)))
614 			set_cpu_cap(c, X86_FEATURE_PEBS);
615 	}
616 
617 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
618 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
619 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
620 
621 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
622 		((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
623 		set_cpu_bug(c, X86_BUG_MONITOR);
624 
625 #ifdef CONFIG_X86_64
626 	if (c->x86 == 15)
627 		c->x86_cache_alignment = c->x86_clflush_size * 2;
628 	if (c->x86 == 6)
629 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
630 #else
631 	/*
632 	 * Names for the Pentium II/Celeron processors
633 	 * detectable only by also checking the cache size.
634 	 * Dixon is NOT a Celeron.
635 	 */
636 	if (c->x86 == 6) {
637 		char *p = NULL;
638 
639 		switch (c->x86_model) {
640 		case 5:
641 			if (l2 == 0)
642 				p = "Celeron (Covington)";
643 			else if (l2 == 256)
644 				p = "Mobile Pentium II (Dixon)";
645 			break;
646 
647 		case 6:
648 			if (l2 == 128)
649 				p = "Celeron (Mendocino)";
650 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
651 				p = "Celeron-A";
652 			break;
653 
654 		case 8:
655 			if (l2 == 128)
656 				p = "Celeron (Coppermine)";
657 			break;
658 		}
659 
660 		if (p)
661 			strcpy(c->x86_model_id, p);
662 	}
663 
664 	if (c->x86 == 15)
665 		set_cpu_cap(c, X86_FEATURE_P4);
666 	if (c->x86 == 6)
667 		set_cpu_cap(c, X86_FEATURE_P3);
668 #endif
669 
670 	/* Work around errata */
671 	srat_detect_node(c);
672 
673 	if (cpu_has(c, X86_FEATURE_VMX))
674 		detect_vmx_virtcap(c);
675 
676 	init_intel_energy_perf(c);
677 
678 	init_intel_misc_features(c);
679 }
680 
681 #ifdef CONFIG_X86_32
682 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
683 {
684 	/*
685 	 * Intel PIII Tualatin. This comes in two flavours.
686 	 * One has 256kb of cache, the other 512. We have no way
687 	 * to determine which, so we use a boottime override
688 	 * for the 512kb model, and assume 256 otherwise.
689 	 */
690 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
691 		size = 256;
692 
693 	/*
694 	 * Intel Quark SoC X1000 contains a 4-way set associative
695 	 * 16K cache with a 16 byte cache line and 256 lines per tag
696 	 */
697 	if ((c->x86 == 5) && (c->x86_model == 9))
698 		size = 16;
699 	return size;
700 }
701 #endif
702 
703 #define TLB_INST_4K	0x01
704 #define TLB_INST_4M	0x02
705 #define TLB_INST_2M_4M	0x03
706 
707 #define TLB_INST_ALL	0x05
708 #define TLB_INST_1G	0x06
709 
710 #define TLB_DATA_4K	0x11
711 #define TLB_DATA_4M	0x12
712 #define TLB_DATA_2M_4M	0x13
713 #define TLB_DATA_4K_4M	0x14
714 
715 #define TLB_DATA_1G	0x16
716 
717 #define TLB_DATA0_4K	0x21
718 #define TLB_DATA0_4M	0x22
719 #define TLB_DATA0_2M_4M	0x23
720 
721 #define STLB_4K		0x41
722 #define STLB_4K_2M	0x42
723 
724 static const struct _tlb_table intel_tlb_table[] = {
725 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
726 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
727 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
728 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
729 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
730 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
731 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages */" },
732 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
733 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
734 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
735 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
736 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
737 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
738 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
739 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
740 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
741 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
742 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
743 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
744 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
745 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
746 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
747 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
748 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
749 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
750 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
751 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
752 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
753 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
754 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
755 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
756 	{ 0xc2, TLB_DATA_2M_4M,		16,	" DTLB 2 MByte/4MByte pages, 4-way associative" },
757 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
758 	{ 0x00, 0, 0 }
759 };
760 
761 static void intel_tlb_lookup(const unsigned char desc)
762 {
763 	unsigned char k;
764 	if (desc == 0)
765 		return;
766 
767 	/* look up this descriptor in the table */
768 	for (k = 0; intel_tlb_table[k].descriptor != desc && \
769 			intel_tlb_table[k].descriptor != 0; k++)
770 		;
771 
772 	if (intel_tlb_table[k].tlb_type == 0)
773 		return;
774 
775 	switch (intel_tlb_table[k].tlb_type) {
776 	case STLB_4K:
777 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
778 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
779 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
780 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
781 		break;
782 	case STLB_4K_2M:
783 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
784 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
785 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
786 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
787 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
788 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
789 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
790 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
791 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
792 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
793 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
794 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
795 		break;
796 	case TLB_INST_ALL:
797 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
798 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
799 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
800 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
801 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
802 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
803 		break;
804 	case TLB_INST_4K:
805 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
806 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
807 		break;
808 	case TLB_INST_4M:
809 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
810 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
811 		break;
812 	case TLB_INST_2M_4M:
813 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
814 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
815 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
816 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
817 		break;
818 	case TLB_DATA_4K:
819 	case TLB_DATA0_4K:
820 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
821 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
822 		break;
823 	case TLB_DATA_4M:
824 	case TLB_DATA0_4M:
825 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
826 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
827 		break;
828 	case TLB_DATA_2M_4M:
829 	case TLB_DATA0_2M_4M:
830 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
831 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
832 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
833 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
834 		break;
835 	case TLB_DATA_4K_4M:
836 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
837 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
838 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
839 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
840 		break;
841 	case TLB_DATA_1G:
842 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
843 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
844 		break;
845 	}
846 }
847 
848 static void intel_detect_tlb(struct cpuinfo_x86 *c)
849 {
850 	int i, j, n;
851 	unsigned int regs[4];
852 	unsigned char *desc = (unsigned char *)regs;
853 
854 	if (c->cpuid_level < 2)
855 		return;
856 
857 	/* Number of times to iterate */
858 	n = cpuid_eax(2) & 0xFF;
859 
860 	for (i = 0 ; i < n ; i++) {
861 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
862 
863 		/* If bit 31 is set, this is an unknown format */
864 		for (j = 0 ; j < 3 ; j++)
865 			if (regs[j] & (1 << 31))
866 				regs[j] = 0;
867 
868 		/* Byte 0 is level count, not a descriptor */
869 		for (j = 1 ; j < 16 ; j++)
870 			intel_tlb_lookup(desc[j]);
871 	}
872 }
873 
874 static const struct cpu_dev intel_cpu_dev = {
875 	.c_vendor	= "Intel",
876 	.c_ident	= { "GenuineIntel" },
877 #ifdef CONFIG_X86_32
878 	.legacy_models = {
879 		{ .family = 4, .model_names =
880 		  {
881 			  [0] = "486 DX-25/33",
882 			  [1] = "486 DX-50",
883 			  [2] = "486 SX",
884 			  [3] = "486 DX/2",
885 			  [4] = "486 SL",
886 			  [5] = "486 SX/2",
887 			  [7] = "486 DX/2-WB",
888 			  [8] = "486 DX/4",
889 			  [9] = "486 DX/4-WB"
890 		  }
891 		},
892 		{ .family = 5, .model_names =
893 		  {
894 			  [0] = "Pentium 60/66 A-step",
895 			  [1] = "Pentium 60/66",
896 			  [2] = "Pentium 75 - 200",
897 			  [3] = "OverDrive PODP5V83",
898 			  [4] = "Pentium MMX",
899 			  [7] = "Mobile Pentium 75 - 200",
900 			  [8] = "Mobile Pentium MMX",
901 			  [9] = "Quark SoC X1000",
902 		  }
903 		},
904 		{ .family = 6, .model_names =
905 		  {
906 			  [0] = "Pentium Pro A-step",
907 			  [1] = "Pentium Pro",
908 			  [3] = "Pentium II (Klamath)",
909 			  [4] = "Pentium II (Deschutes)",
910 			  [5] = "Pentium II (Deschutes)",
911 			  [6] = "Mobile Pentium II",
912 			  [7] = "Pentium III (Katmai)",
913 			  [8] = "Pentium III (Coppermine)",
914 			  [10] = "Pentium III (Cascades)",
915 			  [11] = "Pentium III (Tualatin)",
916 		  }
917 		},
918 		{ .family = 15, .model_names =
919 		  {
920 			  [0] = "Pentium 4 (Unknown)",
921 			  [1] = "Pentium 4 (Willamette)",
922 			  [2] = "Pentium 4 (Northwood)",
923 			  [4] = "Pentium 4 (Foster)",
924 			  [5] = "Pentium 4 (Foster)",
925 		  }
926 		},
927 	},
928 	.legacy_cache_size = intel_size_cache,
929 #endif
930 	.c_detect_tlb	= intel_detect_tlb,
931 	.c_early_init   = early_init_intel,
932 	.c_init		= init_intel,
933 	.c_bsp_resume	= intel_bsp_resume,
934 	.c_x86_vendor	= X86_VENDOR_INTEL,
935 };
936 
937 cpu_dev_register(intel_cpu_dev);
938 
939