xref: /linux-6.15/arch/x86/kernel/cpu/intel.c (revision 60fa7426)
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 	/*
148 	 * We know that the hypervisor lie to us on the microcode version so
149 	 * we may as well hope that it is running the correct version.
150 	 */
151 	if (cpu_has(c, X86_FEATURE_HYPERVISOR))
152 		return false;
153 
154 	for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
155 		if (c->x86_model == spectre_bad_microcodes[i].model &&
156 		    c->x86_stepping == spectre_bad_microcodes[i].stepping)
157 			return (c->microcode <= spectre_bad_microcodes[i].microcode);
158 	}
159 	return false;
160 }
161 
162 static void early_init_intel(struct cpuinfo_x86 *c)
163 {
164 	u64 misc_enable;
165 
166 	/* Unmask CPUID levels if masked: */
167 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
168 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
169 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
170 			c->cpuid_level = cpuid_eax(0);
171 			get_cpu_cap(c);
172 		}
173 	}
174 
175 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
176 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
177 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
178 
179 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
180 		c->microcode = intel_get_microcode_revision();
181 
182 	/* Now if any of them are set, check the blacklist and clear the lot */
183 	if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
184 	     cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
185 	     cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
186 	     cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
187 		pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
188 		setup_clear_cpu_cap(X86_FEATURE_IBRS);
189 		setup_clear_cpu_cap(X86_FEATURE_IBPB);
190 		setup_clear_cpu_cap(X86_FEATURE_STIBP);
191 		setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
192 		setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
193 	}
194 
195 	/*
196 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
197 	 *
198 	 * A race condition between speculative fetches and invalidating
199 	 * a large page.  This is worked around in microcode, but we
200 	 * need the microcode to have already been loaded... so if it is
201 	 * not, recommend a BIOS update and disable large pages.
202 	 */
203 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_stepping <= 2 &&
204 	    c->microcode < 0x20e) {
205 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
206 		clear_cpu_cap(c, X86_FEATURE_PSE);
207 	}
208 
209 #ifdef CONFIG_X86_64
210 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
211 #else
212 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
213 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
214 		c->x86_cache_alignment = 128;
215 #endif
216 
217 	/* CPUID workaround for 0F33/0F34 CPU */
218 	if (c->x86 == 0xF && c->x86_model == 0x3
219 	    && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
220 		c->x86_phys_bits = 36;
221 
222 	/*
223 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
224 	 * with P/T states and does not stop in deep C-states.
225 	 *
226 	 * It is also reliable across cores and sockets. (but not across
227 	 * cabinets - we turn it off in that case explicitly.)
228 	 */
229 	if (c->x86_power & (1 << 8)) {
230 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
231 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
232 	}
233 
234 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
235 	if (c->x86 == 6) {
236 		switch (c->x86_model) {
237 		case 0x27:	/* Penwell */
238 		case 0x35:	/* Cloverview */
239 		case 0x4a:	/* Merrifield */
240 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
241 			break;
242 		default:
243 			break;
244 		}
245 	}
246 
247 	/*
248 	 * There is a known erratum on Pentium III and Core Solo
249 	 * and Core Duo CPUs.
250 	 * " Page with PAT set to WC while associated MTRR is UC
251 	 *   may consolidate to UC "
252 	 * Because of this erratum, it is better to stick with
253 	 * setting WC in MTRR rather than using PAT on these CPUs.
254 	 *
255 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
256 	 */
257 	if (c->x86 == 6 && c->x86_model < 15)
258 		clear_cpu_cap(c, X86_FEATURE_PAT);
259 
260 	/*
261 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
262 	 * clear the fast string and enhanced fast string CPU capabilities.
263 	 */
264 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
265 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
266 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
267 			pr_info("Disabled fast string operations\n");
268 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
269 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
270 		}
271 	}
272 
273 	/*
274 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
275 	 * "The operating system also is required to invalidate (i.e., flush)
276 	 *  the TLB when any changes are made to any of the page table entries.
277 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
278 	 *
279 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
280 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
281 	 * to be modified.
282 	 */
283 	if (c->x86 == 5 && c->x86_model == 9) {
284 		pr_info("Disabling PGE capability bit\n");
285 		setup_clear_cpu_cap(X86_FEATURE_PGE);
286 	}
287 
288 	if (c->cpuid_level >= 0x00000001) {
289 		u32 eax, ebx, ecx, edx;
290 
291 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
292 		/*
293 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
294 		 * apicids which are reserved per package. Store the resulting
295 		 * shift value for the package management code.
296 		 */
297 		if (edx & (1U << 28))
298 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
299 	}
300 
301 	check_mpx_erratum(c);
302 }
303 
304 #ifdef CONFIG_X86_32
305 /*
306  *	Early probe support logic for ppro memory erratum #50
307  *
308  *	This is called before we do cpu ident work
309  */
310 
311 int ppro_with_ram_bug(void)
312 {
313 	/* Uses data from early_cpu_detect now */
314 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
315 	    boot_cpu_data.x86 == 6 &&
316 	    boot_cpu_data.x86_model == 1 &&
317 	    boot_cpu_data.x86_stepping < 8) {
318 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
319 		return 1;
320 	}
321 	return 0;
322 }
323 
324 static void intel_smp_check(struct cpuinfo_x86 *c)
325 {
326 	/* calling is from identify_secondary_cpu() ? */
327 	if (!c->cpu_index)
328 		return;
329 
330 	/*
331 	 * Mask B, Pentium, but not Pentium MMX
332 	 */
333 	if (c->x86 == 5 &&
334 	    c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
335 	    c->x86_model <= 3) {
336 		/*
337 		 * Remember we have B step Pentia with bugs
338 		 */
339 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
340 				    "with B stepping processors.\n");
341 	}
342 }
343 
344 static int forcepae;
345 static int __init forcepae_setup(char *__unused)
346 {
347 	forcepae = 1;
348 	return 1;
349 }
350 __setup("forcepae", forcepae_setup);
351 
352 static void intel_workarounds(struct cpuinfo_x86 *c)
353 {
354 #ifdef CONFIG_X86_F00F_BUG
355 	/*
356 	 * All models of Pentium and Pentium with MMX technology CPUs
357 	 * have the F0 0F bug, which lets nonprivileged users lock up the
358 	 * system. Announce that the fault handler will be checking for it.
359 	 * The Quark is also family 5, but does not have the same bug.
360 	 */
361 	clear_cpu_bug(c, X86_BUG_F00F);
362 	if (c->x86 == 5 && c->x86_model < 9) {
363 		static int f00f_workaround_enabled;
364 
365 		set_cpu_bug(c, X86_BUG_F00F);
366 		if (!f00f_workaround_enabled) {
367 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
368 			f00f_workaround_enabled = 1;
369 		}
370 	}
371 #endif
372 
373 	/*
374 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
375 	 * model 3 mask 3
376 	 */
377 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
378 		clear_cpu_cap(c, X86_FEATURE_SEP);
379 
380 	/*
381 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
382 	 * functionally usable PAE implementation.
383 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
384 	 */
385 	if (forcepae) {
386 		pr_warn("PAE forced!\n");
387 		set_cpu_cap(c, X86_FEATURE_PAE);
388 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
389 	}
390 
391 	/*
392 	 * P4 Xeon erratum 037 workaround.
393 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
394 	 */
395 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
396 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
397 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
398 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
399 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
400 		}
401 	}
402 
403 	/*
404 	 * See if we have a good local APIC by checking for buggy Pentia,
405 	 * i.e. all B steppings and the C2 stepping of P54C when using their
406 	 * integrated APIC (see 11AP erratum in "Pentium Processor
407 	 * Specification Update").
408 	 */
409 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
410 	    (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
411 		set_cpu_bug(c, X86_BUG_11AP);
412 
413 
414 #ifdef CONFIG_X86_INTEL_USERCOPY
415 	/*
416 	 * Set up the preferred alignment for movsl bulk memory moves
417 	 */
418 	switch (c->x86) {
419 	case 4:		/* 486: untested */
420 		break;
421 	case 5:		/* Old Pentia: untested */
422 		break;
423 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
424 		movsl_mask.mask = 7;
425 		break;
426 	case 15:	/* P4 is OK down to 8-byte alignment */
427 		movsl_mask.mask = 7;
428 		break;
429 	}
430 #endif
431 
432 	intel_smp_check(c);
433 }
434 #else
435 static void intel_workarounds(struct cpuinfo_x86 *c)
436 {
437 }
438 #endif
439 
440 static void srat_detect_node(struct cpuinfo_x86 *c)
441 {
442 #ifdef CONFIG_NUMA
443 	unsigned node;
444 	int cpu = smp_processor_id();
445 
446 	/* Don't do the funky fallback heuristics the AMD version employs
447 	   for now. */
448 	node = numa_cpu_node(cpu);
449 	if (node == NUMA_NO_NODE || !node_online(node)) {
450 		/* reuse the value from init_cpu_to_node() */
451 		node = cpu_to_node(cpu);
452 	}
453 	numa_set_node(cpu, node);
454 #endif
455 }
456 
457 /*
458  * find out the number of processor cores on the die
459  */
460 static int intel_num_cpu_cores(struct cpuinfo_x86 *c)
461 {
462 	unsigned int eax, ebx, ecx, edx;
463 
464 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
465 		return 1;
466 
467 	/* Intel has a non-standard dependency on %ecx for this CPUID level. */
468 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
469 	if (eax & 0x1f)
470 		return (eax >> 26) + 1;
471 	else
472 		return 1;
473 }
474 
475 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
476 {
477 	/* Intel VMX MSR indicated features */
478 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
479 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
480 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
481 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
482 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
483 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
484 
485 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
486 
487 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
488 	clear_cpu_cap(c, X86_FEATURE_VNMI);
489 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
490 	clear_cpu_cap(c, X86_FEATURE_EPT);
491 	clear_cpu_cap(c, X86_FEATURE_VPID);
492 
493 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
494 	msr_ctl = vmx_msr_high | vmx_msr_low;
495 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
496 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
497 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
498 		set_cpu_cap(c, X86_FEATURE_VNMI);
499 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
500 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
501 		      vmx_msr_low, vmx_msr_high);
502 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
503 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
504 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
505 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
506 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
507 			set_cpu_cap(c, X86_FEATURE_EPT);
508 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
509 			set_cpu_cap(c, X86_FEATURE_VPID);
510 	}
511 }
512 
513 static void init_intel_energy_perf(struct cpuinfo_x86 *c)
514 {
515 	u64 epb;
516 
517 	/*
518 	 * Initialize MSR_IA32_ENERGY_PERF_BIAS if not already initialized.
519 	 * (x86_energy_perf_policy(8) is available to change it at run-time.)
520 	 */
521 	if (!cpu_has(c, X86_FEATURE_EPB))
522 		return;
523 
524 	rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
525 	if ((epb & 0xF) != ENERGY_PERF_BIAS_PERFORMANCE)
526 		return;
527 
528 	pr_warn_once("ENERGY_PERF_BIAS: Set to 'normal', was 'performance'\n");
529 	pr_warn_once("ENERGY_PERF_BIAS: View and update with x86_energy_perf_policy(8)\n");
530 	epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
531 	wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
532 }
533 
534 static void intel_bsp_resume(struct cpuinfo_x86 *c)
535 {
536 	/*
537 	 * MSR_IA32_ENERGY_PERF_BIAS is lost across suspend/resume,
538 	 * so reinitialize it properly like during bootup:
539 	 */
540 	init_intel_energy_perf(c);
541 }
542 
543 static void init_cpuid_fault(struct cpuinfo_x86 *c)
544 {
545 	u64 msr;
546 
547 	if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
548 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
549 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
550 	}
551 }
552 
553 static void init_intel_misc_features(struct cpuinfo_x86 *c)
554 {
555 	u64 msr;
556 
557 	if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
558 		return;
559 
560 	/* Clear all MISC features */
561 	this_cpu_write(msr_misc_features_shadow, 0);
562 
563 	/* Check features and update capabilities and shadow control bits */
564 	init_cpuid_fault(c);
565 	probe_xeon_phi_r3mwait(c);
566 
567 	msr = this_cpu_read(msr_misc_features_shadow);
568 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
569 }
570 
571 static void init_intel(struct cpuinfo_x86 *c)
572 {
573 	unsigned int l2 = 0;
574 
575 	early_init_intel(c);
576 
577 	intel_workarounds(c);
578 
579 	/*
580 	 * Detect the extended topology information if available. This
581 	 * will reinitialise the initial_apicid which will be used
582 	 * in init_intel_cacheinfo()
583 	 */
584 	detect_extended_topology(c);
585 
586 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
587 		/*
588 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
589 		 * detection.
590 		 */
591 		c->x86_max_cores = intel_num_cpu_cores(c);
592 #ifdef CONFIG_X86_32
593 		detect_ht(c);
594 #endif
595 	}
596 
597 	l2 = init_intel_cacheinfo(c);
598 
599 	/* Detect legacy cache sizes if init_intel_cacheinfo did not */
600 	if (l2 == 0) {
601 		cpu_detect_cache_sizes(c);
602 		l2 = c->x86_cache_size;
603 	}
604 
605 	if (c->cpuid_level > 9) {
606 		unsigned eax = cpuid_eax(10);
607 		/* Check for version and the number of counters */
608 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
609 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
610 	}
611 
612 	if (cpu_has(c, X86_FEATURE_XMM2))
613 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
614 
615 	if (boot_cpu_has(X86_FEATURE_DS)) {
616 		unsigned int l1;
617 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
618 		if (!(l1 & (1<<11)))
619 			set_cpu_cap(c, X86_FEATURE_BTS);
620 		if (!(l1 & (1<<12)))
621 			set_cpu_cap(c, X86_FEATURE_PEBS);
622 	}
623 
624 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
625 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
626 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
627 
628 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
629 		((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
630 		set_cpu_bug(c, X86_BUG_MONITOR);
631 
632 #ifdef CONFIG_X86_64
633 	if (c->x86 == 15)
634 		c->x86_cache_alignment = c->x86_clflush_size * 2;
635 	if (c->x86 == 6)
636 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
637 #else
638 	/*
639 	 * Names for the Pentium II/Celeron processors
640 	 * detectable only by also checking the cache size.
641 	 * Dixon is NOT a Celeron.
642 	 */
643 	if (c->x86 == 6) {
644 		char *p = NULL;
645 
646 		switch (c->x86_model) {
647 		case 5:
648 			if (l2 == 0)
649 				p = "Celeron (Covington)";
650 			else if (l2 == 256)
651 				p = "Mobile Pentium II (Dixon)";
652 			break;
653 
654 		case 6:
655 			if (l2 == 128)
656 				p = "Celeron (Mendocino)";
657 			else if (c->x86_stepping == 0 || c->x86_stepping == 5)
658 				p = "Celeron-A";
659 			break;
660 
661 		case 8:
662 			if (l2 == 128)
663 				p = "Celeron (Coppermine)";
664 			break;
665 		}
666 
667 		if (p)
668 			strcpy(c->x86_model_id, p);
669 	}
670 
671 	if (c->x86 == 15)
672 		set_cpu_cap(c, X86_FEATURE_P4);
673 	if (c->x86 == 6)
674 		set_cpu_cap(c, X86_FEATURE_P3);
675 #endif
676 
677 	/* Work around errata */
678 	srat_detect_node(c);
679 
680 	if (cpu_has(c, X86_FEATURE_VMX))
681 		detect_vmx_virtcap(c);
682 
683 	init_intel_energy_perf(c);
684 
685 	init_intel_misc_features(c);
686 }
687 
688 #ifdef CONFIG_X86_32
689 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
690 {
691 	/*
692 	 * Intel PIII Tualatin. This comes in two flavours.
693 	 * One has 256kb of cache, the other 512. We have no way
694 	 * to determine which, so we use a boottime override
695 	 * for the 512kb model, and assume 256 otherwise.
696 	 */
697 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
698 		size = 256;
699 
700 	/*
701 	 * Intel Quark SoC X1000 contains a 4-way set associative
702 	 * 16K cache with a 16 byte cache line and 256 lines per tag
703 	 */
704 	if ((c->x86 == 5) && (c->x86_model == 9))
705 		size = 16;
706 	return size;
707 }
708 #endif
709 
710 #define TLB_INST_4K	0x01
711 #define TLB_INST_4M	0x02
712 #define TLB_INST_2M_4M	0x03
713 
714 #define TLB_INST_ALL	0x05
715 #define TLB_INST_1G	0x06
716 
717 #define TLB_DATA_4K	0x11
718 #define TLB_DATA_4M	0x12
719 #define TLB_DATA_2M_4M	0x13
720 #define TLB_DATA_4K_4M	0x14
721 
722 #define TLB_DATA_1G	0x16
723 
724 #define TLB_DATA0_4K	0x21
725 #define TLB_DATA0_4M	0x22
726 #define TLB_DATA0_2M_4M	0x23
727 
728 #define STLB_4K		0x41
729 #define STLB_4K_2M	0x42
730 
731 static const struct _tlb_table intel_tlb_table[] = {
732 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
733 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
734 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
735 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
736 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
737 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
738 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages */" },
739 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
740 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
741 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
742 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
743 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
744 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
745 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
746 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
747 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
748 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
749 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
750 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
751 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
752 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
753 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
754 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
755 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
756 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
757 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
758 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
759 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
760 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
761 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
762 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
763 	{ 0xc2, TLB_DATA_2M_4M,		16,	" DTLB 2 MByte/4MByte pages, 4-way associative" },
764 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
765 	{ 0x00, 0, 0 }
766 };
767 
768 static void intel_tlb_lookup(const unsigned char desc)
769 {
770 	unsigned char k;
771 	if (desc == 0)
772 		return;
773 
774 	/* look up this descriptor in the table */
775 	for (k = 0; intel_tlb_table[k].descriptor != desc && \
776 			intel_tlb_table[k].descriptor != 0; k++)
777 		;
778 
779 	if (intel_tlb_table[k].tlb_type == 0)
780 		return;
781 
782 	switch (intel_tlb_table[k].tlb_type) {
783 	case STLB_4K:
784 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
785 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
786 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
787 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
788 		break;
789 	case STLB_4K_2M:
790 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
791 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
792 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
793 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
794 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
795 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
796 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
797 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
798 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
799 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
800 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
801 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
802 		break;
803 	case TLB_INST_ALL:
804 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
805 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
806 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
807 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
808 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
809 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
810 		break;
811 	case TLB_INST_4K:
812 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
813 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
814 		break;
815 	case TLB_INST_4M:
816 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
817 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
818 		break;
819 	case TLB_INST_2M_4M:
820 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
821 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
822 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
823 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
824 		break;
825 	case TLB_DATA_4K:
826 	case TLB_DATA0_4K:
827 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
828 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
829 		break;
830 	case TLB_DATA_4M:
831 	case TLB_DATA0_4M:
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_2M_4M:
836 	case TLB_DATA0_2M_4M:
837 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
838 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
839 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
840 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
841 		break;
842 	case TLB_DATA_4K_4M:
843 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
844 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
845 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
846 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
847 		break;
848 	case TLB_DATA_1G:
849 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
850 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
851 		break;
852 	}
853 }
854 
855 static void intel_detect_tlb(struct cpuinfo_x86 *c)
856 {
857 	int i, j, n;
858 	unsigned int regs[4];
859 	unsigned char *desc = (unsigned char *)regs;
860 
861 	if (c->cpuid_level < 2)
862 		return;
863 
864 	/* Number of times to iterate */
865 	n = cpuid_eax(2) & 0xFF;
866 
867 	for (i = 0 ; i < n ; i++) {
868 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
869 
870 		/* If bit 31 is set, this is an unknown format */
871 		for (j = 0 ; j < 3 ; j++)
872 			if (regs[j] & (1 << 31))
873 				regs[j] = 0;
874 
875 		/* Byte 0 is level count, not a descriptor */
876 		for (j = 1 ; j < 16 ; j++)
877 			intel_tlb_lookup(desc[j]);
878 	}
879 }
880 
881 static const struct cpu_dev intel_cpu_dev = {
882 	.c_vendor	= "Intel",
883 	.c_ident	= { "GenuineIntel" },
884 #ifdef CONFIG_X86_32
885 	.legacy_models = {
886 		{ .family = 4, .model_names =
887 		  {
888 			  [0] = "486 DX-25/33",
889 			  [1] = "486 DX-50",
890 			  [2] = "486 SX",
891 			  [3] = "486 DX/2",
892 			  [4] = "486 SL",
893 			  [5] = "486 SX/2",
894 			  [7] = "486 DX/2-WB",
895 			  [8] = "486 DX/4",
896 			  [9] = "486 DX/4-WB"
897 		  }
898 		},
899 		{ .family = 5, .model_names =
900 		  {
901 			  [0] = "Pentium 60/66 A-step",
902 			  [1] = "Pentium 60/66",
903 			  [2] = "Pentium 75 - 200",
904 			  [3] = "OverDrive PODP5V83",
905 			  [4] = "Pentium MMX",
906 			  [7] = "Mobile Pentium 75 - 200",
907 			  [8] = "Mobile Pentium MMX",
908 			  [9] = "Quark SoC X1000",
909 		  }
910 		},
911 		{ .family = 6, .model_names =
912 		  {
913 			  [0] = "Pentium Pro A-step",
914 			  [1] = "Pentium Pro",
915 			  [3] = "Pentium II (Klamath)",
916 			  [4] = "Pentium II (Deschutes)",
917 			  [5] = "Pentium II (Deschutes)",
918 			  [6] = "Mobile Pentium II",
919 			  [7] = "Pentium III (Katmai)",
920 			  [8] = "Pentium III (Coppermine)",
921 			  [10] = "Pentium III (Cascades)",
922 			  [11] = "Pentium III (Tualatin)",
923 		  }
924 		},
925 		{ .family = 15, .model_names =
926 		  {
927 			  [0] = "Pentium 4 (Unknown)",
928 			  [1] = "Pentium 4 (Willamette)",
929 			  [2] = "Pentium 4 (Northwood)",
930 			  [4] = "Pentium 4 (Foster)",
931 			  [5] = "Pentium 4 (Foster)",
932 		  }
933 		},
934 	},
935 	.legacy_cache_size = intel_size_cache,
936 #endif
937 	.c_detect_tlb	= intel_detect_tlb,
938 	.c_early_init   = early_init_intel,
939 	.c_init		= init_intel,
940 	.c_bsp_resume	= intel_bsp_resume,
941 	.c_x86_vendor	= X86_VENDOR_INTEL,
942 };
943 
944 cpu_dev_register(intel_cpu_dev);
945 
946