xref: /linux/arch/x86/kernel/cpu/intel.c (revision 667ca1ec7c9eb7ac3b80590b6597151b4c2a750b)
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,	0x84 },
120 	{ INTEL_FAM6_KABYLAKE_DESKTOP,	0x0A,	0x84 },
121 	{ INTEL_FAM6_KABYLAKE_DESKTOP,	0x09,	0x84 },
122 	{ INTEL_FAM6_KABYLAKE_MOBILE,	0x0A,	0x84 },
123 	{ INTEL_FAM6_KABYLAKE_MOBILE,	0x09,	0x84 },
124 	{ INTEL_FAM6_SKYLAKE_X,		0x03,	0x0100013e },
125 	{ INTEL_FAM6_SKYLAKE_X,		0x04,	0x0200003c },
126 	{ INTEL_FAM6_SKYLAKE_MOBILE,	0x03,	0xc2 },
127 	{ INTEL_FAM6_SKYLAKE_DESKTOP,	0x03,	0xc2 },
128 	{ INTEL_FAM6_BROADWELL_CORE,	0x04,	0x28 },
129 	{ INTEL_FAM6_BROADWELL_GT3E,	0x01,	0x1b },
130 	{ INTEL_FAM6_BROADWELL_XEON_D,	0x02,	0x14 },
131 	{ INTEL_FAM6_BROADWELL_XEON_D,	0x03,	0x07000011 },
132 	{ INTEL_FAM6_BROADWELL_X,	0x01,	0x0b000025 },
133 	{ INTEL_FAM6_HASWELL_ULT,	0x01,	0x21 },
134 	{ INTEL_FAM6_HASWELL_GT3E,	0x01,	0x18 },
135 	{ INTEL_FAM6_HASWELL_CORE,	0x03,	0x23 },
136 	{ INTEL_FAM6_HASWELL_X,		0x02,	0x3b },
137 	{ INTEL_FAM6_HASWELL_X,		0x04,	0x10 },
138 	{ INTEL_FAM6_IVYBRIDGE_X,	0x04,	0x42a },
139 	/* Updated in the 20180108 release; blacklist until we know otherwise */
140 	{ INTEL_FAM6_ATOM_GEMINI_LAKE,	0x01,	0x22 },
141 	/* Observed in the wild */
142 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x06,	0x61b },
143 	{ INTEL_FAM6_SANDYBRIDGE_X,	0x07,	0x712 },
144 };
145 
146 static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
147 {
148 	int i;
149 
150 	for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
151 		if (c->x86_model == spectre_bad_microcodes[i].model &&
152 		    c->x86_mask == spectre_bad_microcodes[i].stepping)
153 			return (c->microcode <= spectre_bad_microcodes[i].microcode);
154 	}
155 	return false;
156 }
157 
158 static void early_init_intel(struct cpuinfo_x86 *c)
159 {
160 	u64 misc_enable;
161 
162 	/* Unmask CPUID levels if masked: */
163 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
164 		if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
165 				  MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
166 			c->cpuid_level = cpuid_eax(0);
167 			get_cpu_cap(c);
168 		}
169 	}
170 
171 	if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
172 		(c->x86 == 0x6 && c->x86_model >= 0x0e))
173 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
174 
175 	if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
176 		c->microcode = intel_get_microcode_revision();
177 
178 	/*
179 	 * The Intel SPEC_CTRL CPUID bit implies IBRS and IBPB support,
180 	 * and they also have a different bit for STIBP support. Also,
181 	 * a hypervisor might have set the individual AMD bits even on
182 	 * Intel CPUs, for finer-grained selection of what's available.
183 	 */
184 	if (cpu_has(c, X86_FEATURE_SPEC_CTRL)) {
185 		set_cpu_cap(c, X86_FEATURE_IBRS);
186 		set_cpu_cap(c, X86_FEATURE_IBPB);
187 	}
188 	if (cpu_has(c, X86_FEATURE_INTEL_STIBP))
189 		set_cpu_cap(c, X86_FEATURE_STIBP);
190 
191 	/* Now if any of them are set, check the blacklist and clear the lot */
192 	if ((cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
193 	     cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
194 		pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
195 		clear_cpu_cap(c, X86_FEATURE_IBRS);
196 		clear_cpu_cap(c, X86_FEATURE_IBPB);
197 		clear_cpu_cap(c, X86_FEATURE_STIBP);
198 		clear_cpu_cap(c, X86_FEATURE_SPEC_CTRL);
199 		clear_cpu_cap(c, X86_FEATURE_INTEL_STIBP);
200 	}
201 
202 	/*
203 	 * Atom erratum AAE44/AAF40/AAG38/AAH41:
204 	 *
205 	 * A race condition between speculative fetches and invalidating
206 	 * a large page.  This is worked around in microcode, but we
207 	 * need the microcode to have already been loaded... so if it is
208 	 * not, recommend a BIOS update and disable large pages.
209 	 */
210 	if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_mask <= 2 &&
211 	    c->microcode < 0x20e) {
212 		pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
213 		clear_cpu_cap(c, X86_FEATURE_PSE);
214 	}
215 
216 #ifdef CONFIG_X86_64
217 	set_cpu_cap(c, X86_FEATURE_SYSENTER32);
218 #else
219 	/* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
220 	if (c->x86 == 15 && c->x86_cache_alignment == 64)
221 		c->x86_cache_alignment = 128;
222 #endif
223 
224 	/* CPUID workaround for 0F33/0F34 CPU */
225 	if (c->x86 == 0xF && c->x86_model == 0x3
226 	    && (c->x86_mask == 0x3 || c->x86_mask == 0x4))
227 		c->x86_phys_bits = 36;
228 
229 	/*
230 	 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
231 	 * with P/T states and does not stop in deep C-states.
232 	 *
233 	 * It is also reliable across cores and sockets. (but not across
234 	 * cabinets - we turn it off in that case explicitly.)
235 	 */
236 	if (c->x86_power & (1 << 8)) {
237 		set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
238 		set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
239 	}
240 
241 	/* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
242 	if (c->x86 == 6) {
243 		switch (c->x86_model) {
244 		case 0x27:	/* Penwell */
245 		case 0x35:	/* Cloverview */
246 		case 0x4a:	/* Merrifield */
247 			set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
248 			break;
249 		default:
250 			break;
251 		}
252 	}
253 
254 	/*
255 	 * There is a known erratum on Pentium III and Core Solo
256 	 * and Core Duo CPUs.
257 	 * " Page with PAT set to WC while associated MTRR is UC
258 	 *   may consolidate to UC "
259 	 * Because of this erratum, it is better to stick with
260 	 * setting WC in MTRR rather than using PAT on these CPUs.
261 	 *
262 	 * Enable PAT WC only on P4, Core 2 or later CPUs.
263 	 */
264 	if (c->x86 == 6 && c->x86_model < 15)
265 		clear_cpu_cap(c, X86_FEATURE_PAT);
266 
267 	/*
268 	 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
269 	 * clear the fast string and enhanced fast string CPU capabilities.
270 	 */
271 	if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
272 		rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
273 		if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
274 			pr_info("Disabled fast string operations\n");
275 			setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
276 			setup_clear_cpu_cap(X86_FEATURE_ERMS);
277 		}
278 	}
279 
280 	/*
281 	 * Intel Quark Core DevMan_001.pdf section 6.4.11
282 	 * "The operating system also is required to invalidate (i.e., flush)
283 	 *  the TLB when any changes are made to any of the page table entries.
284 	 *  The operating system must reload CR3 to cause the TLB to be flushed"
285 	 *
286 	 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
287 	 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
288 	 * to be modified.
289 	 */
290 	if (c->x86 == 5 && c->x86_model == 9) {
291 		pr_info("Disabling PGE capability bit\n");
292 		setup_clear_cpu_cap(X86_FEATURE_PGE);
293 	}
294 
295 	if (c->cpuid_level >= 0x00000001) {
296 		u32 eax, ebx, ecx, edx;
297 
298 		cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
299 		/*
300 		 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
301 		 * apicids which are reserved per package. Store the resulting
302 		 * shift value for the package management code.
303 		 */
304 		if (edx & (1U << 28))
305 			c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
306 	}
307 
308 	check_mpx_erratum(c);
309 }
310 
311 #ifdef CONFIG_X86_32
312 /*
313  *	Early probe support logic for ppro memory erratum #50
314  *
315  *	This is called before we do cpu ident work
316  */
317 
318 int ppro_with_ram_bug(void)
319 {
320 	/* Uses data from early_cpu_detect now */
321 	if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
322 	    boot_cpu_data.x86 == 6 &&
323 	    boot_cpu_data.x86_model == 1 &&
324 	    boot_cpu_data.x86_mask < 8) {
325 		pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
326 		return 1;
327 	}
328 	return 0;
329 }
330 
331 static void intel_smp_check(struct cpuinfo_x86 *c)
332 {
333 	/* calling is from identify_secondary_cpu() ? */
334 	if (!c->cpu_index)
335 		return;
336 
337 	/*
338 	 * Mask B, Pentium, but not Pentium MMX
339 	 */
340 	if (c->x86 == 5 &&
341 	    c->x86_mask >= 1 && c->x86_mask <= 4 &&
342 	    c->x86_model <= 3) {
343 		/*
344 		 * Remember we have B step Pentia with bugs
345 		 */
346 		WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
347 				    "with B stepping processors.\n");
348 	}
349 }
350 
351 static int forcepae;
352 static int __init forcepae_setup(char *__unused)
353 {
354 	forcepae = 1;
355 	return 1;
356 }
357 __setup("forcepae", forcepae_setup);
358 
359 static void intel_workarounds(struct cpuinfo_x86 *c)
360 {
361 #ifdef CONFIG_X86_F00F_BUG
362 	/*
363 	 * All models of Pentium and Pentium with MMX technology CPUs
364 	 * have the F0 0F bug, which lets nonprivileged users lock up the
365 	 * system. Announce that the fault handler will be checking for it.
366 	 * The Quark is also family 5, but does not have the same bug.
367 	 */
368 	clear_cpu_bug(c, X86_BUG_F00F);
369 	if (c->x86 == 5 && c->x86_model < 9) {
370 		static int f00f_workaround_enabled;
371 
372 		set_cpu_bug(c, X86_BUG_F00F);
373 		if (!f00f_workaround_enabled) {
374 			pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
375 			f00f_workaround_enabled = 1;
376 		}
377 	}
378 #endif
379 
380 	/*
381 	 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
382 	 * model 3 mask 3
383 	 */
384 	if ((c->x86<<8 | c->x86_model<<4 | c->x86_mask) < 0x633)
385 		clear_cpu_cap(c, X86_FEATURE_SEP);
386 
387 	/*
388 	 * PAE CPUID issue: many Pentium M report no PAE but may have a
389 	 * functionally usable PAE implementation.
390 	 * Forcefully enable PAE if kernel parameter "forcepae" is present.
391 	 */
392 	if (forcepae) {
393 		pr_warn("PAE forced!\n");
394 		set_cpu_cap(c, X86_FEATURE_PAE);
395 		add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
396 	}
397 
398 	/*
399 	 * P4 Xeon erratum 037 workaround.
400 	 * Hardware prefetcher may cause stale data to be loaded into the cache.
401 	 */
402 	if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_mask == 1)) {
403 		if (msr_set_bit(MSR_IA32_MISC_ENABLE,
404 				MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
405 			pr_info("CPU: C0 stepping P4 Xeon detected.\n");
406 			pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
407 		}
408 	}
409 
410 	/*
411 	 * See if we have a good local APIC by checking for buggy Pentia,
412 	 * i.e. all B steppings and the C2 stepping of P54C when using their
413 	 * integrated APIC (see 11AP erratum in "Pentium Processor
414 	 * Specification Update").
415 	 */
416 	if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
417 	    (c->x86_mask < 0x6 || c->x86_mask == 0xb))
418 		set_cpu_bug(c, X86_BUG_11AP);
419 
420 
421 #ifdef CONFIG_X86_INTEL_USERCOPY
422 	/*
423 	 * Set up the preferred alignment for movsl bulk memory moves
424 	 */
425 	switch (c->x86) {
426 	case 4:		/* 486: untested */
427 		break;
428 	case 5:		/* Old Pentia: untested */
429 		break;
430 	case 6:		/* PII/PIII only like movsl with 8-byte alignment */
431 		movsl_mask.mask = 7;
432 		break;
433 	case 15:	/* P4 is OK down to 8-byte alignment */
434 		movsl_mask.mask = 7;
435 		break;
436 	}
437 #endif
438 
439 	intel_smp_check(c);
440 }
441 #else
442 static void intel_workarounds(struct cpuinfo_x86 *c)
443 {
444 }
445 #endif
446 
447 static void srat_detect_node(struct cpuinfo_x86 *c)
448 {
449 #ifdef CONFIG_NUMA
450 	unsigned node;
451 	int cpu = smp_processor_id();
452 
453 	/* Don't do the funky fallback heuristics the AMD version employs
454 	   for now. */
455 	node = numa_cpu_node(cpu);
456 	if (node == NUMA_NO_NODE || !node_online(node)) {
457 		/* reuse the value from init_cpu_to_node() */
458 		node = cpu_to_node(cpu);
459 	}
460 	numa_set_node(cpu, node);
461 #endif
462 }
463 
464 /*
465  * find out the number of processor cores on the die
466  */
467 static int intel_num_cpu_cores(struct cpuinfo_x86 *c)
468 {
469 	unsigned int eax, ebx, ecx, edx;
470 
471 	if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
472 		return 1;
473 
474 	/* Intel has a non-standard dependency on %ecx for this CPUID level. */
475 	cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
476 	if (eax & 0x1f)
477 		return (eax >> 26) + 1;
478 	else
479 		return 1;
480 }
481 
482 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
483 {
484 	/* Intel VMX MSR indicated features */
485 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW	0x00200000
486 #define X86_VMX_FEATURE_PROC_CTLS_VNMI		0x00400000
487 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS	0x80000000
488 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC	0x00000001
489 #define X86_VMX_FEATURE_PROC_CTLS2_EPT		0x00000002
490 #define X86_VMX_FEATURE_PROC_CTLS2_VPID		0x00000020
491 
492 	u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
493 
494 	clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
495 	clear_cpu_cap(c, X86_FEATURE_VNMI);
496 	clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
497 	clear_cpu_cap(c, X86_FEATURE_EPT);
498 	clear_cpu_cap(c, X86_FEATURE_VPID);
499 
500 	rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
501 	msr_ctl = vmx_msr_high | vmx_msr_low;
502 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
503 		set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
504 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
505 		set_cpu_cap(c, X86_FEATURE_VNMI);
506 	if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
507 		rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
508 		      vmx_msr_low, vmx_msr_high);
509 		msr_ctl2 = vmx_msr_high | vmx_msr_low;
510 		if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
511 		    (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
512 			set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
513 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
514 			set_cpu_cap(c, X86_FEATURE_EPT);
515 		if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
516 			set_cpu_cap(c, X86_FEATURE_VPID);
517 	}
518 }
519 
520 static void init_intel_energy_perf(struct cpuinfo_x86 *c)
521 {
522 	u64 epb;
523 
524 	/*
525 	 * Initialize MSR_IA32_ENERGY_PERF_BIAS if not already initialized.
526 	 * (x86_energy_perf_policy(8) is available to change it at run-time.)
527 	 */
528 	if (!cpu_has(c, X86_FEATURE_EPB))
529 		return;
530 
531 	rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
532 	if ((epb & 0xF) != ENERGY_PERF_BIAS_PERFORMANCE)
533 		return;
534 
535 	pr_warn_once("ENERGY_PERF_BIAS: Set to 'normal', was 'performance'\n");
536 	pr_warn_once("ENERGY_PERF_BIAS: View and update with x86_energy_perf_policy(8)\n");
537 	epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
538 	wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
539 }
540 
541 static void intel_bsp_resume(struct cpuinfo_x86 *c)
542 {
543 	/*
544 	 * MSR_IA32_ENERGY_PERF_BIAS is lost across suspend/resume,
545 	 * so reinitialize it properly like during bootup:
546 	 */
547 	init_intel_energy_perf(c);
548 }
549 
550 static void init_cpuid_fault(struct cpuinfo_x86 *c)
551 {
552 	u64 msr;
553 
554 	if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
555 		if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
556 			set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
557 	}
558 }
559 
560 static void init_intel_misc_features(struct cpuinfo_x86 *c)
561 {
562 	u64 msr;
563 
564 	if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
565 		return;
566 
567 	/* Clear all MISC features */
568 	this_cpu_write(msr_misc_features_shadow, 0);
569 
570 	/* Check features and update capabilities and shadow control bits */
571 	init_cpuid_fault(c);
572 	probe_xeon_phi_r3mwait(c);
573 
574 	msr = this_cpu_read(msr_misc_features_shadow);
575 	wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
576 }
577 
578 static void init_intel(struct cpuinfo_x86 *c)
579 {
580 	unsigned int l2 = 0;
581 
582 	early_init_intel(c);
583 
584 	intel_workarounds(c);
585 
586 	/*
587 	 * Detect the extended topology information if available. This
588 	 * will reinitialise the initial_apicid which will be used
589 	 * in init_intel_cacheinfo()
590 	 */
591 	detect_extended_topology(c);
592 
593 	if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
594 		/*
595 		 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
596 		 * detection.
597 		 */
598 		c->x86_max_cores = intel_num_cpu_cores(c);
599 #ifdef CONFIG_X86_32
600 		detect_ht(c);
601 #endif
602 	}
603 
604 	l2 = init_intel_cacheinfo(c);
605 
606 	/* Detect legacy cache sizes if init_intel_cacheinfo did not */
607 	if (l2 == 0) {
608 		cpu_detect_cache_sizes(c);
609 		l2 = c->x86_cache_size;
610 	}
611 
612 	if (c->cpuid_level > 9) {
613 		unsigned eax = cpuid_eax(10);
614 		/* Check for version and the number of counters */
615 		if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
616 			set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
617 	}
618 
619 	if (cpu_has(c, X86_FEATURE_XMM2))
620 		set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
621 
622 	if (boot_cpu_has(X86_FEATURE_DS)) {
623 		unsigned int l1;
624 		rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
625 		if (!(l1 & (1<<11)))
626 			set_cpu_cap(c, X86_FEATURE_BTS);
627 		if (!(l1 & (1<<12)))
628 			set_cpu_cap(c, X86_FEATURE_PEBS);
629 	}
630 
631 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
632 	    (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
633 		set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
634 
635 	if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
636 		((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
637 		set_cpu_bug(c, X86_BUG_MONITOR);
638 
639 #ifdef CONFIG_X86_64
640 	if (c->x86 == 15)
641 		c->x86_cache_alignment = c->x86_clflush_size * 2;
642 	if (c->x86 == 6)
643 		set_cpu_cap(c, X86_FEATURE_REP_GOOD);
644 #else
645 	/*
646 	 * Names for the Pentium II/Celeron processors
647 	 * detectable only by also checking the cache size.
648 	 * Dixon is NOT a Celeron.
649 	 */
650 	if (c->x86 == 6) {
651 		char *p = NULL;
652 
653 		switch (c->x86_model) {
654 		case 5:
655 			if (l2 == 0)
656 				p = "Celeron (Covington)";
657 			else if (l2 == 256)
658 				p = "Mobile Pentium II (Dixon)";
659 			break;
660 
661 		case 6:
662 			if (l2 == 128)
663 				p = "Celeron (Mendocino)";
664 			else if (c->x86_mask == 0 || c->x86_mask == 5)
665 				p = "Celeron-A";
666 			break;
667 
668 		case 8:
669 			if (l2 == 128)
670 				p = "Celeron (Coppermine)";
671 			break;
672 		}
673 
674 		if (p)
675 			strcpy(c->x86_model_id, p);
676 	}
677 
678 	if (c->x86 == 15)
679 		set_cpu_cap(c, X86_FEATURE_P4);
680 	if (c->x86 == 6)
681 		set_cpu_cap(c, X86_FEATURE_P3);
682 #endif
683 
684 	/* Work around errata */
685 	srat_detect_node(c);
686 
687 	if (cpu_has(c, X86_FEATURE_VMX))
688 		detect_vmx_virtcap(c);
689 
690 	init_intel_energy_perf(c);
691 
692 	init_intel_misc_features(c);
693 }
694 
695 #ifdef CONFIG_X86_32
696 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
697 {
698 	/*
699 	 * Intel PIII Tualatin. This comes in two flavours.
700 	 * One has 256kb of cache, the other 512. We have no way
701 	 * to determine which, so we use a boottime override
702 	 * for the 512kb model, and assume 256 otherwise.
703 	 */
704 	if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
705 		size = 256;
706 
707 	/*
708 	 * Intel Quark SoC X1000 contains a 4-way set associative
709 	 * 16K cache with a 16 byte cache line and 256 lines per tag
710 	 */
711 	if ((c->x86 == 5) && (c->x86_model == 9))
712 		size = 16;
713 	return size;
714 }
715 #endif
716 
717 #define TLB_INST_4K	0x01
718 #define TLB_INST_4M	0x02
719 #define TLB_INST_2M_4M	0x03
720 
721 #define TLB_INST_ALL	0x05
722 #define TLB_INST_1G	0x06
723 
724 #define TLB_DATA_4K	0x11
725 #define TLB_DATA_4M	0x12
726 #define TLB_DATA_2M_4M	0x13
727 #define TLB_DATA_4K_4M	0x14
728 
729 #define TLB_DATA_1G	0x16
730 
731 #define TLB_DATA0_4K	0x21
732 #define TLB_DATA0_4M	0x22
733 #define TLB_DATA0_2M_4M	0x23
734 
735 #define STLB_4K		0x41
736 #define STLB_4K_2M	0x42
737 
738 static const struct _tlb_table intel_tlb_table[] = {
739 	{ 0x01, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages, 4-way set associative" },
740 	{ 0x02, TLB_INST_4M,		2,	" TLB_INST 4 MByte pages, full associative" },
741 	{ 0x03, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way set associative" },
742 	{ 0x04, TLB_DATA_4M,		8,	" TLB_DATA 4 MByte pages, 4-way set associative" },
743 	{ 0x05, TLB_DATA_4M,		32,	" TLB_DATA 4 MByte pages, 4-way set associative" },
744 	{ 0x0b, TLB_INST_4M,		4,	" TLB_INST 4 MByte pages, 4-way set associative" },
745 	{ 0x4f, TLB_INST_4K,		32,	" TLB_INST 4 KByte pages */" },
746 	{ 0x50, TLB_INST_ALL,		64,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
747 	{ 0x51, TLB_INST_ALL,		128,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
748 	{ 0x52, TLB_INST_ALL,		256,	" TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
749 	{ 0x55, TLB_INST_2M_4M,		7,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
750 	{ 0x56, TLB_DATA0_4M,		16,	" TLB_DATA0 4 MByte pages, 4-way set associative" },
751 	{ 0x57, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, 4-way associative" },
752 	{ 0x59, TLB_DATA0_4K,		16,	" TLB_DATA0 4 KByte pages, fully associative" },
753 	{ 0x5a, TLB_DATA0_2M_4M,	32,	" TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
754 	{ 0x5b, TLB_DATA_4K_4M,		64,	" TLB_DATA 4 KByte and 4 MByte pages" },
755 	{ 0x5c, TLB_DATA_4K_4M,		128,	" TLB_DATA 4 KByte and 4 MByte pages" },
756 	{ 0x5d, TLB_DATA_4K_4M,		256,	" TLB_DATA 4 KByte and 4 MByte pages" },
757 	{ 0x61, TLB_INST_4K,		48,	" TLB_INST 4 KByte pages, full associative" },
758 	{ 0x63, TLB_DATA_1G,		4,	" TLB_DATA 1 GByte pages, 4-way set associative" },
759 	{ 0x76, TLB_INST_2M_4M,		8,	" TLB_INST 2-MByte or 4-MByte pages, fully associative" },
760 	{ 0xb0, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 4-way set associative" },
761 	{ 0xb1, TLB_INST_2M_4M,		4,	" TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
762 	{ 0xb2, TLB_INST_4K,		64,	" TLB_INST 4KByte pages, 4-way set associative" },
763 	{ 0xb3, TLB_DATA_4K,		128,	" TLB_DATA 4 KByte pages, 4-way set associative" },
764 	{ 0xb4, TLB_DATA_4K,		256,	" TLB_DATA 4 KByte pages, 4-way associative" },
765 	{ 0xb5, TLB_INST_4K,		64,	" TLB_INST 4 KByte pages, 8-way set associative" },
766 	{ 0xb6, TLB_INST_4K,		128,	" TLB_INST 4 KByte pages, 8-way set associative" },
767 	{ 0xba, TLB_DATA_4K,		64,	" TLB_DATA 4 KByte pages, 4-way associative" },
768 	{ 0xc0, TLB_DATA_4K_4M,		8,	" TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
769 	{ 0xc1, STLB_4K_2M,		1024,	" STLB 4 KByte and 2 MByte pages, 8-way associative" },
770 	{ 0xc2, TLB_DATA_2M_4M,		16,	" DTLB 2 MByte/4MByte pages, 4-way associative" },
771 	{ 0xca, STLB_4K,		512,	" STLB 4 KByte pages, 4-way associative" },
772 	{ 0x00, 0, 0 }
773 };
774 
775 static void intel_tlb_lookup(const unsigned char desc)
776 {
777 	unsigned char k;
778 	if (desc == 0)
779 		return;
780 
781 	/* look up this descriptor in the table */
782 	for (k = 0; intel_tlb_table[k].descriptor != desc && \
783 			intel_tlb_table[k].descriptor != 0; k++)
784 		;
785 
786 	if (intel_tlb_table[k].tlb_type == 0)
787 		return;
788 
789 	switch (intel_tlb_table[k].tlb_type) {
790 	case STLB_4K:
791 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
792 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
793 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
794 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
795 		break;
796 	case STLB_4K_2M:
797 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
798 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
799 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
800 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
801 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
802 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
803 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
804 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
805 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
806 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
807 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
808 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
809 		break;
810 	case TLB_INST_ALL:
811 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
812 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
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_INST_4K:
819 		if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
820 			tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
821 		break;
822 	case TLB_INST_4M:
823 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
824 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
825 		break;
826 	case TLB_INST_2M_4M:
827 		if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
828 			tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
829 		if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
830 			tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
831 		break;
832 	case TLB_DATA_4K:
833 	case TLB_DATA0_4K:
834 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
835 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
836 		break;
837 	case TLB_DATA_4M:
838 	case TLB_DATA0_4M:
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_2M_4M:
843 	case TLB_DATA0_2M_4M:
844 		if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
845 			tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
846 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
847 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
848 		break;
849 	case TLB_DATA_4K_4M:
850 		if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
851 			tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
852 		if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
853 			tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
854 		break;
855 	case TLB_DATA_1G:
856 		if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
857 			tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
858 		break;
859 	}
860 }
861 
862 static void intel_detect_tlb(struct cpuinfo_x86 *c)
863 {
864 	int i, j, n;
865 	unsigned int regs[4];
866 	unsigned char *desc = (unsigned char *)regs;
867 
868 	if (c->cpuid_level < 2)
869 		return;
870 
871 	/* Number of times to iterate */
872 	n = cpuid_eax(2) & 0xFF;
873 
874 	for (i = 0 ; i < n ; i++) {
875 		cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
876 
877 		/* If bit 31 is set, this is an unknown format */
878 		for (j = 0 ; j < 3 ; j++)
879 			if (regs[j] & (1 << 31))
880 				regs[j] = 0;
881 
882 		/* Byte 0 is level count, not a descriptor */
883 		for (j = 1 ; j < 16 ; j++)
884 			intel_tlb_lookup(desc[j]);
885 	}
886 }
887 
888 static const struct cpu_dev intel_cpu_dev = {
889 	.c_vendor	= "Intel",
890 	.c_ident	= { "GenuineIntel" },
891 #ifdef CONFIG_X86_32
892 	.legacy_models = {
893 		{ .family = 4, .model_names =
894 		  {
895 			  [0] = "486 DX-25/33",
896 			  [1] = "486 DX-50",
897 			  [2] = "486 SX",
898 			  [3] = "486 DX/2",
899 			  [4] = "486 SL",
900 			  [5] = "486 SX/2",
901 			  [7] = "486 DX/2-WB",
902 			  [8] = "486 DX/4",
903 			  [9] = "486 DX/4-WB"
904 		  }
905 		},
906 		{ .family = 5, .model_names =
907 		  {
908 			  [0] = "Pentium 60/66 A-step",
909 			  [1] = "Pentium 60/66",
910 			  [2] = "Pentium 75 - 200",
911 			  [3] = "OverDrive PODP5V83",
912 			  [4] = "Pentium MMX",
913 			  [7] = "Mobile Pentium 75 - 200",
914 			  [8] = "Mobile Pentium MMX",
915 			  [9] = "Quark SoC X1000",
916 		  }
917 		},
918 		{ .family = 6, .model_names =
919 		  {
920 			  [0] = "Pentium Pro A-step",
921 			  [1] = "Pentium Pro",
922 			  [3] = "Pentium II (Klamath)",
923 			  [4] = "Pentium II (Deschutes)",
924 			  [5] = "Pentium II (Deschutes)",
925 			  [6] = "Mobile Pentium II",
926 			  [7] = "Pentium III (Katmai)",
927 			  [8] = "Pentium III (Coppermine)",
928 			  [10] = "Pentium III (Cascades)",
929 			  [11] = "Pentium III (Tualatin)",
930 		  }
931 		},
932 		{ .family = 15, .model_names =
933 		  {
934 			  [0] = "Pentium 4 (Unknown)",
935 			  [1] = "Pentium 4 (Willamette)",
936 			  [2] = "Pentium 4 (Northwood)",
937 			  [4] = "Pentium 4 (Foster)",
938 			  [5] = "Pentium 4 (Foster)",
939 		  }
940 		},
941 	},
942 	.legacy_cache_size = intel_size_cache,
943 #endif
944 	.c_detect_tlb	= intel_detect_tlb,
945 	.c_early_init   = early_init_intel,
946 	.c_init		= init_intel,
947 	.c_bsp_resume	= intel_bsp_resume,
948 	.c_x86_vendor	= X86_VENDOR_INTEL,
949 };
950 
951 cpu_dev_register(intel_cpu_dev);
952 
953