Loading...
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 */
38static int forcempx;
39
40static 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
48void 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
69static bool ring3mwait_disabled __read_mostly;
70
71static int __init ring3mwait_disable(char *__unused)
72{
73 ring3mwait_disabled = true;
74 return 0;
75}
76__setup("ring3mwait=disable", ring3mwait_disable);
77
78static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c)
79{
80 /*
81 * Ring 3 MONITOR/MWAIT feature cannot be detected without
82 * cpu model and family comparison.
83 */
84 if (c->x86 != 6)
85 return;
86 switch (c->x86_model) {
87 case INTEL_FAM6_XEON_PHI_KNL:
88 case INTEL_FAM6_XEON_PHI_KNM:
89 break;
90 default:
91 return;
92 }
93
94 if (ring3mwait_disabled)
95 return;
96
97 set_cpu_cap(c, X86_FEATURE_RING3MWAIT);
98 this_cpu_or(msr_misc_features_shadow,
99 1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT);
100
101 if (c == &boot_cpu_data)
102 ELF_HWCAP2 |= HWCAP2_RING3MWAIT;
103}
104
105/*
106 * Early microcode releases for the Spectre v2 mitigation were broken.
107 * Information taken from;
108 * - https://newsroom.intel.com/wp-content/uploads/sites/11/2018/03/microcode-update-guidance.pdf
109 * - https://kb.vmware.com/s/article/52345
110 * - Microcode revisions observed in the wild
111 * - Release note from 20180108 microcode release
112 */
113struct sku_microcode {
114 u8 model;
115 u8 stepping;
116 u32 microcode;
117};
118static const struct sku_microcode spectre_bad_microcodes[] = {
119 { INTEL_FAM6_KABYLAKE_DESKTOP, 0x0B, 0x80 },
120 { INTEL_FAM6_KABYLAKE_DESKTOP, 0x0A, 0x80 },
121 { INTEL_FAM6_KABYLAKE_DESKTOP, 0x09, 0x80 },
122 { INTEL_FAM6_KABYLAKE_MOBILE, 0x0A, 0x80 },
123 { INTEL_FAM6_KABYLAKE_MOBILE, 0x09, 0x80 },
124 { INTEL_FAM6_SKYLAKE_X, 0x03, 0x0100013e },
125 { INTEL_FAM6_SKYLAKE_X, 0x04, 0x0200003c },
126 { INTEL_FAM6_BROADWELL_CORE, 0x04, 0x28 },
127 { INTEL_FAM6_BROADWELL_GT3E, 0x01, 0x1b },
128 { INTEL_FAM6_BROADWELL_XEON_D, 0x02, 0x14 },
129 { INTEL_FAM6_BROADWELL_XEON_D, 0x03, 0x07000011 },
130 { INTEL_FAM6_BROADWELL_X, 0x01, 0x0b000025 },
131 { INTEL_FAM6_HASWELL_ULT, 0x01, 0x21 },
132 { INTEL_FAM6_HASWELL_GT3E, 0x01, 0x18 },
133 { INTEL_FAM6_HASWELL_CORE, 0x03, 0x23 },
134 { INTEL_FAM6_HASWELL_X, 0x02, 0x3b },
135 { INTEL_FAM6_HASWELL_X, 0x04, 0x10 },
136 { INTEL_FAM6_IVYBRIDGE_X, 0x04, 0x42a },
137 /* Observed in the wild */
138 { INTEL_FAM6_SANDYBRIDGE_X, 0x06, 0x61b },
139 { INTEL_FAM6_SANDYBRIDGE_X, 0x07, 0x712 },
140};
141
142static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
143{
144 int i;
145
146 /*
147 * We know that the hypervisor lie to us on the microcode version so
148 * we may as well hope that it is running the correct version.
149 */
150 if (cpu_has(c, X86_FEATURE_HYPERVISOR))
151 return false;
152
153 for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
154 if (c->x86_model == spectre_bad_microcodes[i].model &&
155 c->x86_stepping == spectre_bad_microcodes[i].stepping)
156 return (c->microcode <= spectre_bad_microcodes[i].microcode);
157 }
158 return false;
159}
160
161static void early_init_intel(struct cpuinfo_x86 *c)
162{
163 u64 misc_enable;
164
165 /* Unmask CPUID levels if masked: */
166 if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
167 if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
168 MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
169 c->cpuid_level = cpuid_eax(0);
170 get_cpu_cap(c);
171 }
172 }
173
174 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
175 (c->x86 == 0x6 && c->x86_model >= 0x0e))
176 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
177
178 if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
179 c->microcode = intel_get_microcode_revision();
180
181 /* Now if any of them are set, check the blacklist and clear the lot */
182 if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
183 cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
184 cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
185 cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
186 pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
187 setup_clear_cpu_cap(X86_FEATURE_IBRS);
188 setup_clear_cpu_cap(X86_FEATURE_IBPB);
189 setup_clear_cpu_cap(X86_FEATURE_STIBP);
190 setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
191 setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL);
192 setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
193 setup_clear_cpu_cap(X86_FEATURE_SSBD);
194 setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD);
195 }
196
197 /*
198 * Atom erratum AAE44/AAF40/AAG38/AAH41:
199 *
200 * A race condition between speculative fetches and invalidating
201 * a large page. This is worked around in microcode, but we
202 * need the microcode to have already been loaded... so if it is
203 * not, recommend a BIOS update and disable large pages.
204 */
205 if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_stepping <= 2 &&
206 c->microcode < 0x20e) {
207 pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
208 clear_cpu_cap(c, X86_FEATURE_PSE);
209 }
210
211#ifdef CONFIG_X86_64
212 set_cpu_cap(c, X86_FEATURE_SYSENTER32);
213#else
214 /* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
215 if (c->x86 == 15 && c->x86_cache_alignment == 64)
216 c->x86_cache_alignment = 128;
217#endif
218
219 /* CPUID workaround for 0F33/0F34 CPU */
220 if (c->x86 == 0xF && c->x86_model == 0x3
221 && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
222 c->x86_phys_bits = 36;
223
224 /*
225 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
226 * with P/T states and does not stop in deep C-states.
227 *
228 * It is also reliable across cores and sockets. (but not across
229 * cabinets - we turn it off in that case explicitly.)
230 */
231 if (c->x86_power & (1 << 8)) {
232 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
233 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
234 }
235
236 /* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
237 if (c->x86 == 6) {
238 switch (c->x86_model) {
239 case 0x27: /* Penwell */
240 case 0x35: /* Cloverview */
241 case 0x4a: /* Merrifield */
242 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
243 break;
244 default:
245 break;
246 }
247 }
248
249 /*
250 * There is a known erratum on Pentium III and Core Solo
251 * and Core Duo CPUs.
252 * " Page with PAT set to WC while associated MTRR is UC
253 * may consolidate to UC "
254 * Because of this erratum, it is better to stick with
255 * setting WC in MTRR rather than using PAT on these CPUs.
256 *
257 * Enable PAT WC only on P4, Core 2 or later CPUs.
258 */
259 if (c->x86 == 6 && c->x86_model < 15)
260 clear_cpu_cap(c, X86_FEATURE_PAT);
261
262 /*
263 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
264 * clear the fast string and enhanced fast string CPU capabilities.
265 */
266 if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
267 rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
268 if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
269 pr_info("Disabled fast string operations\n");
270 setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
271 setup_clear_cpu_cap(X86_FEATURE_ERMS);
272 }
273 }
274
275 /*
276 * Intel Quark Core DevMan_001.pdf section 6.4.11
277 * "The operating system also is required to invalidate (i.e., flush)
278 * the TLB when any changes are made to any of the page table entries.
279 * The operating system must reload CR3 to cause the TLB to be flushed"
280 *
281 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
282 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
283 * to be modified.
284 */
285 if (c->x86 == 5 && c->x86_model == 9) {
286 pr_info("Disabling PGE capability bit\n");
287 setup_clear_cpu_cap(X86_FEATURE_PGE);
288 }
289
290 if (c->cpuid_level >= 0x00000001) {
291 u32 eax, ebx, ecx, edx;
292
293 cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
294 /*
295 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
296 * apicids which are reserved per package. Store the resulting
297 * shift value for the package management code.
298 */
299 if (edx & (1U << 28))
300 c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
301 }
302
303 check_mpx_erratum(c);
304}
305
306#ifdef CONFIG_X86_32
307/*
308 * Early probe support logic for ppro memory erratum #50
309 *
310 * This is called before we do cpu ident work
311 */
312
313int ppro_with_ram_bug(void)
314{
315 /* Uses data from early_cpu_detect now */
316 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
317 boot_cpu_data.x86 == 6 &&
318 boot_cpu_data.x86_model == 1 &&
319 boot_cpu_data.x86_stepping < 8) {
320 pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
321 return 1;
322 }
323 return 0;
324}
325
326static void intel_smp_check(struct cpuinfo_x86 *c)
327{
328 /* calling is from identify_secondary_cpu() ? */
329 if (!c->cpu_index)
330 return;
331
332 /*
333 * Mask B, Pentium, but not Pentium MMX
334 */
335 if (c->x86 == 5 &&
336 c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
337 c->x86_model <= 3) {
338 /*
339 * Remember we have B step Pentia with bugs
340 */
341 WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
342 "with B stepping processors.\n");
343 }
344}
345
346static int forcepae;
347static int __init forcepae_setup(char *__unused)
348{
349 forcepae = 1;
350 return 1;
351}
352__setup("forcepae", forcepae_setup);
353
354static void intel_workarounds(struct cpuinfo_x86 *c)
355{
356#ifdef CONFIG_X86_F00F_BUG
357 /*
358 * All models of Pentium and Pentium with MMX technology CPUs
359 * have the F0 0F bug, which lets nonprivileged users lock up the
360 * system. Announce that the fault handler will be checking for it.
361 * The Quark is also family 5, but does not have the same bug.
362 */
363 clear_cpu_bug(c, X86_BUG_F00F);
364 if (c->x86 == 5 && c->x86_model < 9) {
365 static int f00f_workaround_enabled;
366
367 set_cpu_bug(c, X86_BUG_F00F);
368 if (!f00f_workaround_enabled) {
369 pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
370 f00f_workaround_enabled = 1;
371 }
372 }
373#endif
374
375 /*
376 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
377 * model 3 mask 3
378 */
379 if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
380 clear_cpu_cap(c, X86_FEATURE_SEP);
381
382 /*
383 * PAE CPUID issue: many Pentium M report no PAE but may have a
384 * functionally usable PAE implementation.
385 * Forcefully enable PAE if kernel parameter "forcepae" is present.
386 */
387 if (forcepae) {
388 pr_warn("PAE forced!\n");
389 set_cpu_cap(c, X86_FEATURE_PAE);
390 add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
391 }
392
393 /*
394 * P4 Xeon erratum 037 workaround.
395 * Hardware prefetcher may cause stale data to be loaded into the cache.
396 */
397 if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
398 if (msr_set_bit(MSR_IA32_MISC_ENABLE,
399 MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
400 pr_info("CPU: C0 stepping P4 Xeon detected.\n");
401 pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
402 }
403 }
404
405 /*
406 * See if we have a good local APIC by checking for buggy Pentia,
407 * i.e. all B steppings and the C2 stepping of P54C when using their
408 * integrated APIC (see 11AP erratum in "Pentium Processor
409 * Specification Update").
410 */
411 if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
412 (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
413 set_cpu_bug(c, X86_BUG_11AP);
414
415
416#ifdef CONFIG_X86_INTEL_USERCOPY
417 /*
418 * Set up the preferred alignment for movsl bulk memory moves
419 */
420 switch (c->x86) {
421 case 4: /* 486: untested */
422 break;
423 case 5: /* Old Pentia: untested */
424 break;
425 case 6: /* PII/PIII only like movsl with 8-byte alignment */
426 movsl_mask.mask = 7;
427 break;
428 case 15: /* P4 is OK down to 8-byte alignment */
429 movsl_mask.mask = 7;
430 break;
431 }
432#endif
433
434 intel_smp_check(c);
435}
436#else
437static void intel_workarounds(struct cpuinfo_x86 *c)
438{
439}
440#endif
441
442static void srat_detect_node(struct cpuinfo_x86 *c)
443{
444#ifdef CONFIG_NUMA
445 unsigned node;
446 int cpu = smp_processor_id();
447
448 /* Don't do the funky fallback heuristics the AMD version employs
449 for now. */
450 node = numa_cpu_node(cpu);
451 if (node == NUMA_NO_NODE || !node_online(node)) {
452 /* reuse the value from init_cpu_to_node() */
453 node = cpu_to_node(cpu);
454 }
455 numa_set_node(cpu, node);
456#endif
457}
458
459/*
460 * find out the number of processor cores on the die
461 */
462static int intel_num_cpu_cores(struct cpuinfo_x86 *c)
463{
464 unsigned int eax, ebx, ecx, edx;
465
466 if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
467 return 1;
468
469 /* Intel has a non-standard dependency on %ecx for this CPUID level. */
470 cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
471 if (eax & 0x1f)
472 return (eax >> 26) + 1;
473 else
474 return 1;
475}
476
477static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
478{
479 /* Intel VMX MSR indicated features */
480#define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW 0x00200000
481#define X86_VMX_FEATURE_PROC_CTLS_VNMI 0x00400000
482#define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS 0x80000000
483#define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC 0x00000001
484#define X86_VMX_FEATURE_PROC_CTLS2_EPT 0x00000002
485#define X86_VMX_FEATURE_PROC_CTLS2_VPID 0x00000020
486
487 u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
488
489 clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
490 clear_cpu_cap(c, X86_FEATURE_VNMI);
491 clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
492 clear_cpu_cap(c, X86_FEATURE_EPT);
493 clear_cpu_cap(c, X86_FEATURE_VPID);
494
495 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
496 msr_ctl = vmx_msr_high | vmx_msr_low;
497 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
498 set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
499 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
500 set_cpu_cap(c, X86_FEATURE_VNMI);
501 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
502 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
503 vmx_msr_low, vmx_msr_high);
504 msr_ctl2 = vmx_msr_high | vmx_msr_low;
505 if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
506 (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
507 set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
508 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
509 set_cpu_cap(c, X86_FEATURE_EPT);
510 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
511 set_cpu_cap(c, X86_FEATURE_VPID);
512 }
513}
514
515#define MSR_IA32_TME_ACTIVATE 0x982
516
517/* Helpers to access TME_ACTIVATE MSR */
518#define TME_ACTIVATE_LOCKED(x) (x & 0x1)
519#define TME_ACTIVATE_ENABLED(x) (x & 0x2)
520
521#define TME_ACTIVATE_POLICY(x) ((x >> 4) & 0xf) /* Bits 7:4 */
522#define TME_ACTIVATE_POLICY_AES_XTS_128 0
523
524#define TME_ACTIVATE_KEYID_BITS(x) ((x >> 32) & 0xf) /* Bits 35:32 */
525
526#define TME_ACTIVATE_CRYPTO_ALGS(x) ((x >> 48) & 0xffff) /* Bits 63:48 */
527#define TME_ACTIVATE_CRYPTO_AES_XTS_128 1
528
529/* Values for mktme_status (SW only construct) */
530#define MKTME_ENABLED 0
531#define MKTME_DISABLED 1
532#define MKTME_UNINITIALIZED 2
533static int mktme_status = MKTME_UNINITIALIZED;
534
535static void detect_tme(struct cpuinfo_x86 *c)
536{
537 u64 tme_activate, tme_policy, tme_crypto_algs;
538 int keyid_bits = 0, nr_keyids = 0;
539 static u64 tme_activate_cpu0 = 0;
540
541 rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
542
543 if (mktme_status != MKTME_UNINITIALIZED) {
544 if (tme_activate != tme_activate_cpu0) {
545 /* Broken BIOS? */
546 pr_err_once("x86/tme: configuration is inconsistent between CPUs\n");
547 pr_err_once("x86/tme: MKTME is not usable\n");
548 mktme_status = MKTME_DISABLED;
549
550 /* Proceed. We may need to exclude bits from x86_phys_bits. */
551 }
552 } else {
553 tme_activate_cpu0 = tme_activate;
554 }
555
556 if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
557 pr_info_once("x86/tme: not enabled by BIOS\n");
558 mktme_status = MKTME_DISABLED;
559 return;
560 }
561
562 if (mktme_status != MKTME_UNINITIALIZED)
563 goto detect_keyid_bits;
564
565 pr_info("x86/tme: enabled by BIOS\n");
566
567 tme_policy = TME_ACTIVATE_POLICY(tme_activate);
568 if (tme_policy != TME_ACTIVATE_POLICY_AES_XTS_128)
569 pr_warn("x86/tme: Unknown policy is active: %#llx\n", tme_policy);
570
571 tme_crypto_algs = TME_ACTIVATE_CRYPTO_ALGS(tme_activate);
572 if (!(tme_crypto_algs & TME_ACTIVATE_CRYPTO_AES_XTS_128)) {
573 pr_err("x86/mktme: No known encryption algorithm is supported: %#llx\n",
574 tme_crypto_algs);
575 mktme_status = MKTME_DISABLED;
576 }
577detect_keyid_bits:
578 keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
579 nr_keyids = (1UL << keyid_bits) - 1;
580 if (nr_keyids) {
581 pr_info_once("x86/mktme: enabled by BIOS\n");
582 pr_info_once("x86/mktme: %d KeyIDs available\n", nr_keyids);
583 } else {
584 pr_info_once("x86/mktme: disabled by BIOS\n");
585 }
586
587 if (mktme_status == MKTME_UNINITIALIZED) {
588 /* MKTME is usable */
589 mktme_status = MKTME_ENABLED;
590 }
591
592 /*
593 * KeyID bits effectively lower the number of physical address
594 * bits. Update cpuinfo_x86::x86_phys_bits accordingly.
595 */
596 c->x86_phys_bits -= keyid_bits;
597}
598
599static void init_intel_energy_perf(struct cpuinfo_x86 *c)
600{
601 u64 epb;
602
603 /*
604 * Initialize MSR_IA32_ENERGY_PERF_BIAS if not already initialized.
605 * (x86_energy_perf_policy(8) is available to change it at run-time.)
606 */
607 if (!cpu_has(c, X86_FEATURE_EPB))
608 return;
609
610 rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
611 if ((epb & 0xF) != ENERGY_PERF_BIAS_PERFORMANCE)
612 return;
613
614 pr_warn_once("ENERGY_PERF_BIAS: Set to 'normal', was 'performance'\n");
615 pr_warn_once("ENERGY_PERF_BIAS: View and update with x86_energy_perf_policy(8)\n");
616 epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
617 wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
618}
619
620static void intel_bsp_resume(struct cpuinfo_x86 *c)
621{
622 /*
623 * MSR_IA32_ENERGY_PERF_BIAS is lost across suspend/resume,
624 * so reinitialize it properly like during bootup:
625 */
626 init_intel_energy_perf(c);
627}
628
629static void init_cpuid_fault(struct cpuinfo_x86 *c)
630{
631 u64 msr;
632
633 if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
634 if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
635 set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
636 }
637}
638
639static void init_intel_misc_features(struct cpuinfo_x86 *c)
640{
641 u64 msr;
642
643 if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
644 return;
645
646 /* Clear all MISC features */
647 this_cpu_write(msr_misc_features_shadow, 0);
648
649 /* Check features and update capabilities and shadow control bits */
650 init_cpuid_fault(c);
651 probe_xeon_phi_r3mwait(c);
652
653 msr = this_cpu_read(msr_misc_features_shadow);
654 wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
655}
656
657static void init_intel(struct cpuinfo_x86 *c)
658{
659 unsigned int l2 = 0;
660
661 early_init_intel(c);
662
663 intel_workarounds(c);
664
665 /*
666 * Detect the extended topology information if available. This
667 * will reinitialise the initial_apicid which will be used
668 * in init_intel_cacheinfo()
669 */
670 detect_extended_topology(c);
671
672 if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
673 /*
674 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
675 * detection.
676 */
677 c->x86_max_cores = intel_num_cpu_cores(c);
678#ifdef CONFIG_X86_32
679 detect_ht(c);
680#endif
681 }
682
683 l2 = init_intel_cacheinfo(c);
684
685 /* Detect legacy cache sizes if init_intel_cacheinfo did not */
686 if (l2 == 0) {
687 cpu_detect_cache_sizes(c);
688 l2 = c->x86_cache_size;
689 }
690
691 if (c->cpuid_level > 9) {
692 unsigned eax = cpuid_eax(10);
693 /* Check for version and the number of counters */
694 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
695 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
696 }
697
698 if (cpu_has(c, X86_FEATURE_XMM2))
699 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
700
701 if (boot_cpu_has(X86_FEATURE_DS)) {
702 unsigned int l1;
703 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
704 if (!(l1 & (1<<11)))
705 set_cpu_cap(c, X86_FEATURE_BTS);
706 if (!(l1 & (1<<12)))
707 set_cpu_cap(c, X86_FEATURE_PEBS);
708 }
709
710 if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
711 (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
712 set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
713
714 if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
715 ((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
716 set_cpu_bug(c, X86_BUG_MONITOR);
717
718#ifdef CONFIG_X86_64
719 if (c->x86 == 15)
720 c->x86_cache_alignment = c->x86_clflush_size * 2;
721 if (c->x86 == 6)
722 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
723#else
724 /*
725 * Names for the Pentium II/Celeron processors
726 * detectable only by also checking the cache size.
727 * Dixon is NOT a Celeron.
728 */
729 if (c->x86 == 6) {
730 char *p = NULL;
731
732 switch (c->x86_model) {
733 case 5:
734 if (l2 == 0)
735 p = "Celeron (Covington)";
736 else if (l2 == 256)
737 p = "Mobile Pentium II (Dixon)";
738 break;
739
740 case 6:
741 if (l2 == 128)
742 p = "Celeron (Mendocino)";
743 else if (c->x86_stepping == 0 || c->x86_stepping == 5)
744 p = "Celeron-A";
745 break;
746
747 case 8:
748 if (l2 == 128)
749 p = "Celeron (Coppermine)";
750 break;
751 }
752
753 if (p)
754 strcpy(c->x86_model_id, p);
755 }
756
757 if (c->x86 == 15)
758 set_cpu_cap(c, X86_FEATURE_P4);
759 if (c->x86 == 6)
760 set_cpu_cap(c, X86_FEATURE_P3);
761#endif
762
763 /* Work around errata */
764 srat_detect_node(c);
765
766 if (cpu_has(c, X86_FEATURE_VMX))
767 detect_vmx_virtcap(c);
768
769 if (cpu_has(c, X86_FEATURE_TME))
770 detect_tme(c);
771
772 init_intel_energy_perf(c);
773
774 init_intel_misc_features(c);
775}
776
777#ifdef CONFIG_X86_32
778static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
779{
780 /*
781 * Intel PIII Tualatin. This comes in two flavours.
782 * One has 256kb of cache, the other 512. We have no way
783 * to determine which, so we use a boottime override
784 * for the 512kb model, and assume 256 otherwise.
785 */
786 if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
787 size = 256;
788
789 /*
790 * Intel Quark SoC X1000 contains a 4-way set associative
791 * 16K cache with a 16 byte cache line and 256 lines per tag
792 */
793 if ((c->x86 == 5) && (c->x86_model == 9))
794 size = 16;
795 return size;
796}
797#endif
798
799#define TLB_INST_4K 0x01
800#define TLB_INST_4M 0x02
801#define TLB_INST_2M_4M 0x03
802
803#define TLB_INST_ALL 0x05
804#define TLB_INST_1G 0x06
805
806#define TLB_DATA_4K 0x11
807#define TLB_DATA_4M 0x12
808#define TLB_DATA_2M_4M 0x13
809#define TLB_DATA_4K_4M 0x14
810
811#define TLB_DATA_1G 0x16
812
813#define TLB_DATA0_4K 0x21
814#define TLB_DATA0_4M 0x22
815#define TLB_DATA0_2M_4M 0x23
816
817#define STLB_4K 0x41
818#define STLB_4K_2M 0x42
819
820static const struct _tlb_table intel_tlb_table[] = {
821 { 0x01, TLB_INST_4K, 32, " TLB_INST 4 KByte pages, 4-way set associative" },
822 { 0x02, TLB_INST_4M, 2, " TLB_INST 4 MByte pages, full associative" },
823 { 0x03, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way set associative" },
824 { 0x04, TLB_DATA_4M, 8, " TLB_DATA 4 MByte pages, 4-way set associative" },
825 { 0x05, TLB_DATA_4M, 32, " TLB_DATA 4 MByte pages, 4-way set associative" },
826 { 0x0b, TLB_INST_4M, 4, " TLB_INST 4 MByte pages, 4-way set associative" },
827 { 0x4f, TLB_INST_4K, 32, " TLB_INST 4 KByte pages */" },
828 { 0x50, TLB_INST_ALL, 64, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
829 { 0x51, TLB_INST_ALL, 128, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
830 { 0x52, TLB_INST_ALL, 256, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
831 { 0x55, TLB_INST_2M_4M, 7, " TLB_INST 2-MByte or 4-MByte pages, fully associative" },
832 { 0x56, TLB_DATA0_4M, 16, " TLB_DATA0 4 MByte pages, 4-way set associative" },
833 { 0x57, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, 4-way associative" },
834 { 0x59, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, fully associative" },
835 { 0x5a, TLB_DATA0_2M_4M, 32, " TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
836 { 0x5b, TLB_DATA_4K_4M, 64, " TLB_DATA 4 KByte and 4 MByte pages" },
837 { 0x5c, TLB_DATA_4K_4M, 128, " TLB_DATA 4 KByte and 4 MByte pages" },
838 { 0x5d, TLB_DATA_4K_4M, 256, " TLB_DATA 4 KByte and 4 MByte pages" },
839 { 0x61, TLB_INST_4K, 48, " TLB_INST 4 KByte pages, full associative" },
840 { 0x63, TLB_DATA_1G, 4, " TLB_DATA 1 GByte pages, 4-way set associative" },
841 { 0x6b, TLB_DATA_4K, 256, " TLB_DATA 4 KByte pages, 8-way associative" },
842 { 0x6c, TLB_DATA_2M_4M, 128, " TLB_DATA 2 MByte or 4 MByte pages, 8-way associative" },
843 { 0x6d, TLB_DATA_1G, 16, " TLB_DATA 1 GByte pages, fully associative" },
844 { 0x76, TLB_INST_2M_4M, 8, " TLB_INST 2-MByte or 4-MByte pages, fully associative" },
845 { 0xb0, TLB_INST_4K, 128, " TLB_INST 4 KByte pages, 4-way set associative" },
846 { 0xb1, TLB_INST_2M_4M, 4, " TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
847 { 0xb2, TLB_INST_4K, 64, " TLB_INST 4KByte pages, 4-way set associative" },
848 { 0xb3, TLB_DATA_4K, 128, " TLB_DATA 4 KByte pages, 4-way set associative" },
849 { 0xb4, TLB_DATA_4K, 256, " TLB_DATA 4 KByte pages, 4-way associative" },
850 { 0xb5, TLB_INST_4K, 64, " TLB_INST 4 KByte pages, 8-way set associative" },
851 { 0xb6, TLB_INST_4K, 128, " TLB_INST 4 KByte pages, 8-way set associative" },
852 { 0xba, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way associative" },
853 { 0xc0, TLB_DATA_4K_4M, 8, " TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
854 { 0xc1, STLB_4K_2M, 1024, " STLB 4 KByte and 2 MByte pages, 8-way associative" },
855 { 0xc2, TLB_DATA_2M_4M, 16, " DTLB 2 MByte/4MByte pages, 4-way associative" },
856 { 0xca, STLB_4K, 512, " STLB 4 KByte pages, 4-way associative" },
857 { 0x00, 0, 0 }
858};
859
860static void intel_tlb_lookup(const unsigned char desc)
861{
862 unsigned char k;
863 if (desc == 0)
864 return;
865
866 /* look up this descriptor in the table */
867 for (k = 0; intel_tlb_table[k].descriptor != desc && \
868 intel_tlb_table[k].descriptor != 0; k++)
869 ;
870
871 if (intel_tlb_table[k].tlb_type == 0)
872 return;
873
874 switch (intel_tlb_table[k].tlb_type) {
875 case STLB_4K:
876 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
877 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
878 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
879 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
880 break;
881 case STLB_4K_2M:
882 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
883 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
884 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
885 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
886 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
887 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
888 if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
889 tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
890 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
891 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
892 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
893 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
894 break;
895 case TLB_INST_ALL:
896 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
897 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
898 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
899 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
900 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
901 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
902 break;
903 case TLB_INST_4K:
904 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
905 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
906 break;
907 case TLB_INST_4M:
908 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
909 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
910 break;
911 case TLB_INST_2M_4M:
912 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
913 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
914 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
915 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
916 break;
917 case TLB_DATA_4K:
918 case TLB_DATA0_4K:
919 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
920 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
921 break;
922 case TLB_DATA_4M:
923 case TLB_DATA0_4M:
924 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
925 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
926 break;
927 case TLB_DATA_2M_4M:
928 case TLB_DATA0_2M_4M:
929 if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
930 tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
931 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
932 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
933 break;
934 case TLB_DATA_4K_4M:
935 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
936 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
937 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
938 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
939 break;
940 case TLB_DATA_1G:
941 if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
942 tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
943 break;
944 }
945}
946
947static void intel_detect_tlb(struct cpuinfo_x86 *c)
948{
949 int i, j, n;
950 unsigned int regs[4];
951 unsigned char *desc = (unsigned char *)regs;
952
953 if (c->cpuid_level < 2)
954 return;
955
956 /* Number of times to iterate */
957 n = cpuid_eax(2) & 0xFF;
958
959 for (i = 0 ; i < n ; i++) {
960 cpuid(2, ®s[0], ®s[1], ®s[2], ®s[3]);
961
962 /* If bit 31 is set, this is an unknown format */
963 for (j = 0 ; j < 3 ; j++)
964 if (regs[j] & (1 << 31))
965 regs[j] = 0;
966
967 /* Byte 0 is level count, not a descriptor */
968 for (j = 1 ; j < 16 ; j++)
969 intel_tlb_lookup(desc[j]);
970 }
971}
972
973static const struct cpu_dev intel_cpu_dev = {
974 .c_vendor = "Intel",
975 .c_ident = { "GenuineIntel" },
976#ifdef CONFIG_X86_32
977 .legacy_models = {
978 { .family = 4, .model_names =
979 {
980 [0] = "486 DX-25/33",
981 [1] = "486 DX-50",
982 [2] = "486 SX",
983 [3] = "486 DX/2",
984 [4] = "486 SL",
985 [5] = "486 SX/2",
986 [7] = "486 DX/2-WB",
987 [8] = "486 DX/4",
988 [9] = "486 DX/4-WB"
989 }
990 },
991 { .family = 5, .model_names =
992 {
993 [0] = "Pentium 60/66 A-step",
994 [1] = "Pentium 60/66",
995 [2] = "Pentium 75 - 200",
996 [3] = "OverDrive PODP5V83",
997 [4] = "Pentium MMX",
998 [7] = "Mobile Pentium 75 - 200",
999 [8] = "Mobile Pentium MMX",
1000 [9] = "Quark SoC X1000",
1001 }
1002 },
1003 { .family = 6, .model_names =
1004 {
1005 [0] = "Pentium Pro A-step",
1006 [1] = "Pentium Pro",
1007 [3] = "Pentium II (Klamath)",
1008 [4] = "Pentium II (Deschutes)",
1009 [5] = "Pentium II (Deschutes)",
1010 [6] = "Mobile Pentium II",
1011 [7] = "Pentium III (Katmai)",
1012 [8] = "Pentium III (Coppermine)",
1013 [10] = "Pentium III (Cascades)",
1014 [11] = "Pentium III (Tualatin)",
1015 }
1016 },
1017 { .family = 15, .model_names =
1018 {
1019 [0] = "Pentium 4 (Unknown)",
1020 [1] = "Pentium 4 (Willamette)",
1021 [2] = "Pentium 4 (Northwood)",
1022 [4] = "Pentium 4 (Foster)",
1023 [5] = "Pentium 4 (Foster)",
1024 }
1025 },
1026 },
1027 .legacy_cache_size = intel_size_cache,
1028#endif
1029 .c_detect_tlb = intel_detect_tlb,
1030 .c_early_init = early_init_intel,
1031 .c_init = init_intel,
1032 .c_bsp_resume = intel_bsp_resume,
1033 .c_x86_vendor = X86_VENDOR_INTEL,
1034};
1035
1036cpu_dev_register(intel_cpu_dev);
1037
1// SPDX-License-Identifier: GPL-2.0
2#include <linux/kernel.h>
3#include <linux/pgtable.h>
4
5#include <linux/string.h>
6#include <linux/bitops.h>
7#include <linux/smp.h>
8#include <linux/sched.h>
9#include <linux/sched/clock.h>
10#include <linux/thread_info.h>
11#include <linux/init.h>
12#include <linux/uaccess.h>
13
14#include <asm/cpufeature.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.h>
20#include <asm/hwcap2.h>
21#include <asm/elf.h>
22#include <asm/cpu_device_id.h>
23#include <asm/resctrl.h>
24#include <asm/numa.h>
25#include <asm/thermal.h>
26
27#ifdef CONFIG_X86_64
28#include <linux/topology.h>
29#endif
30
31#include "cpu.h"
32
33#ifdef CONFIG_X86_LOCAL_APIC
34#include <asm/mpspec.h>
35#include <asm/apic.h>
36#endif
37
38/*
39 * Processors which have self-snooping capability can handle conflicting
40 * memory type across CPUs by snooping its own cache. However, there exists
41 * CPU models in which having conflicting memory types still leads to
42 * unpredictable behavior, machine check errors, or hangs. Clear this
43 * feature to prevent its use on machines with known erratas.
44 */
45static void check_memory_type_self_snoop_errata(struct cpuinfo_x86 *c)
46{
47 switch (c->x86_vfm) {
48 case INTEL_CORE_YONAH:
49 case INTEL_CORE2_MEROM:
50 case INTEL_CORE2_MEROM_L:
51 case INTEL_CORE2_PENRYN:
52 case INTEL_CORE2_DUNNINGTON:
53 case INTEL_NEHALEM:
54 case INTEL_NEHALEM_G:
55 case INTEL_NEHALEM_EP:
56 case INTEL_NEHALEM_EX:
57 case INTEL_WESTMERE:
58 case INTEL_WESTMERE_EP:
59 case INTEL_SANDYBRIDGE:
60 setup_clear_cpu_cap(X86_FEATURE_SELFSNOOP);
61 }
62}
63
64static bool ring3mwait_disabled __read_mostly;
65
66static int __init ring3mwait_disable(char *__unused)
67{
68 ring3mwait_disabled = true;
69 return 1;
70}
71__setup("ring3mwait=disable", ring3mwait_disable);
72
73static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c)
74{
75 /*
76 * Ring 3 MONITOR/MWAIT feature cannot be detected without
77 * cpu model and family comparison.
78 */
79 if (c->x86 != 6)
80 return;
81 switch (c->x86_vfm) {
82 case INTEL_XEON_PHI_KNL:
83 case INTEL_XEON_PHI_KNM:
84 break;
85 default:
86 return;
87 }
88
89 if (ring3mwait_disabled)
90 return;
91
92 set_cpu_cap(c, X86_FEATURE_RING3MWAIT);
93 this_cpu_or(msr_misc_features_shadow,
94 1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT);
95
96 if (c == &boot_cpu_data)
97 ELF_HWCAP2 |= HWCAP2_RING3MWAIT;
98}
99
100/*
101 * Early microcode releases for the Spectre v2 mitigation were broken.
102 * Information taken from;
103 * - https://newsroom.intel.com/wp-content/uploads/sites/11/2018/03/microcode-update-guidance.pdf
104 * - https://kb.vmware.com/s/article/52345
105 * - Microcode revisions observed in the wild
106 * - Release note from 20180108 microcode release
107 */
108struct sku_microcode {
109 u32 vfm;
110 u8 stepping;
111 u32 microcode;
112};
113static const struct sku_microcode spectre_bad_microcodes[] = {
114 { INTEL_KABYLAKE, 0x0B, 0x80 },
115 { INTEL_KABYLAKE, 0x0A, 0x80 },
116 { INTEL_KABYLAKE, 0x09, 0x80 },
117 { INTEL_KABYLAKE_L, 0x0A, 0x80 },
118 { INTEL_KABYLAKE_L, 0x09, 0x80 },
119 { INTEL_SKYLAKE_X, 0x03, 0x0100013e },
120 { INTEL_SKYLAKE_X, 0x04, 0x0200003c },
121 { INTEL_BROADWELL, 0x04, 0x28 },
122 { INTEL_BROADWELL_G, 0x01, 0x1b },
123 { INTEL_BROADWELL_D, 0x02, 0x14 },
124 { INTEL_BROADWELL_D, 0x03, 0x07000011 },
125 { INTEL_BROADWELL_X, 0x01, 0x0b000025 },
126 { INTEL_HASWELL_L, 0x01, 0x21 },
127 { INTEL_HASWELL_G, 0x01, 0x18 },
128 { INTEL_HASWELL, 0x03, 0x23 },
129 { INTEL_HASWELL_X, 0x02, 0x3b },
130 { INTEL_HASWELL_X, 0x04, 0x10 },
131 { INTEL_IVYBRIDGE_X, 0x04, 0x42a },
132 /* Observed in the wild */
133 { INTEL_SANDYBRIDGE_X, 0x06, 0x61b },
134 { INTEL_SANDYBRIDGE_X, 0x07, 0x712 },
135};
136
137static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
138{
139 int i;
140
141 /*
142 * We know that the hypervisor lie to us on the microcode version so
143 * we may as well hope that it is running the correct version.
144 */
145 if (cpu_has(c, X86_FEATURE_HYPERVISOR))
146 return false;
147
148 for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
149 if (c->x86_vfm == spectre_bad_microcodes[i].vfm &&
150 c->x86_stepping == spectre_bad_microcodes[i].stepping)
151 return (c->microcode <= spectre_bad_microcodes[i].microcode);
152 }
153 return false;
154}
155
156#define MSR_IA32_TME_ACTIVATE 0x982
157
158/* Helpers to access TME_ACTIVATE MSR */
159#define TME_ACTIVATE_LOCKED(x) (x & 0x1)
160#define TME_ACTIVATE_ENABLED(x) (x & 0x2)
161
162#define TME_ACTIVATE_KEYID_BITS(x) ((x >> 32) & 0xf) /* Bits 35:32 */
163
164static void detect_tme_early(struct cpuinfo_x86 *c)
165{
166 u64 tme_activate;
167 int keyid_bits;
168
169 rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
170
171 if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
172 pr_info_once("x86/tme: not enabled by BIOS\n");
173 clear_cpu_cap(c, X86_FEATURE_TME);
174 return;
175 }
176 pr_info_once("x86/tme: enabled by BIOS\n");
177 keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
178 if (!keyid_bits)
179 return;
180
181 /*
182 * KeyID bits are set by BIOS and can be present regardless
183 * of whether the kernel is using them. They effectively lower
184 * the number of physical address bits.
185 *
186 * Update cpuinfo_x86::x86_phys_bits accordingly.
187 */
188 c->x86_phys_bits -= keyid_bits;
189 pr_info_once("x86/mktme: BIOS enabled: x86_phys_bits reduced by %d\n",
190 keyid_bits);
191}
192
193void intel_unlock_cpuid_leafs(struct cpuinfo_x86 *c)
194{
195 if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
196 return;
197
198 if (c->x86 < 6 || (c->x86 == 6 && c->x86_model < 0xd))
199 return;
200
201 /*
202 * The BIOS can have limited CPUID to leaf 2, which breaks feature
203 * enumeration. Unlock it and update the maximum leaf info.
204 */
205 if (msr_clear_bit(MSR_IA32_MISC_ENABLE, MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0)
206 c->cpuid_level = cpuid_eax(0);
207}
208
209static void early_init_intel(struct cpuinfo_x86 *c)
210{
211 u64 misc_enable;
212
213 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
214 (c->x86 == 0x6 && c->x86_model >= 0x0e))
215 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
216
217 if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
218 c->microcode = intel_get_microcode_revision();
219
220 /* Now if any of them are set, check the blacklist and clear the lot */
221 if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
222 cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
223 cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
224 cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
225 pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
226 setup_clear_cpu_cap(X86_FEATURE_IBRS);
227 setup_clear_cpu_cap(X86_FEATURE_IBPB);
228 setup_clear_cpu_cap(X86_FEATURE_STIBP);
229 setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
230 setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL);
231 setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
232 setup_clear_cpu_cap(X86_FEATURE_SSBD);
233 setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD);
234 }
235
236 /*
237 * Atom erratum AAE44/AAF40/AAG38/AAH41:
238 *
239 * A race condition between speculative fetches and invalidating
240 * a large page. This is worked around in microcode, but we
241 * need the microcode to have already been loaded... so if it is
242 * not, recommend a BIOS update and disable large pages.
243 */
244 if (c->x86_vfm == INTEL_ATOM_BONNELL && c->x86_stepping <= 2 &&
245 c->microcode < 0x20e) {
246 pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
247 clear_cpu_cap(c, X86_FEATURE_PSE);
248 }
249
250#ifdef CONFIG_X86_64
251 set_cpu_cap(c, X86_FEATURE_SYSENTER32);
252#else
253 /* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
254 if (c->x86 == 15 && c->x86_cache_alignment == 64)
255 c->x86_cache_alignment = 128;
256#endif
257
258 /* CPUID workaround for 0F33/0F34 CPU */
259 if (c->x86 == 0xF && c->x86_model == 0x3
260 && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
261 c->x86_phys_bits = 36;
262
263 /*
264 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
265 * with P/T states and does not stop in deep C-states.
266 *
267 * It is also reliable across cores and sockets. (but not across
268 * cabinets - we turn it off in that case explicitly.)
269 */
270 if (c->x86_power & (1 << 8)) {
271 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
272 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
273 }
274
275 /* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
276 switch (c->x86_vfm) {
277 case INTEL_ATOM_SALTWELL_MID:
278 case INTEL_ATOM_SALTWELL_TABLET:
279 case INTEL_ATOM_SILVERMONT_MID:
280 case INTEL_ATOM_AIRMONT_NP:
281 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
282 break;
283 }
284
285 /*
286 * PAT is broken on early family 6 CPUs, the last of which
287 * is "Yonah" where the erratum is named "AN7":
288 *
289 * Page with PAT (Page Attribute Table) Set to USWC
290 * (Uncacheable Speculative Write Combine) While
291 * Associated MTRR (Memory Type Range Register) Is UC
292 * (Uncacheable) May Consolidate to UC
293 *
294 * Disable PAT and fall back to MTRR on these CPUs.
295 */
296 if (c->x86_vfm >= INTEL_PENTIUM_PRO &&
297 c->x86_vfm <= INTEL_CORE_YONAH)
298 clear_cpu_cap(c, X86_FEATURE_PAT);
299
300 /*
301 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
302 * clear the fast string and enhanced fast string CPU capabilities.
303 */
304 if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
305 rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
306 if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
307 pr_info("Disabled fast string operations\n");
308 setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
309 setup_clear_cpu_cap(X86_FEATURE_ERMS);
310 }
311 }
312
313 /*
314 * Intel Quark Core DevMan_001.pdf section 6.4.11
315 * "The operating system also is required to invalidate (i.e., flush)
316 * the TLB when any changes are made to any of the page table entries.
317 * The operating system must reload CR3 to cause the TLB to be flushed"
318 *
319 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
320 * should be false so that __flush_tlb_all() causes CR3 instead of CR4.PGE
321 * to be modified.
322 */
323 if (c->x86_vfm == INTEL_QUARK_X1000) {
324 pr_info("Disabling PGE capability bit\n");
325 setup_clear_cpu_cap(X86_FEATURE_PGE);
326 }
327
328 check_memory_type_self_snoop_errata(c);
329
330 /*
331 * Adjust the number of physical bits early because it affects the
332 * valid bits of the MTRR mask registers.
333 */
334 if (cpu_has(c, X86_FEATURE_TME))
335 detect_tme_early(c);
336}
337
338static void bsp_init_intel(struct cpuinfo_x86 *c)
339{
340 resctrl_cpu_detect(c);
341}
342
343#ifdef CONFIG_X86_32
344/*
345 * Early probe support logic for ppro memory erratum #50
346 *
347 * This is called before we do cpu ident work
348 */
349
350int ppro_with_ram_bug(void)
351{
352 /* Uses data from early_cpu_detect now */
353 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
354 boot_cpu_data.x86 == 6 &&
355 boot_cpu_data.x86_model == 1 &&
356 boot_cpu_data.x86_stepping < 8) {
357 pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
358 return 1;
359 }
360 return 0;
361}
362
363static void intel_smp_check(struct cpuinfo_x86 *c)
364{
365 /* calling is from identify_secondary_cpu() ? */
366 if (!c->cpu_index)
367 return;
368
369 /*
370 * Mask B, Pentium, but not Pentium MMX
371 */
372 if (c->x86 == 5 &&
373 c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
374 c->x86_model <= 3) {
375 /*
376 * Remember we have B step Pentia with bugs
377 */
378 WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
379 "with B stepping processors.\n");
380 }
381}
382
383static int forcepae;
384static int __init forcepae_setup(char *__unused)
385{
386 forcepae = 1;
387 return 1;
388}
389__setup("forcepae", forcepae_setup);
390
391static void intel_workarounds(struct cpuinfo_x86 *c)
392{
393#ifdef CONFIG_X86_F00F_BUG
394 /*
395 * All models of Pentium and Pentium with MMX technology CPUs
396 * have the F0 0F bug, which lets nonprivileged users lock up the
397 * system. Announce that the fault handler will be checking for it.
398 * The Quark is also family 5, but does not have the same bug.
399 */
400 clear_cpu_bug(c, X86_BUG_F00F);
401 if (c->x86 == 5 && c->x86_model < 9) {
402 static int f00f_workaround_enabled;
403
404 set_cpu_bug(c, X86_BUG_F00F);
405 if (!f00f_workaround_enabled) {
406 pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
407 f00f_workaround_enabled = 1;
408 }
409 }
410#endif
411
412 /*
413 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
414 * model 3 mask 3
415 */
416 if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
417 clear_cpu_cap(c, X86_FEATURE_SEP);
418
419 /*
420 * PAE CPUID issue: many Pentium M report no PAE but may have a
421 * functionally usable PAE implementation.
422 * Forcefully enable PAE if kernel parameter "forcepae" is present.
423 */
424 if (forcepae) {
425 pr_warn("PAE forced!\n");
426 set_cpu_cap(c, X86_FEATURE_PAE);
427 add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
428 }
429
430 /*
431 * P4 Xeon erratum 037 workaround.
432 * Hardware prefetcher may cause stale data to be loaded into the cache.
433 */
434 if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
435 if (msr_set_bit(MSR_IA32_MISC_ENABLE,
436 MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
437 pr_info("CPU: C0 stepping P4 Xeon detected.\n");
438 pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
439 }
440 }
441
442 /*
443 * See if we have a good local APIC by checking for buggy Pentia,
444 * i.e. all B steppings and the C2 stepping of P54C when using their
445 * integrated APIC (see 11AP erratum in "Pentium Processor
446 * Specification Update").
447 */
448 if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
449 (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
450 set_cpu_bug(c, X86_BUG_11AP);
451
452
453#ifdef CONFIG_X86_INTEL_USERCOPY
454 /*
455 * Set up the preferred alignment for movsl bulk memory moves
456 */
457 switch (c->x86) {
458 case 4: /* 486: untested */
459 break;
460 case 5: /* Old Pentia: untested */
461 break;
462 case 6: /* PII/PIII only like movsl with 8-byte alignment */
463 movsl_mask.mask = 7;
464 break;
465 case 15: /* P4 is OK down to 8-byte alignment */
466 movsl_mask.mask = 7;
467 break;
468 }
469#endif
470
471 intel_smp_check(c);
472}
473#else
474static void intel_workarounds(struct cpuinfo_x86 *c)
475{
476}
477#endif
478
479static void srat_detect_node(struct cpuinfo_x86 *c)
480{
481#ifdef CONFIG_NUMA
482 unsigned node;
483 int cpu = smp_processor_id();
484
485 /* Don't do the funky fallback heuristics the AMD version employs
486 for now. */
487 node = numa_cpu_node(cpu);
488 if (node == NUMA_NO_NODE || !node_online(node)) {
489 /* reuse the value from init_cpu_to_node() */
490 node = cpu_to_node(cpu);
491 }
492 numa_set_node(cpu, node);
493#endif
494}
495
496static void init_cpuid_fault(struct cpuinfo_x86 *c)
497{
498 u64 msr;
499
500 if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
501 if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
502 set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
503 }
504}
505
506static void init_intel_misc_features(struct cpuinfo_x86 *c)
507{
508 u64 msr;
509
510 if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
511 return;
512
513 /* Clear all MISC features */
514 this_cpu_write(msr_misc_features_shadow, 0);
515
516 /* Check features and update capabilities and shadow control bits */
517 init_cpuid_fault(c);
518 probe_xeon_phi_r3mwait(c);
519
520 msr = this_cpu_read(msr_misc_features_shadow);
521 wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
522}
523
524static void init_intel(struct cpuinfo_x86 *c)
525{
526 early_init_intel(c);
527
528 intel_workarounds(c);
529
530 init_intel_cacheinfo(c);
531
532 if (c->cpuid_level > 9) {
533 unsigned eax = cpuid_eax(10);
534 /* Check for version and the number of counters */
535 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
536 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
537 }
538
539 if (cpu_has(c, X86_FEATURE_XMM2))
540 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
541
542 if (boot_cpu_has(X86_FEATURE_DS)) {
543 unsigned int l1, l2;
544
545 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
546 if (!(l1 & MSR_IA32_MISC_ENABLE_BTS_UNAVAIL))
547 set_cpu_cap(c, X86_FEATURE_BTS);
548 if (!(l1 & MSR_IA32_MISC_ENABLE_PEBS_UNAVAIL))
549 set_cpu_cap(c, X86_FEATURE_PEBS);
550 }
551
552 if (boot_cpu_has(X86_FEATURE_CLFLUSH) &&
553 (c->x86_vfm == INTEL_CORE2_DUNNINGTON ||
554 c->x86_vfm == INTEL_NEHALEM_EX ||
555 c->x86_vfm == INTEL_WESTMERE_EX))
556 set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
557
558 if (boot_cpu_has(X86_FEATURE_MWAIT) &&
559 (c->x86_vfm == INTEL_ATOM_GOLDMONT ||
560 c->x86_vfm == INTEL_LUNARLAKE_M))
561 set_cpu_bug(c, X86_BUG_MONITOR);
562
563#ifdef CONFIG_X86_64
564 if (c->x86 == 15)
565 c->x86_cache_alignment = c->x86_clflush_size * 2;
566 if (c->x86 == 6)
567 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
568#else
569 /*
570 * Names for the Pentium II/Celeron processors
571 * detectable only by also checking the cache size.
572 * Dixon is NOT a Celeron.
573 */
574 if (c->x86 == 6) {
575 unsigned int l2 = c->x86_cache_size;
576 char *p = NULL;
577
578 switch (c->x86_model) {
579 case 5:
580 if (l2 == 0)
581 p = "Celeron (Covington)";
582 else if (l2 == 256)
583 p = "Mobile Pentium II (Dixon)";
584 break;
585
586 case 6:
587 if (l2 == 128)
588 p = "Celeron (Mendocino)";
589 else if (c->x86_stepping == 0 || c->x86_stepping == 5)
590 p = "Celeron-A";
591 break;
592
593 case 8:
594 if (l2 == 128)
595 p = "Celeron (Coppermine)";
596 break;
597 }
598
599 if (p)
600 strcpy(c->x86_model_id, p);
601 }
602
603 if (c->x86 == 15)
604 set_cpu_cap(c, X86_FEATURE_P4);
605 if (c->x86 == 6)
606 set_cpu_cap(c, X86_FEATURE_P3);
607#endif
608
609 /* Work around errata */
610 srat_detect_node(c);
611
612 init_ia32_feat_ctl(c);
613
614 init_intel_misc_features(c);
615
616 split_lock_init();
617
618 intel_init_thermal(c);
619}
620
621#ifdef CONFIG_X86_32
622static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
623{
624 /*
625 * Intel PIII Tualatin. This comes in two flavours.
626 * One has 256kb of cache, the other 512. We have no way
627 * to determine which, so we use a boottime override
628 * for the 512kb model, and assume 256 otherwise.
629 */
630 if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
631 size = 256;
632
633 /*
634 * Intel Quark SoC X1000 contains a 4-way set associative
635 * 16K cache with a 16 byte cache line and 256 lines per tag
636 */
637 if ((c->x86 == 5) && (c->x86_model == 9))
638 size = 16;
639 return size;
640}
641#endif
642
643#define TLB_INST_4K 0x01
644#define TLB_INST_4M 0x02
645#define TLB_INST_2M_4M 0x03
646
647#define TLB_INST_ALL 0x05
648#define TLB_INST_1G 0x06
649
650#define TLB_DATA_4K 0x11
651#define TLB_DATA_4M 0x12
652#define TLB_DATA_2M_4M 0x13
653#define TLB_DATA_4K_4M 0x14
654
655#define TLB_DATA_1G 0x16
656#define TLB_DATA_1G_2M_4M 0x17
657
658#define TLB_DATA0_4K 0x21
659#define TLB_DATA0_4M 0x22
660#define TLB_DATA0_2M_4M 0x23
661
662#define STLB_4K 0x41
663#define STLB_4K_2M 0x42
664
665/*
666 * All of leaf 0x2's one-byte TLB descriptors implies the same number of
667 * entries for their respective TLB types. The 0x63 descriptor is an
668 * exception: it implies 4 dTLB entries for 1GB pages 32 dTLB entries
669 * for 2MB or 4MB pages. Encode descriptor 0x63 dTLB entry count for
670 * 2MB/4MB pages here, as its count for dTLB 1GB pages is already at the
671 * intel_tlb_table[] mapping.
672 */
673#define TLB_0x63_2M_4M_ENTRIES 32
674
675static const struct _tlb_table intel_tlb_table[] = {
676 { 0x01, TLB_INST_4K, 32, " TLB_INST 4 KByte pages, 4-way set associative" },
677 { 0x02, TLB_INST_4M, 2, " TLB_INST 4 MByte pages, full associative" },
678 { 0x03, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way set associative" },
679 { 0x04, TLB_DATA_4M, 8, " TLB_DATA 4 MByte pages, 4-way set associative" },
680 { 0x05, TLB_DATA_4M, 32, " TLB_DATA 4 MByte pages, 4-way set associative" },
681 { 0x0b, TLB_INST_4M, 4, " TLB_INST 4 MByte pages, 4-way set associative" },
682 { 0x4f, TLB_INST_4K, 32, " TLB_INST 4 KByte pages" },
683 { 0x50, TLB_INST_ALL, 64, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
684 { 0x51, TLB_INST_ALL, 128, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
685 { 0x52, TLB_INST_ALL, 256, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
686 { 0x55, TLB_INST_2M_4M, 7, " TLB_INST 2-MByte or 4-MByte pages, fully associative" },
687 { 0x56, TLB_DATA0_4M, 16, " TLB_DATA0 4 MByte pages, 4-way set associative" },
688 { 0x57, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, 4-way associative" },
689 { 0x59, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, fully associative" },
690 { 0x5a, TLB_DATA0_2M_4M, 32, " TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
691 { 0x5b, TLB_DATA_4K_4M, 64, " TLB_DATA 4 KByte and 4 MByte pages" },
692 { 0x5c, TLB_DATA_4K_4M, 128, " TLB_DATA 4 KByte and 4 MByte pages" },
693 { 0x5d, TLB_DATA_4K_4M, 256, " TLB_DATA 4 KByte and 4 MByte pages" },
694 { 0x61, TLB_INST_4K, 48, " TLB_INST 4 KByte pages, full associative" },
695 { 0x63, TLB_DATA_1G_2M_4M, 4, " TLB_DATA 1 GByte pages, 4-way set associative"
696 " (plus 32 entries TLB_DATA 2 MByte or 4 MByte pages, not encoded here)" },
697 { 0x6b, TLB_DATA_4K, 256, " TLB_DATA 4 KByte pages, 8-way associative" },
698 { 0x6c, TLB_DATA_2M_4M, 128, " TLB_DATA 2 MByte or 4 MByte pages, 8-way associative" },
699 { 0x6d, TLB_DATA_1G, 16, " TLB_DATA 1 GByte pages, fully associative" },
700 { 0x76, TLB_INST_2M_4M, 8, " TLB_INST 2-MByte or 4-MByte pages, fully associative" },
701 { 0xb0, TLB_INST_4K, 128, " TLB_INST 4 KByte pages, 4-way set associative" },
702 { 0xb1, TLB_INST_2M_4M, 4, " TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
703 { 0xb2, TLB_INST_4K, 64, " TLB_INST 4KByte pages, 4-way set associative" },
704 { 0xb3, TLB_DATA_4K, 128, " TLB_DATA 4 KByte pages, 4-way set associative" },
705 { 0xb4, TLB_DATA_4K, 256, " TLB_DATA 4 KByte pages, 4-way associative" },
706 { 0xb5, TLB_INST_4K, 64, " TLB_INST 4 KByte pages, 8-way set associative" },
707 { 0xb6, TLB_INST_4K, 128, " TLB_INST 4 KByte pages, 8-way set associative" },
708 { 0xba, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way associative" },
709 { 0xc0, TLB_DATA_4K_4M, 8, " TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
710 { 0xc1, STLB_4K_2M, 1024, " STLB 4 KByte and 2 MByte pages, 8-way associative" },
711 { 0xc2, TLB_DATA_2M_4M, 16, " TLB_DATA 2 MByte/4MByte pages, 4-way associative" },
712 { 0xca, STLB_4K, 512, " STLB 4 KByte pages, 4-way associative" },
713 { 0x00, 0, 0 }
714};
715
716static void intel_tlb_lookup(const unsigned char desc)
717{
718 unsigned char k;
719 if (desc == 0)
720 return;
721
722 /* look up this descriptor in the table */
723 for (k = 0; intel_tlb_table[k].descriptor != desc &&
724 intel_tlb_table[k].descriptor != 0; k++)
725 ;
726
727 if (intel_tlb_table[k].tlb_type == 0)
728 return;
729
730 switch (intel_tlb_table[k].tlb_type) {
731 case STLB_4K:
732 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
733 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
734 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
735 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
736 break;
737 case STLB_4K_2M:
738 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
739 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
740 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
741 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
742 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
743 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
744 if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
745 tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
746 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
747 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
748 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
749 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
750 break;
751 case TLB_INST_ALL:
752 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
753 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
754 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
755 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
756 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
757 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
758 break;
759 case TLB_INST_4K:
760 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
761 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
762 break;
763 case TLB_INST_4M:
764 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
765 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
766 break;
767 case TLB_INST_2M_4M:
768 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
769 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
770 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
771 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
772 break;
773 case TLB_DATA_4K:
774 case TLB_DATA0_4K:
775 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
776 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
777 break;
778 case TLB_DATA_4M:
779 case TLB_DATA0_4M:
780 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
781 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
782 break;
783 case TLB_DATA_2M_4M:
784 case TLB_DATA0_2M_4M:
785 if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
786 tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
787 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
788 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
789 break;
790 case TLB_DATA_4K_4M:
791 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
792 tlb_lld_4k[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_DATA_1G_2M_4M:
797 if (tlb_lld_2m[ENTRIES] < TLB_0x63_2M_4M_ENTRIES)
798 tlb_lld_2m[ENTRIES] = TLB_0x63_2M_4M_ENTRIES;
799 if (tlb_lld_4m[ENTRIES] < TLB_0x63_2M_4M_ENTRIES)
800 tlb_lld_4m[ENTRIES] = TLB_0x63_2M_4M_ENTRIES;
801 fallthrough;
802 case TLB_DATA_1G:
803 if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
804 tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
805 break;
806 }
807}
808
809static void intel_detect_tlb(struct cpuinfo_x86 *c)
810{
811 int i, j, n;
812 unsigned int regs[4];
813 unsigned char *desc = (unsigned char *)regs;
814
815 if (c->cpuid_level < 2)
816 return;
817
818 /* Number of times to iterate */
819 n = cpuid_eax(2) & 0xFF;
820
821 for (i = 0 ; i < n ; i++) {
822 cpuid(2, ®s[0], ®s[1], ®s[2], ®s[3]);
823
824 /* If bit 31 is set, this is an unknown format */
825 for (j = 0 ; j < 4 ; j++)
826 if (regs[j] & (1 << 31))
827 regs[j] = 0;
828
829 /* Byte 0 is level count, not a descriptor */
830 for (j = 1 ; j < 16 ; j++)
831 intel_tlb_lookup(desc[j]);
832 }
833}
834
835static const struct cpu_dev intel_cpu_dev = {
836 .c_vendor = "Intel",
837 .c_ident = { "GenuineIntel" },
838#ifdef CONFIG_X86_32
839 .legacy_models = {
840 { .family = 4, .model_names =
841 {
842 [0] = "486 DX-25/33",
843 [1] = "486 DX-50",
844 [2] = "486 SX",
845 [3] = "486 DX/2",
846 [4] = "486 SL",
847 [5] = "486 SX/2",
848 [7] = "486 DX/2-WB",
849 [8] = "486 DX/4",
850 [9] = "486 DX/4-WB"
851 }
852 },
853 { .family = 5, .model_names =
854 {
855 [0] = "Pentium 60/66 A-step",
856 [1] = "Pentium 60/66",
857 [2] = "Pentium 75 - 200",
858 [3] = "OverDrive PODP5V83",
859 [4] = "Pentium MMX",
860 [7] = "Mobile Pentium 75 - 200",
861 [8] = "Mobile Pentium MMX",
862 [9] = "Quark SoC X1000",
863 }
864 },
865 { .family = 6, .model_names =
866 {
867 [0] = "Pentium Pro A-step",
868 [1] = "Pentium Pro",
869 [3] = "Pentium II (Klamath)",
870 [4] = "Pentium II (Deschutes)",
871 [5] = "Pentium II (Deschutes)",
872 [6] = "Mobile Pentium II",
873 [7] = "Pentium III (Katmai)",
874 [8] = "Pentium III (Coppermine)",
875 [10] = "Pentium III (Cascades)",
876 [11] = "Pentium III (Tualatin)",
877 }
878 },
879 { .family = 15, .model_names =
880 {
881 [0] = "Pentium 4 (Unknown)",
882 [1] = "Pentium 4 (Willamette)",
883 [2] = "Pentium 4 (Northwood)",
884 [4] = "Pentium 4 (Foster)",
885 [5] = "Pentium 4 (Foster)",
886 }
887 },
888 },
889 .legacy_cache_size = intel_size_cache,
890#endif
891 .c_detect_tlb = intel_detect_tlb,
892 .c_early_init = early_init_intel,
893 .c_bsp_init = bsp_init_intel,
894 .c_init = init_intel,
895 .c_x86_vendor = X86_VENDOR_INTEL,
896};
897
898cpu_dev_register(intel_cpu_dev);
899
900#define X86_HYBRID_CPU_TYPE_ID_SHIFT 24
901
902/**
903 * get_this_hybrid_cpu_type() - Get the type of this hybrid CPU
904 *
905 * Returns the CPU type [31:24] (i.e., Atom or Core) of a CPU in
906 * a hybrid processor. If the processor is not hybrid, returns 0.
907 */
908u8 get_this_hybrid_cpu_type(void)
909{
910 if (!cpu_feature_enabled(X86_FEATURE_HYBRID_CPU))
911 return 0;
912
913 return cpuid_eax(0x0000001a) >> X86_HYBRID_CPU_TYPE_ID_SHIFT;
914}
915
916/**
917 * get_this_hybrid_cpu_native_id() - Get the native id of this hybrid CPU
918 *
919 * Returns the uarch native ID [23:0] of a CPU in a hybrid processor.
920 * If the processor is not hybrid, returns 0.
921 */
922u32 get_this_hybrid_cpu_native_id(void)
923{
924 if (!cpu_feature_enabled(X86_FEATURE_HYBRID_CPU))
925 return 0;
926
927 return cpuid_eax(0x0000001a) &
928 (BIT_ULL(X86_HYBRID_CPU_TYPE_ID_SHIFT) - 1);
929}