Linux 4.18.10
[linux/fpc-iii.git] / arch / x86 / kernel / cpu / intel.c
blob3f0abb62161be06864f550d5a53680fa6064f75e
1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/kernel.h>
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>
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>
23 #ifdef CONFIG_X86_64
24 #include <linux/topology.h>
25 #endif
27 #include "cpu.h"
29 #ifdef CONFIG_X86_LOCAL_APIC
30 #include <asm/mpspec.h>
31 #include <asm/apic.h>
32 #endif
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.
38 static int forcempx;
40 static int __init forcempx_setup(char *__unused)
42 forcempx = 1;
44 return 1;
46 __setup("intel-skd-046-workaround=disable", forcempx_setup);
48 void check_mpx_erratum(struct cpuinfo_x86 *c)
50 if (forcempx)
51 return;
53 * Turn off the MPX feature on CPUs where SMEP is not
54 * available or disabled.
56 * Works around Intel Erratum SKD046: "Branch Instructions
57 * May Initialize MPX Bound Registers Incorrectly".
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.
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");
69 static bool ring3mwait_disabled __read_mostly;
71 static int __init ring3mwait_disable(char *__unused)
73 ring3mwait_disabled = true;
74 return 0;
76 __setup("ring3mwait=disable", ring3mwait_disable);
78 static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c)
81 * Ring 3 MONITOR/MWAIT feature cannot be detected without
82 * cpu model and family comparison.
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;
94 if (ring3mwait_disabled)
95 return;
97 set_cpu_cap(c, X86_FEATURE_RING3MWAIT);
98 this_cpu_or(msr_misc_features_shadow,
99 1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT);
101 if (c == &boot_cpu_data)
102 ELF_HWCAP2 |= HWCAP2_RING3MWAIT;
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
113 struct sku_microcode {
114 u8 model;
115 u8 stepping;
116 u32 microcode;
118 static const struct sku_microcode spectre_bad_microcodes[] = {
119 { INTEL_FAM6_KABYLAKE_DESKTOP, 0x0B, 0x80 },
120 { INTEL_FAM6_KABYLAKE_DESKTOP, 0x0A, 0x80 },
121 { INTEL_FAM6_KABYLAKE_DESKTOP, 0x09, 0x80 },
122 { INTEL_FAM6_KABYLAKE_MOBILE, 0x0A, 0x80 },
123 { INTEL_FAM6_KABYLAKE_MOBILE, 0x09, 0x80 },
124 { INTEL_FAM6_SKYLAKE_X, 0x03, 0x0100013e },
125 { INTEL_FAM6_SKYLAKE_X, 0x04, 0x0200003c },
126 { INTEL_FAM6_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 },
142 static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
144 int i;
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.
150 if (cpu_has(c, X86_FEATURE_HYPERVISOR))
151 return false;
153 if (c->x86 != 6)
154 return false;
156 for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
157 if (c->x86_model == spectre_bad_microcodes[i].model &&
158 c->x86_stepping == spectre_bad_microcodes[i].stepping)
159 return (c->microcode <= spectre_bad_microcodes[i].microcode);
161 return false;
164 static void early_init_intel(struct cpuinfo_x86 *c)
166 u64 misc_enable;
168 /* Unmask CPUID levels if masked: */
169 if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
170 if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
171 MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
172 c->cpuid_level = cpuid_eax(0);
173 get_cpu_cap(c);
177 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
178 (c->x86 == 0x6 && c->x86_model >= 0x0e))
179 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
181 if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
182 c->microcode = intel_get_microcode_revision();
184 /* Now if any of them are set, check the blacklist and clear the lot */
185 if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
186 cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
187 cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
188 cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
189 pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
190 setup_clear_cpu_cap(X86_FEATURE_IBRS);
191 setup_clear_cpu_cap(X86_FEATURE_IBPB);
192 setup_clear_cpu_cap(X86_FEATURE_STIBP);
193 setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
194 setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL);
195 setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
196 setup_clear_cpu_cap(X86_FEATURE_SSBD);
197 setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD);
201 * Atom erratum AAE44/AAF40/AAG38/AAH41:
203 * A race condition between speculative fetches and invalidating
204 * a large page. This is worked around in microcode, but we
205 * need the microcode to have already been loaded... so if it is
206 * not, recommend a BIOS update and disable large pages.
208 if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_stepping <= 2 &&
209 c->microcode < 0x20e) {
210 pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
211 clear_cpu_cap(c, X86_FEATURE_PSE);
214 #ifdef CONFIG_X86_64
215 set_cpu_cap(c, X86_FEATURE_SYSENTER32);
216 #else
217 /* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
218 if (c->x86 == 15 && c->x86_cache_alignment == 64)
219 c->x86_cache_alignment = 128;
220 #endif
222 /* CPUID workaround for 0F33/0F34 CPU */
223 if (c->x86 == 0xF && c->x86_model == 0x3
224 && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
225 c->x86_phys_bits = 36;
228 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
229 * with P/T states and does not stop in deep C-states.
231 * It is also reliable across cores and sockets. (but not across
232 * cabinets - we turn it off in that case explicitly.)
234 if (c->x86_power & (1 << 8)) {
235 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
236 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
239 /* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
240 if (c->x86 == 6) {
241 switch (c->x86_model) {
242 case 0x27: /* Penwell */
243 case 0x35: /* Cloverview */
244 case 0x4a: /* Merrifield */
245 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
246 break;
247 default:
248 break;
253 * There is a known erratum on Pentium III and Core Solo
254 * and Core Duo CPUs.
255 * " Page with PAT set to WC while associated MTRR is UC
256 * may consolidate to UC "
257 * Because of this erratum, it is better to stick with
258 * setting WC in MTRR rather than using PAT on these CPUs.
260 * Enable PAT WC only on P4, Core 2 or later CPUs.
262 if (c->x86 == 6 && c->x86_model < 15)
263 clear_cpu_cap(c, X86_FEATURE_PAT);
266 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
267 * clear the fast string and enhanced fast string CPU capabilities.
269 if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
270 rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
271 if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
272 pr_info("Disabled fast string operations\n");
273 setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
274 setup_clear_cpu_cap(X86_FEATURE_ERMS);
279 * Intel Quark Core DevMan_001.pdf section 6.4.11
280 * "The operating system also is required to invalidate (i.e., flush)
281 * the TLB when any changes are made to any of the page table entries.
282 * The operating system must reload CR3 to cause the TLB to be flushed"
284 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
285 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
286 * to be modified.
288 if (c->x86 == 5 && c->x86_model == 9) {
289 pr_info("Disabling PGE capability bit\n");
290 setup_clear_cpu_cap(X86_FEATURE_PGE);
293 if (c->cpuid_level >= 0x00000001) {
294 u32 eax, ebx, ecx, edx;
296 cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
298 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
299 * apicids which are reserved per package. Store the resulting
300 * shift value for the package management code.
302 if (edx & (1U << 28))
303 c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
306 check_mpx_erratum(c);
309 * Get the number of SMT siblings early from the extended topology
310 * leaf, if available. Otherwise try the legacy SMT detection.
312 if (detect_extended_topology_early(c) < 0)
313 detect_ht_early(c);
316 #ifdef CONFIG_X86_32
318 * Early probe support logic for ppro memory erratum #50
320 * This is called before we do cpu ident work
323 int ppro_with_ram_bug(void)
325 /* Uses data from early_cpu_detect now */
326 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
327 boot_cpu_data.x86 == 6 &&
328 boot_cpu_data.x86_model == 1 &&
329 boot_cpu_data.x86_stepping < 8) {
330 pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
331 return 1;
333 return 0;
336 static void intel_smp_check(struct cpuinfo_x86 *c)
338 /* calling is from identify_secondary_cpu() ? */
339 if (!c->cpu_index)
340 return;
343 * Mask B, Pentium, but not Pentium MMX
345 if (c->x86 == 5 &&
346 c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
347 c->x86_model <= 3) {
349 * Remember we have B step Pentia with bugs
351 WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
352 "with B stepping processors.\n");
356 static int forcepae;
357 static int __init forcepae_setup(char *__unused)
359 forcepae = 1;
360 return 1;
362 __setup("forcepae", forcepae_setup);
364 static void intel_workarounds(struct cpuinfo_x86 *c)
366 #ifdef CONFIG_X86_F00F_BUG
368 * All models of Pentium and Pentium with MMX technology CPUs
369 * have the F0 0F bug, which lets nonprivileged users lock up the
370 * system. Announce that the fault handler will be checking for it.
371 * The Quark is also family 5, but does not have the same bug.
373 clear_cpu_bug(c, X86_BUG_F00F);
374 if (c->x86 == 5 && c->x86_model < 9) {
375 static int f00f_workaround_enabled;
377 set_cpu_bug(c, X86_BUG_F00F);
378 if (!f00f_workaround_enabled) {
379 pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
380 f00f_workaround_enabled = 1;
383 #endif
386 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
387 * model 3 mask 3
389 if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
390 clear_cpu_cap(c, X86_FEATURE_SEP);
393 * PAE CPUID issue: many Pentium M report no PAE but may have a
394 * functionally usable PAE implementation.
395 * Forcefully enable PAE if kernel parameter "forcepae" is present.
397 if (forcepae) {
398 pr_warn("PAE forced!\n");
399 set_cpu_cap(c, X86_FEATURE_PAE);
400 add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
404 * P4 Xeon erratum 037 workaround.
405 * Hardware prefetcher may cause stale data to be loaded into the cache.
407 if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
408 if (msr_set_bit(MSR_IA32_MISC_ENABLE,
409 MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
410 pr_info("CPU: C0 stepping P4 Xeon detected.\n");
411 pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
416 * See if we have a good local APIC by checking for buggy Pentia,
417 * i.e. all B steppings and the C2 stepping of P54C when using their
418 * integrated APIC (see 11AP erratum in "Pentium Processor
419 * Specification Update").
421 if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
422 (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
423 set_cpu_bug(c, X86_BUG_11AP);
426 #ifdef CONFIG_X86_INTEL_USERCOPY
428 * Set up the preferred alignment for movsl bulk memory moves
430 switch (c->x86) {
431 case 4: /* 486: untested */
432 break;
433 case 5: /* Old Pentia: untested */
434 break;
435 case 6: /* PII/PIII only like movsl with 8-byte alignment */
436 movsl_mask.mask = 7;
437 break;
438 case 15: /* P4 is OK down to 8-byte alignment */
439 movsl_mask.mask = 7;
440 break;
442 #endif
444 intel_smp_check(c);
446 #else
447 static void intel_workarounds(struct cpuinfo_x86 *c)
450 #endif
452 static void srat_detect_node(struct cpuinfo_x86 *c)
454 #ifdef CONFIG_NUMA
455 unsigned node;
456 int cpu = smp_processor_id();
458 /* Don't do the funky fallback heuristics the AMD version employs
459 for now. */
460 node = numa_cpu_node(cpu);
461 if (node == NUMA_NO_NODE || !node_online(node)) {
462 /* reuse the value from init_cpu_to_node() */
463 node = cpu_to_node(cpu);
465 numa_set_node(cpu, node);
466 #endif
469 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
471 /* Intel VMX MSR indicated features */
472 #define X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW 0x00200000
473 #define X86_VMX_FEATURE_PROC_CTLS_VNMI 0x00400000
474 #define X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS 0x80000000
475 #define X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC 0x00000001
476 #define X86_VMX_FEATURE_PROC_CTLS2_EPT 0x00000002
477 #define X86_VMX_FEATURE_PROC_CTLS2_VPID 0x00000020
479 u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
481 clear_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
482 clear_cpu_cap(c, X86_FEATURE_VNMI);
483 clear_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
484 clear_cpu_cap(c, X86_FEATURE_EPT);
485 clear_cpu_cap(c, X86_FEATURE_VPID);
487 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS, vmx_msr_low, vmx_msr_high);
488 msr_ctl = vmx_msr_high | vmx_msr_low;
489 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW)
490 set_cpu_cap(c, X86_FEATURE_TPR_SHADOW);
491 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_VNMI)
492 set_cpu_cap(c, X86_FEATURE_VNMI);
493 if (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_2ND_CTLS) {
494 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
495 vmx_msr_low, vmx_msr_high);
496 msr_ctl2 = vmx_msr_high | vmx_msr_low;
497 if ((msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VIRT_APIC) &&
498 (msr_ctl & X86_VMX_FEATURE_PROC_CTLS_TPR_SHADOW))
499 set_cpu_cap(c, X86_FEATURE_FLEXPRIORITY);
500 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_EPT)
501 set_cpu_cap(c, X86_FEATURE_EPT);
502 if (msr_ctl2 & X86_VMX_FEATURE_PROC_CTLS2_VPID)
503 set_cpu_cap(c, X86_FEATURE_VPID);
507 #define MSR_IA32_TME_ACTIVATE 0x982
509 /* Helpers to access TME_ACTIVATE MSR */
510 #define TME_ACTIVATE_LOCKED(x) (x & 0x1)
511 #define TME_ACTIVATE_ENABLED(x) (x & 0x2)
513 #define TME_ACTIVATE_POLICY(x) ((x >> 4) & 0xf) /* Bits 7:4 */
514 #define TME_ACTIVATE_POLICY_AES_XTS_128 0
516 #define TME_ACTIVATE_KEYID_BITS(x) ((x >> 32) & 0xf) /* Bits 35:32 */
518 #define TME_ACTIVATE_CRYPTO_ALGS(x) ((x >> 48) & 0xffff) /* Bits 63:48 */
519 #define TME_ACTIVATE_CRYPTO_AES_XTS_128 1
521 /* Values for mktme_status (SW only construct) */
522 #define MKTME_ENABLED 0
523 #define MKTME_DISABLED 1
524 #define MKTME_UNINITIALIZED 2
525 static int mktme_status = MKTME_UNINITIALIZED;
527 static void detect_tme(struct cpuinfo_x86 *c)
529 u64 tme_activate, tme_policy, tme_crypto_algs;
530 int keyid_bits = 0, nr_keyids = 0;
531 static u64 tme_activate_cpu0 = 0;
533 rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
535 if (mktme_status != MKTME_UNINITIALIZED) {
536 if (tme_activate != tme_activate_cpu0) {
537 /* Broken BIOS? */
538 pr_err_once("x86/tme: configuration is inconsistent between CPUs\n");
539 pr_err_once("x86/tme: MKTME is not usable\n");
540 mktme_status = MKTME_DISABLED;
542 /* Proceed. We may need to exclude bits from x86_phys_bits. */
544 } else {
545 tme_activate_cpu0 = tme_activate;
548 if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
549 pr_info_once("x86/tme: not enabled by BIOS\n");
550 mktme_status = MKTME_DISABLED;
551 return;
554 if (mktme_status != MKTME_UNINITIALIZED)
555 goto detect_keyid_bits;
557 pr_info("x86/tme: enabled by BIOS\n");
559 tme_policy = TME_ACTIVATE_POLICY(tme_activate);
560 if (tme_policy != TME_ACTIVATE_POLICY_AES_XTS_128)
561 pr_warn("x86/tme: Unknown policy is active: %#llx\n", tme_policy);
563 tme_crypto_algs = TME_ACTIVATE_CRYPTO_ALGS(tme_activate);
564 if (!(tme_crypto_algs & TME_ACTIVATE_CRYPTO_AES_XTS_128)) {
565 pr_err("x86/mktme: No known encryption algorithm is supported: %#llx\n",
566 tme_crypto_algs);
567 mktme_status = MKTME_DISABLED;
569 detect_keyid_bits:
570 keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
571 nr_keyids = (1UL << keyid_bits) - 1;
572 if (nr_keyids) {
573 pr_info_once("x86/mktme: enabled by BIOS\n");
574 pr_info_once("x86/mktme: %d KeyIDs available\n", nr_keyids);
575 } else {
576 pr_info_once("x86/mktme: disabled by BIOS\n");
579 if (mktme_status == MKTME_UNINITIALIZED) {
580 /* MKTME is usable */
581 mktme_status = MKTME_ENABLED;
585 * KeyID bits effectively lower the number of physical address
586 * bits. Update cpuinfo_x86::x86_phys_bits accordingly.
588 c->x86_phys_bits -= keyid_bits;
591 static void init_intel_energy_perf(struct cpuinfo_x86 *c)
593 u64 epb;
596 * Initialize MSR_IA32_ENERGY_PERF_BIAS if not already initialized.
597 * (x86_energy_perf_policy(8) is available to change it at run-time.)
599 if (!cpu_has(c, X86_FEATURE_EPB))
600 return;
602 rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
603 if ((epb & 0xF) != ENERGY_PERF_BIAS_PERFORMANCE)
604 return;
606 pr_warn_once("ENERGY_PERF_BIAS: Set to 'normal', was 'performance'\n");
607 pr_warn_once("ENERGY_PERF_BIAS: View and update with x86_energy_perf_policy(8)\n");
608 epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
609 wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
612 static void intel_bsp_resume(struct cpuinfo_x86 *c)
615 * MSR_IA32_ENERGY_PERF_BIAS is lost across suspend/resume,
616 * so reinitialize it properly like during bootup:
618 init_intel_energy_perf(c);
621 static void init_cpuid_fault(struct cpuinfo_x86 *c)
623 u64 msr;
625 if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
626 if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
627 set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
631 static void init_intel_misc_features(struct cpuinfo_x86 *c)
633 u64 msr;
635 if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
636 return;
638 /* Clear all MISC features */
639 this_cpu_write(msr_misc_features_shadow, 0);
641 /* Check features and update capabilities and shadow control bits */
642 init_cpuid_fault(c);
643 probe_xeon_phi_r3mwait(c);
645 msr = this_cpu_read(msr_misc_features_shadow);
646 wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
649 static void init_intel(struct cpuinfo_x86 *c)
651 early_init_intel(c);
653 intel_workarounds(c);
656 * Detect the extended topology information if available. This
657 * will reinitialise the initial_apicid which will be used
658 * in init_intel_cacheinfo()
660 detect_extended_topology(c);
662 if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
664 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
665 * detection.
667 detect_num_cpu_cores(c);
668 #ifdef CONFIG_X86_32
669 detect_ht(c);
670 #endif
673 init_intel_cacheinfo(c);
675 if (c->cpuid_level > 9) {
676 unsigned eax = cpuid_eax(10);
677 /* Check for version and the number of counters */
678 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
679 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
682 if (cpu_has(c, X86_FEATURE_XMM2))
683 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
685 if (boot_cpu_has(X86_FEATURE_DS)) {
686 unsigned int l1, l2;
688 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
689 if (!(l1 & (1<<11)))
690 set_cpu_cap(c, X86_FEATURE_BTS);
691 if (!(l1 & (1<<12)))
692 set_cpu_cap(c, X86_FEATURE_PEBS);
695 if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
696 (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
697 set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
699 if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
700 ((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
701 set_cpu_bug(c, X86_BUG_MONITOR);
703 #ifdef CONFIG_X86_64
704 if (c->x86 == 15)
705 c->x86_cache_alignment = c->x86_clflush_size * 2;
706 if (c->x86 == 6)
707 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
708 #else
710 * Names for the Pentium II/Celeron processors
711 * detectable only by also checking the cache size.
712 * Dixon is NOT a Celeron.
714 if (c->x86 == 6) {
715 unsigned int l2 = c->x86_cache_size;
716 char *p = NULL;
718 switch (c->x86_model) {
719 case 5:
720 if (l2 == 0)
721 p = "Celeron (Covington)";
722 else if (l2 == 256)
723 p = "Mobile Pentium II (Dixon)";
724 break;
726 case 6:
727 if (l2 == 128)
728 p = "Celeron (Mendocino)";
729 else if (c->x86_stepping == 0 || c->x86_stepping == 5)
730 p = "Celeron-A";
731 break;
733 case 8:
734 if (l2 == 128)
735 p = "Celeron (Coppermine)";
736 break;
739 if (p)
740 strcpy(c->x86_model_id, p);
743 if (c->x86 == 15)
744 set_cpu_cap(c, X86_FEATURE_P4);
745 if (c->x86 == 6)
746 set_cpu_cap(c, X86_FEATURE_P3);
747 #endif
749 /* Work around errata */
750 srat_detect_node(c);
752 if (cpu_has(c, X86_FEATURE_VMX))
753 detect_vmx_virtcap(c);
755 if (cpu_has(c, X86_FEATURE_TME))
756 detect_tme(c);
758 init_intel_energy_perf(c);
760 init_intel_misc_features(c);
763 #ifdef CONFIG_X86_32
764 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
767 * Intel PIII Tualatin. This comes in two flavours.
768 * One has 256kb of cache, the other 512. We have no way
769 * to determine which, so we use a boottime override
770 * for the 512kb model, and assume 256 otherwise.
772 if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
773 size = 256;
776 * Intel Quark SoC X1000 contains a 4-way set associative
777 * 16K cache with a 16 byte cache line and 256 lines per tag
779 if ((c->x86 == 5) && (c->x86_model == 9))
780 size = 16;
781 return size;
783 #endif
785 #define TLB_INST_4K 0x01
786 #define TLB_INST_4M 0x02
787 #define TLB_INST_2M_4M 0x03
789 #define TLB_INST_ALL 0x05
790 #define TLB_INST_1G 0x06
792 #define TLB_DATA_4K 0x11
793 #define TLB_DATA_4M 0x12
794 #define TLB_DATA_2M_4M 0x13
795 #define TLB_DATA_4K_4M 0x14
797 #define TLB_DATA_1G 0x16
799 #define TLB_DATA0_4K 0x21
800 #define TLB_DATA0_4M 0x22
801 #define TLB_DATA0_2M_4M 0x23
803 #define STLB_4K 0x41
804 #define STLB_4K_2M 0x42
806 static const struct _tlb_table intel_tlb_table[] = {
807 { 0x01, TLB_INST_4K, 32, " TLB_INST 4 KByte pages, 4-way set associative" },
808 { 0x02, TLB_INST_4M, 2, " TLB_INST 4 MByte pages, full associative" },
809 { 0x03, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way set associative" },
810 { 0x04, TLB_DATA_4M, 8, " TLB_DATA 4 MByte pages, 4-way set associative" },
811 { 0x05, TLB_DATA_4M, 32, " TLB_DATA 4 MByte pages, 4-way set associative" },
812 { 0x0b, TLB_INST_4M, 4, " TLB_INST 4 MByte pages, 4-way set associative" },
813 { 0x4f, TLB_INST_4K, 32, " TLB_INST 4 KByte pages */" },
814 { 0x50, TLB_INST_ALL, 64, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
815 { 0x51, TLB_INST_ALL, 128, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
816 { 0x52, TLB_INST_ALL, 256, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
817 { 0x55, TLB_INST_2M_4M, 7, " TLB_INST 2-MByte or 4-MByte pages, fully associative" },
818 { 0x56, TLB_DATA0_4M, 16, " TLB_DATA0 4 MByte pages, 4-way set associative" },
819 { 0x57, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, 4-way associative" },
820 { 0x59, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, fully associative" },
821 { 0x5a, TLB_DATA0_2M_4M, 32, " TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
822 { 0x5b, TLB_DATA_4K_4M, 64, " TLB_DATA 4 KByte and 4 MByte pages" },
823 { 0x5c, TLB_DATA_4K_4M, 128, " TLB_DATA 4 KByte and 4 MByte pages" },
824 { 0x5d, TLB_DATA_4K_4M, 256, " TLB_DATA 4 KByte and 4 MByte pages" },
825 { 0x61, TLB_INST_4K, 48, " TLB_INST 4 KByte pages, full associative" },
826 { 0x63, TLB_DATA_1G, 4, " TLB_DATA 1 GByte pages, 4-way set associative" },
827 { 0x6b, TLB_DATA_4K, 256, " TLB_DATA 4 KByte pages, 8-way associative" },
828 { 0x6c, TLB_DATA_2M_4M, 128, " TLB_DATA 2 MByte or 4 MByte pages, 8-way associative" },
829 { 0x6d, TLB_DATA_1G, 16, " TLB_DATA 1 GByte pages, fully associative" },
830 { 0x76, TLB_INST_2M_4M, 8, " TLB_INST 2-MByte or 4-MByte pages, fully associative" },
831 { 0xb0, TLB_INST_4K, 128, " TLB_INST 4 KByte pages, 4-way set associative" },
832 { 0xb1, TLB_INST_2M_4M, 4, " TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
833 { 0xb2, TLB_INST_4K, 64, " TLB_INST 4KByte pages, 4-way set associative" },
834 { 0xb3, TLB_DATA_4K, 128, " TLB_DATA 4 KByte pages, 4-way set associative" },
835 { 0xb4, TLB_DATA_4K, 256, " TLB_DATA 4 KByte pages, 4-way associative" },
836 { 0xb5, TLB_INST_4K, 64, " TLB_INST 4 KByte pages, 8-way set associative" },
837 { 0xb6, TLB_INST_4K, 128, " TLB_INST 4 KByte pages, 8-way set associative" },
838 { 0xba, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way associative" },
839 { 0xc0, TLB_DATA_4K_4M, 8, " TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
840 { 0xc1, STLB_4K_2M, 1024, " STLB 4 KByte and 2 MByte pages, 8-way associative" },
841 { 0xc2, TLB_DATA_2M_4M, 16, " DTLB 2 MByte/4MByte pages, 4-way associative" },
842 { 0xca, STLB_4K, 512, " STLB 4 KByte pages, 4-way associative" },
843 { 0x00, 0, 0 }
846 static void intel_tlb_lookup(const unsigned char desc)
848 unsigned char k;
849 if (desc == 0)
850 return;
852 /* look up this descriptor in the table */
853 for (k = 0; intel_tlb_table[k].descriptor != desc && \
854 intel_tlb_table[k].descriptor != 0; k++)
857 if (intel_tlb_table[k].tlb_type == 0)
858 return;
860 switch (intel_tlb_table[k].tlb_type) {
861 case STLB_4K:
862 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
863 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
864 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
865 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
866 break;
867 case STLB_4K_2M:
868 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
869 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
870 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
871 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
872 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
873 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
874 if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
875 tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
876 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
877 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
878 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
879 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
880 break;
881 case TLB_INST_ALL:
882 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
883 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
884 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
885 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
886 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
887 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
888 break;
889 case TLB_INST_4K:
890 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
891 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
892 break;
893 case TLB_INST_4M:
894 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
895 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
896 break;
897 case TLB_INST_2M_4M:
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_DATA_4K:
904 case TLB_DATA0_4K:
905 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
906 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
907 break;
908 case TLB_DATA_4M:
909 case TLB_DATA0_4M:
910 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
911 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
912 break;
913 case TLB_DATA_2M_4M:
914 case TLB_DATA0_2M_4M:
915 if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
916 tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
917 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
918 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
919 break;
920 case TLB_DATA_4K_4M:
921 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
922 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
923 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
924 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
925 break;
926 case TLB_DATA_1G:
927 if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
928 tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
929 break;
933 static void intel_detect_tlb(struct cpuinfo_x86 *c)
935 int i, j, n;
936 unsigned int regs[4];
937 unsigned char *desc = (unsigned char *)regs;
939 if (c->cpuid_level < 2)
940 return;
942 /* Number of times to iterate */
943 n = cpuid_eax(2) & 0xFF;
945 for (i = 0 ; i < n ; i++) {
946 cpuid(2, &regs[0], &regs[1], &regs[2], &regs[3]);
948 /* If bit 31 is set, this is an unknown format */
949 for (j = 0 ; j < 3 ; j++)
950 if (regs[j] & (1 << 31))
951 regs[j] = 0;
953 /* Byte 0 is level count, not a descriptor */
954 for (j = 1 ; j < 16 ; j++)
955 intel_tlb_lookup(desc[j]);
959 static const struct cpu_dev intel_cpu_dev = {
960 .c_vendor = "Intel",
961 .c_ident = { "GenuineIntel" },
962 #ifdef CONFIG_X86_32
963 .legacy_models = {
964 { .family = 4, .model_names =
966 [0] = "486 DX-25/33",
967 [1] = "486 DX-50",
968 [2] = "486 SX",
969 [3] = "486 DX/2",
970 [4] = "486 SL",
971 [5] = "486 SX/2",
972 [7] = "486 DX/2-WB",
973 [8] = "486 DX/4",
974 [9] = "486 DX/4-WB"
977 { .family = 5, .model_names =
979 [0] = "Pentium 60/66 A-step",
980 [1] = "Pentium 60/66",
981 [2] = "Pentium 75 - 200",
982 [3] = "OverDrive PODP5V83",
983 [4] = "Pentium MMX",
984 [7] = "Mobile Pentium 75 - 200",
985 [8] = "Mobile Pentium MMX",
986 [9] = "Quark SoC X1000",
989 { .family = 6, .model_names =
991 [0] = "Pentium Pro A-step",
992 [1] = "Pentium Pro",
993 [3] = "Pentium II (Klamath)",
994 [4] = "Pentium II (Deschutes)",
995 [5] = "Pentium II (Deschutes)",
996 [6] = "Mobile Pentium II",
997 [7] = "Pentium III (Katmai)",
998 [8] = "Pentium III (Coppermine)",
999 [10] = "Pentium III (Cascades)",
1000 [11] = "Pentium III (Tualatin)",
1003 { .family = 15, .model_names =
1005 [0] = "Pentium 4 (Unknown)",
1006 [1] = "Pentium 4 (Willamette)",
1007 [2] = "Pentium 4 (Northwood)",
1008 [4] = "Pentium 4 (Foster)",
1009 [5] = "Pentium 4 (Foster)",
1013 .legacy_cache_size = intel_size_cache,
1014 #endif
1015 .c_detect_tlb = intel_detect_tlb,
1016 .c_early_init = early_init_intel,
1017 .c_init = init_intel,
1018 .c_bsp_resume = intel_bsp_resume,
1019 .c_x86_vendor = X86_VENDOR_INTEL,
1022 cpu_dev_register(intel_cpu_dev);