1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/kernel.h>
4 #include <linux/string.h>
5 #include <linux/bitops.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>
18 #include <asm/intel-family.h>
19 #include <asm/microcode_intel.h>
20 #include <asm/hwcap2.h>
24 #include <linux/topology.h>
29 #ifdef CONFIG_X86_LOCAL_APIC
30 #include <asm/mpspec.h>
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.
40 static int __init forcempx_setup(char *__unused)
46 __setup("intel-skd-046-workaround=disable", forcempx_setup);
48 void check_mpx_erratum(struct cpuinfo_x86 *c)
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;
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.
86 switch (c->x86_model) {
87 case INTEL_FAM6_XEON_PHI_KNL:
88 case INTEL_FAM6_XEON_PHI_KNM:
94 if (ring3mwait_disabled)
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 {
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)
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))
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);
161 static void early_init_intel(struct cpuinfo_x86 *c)
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);
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);
178 if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
179 c->microcode = intel_get_microcode_revision();
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);
198 * Atom erratum AAE44/AAF40/AAG38/AAH41:
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.
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);
212 set_cpu_cap(c, X86_FEATURE_SYSENTER32);
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;
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;
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.
228 * It is also reliable across cores and sockets. (but not across
229 * cabinets - we turn it off in that case explicitly.)
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);
236 /* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
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);
250 * There is a known erratum on Pentium III and Core Solo
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.
257 * Enable PAT WC only on P4, Core 2 or later CPUs.
259 if (c->x86 == 6 && c->x86_model < 15)
260 clear_cpu_cap(c, X86_FEATURE_PAT);
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.
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);
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"
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
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);
290 if (c->cpuid_level >= 0x00000001) {
291 u32 eax, ebx, ecx, edx;
293 cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
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.
299 if (edx & (1U << 28))
300 c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
303 check_mpx_erratum(c);
308 * Early probe support logic for ppro memory erratum #50
310 * This is called before we do cpu ident work
313 int ppro_with_ram_bug(void)
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");
326 static void intel_smp_check(struct cpuinfo_x86 *c)
328 /* calling is from identify_secondary_cpu() ? */
333 * Mask B, Pentium, but not Pentium MMX
336 c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
339 * Remember we have B step Pentia with bugs
341 WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
342 "with B stepping processors.\n");
347 static int __init forcepae_setup(char *__unused)
352 __setup("forcepae", forcepae_setup);
354 static void intel_workarounds(struct cpuinfo_x86 *c)
356 #ifdef CONFIG_X86_F00F_BUG
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.
363 clear_cpu_bug(c, X86_BUG_F00F);
364 if (c->x86 == 5 && c->x86_model < 9) {
365 static int f00f_workaround_enabled;
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;
376 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
379 if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
380 clear_cpu_cap(c, X86_FEATURE_SEP);
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.
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);
394 * P4 Xeon erratum 037 workaround.
395 * Hardware prefetcher may cause stale data to be loaded into the cache.
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");
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").
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);
416 #ifdef CONFIG_X86_INTEL_USERCOPY
418 * Set up the preferred alignment for movsl bulk memory moves
421 case 4: /* 486: untested */
423 case 5: /* Old Pentia: untested */
425 case 6: /* PII/PIII only like movsl with 8-byte alignment */
428 case 15: /* P4 is OK down to 8-byte alignment */
437 static void intel_workarounds(struct cpuinfo_x86 *c)
442 static void srat_detect_node(struct cpuinfo_x86 *c)
446 int cpu = smp_processor_id();
448 /* Don't do the funky fallback heuristics the AMD version employs
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);
455 numa_set_node(cpu, node);
460 * find out the number of processor cores on the die
462 static int intel_num_cpu_cores(struct cpuinfo_x86 *c)
464 unsigned int eax, ebx, ecx, edx;
466 if (!IS_ENABLED(CONFIG_SMP) || c->cpuid_level < 4)
469 /* Intel has a non-standard dependency on %ecx for this CPUID level. */
470 cpuid_count(4, 0, &eax, &ebx, &ecx, &edx);
472 return (eax >> 26) + 1;
477 static void detect_vmx_virtcap(struct cpuinfo_x86 *c)
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
487 u32 vmx_msr_low, vmx_msr_high, msr_ctl, msr_ctl2;
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);
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);
515 #define MSR_IA32_TME_ACTIVATE 0x982
517 /* Helpers to access TME_ACTIVATE MSR */
518 #define TME_ACTIVATE_LOCKED(x) (x & 0x1)
519 #define TME_ACTIVATE_ENABLED(x) (x & 0x2)
521 #define TME_ACTIVATE_POLICY(x) ((x >> 4) & 0xf) /* Bits 7:4 */
522 #define TME_ACTIVATE_POLICY_AES_XTS_128 0
524 #define TME_ACTIVATE_KEYID_BITS(x) ((x >> 32) & 0xf) /* Bits 35:32 */
526 #define TME_ACTIVATE_CRYPTO_ALGS(x) ((x >> 48) & 0xffff) /* Bits 63:48 */
527 #define TME_ACTIVATE_CRYPTO_AES_XTS_128 1
529 /* Values for mktme_status (SW only construct) */
530 #define MKTME_ENABLED 0
531 #define MKTME_DISABLED 1
532 #define MKTME_UNINITIALIZED 2
533 static int mktme_status = MKTME_UNINITIALIZED;
535 static void detect_tme(struct cpuinfo_x86 *c)
537 u64 tme_activate, tme_policy, tme_crypto_algs;
538 int keyid_bits = 0, nr_keyids = 0;
539 static u64 tme_activate_cpu0 = 0;
541 rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
543 if (mktme_status != MKTME_UNINITIALIZED) {
544 if (tme_activate != tme_activate_cpu0) {
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;
550 /* Proceed. We may need to exclude bits from x86_phys_bits. */
553 tme_activate_cpu0 = tme_activate;
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;
562 if (mktme_status != MKTME_UNINITIALIZED)
563 goto detect_keyid_bits;
565 pr_info("x86/tme: enabled by BIOS\n");
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);
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",
575 mktme_status = MKTME_DISABLED;
578 keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
579 nr_keyids = (1UL << keyid_bits) - 1;
581 pr_info_once("x86/mktme: enabled by BIOS\n");
582 pr_info_once("x86/mktme: %d KeyIDs available\n", nr_keyids);
584 pr_info_once("x86/mktme: disabled by BIOS\n");
587 if (mktme_status == MKTME_UNINITIALIZED) {
588 /* MKTME is usable */
589 mktme_status = MKTME_ENABLED;
593 * KeyID bits effectively lower the number of physical address
594 * bits. Update cpuinfo_x86::x86_phys_bits accordingly.
596 c->x86_phys_bits -= keyid_bits;
599 static void init_intel_energy_perf(struct cpuinfo_x86 *c)
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.)
607 if (!cpu_has(c, X86_FEATURE_EPB))
610 rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
611 if ((epb & 0xF) != ENERGY_PERF_BIAS_PERFORMANCE)
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);
620 static void intel_bsp_resume(struct cpuinfo_x86 *c)
623 * MSR_IA32_ENERGY_PERF_BIAS is lost across suspend/resume,
624 * so reinitialize it properly like during bootup:
626 init_intel_energy_perf(c);
629 static void init_cpuid_fault(struct cpuinfo_x86 *c)
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);
639 static void init_intel_misc_features(struct cpuinfo_x86 *c)
643 if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
646 /* Clear all MISC features */
647 this_cpu_write(msr_misc_features_shadow, 0);
649 /* Check features and update capabilities and shadow control bits */
651 probe_xeon_phi_r3mwait(c);
653 msr = this_cpu_read(msr_misc_features_shadow);
654 wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
657 static void init_intel(struct cpuinfo_x86 *c)
663 intel_workarounds(c);
666 * Detect the extended topology information if available. This
667 * will reinitialise the initial_apicid which will be used
668 * in init_intel_cacheinfo()
670 detect_extended_topology(c);
672 if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
674 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
677 c->x86_max_cores = intel_num_cpu_cores(c);
683 l2 = init_intel_cacheinfo(c);
685 /* Detect legacy cache sizes if init_intel_cacheinfo did not */
687 cpu_detect_cache_sizes(c);
688 l2 = c->x86_cache_size;
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);
698 if (cpu_has(c, X86_FEATURE_XMM2))
699 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
701 if (boot_cpu_has(X86_FEATURE_DS)) {
703 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
705 set_cpu_cap(c, X86_FEATURE_BTS);
707 set_cpu_cap(c, X86_FEATURE_PEBS);
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);
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);
720 c->x86_cache_alignment = c->x86_clflush_size * 2;
722 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
725 * Names for the Pentium II/Celeron processors
726 * detectable only by also checking the cache size.
727 * Dixon is NOT a Celeron.
732 switch (c->x86_model) {
735 p = "Celeron (Covington)";
737 p = "Mobile Pentium II (Dixon)";
742 p = "Celeron (Mendocino)";
743 else if (c->x86_stepping == 0 || c->x86_stepping == 5)
749 p = "Celeron (Coppermine)";
754 strcpy(c->x86_model_id, p);
758 set_cpu_cap(c, X86_FEATURE_P4);
760 set_cpu_cap(c, X86_FEATURE_P3);
763 /* Work around errata */
766 if (cpu_has(c, X86_FEATURE_VMX))
767 detect_vmx_virtcap(c);
769 if (cpu_has(c, X86_FEATURE_TME))
772 init_intel_energy_perf(c);
774 init_intel_misc_features(c);
778 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
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.
786 if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
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
793 if ((c->x86 == 5) && (c->x86_model == 9))
799 #define TLB_INST_4K 0x01
800 #define TLB_INST_4M 0x02
801 #define TLB_INST_2M_4M 0x03
803 #define TLB_INST_ALL 0x05
804 #define TLB_INST_1G 0x06
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
811 #define TLB_DATA_1G 0x16
813 #define TLB_DATA0_4K 0x21
814 #define TLB_DATA0_4M 0x22
815 #define TLB_DATA0_2M_4M 0x23
818 #define STLB_4K_2M 0x42
820 static 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" },
860 static void intel_tlb_lookup(const unsigned char desc)
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++)
871 if (intel_tlb_table[k].tlb_type == 0)
874 switch (intel_tlb_table[k].tlb_type) {
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;
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;
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;
904 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
905 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
908 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
909 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
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;
919 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
920 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
924 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
925 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
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;
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;
941 if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
942 tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
947 static void intel_detect_tlb(struct cpuinfo_x86 *c)
950 unsigned int regs[4];
951 unsigned char *desc = (unsigned char *)regs;
953 if (c->cpuid_level < 2)
956 /* Number of times to iterate */
957 n = cpuid_eax(2) & 0xFF;
959 for (i = 0 ; i < n ; i++) {
960 cpuid(2, ®s[0], ®s[1], ®s[2], ®s[3]);
962 /* If bit 31 is set, this is an unknown format */
963 for (j = 0 ; j < 3 ; j++)
964 if (regs[j] & (1 << 31))
967 /* Byte 0 is level count, not a descriptor */
968 for (j = 1 ; j < 16 ; j++)
969 intel_tlb_lookup(desc[j]);
973 static const struct cpu_dev intel_cpu_dev = {
975 .c_ident = { "GenuineIntel" },
978 { .family = 4, .model_names =
980 [0] = "486 DX-25/33",
991 { .family = 5, .model_names =
993 [0] = "Pentium 60/66 A-step",
994 [1] = "Pentium 60/66",
995 [2] = "Pentium 75 - 200",
996 [3] = "OverDrive PODP5V83",
998 [7] = "Mobile Pentium 75 - 200",
999 [8] = "Mobile Pentium MMX",
1000 [9] = "Quark SoC X1000",
1003 { .family = 6, .model_names =
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)",
1017 { .family = 15, .model_names =
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)",
1027 .legacy_cache_size = intel_size_cache,
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,
1036 cpu_dev_register(intel_cpu_dev);