1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/kernel.h>
3 #include <linux/pgtable.h>
5 #include <linux/string.h>
6 #include <linux/bitops.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>
14 #include <asm/cpufeature.h>
18 #include <asm/intel-family.h>
19 #include <asm/microcode_intel.h>
20 #include <asm/hwcap2.h>
22 #include <asm/cpu_device_id.h>
23 #include <asm/cmdline.h>
24 #include <asm/traps.h>
25 #include <asm/resctrl.h>
28 #include <linux/topology.h>
33 #ifdef CONFIG_X86_LOCAL_APIC
34 #include <asm/mpspec.h>
38 enum split_lock_detect_state {
45 * Default to sld_off because most systems do not support split lock detection
46 * split_lock_setup() will switch this to sld_warn on systems that support
47 * split lock detect, unless there is a command line override.
49 static enum split_lock_detect_state sld_state __ro_after_init = sld_off;
50 static u64 msr_test_ctrl_cache __ro_after_init;
53 * Processors which have self-snooping capability can handle conflicting
54 * memory type across CPUs by snooping its own cache. However, there exists
55 * CPU models in which having conflicting memory types still leads to
56 * unpredictable behavior, machine check errors, or hangs. Clear this
57 * feature to prevent its use on machines with known erratas.
59 static void check_memory_type_self_snoop_errata(struct cpuinfo_x86 *c)
61 switch (c->x86_model) {
62 case INTEL_FAM6_CORE_YONAH:
63 case INTEL_FAM6_CORE2_MEROM:
64 case INTEL_FAM6_CORE2_MEROM_L:
65 case INTEL_FAM6_CORE2_PENRYN:
66 case INTEL_FAM6_CORE2_DUNNINGTON:
67 case INTEL_FAM6_NEHALEM:
68 case INTEL_FAM6_NEHALEM_G:
69 case INTEL_FAM6_NEHALEM_EP:
70 case INTEL_FAM6_NEHALEM_EX:
71 case INTEL_FAM6_WESTMERE:
72 case INTEL_FAM6_WESTMERE_EP:
73 case INTEL_FAM6_SANDYBRIDGE:
74 setup_clear_cpu_cap(X86_FEATURE_SELFSNOOP);
78 static bool ring3mwait_disabled __read_mostly;
80 static int __init ring3mwait_disable(char *__unused)
82 ring3mwait_disabled = true;
85 __setup("ring3mwait=disable", ring3mwait_disable);
87 static void probe_xeon_phi_r3mwait(struct cpuinfo_x86 *c)
90 * Ring 3 MONITOR/MWAIT feature cannot be detected without
91 * cpu model and family comparison.
95 switch (c->x86_model) {
96 case INTEL_FAM6_XEON_PHI_KNL:
97 case INTEL_FAM6_XEON_PHI_KNM:
103 if (ring3mwait_disabled)
106 set_cpu_cap(c, X86_FEATURE_RING3MWAIT);
107 this_cpu_or(msr_misc_features_shadow,
108 1UL << MSR_MISC_FEATURES_ENABLES_RING3MWAIT_BIT);
110 if (c == &boot_cpu_data)
111 ELF_HWCAP2 |= HWCAP2_RING3MWAIT;
115 * Early microcode releases for the Spectre v2 mitigation were broken.
116 * Information taken from;
117 * - https://newsroom.intel.com/wp-content/uploads/sites/11/2018/03/microcode-update-guidance.pdf
118 * - https://kb.vmware.com/s/article/52345
119 * - Microcode revisions observed in the wild
120 * - Release note from 20180108 microcode release
122 struct sku_microcode {
127 static const struct sku_microcode spectre_bad_microcodes[] = {
128 { INTEL_FAM6_KABYLAKE, 0x0B, 0x80 },
129 { INTEL_FAM6_KABYLAKE, 0x0A, 0x80 },
130 { INTEL_FAM6_KABYLAKE, 0x09, 0x80 },
131 { INTEL_FAM6_KABYLAKE_L, 0x0A, 0x80 },
132 { INTEL_FAM6_KABYLAKE_L, 0x09, 0x80 },
133 { INTEL_FAM6_SKYLAKE_X, 0x03, 0x0100013e },
134 { INTEL_FAM6_SKYLAKE_X, 0x04, 0x0200003c },
135 { INTEL_FAM6_BROADWELL, 0x04, 0x28 },
136 { INTEL_FAM6_BROADWELL_G, 0x01, 0x1b },
137 { INTEL_FAM6_BROADWELL_D, 0x02, 0x14 },
138 { INTEL_FAM6_BROADWELL_D, 0x03, 0x07000011 },
139 { INTEL_FAM6_BROADWELL_X, 0x01, 0x0b000025 },
140 { INTEL_FAM6_HASWELL_L, 0x01, 0x21 },
141 { INTEL_FAM6_HASWELL_G, 0x01, 0x18 },
142 { INTEL_FAM6_HASWELL, 0x03, 0x23 },
143 { INTEL_FAM6_HASWELL_X, 0x02, 0x3b },
144 { INTEL_FAM6_HASWELL_X, 0x04, 0x10 },
145 { INTEL_FAM6_IVYBRIDGE_X, 0x04, 0x42a },
146 /* Observed in the wild */
147 { INTEL_FAM6_SANDYBRIDGE_X, 0x06, 0x61b },
148 { INTEL_FAM6_SANDYBRIDGE_X, 0x07, 0x712 },
151 static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
156 * We know that the hypervisor lie to us on the microcode version so
157 * we may as well hope that it is running the correct version.
159 if (cpu_has(c, X86_FEATURE_HYPERVISOR))
165 for (i = 0; i < ARRAY_SIZE(spectre_bad_microcodes); i++) {
166 if (c->x86_model == spectre_bad_microcodes[i].model &&
167 c->x86_stepping == spectre_bad_microcodes[i].stepping)
168 return (c->microcode <= spectre_bad_microcodes[i].microcode);
173 static void early_init_intel(struct cpuinfo_x86 *c)
177 /* Unmask CPUID levels if masked: */
178 if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
179 if (msr_clear_bit(MSR_IA32_MISC_ENABLE,
180 MSR_IA32_MISC_ENABLE_LIMIT_CPUID_BIT) > 0) {
181 c->cpuid_level = cpuid_eax(0);
186 if ((c->x86 == 0xf && c->x86_model >= 0x03) ||
187 (c->x86 == 0x6 && c->x86_model >= 0x0e))
188 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
190 if (c->x86 >= 6 && !cpu_has(c, X86_FEATURE_IA64))
191 c->microcode = intel_get_microcode_revision();
193 /* Now if any of them are set, check the blacklist and clear the lot */
194 if ((cpu_has(c, X86_FEATURE_SPEC_CTRL) ||
195 cpu_has(c, X86_FEATURE_INTEL_STIBP) ||
196 cpu_has(c, X86_FEATURE_IBRS) || cpu_has(c, X86_FEATURE_IBPB) ||
197 cpu_has(c, X86_FEATURE_STIBP)) && bad_spectre_microcode(c)) {
198 pr_warn("Intel Spectre v2 broken microcode detected; disabling Speculation Control\n");
199 setup_clear_cpu_cap(X86_FEATURE_IBRS);
200 setup_clear_cpu_cap(X86_FEATURE_IBPB);
201 setup_clear_cpu_cap(X86_FEATURE_STIBP);
202 setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL);
203 setup_clear_cpu_cap(X86_FEATURE_MSR_SPEC_CTRL);
204 setup_clear_cpu_cap(X86_FEATURE_INTEL_STIBP);
205 setup_clear_cpu_cap(X86_FEATURE_SSBD);
206 setup_clear_cpu_cap(X86_FEATURE_SPEC_CTRL_SSBD);
210 * Atom erratum AAE44/AAF40/AAG38/AAH41:
212 * A race condition between speculative fetches and invalidating
213 * a large page. This is worked around in microcode, but we
214 * need the microcode to have already been loaded... so if it is
215 * not, recommend a BIOS update and disable large pages.
217 if (c->x86 == 6 && c->x86_model == 0x1c && c->x86_stepping <= 2 &&
218 c->microcode < 0x20e) {
219 pr_warn("Atom PSE erratum detected, BIOS microcode update recommended\n");
220 clear_cpu_cap(c, X86_FEATURE_PSE);
224 set_cpu_cap(c, X86_FEATURE_SYSENTER32);
226 /* Netburst reports 64 bytes clflush size, but does IO in 128 bytes */
227 if (c->x86 == 15 && c->x86_cache_alignment == 64)
228 c->x86_cache_alignment = 128;
231 /* CPUID workaround for 0F33/0F34 CPU */
232 if (c->x86 == 0xF && c->x86_model == 0x3
233 && (c->x86_stepping == 0x3 || c->x86_stepping == 0x4))
234 c->x86_phys_bits = 36;
237 * c->x86_power is 8000_0007 edx. Bit 8 is TSC runs at constant rate
238 * with P/T states and does not stop in deep C-states.
240 * It is also reliable across cores and sockets. (but not across
241 * cabinets - we turn it off in that case explicitly.)
243 if (c->x86_power & (1 << 8)) {
244 set_cpu_cap(c, X86_FEATURE_CONSTANT_TSC);
245 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC);
248 /* Penwell and Cloverview have the TSC which doesn't sleep on S3 */
250 switch (c->x86_model) {
251 case INTEL_FAM6_ATOM_SALTWELL_MID:
252 case INTEL_FAM6_ATOM_SALTWELL_TABLET:
253 case INTEL_FAM6_ATOM_SILVERMONT_MID:
254 case INTEL_FAM6_ATOM_AIRMONT_NP:
255 set_cpu_cap(c, X86_FEATURE_NONSTOP_TSC_S3);
263 * There is a known erratum on Pentium III and Core Solo
265 * " Page with PAT set to WC while associated MTRR is UC
266 * may consolidate to UC "
267 * Because of this erratum, it is better to stick with
268 * setting WC in MTRR rather than using PAT on these CPUs.
270 * Enable PAT WC only on P4, Core 2 or later CPUs.
272 if (c->x86 == 6 && c->x86_model < 15)
273 clear_cpu_cap(c, X86_FEATURE_PAT);
276 * If fast string is not enabled in IA32_MISC_ENABLE for any reason,
277 * clear the fast string and enhanced fast string CPU capabilities.
279 if (c->x86 > 6 || (c->x86 == 6 && c->x86_model >= 0xd)) {
280 rdmsrl(MSR_IA32_MISC_ENABLE, misc_enable);
281 if (!(misc_enable & MSR_IA32_MISC_ENABLE_FAST_STRING)) {
282 pr_info("Disabled fast string operations\n");
283 setup_clear_cpu_cap(X86_FEATURE_REP_GOOD);
284 setup_clear_cpu_cap(X86_FEATURE_ERMS);
289 * Intel Quark Core DevMan_001.pdf section 6.4.11
290 * "The operating system also is required to invalidate (i.e., flush)
291 * the TLB when any changes are made to any of the page table entries.
292 * The operating system must reload CR3 to cause the TLB to be flushed"
294 * As a result, boot_cpu_has(X86_FEATURE_PGE) in arch/x86/include/asm/tlbflush.h
295 * should be false so that __flush_tlb_all() causes CR3 insted of CR4.PGE
298 if (c->x86 == 5 && c->x86_model == 9) {
299 pr_info("Disabling PGE capability bit\n");
300 setup_clear_cpu_cap(X86_FEATURE_PGE);
303 if (c->cpuid_level >= 0x00000001) {
304 u32 eax, ebx, ecx, edx;
306 cpuid(0x00000001, &eax, &ebx, &ecx, &edx);
308 * If HTT (EDX[28]) is set EBX[16:23] contain the number of
309 * apicids which are reserved per package. Store the resulting
310 * shift value for the package management code.
312 if (edx & (1U << 28))
313 c->x86_coreid_bits = get_count_order((ebx >> 16) & 0xff);
316 check_memory_type_self_snoop_errata(c);
319 * Get the number of SMT siblings early from the extended topology
320 * leaf, if available. Otherwise try the legacy SMT detection.
322 if (detect_extended_topology_early(c) < 0)
326 static void bsp_init_intel(struct cpuinfo_x86 *c)
328 resctrl_cpu_detect(c);
333 * Early probe support logic for ppro memory erratum #50
335 * This is called before we do cpu ident work
338 int ppro_with_ram_bug(void)
340 /* Uses data from early_cpu_detect now */
341 if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL &&
342 boot_cpu_data.x86 == 6 &&
343 boot_cpu_data.x86_model == 1 &&
344 boot_cpu_data.x86_stepping < 8) {
345 pr_info("Pentium Pro with Errata#50 detected. Taking evasive action.\n");
351 static void intel_smp_check(struct cpuinfo_x86 *c)
353 /* calling is from identify_secondary_cpu() ? */
358 * Mask B, Pentium, but not Pentium MMX
361 c->x86_stepping >= 1 && c->x86_stepping <= 4 &&
364 * Remember we have B step Pentia with bugs
366 WARN_ONCE(1, "WARNING: SMP operation may be unreliable"
367 "with B stepping processors.\n");
372 static int __init forcepae_setup(char *__unused)
377 __setup("forcepae", forcepae_setup);
379 static void intel_workarounds(struct cpuinfo_x86 *c)
381 #ifdef CONFIG_X86_F00F_BUG
383 * All models of Pentium and Pentium with MMX technology CPUs
384 * have the F0 0F bug, which lets nonprivileged users lock up the
385 * system. Announce that the fault handler will be checking for it.
386 * The Quark is also family 5, but does not have the same bug.
388 clear_cpu_bug(c, X86_BUG_F00F);
389 if (c->x86 == 5 && c->x86_model < 9) {
390 static int f00f_workaround_enabled;
392 set_cpu_bug(c, X86_BUG_F00F);
393 if (!f00f_workaround_enabled) {
394 pr_notice("Intel Pentium with F0 0F bug - workaround enabled.\n");
395 f00f_workaround_enabled = 1;
401 * SEP CPUID bug: Pentium Pro reports SEP but doesn't have it until
404 if ((c->x86<<8 | c->x86_model<<4 | c->x86_stepping) < 0x633)
405 clear_cpu_cap(c, X86_FEATURE_SEP);
408 * PAE CPUID issue: many Pentium M report no PAE but may have a
409 * functionally usable PAE implementation.
410 * Forcefully enable PAE if kernel parameter "forcepae" is present.
413 pr_warn("PAE forced!\n");
414 set_cpu_cap(c, X86_FEATURE_PAE);
415 add_taint(TAINT_CPU_OUT_OF_SPEC, LOCKDEP_NOW_UNRELIABLE);
419 * P4 Xeon erratum 037 workaround.
420 * Hardware prefetcher may cause stale data to be loaded into the cache.
422 if ((c->x86 == 15) && (c->x86_model == 1) && (c->x86_stepping == 1)) {
423 if (msr_set_bit(MSR_IA32_MISC_ENABLE,
424 MSR_IA32_MISC_ENABLE_PREFETCH_DISABLE_BIT) > 0) {
425 pr_info("CPU: C0 stepping P4 Xeon detected.\n");
426 pr_info("CPU: Disabling hardware prefetching (Erratum 037)\n");
431 * See if we have a good local APIC by checking for buggy Pentia,
432 * i.e. all B steppings and the C2 stepping of P54C when using their
433 * integrated APIC (see 11AP erratum in "Pentium Processor
434 * Specification Update").
436 if (boot_cpu_has(X86_FEATURE_APIC) && (c->x86<<8 | c->x86_model<<4) == 0x520 &&
437 (c->x86_stepping < 0x6 || c->x86_stepping == 0xb))
438 set_cpu_bug(c, X86_BUG_11AP);
441 #ifdef CONFIG_X86_INTEL_USERCOPY
443 * Set up the preferred alignment for movsl bulk memory moves
446 case 4: /* 486: untested */
448 case 5: /* Old Pentia: untested */
450 case 6: /* PII/PIII only like movsl with 8-byte alignment */
453 case 15: /* P4 is OK down to 8-byte alignment */
462 static void intel_workarounds(struct cpuinfo_x86 *c)
467 static void srat_detect_node(struct cpuinfo_x86 *c)
471 int cpu = smp_processor_id();
473 /* Don't do the funky fallback heuristics the AMD version employs
475 node = numa_cpu_node(cpu);
476 if (node == NUMA_NO_NODE || !node_online(node)) {
477 /* reuse the value from init_cpu_to_node() */
478 node = cpu_to_node(cpu);
480 numa_set_node(cpu, node);
484 #define MSR_IA32_TME_ACTIVATE 0x982
486 /* Helpers to access TME_ACTIVATE MSR */
487 #define TME_ACTIVATE_LOCKED(x) (x & 0x1)
488 #define TME_ACTIVATE_ENABLED(x) (x & 0x2)
490 #define TME_ACTIVATE_POLICY(x) ((x >> 4) & 0xf) /* Bits 7:4 */
491 #define TME_ACTIVATE_POLICY_AES_XTS_128 0
493 #define TME_ACTIVATE_KEYID_BITS(x) ((x >> 32) & 0xf) /* Bits 35:32 */
495 #define TME_ACTIVATE_CRYPTO_ALGS(x) ((x >> 48) & 0xffff) /* Bits 63:48 */
496 #define TME_ACTIVATE_CRYPTO_AES_XTS_128 1
498 /* Values for mktme_status (SW only construct) */
499 #define MKTME_ENABLED 0
500 #define MKTME_DISABLED 1
501 #define MKTME_UNINITIALIZED 2
502 static int mktme_status = MKTME_UNINITIALIZED;
504 static void detect_tme(struct cpuinfo_x86 *c)
506 u64 tme_activate, tme_policy, tme_crypto_algs;
507 int keyid_bits = 0, nr_keyids = 0;
508 static u64 tme_activate_cpu0 = 0;
510 rdmsrl(MSR_IA32_TME_ACTIVATE, tme_activate);
512 if (mktme_status != MKTME_UNINITIALIZED) {
513 if (tme_activate != tme_activate_cpu0) {
515 pr_err_once("x86/tme: configuration is inconsistent between CPUs\n");
516 pr_err_once("x86/tme: MKTME is not usable\n");
517 mktme_status = MKTME_DISABLED;
519 /* Proceed. We may need to exclude bits from x86_phys_bits. */
522 tme_activate_cpu0 = tme_activate;
525 if (!TME_ACTIVATE_LOCKED(tme_activate) || !TME_ACTIVATE_ENABLED(tme_activate)) {
526 pr_info_once("x86/tme: not enabled by BIOS\n");
527 mktme_status = MKTME_DISABLED;
531 if (mktme_status != MKTME_UNINITIALIZED)
532 goto detect_keyid_bits;
534 pr_info("x86/tme: enabled by BIOS\n");
536 tme_policy = TME_ACTIVATE_POLICY(tme_activate);
537 if (tme_policy != TME_ACTIVATE_POLICY_AES_XTS_128)
538 pr_warn("x86/tme: Unknown policy is active: %#llx\n", tme_policy);
540 tme_crypto_algs = TME_ACTIVATE_CRYPTO_ALGS(tme_activate);
541 if (!(tme_crypto_algs & TME_ACTIVATE_CRYPTO_AES_XTS_128)) {
542 pr_err("x86/mktme: No known encryption algorithm is supported: %#llx\n",
544 mktme_status = MKTME_DISABLED;
547 keyid_bits = TME_ACTIVATE_KEYID_BITS(tme_activate);
548 nr_keyids = (1UL << keyid_bits) - 1;
550 pr_info_once("x86/mktme: enabled by BIOS\n");
551 pr_info_once("x86/mktme: %d KeyIDs available\n", nr_keyids);
553 pr_info_once("x86/mktme: disabled by BIOS\n");
556 if (mktme_status == MKTME_UNINITIALIZED) {
557 /* MKTME is usable */
558 mktme_status = MKTME_ENABLED;
562 * KeyID bits effectively lower the number of physical address
563 * bits. Update cpuinfo_x86::x86_phys_bits accordingly.
565 c->x86_phys_bits -= keyid_bits;
568 static void init_cpuid_fault(struct cpuinfo_x86 *c)
572 if (!rdmsrl_safe(MSR_PLATFORM_INFO, &msr)) {
573 if (msr & MSR_PLATFORM_INFO_CPUID_FAULT)
574 set_cpu_cap(c, X86_FEATURE_CPUID_FAULT);
578 static void init_intel_misc_features(struct cpuinfo_x86 *c)
582 if (rdmsrl_safe(MSR_MISC_FEATURES_ENABLES, &msr))
585 /* Clear all MISC features */
586 this_cpu_write(msr_misc_features_shadow, 0);
588 /* Check features and update capabilities and shadow control bits */
590 probe_xeon_phi_r3mwait(c);
592 msr = this_cpu_read(msr_misc_features_shadow);
593 wrmsrl(MSR_MISC_FEATURES_ENABLES, msr);
596 static void split_lock_init(void);
598 static void init_intel(struct cpuinfo_x86 *c)
602 intel_workarounds(c);
605 * Detect the extended topology information if available. This
606 * will reinitialise the initial_apicid which will be used
607 * in init_intel_cacheinfo()
609 detect_extended_topology(c);
611 if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) {
613 * let's use the legacy cpuid vector 0x1 and 0x4 for topology
616 detect_num_cpu_cores(c);
622 init_intel_cacheinfo(c);
624 if (c->cpuid_level > 9) {
625 unsigned eax = cpuid_eax(10);
626 /* Check for version and the number of counters */
627 if ((eax & 0xff) && (((eax>>8) & 0xff) > 1))
628 set_cpu_cap(c, X86_FEATURE_ARCH_PERFMON);
631 if (cpu_has(c, X86_FEATURE_XMM2))
632 set_cpu_cap(c, X86_FEATURE_LFENCE_RDTSC);
634 if (boot_cpu_has(X86_FEATURE_DS)) {
637 rdmsr(MSR_IA32_MISC_ENABLE, l1, l2);
639 set_cpu_cap(c, X86_FEATURE_BTS);
641 set_cpu_cap(c, X86_FEATURE_PEBS);
644 if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_CLFLUSH) &&
645 (c->x86_model == 29 || c->x86_model == 46 || c->x86_model == 47))
646 set_cpu_bug(c, X86_BUG_CLFLUSH_MONITOR);
648 if (c->x86 == 6 && boot_cpu_has(X86_FEATURE_MWAIT) &&
649 ((c->x86_model == INTEL_FAM6_ATOM_GOLDMONT)))
650 set_cpu_bug(c, X86_BUG_MONITOR);
654 c->x86_cache_alignment = c->x86_clflush_size * 2;
656 set_cpu_cap(c, X86_FEATURE_REP_GOOD);
659 * Names for the Pentium II/Celeron processors
660 * detectable only by also checking the cache size.
661 * Dixon is NOT a Celeron.
664 unsigned int l2 = c->x86_cache_size;
667 switch (c->x86_model) {
670 p = "Celeron (Covington)";
672 p = "Mobile Pentium II (Dixon)";
677 p = "Celeron (Mendocino)";
678 else if (c->x86_stepping == 0 || c->x86_stepping == 5)
684 p = "Celeron (Coppermine)";
689 strcpy(c->x86_model_id, p);
693 set_cpu_cap(c, X86_FEATURE_P4);
695 set_cpu_cap(c, X86_FEATURE_P3);
698 /* Work around errata */
701 init_ia32_feat_ctl(c);
703 if (cpu_has(c, X86_FEATURE_TME))
706 init_intel_misc_features(c);
708 if (tsx_ctrl_state == TSX_CTRL_ENABLE)
710 if (tsx_ctrl_state == TSX_CTRL_DISABLE)
717 static unsigned int intel_size_cache(struct cpuinfo_x86 *c, unsigned int size)
720 * Intel PIII Tualatin. This comes in two flavours.
721 * One has 256kb of cache, the other 512. We have no way
722 * to determine which, so we use a boottime override
723 * for the 512kb model, and assume 256 otherwise.
725 if ((c->x86 == 6) && (c->x86_model == 11) && (size == 0))
729 * Intel Quark SoC X1000 contains a 4-way set associative
730 * 16K cache with a 16 byte cache line and 256 lines per tag
732 if ((c->x86 == 5) && (c->x86_model == 9))
738 #define TLB_INST_4K 0x01
739 #define TLB_INST_4M 0x02
740 #define TLB_INST_2M_4M 0x03
742 #define TLB_INST_ALL 0x05
743 #define TLB_INST_1G 0x06
745 #define TLB_DATA_4K 0x11
746 #define TLB_DATA_4M 0x12
747 #define TLB_DATA_2M_4M 0x13
748 #define TLB_DATA_4K_4M 0x14
750 #define TLB_DATA_1G 0x16
752 #define TLB_DATA0_4K 0x21
753 #define TLB_DATA0_4M 0x22
754 #define TLB_DATA0_2M_4M 0x23
757 #define STLB_4K_2M 0x42
759 static const struct _tlb_table intel_tlb_table[] = {
760 { 0x01, TLB_INST_4K, 32, " TLB_INST 4 KByte pages, 4-way set associative" },
761 { 0x02, TLB_INST_4M, 2, " TLB_INST 4 MByte pages, full associative" },
762 { 0x03, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way set associative" },
763 { 0x04, TLB_DATA_4M, 8, " TLB_DATA 4 MByte pages, 4-way set associative" },
764 { 0x05, TLB_DATA_4M, 32, " TLB_DATA 4 MByte pages, 4-way set associative" },
765 { 0x0b, TLB_INST_4M, 4, " TLB_INST 4 MByte pages, 4-way set associative" },
766 { 0x4f, TLB_INST_4K, 32, " TLB_INST 4 KByte pages" },
767 { 0x50, TLB_INST_ALL, 64, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
768 { 0x51, TLB_INST_ALL, 128, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
769 { 0x52, TLB_INST_ALL, 256, " TLB_INST 4 KByte and 2-MByte or 4-MByte pages" },
770 { 0x55, TLB_INST_2M_4M, 7, " TLB_INST 2-MByte or 4-MByte pages, fully associative" },
771 { 0x56, TLB_DATA0_4M, 16, " TLB_DATA0 4 MByte pages, 4-way set associative" },
772 { 0x57, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, 4-way associative" },
773 { 0x59, TLB_DATA0_4K, 16, " TLB_DATA0 4 KByte pages, fully associative" },
774 { 0x5a, TLB_DATA0_2M_4M, 32, " TLB_DATA0 2-MByte or 4 MByte pages, 4-way set associative" },
775 { 0x5b, TLB_DATA_4K_4M, 64, " TLB_DATA 4 KByte and 4 MByte pages" },
776 { 0x5c, TLB_DATA_4K_4M, 128, " TLB_DATA 4 KByte and 4 MByte pages" },
777 { 0x5d, TLB_DATA_4K_4M, 256, " TLB_DATA 4 KByte and 4 MByte pages" },
778 { 0x61, TLB_INST_4K, 48, " TLB_INST 4 KByte pages, full associative" },
779 { 0x63, TLB_DATA_1G, 4, " TLB_DATA 1 GByte pages, 4-way set associative" },
780 { 0x6b, TLB_DATA_4K, 256, " TLB_DATA 4 KByte pages, 8-way associative" },
781 { 0x6c, TLB_DATA_2M_4M, 128, " TLB_DATA 2 MByte or 4 MByte pages, 8-way associative" },
782 { 0x6d, TLB_DATA_1G, 16, " TLB_DATA 1 GByte pages, fully associative" },
783 { 0x76, TLB_INST_2M_4M, 8, " TLB_INST 2-MByte or 4-MByte pages, fully associative" },
784 { 0xb0, TLB_INST_4K, 128, " TLB_INST 4 KByte pages, 4-way set associative" },
785 { 0xb1, TLB_INST_2M_4M, 4, " TLB_INST 2M pages, 4-way, 8 entries or 4M pages, 4-way entries" },
786 { 0xb2, TLB_INST_4K, 64, " TLB_INST 4KByte pages, 4-way set associative" },
787 { 0xb3, TLB_DATA_4K, 128, " TLB_DATA 4 KByte pages, 4-way set associative" },
788 { 0xb4, TLB_DATA_4K, 256, " TLB_DATA 4 KByte pages, 4-way associative" },
789 { 0xb5, TLB_INST_4K, 64, " TLB_INST 4 KByte pages, 8-way set associative" },
790 { 0xb6, TLB_INST_4K, 128, " TLB_INST 4 KByte pages, 8-way set associative" },
791 { 0xba, TLB_DATA_4K, 64, " TLB_DATA 4 KByte pages, 4-way associative" },
792 { 0xc0, TLB_DATA_4K_4M, 8, " TLB_DATA 4 KByte and 4 MByte pages, 4-way associative" },
793 { 0xc1, STLB_4K_2M, 1024, " STLB 4 KByte and 2 MByte pages, 8-way associative" },
794 { 0xc2, TLB_DATA_2M_4M, 16, " TLB_DATA 2 MByte/4MByte pages, 4-way associative" },
795 { 0xca, STLB_4K, 512, " STLB 4 KByte pages, 4-way associative" },
799 static void intel_tlb_lookup(const unsigned char desc)
805 /* look up this descriptor in the table */
806 for (k = 0; intel_tlb_table[k].descriptor != desc &&
807 intel_tlb_table[k].descriptor != 0; k++)
810 if (intel_tlb_table[k].tlb_type == 0)
813 switch (intel_tlb_table[k].tlb_type) {
815 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
816 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
817 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
818 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
821 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
822 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
823 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
824 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
825 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
826 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
827 if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
828 tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
829 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
830 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
831 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
832 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
835 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
836 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
837 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
838 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
839 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
840 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
843 if (tlb_lli_4k[ENTRIES] < intel_tlb_table[k].entries)
844 tlb_lli_4k[ENTRIES] = intel_tlb_table[k].entries;
847 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
848 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
851 if (tlb_lli_2m[ENTRIES] < intel_tlb_table[k].entries)
852 tlb_lli_2m[ENTRIES] = intel_tlb_table[k].entries;
853 if (tlb_lli_4m[ENTRIES] < intel_tlb_table[k].entries)
854 tlb_lli_4m[ENTRIES] = intel_tlb_table[k].entries;
858 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
859 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
863 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
864 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
867 case TLB_DATA0_2M_4M:
868 if (tlb_lld_2m[ENTRIES] < intel_tlb_table[k].entries)
869 tlb_lld_2m[ENTRIES] = intel_tlb_table[k].entries;
870 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
871 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
874 if (tlb_lld_4k[ENTRIES] < intel_tlb_table[k].entries)
875 tlb_lld_4k[ENTRIES] = intel_tlb_table[k].entries;
876 if (tlb_lld_4m[ENTRIES] < intel_tlb_table[k].entries)
877 tlb_lld_4m[ENTRIES] = intel_tlb_table[k].entries;
880 if (tlb_lld_1g[ENTRIES] < intel_tlb_table[k].entries)
881 tlb_lld_1g[ENTRIES] = intel_tlb_table[k].entries;
886 static void intel_detect_tlb(struct cpuinfo_x86 *c)
889 unsigned int regs[4];
890 unsigned char *desc = (unsigned char *)regs;
892 if (c->cpuid_level < 2)
895 /* Number of times to iterate */
896 n = cpuid_eax(2) & 0xFF;
898 for (i = 0 ; i < n ; i++) {
899 cpuid(2, ®s[0], ®s[1], ®s[2], ®s[3]);
901 /* If bit 31 is set, this is an unknown format */
902 for (j = 0 ; j < 3 ; j++)
903 if (regs[j] & (1 << 31))
906 /* Byte 0 is level count, not a descriptor */
907 for (j = 1 ; j < 16 ; j++)
908 intel_tlb_lookup(desc[j]);
912 static const struct cpu_dev intel_cpu_dev = {
914 .c_ident = { "GenuineIntel" },
917 { .family = 4, .model_names =
919 [0] = "486 DX-25/33",
930 { .family = 5, .model_names =
932 [0] = "Pentium 60/66 A-step",
933 [1] = "Pentium 60/66",
934 [2] = "Pentium 75 - 200",
935 [3] = "OverDrive PODP5V83",
937 [7] = "Mobile Pentium 75 - 200",
938 [8] = "Mobile Pentium MMX",
939 [9] = "Quark SoC X1000",
942 { .family = 6, .model_names =
944 [0] = "Pentium Pro A-step",
946 [3] = "Pentium II (Klamath)",
947 [4] = "Pentium II (Deschutes)",
948 [5] = "Pentium II (Deschutes)",
949 [6] = "Mobile Pentium II",
950 [7] = "Pentium III (Katmai)",
951 [8] = "Pentium III (Coppermine)",
952 [10] = "Pentium III (Cascades)",
953 [11] = "Pentium III (Tualatin)",
956 { .family = 15, .model_names =
958 [0] = "Pentium 4 (Unknown)",
959 [1] = "Pentium 4 (Willamette)",
960 [2] = "Pentium 4 (Northwood)",
961 [4] = "Pentium 4 (Foster)",
962 [5] = "Pentium 4 (Foster)",
966 .legacy_cache_size = intel_size_cache,
968 .c_detect_tlb = intel_detect_tlb,
969 .c_early_init = early_init_intel,
970 .c_bsp_init = bsp_init_intel,
971 .c_init = init_intel,
972 .c_x86_vendor = X86_VENDOR_INTEL,
975 cpu_dev_register(intel_cpu_dev);
978 #define pr_fmt(fmt) "x86/split lock detection: " fmt
980 static const struct {
982 enum split_lock_detect_state state;
983 } sld_options[] __initconst = {
985 { "warn", sld_warn },
986 { "fatal", sld_fatal },
989 static inline bool match_option(const char *arg, int arglen, const char *opt)
991 int len = strlen(opt);
993 return len == arglen && !strncmp(arg, opt, len);
996 static bool split_lock_verify_msr(bool on)
1000 if (rdmsrl_safe(MSR_TEST_CTRL, &ctrl))
1003 ctrl |= MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
1005 ctrl &= ~MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
1006 if (wrmsrl_safe(MSR_TEST_CTRL, ctrl))
1008 rdmsrl(MSR_TEST_CTRL, tmp);
1012 static void __init split_lock_setup(void)
1014 enum split_lock_detect_state state = sld_warn;
1018 if (!split_lock_verify_msr(false)) {
1019 pr_info("MSR access failed: Disabled\n");
1023 ret = cmdline_find_option(boot_command_line, "split_lock_detect",
1026 for (i = 0; i < ARRAY_SIZE(sld_options); i++) {
1027 if (match_option(arg, ret, sld_options[i].option)) {
1028 state = sld_options[i].state;
1036 pr_info("disabled\n");
1039 pr_info("warning about user-space split_locks\n");
1042 pr_info("sending SIGBUS on user-space split_locks\n");
1046 rdmsrl(MSR_TEST_CTRL, msr_test_ctrl_cache);
1048 if (!split_lock_verify_msr(true)) {
1049 pr_info("MSR access failed: Disabled\n");
1054 setup_force_cpu_cap(X86_FEATURE_SPLIT_LOCK_DETECT);
1058 * MSR_TEST_CTRL is per core, but we treat it like a per CPU MSR. Locking
1059 * is not implemented as one thread could undo the setting of the other
1060 * thread immediately after dropping the lock anyway.
1062 static void sld_update_msr(bool on)
1064 u64 test_ctrl_val = msr_test_ctrl_cache;
1067 test_ctrl_val |= MSR_TEST_CTRL_SPLIT_LOCK_DETECT;
1069 wrmsrl(MSR_TEST_CTRL, test_ctrl_val);
1072 static void split_lock_init(void)
1074 split_lock_verify_msr(sld_state != sld_off);
1077 static void split_lock_warn(unsigned long ip)
1079 pr_warn_ratelimited("#AC: %s/%d took a split_lock trap at address: 0x%lx\n",
1080 current->comm, current->pid, ip);
1083 * Disable the split lock detection for this task so it can make
1084 * progress and set TIF_SLD so the detection is re-enabled via
1085 * switch_to_sld() when the task is scheduled out.
1087 sld_update_msr(false);
1088 set_tsk_thread_flag(current, TIF_SLD);
1091 bool handle_guest_split_lock(unsigned long ip)
1093 if (sld_state == sld_warn) {
1094 split_lock_warn(ip);
1098 pr_warn_once("#AC: %s/%d %s split_lock trap at address: 0x%lx\n",
1099 current->comm, current->pid,
1100 sld_state == sld_fatal ? "fatal" : "bogus", ip);
1102 current->thread.error_code = 0;
1103 current->thread.trap_nr = X86_TRAP_AC;
1104 force_sig_fault(SIGBUS, BUS_ADRALN, NULL);
1107 EXPORT_SYMBOL_GPL(handle_guest_split_lock);
1109 bool handle_user_split_lock(struct pt_regs *regs, long error_code)
1111 if ((regs->flags & X86_EFLAGS_AC) || sld_state == sld_fatal)
1113 split_lock_warn(regs->ip);
1118 * This function is called only when switching between tasks with
1119 * different split-lock detection modes. It sets the MSR for the
1120 * mode of the new task. This is right most of the time, but since
1121 * the MSR is shared by hyperthreads on a physical core there can
1122 * be glitches when the two threads need different modes.
1124 void switch_to_sld(unsigned long tifn)
1126 sld_update_msr(!(tifn & _TIF_SLD));
1130 * Bits in the IA32_CORE_CAPABILITIES are not architectural, so they should
1131 * only be trusted if it is confirmed that a CPU model implements a
1132 * specific feature at a particular bit position.
1134 * The possible driver data field values:
1136 * - 0: CPU models that are known to have the per-core split-lock detection
1137 * feature even though they do not enumerate IA32_CORE_CAPABILITIES.
1139 * - 1: CPU models which may enumerate IA32_CORE_CAPABILITIES and if so use
1140 * bit 5 to enumerate the per-core split-lock detection feature.
1142 static const struct x86_cpu_id split_lock_cpu_ids[] __initconst = {
1143 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_X, 0),
1144 X86_MATCH_INTEL_FAM6_MODEL(ICELAKE_L, 0),
1145 X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT, 1),
1146 X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT_D, 1),
1147 X86_MATCH_INTEL_FAM6_MODEL(ATOM_TREMONT_L, 1),
1151 void __init cpu_set_core_cap_bits(struct cpuinfo_x86 *c)
1153 const struct x86_cpu_id *m;
1156 if (boot_cpu_has(X86_FEATURE_HYPERVISOR))
1159 m = x86_match_cpu(split_lock_cpu_ids);
1163 switch (m->driver_data) {
1167 if (!cpu_has(c, X86_FEATURE_CORE_CAPABILITIES))
1169 rdmsrl(MSR_IA32_CORE_CAPS, ia32_core_caps);
1170 if (!(ia32_core_caps & MSR_IA32_CORE_CAPS_SPLIT_LOCK_DETECT))