1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * derived from drivers/kvm/kvm_main.c
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
19 #include <linux/kvm_host.h>
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
61 #include <linux/suspend.h>
63 #include <trace/events/kvm.h>
65 #include <asm/debugreg.h>
70 #include <linux/kernel_stat.h>
71 #include <asm/fpu/api.h>
72 #include <asm/fpu/xcr.h>
73 #include <asm/fpu/xstate.h>
74 #include <asm/pvclock.h>
75 #include <asm/div64.h>
76 #include <asm/irq_remapping.h>
77 #include <asm/mshyperv.h>
78 #include <asm/hypervisor.h>
79 #include <asm/tlbflush.h>
80 #include <asm/intel_pt.h>
81 #include <asm/emulate_prefix.h>
83 #include <clocksource/hyperv_timer.h>
85 #define CREATE_TRACE_POINTS
88 #define MAX_IO_MSRS 256
89 #define KVM_MAX_MCE_BANKS 32
90 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
91 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
93 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
95 #define emul_to_vcpu(ctxt) \
96 ((struct kvm_vcpu *)(ctxt)->vcpu)
99 * - enable syscall per default because its emulated by KVM
100 * - enable LME and LMA per default on 64 bit KVM
104 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
106 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
109 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
111 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
113 #define KVM_CAP_PMU_VALID_MASK KVM_PMU_CAP_DISABLE
115 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
116 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
118 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
119 static void process_nmi(struct kvm_vcpu *vcpu);
120 static void process_smi(struct kvm_vcpu *vcpu);
121 static void enter_smm(struct kvm_vcpu *vcpu);
122 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
123 static void store_regs(struct kvm_vcpu *vcpu);
124 static int sync_regs(struct kvm_vcpu *vcpu);
125 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu);
127 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
128 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
130 struct kvm_x86_ops kvm_x86_ops __read_mostly;
132 #define KVM_X86_OP(func) \
133 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
134 *(((struct kvm_x86_ops *)0)->func));
135 #define KVM_X86_OP_OPTIONAL KVM_X86_OP
136 #define KVM_X86_OP_OPTIONAL_RET0 KVM_X86_OP
137 #include <asm/kvm-x86-ops.h>
138 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
139 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
141 static bool __read_mostly ignore_msrs = 0;
142 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
144 bool __read_mostly report_ignored_msrs = true;
145 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
146 EXPORT_SYMBOL_GPL(report_ignored_msrs);
148 unsigned int min_timer_period_us = 200;
149 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
151 static bool __read_mostly kvmclock_periodic_sync = true;
152 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
154 bool __read_mostly kvm_has_tsc_control;
155 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
156 u32 __read_mostly kvm_max_guest_tsc_khz;
157 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
158 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
159 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
160 u64 __read_mostly kvm_max_tsc_scaling_ratio;
161 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
162 u64 __read_mostly kvm_default_tsc_scaling_ratio;
163 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
164 bool __read_mostly kvm_has_bus_lock_exit;
165 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
167 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
168 static u32 __read_mostly tsc_tolerance_ppm = 250;
169 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
172 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
173 * adaptive tuning starting from default advancement of 1000ns. '0' disables
174 * advancement entirely. Any other value is used as-is and disables adaptive
175 * tuning, i.e. allows privileged userspace to set an exact advancement time.
177 static int __read_mostly lapic_timer_advance_ns = -1;
178 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
180 static bool __read_mostly vector_hashing = true;
181 module_param(vector_hashing, bool, S_IRUGO);
183 bool __read_mostly enable_vmware_backdoor = false;
184 module_param(enable_vmware_backdoor, bool, S_IRUGO);
185 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
187 static bool __read_mostly force_emulation_prefix = false;
188 module_param(force_emulation_prefix, bool, S_IRUGO);
190 int __read_mostly pi_inject_timer = -1;
191 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
193 /* Enable/disable PMU virtualization */
194 bool __read_mostly enable_pmu = true;
195 EXPORT_SYMBOL_GPL(enable_pmu);
196 module_param(enable_pmu, bool, 0444);
198 bool __read_mostly eager_page_split = true;
199 module_param(eager_page_split, bool, 0644);
202 * Restoring the host value for MSRs that are only consumed when running in
203 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
204 * returns to userspace, i.e. the kernel can run with the guest's value.
206 #define KVM_MAX_NR_USER_RETURN_MSRS 16
208 struct kvm_user_return_msrs {
209 struct user_return_notifier urn;
211 struct kvm_user_return_msr_values {
214 } values[KVM_MAX_NR_USER_RETURN_MSRS];
217 u32 __read_mostly kvm_nr_uret_msrs;
218 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
219 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
220 static struct kvm_user_return_msrs __percpu *user_return_msrs;
222 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
223 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
224 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
225 | XFEATURE_MASK_PKRU | XFEATURE_MASK_XTILE)
227 u64 __read_mostly host_efer;
228 EXPORT_SYMBOL_GPL(host_efer);
230 bool __read_mostly allow_smaller_maxphyaddr = 0;
231 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
233 bool __read_mostly enable_apicv = true;
234 EXPORT_SYMBOL_GPL(enable_apicv);
236 u64 __read_mostly host_xss;
237 EXPORT_SYMBOL_GPL(host_xss);
238 u64 __read_mostly supported_xss;
239 EXPORT_SYMBOL_GPL(supported_xss);
241 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
242 KVM_GENERIC_VM_STATS(),
243 STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
244 STATS_DESC_COUNTER(VM, mmu_pte_write),
245 STATS_DESC_COUNTER(VM, mmu_pde_zapped),
246 STATS_DESC_COUNTER(VM, mmu_flooded),
247 STATS_DESC_COUNTER(VM, mmu_recycled),
248 STATS_DESC_COUNTER(VM, mmu_cache_miss),
249 STATS_DESC_ICOUNTER(VM, mmu_unsync),
250 STATS_DESC_ICOUNTER(VM, pages_4k),
251 STATS_DESC_ICOUNTER(VM, pages_2m),
252 STATS_DESC_ICOUNTER(VM, pages_1g),
253 STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
254 STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
255 STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
258 const struct kvm_stats_header kvm_vm_stats_header = {
259 .name_size = KVM_STATS_NAME_SIZE,
260 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
261 .id_offset = sizeof(struct kvm_stats_header),
262 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
263 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
264 sizeof(kvm_vm_stats_desc),
267 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
268 KVM_GENERIC_VCPU_STATS(),
269 STATS_DESC_COUNTER(VCPU, pf_taken),
270 STATS_DESC_COUNTER(VCPU, pf_fixed),
271 STATS_DESC_COUNTER(VCPU, pf_emulate),
272 STATS_DESC_COUNTER(VCPU, pf_spurious),
273 STATS_DESC_COUNTER(VCPU, pf_fast),
274 STATS_DESC_COUNTER(VCPU, pf_mmio_spte_created),
275 STATS_DESC_COUNTER(VCPU, pf_guest),
276 STATS_DESC_COUNTER(VCPU, tlb_flush),
277 STATS_DESC_COUNTER(VCPU, invlpg),
278 STATS_DESC_COUNTER(VCPU, exits),
279 STATS_DESC_COUNTER(VCPU, io_exits),
280 STATS_DESC_COUNTER(VCPU, mmio_exits),
281 STATS_DESC_COUNTER(VCPU, signal_exits),
282 STATS_DESC_COUNTER(VCPU, irq_window_exits),
283 STATS_DESC_COUNTER(VCPU, nmi_window_exits),
284 STATS_DESC_COUNTER(VCPU, l1d_flush),
285 STATS_DESC_COUNTER(VCPU, halt_exits),
286 STATS_DESC_COUNTER(VCPU, request_irq_exits),
287 STATS_DESC_COUNTER(VCPU, irq_exits),
288 STATS_DESC_COUNTER(VCPU, host_state_reload),
289 STATS_DESC_COUNTER(VCPU, fpu_reload),
290 STATS_DESC_COUNTER(VCPU, insn_emulation),
291 STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
292 STATS_DESC_COUNTER(VCPU, hypercalls),
293 STATS_DESC_COUNTER(VCPU, irq_injections),
294 STATS_DESC_COUNTER(VCPU, nmi_injections),
295 STATS_DESC_COUNTER(VCPU, req_event),
296 STATS_DESC_COUNTER(VCPU, nested_run),
297 STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
298 STATS_DESC_COUNTER(VCPU, directed_yield_successful),
299 STATS_DESC_ICOUNTER(VCPU, guest_mode)
302 const struct kvm_stats_header kvm_vcpu_stats_header = {
303 .name_size = KVM_STATS_NAME_SIZE,
304 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
305 .id_offset = sizeof(struct kvm_stats_header),
306 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
307 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
308 sizeof(kvm_vcpu_stats_desc),
311 u64 __read_mostly host_xcr0;
312 u64 __read_mostly supported_xcr0;
313 EXPORT_SYMBOL_GPL(supported_xcr0);
315 static struct kmem_cache *x86_emulator_cache;
318 * When called, it means the previous get/set msr reached an invalid msr.
319 * Return true if we want to ignore/silent this failed msr access.
321 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
323 const char *op = write ? "wrmsr" : "rdmsr";
326 if (report_ignored_msrs)
327 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
332 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
338 static struct kmem_cache *kvm_alloc_emulator_cache(void)
340 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
341 unsigned int size = sizeof(struct x86_emulate_ctxt);
343 return kmem_cache_create_usercopy("x86_emulator", size,
344 __alignof__(struct x86_emulate_ctxt),
345 SLAB_ACCOUNT, useroffset,
346 size - useroffset, NULL);
349 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
351 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
354 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
355 vcpu->arch.apf.gfns[i] = ~0;
358 static void kvm_on_user_return(struct user_return_notifier *urn)
361 struct kvm_user_return_msrs *msrs
362 = container_of(urn, struct kvm_user_return_msrs, urn);
363 struct kvm_user_return_msr_values *values;
367 * Disabling irqs at this point since the following code could be
368 * interrupted and executed through kvm_arch_hardware_disable()
370 local_irq_save(flags);
371 if (msrs->registered) {
372 msrs->registered = false;
373 user_return_notifier_unregister(urn);
375 local_irq_restore(flags);
376 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
377 values = &msrs->values[slot];
378 if (values->host != values->curr) {
379 wrmsrl(kvm_uret_msrs_list[slot], values->host);
380 values->curr = values->host;
385 static int kvm_probe_user_return_msr(u32 msr)
391 ret = rdmsrl_safe(msr, &val);
394 ret = wrmsrl_safe(msr, val);
400 int kvm_add_user_return_msr(u32 msr)
402 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
404 if (kvm_probe_user_return_msr(msr))
407 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
408 return kvm_nr_uret_msrs++;
410 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
412 int kvm_find_user_return_msr(u32 msr)
416 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
417 if (kvm_uret_msrs_list[i] == msr)
422 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
424 static void kvm_user_return_msr_cpu_online(void)
426 unsigned int cpu = smp_processor_id();
427 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
431 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
432 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
433 msrs->values[i].host = value;
434 msrs->values[i].curr = value;
438 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
440 unsigned int cpu = smp_processor_id();
441 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
444 value = (value & mask) | (msrs->values[slot].host & ~mask);
445 if (value == msrs->values[slot].curr)
447 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
451 msrs->values[slot].curr = value;
452 if (!msrs->registered) {
453 msrs->urn.on_user_return = kvm_on_user_return;
454 user_return_notifier_register(&msrs->urn);
455 msrs->registered = true;
459 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
461 static void drop_user_return_notifiers(void)
463 unsigned int cpu = smp_processor_id();
464 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
466 if (msrs->registered)
467 kvm_on_user_return(&msrs->urn);
470 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
472 return vcpu->arch.apic_base;
474 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
476 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
478 return kvm_apic_mode(kvm_get_apic_base(vcpu));
480 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
482 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
484 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
485 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
486 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
487 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
489 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
491 if (!msr_info->host_initiated) {
492 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
494 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
498 kvm_lapic_set_base(vcpu, msr_info->data);
499 kvm_recalculate_apic_map(vcpu->kvm);
502 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
505 * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
507 * Hardware virtualization extension instructions may fault if a reboot turns
508 * off virtualization while processes are running. Usually after catching the
509 * fault we just panic; during reboot instead the instruction is ignored.
511 noinstr void kvm_spurious_fault(void)
513 /* Fault while not rebooting. We want the trace. */
514 BUG_ON(!kvm_rebooting);
516 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
518 #define EXCPT_BENIGN 0
519 #define EXCPT_CONTRIBUTORY 1
522 static int exception_class(int vector)
532 return EXCPT_CONTRIBUTORY;
539 #define EXCPT_FAULT 0
541 #define EXCPT_ABORT 2
542 #define EXCPT_INTERRUPT 3
544 static int exception_type(int vector)
548 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
549 return EXCPT_INTERRUPT;
553 /* #DB is trap, as instruction watchpoints are handled elsewhere */
554 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
557 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
560 /* Reserved exceptions will result in fault */
564 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
566 unsigned nr = vcpu->arch.exception.nr;
567 bool has_payload = vcpu->arch.exception.has_payload;
568 unsigned long payload = vcpu->arch.exception.payload;
576 * "Certain debug exceptions may clear bit 0-3. The
577 * remaining contents of the DR6 register are never
578 * cleared by the processor".
580 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
582 * In order to reflect the #DB exception payload in guest
583 * dr6, three components need to be considered: active low
584 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
586 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
587 * In the target guest dr6:
588 * FIXED_1 bits should always be set.
589 * Active low bits should be cleared if 1-setting in payload.
590 * Active high bits should be set if 1-setting in payload.
592 * Note, the payload is compatible with the pending debug
593 * exceptions/exit qualification under VMX, that active_low bits
594 * are active high in payload.
595 * So they need to be flipped for DR6.
597 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
598 vcpu->arch.dr6 |= payload;
599 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
602 * The #DB payload is defined as compatible with the 'pending
603 * debug exceptions' field under VMX, not DR6. While bit 12 is
604 * defined in the 'pending debug exceptions' field (enabled
605 * breakpoint), it is reserved and must be zero in DR6.
607 vcpu->arch.dr6 &= ~BIT(12);
610 vcpu->arch.cr2 = payload;
614 vcpu->arch.exception.has_payload = false;
615 vcpu->arch.exception.payload = 0;
617 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
619 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
620 unsigned nr, bool has_error, u32 error_code,
621 bool has_payload, unsigned long payload, bool reinject)
626 kvm_make_request(KVM_REQ_EVENT, vcpu);
628 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
632 * On vmentry, vcpu->arch.exception.pending is only
633 * true if an event injection was blocked by
634 * nested_run_pending. In that case, however,
635 * vcpu_enter_guest requests an immediate exit,
636 * and the guest shouldn't proceed far enough to
639 WARN_ON_ONCE(vcpu->arch.exception.pending);
640 vcpu->arch.exception.injected = true;
641 if (WARN_ON_ONCE(has_payload)) {
643 * A reinjected event has already
644 * delivered its payload.
650 vcpu->arch.exception.pending = true;
651 vcpu->arch.exception.injected = false;
653 vcpu->arch.exception.has_error_code = has_error;
654 vcpu->arch.exception.nr = nr;
655 vcpu->arch.exception.error_code = error_code;
656 vcpu->arch.exception.has_payload = has_payload;
657 vcpu->arch.exception.payload = payload;
658 if (!is_guest_mode(vcpu))
659 kvm_deliver_exception_payload(vcpu);
663 /* to check exception */
664 prev_nr = vcpu->arch.exception.nr;
665 if (prev_nr == DF_VECTOR) {
666 /* triple fault -> shutdown */
667 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
670 class1 = exception_class(prev_nr);
671 class2 = exception_class(nr);
672 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
673 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
675 * Generate double fault per SDM Table 5-5. Set
676 * exception.pending = true so that the double fault
677 * can trigger a nested vmexit.
679 vcpu->arch.exception.pending = true;
680 vcpu->arch.exception.injected = false;
681 vcpu->arch.exception.has_error_code = true;
682 vcpu->arch.exception.nr = DF_VECTOR;
683 vcpu->arch.exception.error_code = 0;
684 vcpu->arch.exception.has_payload = false;
685 vcpu->arch.exception.payload = 0;
687 /* replace previous exception with a new one in a hope
688 that instruction re-execution will regenerate lost
693 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
695 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
697 EXPORT_SYMBOL_GPL(kvm_queue_exception);
699 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
701 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
703 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
705 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
706 unsigned long payload)
708 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
710 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
712 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
713 u32 error_code, unsigned long payload)
715 kvm_multiple_exception(vcpu, nr, true, error_code,
716 true, payload, false);
719 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
722 kvm_inject_gp(vcpu, 0);
724 return kvm_skip_emulated_instruction(vcpu);
728 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
730 static int complete_emulated_insn_gp(struct kvm_vcpu *vcpu, int err)
733 kvm_inject_gp(vcpu, 0);
737 return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE | EMULTYPE_SKIP |
738 EMULTYPE_COMPLETE_USER_EXIT);
741 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
743 ++vcpu->stat.pf_guest;
744 vcpu->arch.exception.nested_apf =
745 is_guest_mode(vcpu) && fault->async_page_fault;
746 if (vcpu->arch.exception.nested_apf) {
747 vcpu->arch.apf.nested_apf_token = fault->address;
748 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
750 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
754 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
756 /* Returns true if the page fault was immediately morphed into a VM-Exit. */
757 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
758 struct x86_exception *fault)
760 struct kvm_mmu *fault_mmu;
761 WARN_ON_ONCE(fault->vector != PF_VECTOR);
763 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
767 * Invalidate the TLB entry for the faulting address, if it exists,
768 * else the access will fault indefinitely (and to emulate hardware).
770 if ((fault->error_code & PFERR_PRESENT_MASK) &&
771 !(fault->error_code & PFERR_RSVD_MASK))
772 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
773 fault_mmu->root.hpa);
776 * A workaround for KVM's bad exception handling. If KVM injected an
777 * exception into L2, and L2 encountered a #PF while vectoring the
778 * injected exception, manually check to see if L1 wants to intercept
779 * #PF, otherwise queuing the #PF will lead to #DF or a lost exception.
780 * In all other cases, defer the check to nested_ops->check_events(),
781 * which will correctly handle priority (this does not). Note, other
782 * exceptions, e.g. #GP, are theoretically affected, #PF is simply the
783 * most problematic, e.g. when L0 and L1 are both intercepting #PF for
786 * TODO: Rewrite exception handling to track injected and pending
787 * (VM-Exit) exceptions separately.
789 if (unlikely(vcpu->arch.exception.injected && is_guest_mode(vcpu)) &&
790 kvm_x86_ops.nested_ops->handle_page_fault_workaround(vcpu, fault))
793 fault_mmu->inject_page_fault(vcpu, fault);
796 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
798 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
800 atomic_inc(&vcpu->arch.nmi_queued);
801 kvm_make_request(KVM_REQ_NMI, vcpu);
803 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
805 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
807 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
809 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
811 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
813 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
815 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
818 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
819 * a #GP and return false.
821 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
823 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
825 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
828 EXPORT_SYMBOL_GPL(kvm_require_cpl);
830 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
832 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
835 kvm_queue_exception(vcpu, UD_VECTOR);
838 EXPORT_SYMBOL_GPL(kvm_require_dr);
840 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
842 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
846 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
848 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
850 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
851 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
855 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
858 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
861 real_gpa = kvm_translate_gpa(vcpu, mmu, gfn_to_gpa(pdpt_gfn),
862 PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
863 if (real_gpa == UNMAPPED_GVA)
866 /* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
867 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
868 cr3 & GENMASK(11, 5), sizeof(pdpte));
872 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
873 if ((pdpte[i] & PT_PRESENT_MASK) &&
874 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
880 * Marking VCPU_EXREG_PDPTR dirty doesn't work for !tdp_enabled.
881 * Shadow page roots need to be reconstructed instead.
883 if (!tdp_enabled && memcmp(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)))
884 kvm_mmu_free_roots(vcpu->kvm, mmu, KVM_MMU_ROOT_CURRENT);
886 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
887 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
888 kvm_make_request(KVM_REQ_LOAD_MMU_PGD, vcpu);
889 vcpu->arch.pdptrs_from_userspace = false;
893 EXPORT_SYMBOL_GPL(load_pdptrs);
895 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
897 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
898 kvm_clear_async_pf_completion_queue(vcpu);
899 kvm_async_pf_hash_reset(vcpu);
902 * Clearing CR0.PG is defined to flush the TLB from the guest's
905 if (!(cr0 & X86_CR0_PG))
906 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
909 if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
910 kvm_mmu_reset_context(vcpu);
912 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
913 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
914 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
915 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
917 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
919 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
921 unsigned long old_cr0 = kvm_read_cr0(vcpu);
926 if (cr0 & 0xffffffff00000000UL)
930 cr0 &= ~CR0_RESERVED_BITS;
932 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
935 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
939 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
940 (cr0 & X86_CR0_PG)) {
945 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
950 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
951 is_pae(vcpu) && ((cr0 ^ old_cr0) & X86_CR0_PDPTR_BITS) &&
952 !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
955 if (!(cr0 & X86_CR0_PG) &&
956 (is_64_bit_mode(vcpu) || kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)))
959 static_call(kvm_x86_set_cr0)(vcpu, cr0);
961 kvm_post_set_cr0(vcpu, old_cr0, cr0);
965 EXPORT_SYMBOL_GPL(kvm_set_cr0);
967 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
969 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
971 EXPORT_SYMBOL_GPL(kvm_lmsw);
973 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
975 if (vcpu->arch.guest_state_protected)
978 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
980 if (vcpu->arch.xcr0 != host_xcr0)
981 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
983 if (vcpu->arch.xsaves_enabled &&
984 vcpu->arch.ia32_xss != host_xss)
985 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
988 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
989 if (static_cpu_has(X86_FEATURE_PKU) &&
990 vcpu->arch.pkru != vcpu->arch.host_pkru &&
991 ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
992 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)))
993 write_pkru(vcpu->arch.pkru);
994 #endif /* CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS */
996 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
998 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
1000 if (vcpu->arch.guest_state_protected)
1003 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1004 if (static_cpu_has(X86_FEATURE_PKU) &&
1005 ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
1006 kvm_read_cr4_bits(vcpu, X86_CR4_PKE))) {
1007 vcpu->arch.pkru = rdpkru();
1008 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
1009 write_pkru(vcpu->arch.host_pkru);
1011 #endif /* CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS */
1013 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
1015 if (vcpu->arch.xcr0 != host_xcr0)
1016 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
1018 if (vcpu->arch.xsaves_enabled &&
1019 vcpu->arch.ia32_xss != host_xss)
1020 wrmsrl(MSR_IA32_XSS, host_xss);
1024 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
1026 static inline u64 kvm_guest_supported_xcr0(struct kvm_vcpu *vcpu)
1028 return vcpu->arch.guest_fpu.fpstate->user_xfeatures;
1031 #ifdef CONFIG_X86_64
1032 static inline u64 kvm_guest_supported_xfd(struct kvm_vcpu *vcpu)
1034 return kvm_guest_supported_xcr0(vcpu) & XFEATURE_MASK_USER_DYNAMIC;
1038 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
1041 u64 old_xcr0 = vcpu->arch.xcr0;
1044 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
1045 if (index != XCR_XFEATURE_ENABLED_MASK)
1047 if (!(xcr0 & XFEATURE_MASK_FP))
1049 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
1053 * Do not allow the guest to set bits that we do not support
1054 * saving. However, xcr0 bit 0 is always set, even if the
1055 * emulated CPU does not support XSAVE (see kvm_vcpu_reset()).
1057 valid_bits = kvm_guest_supported_xcr0(vcpu) | XFEATURE_MASK_FP;
1058 if (xcr0 & ~valid_bits)
1061 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
1062 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1065 if (xcr0 & XFEATURE_MASK_AVX512) {
1066 if (!(xcr0 & XFEATURE_MASK_YMM))
1068 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1072 if ((xcr0 & XFEATURE_MASK_XTILE) &&
1073 ((xcr0 & XFEATURE_MASK_XTILE) != XFEATURE_MASK_XTILE))
1076 vcpu->arch.xcr0 = xcr0;
1078 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1079 kvm_update_cpuid_runtime(vcpu);
1083 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1085 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1086 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1087 kvm_inject_gp(vcpu, 0);
1091 return kvm_skip_emulated_instruction(vcpu);
1093 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1095 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1097 if (cr4 & cr4_reserved_bits)
1100 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1103 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1105 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1107 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1109 if ((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS)
1110 kvm_mmu_reset_context(vcpu);
1113 * If CR4.PCIDE is changed 0 -> 1, there is no need to flush the TLB
1114 * according to the SDM; however, stale prev_roots could be reused
1115 * incorrectly in the future after a MOV to CR3 with NOFLUSH=1, so we
1116 * free them all. This is *not* a superset of KVM_REQ_TLB_FLUSH_GUEST
1117 * or KVM_REQ_TLB_FLUSH_CURRENT, because the hardware TLB is not flushed,
1121 (cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE))
1122 kvm_mmu_unload(vcpu);
1125 * The TLB has to be flushed for all PCIDs if any of the following
1126 * (architecturally required) changes happen:
1127 * - CR4.PCIDE is changed from 1 to 0
1128 * - CR4.PGE is toggled
1130 * This is a superset of KVM_REQ_TLB_FLUSH_CURRENT.
1132 if (((cr4 ^ old_cr4) & X86_CR4_PGE) ||
1133 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1134 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1137 * The TLB has to be flushed for the current PCID if any of the
1138 * following (architecturally required) changes happen:
1139 * - CR4.SMEP is changed from 0 to 1
1140 * - CR4.PAE is toggled
1142 else if (((cr4 ^ old_cr4) & X86_CR4_PAE) ||
1143 ((cr4 & X86_CR4_SMEP) && !(old_cr4 & X86_CR4_SMEP)))
1144 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1147 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1149 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1151 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1153 if (!kvm_is_valid_cr4(vcpu, cr4))
1156 if (is_long_mode(vcpu)) {
1157 if (!(cr4 & X86_CR4_PAE))
1159 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1161 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1162 && ((cr4 ^ old_cr4) & X86_CR4_PDPTR_BITS)
1163 && !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
1166 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1167 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1170 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1171 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1175 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1177 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1181 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1183 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1185 struct kvm_mmu *mmu = vcpu->arch.mmu;
1186 unsigned long roots_to_free = 0;
1190 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1191 * this is reachable when running EPT=1 and unrestricted_guest=0, and
1192 * also via the emulator. KVM's TDP page tables are not in the scope of
1193 * the invalidation, but the guest's TLB entries need to be flushed as
1194 * the CPU may have cached entries in its TLB for the target PCID.
1196 if (unlikely(tdp_enabled)) {
1197 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1202 * If neither the current CR3 nor any of the prev_roots use the given
1203 * PCID, then nothing needs to be done here because a resync will
1204 * happen anyway before switching to any other CR3.
1206 if (kvm_get_active_pcid(vcpu) == pcid) {
1207 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1208 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1212 * If PCID is disabled, there is no need to free prev_roots even if the
1213 * PCIDs for them are also 0, because MOV to CR3 always flushes the TLB
1216 if (!kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
1219 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1220 if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1221 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1223 kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
1226 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1228 bool skip_tlb_flush = false;
1229 unsigned long pcid = 0;
1230 #ifdef CONFIG_X86_64
1231 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1234 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1235 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1236 pcid = cr3 & X86_CR3_PCID_MASK;
1240 /* PDPTRs are always reloaded for PAE paging. */
1241 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1242 goto handle_tlb_flush;
1245 * Do not condition the GPA check on long mode, this helper is used to
1246 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1247 * the current vCPU mode is accurate.
1249 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1252 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, cr3))
1255 if (cr3 != kvm_read_cr3(vcpu))
1256 kvm_mmu_new_pgd(vcpu, cr3);
1258 vcpu->arch.cr3 = cr3;
1259 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1260 /* Do not call post_set_cr3, we do not get here for confidential guests. */
1264 * A load of CR3 that flushes the TLB flushes only the current PCID,
1265 * even if PCID is disabled, in which case PCID=0 is flushed. It's a
1266 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1267 * and it's impossible to use a non-zero PCID when PCID is disabled,
1268 * i.e. only PCID=0 can be relevant.
1270 if (!skip_tlb_flush)
1271 kvm_invalidate_pcid(vcpu, pcid);
1275 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1277 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1279 if (cr8 & CR8_RESERVED_BITS)
1281 if (lapic_in_kernel(vcpu))
1282 kvm_lapic_set_tpr(vcpu, cr8);
1284 vcpu->arch.cr8 = cr8;
1287 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1289 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1291 if (lapic_in_kernel(vcpu))
1292 return kvm_lapic_get_cr8(vcpu);
1294 return vcpu->arch.cr8;
1296 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1298 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1302 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1303 for (i = 0; i < KVM_NR_DB_REGS; i++)
1304 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1308 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1312 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1313 dr7 = vcpu->arch.guest_debug_dr7;
1315 dr7 = vcpu->arch.dr7;
1316 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1317 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1318 if (dr7 & DR7_BP_EN_MASK)
1319 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1321 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1323 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1325 u64 fixed = DR6_FIXED_1;
1327 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1330 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1331 fixed |= DR6_BUS_LOCK;
1335 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1337 size_t size = ARRAY_SIZE(vcpu->arch.db);
1341 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1342 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1343 vcpu->arch.eff_db[dr] = val;
1347 if (!kvm_dr6_valid(val))
1349 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1353 if (!kvm_dr7_valid(val))
1355 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1356 kvm_update_dr7(vcpu);
1362 EXPORT_SYMBOL_GPL(kvm_set_dr);
1364 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1366 size_t size = ARRAY_SIZE(vcpu->arch.db);
1370 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1374 *val = vcpu->arch.dr6;
1378 *val = vcpu->arch.dr7;
1382 EXPORT_SYMBOL_GPL(kvm_get_dr);
1384 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1386 u32 ecx = kvm_rcx_read(vcpu);
1389 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1390 kvm_inject_gp(vcpu, 0);
1394 kvm_rax_write(vcpu, (u32)data);
1395 kvm_rdx_write(vcpu, data >> 32);
1396 return kvm_skip_emulated_instruction(vcpu);
1398 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1401 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1402 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1404 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1405 * extract the supported MSRs from the related const lists.
1406 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1407 * capabilities of the host cpu. This capabilities test skips MSRs that are
1408 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1409 * may depend on host virtualization features rather than host cpu features.
1412 static const u32 msrs_to_save_all[] = {
1413 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1415 #ifdef CONFIG_X86_64
1416 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1418 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1419 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1421 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1422 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1423 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1424 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1425 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1426 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1427 MSR_IA32_UMWAIT_CONTROL,
1429 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1430 MSR_ARCH_PERFMON_FIXED_CTR0 + 2,
1431 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1432 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1433 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1434 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1435 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1436 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1437 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1438 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1439 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1440 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1441 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1442 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1443 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1444 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1445 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1446 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1447 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1448 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1449 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1450 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1452 MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1453 MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1454 MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1455 MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1456 MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1457 MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1458 MSR_IA32_XFD, MSR_IA32_XFD_ERR,
1461 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1462 static unsigned num_msrs_to_save;
1464 static const u32 emulated_msrs_all[] = {
1465 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1466 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1467 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1468 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1469 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1470 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1471 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1473 HV_X64_MSR_VP_INDEX,
1474 HV_X64_MSR_VP_RUNTIME,
1475 HV_X64_MSR_SCONTROL,
1476 HV_X64_MSR_STIMER0_CONFIG,
1477 HV_X64_MSR_VP_ASSIST_PAGE,
1478 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1479 HV_X64_MSR_TSC_EMULATION_STATUS,
1480 HV_X64_MSR_SYNDBG_OPTIONS,
1481 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1482 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1483 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1485 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1486 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1488 MSR_IA32_TSC_ADJUST,
1489 MSR_IA32_TSC_DEADLINE,
1490 MSR_IA32_ARCH_CAPABILITIES,
1491 MSR_IA32_PERF_CAPABILITIES,
1492 MSR_IA32_MISC_ENABLE,
1493 MSR_IA32_MCG_STATUS,
1495 MSR_IA32_MCG_EXT_CTL,
1499 MSR_MISC_FEATURES_ENABLES,
1500 MSR_AMD64_VIRT_SPEC_CTRL,
1501 MSR_AMD64_TSC_RATIO,
1506 * The following list leaves out MSRs whose values are determined
1507 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1508 * We always support the "true" VMX control MSRs, even if the host
1509 * processor does not, so I am putting these registers here rather
1510 * than in msrs_to_save_all.
1513 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1514 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1515 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1516 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1518 MSR_IA32_VMX_CR0_FIXED0,
1519 MSR_IA32_VMX_CR4_FIXED0,
1520 MSR_IA32_VMX_VMCS_ENUM,
1521 MSR_IA32_VMX_PROCBASED_CTLS2,
1522 MSR_IA32_VMX_EPT_VPID_CAP,
1523 MSR_IA32_VMX_VMFUNC,
1526 MSR_KVM_POLL_CONTROL,
1529 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1530 static unsigned num_emulated_msrs;
1533 * List of msr numbers which are used to expose MSR-based features that
1534 * can be used by a hypervisor to validate requested CPU features.
1536 static const u32 msr_based_features_all[] = {
1538 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1539 MSR_IA32_VMX_PINBASED_CTLS,
1540 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1541 MSR_IA32_VMX_PROCBASED_CTLS,
1542 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1543 MSR_IA32_VMX_EXIT_CTLS,
1544 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1545 MSR_IA32_VMX_ENTRY_CTLS,
1547 MSR_IA32_VMX_CR0_FIXED0,
1548 MSR_IA32_VMX_CR0_FIXED1,
1549 MSR_IA32_VMX_CR4_FIXED0,
1550 MSR_IA32_VMX_CR4_FIXED1,
1551 MSR_IA32_VMX_VMCS_ENUM,
1552 MSR_IA32_VMX_PROCBASED_CTLS2,
1553 MSR_IA32_VMX_EPT_VPID_CAP,
1554 MSR_IA32_VMX_VMFUNC,
1558 MSR_IA32_ARCH_CAPABILITIES,
1559 MSR_IA32_PERF_CAPABILITIES,
1562 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1563 static unsigned int num_msr_based_features;
1565 static u64 kvm_get_arch_capabilities(void)
1569 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1570 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1573 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1574 * the nested hypervisor runs with NX huge pages. If it is not,
1575 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1576 * L1 guests, so it need not worry about its own (L2) guests.
1578 data |= ARCH_CAP_PSCHANGE_MC_NO;
1581 * If we're doing cache flushes (either "always" or "cond")
1582 * we will do one whenever the guest does a vmlaunch/vmresume.
1583 * If an outer hypervisor is doing the cache flush for us
1584 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1585 * capability to the guest too, and if EPT is disabled we're not
1586 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1587 * require a nested hypervisor to do a flush of its own.
1589 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1590 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1592 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1593 data |= ARCH_CAP_RDCL_NO;
1594 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1595 data |= ARCH_CAP_SSB_NO;
1596 if (!boot_cpu_has_bug(X86_BUG_MDS))
1597 data |= ARCH_CAP_MDS_NO;
1599 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1601 * If RTM=0 because the kernel has disabled TSX, the host might
1602 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1603 * and therefore knows that there cannot be TAA) but keep
1604 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1605 * and we want to allow migrating those guests to tsx=off hosts.
1607 data &= ~ARCH_CAP_TAA_NO;
1608 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1609 data |= ARCH_CAP_TAA_NO;
1612 * Nothing to do here; we emulate TSX_CTRL if present on the
1613 * host so the guest can choose between disabling TSX or
1614 * using VERW to clear CPU buffers.
1621 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1623 switch (msr->index) {
1624 case MSR_IA32_ARCH_CAPABILITIES:
1625 msr->data = kvm_get_arch_capabilities();
1627 case MSR_IA32_UCODE_REV:
1628 rdmsrl_safe(msr->index, &msr->data);
1631 return static_call(kvm_x86_get_msr_feature)(msr);
1636 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1638 struct kvm_msr_entry msr;
1642 r = kvm_get_msr_feature(&msr);
1644 if (r == KVM_MSR_RET_INVALID) {
1645 /* Unconditionally clear the output for simplicity */
1647 if (kvm_msr_ignored_check(index, 0, false))
1659 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1661 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1664 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1667 if (efer & (EFER_LME | EFER_LMA) &&
1668 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1671 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1677 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1679 if (efer & efer_reserved_bits)
1682 return __kvm_valid_efer(vcpu, efer);
1684 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1686 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1688 u64 old_efer = vcpu->arch.efer;
1689 u64 efer = msr_info->data;
1692 if (efer & efer_reserved_bits)
1695 if (!msr_info->host_initiated) {
1696 if (!__kvm_valid_efer(vcpu, efer))
1699 if (is_paging(vcpu) &&
1700 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1705 efer |= vcpu->arch.efer & EFER_LMA;
1707 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1713 if ((efer ^ old_efer) & KVM_MMU_EFER_ROLE_BITS)
1714 kvm_mmu_reset_context(vcpu);
1719 void kvm_enable_efer_bits(u64 mask)
1721 efer_reserved_bits &= ~mask;
1723 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1725 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1727 struct kvm_x86_msr_filter *msr_filter;
1728 struct msr_bitmap_range *ranges;
1729 struct kvm *kvm = vcpu->kvm;
1734 /* x2APIC MSRs do not support filtering. */
1735 if (index >= 0x800 && index <= 0x8ff)
1738 idx = srcu_read_lock(&kvm->srcu);
1740 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1746 allowed = msr_filter->default_allow;
1747 ranges = msr_filter->ranges;
1749 for (i = 0; i < msr_filter->count; i++) {
1750 u32 start = ranges[i].base;
1751 u32 end = start + ranges[i].nmsrs;
1752 u32 flags = ranges[i].flags;
1753 unsigned long *bitmap = ranges[i].bitmap;
1755 if ((index >= start) && (index < end) && (flags & type)) {
1756 allowed = !!test_bit(index - start, bitmap);
1762 srcu_read_unlock(&kvm->srcu, idx);
1766 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1769 * Write @data into the MSR specified by @index. Select MSR specific fault
1770 * checks are bypassed if @host_initiated is %true.
1771 * Returns 0 on success, non-0 otherwise.
1772 * Assumes vcpu_load() was already called.
1774 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1775 bool host_initiated)
1777 struct msr_data msr;
1782 case MSR_KERNEL_GS_BASE:
1785 if (is_noncanonical_address(data, vcpu))
1788 case MSR_IA32_SYSENTER_EIP:
1789 case MSR_IA32_SYSENTER_ESP:
1791 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1792 * non-canonical address is written on Intel but not on
1793 * AMD (which ignores the top 32-bits, because it does
1794 * not implement 64-bit SYSENTER).
1796 * 64-bit code should hence be able to write a non-canonical
1797 * value on AMD. Making the address canonical ensures that
1798 * vmentry does not fail on Intel after writing a non-canonical
1799 * value, and that something deterministic happens if the guest
1800 * invokes 64-bit SYSENTER.
1802 data = __canonical_address(data, vcpu_virt_addr_bits(vcpu));
1805 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1808 if (!host_initiated &&
1809 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1810 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1814 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1815 * incomplete and conflicting architectural behavior. Current
1816 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1817 * reserved and always read as zeros. Enforce Intel's reserved
1818 * bits check if and only if the guest CPU is Intel, and clear
1819 * the bits in all other cases. This ensures cross-vendor
1820 * migration will provide consistent behavior for the guest.
1822 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1831 msr.host_initiated = host_initiated;
1833 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1836 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1837 u32 index, u64 data, bool host_initiated)
1839 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1841 if (ret == KVM_MSR_RET_INVALID)
1842 if (kvm_msr_ignored_check(index, data, true))
1849 * Read the MSR specified by @index into @data. Select MSR specific fault
1850 * checks are bypassed if @host_initiated is %true.
1851 * Returns 0 on success, non-0 otherwise.
1852 * Assumes vcpu_load() was already called.
1854 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1855 bool host_initiated)
1857 struct msr_data msr;
1862 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1865 if (!host_initiated &&
1866 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1867 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1873 msr.host_initiated = host_initiated;
1875 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1881 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1882 u32 index, u64 *data, bool host_initiated)
1884 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1886 if (ret == KVM_MSR_RET_INVALID) {
1887 /* Unconditionally clear *data for simplicity */
1889 if (kvm_msr_ignored_check(index, 0, false))
1896 static int kvm_get_msr_with_filter(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1898 if (!kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1899 return KVM_MSR_RET_FILTERED;
1900 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1903 static int kvm_set_msr_with_filter(struct kvm_vcpu *vcpu, u32 index, u64 data)
1905 if (!kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1906 return KVM_MSR_RET_FILTERED;
1907 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1910 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1912 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1914 EXPORT_SYMBOL_GPL(kvm_get_msr);
1916 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1918 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1920 EXPORT_SYMBOL_GPL(kvm_set_msr);
1922 static void complete_userspace_rdmsr(struct kvm_vcpu *vcpu)
1924 if (!vcpu->run->msr.error) {
1925 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1926 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1930 static int complete_emulated_msr_access(struct kvm_vcpu *vcpu)
1932 return complete_emulated_insn_gp(vcpu, vcpu->run->msr.error);
1935 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1937 complete_userspace_rdmsr(vcpu);
1938 return complete_emulated_msr_access(vcpu);
1941 static int complete_fast_msr_access(struct kvm_vcpu *vcpu)
1943 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1946 static int complete_fast_rdmsr(struct kvm_vcpu *vcpu)
1948 complete_userspace_rdmsr(vcpu);
1949 return complete_fast_msr_access(vcpu);
1952 static u64 kvm_msr_reason(int r)
1955 case KVM_MSR_RET_INVALID:
1956 return KVM_MSR_EXIT_REASON_UNKNOWN;
1957 case KVM_MSR_RET_FILTERED:
1958 return KVM_MSR_EXIT_REASON_FILTER;
1960 return KVM_MSR_EXIT_REASON_INVAL;
1964 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1965 u32 exit_reason, u64 data,
1966 int (*completion)(struct kvm_vcpu *vcpu),
1969 u64 msr_reason = kvm_msr_reason(r);
1971 /* Check if the user wanted to know about this MSR fault */
1972 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1975 vcpu->run->exit_reason = exit_reason;
1976 vcpu->run->msr.error = 0;
1977 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1978 vcpu->run->msr.reason = msr_reason;
1979 vcpu->run->msr.index = index;
1980 vcpu->run->msr.data = data;
1981 vcpu->arch.complete_userspace_io = completion;
1986 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1988 u32 ecx = kvm_rcx_read(vcpu);
1992 r = kvm_get_msr_with_filter(vcpu, ecx, &data);
1995 trace_kvm_msr_read(ecx, data);
1997 kvm_rax_write(vcpu, data & -1u);
1998 kvm_rdx_write(vcpu, (data >> 32) & -1u);
2000 /* MSR read failed? See if we should ask user space */
2001 if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_RDMSR, 0,
2002 complete_fast_rdmsr, r))
2004 trace_kvm_msr_read_ex(ecx);
2007 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
2009 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
2011 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
2013 u32 ecx = kvm_rcx_read(vcpu);
2014 u64 data = kvm_read_edx_eax(vcpu);
2017 r = kvm_set_msr_with_filter(vcpu, ecx, data);
2020 trace_kvm_msr_write(ecx, data);
2022 /* MSR write failed? See if we should ask user space */
2023 if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_WRMSR, data,
2024 complete_fast_msr_access, r))
2026 /* Signal all other negative errors to userspace */
2029 trace_kvm_msr_write_ex(ecx, data);
2032 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
2034 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
2036 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
2038 return kvm_skip_emulated_instruction(vcpu);
2040 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
2042 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
2044 /* Treat an INVD instruction as a NOP and just skip it. */
2045 return kvm_emulate_as_nop(vcpu);
2047 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
2049 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
2051 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
2052 return kvm_emulate_as_nop(vcpu);
2054 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
2056 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
2058 kvm_queue_exception(vcpu, UD_VECTOR);
2061 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
2063 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
2065 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
2066 return kvm_emulate_as_nop(vcpu);
2068 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
2070 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
2072 xfer_to_guest_mode_prepare();
2073 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
2074 xfer_to_guest_mode_work_pending();
2078 * The fast path for frequent and performance sensitive wrmsr emulation,
2079 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
2080 * the latency of virtual IPI by avoiding the expensive bits of transitioning
2081 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
2082 * other cases which must be called after interrupts are enabled on the host.
2084 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
2086 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
2089 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
2090 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
2091 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
2092 ((u32)(data >> 32) != X2APIC_BROADCAST))
2093 return kvm_x2apic_icr_write(vcpu->arch.apic, data);
2098 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2100 if (!kvm_can_use_hv_timer(vcpu))
2103 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2107 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2109 u32 msr = kvm_rcx_read(vcpu);
2111 fastpath_t ret = EXIT_FASTPATH_NONE;
2114 case APIC_BASE_MSR + (APIC_ICR >> 4):
2115 data = kvm_read_edx_eax(vcpu);
2116 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2117 kvm_skip_emulated_instruction(vcpu);
2118 ret = EXIT_FASTPATH_EXIT_HANDLED;
2121 case MSR_IA32_TSC_DEADLINE:
2122 data = kvm_read_edx_eax(vcpu);
2123 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2124 kvm_skip_emulated_instruction(vcpu);
2125 ret = EXIT_FASTPATH_REENTER_GUEST;
2132 if (ret != EXIT_FASTPATH_NONE)
2133 trace_kvm_msr_write(msr, data);
2137 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2140 * Adapt set_msr() to msr_io()'s calling convention
2142 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2144 return kvm_get_msr_ignored_check(vcpu, index, data, true);
2147 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2149 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2152 #ifdef CONFIG_X86_64
2153 struct pvclock_clock {
2163 struct pvclock_gtod_data {
2166 struct pvclock_clock clock; /* extract of a clocksource struct */
2167 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2173 static struct pvclock_gtod_data pvclock_gtod_data;
2175 static void update_pvclock_gtod(struct timekeeper *tk)
2177 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2179 write_seqcount_begin(&vdata->seq);
2181 /* copy pvclock gtod data */
2182 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2183 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2184 vdata->clock.mask = tk->tkr_mono.mask;
2185 vdata->clock.mult = tk->tkr_mono.mult;
2186 vdata->clock.shift = tk->tkr_mono.shift;
2187 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2188 vdata->clock.offset = tk->tkr_mono.base;
2190 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2191 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2192 vdata->raw_clock.mask = tk->tkr_raw.mask;
2193 vdata->raw_clock.mult = tk->tkr_raw.mult;
2194 vdata->raw_clock.shift = tk->tkr_raw.shift;
2195 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2196 vdata->raw_clock.offset = tk->tkr_raw.base;
2198 vdata->wall_time_sec = tk->xtime_sec;
2200 vdata->offs_boot = tk->offs_boot;
2202 write_seqcount_end(&vdata->seq);
2205 static s64 get_kvmclock_base_ns(void)
2207 /* Count up from boot time, but with the frequency of the raw clock. */
2208 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2211 static s64 get_kvmclock_base_ns(void)
2213 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2214 return ktime_get_boottime_ns();
2218 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2222 struct pvclock_wall_clock wc;
2229 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2234 ++version; /* first time write, random junk */
2238 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2242 * The guest calculates current wall clock time by adding
2243 * system time (updated by kvm_guest_time_update below) to the
2244 * wall clock specified here. We do the reverse here.
2246 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2248 wc.nsec = do_div(wall_nsec, 1000000000);
2249 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2250 wc.version = version;
2252 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2255 wc_sec_hi = wall_nsec >> 32;
2256 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2257 &wc_sec_hi, sizeof(wc_sec_hi));
2261 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2264 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2265 bool old_msr, bool host_initiated)
2267 struct kvm_arch *ka = &vcpu->kvm->arch;
2269 if (vcpu->vcpu_id == 0 && !host_initiated) {
2270 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2271 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2273 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2276 vcpu->arch.time = system_time;
2277 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2279 /* we verify if the enable bit is set... */
2280 if (system_time & 1) {
2281 kvm_gfn_to_pfn_cache_init(vcpu->kvm, &vcpu->arch.pv_time, vcpu,
2282 KVM_HOST_USES_PFN, system_time & ~1ULL,
2283 sizeof(struct pvclock_vcpu_time_info));
2285 kvm_gfn_to_pfn_cache_destroy(vcpu->kvm, &vcpu->arch.pv_time);
2291 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2293 do_shl32_div32(dividend, divisor);
2297 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2298 s8 *pshift, u32 *pmultiplier)
2306 scaled64 = scaled_hz;
2307 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2312 tps32 = (uint32_t)tps64;
2313 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2314 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2322 *pmultiplier = div_frac(scaled64, tps32);
2325 #ifdef CONFIG_X86_64
2326 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2329 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2330 static unsigned long max_tsc_khz;
2332 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2334 u64 v = (u64)khz * (1000000 + ppm);
2339 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2341 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2345 /* Guest TSC same frequency as host TSC? */
2347 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2351 /* TSC scaling supported? */
2352 if (!kvm_has_tsc_control) {
2353 if (user_tsc_khz > tsc_khz) {
2354 vcpu->arch.tsc_catchup = 1;
2355 vcpu->arch.tsc_always_catchup = 1;
2358 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2363 /* TSC scaling required - calculate ratio */
2364 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2365 user_tsc_khz, tsc_khz);
2367 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2368 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2373 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2377 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2379 u32 thresh_lo, thresh_hi;
2380 int use_scaling = 0;
2382 /* tsc_khz can be zero if TSC calibration fails */
2383 if (user_tsc_khz == 0) {
2384 /* set tsc_scaling_ratio to a safe value */
2385 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2389 /* Compute a scale to convert nanoseconds in TSC cycles */
2390 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2391 &vcpu->arch.virtual_tsc_shift,
2392 &vcpu->arch.virtual_tsc_mult);
2393 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2396 * Compute the variation in TSC rate which is acceptable
2397 * within the range of tolerance and decide if the
2398 * rate being applied is within that bounds of the hardware
2399 * rate. If so, no scaling or compensation need be done.
2401 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2402 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2403 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2404 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2407 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2410 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2412 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2413 vcpu->arch.virtual_tsc_mult,
2414 vcpu->arch.virtual_tsc_shift);
2415 tsc += vcpu->arch.this_tsc_write;
2419 #ifdef CONFIG_X86_64
2420 static inline int gtod_is_based_on_tsc(int mode)
2422 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2426 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2428 #ifdef CONFIG_X86_64
2430 struct kvm_arch *ka = &vcpu->kvm->arch;
2431 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2433 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2434 atomic_read(&vcpu->kvm->online_vcpus));
2437 * Once the masterclock is enabled, always perform request in
2438 * order to update it.
2440 * In order to enable masterclock, the host clocksource must be TSC
2441 * and the vcpus need to have matched TSCs. When that happens,
2442 * perform request to enable masterclock.
2444 if (ka->use_master_clock ||
2445 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2446 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2448 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2449 atomic_read(&vcpu->kvm->online_vcpus),
2450 ka->use_master_clock, gtod->clock.vclock_mode);
2455 * Multiply tsc by a fixed point number represented by ratio.
2457 * The most significant 64-N bits (mult) of ratio represent the
2458 * integral part of the fixed point number; the remaining N bits
2459 * (frac) represent the fractional part, ie. ratio represents a fixed
2460 * point number (mult + frac * 2^(-N)).
2462 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2464 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2466 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2469 u64 kvm_scale_tsc(u64 tsc, u64 ratio)
2473 if (ratio != kvm_default_tsc_scaling_ratio)
2474 _tsc = __scale_tsc(ratio, tsc);
2478 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2480 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2484 tsc = kvm_scale_tsc(rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2486 return target_tsc - tsc;
2489 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2491 return vcpu->arch.l1_tsc_offset +
2492 kvm_scale_tsc(host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2494 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2496 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2500 if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2501 nested_offset = l1_offset;
2503 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2504 kvm_tsc_scaling_ratio_frac_bits);
2506 nested_offset += l2_offset;
2507 return nested_offset;
2509 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2511 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2513 if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2514 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2515 kvm_tsc_scaling_ratio_frac_bits);
2517 return l1_multiplier;
2519 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2521 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2523 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2524 vcpu->arch.l1_tsc_offset,
2527 vcpu->arch.l1_tsc_offset = l1_offset;
2530 * If we are here because L1 chose not to trap WRMSR to TSC then
2531 * according to the spec this should set L1's TSC (as opposed to
2532 * setting L1's offset for L2).
2534 if (is_guest_mode(vcpu))
2535 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2537 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2538 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2540 vcpu->arch.tsc_offset = l1_offset;
2542 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2545 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2547 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2549 /* Userspace is changing the multiplier while L2 is active */
2550 if (is_guest_mode(vcpu))
2551 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2553 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2555 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2557 if (kvm_has_tsc_control)
2558 static_call(kvm_x86_write_tsc_multiplier)(
2559 vcpu, vcpu->arch.tsc_scaling_ratio);
2562 static inline bool kvm_check_tsc_unstable(void)
2564 #ifdef CONFIG_X86_64
2566 * TSC is marked unstable when we're running on Hyper-V,
2567 * 'TSC page' clocksource is good.
2569 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2572 return check_tsc_unstable();
2576 * Infers attempts to synchronize the guest's tsc from host writes. Sets the
2577 * offset for the vcpu and tracks the TSC matching generation that the vcpu
2580 static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
2581 u64 ns, bool matched)
2583 struct kvm *kvm = vcpu->kvm;
2585 lockdep_assert_held(&kvm->arch.tsc_write_lock);
2588 * We also track th most recent recorded KHZ, write and time to
2589 * allow the matching interval to be extended at each write.
2591 kvm->arch.last_tsc_nsec = ns;
2592 kvm->arch.last_tsc_write = tsc;
2593 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2594 kvm->arch.last_tsc_offset = offset;
2596 vcpu->arch.last_guest_tsc = tsc;
2598 kvm_vcpu_write_tsc_offset(vcpu, offset);
2602 * We split periods of matched TSC writes into generations.
2603 * For each generation, we track the original measured
2604 * nanosecond time, offset, and write, so if TSCs are in
2605 * sync, we can match exact offset, and if not, we can match
2606 * exact software computation in compute_guest_tsc()
2608 * These values are tracked in kvm->arch.cur_xxx variables.
2610 kvm->arch.cur_tsc_generation++;
2611 kvm->arch.cur_tsc_nsec = ns;
2612 kvm->arch.cur_tsc_write = tsc;
2613 kvm->arch.cur_tsc_offset = offset;
2614 kvm->arch.nr_vcpus_matched_tsc = 0;
2615 } else if (vcpu->arch.this_tsc_generation != kvm->arch.cur_tsc_generation) {
2616 kvm->arch.nr_vcpus_matched_tsc++;
2619 /* Keep track of which generation this VCPU has synchronized to */
2620 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2621 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2622 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2624 kvm_track_tsc_matching(vcpu);
2627 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2629 struct kvm *kvm = vcpu->kvm;
2630 u64 offset, ns, elapsed;
2631 unsigned long flags;
2632 bool matched = false;
2633 bool synchronizing = false;
2635 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2636 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2637 ns = get_kvmclock_base_ns();
2638 elapsed = ns - kvm->arch.last_tsc_nsec;
2640 if (vcpu->arch.virtual_tsc_khz) {
2643 * detection of vcpu initialization -- need to sync
2644 * with other vCPUs. This particularly helps to keep
2645 * kvm_clock stable after CPU hotplug
2647 synchronizing = true;
2649 u64 tsc_exp = kvm->arch.last_tsc_write +
2650 nsec_to_cycles(vcpu, elapsed);
2651 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2653 * Special case: TSC write with a small delta (1 second)
2654 * of virtual cycle time against real time is
2655 * interpreted as an attempt to synchronize the CPU.
2657 synchronizing = data < tsc_exp + tsc_hz &&
2658 data + tsc_hz > tsc_exp;
2663 * For a reliable TSC, we can match TSC offsets, and for an unstable
2664 * TSC, we add elapsed time in this computation. We could let the
2665 * compensation code attempt to catch up if we fall behind, but
2666 * it's better to try to match offsets from the beginning.
2668 if (synchronizing &&
2669 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2670 if (!kvm_check_tsc_unstable()) {
2671 offset = kvm->arch.cur_tsc_offset;
2673 u64 delta = nsec_to_cycles(vcpu, elapsed);
2675 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2680 __kvm_synchronize_tsc(vcpu, offset, data, ns, matched);
2681 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2684 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2687 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2688 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2691 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2693 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2694 WARN_ON(adjustment < 0);
2695 adjustment = kvm_scale_tsc((u64) adjustment,
2696 vcpu->arch.l1_tsc_scaling_ratio);
2697 adjust_tsc_offset_guest(vcpu, adjustment);
2700 #ifdef CONFIG_X86_64
2702 static u64 read_tsc(void)
2704 u64 ret = (u64)rdtsc_ordered();
2705 u64 last = pvclock_gtod_data.clock.cycle_last;
2707 if (likely(ret >= last))
2711 * GCC likes to generate cmov here, but this branch is extremely
2712 * predictable (it's just a function of time and the likely is
2713 * very likely) and there's a data dependence, so force GCC
2714 * to generate a branch instead. I don't barrier() because
2715 * we don't actually need a barrier, and if this function
2716 * ever gets inlined it will generate worse code.
2722 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2728 switch (clock->vclock_mode) {
2729 case VDSO_CLOCKMODE_HVCLOCK:
2730 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2732 if (tsc_pg_val != U64_MAX) {
2733 /* TSC page valid */
2734 *mode = VDSO_CLOCKMODE_HVCLOCK;
2735 v = (tsc_pg_val - clock->cycle_last) &
2738 /* TSC page invalid */
2739 *mode = VDSO_CLOCKMODE_NONE;
2742 case VDSO_CLOCKMODE_TSC:
2743 *mode = VDSO_CLOCKMODE_TSC;
2744 *tsc_timestamp = read_tsc();
2745 v = (*tsc_timestamp - clock->cycle_last) &
2749 *mode = VDSO_CLOCKMODE_NONE;
2752 if (*mode == VDSO_CLOCKMODE_NONE)
2753 *tsc_timestamp = v = 0;
2755 return v * clock->mult;
2758 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2760 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2766 seq = read_seqcount_begin(>od->seq);
2767 ns = gtod->raw_clock.base_cycles;
2768 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2769 ns >>= gtod->raw_clock.shift;
2770 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2771 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2777 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2779 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2785 seq = read_seqcount_begin(>od->seq);
2786 ts->tv_sec = gtod->wall_time_sec;
2787 ns = gtod->clock.base_cycles;
2788 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2789 ns >>= gtod->clock.shift;
2790 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2792 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2798 /* returns true if host is using TSC based clocksource */
2799 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2801 /* checked again under seqlock below */
2802 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2805 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2809 /* returns true if host is using TSC based clocksource */
2810 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2813 /* checked again under seqlock below */
2814 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2817 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2823 * Assuming a stable TSC across physical CPUS, and a stable TSC
2824 * across virtual CPUs, the following condition is possible.
2825 * Each numbered line represents an event visible to both
2826 * CPUs at the next numbered event.
2828 * "timespecX" represents host monotonic time. "tscX" represents
2831 * VCPU0 on CPU0 | VCPU1 on CPU1
2833 * 1. read timespec0,tsc0
2834 * 2. | timespec1 = timespec0 + N
2836 * 3. transition to guest | transition to guest
2837 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2838 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2839 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2841 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2844 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2846 * - 0 < N - M => M < N
2848 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2849 * always the case (the difference between two distinct xtime instances
2850 * might be smaller then the difference between corresponding TSC reads,
2851 * when updating guest vcpus pvclock areas).
2853 * To avoid that problem, do not allow visibility of distinct
2854 * system_timestamp/tsc_timestamp values simultaneously: use a master
2855 * copy of host monotonic time values. Update that master copy
2858 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2862 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2864 #ifdef CONFIG_X86_64
2865 struct kvm_arch *ka = &kvm->arch;
2867 bool host_tsc_clocksource, vcpus_matched;
2869 lockdep_assert_held(&kvm->arch.tsc_write_lock);
2870 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2871 atomic_read(&kvm->online_vcpus));
2874 * If the host uses TSC clock, then passthrough TSC as stable
2877 host_tsc_clocksource = kvm_get_time_and_clockread(
2878 &ka->master_kernel_ns,
2879 &ka->master_cycle_now);
2881 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2882 && !ka->backwards_tsc_observed
2883 && !ka->boot_vcpu_runs_old_kvmclock;
2885 if (ka->use_master_clock)
2886 atomic_set(&kvm_guest_has_master_clock, 1);
2888 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2889 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2894 static void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2896 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2899 static void __kvm_start_pvclock_update(struct kvm *kvm)
2901 raw_spin_lock_irq(&kvm->arch.tsc_write_lock);
2902 write_seqcount_begin(&kvm->arch.pvclock_sc);
2905 static void kvm_start_pvclock_update(struct kvm *kvm)
2907 kvm_make_mclock_inprogress_request(kvm);
2909 /* no guest entries from this point */
2910 __kvm_start_pvclock_update(kvm);
2913 static void kvm_end_pvclock_update(struct kvm *kvm)
2915 struct kvm_arch *ka = &kvm->arch;
2916 struct kvm_vcpu *vcpu;
2919 write_seqcount_end(&ka->pvclock_sc);
2920 raw_spin_unlock_irq(&ka->tsc_write_lock);
2921 kvm_for_each_vcpu(i, vcpu, kvm)
2922 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2924 /* guest entries allowed */
2925 kvm_for_each_vcpu(i, vcpu, kvm)
2926 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2929 static void kvm_update_masterclock(struct kvm *kvm)
2931 kvm_hv_request_tsc_page_update(kvm);
2932 kvm_start_pvclock_update(kvm);
2933 pvclock_update_vm_gtod_copy(kvm);
2934 kvm_end_pvclock_update(kvm);
2937 /* Called within read_seqcount_begin/retry for kvm->pvclock_sc. */
2938 static void __get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2940 struct kvm_arch *ka = &kvm->arch;
2941 struct pvclock_vcpu_time_info hv_clock;
2943 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2947 if (ka->use_master_clock && __this_cpu_read(cpu_tsc_khz)) {
2948 #ifdef CONFIG_X86_64
2949 struct timespec64 ts;
2951 if (kvm_get_walltime_and_clockread(&ts, &data->host_tsc)) {
2952 data->realtime = ts.tv_nsec + NSEC_PER_SEC * ts.tv_sec;
2953 data->flags |= KVM_CLOCK_REALTIME | KVM_CLOCK_HOST_TSC;
2956 data->host_tsc = rdtsc();
2958 data->flags |= KVM_CLOCK_TSC_STABLE;
2959 hv_clock.tsc_timestamp = ka->master_cycle_now;
2960 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2961 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2962 &hv_clock.tsc_shift,
2963 &hv_clock.tsc_to_system_mul);
2964 data->clock = __pvclock_read_cycles(&hv_clock, data->host_tsc);
2966 data->clock = get_kvmclock_base_ns() + ka->kvmclock_offset;
2972 static void get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2974 struct kvm_arch *ka = &kvm->arch;
2978 seq = read_seqcount_begin(&ka->pvclock_sc);
2979 __get_kvmclock(kvm, data);
2980 } while (read_seqcount_retry(&ka->pvclock_sc, seq));
2983 u64 get_kvmclock_ns(struct kvm *kvm)
2985 struct kvm_clock_data data;
2987 get_kvmclock(kvm, &data);
2991 static void kvm_setup_guest_pvclock(struct kvm_vcpu *v,
2992 struct gfn_to_pfn_cache *gpc,
2993 unsigned int offset)
2995 struct kvm_vcpu_arch *vcpu = &v->arch;
2996 struct pvclock_vcpu_time_info *guest_hv_clock;
2997 unsigned long flags;
2999 read_lock_irqsave(&gpc->lock, flags);
3000 while (!kvm_gfn_to_pfn_cache_check(v->kvm, gpc, gpc->gpa,
3001 offset + sizeof(*guest_hv_clock))) {
3002 read_unlock_irqrestore(&gpc->lock, flags);
3004 if (kvm_gfn_to_pfn_cache_refresh(v->kvm, gpc, gpc->gpa,
3005 offset + sizeof(*guest_hv_clock)))
3008 read_lock_irqsave(&gpc->lock, flags);
3011 guest_hv_clock = (void *)(gpc->khva + offset);
3014 * This VCPU is paused, but it's legal for a guest to read another
3015 * VCPU's kvmclock, so we really have to follow the specification where
3016 * it says that version is odd if data is being modified, and even after
3020 guest_hv_clock->version = vcpu->hv_clock.version = (guest_hv_clock->version + 1) | 1;
3023 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
3024 vcpu->hv_clock.flags |= (guest_hv_clock->flags & PVCLOCK_GUEST_STOPPED);
3026 if (vcpu->pvclock_set_guest_stopped_request) {
3027 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
3028 vcpu->pvclock_set_guest_stopped_request = false;
3031 memcpy(guest_hv_clock, &vcpu->hv_clock, sizeof(*guest_hv_clock));
3034 guest_hv_clock->version = ++vcpu->hv_clock.version;
3036 mark_page_dirty_in_slot(v->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT);
3037 read_unlock_irqrestore(&gpc->lock, flags);
3039 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
3042 static int kvm_guest_time_update(struct kvm_vcpu *v)
3044 unsigned long flags, tgt_tsc_khz;
3046 struct kvm_vcpu_arch *vcpu = &v->arch;
3047 struct kvm_arch *ka = &v->kvm->arch;
3049 u64 tsc_timestamp, host_tsc;
3051 bool use_master_clock;
3057 * If the host uses TSC clock, then passthrough TSC as stable
3061 seq = read_seqcount_begin(&ka->pvclock_sc);
3062 use_master_clock = ka->use_master_clock;
3063 if (use_master_clock) {
3064 host_tsc = ka->master_cycle_now;
3065 kernel_ns = ka->master_kernel_ns;
3067 } while (read_seqcount_retry(&ka->pvclock_sc, seq));
3069 /* Keep irq disabled to prevent changes to the clock */
3070 local_irq_save(flags);
3071 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
3072 if (unlikely(tgt_tsc_khz == 0)) {
3073 local_irq_restore(flags);
3074 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3077 if (!use_master_clock) {
3079 kernel_ns = get_kvmclock_base_ns();
3082 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
3085 * We may have to catch up the TSC to match elapsed wall clock
3086 * time for two reasons, even if kvmclock is used.
3087 * 1) CPU could have been running below the maximum TSC rate
3088 * 2) Broken TSC compensation resets the base at each VCPU
3089 * entry to avoid unknown leaps of TSC even when running
3090 * again on the same CPU. This may cause apparent elapsed
3091 * time to disappear, and the guest to stand still or run
3094 if (vcpu->tsc_catchup) {
3095 u64 tsc = compute_guest_tsc(v, kernel_ns);
3096 if (tsc > tsc_timestamp) {
3097 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3098 tsc_timestamp = tsc;
3102 local_irq_restore(flags);
3104 /* With all the info we got, fill in the values */
3106 if (kvm_has_tsc_control)
3107 tgt_tsc_khz = kvm_scale_tsc(tgt_tsc_khz,
3108 v->arch.l1_tsc_scaling_ratio);
3110 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3111 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3112 &vcpu->hv_clock.tsc_shift,
3113 &vcpu->hv_clock.tsc_to_system_mul);
3114 vcpu->hw_tsc_khz = tgt_tsc_khz;
3117 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3118 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3119 vcpu->last_guest_tsc = tsc_timestamp;
3121 /* If the host uses TSC clocksource, then it is stable */
3123 if (use_master_clock)
3124 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3126 vcpu->hv_clock.flags = pvclock_flags;
3128 if (vcpu->pv_time.active)
3129 kvm_setup_guest_pvclock(v, &vcpu->pv_time, 0);
3130 if (vcpu->xen.vcpu_info_cache.active)
3131 kvm_setup_guest_pvclock(v, &vcpu->xen.vcpu_info_cache,
3132 offsetof(struct compat_vcpu_info, time));
3133 if (vcpu->xen.vcpu_time_info_cache.active)
3134 kvm_setup_guest_pvclock(v, &vcpu->xen.vcpu_time_info_cache, 0);
3135 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3140 * kvmclock updates which are isolated to a given vcpu, such as
3141 * vcpu->cpu migration, should not allow system_timestamp from
3142 * the rest of the vcpus to remain static. Otherwise ntp frequency
3143 * correction applies to one vcpu's system_timestamp but not
3146 * So in those cases, request a kvmclock update for all vcpus.
3147 * We need to rate-limit these requests though, as they can
3148 * considerably slow guests that have a large number of vcpus.
3149 * The time for a remote vcpu to update its kvmclock is bound
3150 * by the delay we use to rate-limit the updates.
3153 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3155 static void kvmclock_update_fn(struct work_struct *work)
3158 struct delayed_work *dwork = to_delayed_work(work);
3159 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3160 kvmclock_update_work);
3161 struct kvm *kvm = container_of(ka, struct kvm, arch);
3162 struct kvm_vcpu *vcpu;
3164 kvm_for_each_vcpu(i, vcpu, kvm) {
3165 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3166 kvm_vcpu_kick(vcpu);
3170 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3172 struct kvm *kvm = v->kvm;
3174 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3175 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3176 KVMCLOCK_UPDATE_DELAY);
3179 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3181 static void kvmclock_sync_fn(struct work_struct *work)
3183 struct delayed_work *dwork = to_delayed_work(work);
3184 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3185 kvmclock_sync_work);
3186 struct kvm *kvm = container_of(ka, struct kvm, arch);
3188 if (!kvmclock_periodic_sync)
3191 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3192 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3193 KVMCLOCK_SYNC_PERIOD);
3197 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3199 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3201 /* McStatusWrEn enabled? */
3202 if (guest_cpuid_is_amd_or_hygon(vcpu))
3203 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3208 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3210 u64 mcg_cap = vcpu->arch.mcg_cap;
3211 unsigned bank_num = mcg_cap & 0xff;
3212 u32 msr = msr_info->index;
3213 u64 data = msr_info->data;
3216 case MSR_IA32_MCG_STATUS:
3217 vcpu->arch.mcg_status = data;
3219 case MSR_IA32_MCG_CTL:
3220 if (!(mcg_cap & MCG_CTL_P) &&
3221 (data || !msr_info->host_initiated))
3223 if (data != 0 && data != ~(u64)0)
3225 vcpu->arch.mcg_ctl = data;
3228 if (msr >= MSR_IA32_MC0_CTL &&
3229 msr < MSR_IA32_MCx_CTL(bank_num)) {
3230 u32 offset = array_index_nospec(
3231 msr - MSR_IA32_MC0_CTL,
3232 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3234 /* only 0 or all 1s can be written to IA32_MCi_CTL
3235 * some Linux kernels though clear bit 10 in bank 4 to
3236 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3237 * this to avoid an uncatched #GP in the guest
3239 if ((offset & 0x3) == 0 &&
3240 data != 0 && (data | (1 << 10)) != ~(u64)0)
3244 if (!msr_info->host_initiated &&
3245 (offset & 0x3) == 1 && data != 0) {
3246 if (!can_set_mci_status(vcpu))
3250 vcpu->arch.mce_banks[offset] = data;
3258 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3260 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3262 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3265 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3267 gpa_t gpa = data & ~0x3f;
3269 /* Bits 4:5 are reserved, Should be zero */
3273 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3274 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3277 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3278 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3281 if (!lapic_in_kernel(vcpu))
3282 return data ? 1 : 0;
3284 vcpu->arch.apf.msr_en_val = data;
3286 if (!kvm_pv_async_pf_enabled(vcpu)) {
3287 kvm_clear_async_pf_completion_queue(vcpu);
3288 kvm_async_pf_hash_reset(vcpu);
3292 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3296 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3297 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3299 kvm_async_pf_wakeup_all(vcpu);
3304 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3306 /* Bits 8-63 are reserved */
3310 if (!lapic_in_kernel(vcpu))
3313 vcpu->arch.apf.msr_int_val = data;
3315 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3320 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3322 kvm_gfn_to_pfn_cache_destroy(vcpu->kvm, &vcpu->arch.pv_time);
3323 vcpu->arch.time = 0;
3326 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3328 ++vcpu->stat.tlb_flush;
3329 static_call(kvm_x86_flush_tlb_all)(vcpu);
3332 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3334 ++vcpu->stat.tlb_flush;
3338 * A TLB flush on behalf of the guest is equivalent to
3339 * INVPCID(all), toggling CR4.PGE, etc., which requires
3340 * a forced sync of the shadow page tables. Ensure all the
3341 * roots are synced and the guest TLB in hardware is clean.
3343 kvm_mmu_sync_roots(vcpu);
3344 kvm_mmu_sync_prev_roots(vcpu);
3347 static_call(kvm_x86_flush_tlb_guest)(vcpu);
3351 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
3353 ++vcpu->stat.tlb_flush;
3354 static_call(kvm_x86_flush_tlb_current)(vcpu);
3358 * Service "local" TLB flush requests, which are specific to the current MMU
3359 * context. In addition to the generic event handling in vcpu_enter_guest(),
3360 * TLB flushes that are targeted at an MMU context also need to be serviced
3361 * prior before nested VM-Enter/VM-Exit.
3363 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu)
3365 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3366 kvm_vcpu_flush_tlb_current(vcpu);
3368 if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
3369 kvm_vcpu_flush_tlb_guest(vcpu);
3371 EXPORT_SYMBOL_GPL(kvm_service_local_tlb_flush_requests);
3373 static void record_steal_time(struct kvm_vcpu *vcpu)
3375 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
3376 struct kvm_steal_time __user *st;
3377 struct kvm_memslots *slots;
3381 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3382 kvm_xen_runstate_set_running(vcpu);
3386 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3389 if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
3392 slots = kvm_memslots(vcpu->kvm);
3394 if (unlikely(slots->generation != ghc->generation ||
3395 kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
3396 gfn_t gfn = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
3398 /* We rely on the fact that it fits in a single page. */
3399 BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
3401 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gfn, sizeof(*st)) ||
3402 kvm_is_error_hva(ghc->hva) || !ghc->memslot)
3406 st = (struct kvm_steal_time __user *)ghc->hva;
3408 * Doing a TLB flush here, on the guest's behalf, can avoid
3411 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3412 u8 st_preempted = 0;
3415 if (!user_access_begin(st, sizeof(*st)))
3418 asm volatile("1: xchgb %0, %2\n"
3421 _ASM_EXTABLE_UA(1b, 2b)
3422 : "+q" (st_preempted),
3424 "+m" (st->preempted));
3430 vcpu->arch.st.preempted = 0;
3432 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3433 st_preempted & KVM_VCPU_FLUSH_TLB);
3434 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3435 kvm_vcpu_flush_tlb_guest(vcpu);
3437 if (!user_access_begin(st, sizeof(*st)))
3440 if (!user_access_begin(st, sizeof(*st)))
3443 unsafe_put_user(0, &st->preempted, out);
3444 vcpu->arch.st.preempted = 0;
3447 unsafe_get_user(version, &st->version, out);
3449 version += 1; /* first time write, random junk */
3452 unsafe_put_user(version, &st->version, out);
3456 unsafe_get_user(steal, &st->steal, out);
3457 steal += current->sched_info.run_delay -
3458 vcpu->arch.st.last_steal;
3459 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3460 unsafe_put_user(steal, &st->steal, out);
3463 unsafe_put_user(version, &st->version, out);
3468 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
3471 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3474 u32 msr = msr_info->index;
3475 u64 data = msr_info->data;
3477 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3478 return kvm_xen_write_hypercall_page(vcpu, data);
3481 case MSR_AMD64_NB_CFG:
3482 case MSR_IA32_UCODE_WRITE:
3483 case MSR_VM_HSAVE_PA:
3484 case MSR_AMD64_PATCH_LOADER:
3485 case MSR_AMD64_BU_CFG2:
3486 case MSR_AMD64_DC_CFG:
3487 case MSR_F15H_EX_CFG:
3490 case MSR_IA32_UCODE_REV:
3491 if (msr_info->host_initiated)
3492 vcpu->arch.microcode_version = data;
3494 case MSR_IA32_ARCH_CAPABILITIES:
3495 if (!msr_info->host_initiated)
3497 vcpu->arch.arch_capabilities = data;
3499 case MSR_IA32_PERF_CAPABILITIES: {
3500 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3502 if (!msr_info->host_initiated)
3504 if (kvm_get_msr_feature(&msr_ent))
3506 if (data & ~msr_ent.data)
3509 vcpu->arch.perf_capabilities = data;
3514 return set_efer(vcpu, msr_info);
3516 data &= ~(u64)0x40; /* ignore flush filter disable */
3517 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3518 data &= ~(u64)0x8; /* ignore TLB cache disable */
3520 /* Handle McStatusWrEn */
3521 if (data == BIT_ULL(18)) {
3522 vcpu->arch.msr_hwcr = data;
3523 } else if (data != 0) {
3524 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3529 case MSR_FAM10H_MMIO_CONF_BASE:
3531 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3536 case 0x200 ... 0x2ff:
3537 return kvm_mtrr_set_msr(vcpu, msr, data);
3538 case MSR_IA32_APICBASE:
3539 return kvm_set_apic_base(vcpu, msr_info);
3540 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3541 return kvm_x2apic_msr_write(vcpu, msr, data);
3542 case MSR_IA32_TSC_DEADLINE:
3543 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3545 case MSR_IA32_TSC_ADJUST:
3546 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3547 if (!msr_info->host_initiated) {
3548 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3549 adjust_tsc_offset_guest(vcpu, adj);
3550 /* Before back to guest, tsc_timestamp must be adjusted
3551 * as well, otherwise guest's percpu pvclock time could jump.
3553 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3555 vcpu->arch.ia32_tsc_adjust_msr = data;
3558 case MSR_IA32_MISC_ENABLE:
3559 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3560 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3561 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3563 vcpu->arch.ia32_misc_enable_msr = data;
3564 kvm_update_cpuid_runtime(vcpu);
3566 vcpu->arch.ia32_misc_enable_msr = data;
3569 case MSR_IA32_SMBASE:
3570 if (!msr_info->host_initiated)
3572 vcpu->arch.smbase = data;
3574 case MSR_IA32_POWER_CTL:
3575 vcpu->arch.msr_ia32_power_ctl = data;
3578 if (msr_info->host_initiated) {
3579 kvm_synchronize_tsc(vcpu, data);
3581 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3582 adjust_tsc_offset_guest(vcpu, adj);
3583 vcpu->arch.ia32_tsc_adjust_msr += adj;
3587 if (!msr_info->host_initiated &&
3588 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3591 * KVM supports exposing PT to the guest, but does not support
3592 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3593 * XSAVES/XRSTORS to save/restore PT MSRs.
3595 if (data & ~supported_xss)
3597 vcpu->arch.ia32_xss = data;
3598 kvm_update_cpuid_runtime(vcpu);
3601 if (!msr_info->host_initiated)
3603 vcpu->arch.smi_count = data;
3605 case MSR_KVM_WALL_CLOCK_NEW:
3606 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3609 vcpu->kvm->arch.wall_clock = data;
3610 kvm_write_wall_clock(vcpu->kvm, data, 0);
3612 case MSR_KVM_WALL_CLOCK:
3613 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3616 vcpu->kvm->arch.wall_clock = data;
3617 kvm_write_wall_clock(vcpu->kvm, data, 0);
3619 case MSR_KVM_SYSTEM_TIME_NEW:
3620 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3623 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3625 case MSR_KVM_SYSTEM_TIME:
3626 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3629 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3631 case MSR_KVM_ASYNC_PF_EN:
3632 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3635 if (kvm_pv_enable_async_pf(vcpu, data))
3638 case MSR_KVM_ASYNC_PF_INT:
3639 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3642 if (kvm_pv_enable_async_pf_int(vcpu, data))
3645 case MSR_KVM_ASYNC_PF_ACK:
3646 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3649 vcpu->arch.apf.pageready_pending = false;
3650 kvm_check_async_pf_completion(vcpu);
3653 case MSR_KVM_STEAL_TIME:
3654 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3657 if (unlikely(!sched_info_on()))
3660 if (data & KVM_STEAL_RESERVED_MASK)
3663 vcpu->arch.st.msr_val = data;
3665 if (!(data & KVM_MSR_ENABLED))
3668 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3671 case MSR_KVM_PV_EOI_EN:
3672 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3675 if (kvm_lapic_set_pv_eoi(vcpu, data, sizeof(u8)))
3679 case MSR_KVM_POLL_CONTROL:
3680 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3683 /* only enable bit supported */
3684 if (data & (-1ULL << 1))
3687 vcpu->arch.msr_kvm_poll_control = data;
3690 case MSR_IA32_MCG_CTL:
3691 case MSR_IA32_MCG_STATUS:
3692 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3693 return set_msr_mce(vcpu, msr_info);
3695 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3696 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3699 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3700 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3701 if (kvm_pmu_is_valid_msr(vcpu, msr))
3702 return kvm_pmu_set_msr(vcpu, msr_info);
3704 if (pr || data != 0)
3705 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3706 "0x%x data 0x%llx\n", msr, data);
3708 case MSR_K7_CLK_CTL:
3710 * Ignore all writes to this no longer documented MSR.
3711 * Writes are only relevant for old K7 processors,
3712 * all pre-dating SVM, but a recommended workaround from
3713 * AMD for these chips. It is possible to specify the
3714 * affected processor models on the command line, hence
3715 * the need to ignore the workaround.
3718 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3719 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3720 case HV_X64_MSR_SYNDBG_OPTIONS:
3721 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3722 case HV_X64_MSR_CRASH_CTL:
3723 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3724 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3725 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3726 case HV_X64_MSR_TSC_EMULATION_STATUS:
3727 return kvm_hv_set_msr_common(vcpu, msr, data,
3728 msr_info->host_initiated);
3729 case MSR_IA32_BBL_CR_CTL3:
3730 /* Drop writes to this legacy MSR -- see rdmsr
3731 * counterpart for further detail.
3733 if (report_ignored_msrs)
3734 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3737 case MSR_AMD64_OSVW_ID_LENGTH:
3738 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3740 vcpu->arch.osvw.length = data;
3742 case MSR_AMD64_OSVW_STATUS:
3743 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3745 vcpu->arch.osvw.status = data;
3747 case MSR_PLATFORM_INFO:
3748 if (!msr_info->host_initiated ||
3749 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3750 cpuid_fault_enabled(vcpu)))
3752 vcpu->arch.msr_platform_info = data;
3754 case MSR_MISC_FEATURES_ENABLES:
3755 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3756 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3757 !supports_cpuid_fault(vcpu)))
3759 vcpu->arch.msr_misc_features_enables = data;
3761 #ifdef CONFIG_X86_64
3763 if (!msr_info->host_initiated &&
3764 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3767 if (data & ~kvm_guest_supported_xfd(vcpu))
3770 fpu_update_guest_xfd(&vcpu->arch.guest_fpu, data);
3772 case MSR_IA32_XFD_ERR:
3773 if (!msr_info->host_initiated &&
3774 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3777 if (data & ~kvm_guest_supported_xfd(vcpu))
3780 vcpu->arch.guest_fpu.xfd_err = data;
3784 if (kvm_pmu_is_valid_msr(vcpu, msr))
3785 return kvm_pmu_set_msr(vcpu, msr_info);
3786 return KVM_MSR_RET_INVALID;
3790 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3792 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3795 u64 mcg_cap = vcpu->arch.mcg_cap;
3796 unsigned bank_num = mcg_cap & 0xff;
3799 case MSR_IA32_P5_MC_ADDR:
3800 case MSR_IA32_P5_MC_TYPE:
3803 case MSR_IA32_MCG_CAP:
3804 data = vcpu->arch.mcg_cap;
3806 case MSR_IA32_MCG_CTL:
3807 if (!(mcg_cap & MCG_CTL_P) && !host)
3809 data = vcpu->arch.mcg_ctl;
3811 case MSR_IA32_MCG_STATUS:
3812 data = vcpu->arch.mcg_status;
3815 if (msr >= MSR_IA32_MC0_CTL &&
3816 msr < MSR_IA32_MCx_CTL(bank_num)) {
3817 u32 offset = array_index_nospec(
3818 msr - MSR_IA32_MC0_CTL,
3819 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3821 data = vcpu->arch.mce_banks[offset];
3830 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3832 switch (msr_info->index) {
3833 case MSR_IA32_PLATFORM_ID:
3834 case MSR_IA32_EBL_CR_POWERON:
3835 case MSR_IA32_LASTBRANCHFROMIP:
3836 case MSR_IA32_LASTBRANCHTOIP:
3837 case MSR_IA32_LASTINTFROMIP:
3838 case MSR_IA32_LASTINTTOIP:
3839 case MSR_AMD64_SYSCFG:
3840 case MSR_K8_TSEG_ADDR:
3841 case MSR_K8_TSEG_MASK:
3842 case MSR_VM_HSAVE_PA:
3843 case MSR_K8_INT_PENDING_MSG:
3844 case MSR_AMD64_NB_CFG:
3845 case MSR_FAM10H_MMIO_CONF_BASE:
3846 case MSR_AMD64_BU_CFG2:
3847 case MSR_IA32_PERF_CTL:
3848 case MSR_AMD64_DC_CFG:
3849 case MSR_F15H_EX_CFG:
3851 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3852 * limit) MSRs. Just return 0, as we do not want to expose the host
3853 * data here. Do not conditionalize this on CPUID, as KVM does not do
3854 * so for existing CPU-specific MSRs.
3856 case MSR_RAPL_POWER_UNIT:
3857 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3858 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3859 case MSR_PKG_ENERGY_STATUS: /* Total package */
3860 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3863 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3864 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3865 return kvm_pmu_get_msr(vcpu, msr_info);
3866 if (!msr_info->host_initiated)
3870 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3871 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3872 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3873 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3874 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3875 return kvm_pmu_get_msr(vcpu, msr_info);
3878 case MSR_IA32_UCODE_REV:
3879 msr_info->data = vcpu->arch.microcode_version;
3881 case MSR_IA32_ARCH_CAPABILITIES:
3882 if (!msr_info->host_initiated &&
3883 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3885 msr_info->data = vcpu->arch.arch_capabilities;
3887 case MSR_IA32_PERF_CAPABILITIES:
3888 if (!msr_info->host_initiated &&
3889 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3891 msr_info->data = vcpu->arch.perf_capabilities;
3893 case MSR_IA32_POWER_CTL:
3894 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3896 case MSR_IA32_TSC: {
3898 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3899 * even when not intercepted. AMD manual doesn't explicitly
3900 * state this but appears to behave the same.
3902 * On userspace reads and writes, however, we unconditionally
3903 * return L1's TSC value to ensure backwards-compatible
3904 * behavior for migration.
3908 if (msr_info->host_initiated) {
3909 offset = vcpu->arch.l1_tsc_offset;
3910 ratio = vcpu->arch.l1_tsc_scaling_ratio;
3912 offset = vcpu->arch.tsc_offset;
3913 ratio = vcpu->arch.tsc_scaling_ratio;
3916 msr_info->data = kvm_scale_tsc(rdtsc(), ratio) + offset;
3920 case 0x200 ... 0x2ff:
3921 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3922 case 0xcd: /* fsb frequency */
3926 * MSR_EBC_FREQUENCY_ID
3927 * Conservative value valid for even the basic CPU models.
3928 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3929 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3930 * and 266MHz for model 3, or 4. Set Core Clock
3931 * Frequency to System Bus Frequency Ratio to 1 (bits
3932 * 31:24) even though these are only valid for CPU
3933 * models > 2, however guests may end up dividing or
3934 * multiplying by zero otherwise.
3936 case MSR_EBC_FREQUENCY_ID:
3937 msr_info->data = 1 << 24;
3939 case MSR_IA32_APICBASE:
3940 msr_info->data = kvm_get_apic_base(vcpu);
3942 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3943 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3944 case MSR_IA32_TSC_DEADLINE:
3945 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3947 case MSR_IA32_TSC_ADJUST:
3948 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3950 case MSR_IA32_MISC_ENABLE:
3951 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3953 case MSR_IA32_SMBASE:
3954 if (!msr_info->host_initiated)
3956 msr_info->data = vcpu->arch.smbase;
3959 msr_info->data = vcpu->arch.smi_count;
3961 case MSR_IA32_PERF_STATUS:
3962 /* TSC increment by tick */
3963 msr_info->data = 1000ULL;
3964 /* CPU multiplier */
3965 msr_info->data |= (((uint64_t)4ULL) << 40);
3968 msr_info->data = vcpu->arch.efer;
3970 case MSR_KVM_WALL_CLOCK:
3971 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3974 msr_info->data = vcpu->kvm->arch.wall_clock;
3976 case MSR_KVM_WALL_CLOCK_NEW:
3977 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3980 msr_info->data = vcpu->kvm->arch.wall_clock;
3982 case MSR_KVM_SYSTEM_TIME:
3983 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3986 msr_info->data = vcpu->arch.time;
3988 case MSR_KVM_SYSTEM_TIME_NEW:
3989 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3992 msr_info->data = vcpu->arch.time;
3994 case MSR_KVM_ASYNC_PF_EN:
3995 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3998 msr_info->data = vcpu->arch.apf.msr_en_val;
4000 case MSR_KVM_ASYNC_PF_INT:
4001 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
4004 msr_info->data = vcpu->arch.apf.msr_int_val;
4006 case MSR_KVM_ASYNC_PF_ACK:
4007 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
4012 case MSR_KVM_STEAL_TIME:
4013 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
4016 msr_info->data = vcpu->arch.st.msr_val;
4018 case MSR_KVM_PV_EOI_EN:
4019 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
4022 msr_info->data = vcpu->arch.pv_eoi.msr_val;
4024 case MSR_KVM_POLL_CONTROL:
4025 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
4028 msr_info->data = vcpu->arch.msr_kvm_poll_control;
4030 case MSR_IA32_P5_MC_ADDR:
4031 case MSR_IA32_P5_MC_TYPE:
4032 case MSR_IA32_MCG_CAP:
4033 case MSR_IA32_MCG_CTL:
4034 case MSR_IA32_MCG_STATUS:
4035 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
4036 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
4037 msr_info->host_initiated);
4039 if (!msr_info->host_initiated &&
4040 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
4042 msr_info->data = vcpu->arch.ia32_xss;
4044 case MSR_K7_CLK_CTL:
4046 * Provide expected ramp-up count for K7. All other
4047 * are set to zero, indicating minimum divisors for
4050 * This prevents guest kernels on AMD host with CPU
4051 * type 6, model 8 and higher from exploding due to
4052 * the rdmsr failing.
4054 msr_info->data = 0x20000000;
4056 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
4057 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
4058 case HV_X64_MSR_SYNDBG_OPTIONS:
4059 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
4060 case HV_X64_MSR_CRASH_CTL:
4061 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
4062 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
4063 case HV_X64_MSR_TSC_EMULATION_CONTROL:
4064 case HV_X64_MSR_TSC_EMULATION_STATUS:
4065 return kvm_hv_get_msr_common(vcpu,
4066 msr_info->index, &msr_info->data,
4067 msr_info->host_initiated);
4068 case MSR_IA32_BBL_CR_CTL3:
4069 /* This legacy MSR exists but isn't fully documented in current
4070 * silicon. It is however accessed by winxp in very narrow
4071 * scenarios where it sets bit #19, itself documented as
4072 * a "reserved" bit. Best effort attempt to source coherent
4073 * read data here should the balance of the register be
4074 * interpreted by the guest:
4076 * L2 cache control register 3: 64GB range, 256KB size,
4077 * enabled, latency 0x1, configured
4079 msr_info->data = 0xbe702111;
4081 case MSR_AMD64_OSVW_ID_LENGTH:
4082 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4084 msr_info->data = vcpu->arch.osvw.length;
4086 case MSR_AMD64_OSVW_STATUS:
4087 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4089 msr_info->data = vcpu->arch.osvw.status;
4091 case MSR_PLATFORM_INFO:
4092 if (!msr_info->host_initiated &&
4093 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
4095 msr_info->data = vcpu->arch.msr_platform_info;
4097 case MSR_MISC_FEATURES_ENABLES:
4098 msr_info->data = vcpu->arch.msr_misc_features_enables;
4101 msr_info->data = vcpu->arch.msr_hwcr;
4103 #ifdef CONFIG_X86_64
4105 if (!msr_info->host_initiated &&
4106 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4109 msr_info->data = vcpu->arch.guest_fpu.fpstate->xfd;
4111 case MSR_IA32_XFD_ERR:
4112 if (!msr_info->host_initiated &&
4113 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4116 msr_info->data = vcpu->arch.guest_fpu.xfd_err;
4120 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
4121 return kvm_pmu_get_msr(vcpu, msr_info);
4122 return KVM_MSR_RET_INVALID;
4126 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
4129 * Read or write a bunch of msrs. All parameters are kernel addresses.
4131 * @return number of msrs set successfully.
4133 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
4134 struct kvm_msr_entry *entries,
4135 int (*do_msr)(struct kvm_vcpu *vcpu,
4136 unsigned index, u64 *data))
4140 for (i = 0; i < msrs->nmsrs; ++i)
4141 if (do_msr(vcpu, entries[i].index, &entries[i].data))
4148 * Read or write a bunch of msrs. Parameters are user addresses.
4150 * @return number of msrs set successfully.
4152 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
4153 int (*do_msr)(struct kvm_vcpu *vcpu,
4154 unsigned index, u64 *data),
4157 struct kvm_msrs msrs;
4158 struct kvm_msr_entry *entries;
4163 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
4167 if (msrs.nmsrs >= MAX_IO_MSRS)
4170 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
4171 entries = memdup_user(user_msrs->entries, size);
4172 if (IS_ERR(entries)) {
4173 r = PTR_ERR(entries);
4177 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
4182 if (writeback && copy_to_user(user_msrs->entries, entries, size))
4193 static inline bool kvm_can_mwait_in_guest(void)
4195 return boot_cpu_has(X86_FEATURE_MWAIT) &&
4196 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
4197 boot_cpu_has(X86_FEATURE_ARAT);
4200 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4201 struct kvm_cpuid2 __user *cpuid_arg)
4203 struct kvm_cpuid2 cpuid;
4207 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4210 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4215 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4221 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4226 case KVM_CAP_IRQCHIP:
4228 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4229 case KVM_CAP_SET_TSS_ADDR:
4230 case KVM_CAP_EXT_CPUID:
4231 case KVM_CAP_EXT_EMUL_CPUID:
4232 case KVM_CAP_CLOCKSOURCE:
4234 case KVM_CAP_NOP_IO_DELAY:
4235 case KVM_CAP_MP_STATE:
4236 case KVM_CAP_SYNC_MMU:
4237 case KVM_CAP_USER_NMI:
4238 case KVM_CAP_REINJECT_CONTROL:
4239 case KVM_CAP_IRQ_INJECT_STATUS:
4240 case KVM_CAP_IOEVENTFD:
4241 case KVM_CAP_IOEVENTFD_NO_LENGTH:
4243 case KVM_CAP_PIT_STATE2:
4244 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4245 case KVM_CAP_VCPU_EVENTS:
4246 case KVM_CAP_HYPERV:
4247 case KVM_CAP_HYPERV_VAPIC:
4248 case KVM_CAP_HYPERV_SPIN:
4249 case KVM_CAP_HYPERV_SYNIC:
4250 case KVM_CAP_HYPERV_SYNIC2:
4251 case KVM_CAP_HYPERV_VP_INDEX:
4252 case KVM_CAP_HYPERV_EVENTFD:
4253 case KVM_CAP_HYPERV_TLBFLUSH:
4254 case KVM_CAP_HYPERV_SEND_IPI:
4255 case KVM_CAP_HYPERV_CPUID:
4256 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4257 case KVM_CAP_SYS_HYPERV_CPUID:
4258 case KVM_CAP_PCI_SEGMENT:
4259 case KVM_CAP_DEBUGREGS:
4260 case KVM_CAP_X86_ROBUST_SINGLESTEP:
4262 case KVM_CAP_ASYNC_PF:
4263 case KVM_CAP_ASYNC_PF_INT:
4264 case KVM_CAP_GET_TSC_KHZ:
4265 case KVM_CAP_KVMCLOCK_CTRL:
4266 case KVM_CAP_READONLY_MEM:
4267 case KVM_CAP_HYPERV_TIME:
4268 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4269 case KVM_CAP_TSC_DEADLINE_TIMER:
4270 case KVM_CAP_DISABLE_QUIRKS:
4271 case KVM_CAP_SET_BOOT_CPU_ID:
4272 case KVM_CAP_SPLIT_IRQCHIP:
4273 case KVM_CAP_IMMEDIATE_EXIT:
4274 case KVM_CAP_PMU_EVENT_FILTER:
4275 case KVM_CAP_GET_MSR_FEATURES:
4276 case KVM_CAP_MSR_PLATFORM_INFO:
4277 case KVM_CAP_EXCEPTION_PAYLOAD:
4278 case KVM_CAP_SET_GUEST_DEBUG:
4279 case KVM_CAP_LAST_CPU:
4280 case KVM_CAP_X86_USER_SPACE_MSR:
4281 case KVM_CAP_X86_MSR_FILTER:
4282 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4283 #ifdef CONFIG_X86_SGX_KVM
4284 case KVM_CAP_SGX_ATTRIBUTE:
4286 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4287 case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
4288 case KVM_CAP_SREGS2:
4289 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4290 case KVM_CAP_VCPU_ATTRIBUTES:
4291 case KVM_CAP_SYS_ATTRIBUTES:
4293 case KVM_CAP_ENABLE_CAP:
4296 case KVM_CAP_EXIT_HYPERCALL:
4297 r = KVM_EXIT_HYPERCALL_VALID_MASK;
4299 case KVM_CAP_SET_GUEST_DEBUG2:
4300 return KVM_GUESTDBG_VALID_MASK;
4301 #ifdef CONFIG_KVM_XEN
4302 case KVM_CAP_XEN_HVM:
4303 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4304 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4305 KVM_XEN_HVM_CONFIG_SHARED_INFO |
4306 KVM_XEN_HVM_CONFIG_EVTCHN_2LEVEL |
4307 KVM_XEN_HVM_CONFIG_EVTCHN_SEND;
4308 if (sched_info_on())
4309 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4312 case KVM_CAP_SYNC_REGS:
4313 r = KVM_SYNC_X86_VALID_FIELDS;
4315 case KVM_CAP_ADJUST_CLOCK:
4316 r = KVM_CLOCK_VALID_FLAGS;
4318 case KVM_CAP_X86_DISABLE_EXITS:
4319 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4320 KVM_X86_DISABLE_EXITS_CSTATE;
4321 if(kvm_can_mwait_in_guest())
4322 r |= KVM_X86_DISABLE_EXITS_MWAIT;
4324 case KVM_CAP_X86_SMM:
4325 /* SMBASE is usually relocated above 1M on modern chipsets,
4326 * and SMM handlers might indeed rely on 4G segment limits,
4327 * so do not report SMM to be available if real mode is
4328 * emulated via vm86 mode. Still, do not go to great lengths
4329 * to avoid userspace's usage of the feature, because it is a
4330 * fringe case that is not enabled except via specific settings
4331 * of the module parameters.
4333 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4335 case KVM_CAP_NR_VCPUS:
4336 r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
4338 case KVM_CAP_MAX_VCPUS:
4341 case KVM_CAP_MAX_VCPU_ID:
4342 r = KVM_MAX_VCPU_IDS;
4344 case KVM_CAP_PV_MMU: /* obsolete */
4348 r = KVM_MAX_MCE_BANKS;
4351 r = boot_cpu_has(X86_FEATURE_XSAVE);
4353 case KVM_CAP_TSC_CONTROL:
4354 case KVM_CAP_VM_TSC_CONTROL:
4355 r = kvm_has_tsc_control;
4357 case KVM_CAP_X2APIC_API:
4358 r = KVM_X2APIC_API_VALID_FLAGS;
4360 case KVM_CAP_NESTED_STATE:
4361 r = kvm_x86_ops.nested_ops->get_state ?
4362 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4364 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4365 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4367 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4368 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4370 case KVM_CAP_SMALLER_MAXPHYADDR:
4371 r = (int) allow_smaller_maxphyaddr;
4373 case KVM_CAP_STEAL_TIME:
4374 r = sched_info_on();
4376 case KVM_CAP_X86_BUS_LOCK_EXIT:
4377 if (kvm_has_bus_lock_exit)
4378 r = KVM_BUS_LOCK_DETECTION_OFF |
4379 KVM_BUS_LOCK_DETECTION_EXIT;
4383 case KVM_CAP_XSAVE2: {
4384 u64 guest_perm = xstate_get_guest_group_perm();
4386 r = xstate_required_size(supported_xcr0 & guest_perm, false);
4387 if (r < sizeof(struct kvm_xsave))
4388 r = sizeof(struct kvm_xsave);
4390 case KVM_CAP_PMU_CAPABILITY:
4391 r = enable_pmu ? KVM_CAP_PMU_VALID_MASK : 0;
4394 case KVM_CAP_DISABLE_QUIRKS2:
4395 r = KVM_X86_VALID_QUIRKS;
4403 static inline void __user *kvm_get_attr_addr(struct kvm_device_attr *attr)
4405 void __user *uaddr = (void __user*)(unsigned long)attr->addr;
4407 if ((u64)(unsigned long)uaddr != attr->addr)
4408 return ERR_PTR_USR(-EFAULT);
4412 static int kvm_x86_dev_get_attr(struct kvm_device_attr *attr)
4414 u64 __user *uaddr = kvm_get_attr_addr(attr);
4420 return PTR_ERR(uaddr);
4422 switch (attr->attr) {
4423 case KVM_X86_XCOMP_GUEST_SUPP:
4424 if (put_user(supported_xcr0, uaddr))
4433 static int kvm_x86_dev_has_attr(struct kvm_device_attr *attr)
4438 switch (attr->attr) {
4439 case KVM_X86_XCOMP_GUEST_SUPP:
4446 long kvm_arch_dev_ioctl(struct file *filp,
4447 unsigned int ioctl, unsigned long arg)
4449 void __user *argp = (void __user *)arg;
4453 case KVM_GET_MSR_INDEX_LIST: {
4454 struct kvm_msr_list __user *user_msr_list = argp;
4455 struct kvm_msr_list msr_list;
4459 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4462 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4463 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4466 if (n < msr_list.nmsrs)
4469 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4470 num_msrs_to_save * sizeof(u32)))
4472 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4474 num_emulated_msrs * sizeof(u32)))
4479 case KVM_GET_SUPPORTED_CPUID:
4480 case KVM_GET_EMULATED_CPUID: {
4481 struct kvm_cpuid2 __user *cpuid_arg = argp;
4482 struct kvm_cpuid2 cpuid;
4485 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4488 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4494 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4499 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4501 if (copy_to_user(argp, &kvm_mce_cap_supported,
4502 sizeof(kvm_mce_cap_supported)))
4506 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4507 struct kvm_msr_list __user *user_msr_list = argp;
4508 struct kvm_msr_list msr_list;
4512 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4515 msr_list.nmsrs = num_msr_based_features;
4516 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4519 if (n < msr_list.nmsrs)
4522 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4523 num_msr_based_features * sizeof(u32)))
4529 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4531 case KVM_GET_SUPPORTED_HV_CPUID:
4532 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4534 case KVM_GET_DEVICE_ATTR: {
4535 struct kvm_device_attr attr;
4537 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4539 r = kvm_x86_dev_get_attr(&attr);
4542 case KVM_HAS_DEVICE_ATTR: {
4543 struct kvm_device_attr attr;
4545 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4547 r = kvm_x86_dev_has_attr(&attr);
4558 static void wbinvd_ipi(void *garbage)
4563 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4565 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4568 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4570 /* Address WBINVD may be executed by guest */
4571 if (need_emulate_wbinvd(vcpu)) {
4572 if (static_call(kvm_x86_has_wbinvd_exit)())
4573 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4574 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4575 smp_call_function_single(vcpu->cpu,
4576 wbinvd_ipi, NULL, 1);
4579 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4581 /* Save host pkru register if supported */
4582 vcpu->arch.host_pkru = read_pkru();
4584 /* Apply any externally detected TSC adjustments (due to suspend) */
4585 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4586 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4587 vcpu->arch.tsc_offset_adjustment = 0;
4588 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4591 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4592 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4593 rdtsc() - vcpu->arch.last_host_tsc;
4595 mark_tsc_unstable("KVM discovered backwards TSC");
4597 if (kvm_check_tsc_unstable()) {
4598 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4599 vcpu->arch.last_guest_tsc);
4600 kvm_vcpu_write_tsc_offset(vcpu, offset);
4601 vcpu->arch.tsc_catchup = 1;
4604 if (kvm_lapic_hv_timer_in_use(vcpu))
4605 kvm_lapic_restart_hv_timer(vcpu);
4608 * On a host with synchronized TSC, there is no need to update
4609 * kvmclock on vcpu->cpu migration
4611 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4612 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4613 if (vcpu->cpu != cpu)
4614 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4618 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4621 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4623 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
4624 struct kvm_steal_time __user *st;
4625 struct kvm_memslots *slots;
4626 static const u8 preempted = KVM_VCPU_PREEMPTED;
4628 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4631 if (vcpu->arch.st.preempted)
4634 /* This happens on process exit */
4635 if (unlikely(current->mm != vcpu->kvm->mm))
4638 slots = kvm_memslots(vcpu->kvm);
4640 if (unlikely(slots->generation != ghc->generation ||
4641 kvm_is_error_hva(ghc->hva) || !ghc->memslot))
4644 st = (struct kvm_steal_time __user *)ghc->hva;
4645 BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
4647 if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
4648 vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4650 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
4653 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4657 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4658 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4661 * Take the srcu lock as memslots will be accessed to check the gfn
4662 * cache generation against the memslots generation.
4664 idx = srcu_read_lock(&vcpu->kvm->srcu);
4665 if (kvm_xen_msr_enabled(vcpu->kvm))
4666 kvm_xen_runstate_set_preempted(vcpu);
4668 kvm_steal_time_set_preempted(vcpu);
4669 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4671 static_call(kvm_x86_vcpu_put)(vcpu);
4672 vcpu->arch.last_host_tsc = rdtsc();
4675 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4676 struct kvm_lapic_state *s)
4678 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
4680 return kvm_apic_get_state(vcpu, s);
4683 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4684 struct kvm_lapic_state *s)
4688 r = kvm_apic_set_state(vcpu, s);
4691 update_cr8_intercept(vcpu);
4696 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4699 * We can accept userspace's request for interrupt injection
4700 * as long as we have a place to store the interrupt number.
4701 * The actual injection will happen when the CPU is able to
4702 * deliver the interrupt.
4704 if (kvm_cpu_has_extint(vcpu))
4707 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4708 return (!lapic_in_kernel(vcpu) ||
4709 kvm_apic_accept_pic_intr(vcpu));
4712 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4715 * Do not cause an interrupt window exit if an exception
4716 * is pending or an event needs reinjection; userspace
4717 * might want to inject the interrupt manually using KVM_SET_REGS
4718 * or KVM_SET_SREGS. For that to work, we must be at an
4719 * instruction boundary and with no events half-injected.
4721 return (kvm_arch_interrupt_allowed(vcpu) &&
4722 kvm_cpu_accept_dm_intr(vcpu) &&
4723 !kvm_event_needs_reinjection(vcpu) &&
4724 !vcpu->arch.exception.pending);
4727 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4728 struct kvm_interrupt *irq)
4730 if (irq->irq >= KVM_NR_INTERRUPTS)
4733 if (!irqchip_in_kernel(vcpu->kvm)) {
4734 kvm_queue_interrupt(vcpu, irq->irq, false);
4735 kvm_make_request(KVM_REQ_EVENT, vcpu);
4740 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4741 * fail for in-kernel 8259.
4743 if (pic_in_kernel(vcpu->kvm))
4746 if (vcpu->arch.pending_external_vector != -1)
4749 vcpu->arch.pending_external_vector = irq->irq;
4750 kvm_make_request(KVM_REQ_EVENT, vcpu);
4754 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4756 kvm_inject_nmi(vcpu);
4761 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4763 kvm_make_request(KVM_REQ_SMI, vcpu);
4768 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4769 struct kvm_tpr_access_ctl *tac)
4773 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4777 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4781 unsigned bank_num = mcg_cap & 0xff, bank;
4784 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4786 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4789 vcpu->arch.mcg_cap = mcg_cap;
4790 /* Init IA32_MCG_CTL to all 1s */
4791 if (mcg_cap & MCG_CTL_P)
4792 vcpu->arch.mcg_ctl = ~(u64)0;
4793 /* Init IA32_MCi_CTL to all 1s */
4794 for (bank = 0; bank < bank_num; bank++)
4795 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4797 static_call(kvm_x86_setup_mce)(vcpu);
4802 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4803 struct kvm_x86_mce *mce)
4805 u64 mcg_cap = vcpu->arch.mcg_cap;
4806 unsigned bank_num = mcg_cap & 0xff;
4807 u64 *banks = vcpu->arch.mce_banks;
4809 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4812 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4813 * reporting is disabled
4815 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4816 vcpu->arch.mcg_ctl != ~(u64)0)
4818 banks += 4 * mce->bank;
4820 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4821 * reporting is disabled for the bank
4823 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4825 if (mce->status & MCI_STATUS_UC) {
4826 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4827 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4828 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4831 if (banks[1] & MCI_STATUS_VAL)
4832 mce->status |= MCI_STATUS_OVER;
4833 banks[2] = mce->addr;
4834 banks[3] = mce->misc;
4835 vcpu->arch.mcg_status = mce->mcg_status;
4836 banks[1] = mce->status;
4837 kvm_queue_exception(vcpu, MC_VECTOR);
4838 } else if (!(banks[1] & MCI_STATUS_VAL)
4839 || !(banks[1] & MCI_STATUS_UC)) {
4840 if (banks[1] & MCI_STATUS_VAL)
4841 mce->status |= MCI_STATUS_OVER;
4842 banks[2] = mce->addr;
4843 banks[3] = mce->misc;
4844 banks[1] = mce->status;
4846 banks[1] |= MCI_STATUS_OVER;
4850 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4851 struct kvm_vcpu_events *events)
4855 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4859 * In guest mode, payload delivery should be deferred,
4860 * so that the L1 hypervisor can intercept #PF before
4861 * CR2 is modified (or intercept #DB before DR6 is
4862 * modified under nVMX). Unless the per-VM capability,
4863 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4864 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4865 * opportunistically defer the exception payload, deliver it if the
4866 * capability hasn't been requested before processing a
4867 * KVM_GET_VCPU_EVENTS.
4869 if (!vcpu->kvm->arch.exception_payload_enabled &&
4870 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4871 kvm_deliver_exception_payload(vcpu);
4874 * The API doesn't provide the instruction length for software
4875 * exceptions, so don't report them. As long as the guest RIP
4876 * isn't advanced, we should expect to encounter the exception
4879 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4880 events->exception.injected = 0;
4881 events->exception.pending = 0;
4883 events->exception.injected = vcpu->arch.exception.injected;
4884 events->exception.pending = vcpu->arch.exception.pending;
4886 * For ABI compatibility, deliberately conflate
4887 * pending and injected exceptions when
4888 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4890 if (!vcpu->kvm->arch.exception_payload_enabled)
4891 events->exception.injected |=
4892 vcpu->arch.exception.pending;
4894 events->exception.nr = vcpu->arch.exception.nr;
4895 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4896 events->exception.error_code = vcpu->arch.exception.error_code;
4897 events->exception_has_payload = vcpu->arch.exception.has_payload;
4898 events->exception_payload = vcpu->arch.exception.payload;
4900 events->interrupt.injected =
4901 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4902 events->interrupt.nr = vcpu->arch.interrupt.nr;
4903 events->interrupt.soft = 0;
4904 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4906 events->nmi.injected = vcpu->arch.nmi_injected;
4907 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4908 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4909 events->nmi.pad = 0;
4911 events->sipi_vector = 0; /* never valid when reporting to user space */
4913 events->smi.smm = is_smm(vcpu);
4914 events->smi.pending = vcpu->arch.smi_pending;
4915 events->smi.smm_inside_nmi =
4916 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4917 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4919 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4920 | KVM_VCPUEVENT_VALID_SHADOW
4921 | KVM_VCPUEVENT_VALID_SMM);
4922 if (vcpu->kvm->arch.exception_payload_enabled)
4923 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4925 memset(&events->reserved, 0, sizeof(events->reserved));
4928 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4930 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4931 struct kvm_vcpu_events *events)
4933 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4934 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4935 | KVM_VCPUEVENT_VALID_SHADOW
4936 | KVM_VCPUEVENT_VALID_SMM
4937 | KVM_VCPUEVENT_VALID_PAYLOAD))
4940 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4941 if (!vcpu->kvm->arch.exception_payload_enabled)
4943 if (events->exception.pending)
4944 events->exception.injected = 0;
4946 events->exception_has_payload = 0;
4948 events->exception.pending = 0;
4949 events->exception_has_payload = 0;
4952 if ((events->exception.injected || events->exception.pending) &&
4953 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4956 /* INITs are latched while in SMM */
4957 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4958 (events->smi.smm || events->smi.pending) &&
4959 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4963 vcpu->arch.exception.injected = events->exception.injected;
4964 vcpu->arch.exception.pending = events->exception.pending;
4965 vcpu->arch.exception.nr = events->exception.nr;
4966 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4967 vcpu->arch.exception.error_code = events->exception.error_code;
4968 vcpu->arch.exception.has_payload = events->exception_has_payload;
4969 vcpu->arch.exception.payload = events->exception_payload;
4971 vcpu->arch.interrupt.injected = events->interrupt.injected;
4972 vcpu->arch.interrupt.nr = events->interrupt.nr;
4973 vcpu->arch.interrupt.soft = events->interrupt.soft;
4974 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4975 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4976 events->interrupt.shadow);
4978 vcpu->arch.nmi_injected = events->nmi.injected;
4979 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4980 vcpu->arch.nmi_pending = events->nmi.pending;
4981 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4983 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4984 lapic_in_kernel(vcpu))
4985 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4987 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4988 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4989 kvm_x86_ops.nested_ops->leave_nested(vcpu);
4990 kvm_smm_changed(vcpu, events->smi.smm);
4993 vcpu->arch.smi_pending = events->smi.pending;
4995 if (events->smi.smm) {
4996 if (events->smi.smm_inside_nmi)
4997 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4999 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
5002 if (lapic_in_kernel(vcpu)) {
5003 if (events->smi.latched_init)
5004 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
5006 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
5010 kvm_make_request(KVM_REQ_EVENT, vcpu);
5015 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
5016 struct kvm_debugregs *dbgregs)
5020 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
5021 kvm_get_dr(vcpu, 6, &val);
5023 dbgregs->dr7 = vcpu->arch.dr7;
5025 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
5028 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
5029 struct kvm_debugregs *dbgregs)
5034 if (!kvm_dr6_valid(dbgregs->dr6))
5036 if (!kvm_dr7_valid(dbgregs->dr7))
5039 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
5040 kvm_update_dr0123(vcpu);
5041 vcpu->arch.dr6 = dbgregs->dr6;
5042 vcpu->arch.dr7 = dbgregs->dr7;
5043 kvm_update_dr7(vcpu);
5048 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
5049 struct kvm_xsave *guest_xsave)
5051 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5054 fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
5055 guest_xsave->region,
5056 sizeof(guest_xsave->region),
5060 static void kvm_vcpu_ioctl_x86_get_xsave2(struct kvm_vcpu *vcpu,
5061 u8 *state, unsigned int size)
5063 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5066 fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
5067 state, size, vcpu->arch.pkru);
5070 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
5071 struct kvm_xsave *guest_xsave)
5073 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5076 return fpu_copy_uabi_to_guest_fpstate(&vcpu->arch.guest_fpu,
5077 guest_xsave->region,
5078 supported_xcr0, &vcpu->arch.pkru);
5081 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
5082 struct kvm_xcrs *guest_xcrs)
5084 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
5085 guest_xcrs->nr_xcrs = 0;
5089 guest_xcrs->nr_xcrs = 1;
5090 guest_xcrs->flags = 0;
5091 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
5092 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
5095 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
5096 struct kvm_xcrs *guest_xcrs)
5100 if (!boot_cpu_has(X86_FEATURE_XSAVE))
5103 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
5106 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
5107 /* Only support XCR0 currently */
5108 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
5109 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
5110 guest_xcrs->xcrs[i].value);
5119 * kvm_set_guest_paused() indicates to the guest kernel that it has been
5120 * stopped by the hypervisor. This function will be called from the host only.
5121 * EINVAL is returned when the host attempts to set the flag for a guest that
5122 * does not support pv clocks.
5124 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
5126 if (!vcpu->arch.pv_time.active)
5128 vcpu->arch.pvclock_set_guest_stopped_request = true;
5129 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5133 static int kvm_arch_tsc_has_attr(struct kvm_vcpu *vcpu,
5134 struct kvm_device_attr *attr)
5138 switch (attr->attr) {
5139 case KVM_VCPU_TSC_OFFSET:
5149 static int kvm_arch_tsc_get_attr(struct kvm_vcpu *vcpu,
5150 struct kvm_device_attr *attr)
5152 u64 __user *uaddr = kvm_get_attr_addr(attr);
5156 return PTR_ERR(uaddr);
5158 switch (attr->attr) {
5159 case KVM_VCPU_TSC_OFFSET:
5161 if (put_user(vcpu->arch.l1_tsc_offset, uaddr))
5172 static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
5173 struct kvm_device_attr *attr)
5175 u64 __user *uaddr = kvm_get_attr_addr(attr);
5176 struct kvm *kvm = vcpu->kvm;
5180 return PTR_ERR(uaddr);
5182 switch (attr->attr) {
5183 case KVM_VCPU_TSC_OFFSET: {
5184 u64 offset, tsc, ns;
5185 unsigned long flags;
5189 if (get_user(offset, uaddr))
5192 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
5194 matched = (vcpu->arch.virtual_tsc_khz &&
5195 kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
5196 kvm->arch.last_tsc_offset == offset);
5198 tsc = kvm_scale_tsc(rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
5199 ns = get_kvmclock_base_ns();
5201 __kvm_synchronize_tsc(vcpu, offset, tsc, ns, matched);
5202 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
5214 static int kvm_vcpu_ioctl_device_attr(struct kvm_vcpu *vcpu,
5218 struct kvm_device_attr attr;
5221 if (copy_from_user(&attr, argp, sizeof(attr)))
5224 if (attr.group != KVM_VCPU_TSC_CTRL)
5228 case KVM_HAS_DEVICE_ATTR:
5229 r = kvm_arch_tsc_has_attr(vcpu, &attr);
5231 case KVM_GET_DEVICE_ATTR:
5232 r = kvm_arch_tsc_get_attr(vcpu, &attr);
5234 case KVM_SET_DEVICE_ATTR:
5235 r = kvm_arch_tsc_set_attr(vcpu, &attr);
5242 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5243 struct kvm_enable_cap *cap)
5246 uint16_t vmcs_version;
5247 void __user *user_ptr;
5253 case KVM_CAP_HYPERV_SYNIC2:
5258 case KVM_CAP_HYPERV_SYNIC:
5259 if (!irqchip_in_kernel(vcpu->kvm))
5261 return kvm_hv_activate_synic(vcpu, cap->cap ==
5262 KVM_CAP_HYPERV_SYNIC2);
5263 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
5264 if (!kvm_x86_ops.nested_ops->enable_evmcs)
5266 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
5268 user_ptr = (void __user *)(uintptr_t)cap->args[0];
5269 if (copy_to_user(user_ptr, &vmcs_version,
5270 sizeof(vmcs_version)))
5274 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
5275 if (!kvm_x86_ops.enable_direct_tlbflush)
5278 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
5280 case KVM_CAP_HYPERV_ENFORCE_CPUID:
5281 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
5283 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
5284 vcpu->arch.pv_cpuid.enforce = cap->args[0];
5285 if (vcpu->arch.pv_cpuid.enforce)
5286 kvm_update_pv_runtime(vcpu);
5294 long kvm_arch_vcpu_ioctl(struct file *filp,
5295 unsigned int ioctl, unsigned long arg)
5297 struct kvm_vcpu *vcpu = filp->private_data;
5298 void __user *argp = (void __user *)arg;
5301 struct kvm_sregs2 *sregs2;
5302 struct kvm_lapic_state *lapic;
5303 struct kvm_xsave *xsave;
5304 struct kvm_xcrs *xcrs;
5312 case KVM_GET_LAPIC: {
5314 if (!lapic_in_kernel(vcpu))
5316 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5317 GFP_KERNEL_ACCOUNT);
5322 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5326 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5331 case KVM_SET_LAPIC: {
5333 if (!lapic_in_kernel(vcpu))
5335 u.lapic = memdup_user(argp, sizeof(*u.lapic));
5336 if (IS_ERR(u.lapic)) {
5337 r = PTR_ERR(u.lapic);
5341 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5344 case KVM_INTERRUPT: {
5345 struct kvm_interrupt irq;
5348 if (copy_from_user(&irq, argp, sizeof(irq)))
5350 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5354 r = kvm_vcpu_ioctl_nmi(vcpu);
5358 r = kvm_vcpu_ioctl_smi(vcpu);
5361 case KVM_SET_CPUID: {
5362 struct kvm_cpuid __user *cpuid_arg = argp;
5363 struct kvm_cpuid cpuid;
5366 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5368 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5371 case KVM_SET_CPUID2: {
5372 struct kvm_cpuid2 __user *cpuid_arg = argp;
5373 struct kvm_cpuid2 cpuid;
5376 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5378 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5379 cpuid_arg->entries);
5382 case KVM_GET_CPUID2: {
5383 struct kvm_cpuid2 __user *cpuid_arg = argp;
5384 struct kvm_cpuid2 cpuid;
5387 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5389 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5390 cpuid_arg->entries);
5394 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5399 case KVM_GET_MSRS: {
5400 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5401 r = msr_io(vcpu, argp, do_get_msr, 1);
5402 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5405 case KVM_SET_MSRS: {
5406 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5407 r = msr_io(vcpu, argp, do_set_msr, 0);
5408 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5411 case KVM_TPR_ACCESS_REPORTING: {
5412 struct kvm_tpr_access_ctl tac;
5415 if (copy_from_user(&tac, argp, sizeof(tac)))
5417 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5421 if (copy_to_user(argp, &tac, sizeof(tac)))
5426 case KVM_SET_VAPIC_ADDR: {
5427 struct kvm_vapic_addr va;
5431 if (!lapic_in_kernel(vcpu))
5434 if (copy_from_user(&va, argp, sizeof(va)))
5436 idx = srcu_read_lock(&vcpu->kvm->srcu);
5437 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5438 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5441 case KVM_X86_SETUP_MCE: {
5445 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5447 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5450 case KVM_X86_SET_MCE: {
5451 struct kvm_x86_mce mce;
5454 if (copy_from_user(&mce, argp, sizeof(mce)))
5456 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5459 case KVM_GET_VCPU_EVENTS: {
5460 struct kvm_vcpu_events events;
5462 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5465 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5470 case KVM_SET_VCPU_EVENTS: {
5471 struct kvm_vcpu_events events;
5474 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5477 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5480 case KVM_GET_DEBUGREGS: {
5481 struct kvm_debugregs dbgregs;
5483 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5486 if (copy_to_user(argp, &dbgregs,
5487 sizeof(struct kvm_debugregs)))
5492 case KVM_SET_DEBUGREGS: {
5493 struct kvm_debugregs dbgregs;
5496 if (copy_from_user(&dbgregs, argp,
5497 sizeof(struct kvm_debugregs)))
5500 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5503 case KVM_GET_XSAVE: {
5505 if (vcpu->arch.guest_fpu.uabi_size > sizeof(struct kvm_xsave))
5508 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5513 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5516 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5521 case KVM_SET_XSAVE: {
5522 int size = vcpu->arch.guest_fpu.uabi_size;
5524 u.xsave = memdup_user(argp, size);
5525 if (IS_ERR(u.xsave)) {
5526 r = PTR_ERR(u.xsave);
5530 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5534 case KVM_GET_XSAVE2: {
5535 int size = vcpu->arch.guest_fpu.uabi_size;
5537 u.xsave = kzalloc(size, GFP_KERNEL_ACCOUNT);
5542 kvm_vcpu_ioctl_x86_get_xsave2(vcpu, u.buffer, size);
5545 if (copy_to_user(argp, u.xsave, size))
5552 case KVM_GET_XCRS: {
5553 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5558 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5561 if (copy_to_user(argp, u.xcrs,
5562 sizeof(struct kvm_xcrs)))
5567 case KVM_SET_XCRS: {
5568 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5569 if (IS_ERR(u.xcrs)) {
5570 r = PTR_ERR(u.xcrs);
5574 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5577 case KVM_SET_TSC_KHZ: {
5581 user_tsc_khz = (u32)arg;
5583 if (kvm_has_tsc_control &&
5584 user_tsc_khz >= kvm_max_guest_tsc_khz)
5587 if (user_tsc_khz == 0)
5588 user_tsc_khz = tsc_khz;
5590 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5595 case KVM_GET_TSC_KHZ: {
5596 r = vcpu->arch.virtual_tsc_khz;
5599 case KVM_KVMCLOCK_CTRL: {
5600 r = kvm_set_guest_paused(vcpu);
5603 case KVM_ENABLE_CAP: {
5604 struct kvm_enable_cap cap;
5607 if (copy_from_user(&cap, argp, sizeof(cap)))
5609 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5612 case KVM_GET_NESTED_STATE: {
5613 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5617 if (!kvm_x86_ops.nested_ops->get_state)
5620 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5622 if (get_user(user_data_size, &user_kvm_nested_state->size))
5625 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5630 if (r > user_data_size) {
5631 if (put_user(r, &user_kvm_nested_state->size))
5641 case KVM_SET_NESTED_STATE: {
5642 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5643 struct kvm_nested_state kvm_state;
5647 if (!kvm_x86_ops.nested_ops->set_state)
5651 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5655 if (kvm_state.size < sizeof(kvm_state))
5658 if (kvm_state.flags &
5659 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5660 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5661 | KVM_STATE_NESTED_GIF_SET))
5664 /* nested_run_pending implies guest_mode. */
5665 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5666 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5669 idx = srcu_read_lock(&vcpu->kvm->srcu);
5670 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5671 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5674 case KVM_GET_SUPPORTED_HV_CPUID:
5675 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5677 #ifdef CONFIG_KVM_XEN
5678 case KVM_XEN_VCPU_GET_ATTR: {
5679 struct kvm_xen_vcpu_attr xva;
5682 if (copy_from_user(&xva, argp, sizeof(xva)))
5684 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5685 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5689 case KVM_XEN_VCPU_SET_ATTR: {
5690 struct kvm_xen_vcpu_attr xva;
5693 if (copy_from_user(&xva, argp, sizeof(xva)))
5695 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5699 case KVM_GET_SREGS2: {
5700 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5704 __get_sregs2(vcpu, u.sregs2);
5706 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5711 case KVM_SET_SREGS2: {
5712 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5713 if (IS_ERR(u.sregs2)) {
5714 r = PTR_ERR(u.sregs2);
5718 r = __set_sregs2(vcpu, u.sregs2);
5721 case KVM_HAS_DEVICE_ATTR:
5722 case KVM_GET_DEVICE_ATTR:
5723 case KVM_SET_DEVICE_ATTR:
5724 r = kvm_vcpu_ioctl_device_attr(vcpu, ioctl, argp);
5736 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5738 return VM_FAULT_SIGBUS;
5741 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5745 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5747 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5751 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5754 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5757 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5758 unsigned long kvm_nr_mmu_pages)
5760 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5763 mutex_lock(&kvm->slots_lock);
5765 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5766 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5768 mutex_unlock(&kvm->slots_lock);
5772 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5774 return kvm->arch.n_max_mmu_pages;
5777 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5779 struct kvm_pic *pic = kvm->arch.vpic;
5783 switch (chip->chip_id) {
5784 case KVM_IRQCHIP_PIC_MASTER:
5785 memcpy(&chip->chip.pic, &pic->pics[0],
5786 sizeof(struct kvm_pic_state));
5788 case KVM_IRQCHIP_PIC_SLAVE:
5789 memcpy(&chip->chip.pic, &pic->pics[1],
5790 sizeof(struct kvm_pic_state));
5792 case KVM_IRQCHIP_IOAPIC:
5793 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5802 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5804 struct kvm_pic *pic = kvm->arch.vpic;
5808 switch (chip->chip_id) {
5809 case KVM_IRQCHIP_PIC_MASTER:
5810 spin_lock(&pic->lock);
5811 memcpy(&pic->pics[0], &chip->chip.pic,
5812 sizeof(struct kvm_pic_state));
5813 spin_unlock(&pic->lock);
5815 case KVM_IRQCHIP_PIC_SLAVE:
5816 spin_lock(&pic->lock);
5817 memcpy(&pic->pics[1], &chip->chip.pic,
5818 sizeof(struct kvm_pic_state));
5819 spin_unlock(&pic->lock);
5821 case KVM_IRQCHIP_IOAPIC:
5822 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5828 kvm_pic_update_irq(pic);
5832 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5834 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5836 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5838 mutex_lock(&kps->lock);
5839 memcpy(ps, &kps->channels, sizeof(*ps));
5840 mutex_unlock(&kps->lock);
5844 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5847 struct kvm_pit *pit = kvm->arch.vpit;
5849 mutex_lock(&pit->pit_state.lock);
5850 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5851 for (i = 0; i < 3; i++)
5852 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5853 mutex_unlock(&pit->pit_state.lock);
5857 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5859 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5860 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5861 sizeof(ps->channels));
5862 ps->flags = kvm->arch.vpit->pit_state.flags;
5863 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5864 memset(&ps->reserved, 0, sizeof(ps->reserved));
5868 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5872 u32 prev_legacy, cur_legacy;
5873 struct kvm_pit *pit = kvm->arch.vpit;
5875 mutex_lock(&pit->pit_state.lock);
5876 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5877 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5878 if (!prev_legacy && cur_legacy)
5880 memcpy(&pit->pit_state.channels, &ps->channels,
5881 sizeof(pit->pit_state.channels));
5882 pit->pit_state.flags = ps->flags;
5883 for (i = 0; i < 3; i++)
5884 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5886 mutex_unlock(&pit->pit_state.lock);
5890 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5891 struct kvm_reinject_control *control)
5893 struct kvm_pit *pit = kvm->arch.vpit;
5895 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5896 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5897 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5899 mutex_lock(&pit->pit_state.lock);
5900 kvm_pit_set_reinject(pit, control->pit_reinject);
5901 mutex_unlock(&pit->pit_state.lock);
5906 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5910 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5911 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5912 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5915 struct kvm_vcpu *vcpu;
5918 kvm_for_each_vcpu(i, vcpu, kvm)
5919 kvm_vcpu_kick(vcpu);
5922 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5925 if (!irqchip_in_kernel(kvm))
5928 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5929 irq_event->irq, irq_event->level,
5934 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5935 struct kvm_enable_cap *cap)
5943 case KVM_CAP_DISABLE_QUIRKS2:
5945 if (cap->args[0] & ~KVM_X86_VALID_QUIRKS)
5948 case KVM_CAP_DISABLE_QUIRKS:
5949 kvm->arch.disabled_quirks = cap->args[0];
5952 case KVM_CAP_SPLIT_IRQCHIP: {
5953 mutex_lock(&kvm->lock);
5955 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5956 goto split_irqchip_unlock;
5958 if (irqchip_in_kernel(kvm))
5959 goto split_irqchip_unlock;
5960 if (kvm->created_vcpus)
5961 goto split_irqchip_unlock;
5962 r = kvm_setup_empty_irq_routing(kvm);
5964 goto split_irqchip_unlock;
5965 /* Pairs with irqchip_in_kernel. */
5967 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5968 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5969 kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_ABSENT);
5971 split_irqchip_unlock:
5972 mutex_unlock(&kvm->lock);
5975 case KVM_CAP_X2APIC_API:
5977 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5980 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5981 kvm->arch.x2apic_format = true;
5982 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5983 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5987 case KVM_CAP_X86_DISABLE_EXITS:
5989 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5992 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5993 kvm_can_mwait_in_guest())
5994 kvm->arch.mwait_in_guest = true;
5995 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5996 kvm->arch.hlt_in_guest = true;
5997 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5998 kvm->arch.pause_in_guest = true;
5999 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
6000 kvm->arch.cstate_in_guest = true;
6003 case KVM_CAP_MSR_PLATFORM_INFO:
6004 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
6007 case KVM_CAP_EXCEPTION_PAYLOAD:
6008 kvm->arch.exception_payload_enabled = cap->args[0];
6011 case KVM_CAP_X86_USER_SPACE_MSR:
6012 kvm->arch.user_space_msr_mask = cap->args[0];
6015 case KVM_CAP_X86_BUS_LOCK_EXIT:
6017 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
6020 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
6021 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
6024 if (kvm_has_bus_lock_exit &&
6025 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
6026 kvm->arch.bus_lock_detection_enabled = true;
6029 #ifdef CONFIG_X86_SGX_KVM
6030 case KVM_CAP_SGX_ATTRIBUTE: {
6031 unsigned long allowed_attributes = 0;
6033 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
6037 /* KVM only supports the PROVISIONKEY privileged attribute. */
6038 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
6039 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
6040 kvm->arch.sgx_provisioning_allowed = true;
6046 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
6048 if (!kvm_x86_ops.vm_copy_enc_context_from)
6051 r = static_call(kvm_x86_vm_copy_enc_context_from)(kvm, cap->args[0]);
6053 case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
6055 if (!kvm_x86_ops.vm_move_enc_context_from)
6058 r = static_call(kvm_x86_vm_move_enc_context_from)(kvm, cap->args[0]);
6060 case KVM_CAP_EXIT_HYPERCALL:
6061 if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
6065 kvm->arch.hypercall_exit_enabled = cap->args[0];
6068 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
6070 if (cap->args[0] & ~1)
6072 kvm->arch.exit_on_emulation_error = cap->args[0];
6075 case KVM_CAP_PMU_CAPABILITY:
6077 if (!enable_pmu || (cap->args[0] & ~KVM_CAP_PMU_VALID_MASK))
6080 mutex_lock(&kvm->lock);
6081 if (!kvm->created_vcpus) {
6082 kvm->arch.enable_pmu = !(cap->args[0] & KVM_PMU_CAP_DISABLE);
6085 mutex_unlock(&kvm->lock);
6094 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
6096 struct kvm_x86_msr_filter *msr_filter;
6098 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
6102 msr_filter->default_allow = default_allow;
6106 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
6113 for (i = 0; i < msr_filter->count; i++)
6114 kfree(msr_filter->ranges[i].bitmap);
6119 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
6120 struct kvm_msr_filter_range *user_range)
6122 unsigned long *bitmap = NULL;
6125 if (!user_range->nmsrs)
6128 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
6131 if (!user_range->flags)
6134 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
6135 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
6138 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
6140 return PTR_ERR(bitmap);
6142 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
6143 .flags = user_range->flags,
6144 .base = user_range->base,
6145 .nmsrs = user_range->nmsrs,
6149 msr_filter->count++;
6153 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
6155 struct kvm_msr_filter __user *user_msr_filter = argp;
6156 struct kvm_x86_msr_filter *new_filter, *old_filter;
6157 struct kvm_msr_filter filter;
6163 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
6166 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
6167 empty &= !filter.ranges[i].nmsrs;
6169 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
6170 if (empty && !default_allow)
6173 new_filter = kvm_alloc_msr_filter(default_allow);
6177 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
6178 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
6180 kvm_free_msr_filter(new_filter);
6185 mutex_lock(&kvm->lock);
6187 /* The per-VM filter is protected by kvm->lock... */
6188 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
6190 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
6191 synchronize_srcu(&kvm->srcu);
6193 kvm_free_msr_filter(old_filter);
6195 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
6196 mutex_unlock(&kvm->lock);
6201 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
6202 static int kvm_arch_suspend_notifier(struct kvm *kvm)
6204 struct kvm_vcpu *vcpu;
6208 mutex_lock(&kvm->lock);
6209 kvm_for_each_vcpu(i, vcpu, kvm) {
6210 if (!vcpu->arch.pv_time.active)
6213 ret = kvm_set_guest_paused(vcpu);
6215 kvm_err("Failed to pause guest VCPU%d: %d\n",
6216 vcpu->vcpu_id, ret);
6220 mutex_unlock(&kvm->lock);
6222 return ret ? NOTIFY_BAD : NOTIFY_DONE;
6225 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
6228 case PM_HIBERNATION_PREPARE:
6229 case PM_SUSPEND_PREPARE:
6230 return kvm_arch_suspend_notifier(kvm);
6235 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
6237 static int kvm_vm_ioctl_get_clock(struct kvm *kvm, void __user *argp)
6239 struct kvm_clock_data data = { 0 };
6241 get_kvmclock(kvm, &data);
6242 if (copy_to_user(argp, &data, sizeof(data)))
6248 static int kvm_vm_ioctl_set_clock(struct kvm *kvm, void __user *argp)
6250 struct kvm_arch *ka = &kvm->arch;
6251 struct kvm_clock_data data;
6254 if (copy_from_user(&data, argp, sizeof(data)))
6258 * Only KVM_CLOCK_REALTIME is used, but allow passing the
6259 * result of KVM_GET_CLOCK back to KVM_SET_CLOCK.
6261 if (data.flags & ~KVM_CLOCK_VALID_FLAGS)
6264 kvm_hv_request_tsc_page_update(kvm);
6265 kvm_start_pvclock_update(kvm);
6266 pvclock_update_vm_gtod_copy(kvm);
6269 * This pairs with kvm_guest_time_update(): when masterclock is
6270 * in use, we use master_kernel_ns + kvmclock_offset to set
6271 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6272 * is slightly ahead) here we risk going negative on unsigned
6273 * 'system_time' when 'data.clock' is very small.
6275 if (data.flags & KVM_CLOCK_REALTIME) {
6276 u64 now_real_ns = ktime_get_real_ns();
6279 * Avoid stepping the kvmclock backwards.
6281 if (now_real_ns > data.realtime)
6282 data.clock += now_real_ns - data.realtime;
6285 if (ka->use_master_clock)
6286 now_raw_ns = ka->master_kernel_ns;
6288 now_raw_ns = get_kvmclock_base_ns();
6289 ka->kvmclock_offset = data.clock - now_raw_ns;
6290 kvm_end_pvclock_update(kvm);
6294 long kvm_arch_vm_ioctl(struct file *filp,
6295 unsigned int ioctl, unsigned long arg)
6297 struct kvm *kvm = filp->private_data;
6298 void __user *argp = (void __user *)arg;
6301 * This union makes it completely explicit to gcc-3.x
6302 * that these two variables' stack usage should be
6303 * combined, not added together.
6306 struct kvm_pit_state ps;
6307 struct kvm_pit_state2 ps2;
6308 struct kvm_pit_config pit_config;
6312 case KVM_SET_TSS_ADDR:
6313 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
6315 case KVM_SET_IDENTITY_MAP_ADDR: {
6318 mutex_lock(&kvm->lock);
6320 if (kvm->created_vcpus)
6321 goto set_identity_unlock;
6323 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
6324 goto set_identity_unlock;
6325 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
6326 set_identity_unlock:
6327 mutex_unlock(&kvm->lock);
6330 case KVM_SET_NR_MMU_PAGES:
6331 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
6333 case KVM_GET_NR_MMU_PAGES:
6334 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
6336 case KVM_CREATE_IRQCHIP: {
6337 mutex_lock(&kvm->lock);
6340 if (irqchip_in_kernel(kvm))
6341 goto create_irqchip_unlock;
6344 if (kvm->created_vcpus)
6345 goto create_irqchip_unlock;
6347 r = kvm_pic_init(kvm);
6349 goto create_irqchip_unlock;
6351 r = kvm_ioapic_init(kvm);
6353 kvm_pic_destroy(kvm);
6354 goto create_irqchip_unlock;
6357 r = kvm_setup_default_irq_routing(kvm);
6359 kvm_ioapic_destroy(kvm);
6360 kvm_pic_destroy(kvm);
6361 goto create_irqchip_unlock;
6363 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6365 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6366 kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_ABSENT);
6367 create_irqchip_unlock:
6368 mutex_unlock(&kvm->lock);
6371 case KVM_CREATE_PIT:
6372 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6374 case KVM_CREATE_PIT2:
6376 if (copy_from_user(&u.pit_config, argp,
6377 sizeof(struct kvm_pit_config)))
6380 mutex_lock(&kvm->lock);
6383 goto create_pit_unlock;
6385 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6389 mutex_unlock(&kvm->lock);
6391 case KVM_GET_IRQCHIP: {
6392 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6393 struct kvm_irqchip *chip;
6395 chip = memdup_user(argp, sizeof(*chip));
6402 if (!irqchip_kernel(kvm))
6403 goto get_irqchip_out;
6404 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6406 goto get_irqchip_out;
6408 if (copy_to_user(argp, chip, sizeof(*chip)))
6409 goto get_irqchip_out;
6415 case KVM_SET_IRQCHIP: {
6416 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6417 struct kvm_irqchip *chip;
6419 chip = memdup_user(argp, sizeof(*chip));
6426 if (!irqchip_kernel(kvm))
6427 goto set_irqchip_out;
6428 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6435 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6438 if (!kvm->arch.vpit)
6440 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6444 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6451 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6453 mutex_lock(&kvm->lock);
6455 if (!kvm->arch.vpit)
6457 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6459 mutex_unlock(&kvm->lock);
6462 case KVM_GET_PIT2: {
6464 if (!kvm->arch.vpit)
6466 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6470 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6475 case KVM_SET_PIT2: {
6477 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6479 mutex_lock(&kvm->lock);
6481 if (!kvm->arch.vpit)
6483 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6485 mutex_unlock(&kvm->lock);
6488 case KVM_REINJECT_CONTROL: {
6489 struct kvm_reinject_control control;
6491 if (copy_from_user(&control, argp, sizeof(control)))
6494 if (!kvm->arch.vpit)
6496 r = kvm_vm_ioctl_reinject(kvm, &control);
6499 case KVM_SET_BOOT_CPU_ID:
6501 mutex_lock(&kvm->lock);
6502 if (kvm->created_vcpus)
6505 kvm->arch.bsp_vcpu_id = arg;
6506 mutex_unlock(&kvm->lock);
6508 #ifdef CONFIG_KVM_XEN
6509 case KVM_XEN_HVM_CONFIG: {
6510 struct kvm_xen_hvm_config xhc;
6512 if (copy_from_user(&xhc, argp, sizeof(xhc)))
6514 r = kvm_xen_hvm_config(kvm, &xhc);
6517 case KVM_XEN_HVM_GET_ATTR: {
6518 struct kvm_xen_hvm_attr xha;
6521 if (copy_from_user(&xha, argp, sizeof(xha)))
6523 r = kvm_xen_hvm_get_attr(kvm, &xha);
6524 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6528 case KVM_XEN_HVM_SET_ATTR: {
6529 struct kvm_xen_hvm_attr xha;
6532 if (copy_from_user(&xha, argp, sizeof(xha)))
6534 r = kvm_xen_hvm_set_attr(kvm, &xha);
6537 case KVM_XEN_HVM_EVTCHN_SEND: {
6538 struct kvm_irq_routing_xen_evtchn uxe;
6541 if (copy_from_user(&uxe, argp, sizeof(uxe)))
6543 r = kvm_xen_hvm_evtchn_send(kvm, &uxe);
6548 r = kvm_vm_ioctl_set_clock(kvm, argp);
6551 r = kvm_vm_ioctl_get_clock(kvm, argp);
6553 case KVM_SET_TSC_KHZ: {
6557 user_tsc_khz = (u32)arg;
6559 if (kvm_has_tsc_control &&
6560 user_tsc_khz >= kvm_max_guest_tsc_khz)
6563 if (user_tsc_khz == 0)
6564 user_tsc_khz = tsc_khz;
6566 WRITE_ONCE(kvm->arch.default_tsc_khz, user_tsc_khz);
6571 case KVM_GET_TSC_KHZ: {
6572 r = READ_ONCE(kvm->arch.default_tsc_khz);
6575 case KVM_MEMORY_ENCRYPT_OP: {
6577 if (!kvm_x86_ops.mem_enc_ioctl)
6580 r = static_call(kvm_x86_mem_enc_ioctl)(kvm, argp);
6583 case KVM_MEMORY_ENCRYPT_REG_REGION: {
6584 struct kvm_enc_region region;
6587 if (copy_from_user(®ion, argp, sizeof(region)))
6591 if (!kvm_x86_ops.mem_enc_register_region)
6594 r = static_call(kvm_x86_mem_enc_register_region)(kvm, ®ion);
6597 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6598 struct kvm_enc_region region;
6601 if (copy_from_user(®ion, argp, sizeof(region)))
6605 if (!kvm_x86_ops.mem_enc_unregister_region)
6608 r = static_call(kvm_x86_mem_enc_unregister_region)(kvm, ®ion);
6611 case KVM_HYPERV_EVENTFD: {
6612 struct kvm_hyperv_eventfd hvevfd;
6615 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6617 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6620 case KVM_SET_PMU_EVENT_FILTER:
6621 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6623 case KVM_X86_SET_MSR_FILTER:
6624 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6633 static void kvm_init_msr_list(void)
6635 struct x86_pmu_capability x86_pmu;
6639 BUILD_BUG_ON_MSG(KVM_PMC_MAX_FIXED != 3,
6640 "Please update the fixed PMCs in msrs_to_saved_all[]");
6642 perf_get_x86_pmu_capability(&x86_pmu);
6644 num_msrs_to_save = 0;
6645 num_emulated_msrs = 0;
6646 num_msr_based_features = 0;
6648 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6649 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6653 * Even MSRs that are valid in the host may not be exposed
6654 * to the guests in some cases.
6656 switch (msrs_to_save_all[i]) {
6657 case MSR_IA32_BNDCFGS:
6658 if (!kvm_mpx_supported())
6662 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6663 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6666 case MSR_IA32_UMWAIT_CONTROL:
6667 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6670 case MSR_IA32_RTIT_CTL:
6671 case MSR_IA32_RTIT_STATUS:
6672 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6675 case MSR_IA32_RTIT_CR3_MATCH:
6676 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6677 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6680 case MSR_IA32_RTIT_OUTPUT_BASE:
6681 case MSR_IA32_RTIT_OUTPUT_MASK:
6682 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6683 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6684 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6687 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6688 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6689 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6690 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6693 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6694 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6695 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6698 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6699 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6700 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6704 case MSR_IA32_XFD_ERR:
6705 if (!kvm_cpu_cap_has(X86_FEATURE_XFD))
6712 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6715 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6716 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6719 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6722 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6723 struct kvm_msr_entry msr;
6725 msr.index = msr_based_features_all[i];
6726 if (kvm_get_msr_feature(&msr))
6729 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6733 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6741 if (!(lapic_in_kernel(vcpu) &&
6742 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6743 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6754 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6761 if (!(lapic_in_kernel(vcpu) &&
6762 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6764 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6766 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6776 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6777 struct kvm_segment *var, int seg)
6779 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6782 void kvm_get_segment(struct kvm_vcpu *vcpu,
6783 struct kvm_segment *var, int seg)
6785 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6788 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u64 access,
6789 struct x86_exception *exception)
6791 struct kvm_mmu *mmu = vcpu->arch.mmu;
6794 BUG_ON(!mmu_is_nested(vcpu));
6796 /* NPT walks are always user-walks */
6797 access |= PFERR_USER_MASK;
6798 t_gpa = mmu->gva_to_gpa(vcpu, mmu, gpa, access, exception);
6803 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6804 struct x86_exception *exception)
6806 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6808 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6809 return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
6811 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6813 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6814 struct x86_exception *exception)
6816 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6818 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6819 access |= PFERR_FETCH_MASK;
6820 return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
6823 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6824 struct x86_exception *exception)
6826 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6828 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6829 access |= PFERR_WRITE_MASK;
6830 return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
6832 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6834 /* uses this to access any guest's mapped memory without checking CPL */
6835 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6836 struct x86_exception *exception)
6838 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6840 return mmu->gva_to_gpa(vcpu, mmu, gva, 0, exception);
6843 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6844 struct kvm_vcpu *vcpu, u64 access,
6845 struct x86_exception *exception)
6847 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6849 int r = X86EMUL_CONTINUE;
6852 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
6853 unsigned offset = addr & (PAGE_SIZE-1);
6854 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6857 if (gpa == UNMAPPED_GVA)
6858 return X86EMUL_PROPAGATE_FAULT;
6859 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6862 r = X86EMUL_IO_NEEDED;
6874 /* used for instruction fetching */
6875 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6876 gva_t addr, void *val, unsigned int bytes,
6877 struct x86_exception *exception)
6879 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6880 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6881 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6885 /* Inline kvm_read_guest_virt_helper for speed. */
6886 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access|PFERR_FETCH_MASK,
6888 if (unlikely(gpa == UNMAPPED_GVA))
6889 return X86EMUL_PROPAGATE_FAULT;
6891 offset = addr & (PAGE_SIZE-1);
6892 if (WARN_ON(offset + bytes > PAGE_SIZE))
6893 bytes = (unsigned)PAGE_SIZE - offset;
6894 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6896 if (unlikely(ret < 0))
6897 return X86EMUL_IO_NEEDED;
6899 return X86EMUL_CONTINUE;
6902 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6903 gva_t addr, void *val, unsigned int bytes,
6904 struct x86_exception *exception)
6906 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6909 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6910 * is returned, but our callers are not ready for that and they blindly
6911 * call kvm_inject_page_fault. Ensure that they at least do not leak
6912 * uninitialized kernel stack memory into cr2 and error code.
6914 memset(exception, 0, sizeof(*exception));
6915 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6918 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6920 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6921 gva_t addr, void *val, unsigned int bytes,
6922 struct x86_exception *exception, bool system)
6924 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6928 access |= PFERR_IMPLICIT_ACCESS;
6929 else if (static_call(kvm_x86_get_cpl)(vcpu) == 3)
6930 access |= PFERR_USER_MASK;
6932 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6935 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6936 unsigned long addr, void *val, unsigned int bytes)
6938 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6939 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6941 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6944 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6945 struct kvm_vcpu *vcpu, u64 access,
6946 struct x86_exception *exception)
6948 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6950 int r = X86EMUL_CONTINUE;
6953 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
6954 unsigned offset = addr & (PAGE_SIZE-1);
6955 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6958 if (gpa == UNMAPPED_GVA)
6959 return X86EMUL_PROPAGATE_FAULT;
6960 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6962 r = X86EMUL_IO_NEEDED;
6974 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6975 unsigned int bytes, struct x86_exception *exception,
6978 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6979 u64 access = PFERR_WRITE_MASK;
6982 access |= PFERR_IMPLICIT_ACCESS;
6983 else if (static_call(kvm_x86_get_cpl)(vcpu) == 3)
6984 access |= PFERR_USER_MASK;
6986 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6990 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6991 unsigned int bytes, struct x86_exception *exception)
6993 /* kvm_write_guest_virt_system can pull in tons of pages. */
6994 vcpu->arch.l1tf_flush_l1d = true;
6996 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6997 PFERR_WRITE_MASK, exception);
6999 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
7001 static int kvm_can_emulate_insn(struct kvm_vcpu *vcpu, int emul_type,
7002 void *insn, int insn_len)
7004 return static_call(kvm_x86_can_emulate_instruction)(vcpu, emul_type,
7008 int handle_ud(struct kvm_vcpu *vcpu)
7010 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
7011 int emul_type = EMULTYPE_TRAP_UD;
7012 char sig[5]; /* ud2; .ascii "kvm" */
7013 struct x86_exception e;
7015 if (unlikely(!kvm_can_emulate_insn(vcpu, emul_type, NULL, 0)))
7018 if (force_emulation_prefix &&
7019 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
7020 sig, sizeof(sig), &e) == 0 &&
7021 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
7022 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
7023 emul_type = EMULTYPE_TRAP_UD_FORCED;
7026 return kvm_emulate_instruction(vcpu, emul_type);
7028 EXPORT_SYMBOL_GPL(handle_ud);
7030 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
7031 gpa_t gpa, bool write)
7033 /* For APIC access vmexit */
7034 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7037 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
7038 trace_vcpu_match_mmio(gva, gpa, write, true);
7045 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
7046 gpa_t *gpa, struct x86_exception *exception,
7049 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7050 u64 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
7051 | (write ? PFERR_WRITE_MASK : 0);
7054 * currently PKRU is only applied to ept enabled guest so
7055 * there is no pkey in EPT page table for L1 guest or EPT
7056 * shadow page table for L2 guest.
7058 if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
7059 !permission_fault(vcpu, vcpu->arch.walk_mmu,
7060 vcpu->arch.mmio_access, 0, access))) {
7061 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
7062 (gva & (PAGE_SIZE - 1));
7063 trace_vcpu_match_mmio(gva, *gpa, write, false);
7067 *gpa = mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7069 if (*gpa == UNMAPPED_GVA)
7072 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
7075 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
7076 const void *val, int bytes)
7080 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
7083 kvm_page_track_write(vcpu, gpa, val, bytes);
7087 struct read_write_emulator_ops {
7088 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
7090 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
7091 void *val, int bytes);
7092 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
7093 int bytes, void *val);
7094 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
7095 void *val, int bytes);
7099 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
7101 if (vcpu->mmio_read_completed) {
7102 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
7103 vcpu->mmio_fragments[0].gpa, val);
7104 vcpu->mmio_read_completed = 0;
7111 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
7112 void *val, int bytes)
7114 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
7117 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
7118 void *val, int bytes)
7120 return emulator_write_phys(vcpu, gpa, val, bytes);
7123 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
7125 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
7126 return vcpu_mmio_write(vcpu, gpa, bytes, val);
7129 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
7130 void *val, int bytes)
7132 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
7133 return X86EMUL_IO_NEEDED;
7136 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
7137 void *val, int bytes)
7139 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
7141 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
7142 return X86EMUL_CONTINUE;
7145 static const struct read_write_emulator_ops read_emultor = {
7146 .read_write_prepare = read_prepare,
7147 .read_write_emulate = read_emulate,
7148 .read_write_mmio = vcpu_mmio_read,
7149 .read_write_exit_mmio = read_exit_mmio,
7152 static const struct read_write_emulator_ops write_emultor = {
7153 .read_write_emulate = write_emulate,
7154 .read_write_mmio = write_mmio,
7155 .read_write_exit_mmio = write_exit_mmio,
7159 static int emulator_read_write_onepage(unsigned long addr, void *val,
7161 struct x86_exception *exception,
7162 struct kvm_vcpu *vcpu,
7163 const struct read_write_emulator_ops *ops)
7167 bool write = ops->write;
7168 struct kvm_mmio_fragment *frag;
7169 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7172 * If the exit was due to a NPF we may already have a GPA.
7173 * If the GPA is present, use it to avoid the GVA to GPA table walk.
7174 * Note, this cannot be used on string operations since string
7175 * operation using rep will only have the initial GPA from the NPF
7178 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
7179 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
7180 gpa = ctxt->gpa_val;
7181 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
7183 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
7185 return X86EMUL_PROPAGATE_FAULT;
7188 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
7189 return X86EMUL_CONTINUE;
7192 * Is this MMIO handled locally?
7194 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
7195 if (handled == bytes)
7196 return X86EMUL_CONTINUE;
7202 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
7203 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
7207 return X86EMUL_CONTINUE;
7210 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
7212 void *val, unsigned int bytes,
7213 struct x86_exception *exception,
7214 const struct read_write_emulator_ops *ops)
7216 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7220 if (ops->read_write_prepare &&
7221 ops->read_write_prepare(vcpu, val, bytes))
7222 return X86EMUL_CONTINUE;
7224 vcpu->mmio_nr_fragments = 0;
7226 /* Crossing a page boundary? */
7227 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
7230 now = -addr & ~PAGE_MASK;
7231 rc = emulator_read_write_onepage(addr, val, now, exception,
7234 if (rc != X86EMUL_CONTINUE)
7237 if (ctxt->mode != X86EMUL_MODE_PROT64)
7243 rc = emulator_read_write_onepage(addr, val, bytes, exception,
7245 if (rc != X86EMUL_CONTINUE)
7248 if (!vcpu->mmio_nr_fragments)
7251 gpa = vcpu->mmio_fragments[0].gpa;
7253 vcpu->mmio_needed = 1;
7254 vcpu->mmio_cur_fragment = 0;
7256 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
7257 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
7258 vcpu->run->exit_reason = KVM_EXIT_MMIO;
7259 vcpu->run->mmio.phys_addr = gpa;
7261 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
7264 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
7268 struct x86_exception *exception)
7270 return emulator_read_write(ctxt, addr, val, bytes,
7271 exception, &read_emultor);
7274 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
7278 struct x86_exception *exception)
7280 return emulator_read_write(ctxt, addr, (void *)val, bytes,
7281 exception, &write_emultor);
7284 #define emulator_try_cmpxchg_user(t, ptr, old, new) \
7285 (__try_cmpxchg_user((t __user *)(ptr), (t *)(old), *(t *)(new), efault ## t))
7287 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
7292 struct x86_exception *exception)
7294 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7300 /* guests cmpxchg8b have to be emulated atomically */
7301 if (bytes > 8 || (bytes & (bytes - 1)))
7304 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
7306 if (gpa == UNMAPPED_GVA ||
7307 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7311 * Emulate the atomic as a straight write to avoid #AC if SLD is
7312 * enabled in the host and the access splits a cache line.
7314 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
7315 page_line_mask = ~(cache_line_size() - 1);
7317 page_line_mask = PAGE_MASK;
7319 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
7322 hva = kvm_vcpu_gfn_to_hva(vcpu, gpa_to_gfn(gpa));
7323 if (kvm_is_error_hva(hva))
7326 hva += offset_in_page(gpa);
7330 r = emulator_try_cmpxchg_user(u8, hva, old, new);
7333 r = emulator_try_cmpxchg_user(u16, hva, old, new);
7336 r = emulator_try_cmpxchg_user(u32, hva, old, new);
7339 r = emulator_try_cmpxchg_user(u64, hva, old, new);
7346 return X86EMUL_UNHANDLEABLE;
7348 return X86EMUL_CMPXCHG_FAILED;
7350 kvm_page_track_write(vcpu, gpa, new, bytes);
7352 return X86EMUL_CONTINUE;
7355 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
7357 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
7360 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
7364 for (i = 0; i < vcpu->arch.pio.count; i++) {
7365 if (vcpu->arch.pio.in)
7366 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
7367 vcpu->arch.pio.size, pd);
7369 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
7370 vcpu->arch.pio.port, vcpu->arch.pio.size,
7374 pd += vcpu->arch.pio.size;
7379 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
7380 unsigned short port,
7381 unsigned int count, bool in)
7383 vcpu->arch.pio.port = port;
7384 vcpu->arch.pio.in = in;
7385 vcpu->arch.pio.count = count;
7386 vcpu->arch.pio.size = size;
7388 if (!kernel_pio(vcpu, vcpu->arch.pio_data))
7391 vcpu->run->exit_reason = KVM_EXIT_IO;
7392 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
7393 vcpu->run->io.size = size;
7394 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
7395 vcpu->run->io.count = count;
7396 vcpu->run->io.port = port;
7401 static int __emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7402 unsigned short port, unsigned int count)
7404 WARN_ON(vcpu->arch.pio.count);
7405 memset(vcpu->arch.pio_data, 0, size * count);
7406 return emulator_pio_in_out(vcpu, size, port, count, true);
7409 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
7411 int size = vcpu->arch.pio.size;
7412 unsigned count = vcpu->arch.pio.count;
7413 memcpy(val, vcpu->arch.pio_data, size * count);
7414 trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
7415 vcpu->arch.pio.count = 0;
7418 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7419 unsigned short port, void *val, unsigned int count)
7421 if (vcpu->arch.pio.count) {
7423 * Complete a previous iteration that required userspace I/O.
7424 * Note, @count isn't guaranteed to match pio.count as userspace
7425 * can modify ECX before rerunning the vCPU. Ignore any such
7426 * shenanigans as KVM doesn't support modifying the rep count,
7427 * and the emulator ensures @count doesn't overflow the buffer.
7430 int r = __emulator_pio_in(vcpu, size, port, count);
7434 /* Results already available, fall through. */
7437 complete_emulator_pio_in(vcpu, val);
7441 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7442 int size, unsigned short port, void *val,
7445 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
7449 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7450 unsigned short port, const void *val,
7455 memcpy(vcpu->arch.pio_data, val, size * count);
7456 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
7457 ret = emulator_pio_in_out(vcpu, size, port, count, false);
7459 vcpu->arch.pio.count = 0;
7464 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7465 int size, unsigned short port,
7466 const void *val, unsigned int count)
7468 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7471 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7473 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7476 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7478 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7481 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7483 if (!need_emulate_wbinvd(vcpu))
7484 return X86EMUL_CONTINUE;
7486 if (static_call(kvm_x86_has_wbinvd_exit)()) {
7487 int cpu = get_cpu();
7489 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7490 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7491 wbinvd_ipi, NULL, 1);
7493 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7496 return X86EMUL_CONTINUE;
7499 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7501 kvm_emulate_wbinvd_noskip(vcpu);
7502 return kvm_skip_emulated_instruction(vcpu);
7504 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7508 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7510 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7513 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7514 unsigned long *dest)
7516 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7519 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7520 unsigned long value)
7523 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7526 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7528 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7531 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7533 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7534 unsigned long value;
7538 value = kvm_read_cr0(vcpu);
7541 value = vcpu->arch.cr2;
7544 value = kvm_read_cr3(vcpu);
7547 value = kvm_read_cr4(vcpu);
7550 value = kvm_get_cr8(vcpu);
7553 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7560 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7562 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7567 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7570 vcpu->arch.cr2 = val;
7573 res = kvm_set_cr3(vcpu, val);
7576 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7579 res = kvm_set_cr8(vcpu, val);
7582 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7589 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7591 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7594 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7596 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7599 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7601 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7604 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7606 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7609 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7611 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7614 static unsigned long emulator_get_cached_segment_base(
7615 struct x86_emulate_ctxt *ctxt, int seg)
7617 return get_segment_base(emul_to_vcpu(ctxt), seg);
7620 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7621 struct desc_struct *desc, u32 *base3,
7624 struct kvm_segment var;
7626 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7627 *selector = var.selector;
7630 memset(desc, 0, sizeof(*desc));
7638 set_desc_limit(desc, var.limit);
7639 set_desc_base(desc, (unsigned long)var.base);
7640 #ifdef CONFIG_X86_64
7642 *base3 = var.base >> 32;
7644 desc->type = var.type;
7646 desc->dpl = var.dpl;
7647 desc->p = var.present;
7648 desc->avl = var.avl;
7656 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7657 struct desc_struct *desc, u32 base3,
7660 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7661 struct kvm_segment var;
7663 var.selector = selector;
7664 var.base = get_desc_base(desc);
7665 #ifdef CONFIG_X86_64
7666 var.base |= ((u64)base3) << 32;
7668 var.limit = get_desc_limit(desc);
7670 var.limit = (var.limit << 12) | 0xfff;
7671 var.type = desc->type;
7672 var.dpl = desc->dpl;
7677 var.avl = desc->avl;
7678 var.present = desc->p;
7679 var.unusable = !var.present;
7682 kvm_set_segment(vcpu, &var, seg);
7686 static int emulator_get_msr_with_filter(struct x86_emulate_ctxt *ctxt,
7687 u32 msr_index, u64 *pdata)
7689 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7692 r = kvm_get_msr_with_filter(vcpu, msr_index, pdata);
7694 if (r && kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
7695 complete_emulated_rdmsr, r)) {
7696 /* Bounce to user space */
7697 return X86EMUL_IO_NEEDED;
7703 static int emulator_set_msr_with_filter(struct x86_emulate_ctxt *ctxt,
7704 u32 msr_index, u64 data)
7706 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7709 r = kvm_set_msr_with_filter(vcpu, msr_index, data);
7711 if (r && kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_WRMSR, data,
7712 complete_emulated_msr_access, r)) {
7713 /* Bounce to user space */
7714 return X86EMUL_IO_NEEDED;
7720 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7721 u32 msr_index, u64 *pdata)
7723 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
7726 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7727 u32 msr_index, u64 data)
7729 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
7732 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7734 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7736 return vcpu->arch.smbase;
7739 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7741 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7743 vcpu->arch.smbase = smbase;
7746 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7749 if (kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc))
7754 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7755 u32 pmc, u64 *pdata)
7757 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7760 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7762 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7765 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7766 struct x86_instruction_info *info,
7767 enum x86_intercept_stage stage)
7769 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7773 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7774 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7777 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7780 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7782 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7785 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7787 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7790 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7792 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7795 static bool emulator_guest_has_rdpid(struct x86_emulate_ctxt *ctxt)
7797 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_RDPID);
7800 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7802 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7805 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7807 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7810 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7812 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7815 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7817 return emul_to_vcpu(ctxt)->arch.hflags;
7820 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7822 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7824 kvm_smm_changed(vcpu, false);
7827 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7828 const char *smstate)
7830 return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7833 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7835 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7838 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7840 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7843 static const struct x86_emulate_ops emulate_ops = {
7844 .read_gpr = emulator_read_gpr,
7845 .write_gpr = emulator_write_gpr,
7846 .read_std = emulator_read_std,
7847 .write_std = emulator_write_std,
7848 .read_phys = kvm_read_guest_phys_system,
7849 .fetch = kvm_fetch_guest_virt,
7850 .read_emulated = emulator_read_emulated,
7851 .write_emulated = emulator_write_emulated,
7852 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7853 .invlpg = emulator_invlpg,
7854 .pio_in_emulated = emulator_pio_in_emulated,
7855 .pio_out_emulated = emulator_pio_out_emulated,
7856 .get_segment = emulator_get_segment,
7857 .set_segment = emulator_set_segment,
7858 .get_cached_segment_base = emulator_get_cached_segment_base,
7859 .get_gdt = emulator_get_gdt,
7860 .get_idt = emulator_get_idt,
7861 .set_gdt = emulator_set_gdt,
7862 .set_idt = emulator_set_idt,
7863 .get_cr = emulator_get_cr,
7864 .set_cr = emulator_set_cr,
7865 .cpl = emulator_get_cpl,
7866 .get_dr = emulator_get_dr,
7867 .set_dr = emulator_set_dr,
7868 .get_smbase = emulator_get_smbase,
7869 .set_smbase = emulator_set_smbase,
7870 .set_msr_with_filter = emulator_set_msr_with_filter,
7871 .get_msr_with_filter = emulator_get_msr_with_filter,
7872 .set_msr = emulator_set_msr,
7873 .get_msr = emulator_get_msr,
7874 .check_pmc = emulator_check_pmc,
7875 .read_pmc = emulator_read_pmc,
7876 .halt = emulator_halt,
7877 .wbinvd = emulator_wbinvd,
7878 .fix_hypercall = emulator_fix_hypercall,
7879 .intercept = emulator_intercept,
7880 .get_cpuid = emulator_get_cpuid,
7881 .guest_has_long_mode = emulator_guest_has_long_mode,
7882 .guest_has_movbe = emulator_guest_has_movbe,
7883 .guest_has_fxsr = emulator_guest_has_fxsr,
7884 .guest_has_rdpid = emulator_guest_has_rdpid,
7885 .set_nmi_mask = emulator_set_nmi_mask,
7886 .get_hflags = emulator_get_hflags,
7887 .exiting_smm = emulator_exiting_smm,
7888 .leave_smm = emulator_leave_smm,
7889 .triple_fault = emulator_triple_fault,
7890 .set_xcr = emulator_set_xcr,
7893 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7895 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7897 * an sti; sti; sequence only disable interrupts for the first
7898 * instruction. So, if the last instruction, be it emulated or
7899 * not, left the system with the INT_STI flag enabled, it
7900 * means that the last instruction is an sti. We should not
7901 * leave the flag on in this case. The same goes for mov ss
7903 if (int_shadow & mask)
7905 if (unlikely(int_shadow || mask)) {
7906 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7908 kvm_make_request(KVM_REQ_EVENT, vcpu);
7912 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7914 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7915 if (ctxt->exception.vector == PF_VECTOR)
7916 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7918 if (ctxt->exception.error_code_valid)
7919 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7920 ctxt->exception.error_code);
7922 kvm_queue_exception(vcpu, ctxt->exception.vector);
7926 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7928 struct x86_emulate_ctxt *ctxt;
7930 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7932 pr_err("kvm: failed to allocate vcpu's emulator\n");
7937 ctxt->ops = &emulate_ops;
7938 vcpu->arch.emulate_ctxt = ctxt;
7943 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7945 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7948 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7950 ctxt->gpa_available = false;
7951 ctxt->eflags = kvm_get_rflags(vcpu);
7952 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7954 ctxt->eip = kvm_rip_read(vcpu);
7955 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7956 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7957 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7958 cs_db ? X86EMUL_MODE_PROT32 :
7959 X86EMUL_MODE_PROT16;
7960 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7961 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7962 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7964 ctxt->interruptibility = 0;
7965 ctxt->have_exception = false;
7966 ctxt->exception.vector = -1;
7967 ctxt->perm_ok = false;
7969 init_decode_cache(ctxt);
7970 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7973 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7975 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7978 init_emulate_ctxt(vcpu);
7982 ctxt->_eip = ctxt->eip + inc_eip;
7983 ret = emulate_int_real(ctxt, irq);
7985 if (ret != X86EMUL_CONTINUE) {
7986 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7988 ctxt->eip = ctxt->_eip;
7989 kvm_rip_write(vcpu, ctxt->eip);
7990 kvm_set_rflags(vcpu, ctxt->eflags);
7993 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7995 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7996 u8 ndata, u8 *insn_bytes, u8 insn_size)
7998 struct kvm_run *run = vcpu->run;
8003 * Zero the whole array used to retrieve the exit info, as casting to
8004 * u32 for select entries will leave some chunks uninitialized.
8006 memset(&info, 0, sizeof(info));
8008 static_call(kvm_x86_get_exit_info)(vcpu, (u32 *)&info[0], &info[1],
8009 &info[2], (u32 *)&info[3],
8012 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8013 run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
8016 * There's currently space for 13 entries, but 5 are used for the exit
8017 * reason and info. Restrict to 4 to reduce the maintenance burden
8018 * when expanding kvm_run.emulation_failure in the future.
8020 if (WARN_ON_ONCE(ndata > 4))
8023 /* Always include the flags as a 'data' entry. */
8025 run->emulation_failure.flags = 0;
8028 BUILD_BUG_ON((sizeof(run->emulation_failure.insn_size) +
8029 sizeof(run->emulation_failure.insn_bytes) != 16));
8031 run->emulation_failure.flags |=
8032 KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
8033 run->emulation_failure.insn_size = insn_size;
8034 memset(run->emulation_failure.insn_bytes, 0x90,
8035 sizeof(run->emulation_failure.insn_bytes));
8036 memcpy(run->emulation_failure.insn_bytes, insn_bytes, insn_size);
8039 memcpy(&run->internal.data[info_start], info, sizeof(info));
8040 memcpy(&run->internal.data[info_start + ARRAY_SIZE(info)], data,
8041 ndata * sizeof(data[0]));
8043 run->emulation_failure.ndata = info_start + ARRAY_SIZE(info) + ndata;
8046 static void prepare_emulation_ctxt_failure_exit(struct kvm_vcpu *vcpu)
8048 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8050 prepare_emulation_failure_exit(vcpu, NULL, 0, ctxt->fetch.data,
8051 ctxt->fetch.end - ctxt->fetch.data);
8054 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
8057 prepare_emulation_failure_exit(vcpu, data, ndata, NULL, 0);
8059 EXPORT_SYMBOL_GPL(__kvm_prepare_emulation_failure_exit);
8061 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
8063 __kvm_prepare_emulation_failure_exit(vcpu, NULL, 0);
8065 EXPORT_SYMBOL_GPL(kvm_prepare_emulation_failure_exit);
8067 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
8069 struct kvm *kvm = vcpu->kvm;
8071 ++vcpu->stat.insn_emulation_fail;
8072 trace_kvm_emulate_insn_failed(vcpu);
8074 if (emulation_type & EMULTYPE_VMWARE_GP) {
8075 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8079 if (kvm->arch.exit_on_emulation_error ||
8080 (emulation_type & EMULTYPE_SKIP)) {
8081 prepare_emulation_ctxt_failure_exit(vcpu);
8085 kvm_queue_exception(vcpu, UD_VECTOR);
8087 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
8088 prepare_emulation_ctxt_failure_exit(vcpu);
8095 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8096 bool write_fault_to_shadow_pgtable,
8099 gpa_t gpa = cr2_or_gpa;
8102 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
8105 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
8106 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
8109 if (!vcpu->arch.mmu->root_role.direct) {
8111 * Write permission should be allowed since only
8112 * write access need to be emulated.
8114 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
8117 * If the mapping is invalid in guest, let cpu retry
8118 * it to generate fault.
8120 if (gpa == UNMAPPED_GVA)
8125 * Do not retry the unhandleable instruction if it faults on the
8126 * readonly host memory, otherwise it will goto a infinite loop:
8127 * retry instruction -> write #PF -> emulation fail -> retry
8128 * instruction -> ...
8130 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
8133 * If the instruction failed on the error pfn, it can not be fixed,
8134 * report the error to userspace.
8136 if (is_error_noslot_pfn(pfn))
8139 kvm_release_pfn_clean(pfn);
8141 /* The instructions are well-emulated on direct mmu. */
8142 if (vcpu->arch.mmu->root_role.direct) {
8143 unsigned int indirect_shadow_pages;
8145 write_lock(&vcpu->kvm->mmu_lock);
8146 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
8147 write_unlock(&vcpu->kvm->mmu_lock);
8149 if (indirect_shadow_pages)
8150 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8156 * if emulation was due to access to shadowed page table
8157 * and it failed try to unshadow page and re-enter the
8158 * guest to let CPU execute the instruction.
8160 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8163 * If the access faults on its page table, it can not
8164 * be fixed by unprotecting shadow page and it should
8165 * be reported to userspace.
8167 return !write_fault_to_shadow_pgtable;
8170 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
8171 gpa_t cr2_or_gpa, int emulation_type)
8173 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8174 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
8176 last_retry_eip = vcpu->arch.last_retry_eip;
8177 last_retry_addr = vcpu->arch.last_retry_addr;
8180 * If the emulation is caused by #PF and it is non-page_table
8181 * writing instruction, it means the VM-EXIT is caused by shadow
8182 * page protected, we can zap the shadow page and retry this
8183 * instruction directly.
8185 * Note: if the guest uses a non-page-table modifying instruction
8186 * on the PDE that points to the instruction, then we will unmap
8187 * the instruction and go to an infinite loop. So, we cache the
8188 * last retried eip and the last fault address, if we meet the eip
8189 * and the address again, we can break out of the potential infinite
8192 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
8194 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
8197 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
8198 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
8201 if (x86_page_table_writing_insn(ctxt))
8204 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
8207 vcpu->arch.last_retry_eip = ctxt->eip;
8208 vcpu->arch.last_retry_addr = cr2_or_gpa;
8210 if (!vcpu->arch.mmu->root_role.direct)
8211 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
8213 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8218 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
8219 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
8221 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
8223 trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
8226 vcpu->arch.hflags |= HF_SMM_MASK;
8228 vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
8230 /* Process a latched INIT or SMI, if any. */
8231 kvm_make_request(KVM_REQ_EVENT, vcpu);
8234 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
8235 * on SMM exit we still need to reload them from
8238 vcpu->arch.pdptrs_from_userspace = false;
8241 kvm_mmu_reset_context(vcpu);
8244 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
8253 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
8254 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
8259 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
8261 struct kvm_run *kvm_run = vcpu->run;
8263 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
8264 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
8265 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
8266 kvm_run->debug.arch.exception = DB_VECTOR;
8267 kvm_run->exit_reason = KVM_EXIT_DEBUG;
8270 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
8274 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
8276 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8279 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
8283 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8286 * rflags is the old, "raw" value of the flags. The new value has
8287 * not been saved yet.
8289 * This is correct even for TF set by the guest, because "the
8290 * processor will not generate this exception after the instruction
8291 * that sets the TF flag".
8293 if (unlikely(rflags & X86_EFLAGS_TF))
8294 r = kvm_vcpu_do_singlestep(vcpu);
8297 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
8299 static bool kvm_vcpu_check_code_breakpoint(struct kvm_vcpu *vcpu, int *r)
8301 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
8302 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
8303 struct kvm_run *kvm_run = vcpu->run;
8304 unsigned long eip = kvm_get_linear_rip(vcpu);
8305 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8306 vcpu->arch.guest_debug_dr7,
8310 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
8311 kvm_run->debug.arch.pc = eip;
8312 kvm_run->debug.arch.exception = DB_VECTOR;
8313 kvm_run->exit_reason = KVM_EXIT_DEBUG;
8319 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
8320 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
8321 unsigned long eip = kvm_get_linear_rip(vcpu);
8322 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8327 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
8336 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
8338 switch (ctxt->opcode_len) {
8345 case 0xe6: /* OUT */
8349 case 0x6c: /* INS */
8351 case 0x6e: /* OUTS */
8358 case 0x33: /* RDPMC */
8368 * Decode an instruction for emulation. The caller is responsible for handling
8369 * code breakpoints. Note, manually detecting code breakpoints is unnecessary
8370 * (and wrong) when emulating on an intercepted fault-like exception[*], as
8371 * code breakpoints have higher priority and thus have already been done by
8374 * [*] Except #MC, which is higher priority, but KVM should never emulate in
8375 * response to a machine check.
8377 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
8378 void *insn, int insn_len)
8380 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8383 init_emulate_ctxt(vcpu);
8385 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
8387 trace_kvm_emulate_insn_start(vcpu);
8388 ++vcpu->stat.insn_emulation;
8392 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
8394 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8395 int emulation_type, void *insn, int insn_len)
8398 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8399 bool writeback = true;
8400 bool write_fault_to_spt;
8402 if (unlikely(!kvm_can_emulate_insn(vcpu, emulation_type, insn, insn_len)))
8405 vcpu->arch.l1tf_flush_l1d = true;
8408 * Clear write_fault_to_shadow_pgtable here to ensure it is
8411 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
8412 vcpu->arch.write_fault_to_shadow_pgtable = false;
8414 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8415 kvm_clear_exception_queue(vcpu);
8418 * Return immediately if RIP hits a code breakpoint, such #DBs
8419 * are fault-like and are higher priority than any faults on
8420 * the code fetch itself.
8422 if (!(emulation_type & EMULTYPE_SKIP) &&
8423 kvm_vcpu_check_code_breakpoint(vcpu, &r))
8426 r = x86_decode_emulated_instruction(vcpu, emulation_type,
8428 if (r != EMULATION_OK) {
8429 if ((emulation_type & EMULTYPE_TRAP_UD) ||
8430 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
8431 kvm_queue_exception(vcpu, UD_VECTOR);
8434 if (reexecute_instruction(vcpu, cr2_or_gpa,
8438 if (ctxt->have_exception) {
8440 * #UD should result in just EMULATION_FAILED, and trap-like
8441 * exception should not be encountered during decode.
8443 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8444 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8445 inject_emulated_exception(vcpu);
8448 return handle_emulation_failure(vcpu, emulation_type);
8452 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8453 !is_vmware_backdoor_opcode(ctxt)) {
8454 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8459 * EMULTYPE_SKIP without EMULTYPE_COMPLETE_USER_EXIT is intended for
8460 * use *only* by vendor callbacks for kvm_skip_emulated_instruction().
8461 * The caller is responsible for updating interruptibility state and
8462 * injecting single-step #DBs.
8464 if (emulation_type & EMULTYPE_SKIP) {
8465 if (ctxt->mode != X86EMUL_MODE_PROT64)
8466 ctxt->eip = (u32)ctxt->_eip;
8468 ctxt->eip = ctxt->_eip;
8470 if (emulation_type & EMULTYPE_COMPLETE_USER_EXIT) {
8475 kvm_rip_write(vcpu, ctxt->eip);
8476 if (ctxt->eflags & X86_EFLAGS_RF)
8477 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8481 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8484 /* this is needed for vmware backdoor interface to work since it
8485 changes registers values during IO operation */
8486 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8487 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8488 emulator_invalidate_register_cache(ctxt);
8492 if (emulation_type & EMULTYPE_PF) {
8493 /* Save the faulting GPA (cr2) in the address field */
8494 ctxt->exception.address = cr2_or_gpa;
8496 /* With shadow page tables, cr2 contains a GVA or nGPA. */
8497 if (vcpu->arch.mmu->root_role.direct) {
8498 ctxt->gpa_available = true;
8499 ctxt->gpa_val = cr2_or_gpa;
8502 /* Sanitize the address out of an abundance of paranoia. */
8503 ctxt->exception.address = 0;
8506 r = x86_emulate_insn(ctxt);
8508 if (r == EMULATION_INTERCEPTED)
8511 if (r == EMULATION_FAILED) {
8512 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
8516 return handle_emulation_failure(vcpu, emulation_type);
8519 if (ctxt->have_exception) {
8521 if (inject_emulated_exception(vcpu))
8523 } else if (vcpu->arch.pio.count) {
8524 if (!vcpu->arch.pio.in) {
8525 /* FIXME: return into emulator if single-stepping. */
8526 vcpu->arch.pio.count = 0;
8529 vcpu->arch.complete_userspace_io = complete_emulated_pio;
8532 } else if (vcpu->mmio_needed) {
8533 ++vcpu->stat.mmio_exits;
8535 if (!vcpu->mmio_is_write)
8538 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8539 } else if (vcpu->arch.complete_userspace_io) {
8542 } else if (r == EMULATION_RESTART)
8549 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8550 toggle_interruptibility(vcpu, ctxt->interruptibility);
8551 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8552 if (!ctxt->have_exception ||
8553 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
8554 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8555 if (ctxt->is_branch)
8556 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
8557 kvm_rip_write(vcpu, ctxt->eip);
8558 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
8559 r = kvm_vcpu_do_singlestep(vcpu);
8560 static_call_cond(kvm_x86_update_emulated_instruction)(vcpu);
8561 __kvm_set_rflags(vcpu, ctxt->eflags);
8565 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
8566 * do nothing, and it will be requested again as soon as
8567 * the shadow expires. But we still need to check here,
8568 * because POPF has no interrupt shadow.
8570 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
8571 kvm_make_request(KVM_REQ_EVENT, vcpu);
8573 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
8578 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
8580 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
8582 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
8584 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
8585 void *insn, int insn_len)
8587 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
8589 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8591 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8593 vcpu->arch.pio.count = 0;
8597 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8599 vcpu->arch.pio.count = 0;
8601 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8604 return kvm_skip_emulated_instruction(vcpu);
8607 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8608 unsigned short port)
8610 unsigned long val = kvm_rax_read(vcpu);
8611 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8617 * Workaround userspace that relies on old KVM behavior of %rip being
8618 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8621 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8622 vcpu->arch.complete_userspace_io =
8623 complete_fast_pio_out_port_0x7e;
8624 kvm_skip_emulated_instruction(vcpu);
8626 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8627 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8632 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8636 /* We should only ever be called with arch.pio.count equal to 1 */
8637 BUG_ON(vcpu->arch.pio.count != 1);
8639 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8640 vcpu->arch.pio.count = 0;
8644 /* For size less than 4 we merge, else we zero extend */
8645 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8648 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8649 * the copy and tracing
8651 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8652 kvm_rax_write(vcpu, val);
8654 return kvm_skip_emulated_instruction(vcpu);
8657 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8658 unsigned short port)
8663 /* For size less than 4 we merge, else we zero extend */
8664 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8666 ret = emulator_pio_in(vcpu, size, port, &val, 1);
8668 kvm_rax_write(vcpu, val);
8672 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8673 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8678 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8683 ret = kvm_fast_pio_in(vcpu, size, port);
8685 ret = kvm_fast_pio_out(vcpu, size, port);
8686 return ret && kvm_skip_emulated_instruction(vcpu);
8688 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8690 static int kvmclock_cpu_down_prep(unsigned int cpu)
8692 __this_cpu_write(cpu_tsc_khz, 0);
8696 static void tsc_khz_changed(void *data)
8698 struct cpufreq_freqs *freq = data;
8699 unsigned long khz = 0;
8703 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8704 khz = cpufreq_quick_get(raw_smp_processor_id());
8707 __this_cpu_write(cpu_tsc_khz, khz);
8710 #ifdef CONFIG_X86_64
8711 static void kvm_hyperv_tsc_notifier(void)
8716 mutex_lock(&kvm_lock);
8717 list_for_each_entry(kvm, &vm_list, vm_list)
8718 kvm_make_mclock_inprogress_request(kvm);
8720 /* no guest entries from this point */
8721 hyperv_stop_tsc_emulation();
8723 /* TSC frequency always matches when on Hyper-V */
8724 for_each_present_cpu(cpu)
8725 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8726 kvm_max_guest_tsc_khz = tsc_khz;
8728 list_for_each_entry(kvm, &vm_list, vm_list) {
8729 __kvm_start_pvclock_update(kvm);
8730 pvclock_update_vm_gtod_copy(kvm);
8731 kvm_end_pvclock_update(kvm);
8734 mutex_unlock(&kvm_lock);
8738 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8741 struct kvm_vcpu *vcpu;
8746 * We allow guests to temporarily run on slowing clocks,
8747 * provided we notify them after, or to run on accelerating
8748 * clocks, provided we notify them before. Thus time never
8751 * However, we have a problem. We can't atomically update
8752 * the frequency of a given CPU from this function; it is
8753 * merely a notifier, which can be called from any CPU.
8754 * Changing the TSC frequency at arbitrary points in time
8755 * requires a recomputation of local variables related to
8756 * the TSC for each VCPU. We must flag these local variables
8757 * to be updated and be sure the update takes place with the
8758 * new frequency before any guests proceed.
8760 * Unfortunately, the combination of hotplug CPU and frequency
8761 * change creates an intractable locking scenario; the order
8762 * of when these callouts happen is undefined with respect to
8763 * CPU hotplug, and they can race with each other. As such,
8764 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8765 * undefined; you can actually have a CPU frequency change take
8766 * place in between the computation of X and the setting of the
8767 * variable. To protect against this problem, all updates of
8768 * the per_cpu tsc_khz variable are done in an interrupt
8769 * protected IPI, and all callers wishing to update the value
8770 * must wait for a synchronous IPI to complete (which is trivial
8771 * if the caller is on the CPU already). This establishes the
8772 * necessary total order on variable updates.
8774 * Note that because a guest time update may take place
8775 * anytime after the setting of the VCPU's request bit, the
8776 * correct TSC value must be set before the request. However,
8777 * to ensure the update actually makes it to any guest which
8778 * starts running in hardware virtualization between the set
8779 * and the acquisition of the spinlock, we must also ping the
8780 * CPU after setting the request bit.
8784 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8786 mutex_lock(&kvm_lock);
8787 list_for_each_entry(kvm, &vm_list, vm_list) {
8788 kvm_for_each_vcpu(i, vcpu, kvm) {
8789 if (vcpu->cpu != cpu)
8791 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8792 if (vcpu->cpu != raw_smp_processor_id())
8796 mutex_unlock(&kvm_lock);
8798 if (freq->old < freq->new && send_ipi) {
8800 * We upscale the frequency. Must make the guest
8801 * doesn't see old kvmclock values while running with
8802 * the new frequency, otherwise we risk the guest sees
8803 * time go backwards.
8805 * In case we update the frequency for another cpu
8806 * (which might be in guest context) send an interrupt
8807 * to kick the cpu out of guest context. Next time
8808 * guest context is entered kvmclock will be updated,
8809 * so the guest will not see stale values.
8811 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8815 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8818 struct cpufreq_freqs *freq = data;
8821 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8823 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8826 for_each_cpu(cpu, freq->policy->cpus)
8827 __kvmclock_cpufreq_notifier(freq, cpu);
8832 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8833 .notifier_call = kvmclock_cpufreq_notifier
8836 static int kvmclock_cpu_online(unsigned int cpu)
8838 tsc_khz_changed(NULL);
8842 static void kvm_timer_init(void)
8844 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8845 max_tsc_khz = tsc_khz;
8847 if (IS_ENABLED(CONFIG_CPU_FREQ)) {
8848 struct cpufreq_policy *policy;
8852 policy = cpufreq_cpu_get(cpu);
8854 if (policy->cpuinfo.max_freq)
8855 max_tsc_khz = policy->cpuinfo.max_freq;
8856 cpufreq_cpu_put(policy);
8860 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8861 CPUFREQ_TRANSITION_NOTIFIER);
8864 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8865 kvmclock_cpu_online, kvmclock_cpu_down_prep);
8868 #ifdef CONFIG_X86_64
8869 static void pvclock_gtod_update_fn(struct work_struct *work)
8872 struct kvm_vcpu *vcpu;
8875 mutex_lock(&kvm_lock);
8876 list_for_each_entry(kvm, &vm_list, vm_list)
8877 kvm_for_each_vcpu(i, vcpu, kvm)
8878 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8879 atomic_set(&kvm_guest_has_master_clock, 0);
8880 mutex_unlock(&kvm_lock);
8883 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8886 * Indirection to move queue_work() out of the tk_core.seq write held
8887 * region to prevent possible deadlocks against time accessors which
8888 * are invoked with work related locks held.
8890 static void pvclock_irq_work_fn(struct irq_work *w)
8892 queue_work(system_long_wq, &pvclock_gtod_work);
8895 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8898 * Notification about pvclock gtod data update.
8900 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8903 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8904 struct timekeeper *tk = priv;
8906 update_pvclock_gtod(tk);
8909 * Disable master clock if host does not trust, or does not use,
8910 * TSC based clocksource. Delegate queue_work() to irq_work as
8911 * this is invoked with tk_core.seq write held.
8913 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8914 atomic_read(&kvm_guest_has_master_clock) != 0)
8915 irq_work_queue(&pvclock_irq_work);
8919 static struct notifier_block pvclock_gtod_notifier = {
8920 .notifier_call = pvclock_gtod_notify,
8924 int kvm_arch_init(void *opaque)
8926 struct kvm_x86_init_ops *ops = opaque;
8929 if (kvm_x86_ops.hardware_enable) {
8930 pr_err("kvm: already loaded vendor module '%s'\n", kvm_x86_ops.name);
8935 if (!ops->cpu_has_kvm_support()) {
8936 pr_err_ratelimited("kvm: no hardware support for '%s'\n",
8937 ops->runtime_ops->name);
8941 if (ops->disabled_by_bios()) {
8942 pr_err_ratelimited("kvm: support for '%s' disabled by bios\n",
8943 ops->runtime_ops->name);
8949 * KVM explicitly assumes that the guest has an FPU and
8950 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8951 * vCPU's FPU state as a fxregs_state struct.
8953 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8954 printk(KERN_ERR "kvm: inadequate fpu\n");
8959 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8960 pr_err("RT requires X86_FEATURE_CONSTANT_TSC\n");
8967 x86_emulator_cache = kvm_alloc_emulator_cache();
8968 if (!x86_emulator_cache) {
8969 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8973 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8974 if (!user_return_msrs) {
8975 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8976 goto out_free_x86_emulator_cache;
8978 kvm_nr_uret_msrs = 0;
8980 r = kvm_mmu_vendor_module_init();
8982 goto out_free_percpu;
8986 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8987 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8988 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8991 if (pi_inject_timer == -1)
8992 pi_inject_timer = housekeeping_enabled(HK_TYPE_TIMER);
8993 #ifdef CONFIG_X86_64
8994 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8996 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8997 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
9003 free_percpu(user_return_msrs);
9004 out_free_x86_emulator_cache:
9005 kmem_cache_destroy(x86_emulator_cache);
9010 void kvm_arch_exit(void)
9012 #ifdef CONFIG_X86_64
9013 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
9014 clear_hv_tscchange_cb();
9018 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
9019 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
9020 CPUFREQ_TRANSITION_NOTIFIER);
9021 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
9022 #ifdef CONFIG_X86_64
9023 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
9024 irq_work_sync(&pvclock_irq_work);
9025 cancel_work_sync(&pvclock_gtod_work);
9027 kvm_x86_ops.hardware_enable = NULL;
9028 kvm_mmu_vendor_module_exit();
9029 free_percpu(user_return_msrs);
9030 kmem_cache_destroy(x86_emulator_cache);
9031 #ifdef CONFIG_KVM_XEN
9032 static_key_deferred_flush(&kvm_xen_enabled);
9033 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
9037 static int __kvm_emulate_halt(struct kvm_vcpu *vcpu, int state, int reason)
9040 * The vCPU has halted, e.g. executed HLT. Update the run state if the
9041 * local APIC is in-kernel, the run loop will detect the non-runnable
9042 * state and halt the vCPU. Exit to userspace if the local APIC is
9043 * managed by userspace, in which case userspace is responsible for
9044 * handling wake events.
9046 ++vcpu->stat.halt_exits;
9047 if (lapic_in_kernel(vcpu)) {
9048 vcpu->arch.mp_state = state;
9051 vcpu->run->exit_reason = reason;
9056 int kvm_emulate_halt_noskip(struct kvm_vcpu *vcpu)
9058 return __kvm_emulate_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
9060 EXPORT_SYMBOL_GPL(kvm_emulate_halt_noskip);
9062 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
9064 int ret = kvm_skip_emulated_instruction(vcpu);
9066 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
9067 * KVM_EXIT_DEBUG here.
9069 return kvm_emulate_halt_noskip(vcpu) && ret;
9071 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
9073 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
9075 int ret = kvm_skip_emulated_instruction(vcpu);
9077 return __kvm_emulate_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD,
9078 KVM_EXIT_AP_RESET_HOLD) && ret;
9080 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
9082 #ifdef CONFIG_X86_64
9083 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
9084 unsigned long clock_type)
9086 struct kvm_clock_pairing clock_pairing;
9087 struct timespec64 ts;
9091 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
9092 return -KVM_EOPNOTSUPP;
9095 * When tsc is in permanent catchup mode guests won't be able to use
9096 * pvclock_read_retry loop to get consistent view of pvclock
9098 if (vcpu->arch.tsc_always_catchup)
9099 return -KVM_EOPNOTSUPP;
9101 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
9102 return -KVM_EOPNOTSUPP;
9104 clock_pairing.sec = ts.tv_sec;
9105 clock_pairing.nsec = ts.tv_nsec;
9106 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
9107 clock_pairing.flags = 0;
9108 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
9111 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
9112 sizeof(struct kvm_clock_pairing)))
9120 * kvm_pv_kick_cpu_op: Kick a vcpu.
9122 * @apicid - apicid of vcpu to be kicked.
9124 static void kvm_pv_kick_cpu_op(struct kvm *kvm, int apicid)
9126 struct kvm_lapic_irq lapic_irq;
9128 lapic_irq.shorthand = APIC_DEST_NOSHORT;
9129 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
9130 lapic_irq.level = 0;
9131 lapic_irq.dest_id = apicid;
9132 lapic_irq.msi_redir_hint = false;
9134 lapic_irq.delivery_mode = APIC_DM_REMRD;
9135 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
9138 bool kvm_apicv_activated(struct kvm *kvm)
9140 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
9142 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
9144 bool kvm_vcpu_apicv_activated(struct kvm_vcpu *vcpu)
9146 ulong vm_reasons = READ_ONCE(vcpu->kvm->arch.apicv_inhibit_reasons);
9147 ulong vcpu_reasons = static_call(kvm_x86_vcpu_get_apicv_inhibit_reasons)(vcpu);
9149 return (vm_reasons | vcpu_reasons) == 0;
9151 EXPORT_SYMBOL_GPL(kvm_vcpu_apicv_activated);
9153 static void set_or_clear_apicv_inhibit(unsigned long *inhibits,
9154 enum kvm_apicv_inhibit reason, bool set)
9157 __set_bit(reason, inhibits);
9159 __clear_bit(reason, inhibits);
9161 trace_kvm_apicv_inhibit_changed(reason, set, *inhibits);
9164 static void kvm_apicv_init(struct kvm *kvm)
9166 unsigned long *inhibits = &kvm->arch.apicv_inhibit_reasons;
9168 init_rwsem(&kvm->arch.apicv_update_lock);
9170 set_or_clear_apicv_inhibit(inhibits, APICV_INHIBIT_REASON_ABSENT, true);
9173 set_or_clear_apicv_inhibit(inhibits,
9174 APICV_INHIBIT_REASON_DISABLE, true);
9177 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
9179 struct kvm_vcpu *target = NULL;
9180 struct kvm_apic_map *map;
9182 vcpu->stat.directed_yield_attempted++;
9184 if (single_task_running())
9188 map = rcu_dereference(vcpu->kvm->arch.apic_map);
9190 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
9191 target = map->phys_map[dest_id]->vcpu;
9195 if (!target || !READ_ONCE(target->ready))
9198 /* Ignore requests to yield to self */
9202 if (kvm_vcpu_yield_to(target) <= 0)
9205 vcpu->stat.directed_yield_successful++;
9211 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
9213 u64 ret = vcpu->run->hypercall.ret;
9215 if (!is_64_bit_mode(vcpu))
9217 kvm_rax_write(vcpu, ret);
9218 ++vcpu->stat.hypercalls;
9219 return kvm_skip_emulated_instruction(vcpu);
9222 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
9224 unsigned long nr, a0, a1, a2, a3, ret;
9227 if (kvm_xen_hypercall_enabled(vcpu->kvm))
9228 return kvm_xen_hypercall(vcpu);
9230 if (kvm_hv_hypercall_enabled(vcpu))
9231 return kvm_hv_hypercall(vcpu);
9233 nr = kvm_rax_read(vcpu);
9234 a0 = kvm_rbx_read(vcpu);
9235 a1 = kvm_rcx_read(vcpu);
9236 a2 = kvm_rdx_read(vcpu);
9237 a3 = kvm_rsi_read(vcpu);
9239 trace_kvm_hypercall(nr, a0, a1, a2, a3);
9241 op_64_bit = is_64_bit_hypercall(vcpu);
9250 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
9258 case KVM_HC_VAPIC_POLL_IRQ:
9261 case KVM_HC_KICK_CPU:
9262 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
9265 kvm_pv_kick_cpu_op(vcpu->kvm, a1);
9266 kvm_sched_yield(vcpu, a1);
9269 #ifdef CONFIG_X86_64
9270 case KVM_HC_CLOCK_PAIRING:
9271 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
9274 case KVM_HC_SEND_IPI:
9275 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
9278 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
9280 case KVM_HC_SCHED_YIELD:
9281 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
9284 kvm_sched_yield(vcpu, a0);
9287 case KVM_HC_MAP_GPA_RANGE: {
9288 u64 gpa = a0, npages = a1, attrs = a2;
9291 if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
9294 if (!PAGE_ALIGNED(gpa) || !npages ||
9295 gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
9300 vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
9301 vcpu->run->hypercall.nr = KVM_HC_MAP_GPA_RANGE;
9302 vcpu->run->hypercall.args[0] = gpa;
9303 vcpu->run->hypercall.args[1] = npages;
9304 vcpu->run->hypercall.args[2] = attrs;
9305 vcpu->run->hypercall.longmode = op_64_bit;
9306 vcpu->arch.complete_userspace_io = complete_hypercall_exit;
9316 kvm_rax_write(vcpu, ret);
9318 ++vcpu->stat.hypercalls;
9319 return kvm_skip_emulated_instruction(vcpu);
9321 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
9323 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
9325 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
9326 char instruction[3];
9327 unsigned long rip = kvm_rip_read(vcpu);
9330 * If the quirk is disabled, synthesize a #UD and let the guest pick up
9333 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_FIX_HYPERCALL_INSN)) {
9334 ctxt->exception.error_code_valid = false;
9335 ctxt->exception.vector = UD_VECTOR;
9336 ctxt->have_exception = true;
9337 return X86EMUL_PROPAGATE_FAULT;
9340 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
9342 return emulator_write_emulated(ctxt, rip, instruction, 3,
9346 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
9348 return vcpu->run->request_interrupt_window &&
9349 likely(!pic_in_kernel(vcpu->kvm));
9352 /* Called within kvm->srcu read side. */
9353 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
9355 struct kvm_run *kvm_run = vcpu->run;
9357 kvm_run->if_flag = static_call(kvm_x86_get_if_flag)(vcpu);
9358 kvm_run->cr8 = kvm_get_cr8(vcpu);
9359 kvm_run->apic_base = kvm_get_apic_base(vcpu);
9361 kvm_run->ready_for_interrupt_injection =
9362 pic_in_kernel(vcpu->kvm) ||
9363 kvm_vcpu_ready_for_interrupt_injection(vcpu);
9366 kvm_run->flags |= KVM_RUN_X86_SMM;
9369 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
9373 if (!kvm_x86_ops.update_cr8_intercept)
9376 if (!lapic_in_kernel(vcpu))
9379 if (vcpu->arch.apicv_active)
9382 if (!vcpu->arch.apic->vapic_addr)
9383 max_irr = kvm_lapic_find_highest_irr(vcpu);
9390 tpr = kvm_lapic_get_cr8(vcpu);
9392 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
9396 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
9398 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9399 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9403 return kvm_x86_ops.nested_ops->check_events(vcpu);
9406 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
9408 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
9409 vcpu->arch.exception.error_code = false;
9410 static_call(kvm_x86_queue_exception)(vcpu);
9413 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
9416 bool can_inject = true;
9418 /* try to reinject previous events if any */
9420 if (vcpu->arch.exception.injected) {
9421 kvm_inject_exception(vcpu);
9425 * Do not inject an NMI or interrupt if there is a pending
9426 * exception. Exceptions and interrupts are recognized at
9427 * instruction boundaries, i.e. the start of an instruction.
9428 * Trap-like exceptions, e.g. #DB, have higher priority than
9429 * NMIs and interrupts, i.e. traps are recognized before an
9430 * NMI/interrupt that's pending on the same instruction.
9431 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
9432 * priority, but are only generated (pended) during instruction
9433 * execution, i.e. a pending fault-like exception means the
9434 * fault occurred on the *previous* instruction and must be
9435 * serviced prior to recognizing any new events in order to
9436 * fully complete the previous instruction.
9438 else if (!vcpu->arch.exception.pending) {
9439 if (vcpu->arch.nmi_injected) {
9440 static_call(kvm_x86_inject_nmi)(vcpu);
9442 } else if (vcpu->arch.interrupt.injected) {
9443 static_call(kvm_x86_inject_irq)(vcpu);
9448 WARN_ON_ONCE(vcpu->arch.exception.injected &&
9449 vcpu->arch.exception.pending);
9452 * Call check_nested_events() even if we reinjected a previous event
9453 * in order for caller to determine if it should require immediate-exit
9454 * from L2 to L1 due to pending L1 events which require exit
9457 if (is_guest_mode(vcpu)) {
9458 r = kvm_check_nested_events(vcpu);
9463 /* try to inject new event if pending */
9464 if (vcpu->arch.exception.pending) {
9465 trace_kvm_inj_exception(vcpu->arch.exception.nr,
9466 vcpu->arch.exception.has_error_code,
9467 vcpu->arch.exception.error_code);
9469 vcpu->arch.exception.pending = false;
9470 vcpu->arch.exception.injected = true;
9472 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
9473 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
9476 if (vcpu->arch.exception.nr == DB_VECTOR) {
9477 kvm_deliver_exception_payload(vcpu);
9478 if (vcpu->arch.dr7 & DR7_GD) {
9479 vcpu->arch.dr7 &= ~DR7_GD;
9480 kvm_update_dr7(vcpu);
9484 kvm_inject_exception(vcpu);
9488 /* Don't inject interrupts if the user asked to avoid doing so */
9489 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
9493 * Finally, inject interrupt events. If an event cannot be injected
9494 * due to architectural conditions (e.g. IF=0) a window-open exit
9495 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
9496 * and can architecturally be injected, but we cannot do it right now:
9497 * an interrupt could have arrived just now and we have to inject it
9498 * as a vmexit, or there could already an event in the queue, which is
9499 * indicated by can_inject. In that case we request an immediate exit
9500 * in order to make progress and get back here for another iteration.
9501 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
9503 if (vcpu->arch.smi_pending) {
9504 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
9508 vcpu->arch.smi_pending = false;
9509 ++vcpu->arch.smi_count;
9513 static_call(kvm_x86_enable_smi_window)(vcpu);
9516 if (vcpu->arch.nmi_pending) {
9517 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
9521 --vcpu->arch.nmi_pending;
9522 vcpu->arch.nmi_injected = true;
9523 static_call(kvm_x86_inject_nmi)(vcpu);
9525 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
9527 if (vcpu->arch.nmi_pending)
9528 static_call(kvm_x86_enable_nmi_window)(vcpu);
9531 if (kvm_cpu_has_injectable_intr(vcpu)) {
9532 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
9536 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
9537 static_call(kvm_x86_inject_irq)(vcpu);
9538 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
9540 if (kvm_cpu_has_injectable_intr(vcpu))
9541 static_call(kvm_x86_enable_irq_window)(vcpu);
9544 if (is_guest_mode(vcpu) &&
9545 kvm_x86_ops.nested_ops->hv_timer_pending &&
9546 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
9547 *req_immediate_exit = true;
9549 WARN_ON(vcpu->arch.exception.pending);
9554 *req_immediate_exit = true;
9560 static void process_nmi(struct kvm_vcpu *vcpu)
9565 * x86 is limited to one NMI running, and one NMI pending after it.
9566 * If an NMI is already in progress, limit further NMIs to just one.
9567 * Otherwise, allow two (and we'll inject the first one immediately).
9569 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
9572 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
9573 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
9574 kvm_make_request(KVM_REQ_EVENT, vcpu);
9577 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
9580 flags |= seg->g << 23;
9581 flags |= seg->db << 22;
9582 flags |= seg->l << 21;
9583 flags |= seg->avl << 20;
9584 flags |= seg->present << 15;
9585 flags |= seg->dpl << 13;
9586 flags |= seg->s << 12;
9587 flags |= seg->type << 8;
9591 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
9593 struct kvm_segment seg;
9596 kvm_get_segment(vcpu, &seg, n);
9597 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9600 offset = 0x7f84 + n * 12;
9602 offset = 0x7f2c + (n - 3) * 12;
9604 put_smstate(u32, buf, offset + 8, seg.base);
9605 put_smstate(u32, buf, offset + 4, seg.limit);
9606 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9609 #ifdef CONFIG_X86_64
9610 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9612 struct kvm_segment seg;
9616 kvm_get_segment(vcpu, &seg, n);
9617 offset = 0x7e00 + n * 16;
9619 flags = enter_smm_get_segment_flags(&seg) >> 8;
9620 put_smstate(u16, buf, offset, seg.selector);
9621 put_smstate(u16, buf, offset + 2, flags);
9622 put_smstate(u32, buf, offset + 4, seg.limit);
9623 put_smstate(u64, buf, offset + 8, seg.base);
9627 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9630 struct kvm_segment seg;
9634 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9635 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9636 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9637 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9639 for (i = 0; i < 8; i++)
9640 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9642 kvm_get_dr(vcpu, 6, &val);
9643 put_smstate(u32, buf, 0x7fcc, (u32)val);
9644 kvm_get_dr(vcpu, 7, &val);
9645 put_smstate(u32, buf, 0x7fc8, (u32)val);
9647 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9648 put_smstate(u32, buf, 0x7fc4, seg.selector);
9649 put_smstate(u32, buf, 0x7f64, seg.base);
9650 put_smstate(u32, buf, 0x7f60, seg.limit);
9651 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9653 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9654 put_smstate(u32, buf, 0x7fc0, seg.selector);
9655 put_smstate(u32, buf, 0x7f80, seg.base);
9656 put_smstate(u32, buf, 0x7f7c, seg.limit);
9657 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9659 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9660 put_smstate(u32, buf, 0x7f74, dt.address);
9661 put_smstate(u32, buf, 0x7f70, dt.size);
9663 static_call(kvm_x86_get_idt)(vcpu, &dt);
9664 put_smstate(u32, buf, 0x7f58, dt.address);
9665 put_smstate(u32, buf, 0x7f54, dt.size);
9667 for (i = 0; i < 6; i++)
9668 enter_smm_save_seg_32(vcpu, buf, i);
9670 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9673 put_smstate(u32, buf, 0x7efc, 0x00020000);
9674 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9677 #ifdef CONFIG_X86_64
9678 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9681 struct kvm_segment seg;
9685 for (i = 0; i < 16; i++)
9686 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9688 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9689 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9691 kvm_get_dr(vcpu, 6, &val);
9692 put_smstate(u64, buf, 0x7f68, val);
9693 kvm_get_dr(vcpu, 7, &val);
9694 put_smstate(u64, buf, 0x7f60, val);
9696 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9697 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9698 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9700 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9703 put_smstate(u32, buf, 0x7efc, 0x00020064);
9705 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9707 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9708 put_smstate(u16, buf, 0x7e90, seg.selector);
9709 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9710 put_smstate(u32, buf, 0x7e94, seg.limit);
9711 put_smstate(u64, buf, 0x7e98, seg.base);
9713 static_call(kvm_x86_get_idt)(vcpu, &dt);
9714 put_smstate(u32, buf, 0x7e84, dt.size);
9715 put_smstate(u64, buf, 0x7e88, dt.address);
9717 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9718 put_smstate(u16, buf, 0x7e70, seg.selector);
9719 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9720 put_smstate(u32, buf, 0x7e74, seg.limit);
9721 put_smstate(u64, buf, 0x7e78, seg.base);
9723 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9724 put_smstate(u32, buf, 0x7e64, dt.size);
9725 put_smstate(u64, buf, 0x7e68, dt.address);
9727 for (i = 0; i < 6; i++)
9728 enter_smm_save_seg_64(vcpu, buf, i);
9732 static void enter_smm(struct kvm_vcpu *vcpu)
9734 struct kvm_segment cs, ds;
9739 memset(buf, 0, 512);
9740 #ifdef CONFIG_X86_64
9741 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9742 enter_smm_save_state_64(vcpu, buf);
9745 enter_smm_save_state_32(vcpu, buf);
9748 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9749 * state (e.g. leave guest mode) after we've saved the state into the
9750 * SMM state-save area.
9752 static_call(kvm_x86_enter_smm)(vcpu, buf);
9754 kvm_smm_changed(vcpu, true);
9755 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9757 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9758 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9760 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9762 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9763 kvm_rip_write(vcpu, 0x8000);
9765 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9766 static_call(kvm_x86_set_cr0)(vcpu, cr0);
9767 vcpu->arch.cr0 = cr0;
9769 static_call(kvm_x86_set_cr4)(vcpu, 0);
9771 /* Undocumented: IDT limit is set to zero on entry to SMM. */
9772 dt.address = dt.size = 0;
9773 static_call(kvm_x86_set_idt)(vcpu, &dt);
9775 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9777 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9778 cs.base = vcpu->arch.smbase;
9783 cs.limit = ds.limit = 0xffffffff;
9784 cs.type = ds.type = 0x3;
9785 cs.dpl = ds.dpl = 0;
9790 cs.avl = ds.avl = 0;
9791 cs.present = ds.present = 1;
9792 cs.unusable = ds.unusable = 0;
9793 cs.padding = ds.padding = 0;
9795 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9796 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9797 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9798 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9799 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9800 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9802 #ifdef CONFIG_X86_64
9803 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9804 static_call(kvm_x86_set_efer)(vcpu, 0);
9807 kvm_update_cpuid_runtime(vcpu);
9808 kvm_mmu_reset_context(vcpu);
9811 static void process_smi(struct kvm_vcpu *vcpu)
9813 vcpu->arch.smi_pending = true;
9814 kvm_make_request(KVM_REQ_EVENT, vcpu);
9817 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9818 unsigned long *vcpu_bitmap)
9820 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
9823 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9825 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9828 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9832 if (!lapic_in_kernel(vcpu))
9835 down_read(&vcpu->kvm->arch.apicv_update_lock);
9837 activate = kvm_vcpu_apicv_activated(vcpu);
9839 if (vcpu->arch.apicv_active == activate)
9842 vcpu->arch.apicv_active = activate;
9843 kvm_apic_update_apicv(vcpu);
9844 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9847 * When APICv gets disabled, we may still have injected interrupts
9848 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9849 * still active when the interrupt got accepted. Make sure
9850 * inject_pending_event() is called to check for that.
9852 if (!vcpu->arch.apicv_active)
9853 kvm_make_request(KVM_REQ_EVENT, vcpu);
9856 up_read(&vcpu->kvm->arch.apicv_update_lock);
9858 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9860 void __kvm_set_or_clear_apicv_inhibit(struct kvm *kvm,
9861 enum kvm_apicv_inhibit reason, bool set)
9863 unsigned long old, new;
9865 lockdep_assert_held_write(&kvm->arch.apicv_update_lock);
9867 if (!static_call(kvm_x86_check_apicv_inhibit_reasons)(reason))
9870 old = new = kvm->arch.apicv_inhibit_reasons;
9872 set_or_clear_apicv_inhibit(&new, reason, set);
9874 if (!!old != !!new) {
9876 * Kick all vCPUs before setting apicv_inhibit_reasons to avoid
9877 * false positives in the sanity check WARN in svm_vcpu_run().
9878 * This task will wait for all vCPUs to ack the kick IRQ before
9879 * updating apicv_inhibit_reasons, and all other vCPUs will
9880 * block on acquiring apicv_update_lock so that vCPUs can't
9881 * redo svm_vcpu_run() without seeing the new inhibit state.
9883 * Note, holding apicv_update_lock and taking it in the read
9884 * side (handling the request) also prevents other vCPUs from
9885 * servicing the request with a stale apicv_inhibit_reasons.
9887 kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9888 kvm->arch.apicv_inhibit_reasons = new;
9890 unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9891 kvm_zap_gfn_range(kvm, gfn, gfn+1);
9894 kvm->arch.apicv_inhibit_reasons = new;
9898 void kvm_set_or_clear_apicv_inhibit(struct kvm *kvm,
9899 enum kvm_apicv_inhibit reason, bool set)
9904 down_write(&kvm->arch.apicv_update_lock);
9905 __kvm_set_or_clear_apicv_inhibit(kvm, reason, set);
9906 up_write(&kvm->arch.apicv_update_lock);
9908 EXPORT_SYMBOL_GPL(kvm_set_or_clear_apicv_inhibit);
9910 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9912 if (!kvm_apic_present(vcpu))
9915 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9917 if (irqchip_split(vcpu->kvm))
9918 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9920 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9921 if (ioapic_in_kernel(vcpu->kvm))
9922 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9925 if (is_guest_mode(vcpu))
9926 vcpu->arch.load_eoi_exitmap_pending = true;
9928 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9931 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9933 u64 eoi_exit_bitmap[4];
9935 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9938 if (to_hv_vcpu(vcpu)) {
9939 bitmap_or((ulong *)eoi_exit_bitmap,
9940 vcpu->arch.ioapic_handled_vectors,
9941 to_hv_synic(vcpu)->vec_bitmap, 256);
9942 static_call_cond(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9946 static_call_cond(kvm_x86_load_eoi_exitmap)(
9947 vcpu, (u64 *)vcpu->arch.ioapic_handled_vectors);
9950 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9951 unsigned long start, unsigned long end)
9953 unsigned long apic_address;
9956 * The physical address of apic access page is stored in the VMCS.
9957 * Update it when it becomes invalid.
9959 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9960 if (start <= apic_address && apic_address < end)
9961 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9964 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
9966 static_call_cond(kvm_x86_guest_memory_reclaimed)(kvm);
9969 static void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9971 if (!lapic_in_kernel(vcpu))
9974 static_call_cond(kvm_x86_set_apic_access_page_addr)(vcpu);
9977 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9979 smp_send_reschedule(vcpu->cpu);
9981 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9984 * Called within kvm->srcu read side.
9985 * Returns 1 to let vcpu_run() continue the guest execution loop without
9986 * exiting to the userspace. Otherwise, the value will be returned to the
9989 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9993 dm_request_for_irq_injection(vcpu) &&
9994 kvm_cpu_accept_dm_intr(vcpu);
9995 fastpath_t exit_fastpath;
9997 bool req_immediate_exit = false;
9999 /* Forbid vmenter if vcpu dirty ring is soft-full */
10000 if (unlikely(vcpu->kvm->dirty_ring_size &&
10001 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
10002 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
10003 trace_kvm_dirty_ring_exit(vcpu);
10008 if (kvm_request_pending(vcpu)) {
10009 if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu)) {
10013 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
10014 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
10019 if (kvm_check_request(KVM_REQ_MMU_FREE_OBSOLETE_ROOTS, vcpu))
10020 kvm_mmu_free_obsolete_roots(vcpu);
10021 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
10022 __kvm_migrate_timers(vcpu);
10023 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
10024 kvm_update_masterclock(vcpu->kvm);
10025 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
10026 kvm_gen_kvmclock_update(vcpu);
10027 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
10028 r = kvm_guest_time_update(vcpu);
10032 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
10033 kvm_mmu_sync_roots(vcpu);
10034 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
10035 kvm_mmu_load_pgd(vcpu);
10036 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
10037 kvm_vcpu_flush_tlb_all(vcpu);
10039 /* Flushing all ASIDs flushes the current ASID... */
10040 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
10042 kvm_service_local_tlb_flush_requests(vcpu);
10044 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
10045 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
10049 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
10050 if (is_guest_mode(vcpu)) {
10051 kvm_x86_ops.nested_ops->triple_fault(vcpu);
10053 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
10054 vcpu->mmio_needed = 0;
10059 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
10060 /* Page is swapped out. Do synthetic halt */
10061 vcpu->arch.apf.halted = true;
10065 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
10066 record_steal_time(vcpu);
10067 if (kvm_check_request(KVM_REQ_SMI, vcpu))
10069 if (kvm_check_request(KVM_REQ_NMI, vcpu))
10071 if (kvm_check_request(KVM_REQ_PMU, vcpu))
10072 kvm_pmu_handle_event(vcpu);
10073 if (kvm_check_request(KVM_REQ_PMI, vcpu))
10074 kvm_pmu_deliver_pmi(vcpu);
10075 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
10076 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
10077 if (test_bit(vcpu->arch.pending_ioapic_eoi,
10078 vcpu->arch.ioapic_handled_vectors)) {
10079 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
10080 vcpu->run->eoi.vector =
10081 vcpu->arch.pending_ioapic_eoi;
10086 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
10087 vcpu_scan_ioapic(vcpu);
10088 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
10089 vcpu_load_eoi_exitmap(vcpu);
10090 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
10091 kvm_vcpu_reload_apic_access_page(vcpu);
10092 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
10093 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
10094 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
10095 vcpu->run->system_event.ndata = 0;
10099 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
10100 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
10101 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
10102 vcpu->run->system_event.ndata = 0;
10106 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
10107 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
10109 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
10110 vcpu->run->hyperv = hv_vcpu->exit;
10116 * KVM_REQ_HV_STIMER has to be processed after
10117 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
10118 * depend on the guest clock being up-to-date
10120 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
10121 kvm_hv_process_stimers(vcpu);
10122 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
10123 kvm_vcpu_update_apicv(vcpu);
10124 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
10125 kvm_check_async_pf_completion(vcpu);
10126 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
10127 static_call(kvm_x86_msr_filter_changed)(vcpu);
10129 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
10130 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
10133 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
10134 kvm_xen_has_interrupt(vcpu)) {
10135 ++vcpu->stat.req_event;
10136 r = kvm_apic_accept_events(vcpu);
10141 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
10146 r = inject_pending_event(vcpu, &req_immediate_exit);
10152 static_call(kvm_x86_enable_irq_window)(vcpu);
10154 if (kvm_lapic_enabled(vcpu)) {
10155 update_cr8_intercept(vcpu);
10156 kvm_lapic_sync_to_vapic(vcpu);
10160 r = kvm_mmu_reload(vcpu);
10162 goto cancel_injection;
10167 static_call(kvm_x86_prepare_switch_to_guest)(vcpu);
10170 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
10171 * IPI are then delayed after guest entry, which ensures that they
10172 * result in virtual interrupt delivery.
10174 local_irq_disable();
10176 /* Store vcpu->apicv_active before vcpu->mode. */
10177 smp_store_release(&vcpu->mode, IN_GUEST_MODE);
10179 kvm_vcpu_srcu_read_unlock(vcpu);
10182 * 1) We should set ->mode before checking ->requests. Please see
10183 * the comment in kvm_vcpu_exiting_guest_mode().
10185 * 2) For APICv, we should set ->mode before checking PID.ON. This
10186 * pairs with the memory barrier implicit in pi_test_and_set_on
10187 * (see vmx_deliver_posted_interrupt).
10189 * 3) This also orders the write to mode from any reads to the page
10190 * tables done while the VCPU is running. Please see the comment
10191 * in kvm_flush_remote_tlbs.
10193 smp_mb__after_srcu_read_unlock();
10196 * Process pending posted interrupts to handle the case where the
10197 * notification IRQ arrived in the host, or was never sent (because the
10198 * target vCPU wasn't running). Do this regardless of the vCPU's APICv
10199 * status, KVM doesn't update assigned devices when APICv is inhibited,
10200 * i.e. they can post interrupts even if APICv is temporarily disabled.
10202 if (kvm_lapic_enabled(vcpu))
10203 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10205 if (kvm_vcpu_exit_request(vcpu)) {
10206 vcpu->mode = OUTSIDE_GUEST_MODE;
10208 local_irq_enable();
10210 kvm_vcpu_srcu_read_lock(vcpu);
10212 goto cancel_injection;
10215 if (req_immediate_exit) {
10216 kvm_make_request(KVM_REQ_EVENT, vcpu);
10217 static_call(kvm_x86_request_immediate_exit)(vcpu);
10220 fpregs_assert_state_consistent();
10221 if (test_thread_flag(TIF_NEED_FPU_LOAD))
10222 switch_fpu_return();
10224 if (vcpu->arch.guest_fpu.xfd_err)
10225 wrmsrl(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
10227 if (unlikely(vcpu->arch.switch_db_regs)) {
10228 set_debugreg(0, 7);
10229 set_debugreg(vcpu->arch.eff_db[0], 0);
10230 set_debugreg(vcpu->arch.eff_db[1], 1);
10231 set_debugreg(vcpu->arch.eff_db[2], 2);
10232 set_debugreg(vcpu->arch.eff_db[3], 3);
10233 } else if (unlikely(hw_breakpoint_active())) {
10234 set_debugreg(0, 7);
10237 guest_timing_enter_irqoff();
10241 * Assert that vCPU vs. VM APICv state is consistent. An APICv
10242 * update must kick and wait for all vCPUs before toggling the
10243 * per-VM state, and responsing vCPUs must wait for the update
10244 * to complete before servicing KVM_REQ_APICV_UPDATE.
10246 WARN_ON_ONCE(kvm_vcpu_apicv_activated(vcpu) != kvm_vcpu_apicv_active(vcpu));
10248 exit_fastpath = static_call(kvm_x86_vcpu_run)(vcpu);
10249 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
10252 if (kvm_lapic_enabled(vcpu))
10253 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10255 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
10256 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
10262 * Do this here before restoring debug registers on the host. And
10263 * since we do this before handling the vmexit, a DR access vmexit
10264 * can (a) read the correct value of the debug registers, (b) set
10265 * KVM_DEBUGREG_WONT_EXIT again.
10267 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
10268 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
10269 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
10270 kvm_update_dr0123(vcpu);
10271 kvm_update_dr7(vcpu);
10275 * If the guest has used debug registers, at least dr7
10276 * will be disabled while returning to the host.
10277 * If we don't have active breakpoints in the host, we don't
10278 * care about the messed up debug address registers. But if
10279 * we have some of them active, restore the old state.
10281 if (hw_breakpoint_active())
10282 hw_breakpoint_restore();
10284 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
10285 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
10287 vcpu->mode = OUTSIDE_GUEST_MODE;
10291 * Sync xfd before calling handle_exit_irqoff() which may
10292 * rely on the fact that guest_fpu::xfd is up-to-date (e.g.
10293 * in #NM irqoff handler).
10295 if (vcpu->arch.xfd_no_write_intercept)
10296 fpu_sync_guest_vmexit_xfd_state();
10298 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
10300 if (vcpu->arch.guest_fpu.xfd_err)
10301 wrmsrl(MSR_IA32_XFD_ERR, 0);
10304 * Consume any pending interrupts, including the possible source of
10305 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
10306 * An instruction is required after local_irq_enable() to fully unblock
10307 * interrupts on processors that implement an interrupt shadow, the
10308 * stat.exits increment will do nicely.
10310 kvm_before_interrupt(vcpu, KVM_HANDLING_IRQ);
10311 local_irq_enable();
10312 ++vcpu->stat.exits;
10313 local_irq_disable();
10314 kvm_after_interrupt(vcpu);
10317 * Wait until after servicing IRQs to account guest time so that any
10318 * ticks that occurred while running the guest are properly accounted
10319 * to the guest. Waiting until IRQs are enabled degrades the accuracy
10320 * of accounting via context tracking, but the loss of accuracy is
10321 * acceptable for all known use cases.
10323 guest_timing_exit_irqoff();
10325 local_irq_enable();
10328 kvm_vcpu_srcu_read_lock(vcpu);
10331 * Profile KVM exit RIPs:
10333 if (unlikely(prof_on == KVM_PROFILING)) {
10334 unsigned long rip = kvm_rip_read(vcpu);
10335 profile_hit(KVM_PROFILING, (void *)rip);
10338 if (unlikely(vcpu->arch.tsc_always_catchup))
10339 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10341 if (vcpu->arch.apic_attention)
10342 kvm_lapic_sync_from_vapic(vcpu);
10344 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
10348 if (req_immediate_exit)
10349 kvm_make_request(KVM_REQ_EVENT, vcpu);
10350 static_call(kvm_x86_cancel_injection)(vcpu);
10351 if (unlikely(vcpu->arch.apic_attention))
10352 kvm_lapic_sync_from_vapic(vcpu);
10357 /* Called within kvm->srcu read side. */
10358 static inline int vcpu_block(struct kvm_vcpu *vcpu)
10362 if (!kvm_arch_vcpu_runnable(vcpu)) {
10364 * Switch to the software timer before halt-polling/blocking as
10365 * the guest's timer may be a break event for the vCPU, and the
10366 * hypervisor timer runs only when the CPU is in guest mode.
10367 * Switch before halt-polling so that KVM recognizes an expired
10368 * timer before blocking.
10370 hv_timer = kvm_lapic_hv_timer_in_use(vcpu);
10372 kvm_lapic_switch_to_sw_timer(vcpu);
10374 kvm_vcpu_srcu_read_unlock(vcpu);
10375 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10376 kvm_vcpu_halt(vcpu);
10378 kvm_vcpu_block(vcpu);
10379 kvm_vcpu_srcu_read_lock(vcpu);
10382 kvm_lapic_switch_to_hv_timer(vcpu);
10384 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
10388 if (kvm_apic_accept_events(vcpu) < 0)
10390 switch(vcpu->arch.mp_state) {
10391 case KVM_MP_STATE_HALTED:
10392 case KVM_MP_STATE_AP_RESET_HOLD:
10393 vcpu->arch.pv.pv_unhalted = false;
10394 vcpu->arch.mp_state =
10395 KVM_MP_STATE_RUNNABLE;
10397 case KVM_MP_STATE_RUNNABLE:
10398 vcpu->arch.apf.halted = false;
10400 case KVM_MP_STATE_INIT_RECEIVED:
10408 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
10410 if (is_guest_mode(vcpu))
10411 kvm_check_nested_events(vcpu);
10413 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
10414 !vcpu->arch.apf.halted);
10417 /* Called within kvm->srcu read side. */
10418 static int vcpu_run(struct kvm_vcpu *vcpu)
10422 vcpu->arch.l1tf_flush_l1d = true;
10425 if (kvm_vcpu_running(vcpu)) {
10426 r = vcpu_enter_guest(vcpu);
10428 r = vcpu_block(vcpu);
10434 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
10435 if (kvm_xen_has_pending_events(vcpu))
10436 kvm_xen_inject_pending_events(vcpu);
10438 if (kvm_cpu_has_pending_timer(vcpu))
10439 kvm_inject_pending_timer_irqs(vcpu);
10441 if (dm_request_for_irq_injection(vcpu) &&
10442 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
10444 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
10445 ++vcpu->stat.request_irq_exits;
10449 if (__xfer_to_guest_mode_work_pending()) {
10450 kvm_vcpu_srcu_read_unlock(vcpu);
10451 r = xfer_to_guest_mode_handle_work(vcpu);
10452 kvm_vcpu_srcu_read_lock(vcpu);
10461 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
10463 return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
10466 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
10468 BUG_ON(!vcpu->arch.pio.count);
10470 return complete_emulated_io(vcpu);
10474 * Implements the following, as a state machine:
10477 * for each fragment
10478 * for each mmio piece in the fragment
10485 * for each fragment
10486 * for each mmio piece in the fragment
10491 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
10493 struct kvm_run *run = vcpu->run;
10494 struct kvm_mmio_fragment *frag;
10497 BUG_ON(!vcpu->mmio_needed);
10499 /* Complete previous fragment */
10500 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
10501 len = min(8u, frag->len);
10502 if (!vcpu->mmio_is_write)
10503 memcpy(frag->data, run->mmio.data, len);
10505 if (frag->len <= 8) {
10506 /* Switch to the next fragment. */
10508 vcpu->mmio_cur_fragment++;
10510 /* Go forward to the next mmio piece. */
10516 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
10517 vcpu->mmio_needed = 0;
10519 /* FIXME: return into emulator if single-stepping. */
10520 if (vcpu->mmio_is_write)
10522 vcpu->mmio_read_completed = 1;
10523 return complete_emulated_io(vcpu);
10526 run->exit_reason = KVM_EXIT_MMIO;
10527 run->mmio.phys_addr = frag->gpa;
10528 if (vcpu->mmio_is_write)
10529 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
10530 run->mmio.len = min(8u, frag->len);
10531 run->mmio.is_write = vcpu->mmio_is_write;
10532 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
10536 /* Swap (qemu) user FPU context for the guest FPU context. */
10537 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
10539 /* Exclude PKRU, it's restored separately immediately after VM-Exit. */
10540 fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, true);
10544 /* When vcpu_run ends, restore user space FPU context. */
10545 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
10547 fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, false);
10548 ++vcpu->stat.fpu_reload;
10552 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
10554 struct kvm_run *kvm_run = vcpu->run;
10558 kvm_sigset_activate(vcpu);
10559 kvm_run->flags = 0;
10560 kvm_load_guest_fpu(vcpu);
10562 kvm_vcpu_srcu_read_lock(vcpu);
10563 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
10564 if (kvm_run->immediate_exit) {
10569 * It should be impossible for the hypervisor timer to be in
10570 * use before KVM has ever run the vCPU.
10572 WARN_ON_ONCE(kvm_lapic_hv_timer_in_use(vcpu));
10574 kvm_vcpu_srcu_read_unlock(vcpu);
10575 kvm_vcpu_block(vcpu);
10576 kvm_vcpu_srcu_read_lock(vcpu);
10578 if (kvm_apic_accept_events(vcpu) < 0) {
10582 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
10584 if (signal_pending(current)) {
10586 kvm_run->exit_reason = KVM_EXIT_INTR;
10587 ++vcpu->stat.signal_exits;
10592 if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10593 (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10598 if (kvm_run->kvm_dirty_regs) {
10599 r = sync_regs(vcpu);
10604 /* re-sync apic's tpr */
10605 if (!lapic_in_kernel(vcpu)) {
10606 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
10612 if (unlikely(vcpu->arch.complete_userspace_io)) {
10613 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10614 vcpu->arch.complete_userspace_io = NULL;
10619 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10621 if (kvm_run->immediate_exit) {
10626 r = static_call(kvm_x86_vcpu_pre_run)(vcpu);
10630 r = vcpu_run(vcpu);
10633 kvm_put_guest_fpu(vcpu);
10634 if (kvm_run->kvm_valid_regs)
10636 post_kvm_run_save(vcpu);
10637 kvm_vcpu_srcu_read_unlock(vcpu);
10639 kvm_sigset_deactivate(vcpu);
10644 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10646 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10648 * We are here if userspace calls get_regs() in the middle of
10649 * instruction emulation. Registers state needs to be copied
10650 * back from emulation context to vcpu. Userspace shouldn't do
10651 * that usually, but some bad designed PV devices (vmware
10652 * backdoor interface) need this to work
10654 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10655 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10657 regs->rax = kvm_rax_read(vcpu);
10658 regs->rbx = kvm_rbx_read(vcpu);
10659 regs->rcx = kvm_rcx_read(vcpu);
10660 regs->rdx = kvm_rdx_read(vcpu);
10661 regs->rsi = kvm_rsi_read(vcpu);
10662 regs->rdi = kvm_rdi_read(vcpu);
10663 regs->rsp = kvm_rsp_read(vcpu);
10664 regs->rbp = kvm_rbp_read(vcpu);
10665 #ifdef CONFIG_X86_64
10666 regs->r8 = kvm_r8_read(vcpu);
10667 regs->r9 = kvm_r9_read(vcpu);
10668 regs->r10 = kvm_r10_read(vcpu);
10669 regs->r11 = kvm_r11_read(vcpu);
10670 regs->r12 = kvm_r12_read(vcpu);
10671 regs->r13 = kvm_r13_read(vcpu);
10672 regs->r14 = kvm_r14_read(vcpu);
10673 regs->r15 = kvm_r15_read(vcpu);
10676 regs->rip = kvm_rip_read(vcpu);
10677 regs->rflags = kvm_get_rflags(vcpu);
10680 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10683 __get_regs(vcpu, regs);
10688 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10690 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10691 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10693 kvm_rax_write(vcpu, regs->rax);
10694 kvm_rbx_write(vcpu, regs->rbx);
10695 kvm_rcx_write(vcpu, regs->rcx);
10696 kvm_rdx_write(vcpu, regs->rdx);
10697 kvm_rsi_write(vcpu, regs->rsi);
10698 kvm_rdi_write(vcpu, regs->rdi);
10699 kvm_rsp_write(vcpu, regs->rsp);
10700 kvm_rbp_write(vcpu, regs->rbp);
10701 #ifdef CONFIG_X86_64
10702 kvm_r8_write(vcpu, regs->r8);
10703 kvm_r9_write(vcpu, regs->r9);
10704 kvm_r10_write(vcpu, regs->r10);
10705 kvm_r11_write(vcpu, regs->r11);
10706 kvm_r12_write(vcpu, regs->r12);
10707 kvm_r13_write(vcpu, regs->r13);
10708 kvm_r14_write(vcpu, regs->r14);
10709 kvm_r15_write(vcpu, regs->r15);
10712 kvm_rip_write(vcpu, regs->rip);
10713 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10715 vcpu->arch.exception.pending = false;
10717 kvm_make_request(KVM_REQ_EVENT, vcpu);
10720 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10723 __set_regs(vcpu, regs);
10728 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10730 struct desc_ptr dt;
10732 if (vcpu->arch.guest_state_protected)
10733 goto skip_protected_regs;
10735 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10736 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10737 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10738 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10739 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10740 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10742 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10743 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10745 static_call(kvm_x86_get_idt)(vcpu, &dt);
10746 sregs->idt.limit = dt.size;
10747 sregs->idt.base = dt.address;
10748 static_call(kvm_x86_get_gdt)(vcpu, &dt);
10749 sregs->gdt.limit = dt.size;
10750 sregs->gdt.base = dt.address;
10752 sregs->cr2 = vcpu->arch.cr2;
10753 sregs->cr3 = kvm_read_cr3(vcpu);
10755 skip_protected_regs:
10756 sregs->cr0 = kvm_read_cr0(vcpu);
10757 sregs->cr4 = kvm_read_cr4(vcpu);
10758 sregs->cr8 = kvm_get_cr8(vcpu);
10759 sregs->efer = vcpu->arch.efer;
10760 sregs->apic_base = kvm_get_apic_base(vcpu);
10763 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10765 __get_sregs_common(vcpu, sregs);
10767 if (vcpu->arch.guest_state_protected)
10770 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10771 set_bit(vcpu->arch.interrupt.nr,
10772 (unsigned long *)sregs->interrupt_bitmap);
10775 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10779 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10781 if (vcpu->arch.guest_state_protected)
10784 if (is_pae_paging(vcpu)) {
10785 for (i = 0 ; i < 4 ; i++)
10786 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10787 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10791 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10792 struct kvm_sregs *sregs)
10795 __get_sregs(vcpu, sregs);
10800 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10801 struct kvm_mp_state *mp_state)
10806 if (kvm_mpx_supported())
10807 kvm_load_guest_fpu(vcpu);
10809 r = kvm_apic_accept_events(vcpu);
10814 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10815 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10816 vcpu->arch.pv.pv_unhalted)
10817 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10819 mp_state->mp_state = vcpu->arch.mp_state;
10822 if (kvm_mpx_supported())
10823 kvm_put_guest_fpu(vcpu);
10828 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10829 struct kvm_mp_state *mp_state)
10835 if (!lapic_in_kernel(vcpu) &&
10836 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10840 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10841 * INIT state; latched init should be reported using
10842 * KVM_SET_VCPU_EVENTS, so reject it here.
10844 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10845 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10846 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10849 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10850 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10851 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10853 vcpu->arch.mp_state = mp_state->mp_state;
10854 kvm_make_request(KVM_REQ_EVENT, vcpu);
10862 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10863 int reason, bool has_error_code, u32 error_code)
10865 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10868 init_emulate_ctxt(vcpu);
10870 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10871 has_error_code, error_code);
10873 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10874 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10875 vcpu->run->internal.ndata = 0;
10879 kvm_rip_write(vcpu, ctxt->eip);
10880 kvm_set_rflags(vcpu, ctxt->eflags);
10883 EXPORT_SYMBOL_GPL(kvm_task_switch);
10885 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10887 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10889 * When EFER.LME and CR0.PG are set, the processor is in
10890 * 64-bit mode (though maybe in a 32-bit code segment).
10891 * CR4.PAE and EFER.LMA must be set.
10893 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10895 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10899 * Not in 64-bit mode: EFER.LMA is clear and the code
10900 * segment cannot be 64-bit.
10902 if (sregs->efer & EFER_LMA || sregs->cs.l)
10906 return kvm_is_valid_cr4(vcpu, sregs->cr4);
10909 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10910 int *mmu_reset_needed, bool update_pdptrs)
10912 struct msr_data apic_base_msr;
10914 struct desc_ptr dt;
10916 if (!kvm_is_valid_sregs(vcpu, sregs))
10919 apic_base_msr.data = sregs->apic_base;
10920 apic_base_msr.host_initiated = true;
10921 if (kvm_set_apic_base(vcpu, &apic_base_msr))
10924 if (vcpu->arch.guest_state_protected)
10927 dt.size = sregs->idt.limit;
10928 dt.address = sregs->idt.base;
10929 static_call(kvm_x86_set_idt)(vcpu, &dt);
10930 dt.size = sregs->gdt.limit;
10931 dt.address = sregs->gdt.base;
10932 static_call(kvm_x86_set_gdt)(vcpu, &dt);
10934 vcpu->arch.cr2 = sregs->cr2;
10935 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10936 vcpu->arch.cr3 = sregs->cr3;
10937 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
10938 static_call_cond(kvm_x86_post_set_cr3)(vcpu, sregs->cr3);
10940 kvm_set_cr8(vcpu, sregs->cr8);
10942 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10943 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10945 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10946 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10947 vcpu->arch.cr0 = sregs->cr0;
10949 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10950 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10952 if (update_pdptrs) {
10953 idx = srcu_read_lock(&vcpu->kvm->srcu);
10954 if (is_pae_paging(vcpu)) {
10955 load_pdptrs(vcpu, kvm_read_cr3(vcpu));
10956 *mmu_reset_needed = 1;
10958 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10961 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10962 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10963 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10964 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10965 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10966 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10968 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10969 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10971 update_cr8_intercept(vcpu);
10973 /* Older userspace won't unhalt the vcpu on reset. */
10974 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10975 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10976 !is_protmode(vcpu))
10977 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10982 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10984 int pending_vec, max_bits;
10985 int mmu_reset_needed = 0;
10986 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10991 if (mmu_reset_needed)
10992 kvm_mmu_reset_context(vcpu);
10994 max_bits = KVM_NR_INTERRUPTS;
10995 pending_vec = find_first_bit(
10996 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
10998 if (pending_vec < max_bits) {
10999 kvm_queue_interrupt(vcpu, pending_vec, false);
11000 pr_debug("Set back pending irq %d\n", pending_vec);
11001 kvm_make_request(KVM_REQ_EVENT, vcpu);
11006 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
11008 int mmu_reset_needed = 0;
11009 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
11010 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
11011 !(sregs2->efer & EFER_LMA);
11014 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
11017 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
11020 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
11021 &mmu_reset_needed, !valid_pdptrs);
11025 if (valid_pdptrs) {
11026 for (i = 0; i < 4 ; i++)
11027 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
11029 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
11030 mmu_reset_needed = 1;
11031 vcpu->arch.pdptrs_from_userspace = true;
11033 if (mmu_reset_needed)
11034 kvm_mmu_reset_context(vcpu);
11038 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
11039 struct kvm_sregs *sregs)
11044 ret = __set_sregs(vcpu, sregs);
11049 static void kvm_arch_vcpu_guestdbg_update_apicv_inhibit(struct kvm *kvm)
11052 struct kvm_vcpu *vcpu;
11058 down_write(&kvm->arch.apicv_update_lock);
11060 kvm_for_each_vcpu(i, vcpu, kvm) {
11061 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ) {
11066 __kvm_set_or_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_BLOCKIRQ, set);
11067 up_write(&kvm->arch.apicv_update_lock);
11070 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
11071 struct kvm_guest_debug *dbg)
11073 unsigned long rflags;
11076 if (vcpu->arch.guest_state_protected)
11081 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
11083 if (vcpu->arch.exception.pending)
11085 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
11086 kvm_queue_exception(vcpu, DB_VECTOR);
11088 kvm_queue_exception(vcpu, BP_VECTOR);
11092 * Read rflags as long as potentially injected trace flags are still
11095 rflags = kvm_get_rflags(vcpu);
11097 vcpu->guest_debug = dbg->control;
11098 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
11099 vcpu->guest_debug = 0;
11101 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
11102 for (i = 0; i < KVM_NR_DB_REGS; ++i)
11103 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
11104 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
11106 for (i = 0; i < KVM_NR_DB_REGS; i++)
11107 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
11109 kvm_update_dr7(vcpu);
11111 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11112 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
11115 * Trigger an rflags update that will inject or remove the trace
11118 kvm_set_rflags(vcpu, rflags);
11120 static_call(kvm_x86_update_exception_bitmap)(vcpu);
11122 kvm_arch_vcpu_guestdbg_update_apicv_inhibit(vcpu->kvm);
11132 * Translate a guest virtual address to a guest physical address.
11134 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
11135 struct kvm_translation *tr)
11137 unsigned long vaddr = tr->linear_address;
11143 idx = srcu_read_lock(&vcpu->kvm->srcu);
11144 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
11145 srcu_read_unlock(&vcpu->kvm->srcu, idx);
11146 tr->physical_address = gpa;
11147 tr->valid = gpa != UNMAPPED_GVA;
11155 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
11157 struct fxregs_state *fxsave;
11159 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
11164 fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
11165 memcpy(fpu->fpr, fxsave->st_space, 128);
11166 fpu->fcw = fxsave->cwd;
11167 fpu->fsw = fxsave->swd;
11168 fpu->ftwx = fxsave->twd;
11169 fpu->last_opcode = fxsave->fop;
11170 fpu->last_ip = fxsave->rip;
11171 fpu->last_dp = fxsave->rdp;
11172 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
11178 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
11180 struct fxregs_state *fxsave;
11182 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
11187 fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
11189 memcpy(fxsave->st_space, fpu->fpr, 128);
11190 fxsave->cwd = fpu->fcw;
11191 fxsave->swd = fpu->fsw;
11192 fxsave->twd = fpu->ftwx;
11193 fxsave->fop = fpu->last_opcode;
11194 fxsave->rip = fpu->last_ip;
11195 fxsave->rdp = fpu->last_dp;
11196 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
11202 static void store_regs(struct kvm_vcpu *vcpu)
11204 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
11206 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
11207 __get_regs(vcpu, &vcpu->run->s.regs.regs);
11209 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
11210 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
11212 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
11213 kvm_vcpu_ioctl_x86_get_vcpu_events(
11214 vcpu, &vcpu->run->s.regs.events);
11217 static int sync_regs(struct kvm_vcpu *vcpu)
11219 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
11220 __set_regs(vcpu, &vcpu->run->s.regs.regs);
11221 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
11223 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
11224 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
11226 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
11228 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
11229 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
11230 vcpu, &vcpu->run->s.regs.events))
11232 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
11238 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
11240 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
11241 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
11242 "guest TSC will not be reliable\n");
11247 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
11252 vcpu->arch.last_vmentry_cpu = -1;
11253 vcpu->arch.regs_avail = ~0;
11254 vcpu->arch.regs_dirty = ~0;
11256 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
11257 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11259 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
11261 r = kvm_mmu_create(vcpu);
11265 if (irqchip_in_kernel(vcpu->kvm)) {
11266 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
11268 goto fail_mmu_destroy;
11271 * Defer evaluating inhibits until the vCPU is first run, as
11272 * this vCPU will not get notified of any changes until this
11273 * vCPU is visible to other vCPUs (marked online and added to
11274 * the set of vCPUs). Opportunistically mark APICv active as
11275 * VMX in particularly is highly unlikely to have inhibits.
11276 * Ignore the current per-VM APICv state so that vCPU creation
11277 * is guaranteed to run with a deterministic value, the request
11278 * will ensure the vCPU gets the correct state before VM-Entry.
11280 if (enable_apicv) {
11281 vcpu->arch.apicv_active = true;
11282 kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
11285 static_branch_inc(&kvm_has_noapic_vcpu);
11289 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
11291 goto fail_free_lapic;
11292 vcpu->arch.pio_data = page_address(page);
11294 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
11295 GFP_KERNEL_ACCOUNT);
11296 if (!vcpu->arch.mce_banks)
11297 goto fail_free_pio_data;
11298 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
11300 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
11301 GFP_KERNEL_ACCOUNT))
11302 goto fail_free_mce_banks;
11304 if (!alloc_emulate_ctxt(vcpu))
11305 goto free_wbinvd_dirty_mask;
11307 if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
11308 pr_err("kvm: failed to allocate vcpu's fpu\n");
11309 goto free_emulate_ctxt;
11312 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
11313 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
11315 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
11317 kvm_async_pf_hash_reset(vcpu);
11318 kvm_pmu_init(vcpu);
11320 vcpu->arch.pending_external_vector = -1;
11321 vcpu->arch.preempted_in_kernel = false;
11323 #if IS_ENABLED(CONFIG_HYPERV)
11324 vcpu->arch.hv_root_tdp = INVALID_PAGE;
11327 r = static_call(kvm_x86_vcpu_create)(vcpu);
11329 goto free_guest_fpu;
11331 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
11332 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
11333 kvm_xen_init_vcpu(vcpu);
11334 kvm_vcpu_mtrr_init(vcpu);
11336 kvm_set_tsc_khz(vcpu, vcpu->kvm->arch.default_tsc_khz);
11337 kvm_vcpu_reset(vcpu, false);
11338 kvm_init_mmu(vcpu);
11343 fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11345 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11346 free_wbinvd_dirty_mask:
11347 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11348 fail_free_mce_banks:
11349 kfree(vcpu->arch.mce_banks);
11350 fail_free_pio_data:
11351 free_page((unsigned long)vcpu->arch.pio_data);
11353 kvm_free_lapic(vcpu);
11355 kvm_mmu_destroy(vcpu);
11359 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
11361 struct kvm *kvm = vcpu->kvm;
11363 if (mutex_lock_killable(&vcpu->mutex))
11366 kvm_synchronize_tsc(vcpu, 0);
11369 /* poll control enabled by default */
11370 vcpu->arch.msr_kvm_poll_control = 1;
11372 mutex_unlock(&vcpu->mutex);
11374 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
11375 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
11376 KVMCLOCK_SYNC_PERIOD);
11379 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
11383 kvmclock_reset(vcpu);
11385 static_call(kvm_x86_vcpu_free)(vcpu);
11387 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11388 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11389 fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11391 kvm_xen_destroy_vcpu(vcpu);
11392 kvm_hv_vcpu_uninit(vcpu);
11393 kvm_pmu_destroy(vcpu);
11394 kfree(vcpu->arch.mce_banks);
11395 kvm_free_lapic(vcpu);
11396 idx = srcu_read_lock(&vcpu->kvm->srcu);
11397 kvm_mmu_destroy(vcpu);
11398 srcu_read_unlock(&vcpu->kvm->srcu, idx);
11399 free_page((unsigned long)vcpu->arch.pio_data);
11400 kvfree(vcpu->arch.cpuid_entries);
11401 if (!lapic_in_kernel(vcpu))
11402 static_branch_dec(&kvm_has_noapic_vcpu);
11405 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
11407 struct kvm_cpuid_entry2 *cpuid_0x1;
11408 unsigned long old_cr0 = kvm_read_cr0(vcpu);
11409 unsigned long new_cr0;
11412 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
11413 * to handle side effects. RESET emulation hits those flows and relies
11414 * on emulated/virtualized registers, including those that are loaded
11415 * into hardware, to be zeroed at vCPU creation. Use CRs as a sentinel
11416 * to detect improper or missing initialization.
11418 WARN_ON_ONCE(!init_event &&
11419 (old_cr0 || kvm_read_cr3(vcpu) || kvm_read_cr4(vcpu)));
11421 kvm_lapic_reset(vcpu, init_event);
11423 vcpu->arch.hflags = 0;
11425 vcpu->arch.smi_pending = 0;
11426 vcpu->arch.smi_count = 0;
11427 atomic_set(&vcpu->arch.nmi_queued, 0);
11428 vcpu->arch.nmi_pending = 0;
11429 vcpu->arch.nmi_injected = false;
11430 kvm_clear_interrupt_queue(vcpu);
11431 kvm_clear_exception_queue(vcpu);
11433 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
11434 kvm_update_dr0123(vcpu);
11435 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
11436 vcpu->arch.dr7 = DR7_FIXED_1;
11437 kvm_update_dr7(vcpu);
11439 vcpu->arch.cr2 = 0;
11441 kvm_make_request(KVM_REQ_EVENT, vcpu);
11442 vcpu->arch.apf.msr_en_val = 0;
11443 vcpu->arch.apf.msr_int_val = 0;
11444 vcpu->arch.st.msr_val = 0;
11446 kvmclock_reset(vcpu);
11448 kvm_clear_async_pf_completion_queue(vcpu);
11449 kvm_async_pf_hash_reset(vcpu);
11450 vcpu->arch.apf.halted = false;
11452 if (vcpu->arch.guest_fpu.fpstate && kvm_mpx_supported()) {
11453 struct fpstate *fpstate = vcpu->arch.guest_fpu.fpstate;
11456 * To avoid have the INIT path from kvm_apic_has_events() that be
11457 * called with loaded FPU and does not let userspace fix the state.
11460 kvm_put_guest_fpu(vcpu);
11462 fpstate_clear_xstate_component(fpstate, XFEATURE_BNDREGS);
11463 fpstate_clear_xstate_component(fpstate, XFEATURE_BNDCSR);
11466 kvm_load_guest_fpu(vcpu);
11470 kvm_pmu_reset(vcpu);
11471 vcpu->arch.smbase = 0x30000;
11473 vcpu->arch.msr_misc_features_enables = 0;
11475 __kvm_set_xcr(vcpu, 0, XFEATURE_MASK_FP);
11476 __kvm_set_msr(vcpu, MSR_IA32_XSS, 0, true);
11479 /* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
11480 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
11481 kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
11484 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
11485 * if no CPUID match is found. Note, it's impossible to get a match at
11486 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
11487 * i.e. it's impossible for kvm_find_cpuid_entry() to find a valid entry
11488 * on RESET. But, go through the motions in case that's ever remedied.
11490 cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
11491 kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
11493 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
11495 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
11496 kvm_rip_write(vcpu, 0xfff0);
11498 vcpu->arch.cr3 = 0;
11499 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11502 * CR0.CD/NW are set on RESET, preserved on INIT. Note, some versions
11503 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
11504 * (or qualify) that with a footnote stating that CD/NW are preserved.
11506 new_cr0 = X86_CR0_ET;
11508 new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
11510 new_cr0 |= X86_CR0_NW | X86_CR0_CD;
11512 static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
11513 static_call(kvm_x86_set_cr4)(vcpu, 0);
11514 static_call(kvm_x86_set_efer)(vcpu, 0);
11515 static_call(kvm_x86_update_exception_bitmap)(vcpu);
11518 * On the standard CR0/CR4/EFER modification paths, there are several
11519 * complex conditions determining whether the MMU has to be reset and/or
11520 * which PCIDs have to be flushed. However, CR0.WP and the paging-related
11521 * bits in CR4 and EFER are irrelevant if CR0.PG was '0'; and a reset+flush
11522 * is needed anyway if CR0.PG was '1' (which can only happen for INIT, as
11523 * CR0 will be '0' prior to RESET). So we only need to check CR0.PG here.
11525 if (old_cr0 & X86_CR0_PG) {
11526 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11527 kvm_mmu_reset_context(vcpu);
11531 * Intel's SDM states that all TLB entries are flushed on INIT. AMD's
11532 * APM states the TLBs are untouched by INIT, but it also states that
11533 * the TLBs are flushed on "External initialization of the processor."
11534 * Flush the guest TLB regardless of vendor, there is no meaningful
11535 * benefit in relying on the guest to flush the TLB immediately after
11536 * INIT. A spurious TLB flush is benign and likely negligible from a
11537 * performance perspective.
11540 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11542 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
11544 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
11546 struct kvm_segment cs;
11548 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
11549 cs.selector = vector << 8;
11550 cs.base = vector << 12;
11551 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
11552 kvm_rip_write(vcpu, 0);
11554 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
11556 int kvm_arch_hardware_enable(void)
11559 struct kvm_vcpu *vcpu;
11564 bool stable, backwards_tsc = false;
11566 kvm_user_return_msr_cpu_online();
11567 ret = static_call(kvm_x86_hardware_enable)();
11571 local_tsc = rdtsc();
11572 stable = !kvm_check_tsc_unstable();
11573 list_for_each_entry(kvm, &vm_list, vm_list) {
11574 kvm_for_each_vcpu(i, vcpu, kvm) {
11575 if (!stable && vcpu->cpu == smp_processor_id())
11576 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
11577 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
11578 backwards_tsc = true;
11579 if (vcpu->arch.last_host_tsc > max_tsc)
11580 max_tsc = vcpu->arch.last_host_tsc;
11586 * Sometimes, even reliable TSCs go backwards. This happens on
11587 * platforms that reset TSC during suspend or hibernate actions, but
11588 * maintain synchronization. We must compensate. Fortunately, we can
11589 * detect that condition here, which happens early in CPU bringup,
11590 * before any KVM threads can be running. Unfortunately, we can't
11591 * bring the TSCs fully up to date with real time, as we aren't yet far
11592 * enough into CPU bringup that we know how much real time has actually
11593 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
11594 * variables that haven't been updated yet.
11596 * So we simply find the maximum observed TSC above, then record the
11597 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
11598 * the adjustment will be applied. Note that we accumulate
11599 * adjustments, in case multiple suspend cycles happen before some VCPU
11600 * gets a chance to run again. In the event that no KVM threads get a
11601 * chance to run, we will miss the entire elapsed period, as we'll have
11602 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
11603 * loose cycle time. This isn't too big a deal, since the loss will be
11604 * uniform across all VCPUs (not to mention the scenario is extremely
11605 * unlikely). It is possible that a second hibernate recovery happens
11606 * much faster than a first, causing the observed TSC here to be
11607 * smaller; this would require additional padding adjustment, which is
11608 * why we set last_host_tsc to the local tsc observed here.
11610 * N.B. - this code below runs only on platforms with reliable TSC,
11611 * as that is the only way backwards_tsc is set above. Also note
11612 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
11613 * have the same delta_cyc adjustment applied if backwards_tsc
11614 * is detected. Note further, this adjustment is only done once,
11615 * as we reset last_host_tsc on all VCPUs to stop this from being
11616 * called multiple times (one for each physical CPU bringup).
11618 * Platforms with unreliable TSCs don't have to deal with this, they
11619 * will be compensated by the logic in vcpu_load, which sets the TSC to
11620 * catchup mode. This will catchup all VCPUs to real time, but cannot
11621 * guarantee that they stay in perfect synchronization.
11623 if (backwards_tsc) {
11624 u64 delta_cyc = max_tsc - local_tsc;
11625 list_for_each_entry(kvm, &vm_list, vm_list) {
11626 kvm->arch.backwards_tsc_observed = true;
11627 kvm_for_each_vcpu(i, vcpu, kvm) {
11628 vcpu->arch.tsc_offset_adjustment += delta_cyc;
11629 vcpu->arch.last_host_tsc = local_tsc;
11630 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11634 * We have to disable TSC offset matching.. if you were
11635 * booting a VM while issuing an S4 host suspend....
11636 * you may have some problem. Solving this issue is
11637 * left as an exercise to the reader.
11639 kvm->arch.last_tsc_nsec = 0;
11640 kvm->arch.last_tsc_write = 0;
11647 void kvm_arch_hardware_disable(void)
11649 static_call(kvm_x86_hardware_disable)();
11650 drop_user_return_notifiers();
11653 static inline void kvm_ops_update(struct kvm_x86_init_ops *ops)
11655 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11657 #define __KVM_X86_OP(func) \
11658 static_call_update(kvm_x86_##func, kvm_x86_ops.func);
11659 #define KVM_X86_OP(func) \
11660 WARN_ON(!kvm_x86_ops.func); __KVM_X86_OP(func)
11661 #define KVM_X86_OP_OPTIONAL __KVM_X86_OP
11662 #define KVM_X86_OP_OPTIONAL_RET0(func) \
11663 static_call_update(kvm_x86_##func, (void *)kvm_x86_ops.func ? : \
11664 (void *)__static_call_return0);
11665 #include <asm/kvm-x86-ops.h>
11666 #undef __KVM_X86_OP
11668 kvm_pmu_ops_update(ops->pmu_ops);
11671 int kvm_arch_hardware_setup(void *opaque)
11673 struct kvm_x86_init_ops *ops = opaque;
11676 rdmsrl_safe(MSR_EFER, &host_efer);
11678 if (boot_cpu_has(X86_FEATURE_XSAVES))
11679 rdmsrl(MSR_IA32_XSS, host_xss);
11681 r = ops->hardware_setup();
11685 kvm_ops_update(ops);
11687 kvm_register_perf_callbacks(ops->handle_intel_pt_intr);
11689 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11692 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11693 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11694 #undef __kvm_cpu_cap_has
11696 if (kvm_has_tsc_control) {
11698 * Make sure the user can only configure tsc_khz values that
11699 * fit into a signed integer.
11700 * A min value is not calculated because it will always
11701 * be 1 on all machines.
11703 u64 max = min(0x7fffffffULL,
11704 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11705 kvm_max_guest_tsc_khz = max;
11707 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11708 kvm_init_msr_list();
11712 void kvm_arch_hardware_unsetup(void)
11714 kvm_unregister_perf_callbacks();
11716 static_call(kvm_x86_hardware_unsetup)();
11719 int kvm_arch_check_processor_compat(void *opaque)
11721 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11722 struct kvm_x86_init_ops *ops = opaque;
11724 WARN_ON(!irqs_disabled());
11726 if (__cr4_reserved_bits(cpu_has, c) !=
11727 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11730 return ops->check_processor_compatibility();
11733 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11735 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11737 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11739 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11741 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11744 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11745 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11747 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11749 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11751 vcpu->arch.l1tf_flush_l1d = true;
11752 if (pmu->version && unlikely(pmu->event_count)) {
11753 pmu->need_cleanup = true;
11754 kvm_make_request(KVM_REQ_PMU, vcpu);
11756 static_call(kvm_x86_sched_in)(vcpu, cpu);
11759 void kvm_arch_free_vm(struct kvm *kvm)
11761 kfree(to_kvm_hv(kvm)->hv_pa_pg);
11762 __kvm_arch_free_vm(kvm);
11766 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11769 unsigned long flags;
11774 ret = kvm_page_track_init(kvm);
11778 ret = kvm_mmu_init_vm(kvm);
11780 goto out_page_track;
11782 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11783 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11784 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11786 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11787 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11788 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11789 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11790 &kvm->arch.irq_sources_bitmap);
11792 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11793 mutex_init(&kvm->arch.apic_map_lock);
11794 seqcount_raw_spinlock_init(&kvm->arch.pvclock_sc, &kvm->arch.tsc_write_lock);
11795 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11797 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
11798 pvclock_update_vm_gtod_copy(kvm);
11799 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
11801 kvm->arch.default_tsc_khz = max_tsc_khz ? : tsc_khz;
11802 kvm->arch.guest_can_read_msr_platform_info = true;
11803 kvm->arch.enable_pmu = enable_pmu;
11805 #if IS_ENABLED(CONFIG_HYPERV)
11806 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11807 kvm->arch.hv_root_tdp = INVALID_PAGE;
11810 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11811 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11813 kvm_apicv_init(kvm);
11814 kvm_hv_init_vm(kvm);
11815 kvm_xen_init_vm(kvm);
11817 return static_call(kvm_x86_vm_init)(kvm);
11820 kvm_page_track_cleanup(kvm);
11825 int kvm_arch_post_init_vm(struct kvm *kvm)
11827 return kvm_mmu_post_init_vm(kvm);
11830 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11833 kvm_mmu_unload(vcpu);
11837 static void kvm_unload_vcpu_mmus(struct kvm *kvm)
11840 struct kvm_vcpu *vcpu;
11842 kvm_for_each_vcpu(i, vcpu, kvm) {
11843 kvm_clear_async_pf_completion_queue(vcpu);
11844 kvm_unload_vcpu_mmu(vcpu);
11848 void kvm_arch_sync_events(struct kvm *kvm)
11850 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11851 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11856 * __x86_set_memory_region: Setup KVM internal memory slot
11858 * @kvm: the kvm pointer to the VM.
11859 * @id: the slot ID to setup.
11860 * @gpa: the GPA to install the slot (unused when @size == 0).
11861 * @size: the size of the slot. Set to zero to uninstall a slot.
11863 * This function helps to setup a KVM internal memory slot. Specify
11864 * @size > 0 to install a new slot, while @size == 0 to uninstall a
11865 * slot. The return code can be one of the following:
11867 * HVA: on success (uninstall will return a bogus HVA)
11870 * The caller should always use IS_ERR() to check the return value
11871 * before use. Note, the KVM internal memory slots are guaranteed to
11872 * remain valid and unchanged until the VM is destroyed, i.e., the
11873 * GPA->HVA translation will not change. However, the HVA is a user
11874 * address, i.e. its accessibility is not guaranteed, and must be
11875 * accessed via __copy_{to,from}_user().
11877 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11881 unsigned long hva, old_npages;
11882 struct kvm_memslots *slots = kvm_memslots(kvm);
11883 struct kvm_memory_slot *slot;
11885 /* Called with kvm->slots_lock held. */
11886 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11887 return ERR_PTR_USR(-EINVAL);
11889 slot = id_to_memslot(slots, id);
11891 if (slot && slot->npages)
11892 return ERR_PTR_USR(-EEXIST);
11895 * MAP_SHARED to prevent internal slot pages from being moved
11898 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11899 MAP_SHARED | MAP_ANONYMOUS, 0);
11900 if (IS_ERR((void *)hva))
11901 return (void __user *)hva;
11903 if (!slot || !slot->npages)
11906 old_npages = slot->npages;
11907 hva = slot->userspace_addr;
11910 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11911 struct kvm_userspace_memory_region m;
11913 m.slot = id | (i << 16);
11915 m.guest_phys_addr = gpa;
11916 m.userspace_addr = hva;
11917 m.memory_size = size;
11918 r = __kvm_set_memory_region(kvm, &m);
11920 return ERR_PTR_USR(r);
11924 vm_munmap(hva, old_npages * PAGE_SIZE);
11926 return (void __user *)hva;
11928 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11930 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11932 kvm_mmu_pre_destroy_vm(kvm);
11935 void kvm_arch_destroy_vm(struct kvm *kvm)
11937 if (current->mm == kvm->mm) {
11939 * Free memory regions allocated on behalf of userspace,
11940 * unless the memory map has changed due to process exit
11943 mutex_lock(&kvm->slots_lock);
11944 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11946 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11948 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11949 mutex_unlock(&kvm->slots_lock);
11951 kvm_unload_vcpu_mmus(kvm);
11952 static_call_cond(kvm_x86_vm_destroy)(kvm);
11953 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11954 kvm_pic_destroy(kvm);
11955 kvm_ioapic_destroy(kvm);
11956 kvm_destroy_vcpus(kvm);
11957 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11958 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11959 kvm_mmu_uninit_vm(kvm);
11960 kvm_page_track_cleanup(kvm);
11961 kvm_xen_destroy_vm(kvm);
11962 kvm_hv_destroy_vm(kvm);
11965 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11969 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11970 kvfree(slot->arch.rmap[i]);
11971 slot->arch.rmap[i] = NULL;
11975 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11979 memslot_rmap_free(slot);
11981 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11982 kvfree(slot->arch.lpage_info[i - 1]);
11983 slot->arch.lpage_info[i - 1] = NULL;
11986 kvm_page_track_free_memslot(slot);
11989 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages)
11991 const int sz = sizeof(*slot->arch.rmap[0]);
11994 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11996 int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11998 if (slot->arch.rmap[i])
12001 slot->arch.rmap[i] = __vcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
12002 if (!slot->arch.rmap[i]) {
12003 memslot_rmap_free(slot);
12011 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
12012 struct kvm_memory_slot *slot)
12014 unsigned long npages = slot->npages;
12018 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
12019 * old arrays will be freed by __kvm_set_memory_region() if installing
12020 * the new memslot is successful.
12022 memset(&slot->arch, 0, sizeof(slot->arch));
12024 if (kvm_memslots_have_rmaps(kvm)) {
12025 r = memslot_rmap_alloc(slot, npages);
12030 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12031 struct kvm_lpage_info *linfo;
12032 unsigned long ugfn;
12036 lpages = __kvm_mmu_slot_lpages(slot, npages, level);
12038 linfo = __vcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
12042 slot->arch.lpage_info[i - 1] = linfo;
12044 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
12045 linfo[0].disallow_lpage = 1;
12046 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
12047 linfo[lpages - 1].disallow_lpage = 1;
12048 ugfn = slot->userspace_addr >> PAGE_SHIFT;
12050 * If the gfn and userspace address are not aligned wrt each
12051 * other, disable large page support for this slot.
12053 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
12056 for (j = 0; j < lpages; ++j)
12057 linfo[j].disallow_lpage = 1;
12061 if (kvm_page_track_create_memslot(kvm, slot, npages))
12067 memslot_rmap_free(slot);
12069 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12070 kvfree(slot->arch.lpage_info[i - 1]);
12071 slot->arch.lpage_info[i - 1] = NULL;
12076 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
12078 struct kvm_vcpu *vcpu;
12082 * memslots->generation has been incremented.
12083 * mmio generation may have reached its maximum value.
12085 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
12087 /* Force re-initialization of steal_time cache */
12088 kvm_for_each_vcpu(i, vcpu, kvm)
12089 kvm_vcpu_kick(vcpu);
12092 int kvm_arch_prepare_memory_region(struct kvm *kvm,
12093 const struct kvm_memory_slot *old,
12094 struct kvm_memory_slot *new,
12095 enum kvm_mr_change change)
12097 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE) {
12098 if ((new->base_gfn + new->npages - 1) > kvm_mmu_max_gfn())
12101 return kvm_alloc_memslot_metadata(kvm, new);
12104 if (change == KVM_MR_FLAGS_ONLY)
12105 memcpy(&new->arch, &old->arch, sizeof(old->arch));
12106 else if (WARN_ON_ONCE(change != KVM_MR_DELETE))
12113 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
12115 struct kvm_arch *ka = &kvm->arch;
12117 if (!kvm_x86_ops.cpu_dirty_log_size)
12120 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
12121 (!enable && --ka->cpu_dirty_logging_count == 0))
12122 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
12124 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
12127 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
12128 struct kvm_memory_slot *old,
12129 const struct kvm_memory_slot *new,
12130 enum kvm_mr_change change)
12132 u32 old_flags = old ? old->flags : 0;
12133 u32 new_flags = new ? new->flags : 0;
12134 bool log_dirty_pages = new_flags & KVM_MEM_LOG_DIRTY_PAGES;
12137 * Update CPU dirty logging if dirty logging is being toggled. This
12138 * applies to all operations.
12140 if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)
12141 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
12144 * Nothing more to do for RO slots (which can't be dirtied and can't be
12145 * made writable) or CREATE/MOVE/DELETE of a slot.
12147 * For a memslot with dirty logging disabled:
12148 * CREATE: No dirty mappings will already exist.
12149 * MOVE/DELETE: The old mappings will already have been cleaned up by
12150 * kvm_arch_flush_shadow_memslot()
12152 * For a memslot with dirty logging enabled:
12153 * CREATE: No shadow pages exist, thus nothing to write-protect
12154 * and no dirty bits to clear.
12155 * MOVE/DELETE: The old mappings will already have been cleaned up by
12156 * kvm_arch_flush_shadow_memslot().
12158 if ((change != KVM_MR_FLAGS_ONLY) || (new_flags & KVM_MEM_READONLY))
12162 * READONLY and non-flags changes were filtered out above, and the only
12163 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
12164 * logging isn't being toggled on or off.
12166 if (WARN_ON_ONCE(!((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)))
12169 if (!log_dirty_pages) {
12171 * Dirty logging tracks sptes in 4k granularity, meaning that
12172 * large sptes have to be split. If live migration succeeds,
12173 * the guest in the source machine will be destroyed and large
12174 * sptes will be created in the destination. However, if the
12175 * guest continues to run in the source machine (for example if
12176 * live migration fails), small sptes will remain around and
12177 * cause bad performance.
12179 * Scan sptes if dirty logging has been stopped, dropping those
12180 * which can be collapsed into a single large-page spte. Later
12181 * page faults will create the large-page sptes.
12183 kvm_mmu_zap_collapsible_sptes(kvm, new);
12186 * Initially-all-set does not require write protecting any page,
12187 * because they're all assumed to be dirty.
12189 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
12192 if (READ_ONCE(eager_page_split))
12193 kvm_mmu_slot_try_split_huge_pages(kvm, new, PG_LEVEL_4K);
12195 if (kvm_x86_ops.cpu_dirty_log_size) {
12196 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
12197 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
12199 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
12204 void kvm_arch_commit_memory_region(struct kvm *kvm,
12205 struct kvm_memory_slot *old,
12206 const struct kvm_memory_slot *new,
12207 enum kvm_mr_change change)
12209 if (!kvm->arch.n_requested_mmu_pages &&
12210 (change == KVM_MR_CREATE || change == KVM_MR_DELETE)) {
12211 unsigned long nr_mmu_pages;
12213 nr_mmu_pages = kvm->nr_memslot_pages / KVM_MEMSLOT_PAGES_TO_MMU_PAGES_RATIO;
12214 nr_mmu_pages = max(nr_mmu_pages, KVM_MIN_ALLOC_MMU_PAGES);
12215 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
12218 kvm_mmu_slot_apply_flags(kvm, old, new, change);
12220 /* Free the arrays associated with the old memslot. */
12221 if (change == KVM_MR_MOVE)
12222 kvm_arch_free_memslot(kvm, old);
12225 void kvm_arch_flush_shadow_all(struct kvm *kvm)
12227 kvm_mmu_zap_all(kvm);
12230 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
12231 struct kvm_memory_slot *slot)
12233 kvm_page_track_flush_slot(kvm, slot);
12236 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
12238 return (is_guest_mode(vcpu) &&
12239 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
12242 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
12244 if (!list_empty_careful(&vcpu->async_pf.done))
12247 if (kvm_apic_has_events(vcpu))
12250 if (vcpu->arch.pv.pv_unhalted)
12253 if (vcpu->arch.exception.pending)
12256 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12257 (vcpu->arch.nmi_pending &&
12258 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
12261 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
12262 (vcpu->arch.smi_pending &&
12263 static_call(kvm_x86_smi_allowed)(vcpu, false)))
12266 if (kvm_arch_interrupt_allowed(vcpu) &&
12267 (kvm_cpu_has_interrupt(vcpu) ||
12268 kvm_guest_apic_has_interrupt(vcpu)))
12271 if (kvm_hv_has_stimer_pending(vcpu))
12274 if (is_guest_mode(vcpu) &&
12275 kvm_x86_ops.nested_ops->hv_timer_pending &&
12276 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
12279 if (kvm_xen_has_pending_events(vcpu))
12282 if (kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu))
12288 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
12290 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
12293 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
12295 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
12301 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
12303 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
12306 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12307 kvm_test_request(KVM_REQ_SMI, vcpu) ||
12308 kvm_test_request(KVM_REQ_EVENT, vcpu))
12311 return kvm_arch_dy_has_pending_interrupt(vcpu);
12314 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
12316 if (vcpu->arch.guest_state_protected)
12319 return vcpu->arch.preempted_in_kernel;
12322 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
12324 return kvm_rip_read(vcpu);
12327 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
12329 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
12332 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
12334 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
12337 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
12339 /* Can't read the RIP when guest state is protected, just return 0 */
12340 if (vcpu->arch.guest_state_protected)
12343 if (is_64_bit_mode(vcpu))
12344 return kvm_rip_read(vcpu);
12345 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
12346 kvm_rip_read(vcpu));
12348 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
12350 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
12352 return kvm_get_linear_rip(vcpu) == linear_rip;
12354 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
12356 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
12358 unsigned long rflags;
12360 rflags = static_call(kvm_x86_get_rflags)(vcpu);
12361 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
12362 rflags &= ~X86_EFLAGS_TF;
12365 EXPORT_SYMBOL_GPL(kvm_get_rflags);
12367 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12369 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
12370 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
12371 rflags |= X86_EFLAGS_TF;
12372 static_call(kvm_x86_set_rflags)(vcpu, rflags);
12375 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12377 __kvm_set_rflags(vcpu, rflags);
12378 kvm_make_request(KVM_REQ_EVENT, vcpu);
12380 EXPORT_SYMBOL_GPL(kvm_set_rflags);
12382 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
12384 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
12386 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
12389 static inline u32 kvm_async_pf_next_probe(u32 key)
12391 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
12394 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12396 u32 key = kvm_async_pf_hash_fn(gfn);
12398 while (vcpu->arch.apf.gfns[key] != ~0)
12399 key = kvm_async_pf_next_probe(key);
12401 vcpu->arch.apf.gfns[key] = gfn;
12404 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
12407 u32 key = kvm_async_pf_hash_fn(gfn);
12409 for (i = 0; i < ASYNC_PF_PER_VCPU &&
12410 (vcpu->arch.apf.gfns[key] != gfn &&
12411 vcpu->arch.apf.gfns[key] != ~0); i++)
12412 key = kvm_async_pf_next_probe(key);
12417 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12419 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
12422 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12426 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
12428 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
12432 vcpu->arch.apf.gfns[i] = ~0;
12434 j = kvm_async_pf_next_probe(j);
12435 if (vcpu->arch.apf.gfns[j] == ~0)
12437 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
12439 * k lies cyclically in ]i,j]
12441 * |....j i.k.| or |.k..j i...|
12443 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
12444 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
12449 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
12451 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
12453 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
12457 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
12459 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12461 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12462 &token, offset, sizeof(token));
12465 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
12467 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12470 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12471 &val, offset, sizeof(val)))
12477 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
12480 if (!kvm_pv_async_pf_enabled(vcpu))
12483 if (vcpu->arch.apf.send_user_only &&
12484 static_call(kvm_x86_get_cpl)(vcpu) == 0)
12487 if (is_guest_mode(vcpu)) {
12489 * L1 needs to opt into the special #PF vmexits that are
12490 * used to deliver async page faults.
12492 return vcpu->arch.apf.delivery_as_pf_vmexit;
12495 * Play it safe in case the guest temporarily disables paging.
12496 * The real mode IDT in particular is unlikely to have a #PF
12499 return is_paging(vcpu);
12503 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
12505 if (unlikely(!lapic_in_kernel(vcpu) ||
12506 kvm_event_needs_reinjection(vcpu) ||
12507 vcpu->arch.exception.pending))
12510 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
12514 * If interrupts are off we cannot even use an artificial
12517 return kvm_arch_interrupt_allowed(vcpu);
12520 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
12521 struct kvm_async_pf *work)
12523 struct x86_exception fault;
12525 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
12526 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
12528 if (kvm_can_deliver_async_pf(vcpu) &&
12529 !apf_put_user_notpresent(vcpu)) {
12530 fault.vector = PF_VECTOR;
12531 fault.error_code_valid = true;
12532 fault.error_code = 0;
12533 fault.nested_page_fault = false;
12534 fault.address = work->arch.token;
12535 fault.async_page_fault = true;
12536 kvm_inject_page_fault(vcpu, &fault);
12540 * It is not possible to deliver a paravirtualized asynchronous
12541 * page fault, but putting the guest in an artificial halt state
12542 * can be beneficial nevertheless: if an interrupt arrives, we
12543 * can deliver it timely and perhaps the guest will schedule
12544 * another process. When the instruction that triggered a page
12545 * fault is retried, hopefully the page will be ready in the host.
12547 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
12552 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
12553 struct kvm_async_pf *work)
12555 struct kvm_lapic_irq irq = {
12556 .delivery_mode = APIC_DM_FIXED,
12557 .vector = vcpu->arch.apf.vec
12560 if (work->wakeup_all)
12561 work->arch.token = ~0; /* broadcast wakeup */
12563 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
12564 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
12566 if ((work->wakeup_all || work->notpresent_injected) &&
12567 kvm_pv_async_pf_enabled(vcpu) &&
12568 !apf_put_user_ready(vcpu, work->arch.token)) {
12569 vcpu->arch.apf.pageready_pending = true;
12570 kvm_apic_set_irq(vcpu, &irq, NULL);
12573 vcpu->arch.apf.halted = false;
12574 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
12577 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
12579 kvm_make_request(KVM_REQ_APF_READY, vcpu);
12580 if (!vcpu->arch.apf.pageready_pending)
12581 kvm_vcpu_kick(vcpu);
12584 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
12586 if (!kvm_pv_async_pf_enabled(vcpu))
12589 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
12592 void kvm_arch_start_assignment(struct kvm *kvm)
12594 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
12595 static_call_cond(kvm_x86_pi_start_assignment)(kvm);
12597 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
12599 void kvm_arch_end_assignment(struct kvm *kvm)
12601 atomic_dec(&kvm->arch.assigned_device_count);
12603 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
12605 bool kvm_arch_has_assigned_device(struct kvm *kvm)
12607 return atomic_read(&kvm->arch.assigned_device_count);
12609 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
12611 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
12613 atomic_inc(&kvm->arch.noncoherent_dma_count);
12615 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12617 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12619 atomic_dec(&kvm->arch.noncoherent_dma_count);
12621 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12623 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12625 return atomic_read(&kvm->arch.noncoherent_dma_count);
12627 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12629 bool kvm_arch_has_irq_bypass(void)
12634 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12635 struct irq_bypass_producer *prod)
12637 struct kvm_kernel_irqfd *irqfd =
12638 container_of(cons, struct kvm_kernel_irqfd, consumer);
12641 irqfd->producer = prod;
12642 kvm_arch_start_assignment(irqfd->kvm);
12643 ret = static_call(kvm_x86_pi_update_irte)(irqfd->kvm,
12644 prod->irq, irqfd->gsi, 1);
12647 kvm_arch_end_assignment(irqfd->kvm);
12652 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12653 struct irq_bypass_producer *prod)
12656 struct kvm_kernel_irqfd *irqfd =
12657 container_of(cons, struct kvm_kernel_irqfd, consumer);
12659 WARN_ON(irqfd->producer != prod);
12660 irqfd->producer = NULL;
12663 * When producer of consumer is unregistered, we change back to
12664 * remapped mode, so we can re-use the current implementation
12665 * when the irq is masked/disabled or the consumer side (KVM
12666 * int this case doesn't want to receive the interrupts.
12668 ret = static_call(kvm_x86_pi_update_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12670 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12671 " fails: %d\n", irqfd->consumer.token, ret);
12673 kvm_arch_end_assignment(irqfd->kvm);
12676 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12677 uint32_t guest_irq, bool set)
12679 return static_call(kvm_x86_pi_update_irte)(kvm, host_irq, guest_irq, set);
12682 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *old,
12683 struct kvm_kernel_irq_routing_entry *new)
12685 if (new->type != KVM_IRQ_ROUTING_MSI)
12688 return !!memcmp(&old->msi, &new->msi, sizeof(new->msi));
12691 bool kvm_vector_hashing_enabled(void)
12693 return vector_hashing;
12696 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12698 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12700 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12703 int kvm_spec_ctrl_test_value(u64 value)
12706 * test that setting IA32_SPEC_CTRL to given value
12707 * is allowed by the host processor
12711 unsigned long flags;
12714 local_irq_save(flags);
12716 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12718 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12721 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12723 local_irq_restore(flags);
12727 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12729 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12731 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
12732 struct x86_exception fault;
12733 u64 access = error_code &
12734 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12736 if (!(error_code & PFERR_PRESENT_MASK) ||
12737 mmu->gva_to_gpa(vcpu, mmu, gva, access, &fault) != UNMAPPED_GVA) {
12739 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12740 * tables probably do not match the TLB. Just proceed
12741 * with the error code that the processor gave.
12743 fault.vector = PF_VECTOR;
12744 fault.error_code_valid = true;
12745 fault.error_code = error_code;
12746 fault.nested_page_fault = false;
12747 fault.address = gva;
12749 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12751 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12754 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12755 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12756 * indicates whether exit to userspace is needed.
12758 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12759 struct x86_exception *e)
12761 if (r == X86EMUL_PROPAGATE_FAULT) {
12762 kvm_inject_emulated_page_fault(vcpu, e);
12767 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12768 * while handling a VMX instruction KVM could've handled the request
12769 * correctly by exiting to userspace and performing I/O but there
12770 * doesn't seem to be a real use-case behind such requests, just return
12771 * KVM_EXIT_INTERNAL_ERROR for now.
12773 kvm_prepare_emulation_failure_exit(vcpu);
12777 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12779 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12782 struct x86_exception e;
12789 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12790 if (r != X86EMUL_CONTINUE)
12791 return kvm_handle_memory_failure(vcpu, r, &e);
12793 if (operand.pcid >> 12 != 0) {
12794 kvm_inject_gp(vcpu, 0);
12798 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12801 case INVPCID_TYPE_INDIV_ADDR:
12802 if ((!pcid_enabled && (operand.pcid != 0)) ||
12803 is_noncanonical_address(operand.gla, vcpu)) {
12804 kvm_inject_gp(vcpu, 0);
12807 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12808 return kvm_skip_emulated_instruction(vcpu);
12810 case INVPCID_TYPE_SINGLE_CTXT:
12811 if (!pcid_enabled && (operand.pcid != 0)) {
12812 kvm_inject_gp(vcpu, 0);
12816 kvm_invalidate_pcid(vcpu, operand.pcid);
12817 return kvm_skip_emulated_instruction(vcpu);
12819 case INVPCID_TYPE_ALL_NON_GLOBAL:
12821 * Currently, KVM doesn't mark global entries in the shadow
12822 * page tables, so a non-global flush just degenerates to a
12823 * global flush. If needed, we could optimize this later by
12824 * keeping track of global entries in shadow page tables.
12828 case INVPCID_TYPE_ALL_INCL_GLOBAL:
12829 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12830 return kvm_skip_emulated_instruction(vcpu);
12833 kvm_inject_gp(vcpu, 0);
12837 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12839 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12841 struct kvm_run *run = vcpu->run;
12842 struct kvm_mmio_fragment *frag;
12845 BUG_ON(!vcpu->mmio_needed);
12847 /* Complete previous fragment */
12848 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12849 len = min(8u, frag->len);
12850 if (!vcpu->mmio_is_write)
12851 memcpy(frag->data, run->mmio.data, len);
12853 if (frag->len <= 8) {
12854 /* Switch to the next fragment. */
12856 vcpu->mmio_cur_fragment++;
12858 /* Go forward to the next mmio piece. */
12864 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12865 vcpu->mmio_needed = 0;
12867 // VMG change, at this point, we're always done
12868 // RIP has already been advanced
12872 // More MMIO is needed
12873 run->mmio.phys_addr = frag->gpa;
12874 run->mmio.len = min(8u, frag->len);
12875 run->mmio.is_write = vcpu->mmio_is_write;
12876 if (run->mmio.is_write)
12877 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12878 run->exit_reason = KVM_EXIT_MMIO;
12880 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12885 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12889 struct kvm_mmio_fragment *frag;
12894 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12895 if (handled == bytes)
12902 /*TODO: Check if need to increment number of frags */
12903 frag = vcpu->mmio_fragments;
12904 vcpu->mmio_nr_fragments = 1;
12909 vcpu->mmio_needed = 1;
12910 vcpu->mmio_cur_fragment = 0;
12912 vcpu->run->mmio.phys_addr = gpa;
12913 vcpu->run->mmio.len = min(8u, frag->len);
12914 vcpu->run->mmio.is_write = 1;
12915 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12916 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12918 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12922 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12924 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12928 struct kvm_mmio_fragment *frag;
12933 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12934 if (handled == bytes)
12941 /*TODO: Check if need to increment number of frags */
12942 frag = vcpu->mmio_fragments;
12943 vcpu->mmio_nr_fragments = 1;
12948 vcpu->mmio_needed = 1;
12949 vcpu->mmio_cur_fragment = 0;
12951 vcpu->run->mmio.phys_addr = gpa;
12952 vcpu->run->mmio.len = min(8u, frag->len);
12953 vcpu->run->mmio.is_write = 0;
12954 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12956 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12960 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12962 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12963 unsigned int port);
12965 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
12967 int size = vcpu->arch.pio.size;
12968 int port = vcpu->arch.pio.port;
12970 vcpu->arch.pio.count = 0;
12971 if (vcpu->arch.sev_pio_count)
12972 return kvm_sev_es_outs(vcpu, size, port);
12976 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12980 unsigned int count =
12981 min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12982 int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
12984 /* memcpy done already by emulator_pio_out. */
12985 vcpu->arch.sev_pio_count -= count;
12986 vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12990 /* Emulation done by the kernel. */
12991 if (!vcpu->arch.sev_pio_count)
12995 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
12999 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
13000 unsigned int port);
13002 static void advance_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
13004 unsigned count = vcpu->arch.pio.count;
13005 complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
13006 vcpu->arch.sev_pio_count -= count;
13007 vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
13010 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
13012 int size = vcpu->arch.pio.size;
13013 int port = vcpu->arch.pio.port;
13015 advance_sev_es_emulated_ins(vcpu);
13016 if (vcpu->arch.sev_pio_count)
13017 return kvm_sev_es_ins(vcpu, size, port);
13021 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
13025 unsigned int count =
13026 min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
13027 if (!__emulator_pio_in(vcpu, size, port, count))
13030 /* Emulation done by the kernel. */
13031 advance_sev_es_emulated_ins(vcpu);
13032 if (!vcpu->arch.sev_pio_count)
13036 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
13040 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
13041 unsigned int port, void *data, unsigned int count,
13044 vcpu->arch.sev_pio_data = data;
13045 vcpu->arch.sev_pio_count = count;
13046 return in ? kvm_sev_es_ins(vcpu, size, port)
13047 : kvm_sev_es_outs(vcpu, size, port);
13049 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
13051 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
13052 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
13053 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
13054 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
13055 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
13056 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
13057 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
13058 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
13059 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
13060 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
13061 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
13062 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
13063 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
13064 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
13065 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
13066 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
13067 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
13068 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
13069 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
13070 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
13071 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
13072 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
13073 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_kick_vcpu_slowpath);
13074 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_accept_irq);
13075 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
13076 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
13077 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
13078 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
13080 static int __init kvm_x86_init(void)
13082 kvm_mmu_x86_module_init();
13085 module_init(kvm_x86_init);
13087 static void __exit kvm_x86_exit(void)
13090 * If module_init() is implemented, module_exit() must also be
13091 * implemented to allow module unload.
13094 module_exit(kvm_x86_exit);