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 emul_to_vcpu(ctxt) \
94 ((struct kvm_vcpu *)(ctxt)->vcpu)
97 * - enable syscall per default because its emulated by KVM
98 * - enable LME and LMA per default on 64 bit KVM
102 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
104 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
107 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
109 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
111 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
112 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
114 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
115 static void process_nmi(struct kvm_vcpu *vcpu);
116 static void process_smi(struct kvm_vcpu *vcpu);
117 static void enter_smm(struct kvm_vcpu *vcpu);
118 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
119 static void store_regs(struct kvm_vcpu *vcpu);
120 static int sync_regs(struct kvm_vcpu *vcpu);
121 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu);
123 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
124 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
126 struct kvm_x86_ops kvm_x86_ops __read_mostly;
127 EXPORT_SYMBOL_GPL(kvm_x86_ops);
129 #define KVM_X86_OP(func) \
130 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
131 *(((struct kvm_x86_ops *)0)->func));
132 #define KVM_X86_OP_NULL KVM_X86_OP
133 #include <asm/kvm-x86-ops.h>
134 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
135 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
136 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
138 static bool __read_mostly ignore_msrs = 0;
139 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
141 bool __read_mostly report_ignored_msrs = true;
142 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
143 EXPORT_SYMBOL_GPL(report_ignored_msrs);
145 unsigned int min_timer_period_us = 200;
146 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
148 static bool __read_mostly kvmclock_periodic_sync = true;
149 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
151 bool __read_mostly kvm_has_tsc_control;
152 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
153 u32 __read_mostly kvm_max_guest_tsc_khz;
154 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
155 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
156 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
157 u64 __read_mostly kvm_max_tsc_scaling_ratio;
158 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
159 u64 __read_mostly kvm_default_tsc_scaling_ratio;
160 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
161 bool __read_mostly kvm_has_bus_lock_exit;
162 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
164 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
165 static u32 __read_mostly tsc_tolerance_ppm = 250;
166 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
169 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
170 * adaptive tuning starting from default advancement of 1000ns. '0' disables
171 * advancement entirely. Any other value is used as-is and disables adaptive
172 * tuning, i.e. allows privileged userspace to set an exact advancement time.
174 static int __read_mostly lapic_timer_advance_ns = -1;
175 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
177 static bool __read_mostly vector_hashing = true;
178 module_param(vector_hashing, bool, S_IRUGO);
180 bool __read_mostly enable_vmware_backdoor = false;
181 module_param(enable_vmware_backdoor, bool, S_IRUGO);
182 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
184 static bool __read_mostly force_emulation_prefix = false;
185 module_param(force_emulation_prefix, bool, S_IRUGO);
187 int __read_mostly pi_inject_timer = -1;
188 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
191 * Restoring the host value for MSRs that are only consumed when running in
192 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
193 * returns to userspace, i.e. the kernel can run with the guest's value.
195 #define KVM_MAX_NR_USER_RETURN_MSRS 16
197 struct kvm_user_return_msrs {
198 struct user_return_notifier urn;
200 struct kvm_user_return_msr_values {
203 } values[KVM_MAX_NR_USER_RETURN_MSRS];
206 u32 __read_mostly kvm_nr_uret_msrs;
207 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
208 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
209 static struct kvm_user_return_msrs __percpu *user_return_msrs;
211 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
212 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
213 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
214 | XFEATURE_MASK_PKRU)
216 u64 __read_mostly host_efer;
217 EXPORT_SYMBOL_GPL(host_efer);
219 bool __read_mostly allow_smaller_maxphyaddr = 0;
220 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
222 bool __read_mostly enable_apicv = true;
223 EXPORT_SYMBOL_GPL(enable_apicv);
225 u64 __read_mostly host_xss;
226 EXPORT_SYMBOL_GPL(host_xss);
227 u64 __read_mostly supported_xss;
228 EXPORT_SYMBOL_GPL(supported_xss);
230 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
231 KVM_GENERIC_VM_STATS(),
232 STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
233 STATS_DESC_COUNTER(VM, mmu_pte_write),
234 STATS_DESC_COUNTER(VM, mmu_pde_zapped),
235 STATS_DESC_COUNTER(VM, mmu_flooded),
236 STATS_DESC_COUNTER(VM, mmu_recycled),
237 STATS_DESC_COUNTER(VM, mmu_cache_miss),
238 STATS_DESC_ICOUNTER(VM, mmu_unsync),
239 STATS_DESC_ICOUNTER(VM, pages_4k),
240 STATS_DESC_ICOUNTER(VM, pages_2m),
241 STATS_DESC_ICOUNTER(VM, pages_1g),
242 STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
243 STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
244 STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
247 const struct kvm_stats_header kvm_vm_stats_header = {
248 .name_size = KVM_STATS_NAME_SIZE,
249 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
250 .id_offset = sizeof(struct kvm_stats_header),
251 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
252 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
253 sizeof(kvm_vm_stats_desc),
256 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
257 KVM_GENERIC_VCPU_STATS(),
258 STATS_DESC_COUNTER(VCPU, pf_fixed),
259 STATS_DESC_COUNTER(VCPU, pf_guest),
260 STATS_DESC_COUNTER(VCPU, tlb_flush),
261 STATS_DESC_COUNTER(VCPU, invlpg),
262 STATS_DESC_COUNTER(VCPU, exits),
263 STATS_DESC_COUNTER(VCPU, io_exits),
264 STATS_DESC_COUNTER(VCPU, mmio_exits),
265 STATS_DESC_COUNTER(VCPU, signal_exits),
266 STATS_DESC_COUNTER(VCPU, irq_window_exits),
267 STATS_DESC_COUNTER(VCPU, nmi_window_exits),
268 STATS_DESC_COUNTER(VCPU, l1d_flush),
269 STATS_DESC_COUNTER(VCPU, halt_exits),
270 STATS_DESC_COUNTER(VCPU, request_irq_exits),
271 STATS_DESC_COUNTER(VCPU, irq_exits),
272 STATS_DESC_COUNTER(VCPU, host_state_reload),
273 STATS_DESC_COUNTER(VCPU, fpu_reload),
274 STATS_DESC_COUNTER(VCPU, insn_emulation),
275 STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
276 STATS_DESC_COUNTER(VCPU, hypercalls),
277 STATS_DESC_COUNTER(VCPU, irq_injections),
278 STATS_DESC_COUNTER(VCPU, nmi_injections),
279 STATS_DESC_COUNTER(VCPU, req_event),
280 STATS_DESC_COUNTER(VCPU, nested_run),
281 STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
282 STATS_DESC_COUNTER(VCPU, directed_yield_successful),
283 STATS_DESC_ICOUNTER(VCPU, guest_mode)
286 const struct kvm_stats_header kvm_vcpu_stats_header = {
287 .name_size = KVM_STATS_NAME_SIZE,
288 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
289 .id_offset = sizeof(struct kvm_stats_header),
290 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
291 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
292 sizeof(kvm_vcpu_stats_desc),
295 u64 __read_mostly host_xcr0;
296 u64 __read_mostly supported_xcr0;
297 EXPORT_SYMBOL_GPL(supported_xcr0);
299 static struct kmem_cache *x86_emulator_cache;
302 * When called, it means the previous get/set msr reached an invalid msr.
303 * Return true if we want to ignore/silent this failed msr access.
305 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
307 const char *op = write ? "wrmsr" : "rdmsr";
310 if (report_ignored_msrs)
311 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
316 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
322 static struct kmem_cache *kvm_alloc_emulator_cache(void)
324 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
325 unsigned int size = sizeof(struct x86_emulate_ctxt);
327 return kmem_cache_create_usercopy("x86_emulator", size,
328 __alignof__(struct x86_emulate_ctxt),
329 SLAB_ACCOUNT, useroffset,
330 size - useroffset, NULL);
333 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
335 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
338 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
339 vcpu->arch.apf.gfns[i] = ~0;
342 static void kvm_on_user_return(struct user_return_notifier *urn)
345 struct kvm_user_return_msrs *msrs
346 = container_of(urn, struct kvm_user_return_msrs, urn);
347 struct kvm_user_return_msr_values *values;
351 * Disabling irqs at this point since the following code could be
352 * interrupted and executed through kvm_arch_hardware_disable()
354 local_irq_save(flags);
355 if (msrs->registered) {
356 msrs->registered = false;
357 user_return_notifier_unregister(urn);
359 local_irq_restore(flags);
360 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
361 values = &msrs->values[slot];
362 if (values->host != values->curr) {
363 wrmsrl(kvm_uret_msrs_list[slot], values->host);
364 values->curr = values->host;
369 static int kvm_probe_user_return_msr(u32 msr)
375 ret = rdmsrl_safe(msr, &val);
378 ret = wrmsrl_safe(msr, val);
384 int kvm_add_user_return_msr(u32 msr)
386 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
388 if (kvm_probe_user_return_msr(msr))
391 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
392 return kvm_nr_uret_msrs++;
394 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
396 int kvm_find_user_return_msr(u32 msr)
400 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
401 if (kvm_uret_msrs_list[i] == msr)
406 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
408 static void kvm_user_return_msr_cpu_online(void)
410 unsigned int cpu = smp_processor_id();
411 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
415 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
416 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
417 msrs->values[i].host = value;
418 msrs->values[i].curr = value;
422 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
424 unsigned int cpu = smp_processor_id();
425 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
428 value = (value & mask) | (msrs->values[slot].host & ~mask);
429 if (value == msrs->values[slot].curr)
431 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
435 msrs->values[slot].curr = value;
436 if (!msrs->registered) {
437 msrs->urn.on_user_return = kvm_on_user_return;
438 user_return_notifier_register(&msrs->urn);
439 msrs->registered = true;
443 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
445 static void drop_user_return_notifiers(void)
447 unsigned int cpu = smp_processor_id();
448 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
450 if (msrs->registered)
451 kvm_on_user_return(&msrs->urn);
454 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
456 return vcpu->arch.apic_base;
458 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
460 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
462 return kvm_apic_mode(kvm_get_apic_base(vcpu));
464 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
466 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
468 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
469 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
470 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
471 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
473 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
475 if (!msr_info->host_initiated) {
476 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
478 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
482 kvm_lapic_set_base(vcpu, msr_info->data);
483 kvm_recalculate_apic_map(vcpu->kvm);
486 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
489 * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
491 * Hardware virtualization extension instructions may fault if a reboot turns
492 * off virtualization while processes are running. Usually after catching the
493 * fault we just panic; during reboot instead the instruction is ignored.
495 noinstr void kvm_spurious_fault(void)
497 /* Fault while not rebooting. We want the trace. */
498 BUG_ON(!kvm_rebooting);
500 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
502 #define EXCPT_BENIGN 0
503 #define EXCPT_CONTRIBUTORY 1
506 static int exception_class(int vector)
516 return EXCPT_CONTRIBUTORY;
523 #define EXCPT_FAULT 0
525 #define EXCPT_ABORT 2
526 #define EXCPT_INTERRUPT 3
528 static int exception_type(int vector)
532 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
533 return EXCPT_INTERRUPT;
537 /* #DB is trap, as instruction watchpoints are handled elsewhere */
538 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
541 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
544 /* Reserved exceptions will result in fault */
548 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
550 unsigned nr = vcpu->arch.exception.nr;
551 bool has_payload = vcpu->arch.exception.has_payload;
552 unsigned long payload = vcpu->arch.exception.payload;
560 * "Certain debug exceptions may clear bit 0-3. The
561 * remaining contents of the DR6 register are never
562 * cleared by the processor".
564 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
566 * In order to reflect the #DB exception payload in guest
567 * dr6, three components need to be considered: active low
568 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
570 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
571 * In the target guest dr6:
572 * FIXED_1 bits should always be set.
573 * Active low bits should be cleared if 1-setting in payload.
574 * Active high bits should be set if 1-setting in payload.
576 * Note, the payload is compatible with the pending debug
577 * exceptions/exit qualification under VMX, that active_low bits
578 * are active high in payload.
579 * So they need to be flipped for DR6.
581 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
582 vcpu->arch.dr6 |= payload;
583 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
586 * The #DB payload is defined as compatible with the 'pending
587 * debug exceptions' field under VMX, not DR6. While bit 12 is
588 * defined in the 'pending debug exceptions' field (enabled
589 * breakpoint), it is reserved and must be zero in DR6.
591 vcpu->arch.dr6 &= ~BIT(12);
594 vcpu->arch.cr2 = payload;
598 vcpu->arch.exception.has_payload = false;
599 vcpu->arch.exception.payload = 0;
601 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
603 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
604 unsigned nr, bool has_error, u32 error_code,
605 bool has_payload, unsigned long payload, bool reinject)
610 kvm_make_request(KVM_REQ_EVENT, vcpu);
612 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
616 * On vmentry, vcpu->arch.exception.pending is only
617 * true if an event injection was blocked by
618 * nested_run_pending. In that case, however,
619 * vcpu_enter_guest requests an immediate exit,
620 * and the guest shouldn't proceed far enough to
623 WARN_ON_ONCE(vcpu->arch.exception.pending);
624 vcpu->arch.exception.injected = true;
625 if (WARN_ON_ONCE(has_payload)) {
627 * A reinjected event has already
628 * delivered its payload.
634 vcpu->arch.exception.pending = true;
635 vcpu->arch.exception.injected = false;
637 vcpu->arch.exception.has_error_code = has_error;
638 vcpu->arch.exception.nr = nr;
639 vcpu->arch.exception.error_code = error_code;
640 vcpu->arch.exception.has_payload = has_payload;
641 vcpu->arch.exception.payload = payload;
642 if (!is_guest_mode(vcpu))
643 kvm_deliver_exception_payload(vcpu);
647 /* to check exception */
648 prev_nr = vcpu->arch.exception.nr;
649 if (prev_nr == DF_VECTOR) {
650 /* triple fault -> shutdown */
651 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
654 class1 = exception_class(prev_nr);
655 class2 = exception_class(nr);
656 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
657 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
659 * Generate double fault per SDM Table 5-5. Set
660 * exception.pending = true so that the double fault
661 * can trigger a nested vmexit.
663 vcpu->arch.exception.pending = true;
664 vcpu->arch.exception.injected = false;
665 vcpu->arch.exception.has_error_code = true;
666 vcpu->arch.exception.nr = DF_VECTOR;
667 vcpu->arch.exception.error_code = 0;
668 vcpu->arch.exception.has_payload = false;
669 vcpu->arch.exception.payload = 0;
671 /* replace previous exception with a new one in a hope
672 that instruction re-execution will regenerate lost
677 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
679 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
681 EXPORT_SYMBOL_GPL(kvm_queue_exception);
683 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
685 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
687 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
689 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
690 unsigned long payload)
692 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
694 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
696 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
697 u32 error_code, unsigned long payload)
699 kvm_multiple_exception(vcpu, nr, true, error_code,
700 true, payload, false);
703 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
706 kvm_inject_gp(vcpu, 0);
708 return kvm_skip_emulated_instruction(vcpu);
712 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
714 static int complete_emulated_insn_gp(struct kvm_vcpu *vcpu, int err)
717 kvm_inject_gp(vcpu, 0);
721 return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE | EMULTYPE_SKIP |
722 EMULTYPE_COMPLETE_USER_EXIT);
725 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
727 ++vcpu->stat.pf_guest;
728 vcpu->arch.exception.nested_apf =
729 is_guest_mode(vcpu) && fault->async_page_fault;
730 if (vcpu->arch.exception.nested_apf) {
731 vcpu->arch.apf.nested_apf_token = fault->address;
732 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
734 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
738 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
740 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
741 struct x86_exception *fault)
743 struct kvm_mmu *fault_mmu;
744 WARN_ON_ONCE(fault->vector != PF_VECTOR);
746 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
750 * Invalidate the TLB entry for the faulting address, if it exists,
751 * else the access will fault indefinitely (and to emulate hardware).
753 if ((fault->error_code & PFERR_PRESENT_MASK) &&
754 !(fault->error_code & PFERR_RSVD_MASK))
755 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
756 fault_mmu->root_hpa);
758 fault_mmu->inject_page_fault(vcpu, fault);
759 return fault->nested_page_fault;
761 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
763 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
765 atomic_inc(&vcpu->arch.nmi_queued);
766 kvm_make_request(KVM_REQ_NMI, vcpu);
768 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
770 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
772 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
774 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
776 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
778 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
780 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
783 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
784 * a #GP and return false.
786 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
788 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
790 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
793 EXPORT_SYMBOL_GPL(kvm_require_cpl);
795 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
797 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
800 kvm_queue_exception(vcpu, UD_VECTOR);
803 EXPORT_SYMBOL_GPL(kvm_require_dr);
805 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
807 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
811 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
813 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
815 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
816 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
820 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
823 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
826 real_gpa = kvm_translate_gpa(vcpu, mmu, gfn_to_gpa(pdpt_gfn),
827 PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
828 if (real_gpa == UNMAPPED_GVA)
831 /* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
832 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
833 cr3 & GENMASK(11, 5), sizeof(pdpte));
837 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
838 if ((pdpte[i] & PT_PRESENT_MASK) &&
839 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
845 * Marking VCPU_EXREG_PDPTR dirty doesn't work for !tdp_enabled.
846 * Shadow page roots need to be reconstructed instead.
848 if (!tdp_enabled && memcmp(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)))
849 kvm_mmu_free_roots(vcpu, mmu, KVM_MMU_ROOT_CURRENT);
851 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
852 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
853 kvm_make_request(KVM_REQ_LOAD_MMU_PGD, vcpu);
854 vcpu->arch.pdptrs_from_userspace = false;
858 EXPORT_SYMBOL_GPL(load_pdptrs);
860 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
862 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
863 kvm_clear_async_pf_completion_queue(vcpu);
864 kvm_async_pf_hash_reset(vcpu);
867 if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
868 kvm_mmu_reset_context(vcpu);
870 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
871 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
872 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
873 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
875 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
877 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
879 unsigned long old_cr0 = kvm_read_cr0(vcpu);
884 if (cr0 & 0xffffffff00000000UL)
888 cr0 &= ~CR0_RESERVED_BITS;
890 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
893 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
897 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
898 (cr0 & X86_CR0_PG)) {
903 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
908 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
909 is_pae(vcpu) && ((cr0 ^ old_cr0) & X86_CR0_PDPTR_BITS) &&
910 !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
913 if (!(cr0 & X86_CR0_PG) &&
914 (is_64_bit_mode(vcpu) || kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)))
917 static_call(kvm_x86_set_cr0)(vcpu, cr0);
919 kvm_post_set_cr0(vcpu, old_cr0, cr0);
923 EXPORT_SYMBOL_GPL(kvm_set_cr0);
925 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
927 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
929 EXPORT_SYMBOL_GPL(kvm_lmsw);
931 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
933 if (vcpu->arch.guest_state_protected)
936 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
938 if (vcpu->arch.xcr0 != host_xcr0)
939 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
941 if (vcpu->arch.xsaves_enabled &&
942 vcpu->arch.ia32_xss != host_xss)
943 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
946 if (static_cpu_has(X86_FEATURE_PKU) &&
947 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
948 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
949 vcpu->arch.pkru != vcpu->arch.host_pkru)
950 write_pkru(vcpu->arch.pkru);
952 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
954 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
956 if (vcpu->arch.guest_state_protected)
959 if (static_cpu_has(X86_FEATURE_PKU) &&
960 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
961 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
962 vcpu->arch.pkru = rdpkru();
963 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
964 write_pkru(vcpu->arch.host_pkru);
967 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
969 if (vcpu->arch.xcr0 != host_xcr0)
970 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
972 if (vcpu->arch.xsaves_enabled &&
973 vcpu->arch.ia32_xss != host_xss)
974 wrmsrl(MSR_IA32_XSS, host_xss);
978 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
980 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
983 u64 old_xcr0 = vcpu->arch.xcr0;
986 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
987 if (index != XCR_XFEATURE_ENABLED_MASK)
989 if (!(xcr0 & XFEATURE_MASK_FP))
991 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
995 * Do not allow the guest to set bits that we do not support
996 * saving. However, xcr0 bit 0 is always set, even if the
997 * emulated CPU does not support XSAVE (see kvm_vcpu_reset()).
999 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
1000 if (xcr0 & ~valid_bits)
1003 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
1004 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1007 if (xcr0 & XFEATURE_MASK_AVX512) {
1008 if (!(xcr0 & XFEATURE_MASK_YMM))
1010 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1013 vcpu->arch.xcr0 = xcr0;
1015 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1016 kvm_update_cpuid_runtime(vcpu);
1020 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1022 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1023 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1024 kvm_inject_gp(vcpu, 0);
1028 return kvm_skip_emulated_instruction(vcpu);
1030 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1032 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1034 if (cr4 & cr4_reserved_bits)
1037 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1040 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1042 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1044 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1047 * If any role bit is changed, the MMU needs to be reset.
1049 * If CR4.PCIDE is changed 1 -> 0, the guest TLB must be flushed.
1050 * If CR4.PCIDE is changed 0 -> 1, there is no need to flush the TLB
1051 * according to the SDM; however, stale prev_roots could be reused
1052 * incorrectly in the future after a MOV to CR3 with NOFLUSH=1, so we
1053 * free them all. KVM_REQ_MMU_RELOAD is fit for the both cases; it
1054 * is slow, but changing CR4.PCIDE is a rare case.
1056 * If CR4.PGE is changed, the guest TLB must be flushed.
1058 * Note: resetting MMU is a superset of KVM_REQ_MMU_RELOAD and
1059 * KVM_REQ_MMU_RELOAD is a superset of KVM_REQ_TLB_FLUSH_GUEST, hence
1060 * the usage of "else if".
1062 if ((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS)
1063 kvm_mmu_reset_context(vcpu);
1064 else if ((cr4 ^ old_cr4) & X86_CR4_PCIDE)
1065 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
1066 else if ((cr4 ^ old_cr4) & X86_CR4_PGE)
1067 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1069 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1071 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1073 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1075 if (!kvm_is_valid_cr4(vcpu, cr4))
1078 if (is_long_mode(vcpu)) {
1079 if (!(cr4 & X86_CR4_PAE))
1081 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1083 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1084 && ((cr4 ^ old_cr4) & X86_CR4_PDPTR_BITS)
1085 && !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
1088 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1089 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1092 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1093 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1097 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1099 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1103 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1105 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1107 struct kvm_mmu *mmu = vcpu->arch.mmu;
1108 unsigned long roots_to_free = 0;
1112 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1113 * this is reachable when running EPT=1 and unrestricted_guest=0, and
1114 * also via the emulator. KVM's TDP page tables are not in the scope of
1115 * the invalidation, but the guest's TLB entries need to be flushed as
1116 * the CPU may have cached entries in its TLB for the target PCID.
1118 if (unlikely(tdp_enabled)) {
1119 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1124 * If neither the current CR3 nor any of the prev_roots use the given
1125 * PCID, then nothing needs to be done here because a resync will
1126 * happen anyway before switching to any other CR3.
1128 if (kvm_get_active_pcid(vcpu) == pcid) {
1129 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1130 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1134 * If PCID is disabled, there is no need to free prev_roots even if the
1135 * PCIDs for them are also 0, because MOV to CR3 always flushes the TLB
1138 if (!kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
1141 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1142 if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1143 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1145 kvm_mmu_free_roots(vcpu, mmu, roots_to_free);
1148 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1150 bool skip_tlb_flush = false;
1151 unsigned long pcid = 0;
1152 #ifdef CONFIG_X86_64
1153 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1156 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1157 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1158 pcid = cr3 & X86_CR3_PCID_MASK;
1162 /* PDPTRs are always reloaded for PAE paging. */
1163 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1164 goto handle_tlb_flush;
1167 * Do not condition the GPA check on long mode, this helper is used to
1168 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1169 * the current vCPU mode is accurate.
1171 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1174 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, cr3))
1177 if (cr3 != kvm_read_cr3(vcpu))
1178 kvm_mmu_new_pgd(vcpu, cr3);
1180 vcpu->arch.cr3 = cr3;
1181 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1182 /* Do not call post_set_cr3, we do not get here for confidential guests. */
1186 * A load of CR3 that flushes the TLB flushes only the current PCID,
1187 * even if PCID is disabled, in which case PCID=0 is flushed. It's a
1188 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1189 * and it's impossible to use a non-zero PCID when PCID is disabled,
1190 * i.e. only PCID=0 can be relevant.
1192 if (!skip_tlb_flush)
1193 kvm_invalidate_pcid(vcpu, pcid);
1197 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1199 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1201 if (cr8 & CR8_RESERVED_BITS)
1203 if (lapic_in_kernel(vcpu))
1204 kvm_lapic_set_tpr(vcpu, cr8);
1206 vcpu->arch.cr8 = cr8;
1209 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1211 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1213 if (lapic_in_kernel(vcpu))
1214 return kvm_lapic_get_cr8(vcpu);
1216 return vcpu->arch.cr8;
1218 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1220 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1224 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1225 for (i = 0; i < KVM_NR_DB_REGS; i++)
1226 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1230 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1234 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1235 dr7 = vcpu->arch.guest_debug_dr7;
1237 dr7 = vcpu->arch.dr7;
1238 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1239 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1240 if (dr7 & DR7_BP_EN_MASK)
1241 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1243 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1245 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1247 u64 fixed = DR6_FIXED_1;
1249 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1252 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1253 fixed |= DR6_BUS_LOCK;
1257 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1259 size_t size = ARRAY_SIZE(vcpu->arch.db);
1263 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1264 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1265 vcpu->arch.eff_db[dr] = val;
1269 if (!kvm_dr6_valid(val))
1271 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1275 if (!kvm_dr7_valid(val))
1277 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1278 kvm_update_dr7(vcpu);
1284 EXPORT_SYMBOL_GPL(kvm_set_dr);
1286 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1288 size_t size = ARRAY_SIZE(vcpu->arch.db);
1292 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1296 *val = vcpu->arch.dr6;
1300 *val = vcpu->arch.dr7;
1304 EXPORT_SYMBOL_GPL(kvm_get_dr);
1306 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1308 u32 ecx = kvm_rcx_read(vcpu);
1311 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1312 kvm_inject_gp(vcpu, 0);
1316 kvm_rax_write(vcpu, (u32)data);
1317 kvm_rdx_write(vcpu, data >> 32);
1318 return kvm_skip_emulated_instruction(vcpu);
1320 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1323 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1324 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1326 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1327 * extract the supported MSRs from the related const lists.
1328 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1329 * capabilities of the host cpu. This capabilities test skips MSRs that are
1330 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1331 * may depend on host virtualization features rather than host cpu features.
1334 static const u32 msrs_to_save_all[] = {
1335 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1337 #ifdef CONFIG_X86_64
1338 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1340 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1341 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1343 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1344 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1345 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1346 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1347 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1348 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1349 MSR_IA32_UMWAIT_CONTROL,
1351 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1352 MSR_ARCH_PERFMON_FIXED_CTR0 + 2,
1353 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1354 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1355 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1356 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1357 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1358 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1359 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1360 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1361 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1362 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1363 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1364 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1365 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1366 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1367 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1368 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1369 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1370 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1371 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1372 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1374 MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1375 MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1376 MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1377 MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1378 MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1379 MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1382 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1383 static unsigned num_msrs_to_save;
1385 static const u32 emulated_msrs_all[] = {
1386 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1387 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1388 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1389 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1390 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1391 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1392 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1394 HV_X64_MSR_VP_INDEX,
1395 HV_X64_MSR_VP_RUNTIME,
1396 HV_X64_MSR_SCONTROL,
1397 HV_X64_MSR_STIMER0_CONFIG,
1398 HV_X64_MSR_VP_ASSIST_PAGE,
1399 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1400 HV_X64_MSR_TSC_EMULATION_STATUS,
1401 HV_X64_MSR_SYNDBG_OPTIONS,
1402 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1403 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1404 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1406 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1407 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1409 MSR_IA32_TSC_ADJUST,
1410 MSR_IA32_TSC_DEADLINE,
1411 MSR_IA32_ARCH_CAPABILITIES,
1412 MSR_IA32_PERF_CAPABILITIES,
1413 MSR_IA32_MISC_ENABLE,
1414 MSR_IA32_MCG_STATUS,
1416 MSR_IA32_MCG_EXT_CTL,
1420 MSR_MISC_FEATURES_ENABLES,
1421 MSR_AMD64_VIRT_SPEC_CTRL,
1422 MSR_AMD64_TSC_RATIO,
1427 * The following list leaves out MSRs whose values are determined
1428 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1429 * We always support the "true" VMX control MSRs, even if the host
1430 * processor does not, so I am putting these registers here rather
1431 * than in msrs_to_save_all.
1434 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1435 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1436 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1437 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1439 MSR_IA32_VMX_CR0_FIXED0,
1440 MSR_IA32_VMX_CR4_FIXED0,
1441 MSR_IA32_VMX_VMCS_ENUM,
1442 MSR_IA32_VMX_PROCBASED_CTLS2,
1443 MSR_IA32_VMX_EPT_VPID_CAP,
1444 MSR_IA32_VMX_VMFUNC,
1447 MSR_KVM_POLL_CONTROL,
1450 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1451 static unsigned num_emulated_msrs;
1454 * List of msr numbers which are used to expose MSR-based features that
1455 * can be used by a hypervisor to validate requested CPU features.
1457 static const u32 msr_based_features_all[] = {
1459 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1460 MSR_IA32_VMX_PINBASED_CTLS,
1461 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1462 MSR_IA32_VMX_PROCBASED_CTLS,
1463 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1464 MSR_IA32_VMX_EXIT_CTLS,
1465 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1466 MSR_IA32_VMX_ENTRY_CTLS,
1468 MSR_IA32_VMX_CR0_FIXED0,
1469 MSR_IA32_VMX_CR0_FIXED1,
1470 MSR_IA32_VMX_CR4_FIXED0,
1471 MSR_IA32_VMX_CR4_FIXED1,
1472 MSR_IA32_VMX_VMCS_ENUM,
1473 MSR_IA32_VMX_PROCBASED_CTLS2,
1474 MSR_IA32_VMX_EPT_VPID_CAP,
1475 MSR_IA32_VMX_VMFUNC,
1479 MSR_IA32_ARCH_CAPABILITIES,
1480 MSR_IA32_PERF_CAPABILITIES,
1483 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1484 static unsigned int num_msr_based_features;
1486 static u64 kvm_get_arch_capabilities(void)
1490 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1491 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1494 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1495 * the nested hypervisor runs with NX huge pages. If it is not,
1496 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1497 * L1 guests, so it need not worry about its own (L2) guests.
1499 data |= ARCH_CAP_PSCHANGE_MC_NO;
1502 * If we're doing cache flushes (either "always" or "cond")
1503 * we will do one whenever the guest does a vmlaunch/vmresume.
1504 * If an outer hypervisor is doing the cache flush for us
1505 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1506 * capability to the guest too, and if EPT is disabled we're not
1507 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1508 * require a nested hypervisor to do a flush of its own.
1510 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1511 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1513 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1514 data |= ARCH_CAP_RDCL_NO;
1515 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1516 data |= ARCH_CAP_SSB_NO;
1517 if (!boot_cpu_has_bug(X86_BUG_MDS))
1518 data |= ARCH_CAP_MDS_NO;
1520 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1522 * If RTM=0 because the kernel has disabled TSX, the host might
1523 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1524 * and therefore knows that there cannot be TAA) but keep
1525 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1526 * and we want to allow migrating those guests to tsx=off hosts.
1528 data &= ~ARCH_CAP_TAA_NO;
1529 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1530 data |= ARCH_CAP_TAA_NO;
1533 * Nothing to do here; we emulate TSX_CTRL if present on the
1534 * host so the guest can choose between disabling TSX or
1535 * using VERW to clear CPU buffers.
1542 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1544 switch (msr->index) {
1545 case MSR_IA32_ARCH_CAPABILITIES:
1546 msr->data = kvm_get_arch_capabilities();
1548 case MSR_IA32_UCODE_REV:
1549 rdmsrl_safe(msr->index, &msr->data);
1552 return static_call(kvm_x86_get_msr_feature)(msr);
1557 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1559 struct kvm_msr_entry msr;
1563 r = kvm_get_msr_feature(&msr);
1565 if (r == KVM_MSR_RET_INVALID) {
1566 /* Unconditionally clear the output for simplicity */
1568 if (kvm_msr_ignored_check(index, 0, false))
1580 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1582 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1585 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1588 if (efer & (EFER_LME | EFER_LMA) &&
1589 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1592 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1598 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1600 if (efer & efer_reserved_bits)
1603 return __kvm_valid_efer(vcpu, efer);
1605 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1607 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1609 u64 old_efer = vcpu->arch.efer;
1610 u64 efer = msr_info->data;
1613 if (efer & efer_reserved_bits)
1616 if (!msr_info->host_initiated) {
1617 if (!__kvm_valid_efer(vcpu, efer))
1620 if (is_paging(vcpu) &&
1621 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1626 efer |= vcpu->arch.efer & EFER_LMA;
1628 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1634 /* Update reserved bits */
1635 if ((efer ^ old_efer) & EFER_NX)
1636 kvm_mmu_reset_context(vcpu);
1641 void kvm_enable_efer_bits(u64 mask)
1643 efer_reserved_bits &= ~mask;
1645 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1647 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1649 struct kvm_x86_msr_filter *msr_filter;
1650 struct msr_bitmap_range *ranges;
1651 struct kvm *kvm = vcpu->kvm;
1656 /* x2APIC MSRs do not support filtering. */
1657 if (index >= 0x800 && index <= 0x8ff)
1660 idx = srcu_read_lock(&kvm->srcu);
1662 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1668 allowed = msr_filter->default_allow;
1669 ranges = msr_filter->ranges;
1671 for (i = 0; i < msr_filter->count; i++) {
1672 u32 start = ranges[i].base;
1673 u32 end = start + ranges[i].nmsrs;
1674 u32 flags = ranges[i].flags;
1675 unsigned long *bitmap = ranges[i].bitmap;
1677 if ((index >= start) && (index < end) && (flags & type)) {
1678 allowed = !!test_bit(index - start, bitmap);
1684 srcu_read_unlock(&kvm->srcu, idx);
1688 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1691 * Write @data into the MSR specified by @index. Select MSR specific fault
1692 * checks are bypassed if @host_initiated is %true.
1693 * Returns 0 on success, non-0 otherwise.
1694 * Assumes vcpu_load() was already called.
1696 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1697 bool host_initiated)
1699 struct msr_data msr;
1701 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1702 return KVM_MSR_RET_FILTERED;
1707 case MSR_KERNEL_GS_BASE:
1710 if (is_noncanonical_address(data, vcpu))
1713 case MSR_IA32_SYSENTER_EIP:
1714 case MSR_IA32_SYSENTER_ESP:
1716 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1717 * non-canonical address is written on Intel but not on
1718 * AMD (which ignores the top 32-bits, because it does
1719 * not implement 64-bit SYSENTER).
1721 * 64-bit code should hence be able to write a non-canonical
1722 * value on AMD. Making the address canonical ensures that
1723 * vmentry does not fail on Intel after writing a non-canonical
1724 * value, and that something deterministic happens if the guest
1725 * invokes 64-bit SYSENTER.
1727 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1730 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1733 if (!host_initiated &&
1734 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1735 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1739 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1740 * incomplete and conflicting architectural behavior. Current
1741 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1742 * reserved and always read as zeros. Enforce Intel's reserved
1743 * bits check if and only if the guest CPU is Intel, and clear
1744 * the bits in all other cases. This ensures cross-vendor
1745 * migration will provide consistent behavior for the guest.
1747 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1756 msr.host_initiated = host_initiated;
1758 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1761 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1762 u32 index, u64 data, bool host_initiated)
1764 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1766 if (ret == KVM_MSR_RET_INVALID)
1767 if (kvm_msr_ignored_check(index, data, true))
1774 * Read the MSR specified by @index into @data. Select MSR specific fault
1775 * checks are bypassed if @host_initiated is %true.
1776 * Returns 0 on success, non-0 otherwise.
1777 * Assumes vcpu_load() was already called.
1779 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1780 bool host_initiated)
1782 struct msr_data msr;
1785 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1786 return KVM_MSR_RET_FILTERED;
1790 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1793 if (!host_initiated &&
1794 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1795 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1801 msr.host_initiated = host_initiated;
1803 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1809 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1810 u32 index, u64 *data, bool host_initiated)
1812 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1814 if (ret == KVM_MSR_RET_INVALID) {
1815 /* Unconditionally clear *data for simplicity */
1817 if (kvm_msr_ignored_check(index, 0, false))
1824 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1826 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1828 EXPORT_SYMBOL_GPL(kvm_get_msr);
1830 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1832 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1834 EXPORT_SYMBOL_GPL(kvm_set_msr);
1836 static void complete_userspace_rdmsr(struct kvm_vcpu *vcpu)
1838 if (!vcpu->run->msr.error) {
1839 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1840 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1844 static int complete_emulated_msr_access(struct kvm_vcpu *vcpu)
1846 return complete_emulated_insn_gp(vcpu, vcpu->run->msr.error);
1849 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1851 complete_userspace_rdmsr(vcpu);
1852 return complete_emulated_msr_access(vcpu);
1855 static int complete_fast_msr_access(struct kvm_vcpu *vcpu)
1857 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1860 static int complete_fast_rdmsr(struct kvm_vcpu *vcpu)
1862 complete_userspace_rdmsr(vcpu);
1863 return complete_fast_msr_access(vcpu);
1866 static u64 kvm_msr_reason(int r)
1869 case KVM_MSR_RET_INVALID:
1870 return KVM_MSR_EXIT_REASON_UNKNOWN;
1871 case KVM_MSR_RET_FILTERED:
1872 return KVM_MSR_EXIT_REASON_FILTER;
1874 return KVM_MSR_EXIT_REASON_INVAL;
1878 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1879 u32 exit_reason, u64 data,
1880 int (*completion)(struct kvm_vcpu *vcpu),
1883 u64 msr_reason = kvm_msr_reason(r);
1885 /* Check if the user wanted to know about this MSR fault */
1886 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1889 vcpu->run->exit_reason = exit_reason;
1890 vcpu->run->msr.error = 0;
1891 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1892 vcpu->run->msr.reason = msr_reason;
1893 vcpu->run->msr.index = index;
1894 vcpu->run->msr.data = data;
1895 vcpu->arch.complete_userspace_io = completion;
1900 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1902 u32 ecx = kvm_rcx_read(vcpu);
1906 r = kvm_get_msr(vcpu, ecx, &data);
1909 trace_kvm_msr_read(ecx, data);
1911 kvm_rax_write(vcpu, data & -1u);
1912 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1914 /* MSR read failed? See if we should ask user space */
1915 if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_RDMSR, 0,
1916 complete_fast_rdmsr, r))
1918 trace_kvm_msr_read_ex(ecx);
1921 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1923 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1925 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1927 u32 ecx = kvm_rcx_read(vcpu);
1928 u64 data = kvm_read_edx_eax(vcpu);
1931 r = kvm_set_msr(vcpu, ecx, data);
1934 trace_kvm_msr_write(ecx, data);
1936 /* MSR write failed? See if we should ask user space */
1937 if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_WRMSR, data,
1938 complete_fast_msr_access, r))
1940 /* Signal all other negative errors to userspace */
1943 trace_kvm_msr_write_ex(ecx, data);
1946 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1948 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1950 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1952 return kvm_skip_emulated_instruction(vcpu);
1954 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1956 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1958 /* Treat an INVD instruction as a NOP and just skip it. */
1959 return kvm_emulate_as_nop(vcpu);
1961 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1963 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1965 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1966 return kvm_emulate_as_nop(vcpu);
1968 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1970 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1972 kvm_queue_exception(vcpu, UD_VECTOR);
1975 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1977 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1979 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1980 return kvm_emulate_as_nop(vcpu);
1982 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1984 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1986 xfer_to_guest_mode_prepare();
1987 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1988 xfer_to_guest_mode_work_pending();
1992 * The fast path for frequent and performance sensitive wrmsr emulation,
1993 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1994 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1995 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1996 * other cases which must be called after interrupts are enabled on the host.
1998 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
2000 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
2003 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
2004 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
2005 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
2006 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
2009 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
2010 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
2011 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
2012 trace_kvm_apic_write(APIC_ICR, (u32)data);
2019 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2021 if (!kvm_can_use_hv_timer(vcpu))
2024 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2028 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2030 u32 msr = kvm_rcx_read(vcpu);
2032 fastpath_t ret = EXIT_FASTPATH_NONE;
2035 case APIC_BASE_MSR + (APIC_ICR >> 4):
2036 data = kvm_read_edx_eax(vcpu);
2037 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2038 kvm_skip_emulated_instruction(vcpu);
2039 ret = EXIT_FASTPATH_EXIT_HANDLED;
2042 case MSR_IA32_TSC_DEADLINE:
2043 data = kvm_read_edx_eax(vcpu);
2044 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2045 kvm_skip_emulated_instruction(vcpu);
2046 ret = EXIT_FASTPATH_REENTER_GUEST;
2053 if (ret != EXIT_FASTPATH_NONE)
2054 trace_kvm_msr_write(msr, data);
2058 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2061 * Adapt set_msr() to msr_io()'s calling convention
2063 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2065 return kvm_get_msr_ignored_check(vcpu, index, data, true);
2068 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2070 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2073 #ifdef CONFIG_X86_64
2074 struct pvclock_clock {
2084 struct pvclock_gtod_data {
2087 struct pvclock_clock clock; /* extract of a clocksource struct */
2088 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2094 static struct pvclock_gtod_data pvclock_gtod_data;
2096 static void update_pvclock_gtod(struct timekeeper *tk)
2098 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2100 write_seqcount_begin(&vdata->seq);
2102 /* copy pvclock gtod data */
2103 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2104 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2105 vdata->clock.mask = tk->tkr_mono.mask;
2106 vdata->clock.mult = tk->tkr_mono.mult;
2107 vdata->clock.shift = tk->tkr_mono.shift;
2108 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2109 vdata->clock.offset = tk->tkr_mono.base;
2111 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2112 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2113 vdata->raw_clock.mask = tk->tkr_raw.mask;
2114 vdata->raw_clock.mult = tk->tkr_raw.mult;
2115 vdata->raw_clock.shift = tk->tkr_raw.shift;
2116 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2117 vdata->raw_clock.offset = tk->tkr_raw.base;
2119 vdata->wall_time_sec = tk->xtime_sec;
2121 vdata->offs_boot = tk->offs_boot;
2123 write_seqcount_end(&vdata->seq);
2126 static s64 get_kvmclock_base_ns(void)
2128 /* Count up from boot time, but with the frequency of the raw clock. */
2129 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2132 static s64 get_kvmclock_base_ns(void)
2134 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2135 return ktime_get_boottime_ns();
2139 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2143 struct pvclock_wall_clock wc;
2150 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2155 ++version; /* first time write, random junk */
2159 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2163 * The guest calculates current wall clock time by adding
2164 * system time (updated by kvm_guest_time_update below) to the
2165 * wall clock specified here. We do the reverse here.
2167 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2169 wc.nsec = do_div(wall_nsec, 1000000000);
2170 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2171 wc.version = version;
2173 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2176 wc_sec_hi = wall_nsec >> 32;
2177 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2178 &wc_sec_hi, sizeof(wc_sec_hi));
2182 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2185 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2186 bool old_msr, bool host_initiated)
2188 struct kvm_arch *ka = &vcpu->kvm->arch;
2190 if (vcpu->vcpu_id == 0 && !host_initiated) {
2191 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2192 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2194 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2197 vcpu->arch.time = system_time;
2198 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2200 /* we verify if the enable bit is set... */
2201 vcpu->arch.pv_time_enabled = false;
2202 if (!(system_time & 1))
2205 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2206 &vcpu->arch.pv_time, system_time & ~1ULL,
2207 sizeof(struct pvclock_vcpu_time_info)))
2208 vcpu->arch.pv_time_enabled = true;
2213 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2215 do_shl32_div32(dividend, divisor);
2219 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2220 s8 *pshift, u32 *pmultiplier)
2228 scaled64 = scaled_hz;
2229 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2234 tps32 = (uint32_t)tps64;
2235 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2236 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2244 *pmultiplier = div_frac(scaled64, tps32);
2247 #ifdef CONFIG_X86_64
2248 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2251 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2252 static unsigned long max_tsc_khz;
2254 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2256 u64 v = (u64)khz * (1000000 + ppm);
2261 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2263 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2267 /* Guest TSC same frequency as host TSC? */
2269 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2273 /* TSC scaling supported? */
2274 if (!kvm_has_tsc_control) {
2275 if (user_tsc_khz > tsc_khz) {
2276 vcpu->arch.tsc_catchup = 1;
2277 vcpu->arch.tsc_always_catchup = 1;
2280 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2285 /* TSC scaling required - calculate ratio */
2286 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2287 user_tsc_khz, tsc_khz);
2289 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2290 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2295 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2299 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2301 u32 thresh_lo, thresh_hi;
2302 int use_scaling = 0;
2304 /* tsc_khz can be zero if TSC calibration fails */
2305 if (user_tsc_khz == 0) {
2306 /* set tsc_scaling_ratio to a safe value */
2307 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_default_tsc_scaling_ratio);
2311 /* Compute a scale to convert nanoseconds in TSC cycles */
2312 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2313 &vcpu->arch.virtual_tsc_shift,
2314 &vcpu->arch.virtual_tsc_mult);
2315 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2318 * Compute the variation in TSC rate which is acceptable
2319 * within the range of tolerance and decide if the
2320 * rate being applied is within that bounds of the hardware
2321 * rate. If so, no scaling or compensation need be done.
2323 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2324 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2325 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2326 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2329 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2332 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2334 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2335 vcpu->arch.virtual_tsc_mult,
2336 vcpu->arch.virtual_tsc_shift);
2337 tsc += vcpu->arch.this_tsc_write;
2341 static inline int gtod_is_based_on_tsc(int mode)
2343 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2346 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2348 #ifdef CONFIG_X86_64
2350 struct kvm_arch *ka = &vcpu->kvm->arch;
2351 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2353 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2354 atomic_read(&vcpu->kvm->online_vcpus));
2357 * Once the masterclock is enabled, always perform request in
2358 * order to update it.
2360 * In order to enable masterclock, the host clocksource must be TSC
2361 * and the vcpus need to have matched TSCs. When that happens,
2362 * perform request to enable masterclock.
2364 if (ka->use_master_clock ||
2365 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2366 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2368 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2369 atomic_read(&vcpu->kvm->online_vcpus),
2370 ka->use_master_clock, gtod->clock.vclock_mode);
2375 * Multiply tsc by a fixed point number represented by ratio.
2377 * The most significant 64-N bits (mult) of ratio represent the
2378 * integral part of the fixed point number; the remaining N bits
2379 * (frac) represent the fractional part, ie. ratio represents a fixed
2380 * point number (mult + frac * 2^(-N)).
2382 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2384 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2386 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2389 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, u64 ratio)
2393 if (ratio != kvm_default_tsc_scaling_ratio)
2394 _tsc = __scale_tsc(ratio, tsc);
2398 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2400 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2404 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2406 return target_tsc - tsc;
2409 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2411 return vcpu->arch.l1_tsc_offset +
2412 kvm_scale_tsc(vcpu, host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2414 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2416 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2420 if (l2_multiplier == kvm_default_tsc_scaling_ratio)
2421 nested_offset = l1_offset;
2423 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2424 kvm_tsc_scaling_ratio_frac_bits);
2426 nested_offset += l2_offset;
2427 return nested_offset;
2429 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2431 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2433 if (l2_multiplier != kvm_default_tsc_scaling_ratio)
2434 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2435 kvm_tsc_scaling_ratio_frac_bits);
2437 return l1_multiplier;
2439 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2441 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2443 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2444 vcpu->arch.l1_tsc_offset,
2447 vcpu->arch.l1_tsc_offset = l1_offset;
2450 * If we are here because L1 chose not to trap WRMSR to TSC then
2451 * according to the spec this should set L1's TSC (as opposed to
2452 * setting L1's offset for L2).
2454 if (is_guest_mode(vcpu))
2455 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2457 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2458 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2460 vcpu->arch.tsc_offset = l1_offset;
2462 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2465 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2467 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2469 /* Userspace is changing the multiplier while L2 is active */
2470 if (is_guest_mode(vcpu))
2471 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2473 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2475 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2477 if (kvm_has_tsc_control)
2478 static_call(kvm_x86_write_tsc_multiplier)(
2479 vcpu, vcpu->arch.tsc_scaling_ratio);
2482 static inline bool kvm_check_tsc_unstable(void)
2484 #ifdef CONFIG_X86_64
2486 * TSC is marked unstable when we're running on Hyper-V,
2487 * 'TSC page' clocksource is good.
2489 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2492 return check_tsc_unstable();
2496 * Infers attempts to synchronize the guest's tsc from host writes. Sets the
2497 * offset for the vcpu and tracks the TSC matching generation that the vcpu
2500 static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
2501 u64 ns, bool matched)
2503 struct kvm *kvm = vcpu->kvm;
2505 lockdep_assert_held(&kvm->arch.tsc_write_lock);
2508 * We also track th most recent recorded KHZ, write and time to
2509 * allow the matching interval to be extended at each write.
2511 kvm->arch.last_tsc_nsec = ns;
2512 kvm->arch.last_tsc_write = tsc;
2513 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2514 kvm->arch.last_tsc_offset = offset;
2516 vcpu->arch.last_guest_tsc = tsc;
2518 kvm_vcpu_write_tsc_offset(vcpu, offset);
2522 * We split periods of matched TSC writes into generations.
2523 * For each generation, we track the original measured
2524 * nanosecond time, offset, and write, so if TSCs are in
2525 * sync, we can match exact offset, and if not, we can match
2526 * exact software computation in compute_guest_tsc()
2528 * These values are tracked in kvm->arch.cur_xxx variables.
2530 kvm->arch.cur_tsc_generation++;
2531 kvm->arch.cur_tsc_nsec = ns;
2532 kvm->arch.cur_tsc_write = tsc;
2533 kvm->arch.cur_tsc_offset = offset;
2534 kvm->arch.nr_vcpus_matched_tsc = 0;
2535 } else if (vcpu->arch.this_tsc_generation != kvm->arch.cur_tsc_generation) {
2536 kvm->arch.nr_vcpus_matched_tsc++;
2539 /* Keep track of which generation this VCPU has synchronized to */
2540 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2541 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2542 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2544 kvm_track_tsc_matching(vcpu);
2547 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2549 struct kvm *kvm = vcpu->kvm;
2550 u64 offset, ns, elapsed;
2551 unsigned long flags;
2552 bool matched = false;
2553 bool synchronizing = false;
2555 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2556 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2557 ns = get_kvmclock_base_ns();
2558 elapsed = ns - kvm->arch.last_tsc_nsec;
2560 if (vcpu->arch.virtual_tsc_khz) {
2563 * detection of vcpu initialization -- need to sync
2564 * with other vCPUs. This particularly helps to keep
2565 * kvm_clock stable after CPU hotplug
2567 synchronizing = true;
2569 u64 tsc_exp = kvm->arch.last_tsc_write +
2570 nsec_to_cycles(vcpu, elapsed);
2571 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2573 * Special case: TSC write with a small delta (1 second)
2574 * of virtual cycle time against real time is
2575 * interpreted as an attempt to synchronize the CPU.
2577 synchronizing = data < tsc_exp + tsc_hz &&
2578 data + tsc_hz > tsc_exp;
2583 * For a reliable TSC, we can match TSC offsets, and for an unstable
2584 * TSC, we add elapsed time in this computation. We could let the
2585 * compensation code attempt to catch up if we fall behind, but
2586 * it's better to try to match offsets from the beginning.
2588 if (synchronizing &&
2589 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2590 if (!kvm_check_tsc_unstable()) {
2591 offset = kvm->arch.cur_tsc_offset;
2593 u64 delta = nsec_to_cycles(vcpu, elapsed);
2595 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2600 __kvm_synchronize_tsc(vcpu, offset, data, ns, matched);
2601 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2604 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2607 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2608 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2611 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2613 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2614 WARN_ON(adjustment < 0);
2615 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment,
2616 vcpu->arch.l1_tsc_scaling_ratio);
2617 adjust_tsc_offset_guest(vcpu, adjustment);
2620 #ifdef CONFIG_X86_64
2622 static u64 read_tsc(void)
2624 u64 ret = (u64)rdtsc_ordered();
2625 u64 last = pvclock_gtod_data.clock.cycle_last;
2627 if (likely(ret >= last))
2631 * GCC likes to generate cmov here, but this branch is extremely
2632 * predictable (it's just a function of time and the likely is
2633 * very likely) and there's a data dependence, so force GCC
2634 * to generate a branch instead. I don't barrier() because
2635 * we don't actually need a barrier, and if this function
2636 * ever gets inlined it will generate worse code.
2642 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2648 switch (clock->vclock_mode) {
2649 case VDSO_CLOCKMODE_HVCLOCK:
2650 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2652 if (tsc_pg_val != U64_MAX) {
2653 /* TSC page valid */
2654 *mode = VDSO_CLOCKMODE_HVCLOCK;
2655 v = (tsc_pg_val - clock->cycle_last) &
2658 /* TSC page invalid */
2659 *mode = VDSO_CLOCKMODE_NONE;
2662 case VDSO_CLOCKMODE_TSC:
2663 *mode = VDSO_CLOCKMODE_TSC;
2664 *tsc_timestamp = read_tsc();
2665 v = (*tsc_timestamp - clock->cycle_last) &
2669 *mode = VDSO_CLOCKMODE_NONE;
2672 if (*mode == VDSO_CLOCKMODE_NONE)
2673 *tsc_timestamp = v = 0;
2675 return v * clock->mult;
2678 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2680 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2686 seq = read_seqcount_begin(>od->seq);
2687 ns = gtod->raw_clock.base_cycles;
2688 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2689 ns >>= gtod->raw_clock.shift;
2690 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2691 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2697 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2699 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2705 seq = read_seqcount_begin(>od->seq);
2706 ts->tv_sec = gtod->wall_time_sec;
2707 ns = gtod->clock.base_cycles;
2708 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2709 ns >>= gtod->clock.shift;
2710 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2712 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2718 /* returns true if host is using TSC based clocksource */
2719 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2721 /* checked again under seqlock below */
2722 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2725 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2729 /* returns true if host is using TSC based clocksource */
2730 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2733 /* checked again under seqlock below */
2734 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2737 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2743 * Assuming a stable TSC across physical CPUS, and a stable TSC
2744 * across virtual CPUs, the following condition is possible.
2745 * Each numbered line represents an event visible to both
2746 * CPUs at the next numbered event.
2748 * "timespecX" represents host monotonic time. "tscX" represents
2751 * VCPU0 on CPU0 | VCPU1 on CPU1
2753 * 1. read timespec0,tsc0
2754 * 2. | timespec1 = timespec0 + N
2756 * 3. transition to guest | transition to guest
2757 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2758 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2759 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2761 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2764 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2766 * - 0 < N - M => M < N
2768 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2769 * always the case (the difference between two distinct xtime instances
2770 * might be smaller then the difference between corresponding TSC reads,
2771 * when updating guest vcpus pvclock areas).
2773 * To avoid that problem, do not allow visibility of distinct
2774 * system_timestamp/tsc_timestamp values simultaneously: use a master
2775 * copy of host monotonic time values. Update that master copy
2778 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2782 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2784 #ifdef CONFIG_X86_64
2785 struct kvm_arch *ka = &kvm->arch;
2787 bool host_tsc_clocksource, vcpus_matched;
2789 lockdep_assert_held(&kvm->arch.tsc_write_lock);
2790 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2791 atomic_read(&kvm->online_vcpus));
2794 * If the host uses TSC clock, then passthrough TSC as stable
2797 host_tsc_clocksource = kvm_get_time_and_clockread(
2798 &ka->master_kernel_ns,
2799 &ka->master_cycle_now);
2801 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2802 && !ka->backwards_tsc_observed
2803 && !ka->boot_vcpu_runs_old_kvmclock;
2805 if (ka->use_master_clock)
2806 atomic_set(&kvm_guest_has_master_clock, 1);
2808 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2809 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2814 static void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2816 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2819 static void __kvm_start_pvclock_update(struct kvm *kvm)
2821 raw_spin_lock_irq(&kvm->arch.tsc_write_lock);
2822 write_seqcount_begin(&kvm->arch.pvclock_sc);
2825 static void kvm_start_pvclock_update(struct kvm *kvm)
2827 kvm_make_mclock_inprogress_request(kvm);
2829 /* no guest entries from this point */
2830 __kvm_start_pvclock_update(kvm);
2833 static void kvm_end_pvclock_update(struct kvm *kvm)
2835 struct kvm_arch *ka = &kvm->arch;
2836 struct kvm_vcpu *vcpu;
2839 write_seqcount_end(&ka->pvclock_sc);
2840 raw_spin_unlock_irq(&ka->tsc_write_lock);
2841 kvm_for_each_vcpu(i, vcpu, kvm)
2842 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2844 /* guest entries allowed */
2845 kvm_for_each_vcpu(i, vcpu, kvm)
2846 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2849 static void kvm_update_masterclock(struct kvm *kvm)
2851 kvm_hv_invalidate_tsc_page(kvm);
2852 kvm_start_pvclock_update(kvm);
2853 pvclock_update_vm_gtod_copy(kvm);
2854 kvm_end_pvclock_update(kvm);
2857 /* Called within read_seqcount_begin/retry for kvm->pvclock_sc. */
2858 static void __get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2860 struct kvm_arch *ka = &kvm->arch;
2861 struct pvclock_vcpu_time_info hv_clock;
2863 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2867 if (ka->use_master_clock && __this_cpu_read(cpu_tsc_khz)) {
2868 #ifdef CONFIG_X86_64
2869 struct timespec64 ts;
2871 if (kvm_get_walltime_and_clockread(&ts, &data->host_tsc)) {
2872 data->realtime = ts.tv_nsec + NSEC_PER_SEC * ts.tv_sec;
2873 data->flags |= KVM_CLOCK_REALTIME | KVM_CLOCK_HOST_TSC;
2876 data->host_tsc = rdtsc();
2878 data->flags |= KVM_CLOCK_TSC_STABLE;
2879 hv_clock.tsc_timestamp = ka->master_cycle_now;
2880 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2881 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2882 &hv_clock.tsc_shift,
2883 &hv_clock.tsc_to_system_mul);
2884 data->clock = __pvclock_read_cycles(&hv_clock, data->host_tsc);
2886 data->clock = get_kvmclock_base_ns() + ka->kvmclock_offset;
2892 static void get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2894 struct kvm_arch *ka = &kvm->arch;
2898 seq = read_seqcount_begin(&ka->pvclock_sc);
2899 __get_kvmclock(kvm, data);
2900 } while (read_seqcount_retry(&ka->pvclock_sc, seq));
2903 u64 get_kvmclock_ns(struct kvm *kvm)
2905 struct kvm_clock_data data;
2907 get_kvmclock(kvm, &data);
2911 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2912 struct gfn_to_hva_cache *cache,
2913 unsigned int offset)
2915 struct kvm_vcpu_arch *vcpu = &v->arch;
2916 struct pvclock_vcpu_time_info guest_hv_clock;
2918 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2919 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2922 /* This VCPU is paused, but it's legal for a guest to read another
2923 * VCPU's kvmclock, so we really have to follow the specification where
2924 * it says that version is odd if data is being modified, and even after
2927 * Version field updates must be kept separate. This is because
2928 * kvm_write_guest_cached might use a "rep movs" instruction, and
2929 * writes within a string instruction are weakly ordered. So there
2930 * are three writes overall.
2932 * As a small optimization, only write the version field in the first
2933 * and third write. The vcpu->pv_time cache is still valid, because the
2934 * version field is the first in the struct.
2936 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2938 if (guest_hv_clock.version & 1)
2939 ++guest_hv_clock.version; /* first time write, random junk */
2941 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2942 kvm_write_guest_offset_cached(v->kvm, cache,
2943 &vcpu->hv_clock, offset,
2944 sizeof(vcpu->hv_clock.version));
2948 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2949 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2951 if (vcpu->pvclock_set_guest_stopped_request) {
2952 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2953 vcpu->pvclock_set_guest_stopped_request = false;
2956 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2958 kvm_write_guest_offset_cached(v->kvm, cache,
2959 &vcpu->hv_clock, offset,
2960 sizeof(vcpu->hv_clock));
2964 vcpu->hv_clock.version++;
2965 kvm_write_guest_offset_cached(v->kvm, cache,
2966 &vcpu->hv_clock, offset,
2967 sizeof(vcpu->hv_clock.version));
2970 static int kvm_guest_time_update(struct kvm_vcpu *v)
2972 unsigned long flags, tgt_tsc_khz;
2974 struct kvm_vcpu_arch *vcpu = &v->arch;
2975 struct kvm_arch *ka = &v->kvm->arch;
2977 u64 tsc_timestamp, host_tsc;
2979 bool use_master_clock;
2985 * If the host uses TSC clock, then passthrough TSC as stable
2989 seq = read_seqcount_begin(&ka->pvclock_sc);
2990 use_master_clock = ka->use_master_clock;
2991 if (use_master_clock) {
2992 host_tsc = ka->master_cycle_now;
2993 kernel_ns = ka->master_kernel_ns;
2995 } while (read_seqcount_retry(&ka->pvclock_sc, seq));
2997 /* Keep irq disabled to prevent changes to the clock */
2998 local_irq_save(flags);
2999 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
3000 if (unlikely(tgt_tsc_khz == 0)) {
3001 local_irq_restore(flags);
3002 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3005 if (!use_master_clock) {
3007 kernel_ns = get_kvmclock_base_ns();
3010 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
3013 * We may have to catch up the TSC to match elapsed wall clock
3014 * time for two reasons, even if kvmclock is used.
3015 * 1) CPU could have been running below the maximum TSC rate
3016 * 2) Broken TSC compensation resets the base at each VCPU
3017 * entry to avoid unknown leaps of TSC even when running
3018 * again on the same CPU. This may cause apparent elapsed
3019 * time to disappear, and the guest to stand still or run
3022 if (vcpu->tsc_catchup) {
3023 u64 tsc = compute_guest_tsc(v, kernel_ns);
3024 if (tsc > tsc_timestamp) {
3025 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3026 tsc_timestamp = tsc;
3030 local_irq_restore(flags);
3032 /* With all the info we got, fill in the values */
3034 if (kvm_has_tsc_control)
3035 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz,
3036 v->arch.l1_tsc_scaling_ratio);
3038 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3039 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3040 &vcpu->hv_clock.tsc_shift,
3041 &vcpu->hv_clock.tsc_to_system_mul);
3042 vcpu->hw_tsc_khz = tgt_tsc_khz;
3045 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3046 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3047 vcpu->last_guest_tsc = tsc_timestamp;
3049 /* If the host uses TSC clocksource, then it is stable */
3051 if (use_master_clock)
3052 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3054 vcpu->hv_clock.flags = pvclock_flags;
3056 if (vcpu->pv_time_enabled)
3057 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
3058 if (vcpu->xen.vcpu_info_set)
3059 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
3060 offsetof(struct compat_vcpu_info, time));
3061 if (vcpu->xen.vcpu_time_info_set)
3062 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
3064 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3069 * kvmclock updates which are isolated to a given vcpu, such as
3070 * vcpu->cpu migration, should not allow system_timestamp from
3071 * the rest of the vcpus to remain static. Otherwise ntp frequency
3072 * correction applies to one vcpu's system_timestamp but not
3075 * So in those cases, request a kvmclock update for all vcpus.
3076 * We need to rate-limit these requests though, as they can
3077 * considerably slow guests that have a large number of vcpus.
3078 * The time for a remote vcpu to update its kvmclock is bound
3079 * by the delay we use to rate-limit the updates.
3082 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3084 static void kvmclock_update_fn(struct work_struct *work)
3087 struct delayed_work *dwork = to_delayed_work(work);
3088 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3089 kvmclock_update_work);
3090 struct kvm *kvm = container_of(ka, struct kvm, arch);
3091 struct kvm_vcpu *vcpu;
3093 kvm_for_each_vcpu(i, vcpu, kvm) {
3094 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3095 kvm_vcpu_kick(vcpu);
3099 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3101 struct kvm *kvm = v->kvm;
3103 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3104 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3105 KVMCLOCK_UPDATE_DELAY);
3108 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3110 static void kvmclock_sync_fn(struct work_struct *work)
3112 struct delayed_work *dwork = to_delayed_work(work);
3113 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3114 kvmclock_sync_work);
3115 struct kvm *kvm = container_of(ka, struct kvm, arch);
3117 if (!kvmclock_periodic_sync)
3120 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3121 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3122 KVMCLOCK_SYNC_PERIOD);
3126 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3128 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3130 /* McStatusWrEn enabled? */
3131 if (guest_cpuid_is_amd_or_hygon(vcpu))
3132 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3137 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3139 u64 mcg_cap = vcpu->arch.mcg_cap;
3140 unsigned bank_num = mcg_cap & 0xff;
3141 u32 msr = msr_info->index;
3142 u64 data = msr_info->data;
3145 case MSR_IA32_MCG_STATUS:
3146 vcpu->arch.mcg_status = data;
3148 case MSR_IA32_MCG_CTL:
3149 if (!(mcg_cap & MCG_CTL_P) &&
3150 (data || !msr_info->host_initiated))
3152 if (data != 0 && data != ~(u64)0)
3154 vcpu->arch.mcg_ctl = data;
3157 if (msr >= MSR_IA32_MC0_CTL &&
3158 msr < MSR_IA32_MCx_CTL(bank_num)) {
3159 u32 offset = array_index_nospec(
3160 msr - MSR_IA32_MC0_CTL,
3161 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3163 /* only 0 or all 1s can be written to IA32_MCi_CTL
3164 * some Linux kernels though clear bit 10 in bank 4 to
3165 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
3166 * this to avoid an uncatched #GP in the guest
3168 if ((offset & 0x3) == 0 &&
3169 data != 0 && (data | (1 << 10)) != ~(u64)0)
3173 if (!msr_info->host_initiated &&
3174 (offset & 0x3) == 1 && data != 0) {
3175 if (!can_set_mci_status(vcpu))
3179 vcpu->arch.mce_banks[offset] = data;
3187 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3189 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3191 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3194 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3196 gpa_t gpa = data & ~0x3f;
3198 /* Bits 4:5 are reserved, Should be zero */
3202 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3203 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3206 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3207 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3210 if (!lapic_in_kernel(vcpu))
3211 return data ? 1 : 0;
3213 vcpu->arch.apf.msr_en_val = data;
3215 if (!kvm_pv_async_pf_enabled(vcpu)) {
3216 kvm_clear_async_pf_completion_queue(vcpu);
3217 kvm_async_pf_hash_reset(vcpu);
3221 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3225 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3226 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3228 kvm_async_pf_wakeup_all(vcpu);
3233 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3235 /* Bits 8-63 are reserved */
3239 if (!lapic_in_kernel(vcpu))
3242 vcpu->arch.apf.msr_int_val = data;
3244 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3249 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3251 vcpu->arch.pv_time_enabled = false;
3252 vcpu->arch.time = 0;
3255 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3257 ++vcpu->stat.tlb_flush;
3258 static_call(kvm_x86_tlb_flush_all)(vcpu);
3261 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3263 ++vcpu->stat.tlb_flush;
3267 * A TLB flush on behalf of the guest is equivalent to
3268 * INVPCID(all), toggling CR4.PGE, etc., which requires
3269 * a forced sync of the shadow page tables. Ensure all the
3270 * roots are synced and the guest TLB in hardware is clean.
3272 kvm_mmu_sync_roots(vcpu);
3273 kvm_mmu_sync_prev_roots(vcpu);
3276 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3280 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
3282 ++vcpu->stat.tlb_flush;
3283 static_call(kvm_x86_tlb_flush_current)(vcpu);
3287 * Service "local" TLB flush requests, which are specific to the current MMU
3288 * context. In addition to the generic event handling in vcpu_enter_guest(),
3289 * TLB flushes that are targeted at an MMU context also need to be serviced
3290 * prior before nested VM-Enter/VM-Exit.
3292 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu)
3294 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3295 kvm_vcpu_flush_tlb_current(vcpu);
3297 if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
3298 kvm_vcpu_flush_tlb_guest(vcpu);
3300 EXPORT_SYMBOL_GPL(kvm_service_local_tlb_flush_requests);
3302 static void record_steal_time(struct kvm_vcpu *vcpu)
3304 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
3305 struct kvm_steal_time __user *st;
3306 struct kvm_memslots *slots;
3310 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3311 kvm_xen_runstate_set_running(vcpu);
3315 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3318 if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
3321 slots = kvm_memslots(vcpu->kvm);
3323 if (unlikely(slots->generation != ghc->generation ||
3324 kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
3325 gfn_t gfn = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
3327 /* We rely on the fact that it fits in a single page. */
3328 BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
3330 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gfn, sizeof(*st)) ||
3331 kvm_is_error_hva(ghc->hva) || !ghc->memslot)
3335 st = (struct kvm_steal_time __user *)ghc->hva;
3337 * Doing a TLB flush here, on the guest's behalf, can avoid
3340 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3341 u8 st_preempted = 0;
3344 if (!user_access_begin(st, sizeof(*st)))
3347 asm volatile("1: xchgb %0, %2\n"
3350 _ASM_EXTABLE_UA(1b, 2b)
3351 : "+q" (st_preempted),
3353 "+m" (st->preempted));
3359 vcpu->arch.st.preempted = 0;
3361 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3362 st_preempted & KVM_VCPU_FLUSH_TLB);
3363 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3364 kvm_vcpu_flush_tlb_guest(vcpu);
3366 if (!user_access_begin(st, sizeof(*st)))
3369 if (!user_access_begin(st, sizeof(*st)))
3372 unsafe_put_user(0, &st->preempted, out);
3373 vcpu->arch.st.preempted = 0;
3376 unsafe_get_user(version, &st->version, out);
3378 version += 1; /* first time write, random junk */
3381 unsafe_put_user(version, &st->version, out);
3385 unsafe_get_user(steal, &st->steal, out);
3386 steal += current->sched_info.run_delay -
3387 vcpu->arch.st.last_steal;
3388 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3389 unsafe_put_user(steal, &st->steal, out);
3392 unsafe_put_user(version, &st->version, out);
3397 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
3400 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3403 u32 msr = msr_info->index;
3404 u64 data = msr_info->data;
3406 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3407 return kvm_xen_write_hypercall_page(vcpu, data);
3410 case MSR_AMD64_NB_CFG:
3411 case MSR_IA32_UCODE_WRITE:
3412 case MSR_VM_HSAVE_PA:
3413 case MSR_AMD64_PATCH_LOADER:
3414 case MSR_AMD64_BU_CFG2:
3415 case MSR_AMD64_DC_CFG:
3416 case MSR_F15H_EX_CFG:
3419 case MSR_IA32_UCODE_REV:
3420 if (msr_info->host_initiated)
3421 vcpu->arch.microcode_version = data;
3423 case MSR_IA32_ARCH_CAPABILITIES:
3424 if (!msr_info->host_initiated)
3426 vcpu->arch.arch_capabilities = data;
3428 case MSR_IA32_PERF_CAPABILITIES: {
3429 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3431 if (!msr_info->host_initiated)
3433 if (kvm_get_msr_feature(&msr_ent))
3435 if (data & ~msr_ent.data)
3438 vcpu->arch.perf_capabilities = data;
3443 return set_efer(vcpu, msr_info);
3445 data &= ~(u64)0x40; /* ignore flush filter disable */
3446 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3447 data &= ~(u64)0x8; /* ignore TLB cache disable */
3449 /* Handle McStatusWrEn */
3450 if (data == BIT_ULL(18)) {
3451 vcpu->arch.msr_hwcr = data;
3452 } else if (data != 0) {
3453 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3458 case MSR_FAM10H_MMIO_CONF_BASE:
3460 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3465 case 0x200 ... 0x2ff:
3466 return kvm_mtrr_set_msr(vcpu, msr, data);
3467 case MSR_IA32_APICBASE:
3468 return kvm_set_apic_base(vcpu, msr_info);
3469 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3470 return kvm_x2apic_msr_write(vcpu, msr, data);
3471 case MSR_IA32_TSC_DEADLINE:
3472 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3474 case MSR_IA32_TSC_ADJUST:
3475 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3476 if (!msr_info->host_initiated) {
3477 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3478 adjust_tsc_offset_guest(vcpu, adj);
3479 /* Before back to guest, tsc_timestamp must be adjusted
3480 * as well, otherwise guest's percpu pvclock time could jump.
3482 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3484 vcpu->arch.ia32_tsc_adjust_msr = data;
3487 case MSR_IA32_MISC_ENABLE:
3488 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3489 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3490 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3492 vcpu->arch.ia32_misc_enable_msr = data;
3493 kvm_update_cpuid_runtime(vcpu);
3495 vcpu->arch.ia32_misc_enable_msr = data;
3498 case MSR_IA32_SMBASE:
3499 if (!msr_info->host_initiated)
3501 vcpu->arch.smbase = data;
3503 case MSR_IA32_POWER_CTL:
3504 vcpu->arch.msr_ia32_power_ctl = data;
3507 if (msr_info->host_initiated) {
3508 kvm_synchronize_tsc(vcpu, data);
3510 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3511 adjust_tsc_offset_guest(vcpu, adj);
3512 vcpu->arch.ia32_tsc_adjust_msr += adj;
3516 if (!msr_info->host_initiated &&
3517 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3520 * KVM supports exposing PT to the guest, but does not support
3521 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3522 * XSAVES/XRSTORS to save/restore PT MSRs.
3524 if (data & ~supported_xss)
3526 vcpu->arch.ia32_xss = data;
3529 if (!msr_info->host_initiated)
3531 vcpu->arch.smi_count = data;
3533 case MSR_KVM_WALL_CLOCK_NEW:
3534 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3537 vcpu->kvm->arch.wall_clock = data;
3538 kvm_write_wall_clock(vcpu->kvm, data, 0);
3540 case MSR_KVM_WALL_CLOCK:
3541 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3544 vcpu->kvm->arch.wall_clock = data;
3545 kvm_write_wall_clock(vcpu->kvm, data, 0);
3547 case MSR_KVM_SYSTEM_TIME_NEW:
3548 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3551 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3553 case MSR_KVM_SYSTEM_TIME:
3554 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3557 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3559 case MSR_KVM_ASYNC_PF_EN:
3560 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3563 if (kvm_pv_enable_async_pf(vcpu, data))
3566 case MSR_KVM_ASYNC_PF_INT:
3567 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3570 if (kvm_pv_enable_async_pf_int(vcpu, data))
3573 case MSR_KVM_ASYNC_PF_ACK:
3574 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3577 vcpu->arch.apf.pageready_pending = false;
3578 kvm_check_async_pf_completion(vcpu);
3581 case MSR_KVM_STEAL_TIME:
3582 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3585 if (unlikely(!sched_info_on()))
3588 if (data & KVM_STEAL_RESERVED_MASK)
3591 vcpu->arch.st.msr_val = data;
3593 if (!(data & KVM_MSR_ENABLED))
3596 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3599 case MSR_KVM_PV_EOI_EN:
3600 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3603 if (kvm_lapic_set_pv_eoi(vcpu, data, sizeof(u8)))
3607 case MSR_KVM_POLL_CONTROL:
3608 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3611 /* only enable bit supported */
3612 if (data & (-1ULL << 1))
3615 vcpu->arch.msr_kvm_poll_control = data;
3618 case MSR_IA32_MCG_CTL:
3619 case MSR_IA32_MCG_STATUS:
3620 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3621 return set_msr_mce(vcpu, msr_info);
3623 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3624 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3627 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3628 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3629 if (kvm_pmu_is_valid_msr(vcpu, msr))
3630 return kvm_pmu_set_msr(vcpu, msr_info);
3632 if (pr || data != 0)
3633 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3634 "0x%x data 0x%llx\n", msr, data);
3636 case MSR_K7_CLK_CTL:
3638 * Ignore all writes to this no longer documented MSR.
3639 * Writes are only relevant for old K7 processors,
3640 * all pre-dating SVM, but a recommended workaround from
3641 * AMD for these chips. It is possible to specify the
3642 * affected processor models on the command line, hence
3643 * the need to ignore the workaround.
3646 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3647 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3648 case HV_X64_MSR_SYNDBG_OPTIONS:
3649 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3650 case HV_X64_MSR_CRASH_CTL:
3651 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3652 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3653 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3654 case HV_X64_MSR_TSC_EMULATION_STATUS:
3655 return kvm_hv_set_msr_common(vcpu, msr, data,
3656 msr_info->host_initiated);
3657 case MSR_IA32_BBL_CR_CTL3:
3658 /* Drop writes to this legacy MSR -- see rdmsr
3659 * counterpart for further detail.
3661 if (report_ignored_msrs)
3662 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3665 case MSR_AMD64_OSVW_ID_LENGTH:
3666 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3668 vcpu->arch.osvw.length = data;
3670 case MSR_AMD64_OSVW_STATUS:
3671 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3673 vcpu->arch.osvw.status = data;
3675 case MSR_PLATFORM_INFO:
3676 if (!msr_info->host_initiated ||
3677 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3678 cpuid_fault_enabled(vcpu)))
3680 vcpu->arch.msr_platform_info = data;
3682 case MSR_MISC_FEATURES_ENABLES:
3683 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3684 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3685 !supports_cpuid_fault(vcpu)))
3687 vcpu->arch.msr_misc_features_enables = data;
3690 if (kvm_pmu_is_valid_msr(vcpu, msr))
3691 return kvm_pmu_set_msr(vcpu, msr_info);
3692 return KVM_MSR_RET_INVALID;
3696 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3698 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3701 u64 mcg_cap = vcpu->arch.mcg_cap;
3702 unsigned bank_num = mcg_cap & 0xff;
3705 case MSR_IA32_P5_MC_ADDR:
3706 case MSR_IA32_P5_MC_TYPE:
3709 case MSR_IA32_MCG_CAP:
3710 data = vcpu->arch.mcg_cap;
3712 case MSR_IA32_MCG_CTL:
3713 if (!(mcg_cap & MCG_CTL_P) && !host)
3715 data = vcpu->arch.mcg_ctl;
3717 case MSR_IA32_MCG_STATUS:
3718 data = vcpu->arch.mcg_status;
3721 if (msr >= MSR_IA32_MC0_CTL &&
3722 msr < MSR_IA32_MCx_CTL(bank_num)) {
3723 u32 offset = array_index_nospec(
3724 msr - MSR_IA32_MC0_CTL,
3725 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3727 data = vcpu->arch.mce_banks[offset];
3736 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3738 switch (msr_info->index) {
3739 case MSR_IA32_PLATFORM_ID:
3740 case MSR_IA32_EBL_CR_POWERON:
3741 case MSR_IA32_LASTBRANCHFROMIP:
3742 case MSR_IA32_LASTBRANCHTOIP:
3743 case MSR_IA32_LASTINTFROMIP:
3744 case MSR_IA32_LASTINTTOIP:
3745 case MSR_AMD64_SYSCFG:
3746 case MSR_K8_TSEG_ADDR:
3747 case MSR_K8_TSEG_MASK:
3748 case MSR_VM_HSAVE_PA:
3749 case MSR_K8_INT_PENDING_MSG:
3750 case MSR_AMD64_NB_CFG:
3751 case MSR_FAM10H_MMIO_CONF_BASE:
3752 case MSR_AMD64_BU_CFG2:
3753 case MSR_IA32_PERF_CTL:
3754 case MSR_AMD64_DC_CFG:
3755 case MSR_F15H_EX_CFG:
3757 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3758 * limit) MSRs. Just return 0, as we do not want to expose the host
3759 * data here. Do not conditionalize this on CPUID, as KVM does not do
3760 * so for existing CPU-specific MSRs.
3762 case MSR_RAPL_POWER_UNIT:
3763 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3764 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3765 case MSR_PKG_ENERGY_STATUS: /* Total package */
3766 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3769 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3770 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3771 return kvm_pmu_get_msr(vcpu, msr_info);
3772 if (!msr_info->host_initiated)
3776 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3777 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3778 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3779 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3780 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3781 return kvm_pmu_get_msr(vcpu, msr_info);
3784 case MSR_IA32_UCODE_REV:
3785 msr_info->data = vcpu->arch.microcode_version;
3787 case MSR_IA32_ARCH_CAPABILITIES:
3788 if (!msr_info->host_initiated &&
3789 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3791 msr_info->data = vcpu->arch.arch_capabilities;
3793 case MSR_IA32_PERF_CAPABILITIES:
3794 if (!msr_info->host_initiated &&
3795 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3797 msr_info->data = vcpu->arch.perf_capabilities;
3799 case MSR_IA32_POWER_CTL:
3800 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3802 case MSR_IA32_TSC: {
3804 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3805 * even when not intercepted. AMD manual doesn't explicitly
3806 * state this but appears to behave the same.
3808 * On userspace reads and writes, however, we unconditionally
3809 * return L1's TSC value to ensure backwards-compatible
3810 * behavior for migration.
3814 if (msr_info->host_initiated) {
3815 offset = vcpu->arch.l1_tsc_offset;
3816 ratio = vcpu->arch.l1_tsc_scaling_ratio;
3818 offset = vcpu->arch.tsc_offset;
3819 ratio = vcpu->arch.tsc_scaling_ratio;
3822 msr_info->data = kvm_scale_tsc(vcpu, rdtsc(), ratio) + offset;
3826 case 0x200 ... 0x2ff:
3827 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3828 case 0xcd: /* fsb frequency */
3832 * MSR_EBC_FREQUENCY_ID
3833 * Conservative value valid for even the basic CPU models.
3834 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3835 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3836 * and 266MHz for model 3, or 4. Set Core Clock
3837 * Frequency to System Bus Frequency Ratio to 1 (bits
3838 * 31:24) even though these are only valid for CPU
3839 * models > 2, however guests may end up dividing or
3840 * multiplying by zero otherwise.
3842 case MSR_EBC_FREQUENCY_ID:
3843 msr_info->data = 1 << 24;
3845 case MSR_IA32_APICBASE:
3846 msr_info->data = kvm_get_apic_base(vcpu);
3848 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3849 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3850 case MSR_IA32_TSC_DEADLINE:
3851 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3853 case MSR_IA32_TSC_ADJUST:
3854 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3856 case MSR_IA32_MISC_ENABLE:
3857 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3859 case MSR_IA32_SMBASE:
3860 if (!msr_info->host_initiated)
3862 msr_info->data = vcpu->arch.smbase;
3865 msr_info->data = vcpu->arch.smi_count;
3867 case MSR_IA32_PERF_STATUS:
3868 /* TSC increment by tick */
3869 msr_info->data = 1000ULL;
3870 /* CPU multiplier */
3871 msr_info->data |= (((uint64_t)4ULL) << 40);
3874 msr_info->data = vcpu->arch.efer;
3876 case MSR_KVM_WALL_CLOCK:
3877 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3880 msr_info->data = vcpu->kvm->arch.wall_clock;
3882 case MSR_KVM_WALL_CLOCK_NEW:
3883 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3886 msr_info->data = vcpu->kvm->arch.wall_clock;
3888 case MSR_KVM_SYSTEM_TIME:
3889 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3892 msr_info->data = vcpu->arch.time;
3894 case MSR_KVM_SYSTEM_TIME_NEW:
3895 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3898 msr_info->data = vcpu->arch.time;
3900 case MSR_KVM_ASYNC_PF_EN:
3901 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3904 msr_info->data = vcpu->arch.apf.msr_en_val;
3906 case MSR_KVM_ASYNC_PF_INT:
3907 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3910 msr_info->data = vcpu->arch.apf.msr_int_val;
3912 case MSR_KVM_ASYNC_PF_ACK:
3913 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3918 case MSR_KVM_STEAL_TIME:
3919 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3922 msr_info->data = vcpu->arch.st.msr_val;
3924 case MSR_KVM_PV_EOI_EN:
3925 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3928 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3930 case MSR_KVM_POLL_CONTROL:
3931 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3934 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3936 case MSR_IA32_P5_MC_ADDR:
3937 case MSR_IA32_P5_MC_TYPE:
3938 case MSR_IA32_MCG_CAP:
3939 case MSR_IA32_MCG_CTL:
3940 case MSR_IA32_MCG_STATUS:
3941 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3942 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3943 msr_info->host_initiated);
3945 if (!msr_info->host_initiated &&
3946 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3948 msr_info->data = vcpu->arch.ia32_xss;
3950 case MSR_K7_CLK_CTL:
3952 * Provide expected ramp-up count for K7. All other
3953 * are set to zero, indicating minimum divisors for
3956 * This prevents guest kernels on AMD host with CPU
3957 * type 6, model 8 and higher from exploding due to
3958 * the rdmsr failing.
3960 msr_info->data = 0x20000000;
3962 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3963 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3964 case HV_X64_MSR_SYNDBG_OPTIONS:
3965 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3966 case HV_X64_MSR_CRASH_CTL:
3967 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3968 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3969 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3970 case HV_X64_MSR_TSC_EMULATION_STATUS:
3971 return kvm_hv_get_msr_common(vcpu,
3972 msr_info->index, &msr_info->data,
3973 msr_info->host_initiated);
3974 case MSR_IA32_BBL_CR_CTL3:
3975 /* This legacy MSR exists but isn't fully documented in current
3976 * silicon. It is however accessed by winxp in very narrow
3977 * scenarios where it sets bit #19, itself documented as
3978 * a "reserved" bit. Best effort attempt to source coherent
3979 * read data here should the balance of the register be
3980 * interpreted by the guest:
3982 * L2 cache control register 3: 64GB range, 256KB size,
3983 * enabled, latency 0x1, configured
3985 msr_info->data = 0xbe702111;
3987 case MSR_AMD64_OSVW_ID_LENGTH:
3988 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3990 msr_info->data = vcpu->arch.osvw.length;
3992 case MSR_AMD64_OSVW_STATUS:
3993 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3995 msr_info->data = vcpu->arch.osvw.status;
3997 case MSR_PLATFORM_INFO:
3998 if (!msr_info->host_initiated &&
3999 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
4001 msr_info->data = vcpu->arch.msr_platform_info;
4003 case MSR_MISC_FEATURES_ENABLES:
4004 msr_info->data = vcpu->arch.msr_misc_features_enables;
4007 msr_info->data = vcpu->arch.msr_hwcr;
4010 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
4011 return kvm_pmu_get_msr(vcpu, msr_info);
4012 return KVM_MSR_RET_INVALID;
4016 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
4019 * Read or write a bunch of msrs. All parameters are kernel addresses.
4021 * @return number of msrs set successfully.
4023 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
4024 struct kvm_msr_entry *entries,
4025 int (*do_msr)(struct kvm_vcpu *vcpu,
4026 unsigned index, u64 *data))
4030 for (i = 0; i < msrs->nmsrs; ++i)
4031 if (do_msr(vcpu, entries[i].index, &entries[i].data))
4038 * Read or write a bunch of msrs. Parameters are user addresses.
4040 * @return number of msrs set successfully.
4042 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
4043 int (*do_msr)(struct kvm_vcpu *vcpu,
4044 unsigned index, u64 *data),
4047 struct kvm_msrs msrs;
4048 struct kvm_msr_entry *entries;
4053 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
4057 if (msrs.nmsrs >= MAX_IO_MSRS)
4060 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
4061 entries = memdup_user(user_msrs->entries, size);
4062 if (IS_ERR(entries)) {
4063 r = PTR_ERR(entries);
4067 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
4072 if (writeback && copy_to_user(user_msrs->entries, entries, size))
4083 static inline bool kvm_can_mwait_in_guest(void)
4085 return boot_cpu_has(X86_FEATURE_MWAIT) &&
4086 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
4087 boot_cpu_has(X86_FEATURE_ARAT);
4090 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4091 struct kvm_cpuid2 __user *cpuid_arg)
4093 struct kvm_cpuid2 cpuid;
4097 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4100 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4105 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4111 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4116 case KVM_CAP_IRQCHIP:
4118 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4119 case KVM_CAP_SET_TSS_ADDR:
4120 case KVM_CAP_EXT_CPUID:
4121 case KVM_CAP_EXT_EMUL_CPUID:
4122 case KVM_CAP_CLOCKSOURCE:
4124 case KVM_CAP_NOP_IO_DELAY:
4125 case KVM_CAP_MP_STATE:
4126 case KVM_CAP_SYNC_MMU:
4127 case KVM_CAP_USER_NMI:
4128 case KVM_CAP_REINJECT_CONTROL:
4129 case KVM_CAP_IRQ_INJECT_STATUS:
4130 case KVM_CAP_IOEVENTFD:
4131 case KVM_CAP_IOEVENTFD_NO_LENGTH:
4133 case KVM_CAP_PIT_STATE2:
4134 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4135 case KVM_CAP_VCPU_EVENTS:
4136 case KVM_CAP_HYPERV:
4137 case KVM_CAP_HYPERV_VAPIC:
4138 case KVM_CAP_HYPERV_SPIN:
4139 case KVM_CAP_HYPERV_SYNIC:
4140 case KVM_CAP_HYPERV_SYNIC2:
4141 case KVM_CAP_HYPERV_VP_INDEX:
4142 case KVM_CAP_HYPERV_EVENTFD:
4143 case KVM_CAP_HYPERV_TLBFLUSH:
4144 case KVM_CAP_HYPERV_SEND_IPI:
4145 case KVM_CAP_HYPERV_CPUID:
4146 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4147 case KVM_CAP_SYS_HYPERV_CPUID:
4148 case KVM_CAP_PCI_SEGMENT:
4149 case KVM_CAP_DEBUGREGS:
4150 case KVM_CAP_X86_ROBUST_SINGLESTEP:
4152 case KVM_CAP_ASYNC_PF:
4153 case KVM_CAP_ASYNC_PF_INT:
4154 case KVM_CAP_GET_TSC_KHZ:
4155 case KVM_CAP_KVMCLOCK_CTRL:
4156 case KVM_CAP_READONLY_MEM:
4157 case KVM_CAP_HYPERV_TIME:
4158 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4159 case KVM_CAP_TSC_DEADLINE_TIMER:
4160 case KVM_CAP_DISABLE_QUIRKS:
4161 case KVM_CAP_SET_BOOT_CPU_ID:
4162 case KVM_CAP_SPLIT_IRQCHIP:
4163 case KVM_CAP_IMMEDIATE_EXIT:
4164 case KVM_CAP_PMU_EVENT_FILTER:
4165 case KVM_CAP_GET_MSR_FEATURES:
4166 case KVM_CAP_MSR_PLATFORM_INFO:
4167 case KVM_CAP_EXCEPTION_PAYLOAD:
4168 case KVM_CAP_SET_GUEST_DEBUG:
4169 case KVM_CAP_LAST_CPU:
4170 case KVM_CAP_X86_USER_SPACE_MSR:
4171 case KVM_CAP_X86_MSR_FILTER:
4172 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4173 #ifdef CONFIG_X86_SGX_KVM
4174 case KVM_CAP_SGX_ATTRIBUTE:
4176 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4177 case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
4178 case KVM_CAP_SREGS2:
4179 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4180 case KVM_CAP_VCPU_ATTRIBUTES:
4183 case KVM_CAP_EXIT_HYPERCALL:
4184 r = KVM_EXIT_HYPERCALL_VALID_MASK;
4186 case KVM_CAP_SET_GUEST_DEBUG2:
4187 return KVM_GUESTDBG_VALID_MASK;
4188 #ifdef CONFIG_KVM_XEN
4189 case KVM_CAP_XEN_HVM:
4190 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4191 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4192 KVM_XEN_HVM_CONFIG_SHARED_INFO |
4193 KVM_XEN_HVM_CONFIG_EVTCHN_2LEVEL;
4194 if (sched_info_on())
4195 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
4198 case KVM_CAP_SYNC_REGS:
4199 r = KVM_SYNC_X86_VALID_FIELDS;
4201 case KVM_CAP_ADJUST_CLOCK:
4202 r = KVM_CLOCK_VALID_FLAGS;
4204 case KVM_CAP_X86_DISABLE_EXITS:
4205 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4206 KVM_X86_DISABLE_EXITS_CSTATE;
4207 if(kvm_can_mwait_in_guest())
4208 r |= KVM_X86_DISABLE_EXITS_MWAIT;
4210 case KVM_CAP_X86_SMM:
4211 /* SMBASE is usually relocated above 1M on modern chipsets,
4212 * and SMM handlers might indeed rely on 4G segment limits,
4213 * so do not report SMM to be available if real mode is
4214 * emulated via vm86 mode. Still, do not go to great lengths
4215 * to avoid userspace's usage of the feature, because it is a
4216 * fringe case that is not enabled except via specific settings
4217 * of the module parameters.
4219 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4222 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
4224 case KVM_CAP_NR_VCPUS:
4225 r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
4227 case KVM_CAP_MAX_VCPUS:
4230 case KVM_CAP_MAX_VCPU_ID:
4231 r = KVM_MAX_VCPU_IDS;
4233 case KVM_CAP_PV_MMU: /* obsolete */
4237 r = KVM_MAX_MCE_BANKS;
4240 r = boot_cpu_has(X86_FEATURE_XSAVE);
4242 case KVM_CAP_TSC_CONTROL:
4243 r = kvm_has_tsc_control;
4245 case KVM_CAP_X2APIC_API:
4246 r = KVM_X2APIC_API_VALID_FLAGS;
4248 case KVM_CAP_NESTED_STATE:
4249 r = kvm_x86_ops.nested_ops->get_state ?
4250 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4252 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4253 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
4255 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4256 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4258 case KVM_CAP_SMALLER_MAXPHYADDR:
4259 r = (int) allow_smaller_maxphyaddr;
4261 case KVM_CAP_STEAL_TIME:
4262 r = sched_info_on();
4264 case KVM_CAP_X86_BUS_LOCK_EXIT:
4265 if (kvm_has_bus_lock_exit)
4266 r = KVM_BUS_LOCK_DETECTION_OFF |
4267 KVM_BUS_LOCK_DETECTION_EXIT;
4278 long kvm_arch_dev_ioctl(struct file *filp,
4279 unsigned int ioctl, unsigned long arg)
4281 void __user *argp = (void __user *)arg;
4285 case KVM_GET_MSR_INDEX_LIST: {
4286 struct kvm_msr_list __user *user_msr_list = argp;
4287 struct kvm_msr_list msr_list;
4291 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4294 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4295 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4298 if (n < msr_list.nmsrs)
4301 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4302 num_msrs_to_save * sizeof(u32)))
4304 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4306 num_emulated_msrs * sizeof(u32)))
4311 case KVM_GET_SUPPORTED_CPUID:
4312 case KVM_GET_EMULATED_CPUID: {
4313 struct kvm_cpuid2 __user *cpuid_arg = argp;
4314 struct kvm_cpuid2 cpuid;
4317 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4320 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4326 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4331 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4333 if (copy_to_user(argp, &kvm_mce_cap_supported,
4334 sizeof(kvm_mce_cap_supported)))
4338 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4339 struct kvm_msr_list __user *user_msr_list = argp;
4340 struct kvm_msr_list msr_list;
4344 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4347 msr_list.nmsrs = num_msr_based_features;
4348 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4351 if (n < msr_list.nmsrs)
4354 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4355 num_msr_based_features * sizeof(u32)))
4361 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4363 case KVM_GET_SUPPORTED_HV_CPUID:
4364 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4374 static void wbinvd_ipi(void *garbage)
4379 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4381 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4384 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4386 /* Address WBINVD may be executed by guest */
4387 if (need_emulate_wbinvd(vcpu)) {
4388 if (static_call(kvm_x86_has_wbinvd_exit)())
4389 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4390 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4391 smp_call_function_single(vcpu->cpu,
4392 wbinvd_ipi, NULL, 1);
4395 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4397 /* Save host pkru register if supported */
4398 vcpu->arch.host_pkru = read_pkru();
4400 /* Apply any externally detected TSC adjustments (due to suspend) */
4401 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4402 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4403 vcpu->arch.tsc_offset_adjustment = 0;
4404 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4407 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4408 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4409 rdtsc() - vcpu->arch.last_host_tsc;
4411 mark_tsc_unstable("KVM discovered backwards TSC");
4413 if (kvm_check_tsc_unstable()) {
4414 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4415 vcpu->arch.last_guest_tsc);
4416 kvm_vcpu_write_tsc_offset(vcpu, offset);
4417 vcpu->arch.tsc_catchup = 1;
4420 if (kvm_lapic_hv_timer_in_use(vcpu))
4421 kvm_lapic_restart_hv_timer(vcpu);
4424 * On a host with synchronized TSC, there is no need to update
4425 * kvmclock on vcpu->cpu migration
4427 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4428 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4429 if (vcpu->cpu != cpu)
4430 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4434 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4437 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4439 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
4440 struct kvm_steal_time __user *st;
4441 struct kvm_memslots *slots;
4442 static const u8 preempted = KVM_VCPU_PREEMPTED;
4444 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4447 if (vcpu->arch.st.preempted)
4450 /* This happens on process exit */
4451 if (unlikely(current->mm != vcpu->kvm->mm))
4454 slots = kvm_memslots(vcpu->kvm);
4456 if (unlikely(slots->generation != ghc->generation ||
4457 kvm_is_error_hva(ghc->hva) || !ghc->memslot))
4460 st = (struct kvm_steal_time __user *)ghc->hva;
4461 BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
4463 if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
4464 vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4466 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
4469 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4473 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4474 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4477 * Take the srcu lock as memslots will be accessed to check the gfn
4478 * cache generation against the memslots generation.
4480 idx = srcu_read_lock(&vcpu->kvm->srcu);
4481 if (kvm_xen_msr_enabled(vcpu->kvm))
4482 kvm_xen_runstate_set_preempted(vcpu);
4484 kvm_steal_time_set_preempted(vcpu);
4485 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4487 static_call(kvm_x86_vcpu_put)(vcpu);
4488 vcpu->arch.last_host_tsc = rdtsc();
4491 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4492 struct kvm_lapic_state *s)
4494 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
4496 return kvm_apic_get_state(vcpu, s);
4499 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4500 struct kvm_lapic_state *s)
4504 r = kvm_apic_set_state(vcpu, s);
4507 update_cr8_intercept(vcpu);
4512 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4515 * We can accept userspace's request for interrupt injection
4516 * as long as we have a place to store the interrupt number.
4517 * The actual injection will happen when the CPU is able to
4518 * deliver the interrupt.
4520 if (kvm_cpu_has_extint(vcpu))
4523 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4524 return (!lapic_in_kernel(vcpu) ||
4525 kvm_apic_accept_pic_intr(vcpu));
4528 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4531 * Do not cause an interrupt window exit if an exception
4532 * is pending or an event needs reinjection; userspace
4533 * might want to inject the interrupt manually using KVM_SET_REGS
4534 * or KVM_SET_SREGS. For that to work, we must be at an
4535 * instruction boundary and with no events half-injected.
4537 return (kvm_arch_interrupt_allowed(vcpu) &&
4538 kvm_cpu_accept_dm_intr(vcpu) &&
4539 !kvm_event_needs_reinjection(vcpu) &&
4540 !vcpu->arch.exception.pending);
4543 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4544 struct kvm_interrupt *irq)
4546 if (irq->irq >= KVM_NR_INTERRUPTS)
4549 if (!irqchip_in_kernel(vcpu->kvm)) {
4550 kvm_queue_interrupt(vcpu, irq->irq, false);
4551 kvm_make_request(KVM_REQ_EVENT, vcpu);
4556 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4557 * fail for in-kernel 8259.
4559 if (pic_in_kernel(vcpu->kvm))
4562 if (vcpu->arch.pending_external_vector != -1)
4565 vcpu->arch.pending_external_vector = irq->irq;
4566 kvm_make_request(KVM_REQ_EVENT, vcpu);
4570 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4572 kvm_inject_nmi(vcpu);
4577 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4579 kvm_make_request(KVM_REQ_SMI, vcpu);
4584 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4585 struct kvm_tpr_access_ctl *tac)
4589 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4593 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4597 unsigned bank_num = mcg_cap & 0xff, bank;
4600 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4602 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4605 vcpu->arch.mcg_cap = mcg_cap;
4606 /* Init IA32_MCG_CTL to all 1s */
4607 if (mcg_cap & MCG_CTL_P)
4608 vcpu->arch.mcg_ctl = ~(u64)0;
4609 /* Init IA32_MCi_CTL to all 1s */
4610 for (bank = 0; bank < bank_num; bank++)
4611 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4613 static_call(kvm_x86_setup_mce)(vcpu);
4618 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4619 struct kvm_x86_mce *mce)
4621 u64 mcg_cap = vcpu->arch.mcg_cap;
4622 unsigned bank_num = mcg_cap & 0xff;
4623 u64 *banks = vcpu->arch.mce_banks;
4625 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4628 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4629 * reporting is disabled
4631 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4632 vcpu->arch.mcg_ctl != ~(u64)0)
4634 banks += 4 * mce->bank;
4636 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4637 * reporting is disabled for the bank
4639 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4641 if (mce->status & MCI_STATUS_UC) {
4642 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4643 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4644 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4647 if (banks[1] & MCI_STATUS_VAL)
4648 mce->status |= MCI_STATUS_OVER;
4649 banks[2] = mce->addr;
4650 banks[3] = mce->misc;
4651 vcpu->arch.mcg_status = mce->mcg_status;
4652 banks[1] = mce->status;
4653 kvm_queue_exception(vcpu, MC_VECTOR);
4654 } else if (!(banks[1] & MCI_STATUS_VAL)
4655 || !(banks[1] & MCI_STATUS_UC)) {
4656 if (banks[1] & MCI_STATUS_VAL)
4657 mce->status |= MCI_STATUS_OVER;
4658 banks[2] = mce->addr;
4659 banks[3] = mce->misc;
4660 banks[1] = mce->status;
4662 banks[1] |= MCI_STATUS_OVER;
4666 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4667 struct kvm_vcpu_events *events)
4671 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4675 * In guest mode, payload delivery should be deferred,
4676 * so that the L1 hypervisor can intercept #PF before
4677 * CR2 is modified (or intercept #DB before DR6 is
4678 * modified under nVMX). Unless the per-VM capability,
4679 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4680 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4681 * opportunistically defer the exception payload, deliver it if the
4682 * capability hasn't been requested before processing a
4683 * KVM_GET_VCPU_EVENTS.
4685 if (!vcpu->kvm->arch.exception_payload_enabled &&
4686 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4687 kvm_deliver_exception_payload(vcpu);
4690 * The API doesn't provide the instruction length for software
4691 * exceptions, so don't report them. As long as the guest RIP
4692 * isn't advanced, we should expect to encounter the exception
4695 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4696 events->exception.injected = 0;
4697 events->exception.pending = 0;
4699 events->exception.injected = vcpu->arch.exception.injected;
4700 events->exception.pending = vcpu->arch.exception.pending;
4702 * For ABI compatibility, deliberately conflate
4703 * pending and injected exceptions when
4704 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4706 if (!vcpu->kvm->arch.exception_payload_enabled)
4707 events->exception.injected |=
4708 vcpu->arch.exception.pending;
4710 events->exception.nr = vcpu->arch.exception.nr;
4711 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4712 events->exception.error_code = vcpu->arch.exception.error_code;
4713 events->exception_has_payload = vcpu->arch.exception.has_payload;
4714 events->exception_payload = vcpu->arch.exception.payload;
4716 events->interrupt.injected =
4717 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4718 events->interrupt.nr = vcpu->arch.interrupt.nr;
4719 events->interrupt.soft = 0;
4720 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4722 events->nmi.injected = vcpu->arch.nmi_injected;
4723 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4724 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4725 events->nmi.pad = 0;
4727 events->sipi_vector = 0; /* never valid when reporting to user space */
4729 events->smi.smm = is_smm(vcpu);
4730 events->smi.pending = vcpu->arch.smi_pending;
4731 events->smi.smm_inside_nmi =
4732 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4733 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4735 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4736 | KVM_VCPUEVENT_VALID_SHADOW
4737 | KVM_VCPUEVENT_VALID_SMM);
4738 if (vcpu->kvm->arch.exception_payload_enabled)
4739 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4741 memset(&events->reserved, 0, sizeof(events->reserved));
4744 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm);
4746 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4747 struct kvm_vcpu_events *events)
4749 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4750 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4751 | KVM_VCPUEVENT_VALID_SHADOW
4752 | KVM_VCPUEVENT_VALID_SMM
4753 | KVM_VCPUEVENT_VALID_PAYLOAD))
4756 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4757 if (!vcpu->kvm->arch.exception_payload_enabled)
4759 if (events->exception.pending)
4760 events->exception.injected = 0;
4762 events->exception_has_payload = 0;
4764 events->exception.pending = 0;
4765 events->exception_has_payload = 0;
4768 if ((events->exception.injected || events->exception.pending) &&
4769 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4772 /* INITs are latched while in SMM */
4773 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4774 (events->smi.smm || events->smi.pending) &&
4775 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4779 vcpu->arch.exception.injected = events->exception.injected;
4780 vcpu->arch.exception.pending = events->exception.pending;
4781 vcpu->arch.exception.nr = events->exception.nr;
4782 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4783 vcpu->arch.exception.error_code = events->exception.error_code;
4784 vcpu->arch.exception.has_payload = events->exception_has_payload;
4785 vcpu->arch.exception.payload = events->exception_payload;
4787 vcpu->arch.interrupt.injected = events->interrupt.injected;
4788 vcpu->arch.interrupt.nr = events->interrupt.nr;
4789 vcpu->arch.interrupt.soft = events->interrupt.soft;
4790 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4791 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4792 events->interrupt.shadow);
4794 vcpu->arch.nmi_injected = events->nmi.injected;
4795 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4796 vcpu->arch.nmi_pending = events->nmi.pending;
4797 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4799 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4800 lapic_in_kernel(vcpu))
4801 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4803 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4804 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm)
4805 kvm_smm_changed(vcpu, events->smi.smm);
4807 vcpu->arch.smi_pending = events->smi.pending;
4809 if (events->smi.smm) {
4810 if (events->smi.smm_inside_nmi)
4811 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4813 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4816 if (lapic_in_kernel(vcpu)) {
4817 if (events->smi.latched_init)
4818 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4820 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4824 kvm_make_request(KVM_REQ_EVENT, vcpu);
4829 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4830 struct kvm_debugregs *dbgregs)
4834 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4835 kvm_get_dr(vcpu, 6, &val);
4837 dbgregs->dr7 = vcpu->arch.dr7;
4839 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4842 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4843 struct kvm_debugregs *dbgregs)
4848 if (!kvm_dr6_valid(dbgregs->dr6))
4850 if (!kvm_dr7_valid(dbgregs->dr7))
4853 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4854 kvm_update_dr0123(vcpu);
4855 vcpu->arch.dr6 = dbgregs->dr6;
4856 vcpu->arch.dr7 = dbgregs->dr7;
4857 kvm_update_dr7(vcpu);
4862 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4863 struct kvm_xsave *guest_xsave)
4865 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4868 fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
4869 guest_xsave->region,
4870 sizeof(guest_xsave->region),
4874 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4875 struct kvm_xsave *guest_xsave)
4877 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
4880 return fpu_copy_uabi_to_guest_fpstate(&vcpu->arch.guest_fpu,
4881 guest_xsave->region,
4882 supported_xcr0, &vcpu->arch.pkru);
4885 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4886 struct kvm_xcrs *guest_xcrs)
4888 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4889 guest_xcrs->nr_xcrs = 0;
4893 guest_xcrs->nr_xcrs = 1;
4894 guest_xcrs->flags = 0;
4895 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4896 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4899 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4900 struct kvm_xcrs *guest_xcrs)
4904 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4907 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4910 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4911 /* Only support XCR0 currently */
4912 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4913 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4914 guest_xcrs->xcrs[i].value);
4923 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4924 * stopped by the hypervisor. This function will be called from the host only.
4925 * EINVAL is returned when the host attempts to set the flag for a guest that
4926 * does not support pv clocks.
4928 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4930 if (!vcpu->arch.pv_time_enabled)
4932 vcpu->arch.pvclock_set_guest_stopped_request = true;
4933 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4937 static int kvm_arch_tsc_has_attr(struct kvm_vcpu *vcpu,
4938 struct kvm_device_attr *attr)
4942 switch (attr->attr) {
4943 case KVM_VCPU_TSC_OFFSET:
4953 static int kvm_arch_tsc_get_attr(struct kvm_vcpu *vcpu,
4954 struct kvm_device_attr *attr)
4956 u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
4959 if ((u64)(unsigned long)uaddr != attr->addr)
4962 switch (attr->attr) {
4963 case KVM_VCPU_TSC_OFFSET:
4965 if (put_user(vcpu->arch.l1_tsc_offset, uaddr))
4976 static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
4977 struct kvm_device_attr *attr)
4979 u64 __user *uaddr = (u64 __user *)(unsigned long)attr->addr;
4980 struct kvm *kvm = vcpu->kvm;
4983 if ((u64)(unsigned long)uaddr != attr->addr)
4986 switch (attr->attr) {
4987 case KVM_VCPU_TSC_OFFSET: {
4988 u64 offset, tsc, ns;
4989 unsigned long flags;
4993 if (get_user(offset, uaddr))
4996 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
4998 matched = (vcpu->arch.virtual_tsc_khz &&
4999 kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
5000 kvm->arch.last_tsc_offset == offset);
5002 tsc = kvm_scale_tsc(vcpu, rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
5003 ns = get_kvmclock_base_ns();
5005 __kvm_synchronize_tsc(vcpu, offset, tsc, ns, matched);
5006 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
5018 static int kvm_vcpu_ioctl_device_attr(struct kvm_vcpu *vcpu,
5022 struct kvm_device_attr attr;
5025 if (copy_from_user(&attr, argp, sizeof(attr)))
5028 if (attr.group != KVM_VCPU_TSC_CTRL)
5032 case KVM_HAS_DEVICE_ATTR:
5033 r = kvm_arch_tsc_has_attr(vcpu, &attr);
5035 case KVM_GET_DEVICE_ATTR:
5036 r = kvm_arch_tsc_get_attr(vcpu, &attr);
5038 case KVM_SET_DEVICE_ATTR:
5039 r = kvm_arch_tsc_set_attr(vcpu, &attr);
5046 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5047 struct kvm_enable_cap *cap)
5050 uint16_t vmcs_version;
5051 void __user *user_ptr;
5057 case KVM_CAP_HYPERV_SYNIC2:
5062 case KVM_CAP_HYPERV_SYNIC:
5063 if (!irqchip_in_kernel(vcpu->kvm))
5065 return kvm_hv_activate_synic(vcpu, cap->cap ==
5066 KVM_CAP_HYPERV_SYNIC2);
5067 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
5068 if (!kvm_x86_ops.nested_ops->enable_evmcs)
5070 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
5072 user_ptr = (void __user *)(uintptr_t)cap->args[0];
5073 if (copy_to_user(user_ptr, &vmcs_version,
5074 sizeof(vmcs_version)))
5078 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
5079 if (!kvm_x86_ops.enable_direct_tlbflush)
5082 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
5084 case KVM_CAP_HYPERV_ENFORCE_CPUID:
5085 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
5087 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
5088 vcpu->arch.pv_cpuid.enforce = cap->args[0];
5089 if (vcpu->arch.pv_cpuid.enforce)
5090 kvm_update_pv_runtime(vcpu);
5098 long kvm_arch_vcpu_ioctl(struct file *filp,
5099 unsigned int ioctl, unsigned long arg)
5101 struct kvm_vcpu *vcpu = filp->private_data;
5102 void __user *argp = (void __user *)arg;
5105 struct kvm_sregs2 *sregs2;
5106 struct kvm_lapic_state *lapic;
5107 struct kvm_xsave *xsave;
5108 struct kvm_xcrs *xcrs;
5116 case KVM_GET_LAPIC: {
5118 if (!lapic_in_kernel(vcpu))
5120 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5121 GFP_KERNEL_ACCOUNT);
5126 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5130 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5135 case KVM_SET_LAPIC: {
5137 if (!lapic_in_kernel(vcpu))
5139 u.lapic = memdup_user(argp, sizeof(*u.lapic));
5140 if (IS_ERR(u.lapic)) {
5141 r = PTR_ERR(u.lapic);
5145 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5148 case KVM_INTERRUPT: {
5149 struct kvm_interrupt irq;
5152 if (copy_from_user(&irq, argp, sizeof(irq)))
5154 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5158 r = kvm_vcpu_ioctl_nmi(vcpu);
5162 r = kvm_vcpu_ioctl_smi(vcpu);
5165 case KVM_SET_CPUID: {
5166 struct kvm_cpuid __user *cpuid_arg = argp;
5167 struct kvm_cpuid cpuid;
5170 * KVM does not correctly handle changing guest CPUID after KVM_RUN, as
5171 * MAXPHYADDR, GBPAGES support, AMD reserved bit behavior, etc.. aren't
5172 * tracked in kvm_mmu_page_role. As a result, KVM may miss guest page
5173 * faults due to reusing SPs/SPTEs. In practice no sane VMM mucks with
5174 * the core vCPU model on the fly, so fail.
5177 if (vcpu->arch.last_vmentry_cpu != -1)
5181 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5183 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5186 case KVM_SET_CPUID2: {
5187 struct kvm_cpuid2 __user *cpuid_arg = argp;
5188 struct kvm_cpuid2 cpuid;
5191 * KVM_SET_CPUID{,2} after KVM_RUN is forbidded, see the comment in
5192 * KVM_SET_CPUID case above.
5195 if (vcpu->arch.last_vmentry_cpu != -1)
5199 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5201 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5202 cpuid_arg->entries);
5205 case KVM_GET_CPUID2: {
5206 struct kvm_cpuid2 __user *cpuid_arg = argp;
5207 struct kvm_cpuid2 cpuid;
5210 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5212 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5213 cpuid_arg->entries);
5217 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5222 case KVM_GET_MSRS: {
5223 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5224 r = msr_io(vcpu, argp, do_get_msr, 1);
5225 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5228 case KVM_SET_MSRS: {
5229 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5230 r = msr_io(vcpu, argp, do_set_msr, 0);
5231 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5234 case KVM_TPR_ACCESS_REPORTING: {
5235 struct kvm_tpr_access_ctl tac;
5238 if (copy_from_user(&tac, argp, sizeof(tac)))
5240 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5244 if (copy_to_user(argp, &tac, sizeof(tac)))
5249 case KVM_SET_VAPIC_ADDR: {
5250 struct kvm_vapic_addr va;
5254 if (!lapic_in_kernel(vcpu))
5257 if (copy_from_user(&va, argp, sizeof(va)))
5259 idx = srcu_read_lock(&vcpu->kvm->srcu);
5260 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5261 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5264 case KVM_X86_SETUP_MCE: {
5268 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5270 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5273 case KVM_X86_SET_MCE: {
5274 struct kvm_x86_mce mce;
5277 if (copy_from_user(&mce, argp, sizeof(mce)))
5279 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5282 case KVM_GET_VCPU_EVENTS: {
5283 struct kvm_vcpu_events events;
5285 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5288 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5293 case KVM_SET_VCPU_EVENTS: {
5294 struct kvm_vcpu_events events;
5297 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5300 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5303 case KVM_GET_DEBUGREGS: {
5304 struct kvm_debugregs dbgregs;
5306 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5309 if (copy_to_user(argp, &dbgregs,
5310 sizeof(struct kvm_debugregs)))
5315 case KVM_SET_DEBUGREGS: {
5316 struct kvm_debugregs dbgregs;
5319 if (copy_from_user(&dbgregs, argp,
5320 sizeof(struct kvm_debugregs)))
5323 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5326 case KVM_GET_XSAVE: {
5327 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5332 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5335 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5340 case KVM_SET_XSAVE: {
5341 u.xsave = memdup_user(argp, sizeof(*u.xsave));
5342 if (IS_ERR(u.xsave)) {
5343 r = PTR_ERR(u.xsave);
5347 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5350 case KVM_GET_XCRS: {
5351 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5356 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5359 if (copy_to_user(argp, u.xcrs,
5360 sizeof(struct kvm_xcrs)))
5365 case KVM_SET_XCRS: {
5366 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5367 if (IS_ERR(u.xcrs)) {
5368 r = PTR_ERR(u.xcrs);
5372 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5375 case KVM_SET_TSC_KHZ: {
5379 user_tsc_khz = (u32)arg;
5381 if (kvm_has_tsc_control &&
5382 user_tsc_khz >= kvm_max_guest_tsc_khz)
5385 if (user_tsc_khz == 0)
5386 user_tsc_khz = tsc_khz;
5388 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5393 case KVM_GET_TSC_KHZ: {
5394 r = vcpu->arch.virtual_tsc_khz;
5397 case KVM_KVMCLOCK_CTRL: {
5398 r = kvm_set_guest_paused(vcpu);
5401 case KVM_ENABLE_CAP: {
5402 struct kvm_enable_cap cap;
5405 if (copy_from_user(&cap, argp, sizeof(cap)))
5407 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5410 case KVM_GET_NESTED_STATE: {
5411 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5415 if (!kvm_x86_ops.nested_ops->get_state)
5418 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5420 if (get_user(user_data_size, &user_kvm_nested_state->size))
5423 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5428 if (r > user_data_size) {
5429 if (put_user(r, &user_kvm_nested_state->size))
5439 case KVM_SET_NESTED_STATE: {
5440 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5441 struct kvm_nested_state kvm_state;
5445 if (!kvm_x86_ops.nested_ops->set_state)
5449 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5453 if (kvm_state.size < sizeof(kvm_state))
5456 if (kvm_state.flags &
5457 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5458 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5459 | KVM_STATE_NESTED_GIF_SET))
5462 /* nested_run_pending implies guest_mode. */
5463 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5464 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5467 idx = srcu_read_lock(&vcpu->kvm->srcu);
5468 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5469 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5472 case KVM_GET_SUPPORTED_HV_CPUID:
5473 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5475 #ifdef CONFIG_KVM_XEN
5476 case KVM_XEN_VCPU_GET_ATTR: {
5477 struct kvm_xen_vcpu_attr xva;
5480 if (copy_from_user(&xva, argp, sizeof(xva)))
5482 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5483 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5487 case KVM_XEN_VCPU_SET_ATTR: {
5488 struct kvm_xen_vcpu_attr xva;
5491 if (copy_from_user(&xva, argp, sizeof(xva)))
5493 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5497 case KVM_GET_SREGS2: {
5498 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5502 __get_sregs2(vcpu, u.sregs2);
5504 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5509 case KVM_SET_SREGS2: {
5510 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5511 if (IS_ERR(u.sregs2)) {
5512 r = PTR_ERR(u.sregs2);
5516 r = __set_sregs2(vcpu, u.sregs2);
5519 case KVM_HAS_DEVICE_ATTR:
5520 case KVM_GET_DEVICE_ATTR:
5521 case KVM_SET_DEVICE_ATTR:
5522 r = kvm_vcpu_ioctl_device_attr(vcpu, ioctl, argp);
5534 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5536 return VM_FAULT_SIGBUS;
5539 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5543 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5545 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5549 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5552 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5555 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5556 unsigned long kvm_nr_mmu_pages)
5558 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5561 mutex_lock(&kvm->slots_lock);
5563 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5564 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5566 mutex_unlock(&kvm->slots_lock);
5570 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5572 return kvm->arch.n_max_mmu_pages;
5575 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5577 struct kvm_pic *pic = kvm->arch.vpic;
5581 switch (chip->chip_id) {
5582 case KVM_IRQCHIP_PIC_MASTER:
5583 memcpy(&chip->chip.pic, &pic->pics[0],
5584 sizeof(struct kvm_pic_state));
5586 case KVM_IRQCHIP_PIC_SLAVE:
5587 memcpy(&chip->chip.pic, &pic->pics[1],
5588 sizeof(struct kvm_pic_state));
5590 case KVM_IRQCHIP_IOAPIC:
5591 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5600 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5602 struct kvm_pic *pic = kvm->arch.vpic;
5606 switch (chip->chip_id) {
5607 case KVM_IRQCHIP_PIC_MASTER:
5608 spin_lock(&pic->lock);
5609 memcpy(&pic->pics[0], &chip->chip.pic,
5610 sizeof(struct kvm_pic_state));
5611 spin_unlock(&pic->lock);
5613 case KVM_IRQCHIP_PIC_SLAVE:
5614 spin_lock(&pic->lock);
5615 memcpy(&pic->pics[1], &chip->chip.pic,
5616 sizeof(struct kvm_pic_state));
5617 spin_unlock(&pic->lock);
5619 case KVM_IRQCHIP_IOAPIC:
5620 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5626 kvm_pic_update_irq(pic);
5630 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5632 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5634 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5636 mutex_lock(&kps->lock);
5637 memcpy(ps, &kps->channels, sizeof(*ps));
5638 mutex_unlock(&kps->lock);
5642 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5645 struct kvm_pit *pit = kvm->arch.vpit;
5647 mutex_lock(&pit->pit_state.lock);
5648 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5649 for (i = 0; i < 3; i++)
5650 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5651 mutex_unlock(&pit->pit_state.lock);
5655 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5657 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5658 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5659 sizeof(ps->channels));
5660 ps->flags = kvm->arch.vpit->pit_state.flags;
5661 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5662 memset(&ps->reserved, 0, sizeof(ps->reserved));
5666 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5670 u32 prev_legacy, cur_legacy;
5671 struct kvm_pit *pit = kvm->arch.vpit;
5673 mutex_lock(&pit->pit_state.lock);
5674 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5675 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5676 if (!prev_legacy && cur_legacy)
5678 memcpy(&pit->pit_state.channels, &ps->channels,
5679 sizeof(pit->pit_state.channels));
5680 pit->pit_state.flags = ps->flags;
5681 for (i = 0; i < 3; i++)
5682 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5684 mutex_unlock(&pit->pit_state.lock);
5688 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5689 struct kvm_reinject_control *control)
5691 struct kvm_pit *pit = kvm->arch.vpit;
5693 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5694 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5695 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5697 mutex_lock(&pit->pit_state.lock);
5698 kvm_pit_set_reinject(pit, control->pit_reinject);
5699 mutex_unlock(&pit->pit_state.lock);
5704 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5708 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5709 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5710 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5713 struct kvm_vcpu *vcpu;
5716 kvm_for_each_vcpu(i, vcpu, kvm)
5717 kvm_vcpu_kick(vcpu);
5720 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5723 if (!irqchip_in_kernel(kvm))
5726 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5727 irq_event->irq, irq_event->level,
5732 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5733 struct kvm_enable_cap *cap)
5741 case KVM_CAP_DISABLE_QUIRKS:
5742 kvm->arch.disabled_quirks = cap->args[0];
5745 case KVM_CAP_SPLIT_IRQCHIP: {
5746 mutex_lock(&kvm->lock);
5748 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5749 goto split_irqchip_unlock;
5751 if (irqchip_in_kernel(kvm))
5752 goto split_irqchip_unlock;
5753 if (kvm->created_vcpus)
5754 goto split_irqchip_unlock;
5755 r = kvm_setup_empty_irq_routing(kvm);
5757 goto split_irqchip_unlock;
5758 /* Pairs with irqchip_in_kernel. */
5760 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5761 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5762 kvm_request_apicv_update(kvm, true, APICV_INHIBIT_REASON_ABSENT);
5764 split_irqchip_unlock:
5765 mutex_unlock(&kvm->lock);
5768 case KVM_CAP_X2APIC_API:
5770 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5773 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5774 kvm->arch.x2apic_format = true;
5775 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5776 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5780 case KVM_CAP_X86_DISABLE_EXITS:
5782 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5785 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5786 kvm_can_mwait_in_guest())
5787 kvm->arch.mwait_in_guest = true;
5788 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5789 kvm->arch.hlt_in_guest = true;
5790 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5791 kvm->arch.pause_in_guest = true;
5792 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5793 kvm->arch.cstate_in_guest = true;
5796 case KVM_CAP_MSR_PLATFORM_INFO:
5797 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5800 case KVM_CAP_EXCEPTION_PAYLOAD:
5801 kvm->arch.exception_payload_enabled = cap->args[0];
5804 case KVM_CAP_X86_USER_SPACE_MSR:
5805 kvm->arch.user_space_msr_mask = cap->args[0];
5808 case KVM_CAP_X86_BUS_LOCK_EXIT:
5810 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5813 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5814 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5817 if (kvm_has_bus_lock_exit &&
5818 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5819 kvm->arch.bus_lock_detection_enabled = true;
5822 #ifdef CONFIG_X86_SGX_KVM
5823 case KVM_CAP_SGX_ATTRIBUTE: {
5824 unsigned long allowed_attributes = 0;
5826 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5830 /* KVM only supports the PROVISIONKEY privileged attribute. */
5831 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5832 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5833 kvm->arch.sgx_provisioning_allowed = true;
5839 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5841 if (kvm_x86_ops.vm_copy_enc_context_from)
5842 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5844 case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
5846 if (kvm_x86_ops.vm_move_enc_context_from)
5847 r = kvm_x86_ops.vm_move_enc_context_from(
5850 case KVM_CAP_EXIT_HYPERCALL:
5851 if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
5855 kvm->arch.hypercall_exit_enabled = cap->args[0];
5858 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
5860 if (cap->args[0] & ~1)
5862 kvm->arch.exit_on_emulation_error = cap->args[0];
5872 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5874 struct kvm_x86_msr_filter *msr_filter;
5876 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5880 msr_filter->default_allow = default_allow;
5884 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5891 for (i = 0; i < msr_filter->count; i++)
5892 kfree(msr_filter->ranges[i].bitmap);
5897 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5898 struct kvm_msr_filter_range *user_range)
5900 unsigned long *bitmap = NULL;
5903 if (!user_range->nmsrs)
5906 if (user_range->flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE))
5909 if (!user_range->flags)
5912 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5913 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5916 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5918 return PTR_ERR(bitmap);
5920 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
5921 .flags = user_range->flags,
5922 .base = user_range->base,
5923 .nmsrs = user_range->nmsrs,
5927 msr_filter->count++;
5931 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5933 struct kvm_msr_filter __user *user_msr_filter = argp;
5934 struct kvm_x86_msr_filter *new_filter, *old_filter;
5935 struct kvm_msr_filter filter;
5941 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5944 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5945 empty &= !filter.ranges[i].nmsrs;
5947 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5948 if (empty && !default_allow)
5951 new_filter = kvm_alloc_msr_filter(default_allow);
5955 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5956 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5958 kvm_free_msr_filter(new_filter);
5963 mutex_lock(&kvm->lock);
5965 /* The per-VM filter is protected by kvm->lock... */
5966 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5968 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5969 synchronize_srcu(&kvm->srcu);
5971 kvm_free_msr_filter(old_filter);
5973 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5974 mutex_unlock(&kvm->lock);
5979 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
5980 static int kvm_arch_suspend_notifier(struct kvm *kvm)
5982 struct kvm_vcpu *vcpu;
5986 mutex_lock(&kvm->lock);
5987 kvm_for_each_vcpu(i, vcpu, kvm) {
5988 if (!vcpu->arch.pv_time_enabled)
5991 ret = kvm_set_guest_paused(vcpu);
5993 kvm_err("Failed to pause guest VCPU%d: %d\n",
5994 vcpu->vcpu_id, ret);
5998 mutex_unlock(&kvm->lock);
6000 return ret ? NOTIFY_BAD : NOTIFY_DONE;
6003 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
6006 case PM_HIBERNATION_PREPARE:
6007 case PM_SUSPEND_PREPARE:
6008 return kvm_arch_suspend_notifier(kvm);
6013 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
6015 static int kvm_vm_ioctl_get_clock(struct kvm *kvm, void __user *argp)
6017 struct kvm_clock_data data = { 0 };
6019 get_kvmclock(kvm, &data);
6020 if (copy_to_user(argp, &data, sizeof(data)))
6026 static int kvm_vm_ioctl_set_clock(struct kvm *kvm, void __user *argp)
6028 struct kvm_arch *ka = &kvm->arch;
6029 struct kvm_clock_data data;
6032 if (copy_from_user(&data, argp, sizeof(data)))
6036 * Only KVM_CLOCK_REALTIME is used, but allow passing the
6037 * result of KVM_GET_CLOCK back to KVM_SET_CLOCK.
6039 if (data.flags & ~KVM_CLOCK_VALID_FLAGS)
6042 kvm_hv_invalidate_tsc_page(kvm);
6043 kvm_start_pvclock_update(kvm);
6044 pvclock_update_vm_gtod_copy(kvm);
6047 * This pairs with kvm_guest_time_update(): when masterclock is
6048 * in use, we use master_kernel_ns + kvmclock_offset to set
6049 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6050 * is slightly ahead) here we risk going negative on unsigned
6051 * 'system_time' when 'data.clock' is very small.
6053 if (data.flags & KVM_CLOCK_REALTIME) {
6054 u64 now_real_ns = ktime_get_real_ns();
6057 * Avoid stepping the kvmclock backwards.
6059 if (now_real_ns > data.realtime)
6060 data.clock += now_real_ns - data.realtime;
6063 if (ka->use_master_clock)
6064 now_raw_ns = ka->master_kernel_ns;
6066 now_raw_ns = get_kvmclock_base_ns();
6067 ka->kvmclock_offset = data.clock - now_raw_ns;
6068 kvm_end_pvclock_update(kvm);
6072 long kvm_arch_vm_ioctl(struct file *filp,
6073 unsigned int ioctl, unsigned long arg)
6075 struct kvm *kvm = filp->private_data;
6076 void __user *argp = (void __user *)arg;
6079 * This union makes it completely explicit to gcc-3.x
6080 * that these two variables' stack usage should be
6081 * combined, not added together.
6084 struct kvm_pit_state ps;
6085 struct kvm_pit_state2 ps2;
6086 struct kvm_pit_config pit_config;
6090 case KVM_SET_TSS_ADDR:
6091 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
6093 case KVM_SET_IDENTITY_MAP_ADDR: {
6096 mutex_lock(&kvm->lock);
6098 if (kvm->created_vcpus)
6099 goto set_identity_unlock;
6101 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
6102 goto set_identity_unlock;
6103 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
6104 set_identity_unlock:
6105 mutex_unlock(&kvm->lock);
6108 case KVM_SET_NR_MMU_PAGES:
6109 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
6111 case KVM_GET_NR_MMU_PAGES:
6112 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
6114 case KVM_CREATE_IRQCHIP: {
6115 mutex_lock(&kvm->lock);
6118 if (irqchip_in_kernel(kvm))
6119 goto create_irqchip_unlock;
6122 if (kvm->created_vcpus)
6123 goto create_irqchip_unlock;
6125 r = kvm_pic_init(kvm);
6127 goto create_irqchip_unlock;
6129 r = kvm_ioapic_init(kvm);
6131 kvm_pic_destroy(kvm);
6132 goto create_irqchip_unlock;
6135 r = kvm_setup_default_irq_routing(kvm);
6137 kvm_ioapic_destroy(kvm);
6138 kvm_pic_destroy(kvm);
6139 goto create_irqchip_unlock;
6141 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6143 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6144 kvm_request_apicv_update(kvm, true, APICV_INHIBIT_REASON_ABSENT);
6145 create_irqchip_unlock:
6146 mutex_unlock(&kvm->lock);
6149 case KVM_CREATE_PIT:
6150 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6152 case KVM_CREATE_PIT2:
6154 if (copy_from_user(&u.pit_config, argp,
6155 sizeof(struct kvm_pit_config)))
6158 mutex_lock(&kvm->lock);
6161 goto create_pit_unlock;
6163 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6167 mutex_unlock(&kvm->lock);
6169 case KVM_GET_IRQCHIP: {
6170 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6171 struct kvm_irqchip *chip;
6173 chip = memdup_user(argp, sizeof(*chip));
6180 if (!irqchip_kernel(kvm))
6181 goto get_irqchip_out;
6182 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6184 goto get_irqchip_out;
6186 if (copy_to_user(argp, chip, sizeof(*chip)))
6187 goto get_irqchip_out;
6193 case KVM_SET_IRQCHIP: {
6194 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6195 struct kvm_irqchip *chip;
6197 chip = memdup_user(argp, sizeof(*chip));
6204 if (!irqchip_kernel(kvm))
6205 goto set_irqchip_out;
6206 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6213 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6216 if (!kvm->arch.vpit)
6218 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6222 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6229 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6231 mutex_lock(&kvm->lock);
6233 if (!kvm->arch.vpit)
6235 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6237 mutex_unlock(&kvm->lock);
6240 case KVM_GET_PIT2: {
6242 if (!kvm->arch.vpit)
6244 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6248 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6253 case KVM_SET_PIT2: {
6255 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6257 mutex_lock(&kvm->lock);
6259 if (!kvm->arch.vpit)
6261 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6263 mutex_unlock(&kvm->lock);
6266 case KVM_REINJECT_CONTROL: {
6267 struct kvm_reinject_control control;
6269 if (copy_from_user(&control, argp, sizeof(control)))
6272 if (!kvm->arch.vpit)
6274 r = kvm_vm_ioctl_reinject(kvm, &control);
6277 case KVM_SET_BOOT_CPU_ID:
6279 mutex_lock(&kvm->lock);
6280 if (kvm->created_vcpus)
6283 kvm->arch.bsp_vcpu_id = arg;
6284 mutex_unlock(&kvm->lock);
6286 #ifdef CONFIG_KVM_XEN
6287 case KVM_XEN_HVM_CONFIG: {
6288 struct kvm_xen_hvm_config xhc;
6290 if (copy_from_user(&xhc, argp, sizeof(xhc)))
6292 r = kvm_xen_hvm_config(kvm, &xhc);
6295 case KVM_XEN_HVM_GET_ATTR: {
6296 struct kvm_xen_hvm_attr xha;
6299 if (copy_from_user(&xha, argp, sizeof(xha)))
6301 r = kvm_xen_hvm_get_attr(kvm, &xha);
6302 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6306 case KVM_XEN_HVM_SET_ATTR: {
6307 struct kvm_xen_hvm_attr xha;
6310 if (copy_from_user(&xha, argp, sizeof(xha)))
6312 r = kvm_xen_hvm_set_attr(kvm, &xha);
6317 r = kvm_vm_ioctl_set_clock(kvm, argp);
6320 r = kvm_vm_ioctl_get_clock(kvm, argp);
6322 case KVM_MEMORY_ENCRYPT_OP: {
6324 if (kvm_x86_ops.mem_enc_op)
6325 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
6328 case KVM_MEMORY_ENCRYPT_REG_REGION: {
6329 struct kvm_enc_region region;
6332 if (copy_from_user(®ion, argp, sizeof(region)))
6336 if (kvm_x86_ops.mem_enc_reg_region)
6337 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
6340 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6341 struct kvm_enc_region region;
6344 if (copy_from_user(®ion, argp, sizeof(region)))
6348 if (kvm_x86_ops.mem_enc_unreg_region)
6349 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
6352 case KVM_HYPERV_EVENTFD: {
6353 struct kvm_hyperv_eventfd hvevfd;
6356 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6358 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6361 case KVM_SET_PMU_EVENT_FILTER:
6362 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6364 case KVM_X86_SET_MSR_FILTER:
6365 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
6374 static void kvm_init_msr_list(void)
6376 struct x86_pmu_capability x86_pmu;
6380 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
6381 "Please update the fixed PMCs in msrs_to_saved_all[]");
6383 perf_get_x86_pmu_capability(&x86_pmu);
6385 num_msrs_to_save = 0;
6386 num_emulated_msrs = 0;
6387 num_msr_based_features = 0;
6389 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
6390 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
6394 * Even MSRs that are valid in the host may not be exposed
6395 * to the guests in some cases.
6397 switch (msrs_to_save_all[i]) {
6398 case MSR_IA32_BNDCFGS:
6399 if (!kvm_mpx_supported())
6403 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
6404 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
6407 case MSR_IA32_UMWAIT_CONTROL:
6408 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
6411 case MSR_IA32_RTIT_CTL:
6412 case MSR_IA32_RTIT_STATUS:
6413 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
6416 case MSR_IA32_RTIT_CR3_MATCH:
6417 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6418 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
6421 case MSR_IA32_RTIT_OUTPUT_BASE:
6422 case MSR_IA32_RTIT_OUTPUT_MASK:
6423 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6424 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
6425 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
6428 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
6429 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
6430 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
6431 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
6434 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
6435 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
6436 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6439 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
6440 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
6441 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
6448 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
6451 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
6452 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
6455 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
6458 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
6459 struct kvm_msr_entry msr;
6461 msr.index = msr_based_features_all[i];
6462 if (kvm_get_msr_feature(&msr))
6465 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6469 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6477 if (!(lapic_in_kernel(vcpu) &&
6478 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6479 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6490 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6497 if (!(lapic_in_kernel(vcpu) &&
6498 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6500 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6502 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6512 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6513 struct kvm_segment *var, int seg)
6515 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6518 void kvm_get_segment(struct kvm_vcpu *vcpu,
6519 struct kvm_segment *var, int seg)
6521 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6524 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6525 struct x86_exception *exception)
6527 struct kvm_mmu *mmu = vcpu->arch.mmu;
6530 BUG_ON(!mmu_is_nested(vcpu));
6532 /* NPT walks are always user-walks */
6533 access |= PFERR_USER_MASK;
6534 t_gpa = mmu->gva_to_gpa(vcpu, mmu, gpa, access, exception);
6539 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6540 struct x86_exception *exception)
6542 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6544 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6545 return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
6547 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6549 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6550 struct x86_exception *exception)
6552 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6554 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6555 access |= PFERR_FETCH_MASK;
6556 return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
6559 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6560 struct x86_exception *exception)
6562 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6564 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6565 access |= PFERR_WRITE_MASK;
6566 return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
6568 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6570 /* uses this to access any guest's mapped memory without checking CPL */
6571 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6572 struct x86_exception *exception)
6574 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6576 return mmu->gva_to_gpa(vcpu, mmu, gva, 0, exception);
6579 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6580 struct kvm_vcpu *vcpu, u32 access,
6581 struct x86_exception *exception)
6583 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6585 int r = X86EMUL_CONTINUE;
6588 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
6589 unsigned offset = addr & (PAGE_SIZE-1);
6590 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6593 if (gpa == UNMAPPED_GVA)
6594 return X86EMUL_PROPAGATE_FAULT;
6595 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6598 r = X86EMUL_IO_NEEDED;
6610 /* used for instruction fetching */
6611 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6612 gva_t addr, void *val, unsigned int bytes,
6613 struct x86_exception *exception)
6615 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6616 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6617 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6621 /* Inline kvm_read_guest_virt_helper for speed. */
6622 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access|PFERR_FETCH_MASK,
6624 if (unlikely(gpa == UNMAPPED_GVA))
6625 return X86EMUL_PROPAGATE_FAULT;
6627 offset = addr & (PAGE_SIZE-1);
6628 if (WARN_ON(offset + bytes > PAGE_SIZE))
6629 bytes = (unsigned)PAGE_SIZE - offset;
6630 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6632 if (unlikely(ret < 0))
6633 return X86EMUL_IO_NEEDED;
6635 return X86EMUL_CONTINUE;
6638 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6639 gva_t addr, void *val, unsigned int bytes,
6640 struct x86_exception *exception)
6642 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6645 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6646 * is returned, but our callers are not ready for that and they blindly
6647 * call kvm_inject_page_fault. Ensure that they at least do not leak
6648 * uninitialized kernel stack memory into cr2 and error code.
6650 memset(exception, 0, sizeof(*exception));
6651 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6654 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6656 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6657 gva_t addr, void *val, unsigned int bytes,
6658 struct x86_exception *exception, bool system)
6660 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6663 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6664 access |= PFERR_USER_MASK;
6666 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6669 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6670 unsigned long addr, void *val, unsigned int bytes)
6672 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6673 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6675 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6678 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6679 struct kvm_vcpu *vcpu, u32 access,
6680 struct x86_exception *exception)
6682 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6684 int r = X86EMUL_CONTINUE;
6687 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
6688 unsigned offset = addr & (PAGE_SIZE-1);
6689 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6692 if (gpa == UNMAPPED_GVA)
6693 return X86EMUL_PROPAGATE_FAULT;
6694 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6696 r = X86EMUL_IO_NEEDED;
6708 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6709 unsigned int bytes, struct x86_exception *exception,
6712 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6713 u32 access = PFERR_WRITE_MASK;
6715 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6716 access |= PFERR_USER_MASK;
6718 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6722 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6723 unsigned int bytes, struct x86_exception *exception)
6725 /* kvm_write_guest_virt_system can pull in tons of pages. */
6726 vcpu->arch.l1tf_flush_l1d = true;
6728 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6729 PFERR_WRITE_MASK, exception);
6731 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6733 int handle_ud(struct kvm_vcpu *vcpu)
6735 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6736 int emul_type = EMULTYPE_TRAP_UD;
6737 char sig[5]; /* ud2; .ascii "kvm" */
6738 struct x86_exception e;
6740 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6743 if (force_emulation_prefix &&
6744 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6745 sig, sizeof(sig), &e) == 0 &&
6746 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6747 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6748 emul_type = EMULTYPE_TRAP_UD_FORCED;
6751 return kvm_emulate_instruction(vcpu, emul_type);
6753 EXPORT_SYMBOL_GPL(handle_ud);
6755 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6756 gpa_t gpa, bool write)
6758 /* For APIC access vmexit */
6759 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6762 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6763 trace_vcpu_match_mmio(gva, gpa, write, true);
6770 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6771 gpa_t *gpa, struct x86_exception *exception,
6774 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
6775 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6776 | (write ? PFERR_WRITE_MASK : 0);
6779 * currently PKRU is only applied to ept enabled guest so
6780 * there is no pkey in EPT page table for L1 guest or EPT
6781 * shadow page table for L2 guest.
6783 if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
6784 !permission_fault(vcpu, vcpu->arch.walk_mmu,
6785 vcpu->arch.mmio_access, 0, access))) {
6786 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6787 (gva & (PAGE_SIZE - 1));
6788 trace_vcpu_match_mmio(gva, *gpa, write, false);
6792 *gpa = mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
6794 if (*gpa == UNMAPPED_GVA)
6797 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6800 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6801 const void *val, int bytes)
6805 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6808 kvm_page_track_write(vcpu, gpa, val, bytes);
6812 struct read_write_emulator_ops {
6813 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6815 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6816 void *val, int bytes);
6817 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6818 int bytes, void *val);
6819 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6820 void *val, int bytes);
6824 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6826 if (vcpu->mmio_read_completed) {
6827 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6828 vcpu->mmio_fragments[0].gpa, val);
6829 vcpu->mmio_read_completed = 0;
6836 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6837 void *val, int bytes)
6839 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6842 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6843 void *val, int bytes)
6845 return emulator_write_phys(vcpu, gpa, val, bytes);
6848 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6850 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6851 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6854 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6855 void *val, int bytes)
6857 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6858 return X86EMUL_IO_NEEDED;
6861 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6862 void *val, int bytes)
6864 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6866 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6867 return X86EMUL_CONTINUE;
6870 static const struct read_write_emulator_ops read_emultor = {
6871 .read_write_prepare = read_prepare,
6872 .read_write_emulate = read_emulate,
6873 .read_write_mmio = vcpu_mmio_read,
6874 .read_write_exit_mmio = read_exit_mmio,
6877 static const struct read_write_emulator_ops write_emultor = {
6878 .read_write_emulate = write_emulate,
6879 .read_write_mmio = write_mmio,
6880 .read_write_exit_mmio = write_exit_mmio,
6884 static int emulator_read_write_onepage(unsigned long addr, void *val,
6886 struct x86_exception *exception,
6887 struct kvm_vcpu *vcpu,
6888 const struct read_write_emulator_ops *ops)
6892 bool write = ops->write;
6893 struct kvm_mmio_fragment *frag;
6894 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6897 * If the exit was due to a NPF we may already have a GPA.
6898 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6899 * Note, this cannot be used on string operations since string
6900 * operation using rep will only have the initial GPA from the NPF
6903 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6904 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6905 gpa = ctxt->gpa_val;
6906 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6908 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6910 return X86EMUL_PROPAGATE_FAULT;
6913 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6914 return X86EMUL_CONTINUE;
6917 * Is this MMIO handled locally?
6919 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6920 if (handled == bytes)
6921 return X86EMUL_CONTINUE;
6927 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6928 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6932 return X86EMUL_CONTINUE;
6935 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6937 void *val, unsigned int bytes,
6938 struct x86_exception *exception,
6939 const struct read_write_emulator_ops *ops)
6941 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6945 if (ops->read_write_prepare &&
6946 ops->read_write_prepare(vcpu, val, bytes))
6947 return X86EMUL_CONTINUE;
6949 vcpu->mmio_nr_fragments = 0;
6951 /* Crossing a page boundary? */
6952 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6955 now = -addr & ~PAGE_MASK;
6956 rc = emulator_read_write_onepage(addr, val, now, exception,
6959 if (rc != X86EMUL_CONTINUE)
6962 if (ctxt->mode != X86EMUL_MODE_PROT64)
6968 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6970 if (rc != X86EMUL_CONTINUE)
6973 if (!vcpu->mmio_nr_fragments)
6976 gpa = vcpu->mmio_fragments[0].gpa;
6978 vcpu->mmio_needed = 1;
6979 vcpu->mmio_cur_fragment = 0;
6981 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6982 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6983 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6984 vcpu->run->mmio.phys_addr = gpa;
6986 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6989 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6993 struct x86_exception *exception)
6995 return emulator_read_write(ctxt, addr, val, bytes,
6996 exception, &read_emultor);
6999 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
7003 struct x86_exception *exception)
7005 return emulator_read_write(ctxt, addr, (void *)val, bytes,
7006 exception, &write_emultor);
7009 #define CMPXCHG_TYPE(t, ptr, old, new) \
7010 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
7012 #ifdef CONFIG_X86_64
7013 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
7015 # define CMPXCHG64(ptr, old, new) \
7016 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
7019 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
7024 struct x86_exception *exception)
7026 struct kvm_host_map map;
7027 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7033 /* guests cmpxchg8b have to be emulated atomically */
7034 if (bytes > 8 || (bytes & (bytes - 1)))
7037 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
7039 if (gpa == UNMAPPED_GVA ||
7040 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7044 * Emulate the atomic as a straight write to avoid #AC if SLD is
7045 * enabled in the host and the access splits a cache line.
7047 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
7048 page_line_mask = ~(cache_line_size() - 1);
7050 page_line_mask = PAGE_MASK;
7052 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
7055 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
7058 kaddr = map.hva + offset_in_page(gpa);
7062 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
7065 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
7068 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
7071 exchanged = CMPXCHG64(kaddr, old, new);
7077 kvm_vcpu_unmap(vcpu, &map, true);
7080 return X86EMUL_CMPXCHG_FAILED;
7082 kvm_page_track_write(vcpu, gpa, new, bytes);
7084 return X86EMUL_CONTINUE;
7087 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
7089 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
7092 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
7096 for (i = 0; i < vcpu->arch.pio.count; i++) {
7097 if (vcpu->arch.pio.in)
7098 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
7099 vcpu->arch.pio.size, pd);
7101 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
7102 vcpu->arch.pio.port, vcpu->arch.pio.size,
7106 pd += vcpu->arch.pio.size;
7111 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
7112 unsigned short port,
7113 unsigned int count, bool in)
7115 vcpu->arch.pio.port = port;
7116 vcpu->arch.pio.in = in;
7117 vcpu->arch.pio.count = count;
7118 vcpu->arch.pio.size = size;
7120 if (!kernel_pio(vcpu, vcpu->arch.pio_data))
7123 vcpu->run->exit_reason = KVM_EXIT_IO;
7124 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
7125 vcpu->run->io.size = size;
7126 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
7127 vcpu->run->io.count = count;
7128 vcpu->run->io.port = port;
7133 static int __emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7134 unsigned short port, unsigned int count)
7136 WARN_ON(vcpu->arch.pio.count);
7137 memset(vcpu->arch.pio_data, 0, size * count);
7138 return emulator_pio_in_out(vcpu, size, port, count, true);
7141 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
7143 int size = vcpu->arch.pio.size;
7144 unsigned count = vcpu->arch.pio.count;
7145 memcpy(val, vcpu->arch.pio_data, size * count);
7146 trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
7147 vcpu->arch.pio.count = 0;
7150 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7151 unsigned short port, void *val, unsigned int count)
7153 if (vcpu->arch.pio.count) {
7155 * Complete a previous iteration that required userspace I/O.
7156 * Note, @count isn't guaranteed to match pio.count as userspace
7157 * can modify ECX before rerunning the vCPU. Ignore any such
7158 * shenanigans as KVM doesn't support modifying the rep count,
7159 * and the emulator ensures @count doesn't overflow the buffer.
7162 int r = __emulator_pio_in(vcpu, size, port, count);
7166 /* Results already available, fall through. */
7169 complete_emulator_pio_in(vcpu, val);
7173 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7174 int size, unsigned short port, void *val,
7177 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
7181 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7182 unsigned short port, const void *val,
7187 memcpy(vcpu->arch.pio_data, val, size * count);
7188 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
7189 ret = emulator_pio_in_out(vcpu, size, port, count, false);
7191 vcpu->arch.pio.count = 0;
7196 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7197 int size, unsigned short port,
7198 const void *val, unsigned int count)
7200 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7203 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7205 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7208 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7210 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7213 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7215 if (!need_emulate_wbinvd(vcpu))
7216 return X86EMUL_CONTINUE;
7218 if (static_call(kvm_x86_has_wbinvd_exit)()) {
7219 int cpu = get_cpu();
7221 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7222 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7223 wbinvd_ipi, NULL, 1);
7225 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7228 return X86EMUL_CONTINUE;
7231 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7233 kvm_emulate_wbinvd_noskip(vcpu);
7234 return kvm_skip_emulated_instruction(vcpu);
7236 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7240 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7242 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7245 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7246 unsigned long *dest)
7248 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7251 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7252 unsigned long value)
7255 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7258 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7260 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7263 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7265 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7266 unsigned long value;
7270 value = kvm_read_cr0(vcpu);
7273 value = vcpu->arch.cr2;
7276 value = kvm_read_cr3(vcpu);
7279 value = kvm_read_cr4(vcpu);
7282 value = kvm_get_cr8(vcpu);
7285 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7292 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7294 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7299 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7302 vcpu->arch.cr2 = val;
7305 res = kvm_set_cr3(vcpu, val);
7308 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7311 res = kvm_set_cr8(vcpu, val);
7314 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7321 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7323 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7326 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7328 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7331 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7333 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7336 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7338 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7341 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7343 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7346 static unsigned long emulator_get_cached_segment_base(
7347 struct x86_emulate_ctxt *ctxt, int seg)
7349 return get_segment_base(emul_to_vcpu(ctxt), seg);
7352 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7353 struct desc_struct *desc, u32 *base3,
7356 struct kvm_segment var;
7358 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7359 *selector = var.selector;
7362 memset(desc, 0, sizeof(*desc));
7370 set_desc_limit(desc, var.limit);
7371 set_desc_base(desc, (unsigned long)var.base);
7372 #ifdef CONFIG_X86_64
7374 *base3 = var.base >> 32;
7376 desc->type = var.type;
7378 desc->dpl = var.dpl;
7379 desc->p = var.present;
7380 desc->avl = var.avl;
7388 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
7389 struct desc_struct *desc, u32 base3,
7392 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7393 struct kvm_segment var;
7395 var.selector = selector;
7396 var.base = get_desc_base(desc);
7397 #ifdef CONFIG_X86_64
7398 var.base |= ((u64)base3) << 32;
7400 var.limit = get_desc_limit(desc);
7402 var.limit = (var.limit << 12) | 0xfff;
7403 var.type = desc->type;
7404 var.dpl = desc->dpl;
7409 var.avl = desc->avl;
7410 var.present = desc->p;
7411 var.unusable = !var.present;
7414 kvm_set_segment(vcpu, &var, seg);
7418 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
7419 u32 msr_index, u64 *pdata)
7421 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7424 r = kvm_get_msr(vcpu, msr_index, pdata);
7426 if (r && kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
7427 complete_emulated_rdmsr, r)) {
7428 /* Bounce to user space */
7429 return X86EMUL_IO_NEEDED;
7435 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
7436 u32 msr_index, u64 data)
7438 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7441 r = kvm_set_msr(vcpu, msr_index, data);
7443 if (r && kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_WRMSR, data,
7444 complete_emulated_msr_access, r)) {
7445 /* Bounce to user space */
7446 return X86EMUL_IO_NEEDED;
7452 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
7454 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7456 return vcpu->arch.smbase;
7459 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
7461 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7463 vcpu->arch.smbase = smbase;
7466 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
7469 if (kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc))
7474 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
7475 u32 pmc, u64 *pdata)
7477 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
7480 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
7482 emul_to_vcpu(ctxt)->arch.halt_request = 1;
7485 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
7486 struct x86_instruction_info *info,
7487 enum x86_intercept_stage stage)
7489 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
7493 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
7494 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
7497 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
7500 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
7502 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7505 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7507 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7510 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7512 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7515 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7517 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
7520 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7522 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
7525 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7527 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7530 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7532 return emul_to_vcpu(ctxt)->arch.hflags;
7535 static void emulator_exiting_smm(struct x86_emulate_ctxt *ctxt)
7537 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7539 kvm_smm_changed(vcpu, false);
7542 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt,
7543 const char *smstate)
7545 return static_call(kvm_x86_leave_smm)(emul_to_vcpu(ctxt), smstate);
7548 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
7550 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
7553 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7555 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7558 static const struct x86_emulate_ops emulate_ops = {
7559 .read_gpr = emulator_read_gpr,
7560 .write_gpr = emulator_write_gpr,
7561 .read_std = emulator_read_std,
7562 .write_std = emulator_write_std,
7563 .read_phys = kvm_read_guest_phys_system,
7564 .fetch = kvm_fetch_guest_virt,
7565 .read_emulated = emulator_read_emulated,
7566 .write_emulated = emulator_write_emulated,
7567 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7568 .invlpg = emulator_invlpg,
7569 .pio_in_emulated = emulator_pio_in_emulated,
7570 .pio_out_emulated = emulator_pio_out_emulated,
7571 .get_segment = emulator_get_segment,
7572 .set_segment = emulator_set_segment,
7573 .get_cached_segment_base = emulator_get_cached_segment_base,
7574 .get_gdt = emulator_get_gdt,
7575 .get_idt = emulator_get_idt,
7576 .set_gdt = emulator_set_gdt,
7577 .set_idt = emulator_set_idt,
7578 .get_cr = emulator_get_cr,
7579 .set_cr = emulator_set_cr,
7580 .cpl = emulator_get_cpl,
7581 .get_dr = emulator_get_dr,
7582 .set_dr = emulator_set_dr,
7583 .get_smbase = emulator_get_smbase,
7584 .set_smbase = emulator_set_smbase,
7585 .set_msr = emulator_set_msr,
7586 .get_msr = emulator_get_msr,
7587 .check_pmc = emulator_check_pmc,
7588 .read_pmc = emulator_read_pmc,
7589 .halt = emulator_halt,
7590 .wbinvd = emulator_wbinvd,
7591 .fix_hypercall = emulator_fix_hypercall,
7592 .intercept = emulator_intercept,
7593 .get_cpuid = emulator_get_cpuid,
7594 .guest_has_long_mode = emulator_guest_has_long_mode,
7595 .guest_has_movbe = emulator_guest_has_movbe,
7596 .guest_has_fxsr = emulator_guest_has_fxsr,
7597 .set_nmi_mask = emulator_set_nmi_mask,
7598 .get_hflags = emulator_get_hflags,
7599 .exiting_smm = emulator_exiting_smm,
7600 .leave_smm = emulator_leave_smm,
7601 .triple_fault = emulator_triple_fault,
7602 .set_xcr = emulator_set_xcr,
7605 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7607 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7609 * an sti; sti; sequence only disable interrupts for the first
7610 * instruction. So, if the last instruction, be it emulated or
7611 * not, left the system with the INT_STI flag enabled, it
7612 * means that the last instruction is an sti. We should not
7613 * leave the flag on in this case. The same goes for mov ss
7615 if (int_shadow & mask)
7617 if (unlikely(int_shadow || mask)) {
7618 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7620 kvm_make_request(KVM_REQ_EVENT, vcpu);
7624 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7626 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7627 if (ctxt->exception.vector == PF_VECTOR)
7628 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7630 if (ctxt->exception.error_code_valid)
7631 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7632 ctxt->exception.error_code);
7634 kvm_queue_exception(vcpu, ctxt->exception.vector);
7638 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7640 struct x86_emulate_ctxt *ctxt;
7642 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7644 pr_err("kvm: failed to allocate vcpu's emulator\n");
7649 ctxt->ops = &emulate_ops;
7650 vcpu->arch.emulate_ctxt = ctxt;
7655 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7657 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7660 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7662 ctxt->gpa_available = false;
7663 ctxt->eflags = kvm_get_rflags(vcpu);
7664 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7666 ctxt->eip = kvm_rip_read(vcpu);
7667 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7668 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7669 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7670 cs_db ? X86EMUL_MODE_PROT32 :
7671 X86EMUL_MODE_PROT16;
7672 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7673 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7674 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7676 ctxt->interruptibility = 0;
7677 ctxt->have_exception = false;
7678 ctxt->exception.vector = -1;
7679 ctxt->perm_ok = false;
7681 init_decode_cache(ctxt);
7682 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7685 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7687 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7690 init_emulate_ctxt(vcpu);
7694 ctxt->_eip = ctxt->eip + inc_eip;
7695 ret = emulate_int_real(ctxt, irq);
7697 if (ret != X86EMUL_CONTINUE) {
7698 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7700 ctxt->eip = ctxt->_eip;
7701 kvm_rip_write(vcpu, ctxt->eip);
7702 kvm_set_rflags(vcpu, ctxt->eflags);
7705 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7707 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7708 u8 ndata, u8 *insn_bytes, u8 insn_size)
7710 struct kvm_run *run = vcpu->run;
7715 * Zero the whole array used to retrieve the exit info, as casting to
7716 * u32 for select entries will leave some chunks uninitialized.
7718 memset(&info, 0, sizeof(info));
7720 static_call(kvm_x86_get_exit_info)(vcpu, (u32 *)&info[0], &info[1],
7721 &info[2], (u32 *)&info[3],
7724 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7725 run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
7728 * There's currently space for 13 entries, but 5 are used for the exit
7729 * reason and info. Restrict to 4 to reduce the maintenance burden
7730 * when expanding kvm_run.emulation_failure in the future.
7732 if (WARN_ON_ONCE(ndata > 4))
7735 /* Always include the flags as a 'data' entry. */
7737 run->emulation_failure.flags = 0;
7740 BUILD_BUG_ON((sizeof(run->emulation_failure.insn_size) +
7741 sizeof(run->emulation_failure.insn_bytes) != 16));
7743 run->emulation_failure.flags |=
7744 KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
7745 run->emulation_failure.insn_size = insn_size;
7746 memset(run->emulation_failure.insn_bytes, 0x90,
7747 sizeof(run->emulation_failure.insn_bytes));
7748 memcpy(run->emulation_failure.insn_bytes, insn_bytes, insn_size);
7751 memcpy(&run->internal.data[info_start], info, sizeof(info));
7752 memcpy(&run->internal.data[info_start + ARRAY_SIZE(info)], data,
7753 ndata * sizeof(data[0]));
7755 run->emulation_failure.ndata = info_start + ARRAY_SIZE(info) + ndata;
7758 static void prepare_emulation_ctxt_failure_exit(struct kvm_vcpu *vcpu)
7760 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7762 prepare_emulation_failure_exit(vcpu, NULL, 0, ctxt->fetch.data,
7763 ctxt->fetch.end - ctxt->fetch.data);
7766 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
7769 prepare_emulation_failure_exit(vcpu, data, ndata, NULL, 0);
7771 EXPORT_SYMBOL_GPL(__kvm_prepare_emulation_failure_exit);
7773 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
7775 __kvm_prepare_emulation_failure_exit(vcpu, NULL, 0);
7777 EXPORT_SYMBOL_GPL(kvm_prepare_emulation_failure_exit);
7779 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7781 struct kvm *kvm = vcpu->kvm;
7783 ++vcpu->stat.insn_emulation_fail;
7784 trace_kvm_emulate_insn_failed(vcpu);
7786 if (emulation_type & EMULTYPE_VMWARE_GP) {
7787 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7791 if (kvm->arch.exit_on_emulation_error ||
7792 (emulation_type & EMULTYPE_SKIP)) {
7793 prepare_emulation_ctxt_failure_exit(vcpu);
7797 kvm_queue_exception(vcpu, UD_VECTOR);
7799 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7800 prepare_emulation_ctxt_failure_exit(vcpu);
7807 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7808 bool write_fault_to_shadow_pgtable,
7811 gpa_t gpa = cr2_or_gpa;
7814 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7817 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7818 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7821 if (!vcpu->arch.mmu->direct_map) {
7823 * Write permission should be allowed since only
7824 * write access need to be emulated.
7826 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7829 * If the mapping is invalid in guest, let cpu retry
7830 * it to generate fault.
7832 if (gpa == UNMAPPED_GVA)
7837 * Do not retry the unhandleable instruction if it faults on the
7838 * readonly host memory, otherwise it will goto a infinite loop:
7839 * retry instruction -> write #PF -> emulation fail -> retry
7840 * instruction -> ...
7842 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7845 * If the instruction failed on the error pfn, it can not be fixed,
7846 * report the error to userspace.
7848 if (is_error_noslot_pfn(pfn))
7851 kvm_release_pfn_clean(pfn);
7853 /* The instructions are well-emulated on direct mmu. */
7854 if (vcpu->arch.mmu->direct_map) {
7855 unsigned int indirect_shadow_pages;
7857 write_lock(&vcpu->kvm->mmu_lock);
7858 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7859 write_unlock(&vcpu->kvm->mmu_lock);
7861 if (indirect_shadow_pages)
7862 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7868 * if emulation was due to access to shadowed page table
7869 * and it failed try to unshadow page and re-enter the
7870 * guest to let CPU execute the instruction.
7872 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7875 * If the access faults on its page table, it can not
7876 * be fixed by unprotecting shadow page and it should
7877 * be reported to userspace.
7879 return !write_fault_to_shadow_pgtable;
7882 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7883 gpa_t cr2_or_gpa, int emulation_type)
7885 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7886 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7888 last_retry_eip = vcpu->arch.last_retry_eip;
7889 last_retry_addr = vcpu->arch.last_retry_addr;
7892 * If the emulation is caused by #PF and it is non-page_table
7893 * writing instruction, it means the VM-EXIT is caused by shadow
7894 * page protected, we can zap the shadow page and retry this
7895 * instruction directly.
7897 * Note: if the guest uses a non-page-table modifying instruction
7898 * on the PDE that points to the instruction, then we will unmap
7899 * the instruction and go to an infinite loop. So, we cache the
7900 * last retried eip and the last fault address, if we meet the eip
7901 * and the address again, we can break out of the potential infinite
7904 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7906 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7909 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7910 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7913 if (x86_page_table_writing_insn(ctxt))
7916 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7919 vcpu->arch.last_retry_eip = ctxt->eip;
7920 vcpu->arch.last_retry_addr = cr2_or_gpa;
7922 if (!vcpu->arch.mmu->direct_map)
7923 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7925 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7930 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7931 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7933 static void kvm_smm_changed(struct kvm_vcpu *vcpu, bool entering_smm)
7935 trace_kvm_smm_transition(vcpu->vcpu_id, vcpu->arch.smbase, entering_smm);
7938 vcpu->arch.hflags |= HF_SMM_MASK;
7940 vcpu->arch.hflags &= ~(HF_SMM_MASK | HF_SMM_INSIDE_NMI_MASK);
7942 /* Process a latched INIT or SMI, if any. */
7943 kvm_make_request(KVM_REQ_EVENT, vcpu);
7946 * Even if KVM_SET_SREGS2 loaded PDPTRs out of band,
7947 * on SMM exit we still need to reload them from
7950 vcpu->arch.pdptrs_from_userspace = false;
7953 kvm_mmu_reset_context(vcpu);
7956 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7965 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7966 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7971 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7973 struct kvm_run *kvm_run = vcpu->run;
7975 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7976 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7977 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7978 kvm_run->debug.arch.exception = DB_VECTOR;
7979 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7982 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7986 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7988 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7991 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7995 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
7998 * rflags is the old, "raw" value of the flags. The new value has
7999 * not been saved yet.
8001 * This is correct even for TF set by the guest, because "the
8002 * processor will not generate this exception after the instruction
8003 * that sets the TF flag".
8005 if (unlikely(rflags & X86_EFLAGS_TF))
8006 r = kvm_vcpu_do_singlestep(vcpu);
8009 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
8011 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
8013 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
8014 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
8015 struct kvm_run *kvm_run = vcpu->run;
8016 unsigned long eip = kvm_get_linear_rip(vcpu);
8017 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8018 vcpu->arch.guest_debug_dr7,
8022 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
8023 kvm_run->debug.arch.pc = eip;
8024 kvm_run->debug.arch.exception = DB_VECTOR;
8025 kvm_run->exit_reason = KVM_EXIT_DEBUG;
8031 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
8032 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
8033 unsigned long eip = kvm_get_linear_rip(vcpu);
8034 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8039 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
8048 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
8050 switch (ctxt->opcode_len) {
8057 case 0xe6: /* OUT */
8061 case 0x6c: /* INS */
8063 case 0x6e: /* OUTS */
8070 case 0x33: /* RDPMC */
8080 * Decode to be emulated instruction. Return EMULATION_OK if success.
8082 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
8083 void *insn, int insn_len)
8085 int r = EMULATION_OK;
8086 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8088 init_emulate_ctxt(vcpu);
8091 * We will reenter on the same instruction since we do not set
8092 * complete_userspace_io. This does not handle watchpoints yet,
8093 * those would be handled in the emulate_ops.
8095 if (!(emulation_type & EMULTYPE_SKIP) &&
8096 kvm_vcpu_check_breakpoint(vcpu, &r))
8099 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
8101 trace_kvm_emulate_insn_start(vcpu);
8102 ++vcpu->stat.insn_emulation;
8106 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
8108 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8109 int emulation_type, void *insn, int insn_len)
8112 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8113 bool writeback = true;
8114 bool write_fault_to_spt;
8116 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
8119 vcpu->arch.l1tf_flush_l1d = true;
8122 * Clear write_fault_to_shadow_pgtable here to ensure it is
8125 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
8126 vcpu->arch.write_fault_to_shadow_pgtable = false;
8128 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8129 kvm_clear_exception_queue(vcpu);
8131 r = x86_decode_emulated_instruction(vcpu, emulation_type,
8133 if (r != EMULATION_OK) {
8134 if ((emulation_type & EMULTYPE_TRAP_UD) ||
8135 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
8136 kvm_queue_exception(vcpu, UD_VECTOR);
8139 if (reexecute_instruction(vcpu, cr2_or_gpa,
8143 if (ctxt->have_exception) {
8145 * #UD should result in just EMULATION_FAILED, and trap-like
8146 * exception should not be encountered during decode.
8148 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8149 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8150 inject_emulated_exception(vcpu);
8153 return handle_emulation_failure(vcpu, emulation_type);
8157 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8158 !is_vmware_backdoor_opcode(ctxt)) {
8159 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8164 * EMULTYPE_SKIP without EMULTYPE_COMPLETE_USER_EXIT is intended for
8165 * use *only* by vendor callbacks for kvm_skip_emulated_instruction().
8166 * The caller is responsible for updating interruptibility state and
8167 * injecting single-step #DBs.
8169 if (emulation_type & EMULTYPE_SKIP) {
8170 if (ctxt->mode != X86EMUL_MODE_PROT64)
8171 ctxt->eip = (u32)ctxt->_eip;
8173 ctxt->eip = ctxt->_eip;
8175 if (emulation_type & EMULTYPE_COMPLETE_USER_EXIT) {
8180 kvm_rip_write(vcpu, ctxt->eip);
8181 if (ctxt->eflags & X86_EFLAGS_RF)
8182 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8186 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8189 /* this is needed for vmware backdoor interface to work since it
8190 changes registers values during IO operation */
8191 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8192 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8193 emulator_invalidate_register_cache(ctxt);
8197 if (emulation_type & EMULTYPE_PF) {
8198 /* Save the faulting GPA (cr2) in the address field */
8199 ctxt->exception.address = cr2_or_gpa;
8201 /* With shadow page tables, cr2 contains a GVA or nGPA. */
8202 if (vcpu->arch.mmu->direct_map) {
8203 ctxt->gpa_available = true;
8204 ctxt->gpa_val = cr2_or_gpa;
8207 /* Sanitize the address out of an abundance of paranoia. */
8208 ctxt->exception.address = 0;
8211 r = x86_emulate_insn(ctxt);
8213 if (r == EMULATION_INTERCEPTED)
8216 if (r == EMULATION_FAILED) {
8217 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
8221 return handle_emulation_failure(vcpu, emulation_type);
8224 if (ctxt->have_exception) {
8226 if (inject_emulated_exception(vcpu))
8228 } else if (vcpu->arch.pio.count) {
8229 if (!vcpu->arch.pio.in) {
8230 /* FIXME: return into emulator if single-stepping. */
8231 vcpu->arch.pio.count = 0;
8234 vcpu->arch.complete_userspace_io = complete_emulated_pio;
8237 } else if (vcpu->mmio_needed) {
8238 ++vcpu->stat.mmio_exits;
8240 if (!vcpu->mmio_is_write)
8243 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8244 } else if (vcpu->arch.complete_userspace_io) {
8247 } else if (r == EMULATION_RESTART)
8254 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8255 toggle_interruptibility(vcpu, ctxt->interruptibility);
8256 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8257 if (!ctxt->have_exception ||
8258 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
8259 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8260 if (ctxt->is_branch)
8261 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
8262 kvm_rip_write(vcpu, ctxt->eip);
8263 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
8264 r = kvm_vcpu_do_singlestep(vcpu);
8265 if (kvm_x86_ops.update_emulated_instruction)
8266 static_call(kvm_x86_update_emulated_instruction)(vcpu);
8267 __kvm_set_rflags(vcpu, ctxt->eflags);
8271 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
8272 * do nothing, and it will be requested again as soon as
8273 * the shadow expires. But we still need to check here,
8274 * because POPF has no interrupt shadow.
8276 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
8277 kvm_make_request(KVM_REQ_EVENT, vcpu);
8279 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
8284 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
8286 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
8288 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
8290 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
8291 void *insn, int insn_len)
8293 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
8295 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8297 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8299 vcpu->arch.pio.count = 0;
8303 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8305 vcpu->arch.pio.count = 0;
8307 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8310 return kvm_skip_emulated_instruction(vcpu);
8313 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8314 unsigned short port)
8316 unsigned long val = kvm_rax_read(vcpu);
8317 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8323 * Workaround userspace that relies on old KVM behavior of %rip being
8324 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8327 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8328 vcpu->arch.complete_userspace_io =
8329 complete_fast_pio_out_port_0x7e;
8330 kvm_skip_emulated_instruction(vcpu);
8332 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8333 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8338 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8342 /* We should only ever be called with arch.pio.count equal to 1 */
8343 BUG_ON(vcpu->arch.pio.count != 1);
8345 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8346 vcpu->arch.pio.count = 0;
8350 /* For size less than 4 we merge, else we zero extend */
8351 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
8354 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
8355 * the copy and tracing
8357 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
8358 kvm_rax_write(vcpu, val);
8360 return kvm_skip_emulated_instruction(vcpu);
8363 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
8364 unsigned short port)
8369 /* For size less than 4 we merge, else we zero extend */
8370 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
8372 ret = emulator_pio_in(vcpu, size, port, &val, 1);
8374 kvm_rax_write(vcpu, val);
8378 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8379 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
8384 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
8389 ret = kvm_fast_pio_in(vcpu, size, port);
8391 ret = kvm_fast_pio_out(vcpu, size, port);
8392 return ret && kvm_skip_emulated_instruction(vcpu);
8394 EXPORT_SYMBOL_GPL(kvm_fast_pio);
8396 static int kvmclock_cpu_down_prep(unsigned int cpu)
8398 __this_cpu_write(cpu_tsc_khz, 0);
8402 static void tsc_khz_changed(void *data)
8404 struct cpufreq_freqs *freq = data;
8405 unsigned long khz = 0;
8409 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8410 khz = cpufreq_quick_get(raw_smp_processor_id());
8413 __this_cpu_write(cpu_tsc_khz, khz);
8416 #ifdef CONFIG_X86_64
8417 static void kvm_hyperv_tsc_notifier(void)
8422 mutex_lock(&kvm_lock);
8423 list_for_each_entry(kvm, &vm_list, vm_list)
8424 kvm_make_mclock_inprogress_request(kvm);
8426 /* no guest entries from this point */
8427 hyperv_stop_tsc_emulation();
8429 /* TSC frequency always matches when on Hyper-V */
8430 for_each_present_cpu(cpu)
8431 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
8432 kvm_max_guest_tsc_khz = tsc_khz;
8434 list_for_each_entry(kvm, &vm_list, vm_list) {
8435 __kvm_start_pvclock_update(kvm);
8436 pvclock_update_vm_gtod_copy(kvm);
8437 kvm_end_pvclock_update(kvm);
8440 mutex_unlock(&kvm_lock);
8444 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
8447 struct kvm_vcpu *vcpu;
8452 * We allow guests to temporarily run on slowing clocks,
8453 * provided we notify them after, or to run on accelerating
8454 * clocks, provided we notify them before. Thus time never
8457 * However, we have a problem. We can't atomically update
8458 * the frequency of a given CPU from this function; it is
8459 * merely a notifier, which can be called from any CPU.
8460 * Changing the TSC frequency at arbitrary points in time
8461 * requires a recomputation of local variables related to
8462 * the TSC for each VCPU. We must flag these local variables
8463 * to be updated and be sure the update takes place with the
8464 * new frequency before any guests proceed.
8466 * Unfortunately, the combination of hotplug CPU and frequency
8467 * change creates an intractable locking scenario; the order
8468 * of when these callouts happen is undefined with respect to
8469 * CPU hotplug, and they can race with each other. As such,
8470 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
8471 * undefined; you can actually have a CPU frequency change take
8472 * place in between the computation of X and the setting of the
8473 * variable. To protect against this problem, all updates of
8474 * the per_cpu tsc_khz variable are done in an interrupt
8475 * protected IPI, and all callers wishing to update the value
8476 * must wait for a synchronous IPI to complete (which is trivial
8477 * if the caller is on the CPU already). This establishes the
8478 * necessary total order on variable updates.
8480 * Note that because a guest time update may take place
8481 * anytime after the setting of the VCPU's request bit, the
8482 * correct TSC value must be set before the request. However,
8483 * to ensure the update actually makes it to any guest which
8484 * starts running in hardware virtualization between the set
8485 * and the acquisition of the spinlock, we must also ping the
8486 * CPU after setting the request bit.
8490 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8492 mutex_lock(&kvm_lock);
8493 list_for_each_entry(kvm, &vm_list, vm_list) {
8494 kvm_for_each_vcpu(i, vcpu, kvm) {
8495 if (vcpu->cpu != cpu)
8497 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8498 if (vcpu->cpu != raw_smp_processor_id())
8502 mutex_unlock(&kvm_lock);
8504 if (freq->old < freq->new && send_ipi) {
8506 * We upscale the frequency. Must make the guest
8507 * doesn't see old kvmclock values while running with
8508 * the new frequency, otherwise we risk the guest sees
8509 * time go backwards.
8511 * In case we update the frequency for another cpu
8512 * (which might be in guest context) send an interrupt
8513 * to kick the cpu out of guest context. Next time
8514 * guest context is entered kvmclock will be updated,
8515 * so the guest will not see stale values.
8517 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
8521 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
8524 struct cpufreq_freqs *freq = data;
8527 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
8529 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
8532 for_each_cpu(cpu, freq->policy->cpus)
8533 __kvmclock_cpufreq_notifier(freq, cpu);
8538 static struct notifier_block kvmclock_cpufreq_notifier_block = {
8539 .notifier_call = kvmclock_cpufreq_notifier
8542 static int kvmclock_cpu_online(unsigned int cpu)
8544 tsc_khz_changed(NULL);
8548 static void kvm_timer_init(void)
8550 max_tsc_khz = tsc_khz;
8552 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
8553 #ifdef CONFIG_CPU_FREQ
8554 struct cpufreq_policy *policy;
8558 policy = cpufreq_cpu_get(cpu);
8560 if (policy->cpuinfo.max_freq)
8561 max_tsc_khz = policy->cpuinfo.max_freq;
8562 cpufreq_cpu_put(policy);
8566 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
8567 CPUFREQ_TRANSITION_NOTIFIER);
8570 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
8571 kvmclock_cpu_online, kvmclock_cpu_down_prep);
8574 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
8575 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
8577 int kvm_is_in_guest(void)
8579 return __this_cpu_read(current_vcpu) != NULL;
8582 static int kvm_is_user_mode(void)
8586 if (__this_cpu_read(current_vcpu))
8587 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
8589 return user_mode != 0;
8592 static unsigned long kvm_get_guest_ip(void)
8594 unsigned long ip = 0;
8596 if (__this_cpu_read(current_vcpu))
8597 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8602 static void kvm_handle_intel_pt_intr(void)
8604 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8606 kvm_make_request(KVM_REQ_PMI, vcpu);
8607 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8608 (unsigned long *)&vcpu->arch.pmu.global_status);
8611 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8612 .is_in_guest = kvm_is_in_guest,
8613 .is_user_mode = kvm_is_user_mode,
8614 .get_guest_ip = kvm_get_guest_ip,
8615 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8618 #ifdef CONFIG_X86_64
8619 static void pvclock_gtod_update_fn(struct work_struct *work)
8622 struct kvm_vcpu *vcpu;
8625 mutex_lock(&kvm_lock);
8626 list_for_each_entry(kvm, &vm_list, vm_list)
8627 kvm_for_each_vcpu(i, vcpu, kvm)
8628 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8629 atomic_set(&kvm_guest_has_master_clock, 0);
8630 mutex_unlock(&kvm_lock);
8633 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8636 * Indirection to move queue_work() out of the tk_core.seq write held
8637 * region to prevent possible deadlocks against time accessors which
8638 * are invoked with work related locks held.
8640 static void pvclock_irq_work_fn(struct irq_work *w)
8642 queue_work(system_long_wq, &pvclock_gtod_work);
8645 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
8648 * Notification about pvclock gtod data update.
8650 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8653 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8654 struct timekeeper *tk = priv;
8656 update_pvclock_gtod(tk);
8659 * Disable master clock if host does not trust, or does not use,
8660 * TSC based clocksource. Delegate queue_work() to irq_work as
8661 * this is invoked with tk_core.seq write held.
8663 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8664 atomic_read(&kvm_guest_has_master_clock) != 0)
8665 irq_work_queue(&pvclock_irq_work);
8669 static struct notifier_block pvclock_gtod_notifier = {
8670 .notifier_call = pvclock_gtod_notify,
8674 int kvm_arch_init(void *opaque)
8676 struct kvm_x86_init_ops *ops = opaque;
8679 if (kvm_x86_ops.hardware_enable) {
8680 pr_err("kvm: already loaded vendor module '%s'\n", kvm_x86_ops.name);
8685 if (!ops->cpu_has_kvm_support()) {
8686 pr_err_ratelimited("kvm: no hardware support for '%s'\n",
8687 ops->runtime_ops->name);
8691 if (ops->disabled_by_bios()) {
8692 pr_err_ratelimited("kvm: support for '%s' disabled by bios\n",
8693 ops->runtime_ops->name);
8699 * KVM explicitly assumes that the guest has an FPU and
8700 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8701 * vCPU's FPU state as a fxregs_state struct.
8703 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8704 printk(KERN_ERR "kvm: inadequate fpu\n");
8711 x86_emulator_cache = kvm_alloc_emulator_cache();
8712 if (!x86_emulator_cache) {
8713 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8717 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8718 if (!user_return_msrs) {
8719 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8720 goto out_free_x86_emulator_cache;
8722 kvm_nr_uret_msrs = 0;
8724 r = kvm_mmu_module_init();
8726 goto out_free_percpu;
8730 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8732 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8733 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8734 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8737 if (pi_inject_timer == -1)
8738 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8739 #ifdef CONFIG_X86_64
8740 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8742 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8743 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8749 free_percpu(user_return_msrs);
8750 out_free_x86_emulator_cache:
8751 kmem_cache_destroy(x86_emulator_cache);
8756 void kvm_arch_exit(void)
8758 #ifdef CONFIG_X86_64
8759 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8760 clear_hv_tscchange_cb();
8763 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8765 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8766 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8767 CPUFREQ_TRANSITION_NOTIFIER);
8768 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8769 #ifdef CONFIG_X86_64
8770 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8771 irq_work_sync(&pvclock_irq_work);
8772 cancel_work_sync(&pvclock_gtod_work);
8774 kvm_x86_ops.hardware_enable = NULL;
8775 kvm_mmu_module_exit();
8776 free_percpu(user_return_msrs);
8777 kmem_cache_destroy(x86_emulator_cache);
8778 #ifdef CONFIG_KVM_XEN
8779 static_key_deferred_flush(&kvm_xen_enabled);
8780 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8784 static int __kvm_emulate_halt(struct kvm_vcpu *vcpu, int state, int reason)
8787 * The vCPU has halted, e.g. executed HLT. Update the run state if the
8788 * local APIC is in-kernel, the run loop will detect the non-runnable
8789 * state and halt the vCPU. Exit to userspace if the local APIC is
8790 * managed by userspace, in which case userspace is responsible for
8791 * handling wake events.
8793 ++vcpu->stat.halt_exits;
8794 if (lapic_in_kernel(vcpu)) {
8795 vcpu->arch.mp_state = state;
8798 vcpu->run->exit_reason = reason;
8803 int kvm_emulate_halt_noskip(struct kvm_vcpu *vcpu)
8805 return __kvm_emulate_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8807 EXPORT_SYMBOL_GPL(kvm_emulate_halt_noskip);
8809 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8811 int ret = kvm_skip_emulated_instruction(vcpu);
8813 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8814 * KVM_EXIT_DEBUG here.
8816 return kvm_emulate_halt_noskip(vcpu) && ret;
8818 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8820 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8822 int ret = kvm_skip_emulated_instruction(vcpu);
8824 return __kvm_emulate_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD,
8825 KVM_EXIT_AP_RESET_HOLD) && ret;
8827 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8829 #ifdef CONFIG_X86_64
8830 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8831 unsigned long clock_type)
8833 struct kvm_clock_pairing clock_pairing;
8834 struct timespec64 ts;
8838 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8839 return -KVM_EOPNOTSUPP;
8841 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8842 return -KVM_EOPNOTSUPP;
8844 clock_pairing.sec = ts.tv_sec;
8845 clock_pairing.nsec = ts.tv_nsec;
8846 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8847 clock_pairing.flags = 0;
8848 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8851 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8852 sizeof(struct kvm_clock_pairing)))
8860 * kvm_pv_kick_cpu_op: Kick a vcpu.
8862 * @apicid - apicid of vcpu to be kicked.
8864 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8866 struct kvm_lapic_irq lapic_irq;
8868 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8869 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8870 lapic_irq.level = 0;
8871 lapic_irq.dest_id = apicid;
8872 lapic_irq.msi_redir_hint = false;
8874 lapic_irq.delivery_mode = APIC_DM_REMRD;
8875 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8878 bool kvm_apicv_activated(struct kvm *kvm)
8880 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8882 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8884 static void kvm_apicv_init(struct kvm *kvm)
8886 init_rwsem(&kvm->arch.apicv_update_lock);
8888 set_bit(APICV_INHIBIT_REASON_ABSENT,
8889 &kvm->arch.apicv_inhibit_reasons);
8891 set_bit(APICV_INHIBIT_REASON_DISABLE,
8892 &kvm->arch.apicv_inhibit_reasons);
8895 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8897 struct kvm_vcpu *target = NULL;
8898 struct kvm_apic_map *map;
8900 vcpu->stat.directed_yield_attempted++;
8902 if (single_task_running())
8906 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8908 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8909 target = map->phys_map[dest_id]->vcpu;
8913 if (!target || !READ_ONCE(target->ready))
8916 /* Ignore requests to yield to self */
8920 if (kvm_vcpu_yield_to(target) <= 0)
8923 vcpu->stat.directed_yield_successful++;
8929 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
8931 u64 ret = vcpu->run->hypercall.ret;
8933 if (!is_64_bit_mode(vcpu))
8935 kvm_rax_write(vcpu, ret);
8936 ++vcpu->stat.hypercalls;
8937 return kvm_skip_emulated_instruction(vcpu);
8940 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8942 unsigned long nr, a0, a1, a2, a3, ret;
8945 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8946 return kvm_xen_hypercall(vcpu);
8948 if (kvm_hv_hypercall_enabled(vcpu))
8949 return kvm_hv_hypercall(vcpu);
8951 nr = kvm_rax_read(vcpu);
8952 a0 = kvm_rbx_read(vcpu);
8953 a1 = kvm_rcx_read(vcpu);
8954 a2 = kvm_rdx_read(vcpu);
8955 a3 = kvm_rsi_read(vcpu);
8957 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8959 op_64_bit = is_64_bit_hypercall(vcpu);
8968 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8976 case KVM_HC_VAPIC_POLL_IRQ:
8979 case KVM_HC_KICK_CPU:
8980 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8983 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8984 kvm_sched_yield(vcpu, a1);
8987 #ifdef CONFIG_X86_64
8988 case KVM_HC_CLOCK_PAIRING:
8989 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8992 case KVM_HC_SEND_IPI:
8993 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8996 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8998 case KVM_HC_SCHED_YIELD:
8999 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
9002 kvm_sched_yield(vcpu, a0);
9005 case KVM_HC_MAP_GPA_RANGE: {
9006 u64 gpa = a0, npages = a1, attrs = a2;
9009 if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
9012 if (!PAGE_ALIGNED(gpa) || !npages ||
9013 gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
9018 vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
9019 vcpu->run->hypercall.nr = KVM_HC_MAP_GPA_RANGE;
9020 vcpu->run->hypercall.args[0] = gpa;
9021 vcpu->run->hypercall.args[1] = npages;
9022 vcpu->run->hypercall.args[2] = attrs;
9023 vcpu->run->hypercall.longmode = op_64_bit;
9024 vcpu->arch.complete_userspace_io = complete_hypercall_exit;
9034 kvm_rax_write(vcpu, ret);
9036 ++vcpu->stat.hypercalls;
9037 return kvm_skip_emulated_instruction(vcpu);
9039 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
9041 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
9043 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
9044 char instruction[3];
9045 unsigned long rip = kvm_rip_read(vcpu);
9047 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
9049 return emulator_write_emulated(ctxt, rip, instruction, 3,
9053 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
9055 return vcpu->run->request_interrupt_window &&
9056 likely(!pic_in_kernel(vcpu->kvm));
9059 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
9061 struct kvm_run *kvm_run = vcpu->run;
9063 kvm_run->if_flag = static_call(kvm_x86_get_if_flag)(vcpu);
9064 kvm_run->cr8 = kvm_get_cr8(vcpu);
9065 kvm_run->apic_base = kvm_get_apic_base(vcpu);
9068 * The call to kvm_ready_for_interrupt_injection() may end up in
9069 * kvm_xen_has_interrupt() which may require the srcu lock to be
9070 * held, to protect against changes in the vcpu_info address.
9072 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9073 kvm_run->ready_for_interrupt_injection =
9074 pic_in_kernel(vcpu->kvm) ||
9075 kvm_vcpu_ready_for_interrupt_injection(vcpu);
9076 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9079 kvm_run->flags |= KVM_RUN_X86_SMM;
9082 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
9086 if (!kvm_x86_ops.update_cr8_intercept)
9089 if (!lapic_in_kernel(vcpu))
9092 if (vcpu->arch.apicv_active)
9095 if (!vcpu->arch.apic->vapic_addr)
9096 max_irr = kvm_lapic_find_highest_irr(vcpu);
9103 tpr = kvm_lapic_get_cr8(vcpu);
9105 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
9109 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
9111 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9112 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9116 return kvm_x86_ops.nested_ops->check_events(vcpu);
9119 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
9121 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
9122 vcpu->arch.exception.error_code = false;
9123 static_call(kvm_x86_queue_exception)(vcpu);
9126 static int inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
9129 bool can_inject = true;
9131 /* try to reinject previous events if any */
9133 if (vcpu->arch.exception.injected) {
9134 kvm_inject_exception(vcpu);
9138 * Do not inject an NMI or interrupt if there is a pending
9139 * exception. Exceptions and interrupts are recognized at
9140 * instruction boundaries, i.e. the start of an instruction.
9141 * Trap-like exceptions, e.g. #DB, have higher priority than
9142 * NMIs and interrupts, i.e. traps are recognized before an
9143 * NMI/interrupt that's pending on the same instruction.
9144 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
9145 * priority, but are only generated (pended) during instruction
9146 * execution, i.e. a pending fault-like exception means the
9147 * fault occurred on the *previous* instruction and must be
9148 * serviced prior to recognizing any new events in order to
9149 * fully complete the previous instruction.
9151 else if (!vcpu->arch.exception.pending) {
9152 if (vcpu->arch.nmi_injected) {
9153 static_call(kvm_x86_set_nmi)(vcpu);
9155 } else if (vcpu->arch.interrupt.injected) {
9156 static_call(kvm_x86_set_irq)(vcpu);
9161 WARN_ON_ONCE(vcpu->arch.exception.injected &&
9162 vcpu->arch.exception.pending);
9165 * Call check_nested_events() even if we reinjected a previous event
9166 * in order for caller to determine if it should require immediate-exit
9167 * from L2 to L1 due to pending L1 events which require exit
9170 if (is_guest_mode(vcpu)) {
9171 r = kvm_check_nested_events(vcpu);
9176 /* try to inject new event if pending */
9177 if (vcpu->arch.exception.pending) {
9178 trace_kvm_inj_exception(vcpu->arch.exception.nr,
9179 vcpu->arch.exception.has_error_code,
9180 vcpu->arch.exception.error_code);
9182 vcpu->arch.exception.pending = false;
9183 vcpu->arch.exception.injected = true;
9185 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
9186 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
9189 if (vcpu->arch.exception.nr == DB_VECTOR) {
9190 kvm_deliver_exception_payload(vcpu);
9191 if (vcpu->arch.dr7 & DR7_GD) {
9192 vcpu->arch.dr7 &= ~DR7_GD;
9193 kvm_update_dr7(vcpu);
9197 kvm_inject_exception(vcpu);
9201 /* Don't inject interrupts if the user asked to avoid doing so */
9202 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
9206 * Finally, inject interrupt events. If an event cannot be injected
9207 * due to architectural conditions (e.g. IF=0) a window-open exit
9208 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
9209 * and can architecturally be injected, but we cannot do it right now:
9210 * an interrupt could have arrived just now and we have to inject it
9211 * as a vmexit, or there could already an event in the queue, which is
9212 * indicated by can_inject. In that case we request an immediate exit
9213 * in order to make progress and get back here for another iteration.
9214 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
9216 if (vcpu->arch.smi_pending) {
9217 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
9221 vcpu->arch.smi_pending = false;
9222 ++vcpu->arch.smi_count;
9226 static_call(kvm_x86_enable_smi_window)(vcpu);
9229 if (vcpu->arch.nmi_pending) {
9230 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
9234 --vcpu->arch.nmi_pending;
9235 vcpu->arch.nmi_injected = true;
9236 static_call(kvm_x86_set_nmi)(vcpu);
9238 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
9240 if (vcpu->arch.nmi_pending)
9241 static_call(kvm_x86_enable_nmi_window)(vcpu);
9244 if (kvm_cpu_has_injectable_intr(vcpu)) {
9245 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
9249 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
9250 static_call(kvm_x86_set_irq)(vcpu);
9251 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
9253 if (kvm_cpu_has_injectable_intr(vcpu))
9254 static_call(kvm_x86_enable_irq_window)(vcpu);
9257 if (is_guest_mode(vcpu) &&
9258 kvm_x86_ops.nested_ops->hv_timer_pending &&
9259 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
9260 *req_immediate_exit = true;
9262 WARN_ON(vcpu->arch.exception.pending);
9267 *req_immediate_exit = true;
9273 static void process_nmi(struct kvm_vcpu *vcpu)
9278 * x86 is limited to one NMI running, and one NMI pending after it.
9279 * If an NMI is already in progress, limit further NMIs to just one.
9280 * Otherwise, allow two (and we'll inject the first one immediately).
9282 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
9285 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
9286 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
9287 kvm_make_request(KVM_REQ_EVENT, vcpu);
9290 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
9293 flags |= seg->g << 23;
9294 flags |= seg->db << 22;
9295 flags |= seg->l << 21;
9296 flags |= seg->avl << 20;
9297 flags |= seg->present << 15;
9298 flags |= seg->dpl << 13;
9299 flags |= seg->s << 12;
9300 flags |= seg->type << 8;
9304 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
9306 struct kvm_segment seg;
9309 kvm_get_segment(vcpu, &seg, n);
9310 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
9313 offset = 0x7f84 + n * 12;
9315 offset = 0x7f2c + (n - 3) * 12;
9317 put_smstate(u32, buf, offset + 8, seg.base);
9318 put_smstate(u32, buf, offset + 4, seg.limit);
9319 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
9322 #ifdef CONFIG_X86_64
9323 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
9325 struct kvm_segment seg;
9329 kvm_get_segment(vcpu, &seg, n);
9330 offset = 0x7e00 + n * 16;
9332 flags = enter_smm_get_segment_flags(&seg) >> 8;
9333 put_smstate(u16, buf, offset, seg.selector);
9334 put_smstate(u16, buf, offset + 2, flags);
9335 put_smstate(u32, buf, offset + 4, seg.limit);
9336 put_smstate(u64, buf, offset + 8, seg.base);
9340 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
9343 struct kvm_segment seg;
9347 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
9348 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
9349 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
9350 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
9352 for (i = 0; i < 8; i++)
9353 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read_raw(vcpu, i));
9355 kvm_get_dr(vcpu, 6, &val);
9356 put_smstate(u32, buf, 0x7fcc, (u32)val);
9357 kvm_get_dr(vcpu, 7, &val);
9358 put_smstate(u32, buf, 0x7fc8, (u32)val);
9360 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9361 put_smstate(u32, buf, 0x7fc4, seg.selector);
9362 put_smstate(u32, buf, 0x7f64, seg.base);
9363 put_smstate(u32, buf, 0x7f60, seg.limit);
9364 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
9366 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9367 put_smstate(u32, buf, 0x7fc0, seg.selector);
9368 put_smstate(u32, buf, 0x7f80, seg.base);
9369 put_smstate(u32, buf, 0x7f7c, seg.limit);
9370 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
9372 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9373 put_smstate(u32, buf, 0x7f74, dt.address);
9374 put_smstate(u32, buf, 0x7f70, dt.size);
9376 static_call(kvm_x86_get_idt)(vcpu, &dt);
9377 put_smstate(u32, buf, 0x7f58, dt.address);
9378 put_smstate(u32, buf, 0x7f54, dt.size);
9380 for (i = 0; i < 6; i++)
9381 enter_smm_save_seg_32(vcpu, buf, i);
9383 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
9386 put_smstate(u32, buf, 0x7efc, 0x00020000);
9387 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
9390 #ifdef CONFIG_X86_64
9391 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
9394 struct kvm_segment seg;
9398 for (i = 0; i < 16; i++)
9399 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read_raw(vcpu, i));
9401 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
9402 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
9404 kvm_get_dr(vcpu, 6, &val);
9405 put_smstate(u64, buf, 0x7f68, val);
9406 kvm_get_dr(vcpu, 7, &val);
9407 put_smstate(u64, buf, 0x7f60, val);
9409 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
9410 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
9411 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
9413 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
9416 put_smstate(u32, buf, 0x7efc, 0x00020064);
9418 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
9420 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
9421 put_smstate(u16, buf, 0x7e90, seg.selector);
9422 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
9423 put_smstate(u32, buf, 0x7e94, seg.limit);
9424 put_smstate(u64, buf, 0x7e98, seg.base);
9426 static_call(kvm_x86_get_idt)(vcpu, &dt);
9427 put_smstate(u32, buf, 0x7e84, dt.size);
9428 put_smstate(u64, buf, 0x7e88, dt.address);
9430 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
9431 put_smstate(u16, buf, 0x7e70, seg.selector);
9432 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
9433 put_smstate(u32, buf, 0x7e74, seg.limit);
9434 put_smstate(u64, buf, 0x7e78, seg.base);
9436 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9437 put_smstate(u32, buf, 0x7e64, dt.size);
9438 put_smstate(u64, buf, 0x7e68, dt.address);
9440 for (i = 0; i < 6; i++)
9441 enter_smm_save_seg_64(vcpu, buf, i);
9445 static void enter_smm(struct kvm_vcpu *vcpu)
9447 struct kvm_segment cs, ds;
9452 memset(buf, 0, 512);
9453 #ifdef CONFIG_X86_64
9454 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9455 enter_smm_save_state_64(vcpu, buf);
9458 enter_smm_save_state_32(vcpu, buf);
9461 * Give enter_smm() a chance to make ISA-specific changes to the vCPU
9462 * state (e.g. leave guest mode) after we've saved the state into the
9463 * SMM state-save area.
9465 static_call(kvm_x86_enter_smm)(vcpu, buf);
9467 kvm_smm_changed(vcpu, true);
9468 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
9470 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
9471 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
9473 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
9475 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
9476 kvm_rip_write(vcpu, 0x8000);
9478 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
9479 static_call(kvm_x86_set_cr0)(vcpu, cr0);
9480 vcpu->arch.cr0 = cr0;
9482 static_call(kvm_x86_set_cr4)(vcpu, 0);
9484 /* Undocumented: IDT limit is set to zero on entry to SMM. */
9485 dt.address = dt.size = 0;
9486 static_call(kvm_x86_set_idt)(vcpu, &dt);
9488 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
9490 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
9491 cs.base = vcpu->arch.smbase;
9496 cs.limit = ds.limit = 0xffffffff;
9497 cs.type = ds.type = 0x3;
9498 cs.dpl = ds.dpl = 0;
9503 cs.avl = ds.avl = 0;
9504 cs.present = ds.present = 1;
9505 cs.unusable = ds.unusable = 0;
9506 cs.padding = ds.padding = 0;
9508 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9509 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
9510 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
9511 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
9512 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
9513 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
9515 #ifdef CONFIG_X86_64
9516 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
9517 static_call(kvm_x86_set_efer)(vcpu, 0);
9520 kvm_update_cpuid_runtime(vcpu);
9521 kvm_mmu_reset_context(vcpu);
9524 static void process_smi(struct kvm_vcpu *vcpu)
9526 vcpu->arch.smi_pending = true;
9527 kvm_make_request(KVM_REQ_EVENT, vcpu);
9530 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
9531 unsigned long *vcpu_bitmap)
9533 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
9536 void kvm_make_scan_ioapic_request(struct kvm *kvm)
9538 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
9541 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
9545 if (!lapic_in_kernel(vcpu))
9548 down_read(&vcpu->kvm->arch.apicv_update_lock);
9550 activate = kvm_apicv_activated(vcpu->kvm);
9551 if (vcpu->arch.apicv_active == activate)
9554 vcpu->arch.apicv_active = activate;
9555 kvm_apic_update_apicv(vcpu);
9556 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
9559 * When APICv gets disabled, we may still have injected interrupts
9560 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
9561 * still active when the interrupt got accepted. Make sure
9562 * inject_pending_event() is called to check for that.
9564 if (!vcpu->arch.apicv_active)
9565 kvm_make_request(KVM_REQ_EVENT, vcpu);
9568 up_read(&vcpu->kvm->arch.apicv_update_lock);
9570 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
9572 void __kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9574 unsigned long old, new;
9576 lockdep_assert_held_write(&kvm->arch.apicv_update_lock);
9578 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
9579 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
9582 old = new = kvm->arch.apicv_inhibit_reasons;
9585 __clear_bit(bit, &new);
9587 __set_bit(bit, &new);
9589 if (!!old != !!new) {
9590 trace_kvm_apicv_update_request(activate, bit);
9592 * Kick all vCPUs before setting apicv_inhibit_reasons to avoid
9593 * false positives in the sanity check WARN in svm_vcpu_run().
9594 * This task will wait for all vCPUs to ack the kick IRQ before
9595 * updating apicv_inhibit_reasons, and all other vCPUs will
9596 * block on acquiring apicv_update_lock so that vCPUs can't
9597 * redo svm_vcpu_run() without seeing the new inhibit state.
9599 * Note, holding apicv_update_lock and taking it in the read
9600 * side (handling the request) also prevents other vCPUs from
9601 * servicing the request with a stale apicv_inhibit_reasons.
9603 kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
9604 kvm->arch.apicv_inhibit_reasons = new;
9606 unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
9607 kvm_zap_gfn_range(kvm, gfn, gfn+1);
9610 kvm->arch.apicv_inhibit_reasons = new;
9612 EXPORT_SYMBOL_GPL(__kvm_request_apicv_update);
9614 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
9616 down_write(&kvm->arch.apicv_update_lock);
9617 __kvm_request_apicv_update(kvm, activate, bit);
9618 up_write(&kvm->arch.apicv_update_lock);
9620 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
9622 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
9624 if (!kvm_apic_present(vcpu))
9627 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
9629 if (irqchip_split(vcpu->kvm))
9630 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
9632 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9633 if (ioapic_in_kernel(vcpu->kvm))
9634 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
9637 if (is_guest_mode(vcpu))
9638 vcpu->arch.load_eoi_exitmap_pending = true;
9640 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
9643 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
9645 u64 eoi_exit_bitmap[4];
9647 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
9650 if (to_hv_vcpu(vcpu)) {
9651 bitmap_or((ulong *)eoi_exit_bitmap,
9652 vcpu->arch.ioapic_handled_vectors,
9653 to_hv_synic(vcpu)->vec_bitmap, 256);
9654 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
9658 static_call(kvm_x86_load_eoi_exitmap)(
9659 vcpu, (u64 *)vcpu->arch.ioapic_handled_vectors);
9662 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
9663 unsigned long start, unsigned long end)
9665 unsigned long apic_address;
9668 * The physical address of apic access page is stored in the VMCS.
9669 * Update it when it becomes invalid.
9671 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9672 if (start <= apic_address && apic_address < end)
9673 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9676 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9678 if (!lapic_in_kernel(vcpu))
9681 if (!kvm_x86_ops.set_apic_access_page_addr)
9684 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9687 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9689 smp_send_reschedule(vcpu->cpu);
9691 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9694 * Returns 1 to let vcpu_run() continue the guest execution loop without
9695 * exiting to the userspace. Otherwise, the value will be returned to the
9698 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9702 dm_request_for_irq_injection(vcpu) &&
9703 kvm_cpu_accept_dm_intr(vcpu);
9704 fastpath_t exit_fastpath;
9706 bool req_immediate_exit = false;
9708 /* Forbid vmenter if vcpu dirty ring is soft-full */
9709 if (unlikely(vcpu->kvm->dirty_ring_size &&
9710 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9711 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9712 trace_kvm_dirty_ring_exit(vcpu);
9717 if (kvm_request_pending(vcpu)) {
9718 if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu)) {
9722 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9723 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9728 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9729 kvm_mmu_unload(vcpu);
9730 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9731 __kvm_migrate_timers(vcpu);
9732 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9733 kvm_update_masterclock(vcpu->kvm);
9734 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9735 kvm_gen_kvmclock_update(vcpu);
9736 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9737 r = kvm_guest_time_update(vcpu);
9741 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9742 kvm_mmu_sync_roots(vcpu);
9743 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9744 kvm_mmu_load_pgd(vcpu);
9745 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9746 kvm_vcpu_flush_tlb_all(vcpu);
9748 /* Flushing all ASIDs flushes the current ASID... */
9749 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9751 kvm_service_local_tlb_flush_requests(vcpu);
9753 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9754 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9758 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9759 if (is_guest_mode(vcpu)) {
9760 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9762 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9763 vcpu->mmio_needed = 0;
9768 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9769 /* Page is swapped out. Do synthetic halt */
9770 vcpu->arch.apf.halted = true;
9774 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9775 record_steal_time(vcpu);
9776 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9778 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9780 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9781 kvm_pmu_handle_event(vcpu);
9782 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9783 kvm_pmu_deliver_pmi(vcpu);
9784 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9785 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9786 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9787 vcpu->arch.ioapic_handled_vectors)) {
9788 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9789 vcpu->run->eoi.vector =
9790 vcpu->arch.pending_ioapic_eoi;
9795 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9796 vcpu_scan_ioapic(vcpu);
9797 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9798 vcpu_load_eoi_exitmap(vcpu);
9799 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9800 kvm_vcpu_reload_apic_access_page(vcpu);
9801 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9802 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9803 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9807 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9808 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9809 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9813 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9814 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9816 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9817 vcpu->run->hyperv = hv_vcpu->exit;
9823 * KVM_REQ_HV_STIMER has to be processed after
9824 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9825 * depend on the guest clock being up-to-date
9827 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9828 kvm_hv_process_stimers(vcpu);
9829 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9830 kvm_vcpu_update_apicv(vcpu);
9831 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9832 kvm_check_async_pf_completion(vcpu);
9833 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9834 static_call(kvm_x86_msr_filter_changed)(vcpu);
9836 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9837 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9840 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9841 kvm_xen_has_interrupt(vcpu)) {
9842 ++vcpu->stat.req_event;
9843 r = kvm_apic_accept_events(vcpu);
9848 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9853 r = inject_pending_event(vcpu, &req_immediate_exit);
9859 static_call(kvm_x86_enable_irq_window)(vcpu);
9861 if (kvm_lapic_enabled(vcpu)) {
9862 update_cr8_intercept(vcpu);
9863 kvm_lapic_sync_to_vapic(vcpu);
9867 r = kvm_mmu_reload(vcpu);
9869 goto cancel_injection;
9874 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9877 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9878 * IPI are then delayed after guest entry, which ensures that they
9879 * result in virtual interrupt delivery.
9881 local_irq_disable();
9882 vcpu->mode = IN_GUEST_MODE;
9884 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9887 * 1) We should set ->mode before checking ->requests. Please see
9888 * the comment in kvm_vcpu_exiting_guest_mode().
9890 * 2) For APICv, we should set ->mode before checking PID.ON. This
9891 * pairs with the memory barrier implicit in pi_test_and_set_on
9892 * (see vmx_deliver_posted_interrupt).
9894 * 3) This also orders the write to mode from any reads to the page
9895 * tables done while the VCPU is running. Please see the comment
9896 * in kvm_flush_remote_tlbs.
9898 smp_mb__after_srcu_read_unlock();
9901 * This handles the case where a posted interrupt was
9902 * notified with kvm_vcpu_kick. Assigned devices can
9903 * use the POSTED_INTR_VECTOR even if APICv is disabled,
9904 * so do it even if APICv is disabled on this vCPU.
9906 if (kvm_lapic_enabled(vcpu))
9907 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9909 if (kvm_vcpu_exit_request(vcpu)) {
9910 vcpu->mode = OUTSIDE_GUEST_MODE;
9914 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9916 goto cancel_injection;
9919 if (req_immediate_exit) {
9920 kvm_make_request(KVM_REQ_EVENT, vcpu);
9921 static_call(kvm_x86_request_immediate_exit)(vcpu);
9924 fpregs_assert_state_consistent();
9925 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9926 switch_fpu_return();
9928 if (unlikely(vcpu->arch.switch_db_regs)) {
9930 set_debugreg(vcpu->arch.eff_db[0], 0);
9931 set_debugreg(vcpu->arch.eff_db[1], 1);
9932 set_debugreg(vcpu->arch.eff_db[2], 2);
9933 set_debugreg(vcpu->arch.eff_db[3], 3);
9934 } else if (unlikely(hw_breakpoint_active())) {
9940 * Assert that vCPU vs. VM APICv state is consistent. An APICv
9941 * update must kick and wait for all vCPUs before toggling the
9942 * per-VM state, and responsing vCPUs must wait for the update
9943 * to complete before servicing KVM_REQ_APICV_UPDATE.
9945 WARN_ON_ONCE(kvm_apicv_activated(vcpu->kvm) != kvm_vcpu_apicv_active(vcpu));
9947 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9948 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9951 if (kvm_lapic_enabled(vcpu))
9952 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
9954 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9955 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9961 * Do this here before restoring debug registers on the host. And
9962 * since we do this before handling the vmexit, a DR access vmexit
9963 * can (a) read the correct value of the debug registers, (b) set
9964 * KVM_DEBUGREG_WONT_EXIT again.
9966 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9967 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9968 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9969 kvm_update_dr0123(vcpu);
9970 kvm_update_dr7(vcpu);
9974 * If the guest has used debug registers, at least dr7
9975 * will be disabled while returning to the host.
9976 * If we don't have active breakpoints in the host, we don't
9977 * care about the messed up debug address registers. But if
9978 * we have some of them active, restore the old state.
9980 if (hw_breakpoint_active())
9981 hw_breakpoint_restore();
9983 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9984 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9986 vcpu->mode = OUTSIDE_GUEST_MODE;
9989 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9992 * Consume any pending interrupts, including the possible source of
9993 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9994 * An instruction is required after local_irq_enable() to fully unblock
9995 * interrupts on processors that implement an interrupt shadow, the
9996 * stat.exits increment will do nicely.
9998 kvm_before_interrupt(vcpu);
10000 ++vcpu->stat.exits;
10001 local_irq_disable();
10002 kvm_after_interrupt(vcpu);
10005 * Wait until after servicing IRQs to account guest time so that any
10006 * ticks that occurred while running the guest are properly accounted
10007 * to the guest. Waiting until IRQs are enabled degrades the accuracy
10008 * of accounting via context tracking, but the loss of accuracy is
10009 * acceptable for all known use cases.
10011 vtime_account_guest_exit();
10013 if (lapic_in_kernel(vcpu)) {
10014 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
10015 if (delta != S64_MIN) {
10016 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
10017 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
10021 local_irq_enable();
10024 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
10027 * Profile KVM exit RIPs:
10029 if (unlikely(prof_on == KVM_PROFILING)) {
10030 unsigned long rip = kvm_rip_read(vcpu);
10031 profile_hit(KVM_PROFILING, (void *)rip);
10034 if (unlikely(vcpu->arch.tsc_always_catchup))
10035 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10037 if (vcpu->arch.apic_attention)
10038 kvm_lapic_sync_from_vapic(vcpu);
10040 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
10044 if (req_immediate_exit)
10045 kvm_make_request(KVM_REQ_EVENT, vcpu);
10046 static_call(kvm_x86_cancel_injection)(vcpu);
10047 if (unlikely(vcpu->arch.apic_attention))
10048 kvm_lapic_sync_from_vapic(vcpu);
10053 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
10055 if (!kvm_arch_vcpu_runnable(vcpu) &&
10056 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
10057 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10058 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10059 kvm_vcpu_halt(vcpu);
10061 kvm_vcpu_block(vcpu);
10062 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10064 if (kvm_x86_ops.post_block)
10065 static_call(kvm_x86_post_block)(vcpu);
10067 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
10071 if (kvm_apic_accept_events(vcpu) < 0)
10073 switch(vcpu->arch.mp_state) {
10074 case KVM_MP_STATE_HALTED:
10075 case KVM_MP_STATE_AP_RESET_HOLD:
10076 vcpu->arch.pv.pv_unhalted = false;
10077 vcpu->arch.mp_state =
10078 KVM_MP_STATE_RUNNABLE;
10080 case KVM_MP_STATE_RUNNABLE:
10081 vcpu->arch.apf.halted = false;
10083 case KVM_MP_STATE_INIT_RECEIVED:
10091 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
10093 if (is_guest_mode(vcpu))
10094 kvm_check_nested_events(vcpu);
10096 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
10097 !vcpu->arch.apf.halted);
10100 static int vcpu_run(struct kvm_vcpu *vcpu)
10103 struct kvm *kvm = vcpu->kvm;
10105 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10106 vcpu->arch.l1tf_flush_l1d = true;
10109 if (kvm_vcpu_running(vcpu)) {
10110 r = vcpu_enter_guest(vcpu);
10112 r = vcpu_block(kvm, vcpu);
10118 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
10119 if (kvm_cpu_has_pending_timer(vcpu))
10120 kvm_inject_pending_timer_irqs(vcpu);
10122 if (dm_request_for_irq_injection(vcpu) &&
10123 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
10125 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
10126 ++vcpu->stat.request_irq_exits;
10130 if (__xfer_to_guest_mode_work_pending()) {
10131 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10132 r = xfer_to_guest_mode_handle_work(vcpu);
10135 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
10139 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
10144 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
10148 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
10149 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
10150 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
10154 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
10156 BUG_ON(!vcpu->arch.pio.count);
10158 return complete_emulated_io(vcpu);
10162 * Implements the following, as a state machine:
10165 * for each fragment
10166 * for each mmio piece in the fragment
10173 * for each fragment
10174 * for each mmio piece in the fragment
10179 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
10181 struct kvm_run *run = vcpu->run;
10182 struct kvm_mmio_fragment *frag;
10185 BUG_ON(!vcpu->mmio_needed);
10187 /* Complete previous fragment */
10188 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
10189 len = min(8u, frag->len);
10190 if (!vcpu->mmio_is_write)
10191 memcpy(frag->data, run->mmio.data, len);
10193 if (frag->len <= 8) {
10194 /* Switch to the next fragment. */
10196 vcpu->mmio_cur_fragment++;
10198 /* Go forward to the next mmio piece. */
10204 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
10205 vcpu->mmio_needed = 0;
10207 /* FIXME: return into emulator if single-stepping. */
10208 if (vcpu->mmio_is_write)
10210 vcpu->mmio_read_completed = 1;
10211 return complete_emulated_io(vcpu);
10214 run->exit_reason = KVM_EXIT_MMIO;
10215 run->mmio.phys_addr = frag->gpa;
10216 if (vcpu->mmio_is_write)
10217 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
10218 run->mmio.len = min(8u, frag->len);
10219 run->mmio.is_write = vcpu->mmio_is_write;
10220 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
10224 /* Swap (qemu) user FPU context for the guest FPU context. */
10225 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
10228 * Exclude PKRU from restore as restored separately in
10229 * kvm_x86_ops.run().
10231 fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, true);
10235 /* When vcpu_run ends, restore user space FPU context. */
10236 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
10238 fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, false);
10239 ++vcpu->stat.fpu_reload;
10243 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
10245 struct kvm_run *kvm_run = vcpu->run;
10249 kvm_sigset_activate(vcpu);
10250 kvm_run->flags = 0;
10251 kvm_load_guest_fpu(vcpu);
10253 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
10254 if (kvm_run->immediate_exit) {
10258 kvm_vcpu_block(vcpu);
10259 if (kvm_apic_accept_events(vcpu) < 0) {
10263 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
10265 if (signal_pending(current)) {
10267 kvm_run->exit_reason = KVM_EXIT_INTR;
10268 ++vcpu->stat.signal_exits;
10273 if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10274 (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10279 if (kvm_run->kvm_dirty_regs) {
10280 r = sync_regs(vcpu);
10285 /* re-sync apic's tpr */
10286 if (!lapic_in_kernel(vcpu)) {
10287 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
10293 if (unlikely(vcpu->arch.complete_userspace_io)) {
10294 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
10295 vcpu->arch.complete_userspace_io = NULL;
10300 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
10302 if (kvm_run->immediate_exit)
10305 r = vcpu_run(vcpu);
10308 kvm_put_guest_fpu(vcpu);
10309 if (kvm_run->kvm_valid_regs)
10311 post_kvm_run_save(vcpu);
10312 kvm_sigset_deactivate(vcpu);
10318 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10320 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
10322 * We are here if userspace calls get_regs() in the middle of
10323 * instruction emulation. Registers state needs to be copied
10324 * back from emulation context to vcpu. Userspace shouldn't do
10325 * that usually, but some bad designed PV devices (vmware
10326 * backdoor interface) need this to work
10328 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
10329 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10331 regs->rax = kvm_rax_read(vcpu);
10332 regs->rbx = kvm_rbx_read(vcpu);
10333 regs->rcx = kvm_rcx_read(vcpu);
10334 regs->rdx = kvm_rdx_read(vcpu);
10335 regs->rsi = kvm_rsi_read(vcpu);
10336 regs->rdi = kvm_rdi_read(vcpu);
10337 regs->rsp = kvm_rsp_read(vcpu);
10338 regs->rbp = kvm_rbp_read(vcpu);
10339 #ifdef CONFIG_X86_64
10340 regs->r8 = kvm_r8_read(vcpu);
10341 regs->r9 = kvm_r9_read(vcpu);
10342 regs->r10 = kvm_r10_read(vcpu);
10343 regs->r11 = kvm_r11_read(vcpu);
10344 regs->r12 = kvm_r12_read(vcpu);
10345 regs->r13 = kvm_r13_read(vcpu);
10346 regs->r14 = kvm_r14_read(vcpu);
10347 regs->r15 = kvm_r15_read(vcpu);
10350 regs->rip = kvm_rip_read(vcpu);
10351 regs->rflags = kvm_get_rflags(vcpu);
10354 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10357 __get_regs(vcpu, regs);
10362 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10364 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
10365 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
10367 kvm_rax_write(vcpu, regs->rax);
10368 kvm_rbx_write(vcpu, regs->rbx);
10369 kvm_rcx_write(vcpu, regs->rcx);
10370 kvm_rdx_write(vcpu, regs->rdx);
10371 kvm_rsi_write(vcpu, regs->rsi);
10372 kvm_rdi_write(vcpu, regs->rdi);
10373 kvm_rsp_write(vcpu, regs->rsp);
10374 kvm_rbp_write(vcpu, regs->rbp);
10375 #ifdef CONFIG_X86_64
10376 kvm_r8_write(vcpu, regs->r8);
10377 kvm_r9_write(vcpu, regs->r9);
10378 kvm_r10_write(vcpu, regs->r10);
10379 kvm_r11_write(vcpu, regs->r11);
10380 kvm_r12_write(vcpu, regs->r12);
10381 kvm_r13_write(vcpu, regs->r13);
10382 kvm_r14_write(vcpu, regs->r14);
10383 kvm_r15_write(vcpu, regs->r15);
10386 kvm_rip_write(vcpu, regs->rip);
10387 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
10389 vcpu->arch.exception.pending = false;
10391 kvm_make_request(KVM_REQ_EVENT, vcpu);
10394 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
10397 __set_regs(vcpu, regs);
10402 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
10404 struct kvm_segment cs;
10406 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10410 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
10412 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10414 struct desc_ptr dt;
10416 if (vcpu->arch.guest_state_protected)
10417 goto skip_protected_regs;
10419 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10420 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10421 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10422 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10423 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10424 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10426 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10427 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10429 static_call(kvm_x86_get_idt)(vcpu, &dt);
10430 sregs->idt.limit = dt.size;
10431 sregs->idt.base = dt.address;
10432 static_call(kvm_x86_get_gdt)(vcpu, &dt);
10433 sregs->gdt.limit = dt.size;
10434 sregs->gdt.base = dt.address;
10436 sregs->cr2 = vcpu->arch.cr2;
10437 sregs->cr3 = kvm_read_cr3(vcpu);
10439 skip_protected_regs:
10440 sregs->cr0 = kvm_read_cr0(vcpu);
10441 sregs->cr4 = kvm_read_cr4(vcpu);
10442 sregs->cr8 = kvm_get_cr8(vcpu);
10443 sregs->efer = vcpu->arch.efer;
10444 sregs->apic_base = kvm_get_apic_base(vcpu);
10447 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10449 __get_sregs_common(vcpu, sregs);
10451 if (vcpu->arch.guest_state_protected)
10454 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
10455 set_bit(vcpu->arch.interrupt.nr,
10456 (unsigned long *)sregs->interrupt_bitmap);
10459 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10463 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
10465 if (vcpu->arch.guest_state_protected)
10468 if (is_pae_paging(vcpu)) {
10469 for (i = 0 ; i < 4 ; i++)
10470 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
10471 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
10475 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
10476 struct kvm_sregs *sregs)
10479 __get_sregs(vcpu, sregs);
10484 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
10485 struct kvm_mp_state *mp_state)
10490 if (kvm_mpx_supported())
10491 kvm_load_guest_fpu(vcpu);
10493 r = kvm_apic_accept_events(vcpu);
10498 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
10499 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
10500 vcpu->arch.pv.pv_unhalted)
10501 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
10503 mp_state->mp_state = vcpu->arch.mp_state;
10506 if (kvm_mpx_supported())
10507 kvm_put_guest_fpu(vcpu);
10512 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
10513 struct kvm_mp_state *mp_state)
10519 if (!lapic_in_kernel(vcpu) &&
10520 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
10524 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
10525 * INIT state; latched init should be reported using
10526 * KVM_SET_VCPU_EVENTS, so reject it here.
10528 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
10529 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
10530 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
10533 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
10534 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
10535 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
10537 vcpu->arch.mp_state = mp_state->mp_state;
10538 kvm_make_request(KVM_REQ_EVENT, vcpu);
10546 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
10547 int reason, bool has_error_code, u32 error_code)
10549 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
10552 init_emulate_ctxt(vcpu);
10554 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
10555 has_error_code, error_code);
10557 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
10558 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
10559 vcpu->run->internal.ndata = 0;
10563 kvm_rip_write(vcpu, ctxt->eip);
10564 kvm_set_rflags(vcpu, ctxt->eflags);
10567 EXPORT_SYMBOL_GPL(kvm_task_switch);
10569 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10571 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
10573 * When EFER.LME and CR0.PG are set, the processor is in
10574 * 64-bit mode (though maybe in a 32-bit code segment).
10575 * CR4.PAE and EFER.LMA must be set.
10577 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
10579 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
10583 * Not in 64-bit mode: EFER.LMA is clear and the code
10584 * segment cannot be 64-bit.
10586 if (sregs->efer & EFER_LMA || sregs->cs.l)
10590 return kvm_is_valid_cr4(vcpu, sregs->cr4);
10593 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
10594 int *mmu_reset_needed, bool update_pdptrs)
10596 struct msr_data apic_base_msr;
10598 struct desc_ptr dt;
10600 if (!kvm_is_valid_sregs(vcpu, sregs))
10603 apic_base_msr.data = sregs->apic_base;
10604 apic_base_msr.host_initiated = true;
10605 if (kvm_set_apic_base(vcpu, &apic_base_msr))
10608 if (vcpu->arch.guest_state_protected)
10611 dt.size = sregs->idt.limit;
10612 dt.address = sregs->idt.base;
10613 static_call(kvm_x86_set_idt)(vcpu, &dt);
10614 dt.size = sregs->gdt.limit;
10615 dt.address = sregs->gdt.base;
10616 static_call(kvm_x86_set_gdt)(vcpu, &dt);
10618 vcpu->arch.cr2 = sregs->cr2;
10619 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
10620 vcpu->arch.cr3 = sregs->cr3;
10621 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
10622 static_call_cond(kvm_x86_post_set_cr3)(vcpu, sregs->cr3);
10624 kvm_set_cr8(vcpu, sregs->cr8);
10626 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
10627 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
10629 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
10630 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
10631 vcpu->arch.cr0 = sregs->cr0;
10633 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
10634 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
10636 if (update_pdptrs) {
10637 idx = srcu_read_lock(&vcpu->kvm->srcu);
10638 if (is_pae_paging(vcpu)) {
10639 load_pdptrs(vcpu, kvm_read_cr3(vcpu));
10640 *mmu_reset_needed = 1;
10642 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10645 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
10646 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
10647 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
10648 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
10649 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
10650 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
10652 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
10653 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
10655 update_cr8_intercept(vcpu);
10657 /* Older userspace won't unhalt the vcpu on reset. */
10658 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
10659 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
10660 !is_protmode(vcpu))
10661 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10666 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
10668 int pending_vec, max_bits;
10669 int mmu_reset_needed = 0;
10670 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
10675 if (mmu_reset_needed)
10676 kvm_mmu_reset_context(vcpu);
10678 max_bits = KVM_NR_INTERRUPTS;
10679 pending_vec = find_first_bit(
10680 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
10682 if (pending_vec < max_bits) {
10683 kvm_queue_interrupt(vcpu, pending_vec, false);
10684 pr_debug("Set back pending irq %d\n", pending_vec);
10685 kvm_make_request(KVM_REQ_EVENT, vcpu);
10690 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
10692 int mmu_reset_needed = 0;
10693 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
10694 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
10695 !(sregs2->efer & EFER_LMA);
10698 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
10701 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
10704 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
10705 &mmu_reset_needed, !valid_pdptrs);
10709 if (valid_pdptrs) {
10710 for (i = 0; i < 4 ; i++)
10711 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
10713 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
10714 mmu_reset_needed = 1;
10715 vcpu->arch.pdptrs_from_userspace = true;
10717 if (mmu_reset_needed)
10718 kvm_mmu_reset_context(vcpu);
10722 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
10723 struct kvm_sregs *sregs)
10728 ret = __set_sregs(vcpu, sregs);
10733 static void kvm_arch_vcpu_guestdbg_update_apicv_inhibit(struct kvm *kvm)
10735 bool inhibit = false;
10736 struct kvm_vcpu *vcpu;
10739 down_write(&kvm->arch.apicv_update_lock);
10741 kvm_for_each_vcpu(i, vcpu, kvm) {
10742 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ) {
10747 __kvm_request_apicv_update(kvm, !inhibit, APICV_INHIBIT_REASON_BLOCKIRQ);
10748 up_write(&kvm->arch.apicv_update_lock);
10751 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
10752 struct kvm_guest_debug *dbg)
10754 unsigned long rflags;
10757 if (vcpu->arch.guest_state_protected)
10762 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10764 if (vcpu->arch.exception.pending)
10766 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10767 kvm_queue_exception(vcpu, DB_VECTOR);
10769 kvm_queue_exception(vcpu, BP_VECTOR);
10773 * Read rflags as long as potentially injected trace flags are still
10776 rflags = kvm_get_rflags(vcpu);
10778 vcpu->guest_debug = dbg->control;
10779 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10780 vcpu->guest_debug = 0;
10782 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10783 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10784 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10785 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10787 for (i = 0; i < KVM_NR_DB_REGS; i++)
10788 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10790 kvm_update_dr7(vcpu);
10792 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10793 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
10796 * Trigger an rflags update that will inject or remove the trace
10799 kvm_set_rflags(vcpu, rflags);
10801 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10803 kvm_arch_vcpu_guestdbg_update_apicv_inhibit(vcpu->kvm);
10813 * Translate a guest virtual address to a guest physical address.
10815 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10816 struct kvm_translation *tr)
10818 unsigned long vaddr = tr->linear_address;
10824 idx = srcu_read_lock(&vcpu->kvm->srcu);
10825 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10826 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10827 tr->physical_address = gpa;
10828 tr->valid = gpa != UNMAPPED_GVA;
10836 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10838 struct fxregs_state *fxsave;
10840 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10845 fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10846 memcpy(fpu->fpr, fxsave->st_space, 128);
10847 fpu->fcw = fxsave->cwd;
10848 fpu->fsw = fxsave->swd;
10849 fpu->ftwx = fxsave->twd;
10850 fpu->last_opcode = fxsave->fop;
10851 fpu->last_ip = fxsave->rip;
10852 fpu->last_dp = fxsave->rdp;
10853 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10859 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10861 struct fxregs_state *fxsave;
10863 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
10868 fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
10870 memcpy(fxsave->st_space, fpu->fpr, 128);
10871 fxsave->cwd = fpu->fcw;
10872 fxsave->swd = fpu->fsw;
10873 fxsave->twd = fpu->ftwx;
10874 fxsave->fop = fpu->last_opcode;
10875 fxsave->rip = fpu->last_ip;
10876 fxsave->rdp = fpu->last_dp;
10877 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10883 static void store_regs(struct kvm_vcpu *vcpu)
10885 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10887 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10888 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10890 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10891 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10893 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10894 kvm_vcpu_ioctl_x86_get_vcpu_events(
10895 vcpu, &vcpu->run->s.regs.events);
10898 static int sync_regs(struct kvm_vcpu *vcpu)
10900 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10901 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10902 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10904 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10905 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10907 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10909 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10910 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10911 vcpu, &vcpu->run->s.regs.events))
10913 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10919 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10921 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10922 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10923 "guest TSC will not be reliable\n");
10928 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10933 vcpu->arch.last_vmentry_cpu = -1;
10934 vcpu->arch.regs_avail = ~0;
10935 vcpu->arch.regs_dirty = ~0;
10937 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10938 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10940 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10942 r = kvm_mmu_create(vcpu);
10946 if (irqchip_in_kernel(vcpu->kvm)) {
10947 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10949 goto fail_mmu_destroy;
10950 if (kvm_apicv_activated(vcpu->kvm))
10951 vcpu->arch.apicv_active = true;
10953 static_branch_inc(&kvm_has_noapic_vcpu);
10957 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10959 goto fail_free_lapic;
10960 vcpu->arch.pio_data = page_address(page);
10962 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10963 GFP_KERNEL_ACCOUNT);
10964 if (!vcpu->arch.mce_banks)
10965 goto fail_free_pio_data;
10966 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10968 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10969 GFP_KERNEL_ACCOUNT))
10970 goto fail_free_mce_banks;
10972 if (!alloc_emulate_ctxt(vcpu))
10973 goto free_wbinvd_dirty_mask;
10975 if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
10976 pr_err("kvm: failed to allocate vcpu's fpu\n");
10977 goto free_emulate_ctxt;
10980 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10981 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10983 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10985 kvm_async_pf_hash_reset(vcpu);
10986 kvm_pmu_init(vcpu);
10988 vcpu->arch.pending_external_vector = -1;
10989 vcpu->arch.preempted_in_kernel = false;
10991 #if IS_ENABLED(CONFIG_HYPERV)
10992 vcpu->arch.hv_root_tdp = INVALID_PAGE;
10995 r = static_call(kvm_x86_vcpu_create)(vcpu);
10997 goto free_guest_fpu;
10999 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
11000 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
11001 kvm_vcpu_mtrr_init(vcpu);
11003 kvm_set_tsc_khz(vcpu, max_tsc_khz);
11004 kvm_vcpu_reset(vcpu, false);
11005 kvm_init_mmu(vcpu);
11010 fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11012 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11013 free_wbinvd_dirty_mask:
11014 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11015 fail_free_mce_banks:
11016 kfree(vcpu->arch.mce_banks);
11017 fail_free_pio_data:
11018 free_page((unsigned long)vcpu->arch.pio_data);
11020 kvm_free_lapic(vcpu);
11022 kvm_mmu_destroy(vcpu);
11026 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
11028 struct kvm *kvm = vcpu->kvm;
11030 if (mutex_lock_killable(&vcpu->mutex))
11033 kvm_synchronize_tsc(vcpu, 0);
11036 /* poll control enabled by default */
11037 vcpu->arch.msr_kvm_poll_control = 1;
11039 mutex_unlock(&vcpu->mutex);
11041 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
11042 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
11043 KVMCLOCK_SYNC_PERIOD);
11046 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
11050 kvmclock_reset(vcpu);
11052 static_call(kvm_x86_vcpu_free)(vcpu);
11054 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11055 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11056 fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11058 kvm_hv_vcpu_uninit(vcpu);
11059 kvm_pmu_destroy(vcpu);
11060 kfree(vcpu->arch.mce_banks);
11061 kvm_free_lapic(vcpu);
11062 idx = srcu_read_lock(&vcpu->kvm->srcu);
11063 kvm_mmu_destroy(vcpu);
11064 srcu_read_unlock(&vcpu->kvm->srcu, idx);
11065 free_page((unsigned long)vcpu->arch.pio_data);
11066 kvfree(vcpu->arch.cpuid_entries);
11067 if (!lapic_in_kernel(vcpu))
11068 static_branch_dec(&kvm_has_noapic_vcpu);
11071 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
11073 struct kvm_cpuid_entry2 *cpuid_0x1;
11074 unsigned long old_cr0 = kvm_read_cr0(vcpu);
11075 unsigned long new_cr0;
11078 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
11079 * to handle side effects. RESET emulation hits those flows and relies
11080 * on emulated/virtualized registers, including those that are loaded
11081 * into hardware, to be zeroed at vCPU creation. Use CRs as a sentinel
11082 * to detect improper or missing initialization.
11084 WARN_ON_ONCE(!init_event &&
11085 (old_cr0 || kvm_read_cr3(vcpu) || kvm_read_cr4(vcpu)));
11087 kvm_lapic_reset(vcpu, init_event);
11089 vcpu->arch.hflags = 0;
11091 vcpu->arch.smi_pending = 0;
11092 vcpu->arch.smi_count = 0;
11093 atomic_set(&vcpu->arch.nmi_queued, 0);
11094 vcpu->arch.nmi_pending = 0;
11095 vcpu->arch.nmi_injected = false;
11096 kvm_clear_interrupt_queue(vcpu);
11097 kvm_clear_exception_queue(vcpu);
11099 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
11100 kvm_update_dr0123(vcpu);
11101 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
11102 vcpu->arch.dr7 = DR7_FIXED_1;
11103 kvm_update_dr7(vcpu);
11105 vcpu->arch.cr2 = 0;
11107 kvm_make_request(KVM_REQ_EVENT, vcpu);
11108 vcpu->arch.apf.msr_en_val = 0;
11109 vcpu->arch.apf.msr_int_val = 0;
11110 vcpu->arch.st.msr_val = 0;
11112 kvmclock_reset(vcpu);
11114 kvm_clear_async_pf_completion_queue(vcpu);
11115 kvm_async_pf_hash_reset(vcpu);
11116 vcpu->arch.apf.halted = false;
11118 if (vcpu->arch.guest_fpu.fpstate && kvm_mpx_supported()) {
11119 struct fpstate *fpstate = vcpu->arch.guest_fpu.fpstate;
11122 * To avoid have the INIT path from kvm_apic_has_events() that be
11123 * called with loaded FPU and does not let userspace fix the state.
11126 kvm_put_guest_fpu(vcpu);
11128 fpstate_clear_xstate_component(fpstate, XFEATURE_BNDREGS);
11129 fpstate_clear_xstate_component(fpstate, XFEATURE_BNDCSR);
11132 kvm_load_guest_fpu(vcpu);
11136 kvm_pmu_reset(vcpu);
11137 vcpu->arch.smbase = 0x30000;
11139 vcpu->arch.msr_misc_features_enables = 0;
11141 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
11144 /* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
11145 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
11146 kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
11149 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
11150 * if no CPUID match is found. Note, it's impossible to get a match at
11151 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
11152 * i.e. it's impossible for kvm_find_cpuid_entry() to find a valid entry
11153 * on RESET. But, go through the motions in case that's ever remedied.
11155 cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1, 0);
11156 kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
11158 vcpu->arch.ia32_xss = 0;
11160 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
11162 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
11163 kvm_rip_write(vcpu, 0xfff0);
11165 vcpu->arch.cr3 = 0;
11166 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11169 * CR0.CD/NW are set on RESET, preserved on INIT. Note, some versions
11170 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
11171 * (or qualify) that with a footnote stating that CD/NW are preserved.
11173 new_cr0 = X86_CR0_ET;
11175 new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
11177 new_cr0 |= X86_CR0_NW | X86_CR0_CD;
11179 static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
11180 static_call(kvm_x86_set_cr4)(vcpu, 0);
11181 static_call(kvm_x86_set_efer)(vcpu, 0);
11182 static_call(kvm_x86_update_exception_bitmap)(vcpu);
11185 * Reset the MMU context if paging was enabled prior to INIT (which is
11186 * implied if CR0.PG=1 as CR0 will be '0' prior to RESET). Unlike the
11187 * standard CR0/CR4/EFER modification paths, only CR0.PG needs to be
11188 * checked because it is unconditionally cleared on INIT and all other
11189 * paging related bits are ignored if paging is disabled, i.e. CR0.WP,
11190 * CR4, and EFER changes are all irrelevant if CR0.PG was '0'.
11192 if (old_cr0 & X86_CR0_PG)
11193 kvm_mmu_reset_context(vcpu);
11196 * Intel's SDM states that all TLB entries are flushed on INIT. AMD's
11197 * APM states the TLBs are untouched by INIT, but it also states that
11198 * the TLBs are flushed on "External initialization of the processor."
11199 * Flush the guest TLB regardless of vendor, there is no meaningful
11200 * benefit in relying on the guest to flush the TLB immediately after
11201 * INIT. A spurious TLB flush is benign and likely negligible from a
11202 * performance perspective.
11205 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11207 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
11209 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
11211 struct kvm_segment cs;
11213 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
11214 cs.selector = vector << 8;
11215 cs.base = vector << 12;
11216 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
11217 kvm_rip_write(vcpu, 0);
11219 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
11221 int kvm_arch_hardware_enable(void)
11224 struct kvm_vcpu *vcpu;
11229 bool stable, backwards_tsc = false;
11231 kvm_user_return_msr_cpu_online();
11232 ret = static_call(kvm_x86_hardware_enable)();
11236 local_tsc = rdtsc();
11237 stable = !kvm_check_tsc_unstable();
11238 list_for_each_entry(kvm, &vm_list, vm_list) {
11239 kvm_for_each_vcpu(i, vcpu, kvm) {
11240 if (!stable && vcpu->cpu == smp_processor_id())
11241 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
11242 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
11243 backwards_tsc = true;
11244 if (vcpu->arch.last_host_tsc > max_tsc)
11245 max_tsc = vcpu->arch.last_host_tsc;
11251 * Sometimes, even reliable TSCs go backwards. This happens on
11252 * platforms that reset TSC during suspend or hibernate actions, but
11253 * maintain synchronization. We must compensate. Fortunately, we can
11254 * detect that condition here, which happens early in CPU bringup,
11255 * before any KVM threads can be running. Unfortunately, we can't
11256 * bring the TSCs fully up to date with real time, as we aren't yet far
11257 * enough into CPU bringup that we know how much real time has actually
11258 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
11259 * variables that haven't been updated yet.
11261 * So we simply find the maximum observed TSC above, then record the
11262 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
11263 * the adjustment will be applied. Note that we accumulate
11264 * adjustments, in case multiple suspend cycles happen before some VCPU
11265 * gets a chance to run again. In the event that no KVM threads get a
11266 * chance to run, we will miss the entire elapsed period, as we'll have
11267 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
11268 * loose cycle time. This isn't too big a deal, since the loss will be
11269 * uniform across all VCPUs (not to mention the scenario is extremely
11270 * unlikely). It is possible that a second hibernate recovery happens
11271 * much faster than a first, causing the observed TSC here to be
11272 * smaller; this would require additional padding adjustment, which is
11273 * why we set last_host_tsc to the local tsc observed here.
11275 * N.B. - this code below runs only on platforms with reliable TSC,
11276 * as that is the only way backwards_tsc is set above. Also note
11277 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
11278 * have the same delta_cyc adjustment applied if backwards_tsc
11279 * is detected. Note further, this adjustment is only done once,
11280 * as we reset last_host_tsc on all VCPUs to stop this from being
11281 * called multiple times (one for each physical CPU bringup).
11283 * Platforms with unreliable TSCs don't have to deal with this, they
11284 * will be compensated by the logic in vcpu_load, which sets the TSC to
11285 * catchup mode. This will catchup all VCPUs to real time, but cannot
11286 * guarantee that they stay in perfect synchronization.
11288 if (backwards_tsc) {
11289 u64 delta_cyc = max_tsc - local_tsc;
11290 list_for_each_entry(kvm, &vm_list, vm_list) {
11291 kvm->arch.backwards_tsc_observed = true;
11292 kvm_for_each_vcpu(i, vcpu, kvm) {
11293 vcpu->arch.tsc_offset_adjustment += delta_cyc;
11294 vcpu->arch.last_host_tsc = local_tsc;
11295 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
11299 * We have to disable TSC offset matching.. if you were
11300 * booting a VM while issuing an S4 host suspend....
11301 * you may have some problem. Solving this issue is
11302 * left as an exercise to the reader.
11304 kvm->arch.last_tsc_nsec = 0;
11305 kvm->arch.last_tsc_write = 0;
11312 void kvm_arch_hardware_disable(void)
11314 static_call(kvm_x86_hardware_disable)();
11315 drop_user_return_notifiers();
11318 int kvm_arch_hardware_setup(void *opaque)
11320 struct kvm_x86_init_ops *ops = opaque;
11323 rdmsrl_safe(MSR_EFER, &host_efer);
11325 if (boot_cpu_has(X86_FEATURE_XSAVES))
11326 rdmsrl(MSR_IA32_XSS, host_xss);
11328 r = ops->hardware_setup();
11332 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
11333 kvm_ops_static_call_update();
11335 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
11338 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
11339 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
11340 #undef __kvm_cpu_cap_has
11342 if (kvm_has_tsc_control) {
11344 * Make sure the user can only configure tsc_khz values that
11345 * fit into a signed integer.
11346 * A min value is not calculated because it will always
11347 * be 1 on all machines.
11349 u64 max = min(0x7fffffffULL,
11350 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
11351 kvm_max_guest_tsc_khz = max;
11353 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
11356 kvm_init_msr_list();
11360 void kvm_arch_hardware_unsetup(void)
11362 static_call(kvm_x86_hardware_unsetup)();
11365 int kvm_arch_check_processor_compat(void *opaque)
11367 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
11368 struct kvm_x86_init_ops *ops = opaque;
11370 WARN_ON(!irqs_disabled());
11372 if (__cr4_reserved_bits(cpu_has, c) !=
11373 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
11376 return ops->check_processor_compatibility();
11379 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
11381 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
11383 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
11385 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
11387 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
11390 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
11391 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
11393 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
11395 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
11397 vcpu->arch.l1tf_flush_l1d = true;
11398 if (pmu->version && unlikely(pmu->event_count)) {
11399 pmu->need_cleanup = true;
11400 kvm_make_request(KVM_REQ_PMU, vcpu);
11402 static_call(kvm_x86_sched_in)(vcpu, cpu);
11405 void kvm_arch_free_vm(struct kvm *kvm)
11407 kfree(to_kvm_hv(kvm)->hv_pa_pg);
11408 __kvm_arch_free_vm(kvm);
11412 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
11415 unsigned long flags;
11420 ret = kvm_page_track_init(kvm);
11424 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
11425 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
11426 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
11427 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
11428 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
11429 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
11431 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
11432 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
11433 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
11434 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
11435 &kvm->arch.irq_sources_bitmap);
11437 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
11438 mutex_init(&kvm->arch.apic_map_lock);
11439 seqcount_raw_spinlock_init(&kvm->arch.pvclock_sc, &kvm->arch.tsc_write_lock);
11440 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
11442 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
11443 pvclock_update_vm_gtod_copy(kvm);
11444 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
11446 kvm->arch.guest_can_read_msr_platform_info = true;
11448 #if IS_ENABLED(CONFIG_HYPERV)
11449 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
11450 kvm->arch.hv_root_tdp = INVALID_PAGE;
11453 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
11454 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
11456 kvm_apicv_init(kvm);
11457 kvm_hv_init_vm(kvm);
11458 kvm_mmu_init_vm(kvm);
11459 kvm_xen_init_vm(kvm);
11461 return static_call(kvm_x86_vm_init)(kvm);
11464 int kvm_arch_post_init_vm(struct kvm *kvm)
11466 return kvm_mmu_post_init_vm(kvm);
11469 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
11472 kvm_mmu_unload(vcpu);
11476 static void kvm_free_vcpus(struct kvm *kvm)
11479 struct kvm_vcpu *vcpu;
11482 * Unpin any mmu pages first.
11484 kvm_for_each_vcpu(i, vcpu, kvm) {
11485 kvm_clear_async_pf_completion_queue(vcpu);
11486 kvm_unload_vcpu_mmu(vcpu);
11489 kvm_destroy_vcpus(kvm);
11492 void kvm_arch_sync_events(struct kvm *kvm)
11494 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
11495 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
11499 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
11502 * __x86_set_memory_region: Setup KVM internal memory slot
11504 * @kvm: the kvm pointer to the VM.
11505 * @id: the slot ID to setup.
11506 * @gpa: the GPA to install the slot (unused when @size == 0).
11507 * @size: the size of the slot. Set to zero to uninstall a slot.
11509 * This function helps to setup a KVM internal memory slot. Specify
11510 * @size > 0 to install a new slot, while @size == 0 to uninstall a
11511 * slot. The return code can be one of the following:
11513 * HVA: on success (uninstall will return a bogus HVA)
11516 * The caller should always use IS_ERR() to check the return value
11517 * before use. Note, the KVM internal memory slots are guaranteed to
11518 * remain valid and unchanged until the VM is destroyed, i.e., the
11519 * GPA->HVA translation will not change. However, the HVA is a user
11520 * address, i.e. its accessibility is not guaranteed, and must be
11521 * accessed via __copy_{to,from}_user().
11523 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
11527 unsigned long hva, old_npages;
11528 struct kvm_memslots *slots = kvm_memslots(kvm);
11529 struct kvm_memory_slot *slot;
11531 /* Called with kvm->slots_lock held. */
11532 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
11533 return ERR_PTR_USR(-EINVAL);
11535 slot = id_to_memslot(slots, id);
11537 if (slot && slot->npages)
11538 return ERR_PTR_USR(-EEXIST);
11541 * MAP_SHARED to prevent internal slot pages from being moved
11544 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
11545 MAP_SHARED | MAP_ANONYMOUS, 0);
11546 if (IS_ERR((void *)hva))
11547 return (void __user *)hva;
11549 if (!slot || !slot->npages)
11552 old_npages = slot->npages;
11553 hva = slot->userspace_addr;
11556 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
11557 struct kvm_userspace_memory_region m;
11559 m.slot = id | (i << 16);
11561 m.guest_phys_addr = gpa;
11562 m.userspace_addr = hva;
11563 m.memory_size = size;
11564 r = __kvm_set_memory_region(kvm, &m);
11566 return ERR_PTR_USR(r);
11570 vm_munmap(hva, old_npages * PAGE_SIZE);
11572 return (void __user *)hva;
11574 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
11576 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
11578 kvm_mmu_pre_destroy_vm(kvm);
11581 void kvm_arch_destroy_vm(struct kvm *kvm)
11583 if (current->mm == kvm->mm) {
11585 * Free memory regions allocated on behalf of userspace,
11586 * unless the the memory map has changed due to process exit
11589 mutex_lock(&kvm->slots_lock);
11590 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
11592 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
11594 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
11595 mutex_unlock(&kvm->slots_lock);
11597 static_call_cond(kvm_x86_vm_destroy)(kvm);
11598 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
11599 kvm_pic_destroy(kvm);
11600 kvm_ioapic_destroy(kvm);
11601 kvm_free_vcpus(kvm);
11602 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
11603 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
11604 kvm_mmu_uninit_vm(kvm);
11605 kvm_page_track_cleanup(kvm);
11606 kvm_xen_destroy_vm(kvm);
11607 kvm_hv_destroy_vm(kvm);
11610 static void memslot_rmap_free(struct kvm_memory_slot *slot)
11614 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11615 kvfree(slot->arch.rmap[i]);
11616 slot->arch.rmap[i] = NULL;
11620 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
11624 memslot_rmap_free(slot);
11626 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11627 kvfree(slot->arch.lpage_info[i - 1]);
11628 slot->arch.lpage_info[i - 1] = NULL;
11631 kvm_page_track_free_memslot(slot);
11634 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages)
11636 const int sz = sizeof(*slot->arch.rmap[0]);
11639 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
11641 int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11643 if (slot->arch.rmap[i])
11646 slot->arch.rmap[i] = kvcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
11647 if (!slot->arch.rmap[i]) {
11648 memslot_rmap_free(slot);
11656 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
11657 struct kvm_memory_slot *slot)
11659 unsigned long npages = slot->npages;
11663 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
11664 * old arrays will be freed by __kvm_set_memory_region() if installing
11665 * the new memslot is successful.
11667 memset(&slot->arch, 0, sizeof(slot->arch));
11669 if (kvm_memslots_have_rmaps(kvm)) {
11670 r = memslot_rmap_alloc(slot, npages);
11675 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11676 struct kvm_lpage_info *linfo;
11677 unsigned long ugfn;
11681 lpages = __kvm_mmu_slot_lpages(slot, npages, level);
11683 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
11687 slot->arch.lpage_info[i - 1] = linfo;
11689 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
11690 linfo[0].disallow_lpage = 1;
11691 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
11692 linfo[lpages - 1].disallow_lpage = 1;
11693 ugfn = slot->userspace_addr >> PAGE_SHIFT;
11695 * If the gfn and userspace address are not aligned wrt each
11696 * other, disable large page support for this slot.
11698 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
11701 for (j = 0; j < lpages; ++j)
11702 linfo[j].disallow_lpage = 1;
11706 if (kvm_page_track_create_memslot(kvm, slot, npages))
11712 memslot_rmap_free(slot);
11714 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
11715 kvfree(slot->arch.lpage_info[i - 1]);
11716 slot->arch.lpage_info[i - 1] = NULL;
11721 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
11723 struct kvm_vcpu *vcpu;
11727 * memslots->generation has been incremented.
11728 * mmio generation may have reached its maximum value.
11730 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
11732 /* Force re-initialization of steal_time cache */
11733 kvm_for_each_vcpu(i, vcpu, kvm)
11734 kvm_vcpu_kick(vcpu);
11737 int kvm_arch_prepare_memory_region(struct kvm *kvm,
11738 const struct kvm_memory_slot *old,
11739 struct kvm_memory_slot *new,
11740 enum kvm_mr_change change)
11742 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
11743 return kvm_alloc_memslot_metadata(kvm, new);
11745 if (change == KVM_MR_FLAGS_ONLY)
11746 memcpy(&new->arch, &old->arch, sizeof(old->arch));
11747 else if (WARN_ON_ONCE(change != KVM_MR_DELETE))
11754 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
11756 struct kvm_arch *ka = &kvm->arch;
11758 if (!kvm_x86_ops.cpu_dirty_log_size)
11761 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
11762 (!enable && --ka->cpu_dirty_logging_count == 0))
11763 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
11765 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
11768 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
11769 struct kvm_memory_slot *old,
11770 const struct kvm_memory_slot *new,
11771 enum kvm_mr_change change)
11773 u32 old_flags = old ? old->flags : 0;
11774 u32 new_flags = new ? new->flags : 0;
11775 bool log_dirty_pages = new_flags & KVM_MEM_LOG_DIRTY_PAGES;
11778 * Update CPU dirty logging if dirty logging is being toggled. This
11779 * applies to all operations.
11781 if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)
11782 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
11785 * Nothing more to do for RO slots (which can't be dirtied and can't be
11786 * made writable) or CREATE/MOVE/DELETE of a slot.
11788 * For a memslot with dirty logging disabled:
11789 * CREATE: No dirty mappings will already exist.
11790 * MOVE/DELETE: The old mappings will already have been cleaned up by
11791 * kvm_arch_flush_shadow_memslot()
11793 * For a memslot with dirty logging enabled:
11794 * CREATE: No shadow pages exist, thus nothing to write-protect
11795 * and no dirty bits to clear.
11796 * MOVE/DELETE: The old mappings will already have been cleaned up by
11797 * kvm_arch_flush_shadow_memslot().
11799 if ((change != KVM_MR_FLAGS_ONLY) || (new_flags & KVM_MEM_READONLY))
11803 * READONLY and non-flags changes were filtered out above, and the only
11804 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
11805 * logging isn't being toggled on or off.
11807 if (WARN_ON_ONCE(!((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)))
11810 if (!log_dirty_pages) {
11812 * Dirty logging tracks sptes in 4k granularity, meaning that
11813 * large sptes have to be split. If live migration succeeds,
11814 * the guest in the source machine will be destroyed and large
11815 * sptes will be created in the destination. However, if the
11816 * guest continues to run in the source machine (for example if
11817 * live migration fails), small sptes will remain around and
11818 * cause bad performance.
11820 * Scan sptes if dirty logging has been stopped, dropping those
11821 * which can be collapsed into a single large-page spte. Later
11822 * page faults will create the large-page sptes.
11824 kvm_mmu_zap_collapsible_sptes(kvm, new);
11827 * Initially-all-set does not require write protecting any page,
11828 * because they're all assumed to be dirty.
11830 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
11833 if (kvm_x86_ops.cpu_dirty_log_size) {
11834 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11835 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
11837 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
11842 void kvm_arch_commit_memory_region(struct kvm *kvm,
11843 struct kvm_memory_slot *old,
11844 const struct kvm_memory_slot *new,
11845 enum kvm_mr_change change)
11847 if (!kvm->arch.n_requested_mmu_pages &&
11848 (change == KVM_MR_CREATE || change == KVM_MR_DELETE)) {
11849 unsigned long nr_mmu_pages;
11851 nr_mmu_pages = kvm->nr_memslot_pages / KVM_MEMSLOT_PAGES_TO_MMU_PAGES_RATIO;
11852 nr_mmu_pages = max(nr_mmu_pages, KVM_MIN_ALLOC_MMU_PAGES);
11853 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
11856 kvm_mmu_slot_apply_flags(kvm, old, new, change);
11858 /* Free the arrays associated with the old memslot. */
11859 if (change == KVM_MR_MOVE)
11860 kvm_arch_free_memslot(kvm, old);
11863 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11865 kvm_mmu_zap_all(kvm);
11868 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11869 struct kvm_memory_slot *slot)
11871 kvm_page_track_flush_slot(kvm, slot);
11874 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11876 return (is_guest_mode(vcpu) &&
11877 kvm_x86_ops.guest_apic_has_interrupt &&
11878 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11881 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11883 if (!list_empty_careful(&vcpu->async_pf.done))
11886 if (kvm_apic_has_events(vcpu))
11889 if (vcpu->arch.pv.pv_unhalted)
11892 if (vcpu->arch.exception.pending)
11895 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11896 (vcpu->arch.nmi_pending &&
11897 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11900 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11901 (vcpu->arch.smi_pending &&
11902 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11905 if (kvm_arch_interrupt_allowed(vcpu) &&
11906 (kvm_cpu_has_interrupt(vcpu) ||
11907 kvm_guest_apic_has_interrupt(vcpu)))
11910 if (kvm_hv_has_stimer_pending(vcpu))
11913 if (is_guest_mode(vcpu) &&
11914 kvm_x86_ops.nested_ops->hv_timer_pending &&
11915 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11921 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11923 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11926 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11928 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11934 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11936 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11939 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11940 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11941 kvm_test_request(KVM_REQ_EVENT, vcpu))
11944 return kvm_arch_dy_has_pending_interrupt(vcpu);
11947 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11949 if (vcpu->arch.guest_state_protected)
11952 return vcpu->arch.preempted_in_kernel;
11955 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11957 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11960 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11962 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11965 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11967 /* Can't read the RIP when guest state is protected, just return 0 */
11968 if (vcpu->arch.guest_state_protected)
11971 if (is_64_bit_mode(vcpu))
11972 return kvm_rip_read(vcpu);
11973 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11974 kvm_rip_read(vcpu));
11976 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11978 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11980 return kvm_get_linear_rip(vcpu) == linear_rip;
11982 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11984 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11986 unsigned long rflags;
11988 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11989 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11990 rflags &= ~X86_EFLAGS_TF;
11993 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11995 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11997 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11998 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11999 rflags |= X86_EFLAGS_TF;
12000 static_call(kvm_x86_set_rflags)(vcpu, rflags);
12003 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12005 __kvm_set_rflags(vcpu, rflags);
12006 kvm_make_request(KVM_REQ_EVENT, vcpu);
12008 EXPORT_SYMBOL_GPL(kvm_set_rflags);
12010 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
12014 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
12018 r = kvm_mmu_reload(vcpu);
12022 if (!vcpu->arch.mmu->direct_map &&
12023 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
12026 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
12029 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
12031 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
12033 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
12036 static inline u32 kvm_async_pf_next_probe(u32 key)
12038 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
12041 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12043 u32 key = kvm_async_pf_hash_fn(gfn);
12045 while (vcpu->arch.apf.gfns[key] != ~0)
12046 key = kvm_async_pf_next_probe(key);
12048 vcpu->arch.apf.gfns[key] = gfn;
12051 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
12054 u32 key = kvm_async_pf_hash_fn(gfn);
12056 for (i = 0; i < ASYNC_PF_PER_VCPU &&
12057 (vcpu->arch.apf.gfns[key] != gfn &&
12058 vcpu->arch.apf.gfns[key] != ~0); i++)
12059 key = kvm_async_pf_next_probe(key);
12064 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12066 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
12069 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12073 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
12075 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
12079 vcpu->arch.apf.gfns[i] = ~0;
12081 j = kvm_async_pf_next_probe(j);
12082 if (vcpu->arch.apf.gfns[j] == ~0)
12084 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
12086 * k lies cyclically in ]i,j]
12088 * |....j i.k.| or |.k..j i...|
12090 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
12091 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
12096 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
12098 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
12100 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
12104 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
12106 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12108 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12109 &token, offset, sizeof(token));
12112 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
12114 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12117 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12118 &val, offset, sizeof(val)))
12124 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
12126 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
12129 if (!kvm_pv_async_pf_enabled(vcpu) ||
12130 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
12136 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
12138 if (unlikely(!lapic_in_kernel(vcpu) ||
12139 kvm_event_needs_reinjection(vcpu) ||
12140 vcpu->arch.exception.pending))
12143 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
12147 * If interrupts are off we cannot even use an artificial
12150 return kvm_arch_interrupt_allowed(vcpu);
12153 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
12154 struct kvm_async_pf *work)
12156 struct x86_exception fault;
12158 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
12159 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
12161 if (kvm_can_deliver_async_pf(vcpu) &&
12162 !apf_put_user_notpresent(vcpu)) {
12163 fault.vector = PF_VECTOR;
12164 fault.error_code_valid = true;
12165 fault.error_code = 0;
12166 fault.nested_page_fault = false;
12167 fault.address = work->arch.token;
12168 fault.async_page_fault = true;
12169 kvm_inject_page_fault(vcpu, &fault);
12173 * It is not possible to deliver a paravirtualized asynchronous
12174 * page fault, but putting the guest in an artificial halt state
12175 * can be beneficial nevertheless: if an interrupt arrives, we
12176 * can deliver it timely and perhaps the guest will schedule
12177 * another process. When the instruction that triggered a page
12178 * fault is retried, hopefully the page will be ready in the host.
12180 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
12185 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
12186 struct kvm_async_pf *work)
12188 struct kvm_lapic_irq irq = {
12189 .delivery_mode = APIC_DM_FIXED,
12190 .vector = vcpu->arch.apf.vec
12193 if (work->wakeup_all)
12194 work->arch.token = ~0; /* broadcast wakeup */
12196 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
12197 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
12199 if ((work->wakeup_all || work->notpresent_injected) &&
12200 kvm_pv_async_pf_enabled(vcpu) &&
12201 !apf_put_user_ready(vcpu, work->arch.token)) {
12202 vcpu->arch.apf.pageready_pending = true;
12203 kvm_apic_set_irq(vcpu, &irq, NULL);
12206 vcpu->arch.apf.halted = false;
12207 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
12210 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
12212 kvm_make_request(KVM_REQ_APF_READY, vcpu);
12213 if (!vcpu->arch.apf.pageready_pending)
12214 kvm_vcpu_kick(vcpu);
12217 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
12219 if (!kvm_pv_async_pf_enabled(vcpu))
12222 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
12225 void kvm_arch_start_assignment(struct kvm *kvm)
12227 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
12228 static_call_cond(kvm_x86_start_assignment)(kvm);
12230 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
12232 void kvm_arch_end_assignment(struct kvm *kvm)
12234 atomic_dec(&kvm->arch.assigned_device_count);
12236 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
12238 bool kvm_arch_has_assigned_device(struct kvm *kvm)
12240 return atomic_read(&kvm->arch.assigned_device_count);
12242 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
12244 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
12246 atomic_inc(&kvm->arch.noncoherent_dma_count);
12248 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
12250 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
12252 atomic_dec(&kvm->arch.noncoherent_dma_count);
12254 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
12256 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
12258 return atomic_read(&kvm->arch.noncoherent_dma_count);
12260 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
12262 bool kvm_arch_has_irq_bypass(void)
12267 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
12268 struct irq_bypass_producer *prod)
12270 struct kvm_kernel_irqfd *irqfd =
12271 container_of(cons, struct kvm_kernel_irqfd, consumer);
12274 irqfd->producer = prod;
12275 kvm_arch_start_assignment(irqfd->kvm);
12276 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
12277 prod->irq, irqfd->gsi, 1);
12280 kvm_arch_end_assignment(irqfd->kvm);
12285 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
12286 struct irq_bypass_producer *prod)
12289 struct kvm_kernel_irqfd *irqfd =
12290 container_of(cons, struct kvm_kernel_irqfd, consumer);
12292 WARN_ON(irqfd->producer != prod);
12293 irqfd->producer = NULL;
12296 * When producer of consumer is unregistered, we change back to
12297 * remapped mode, so we can re-use the current implementation
12298 * when the irq is masked/disabled or the consumer side (KVM
12299 * int this case doesn't want to receive the interrupts.
12301 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
12303 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
12304 " fails: %d\n", irqfd->consumer.token, ret);
12306 kvm_arch_end_assignment(irqfd->kvm);
12309 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
12310 uint32_t guest_irq, bool set)
12312 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
12315 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *old,
12316 struct kvm_kernel_irq_routing_entry *new)
12318 if (new->type != KVM_IRQ_ROUTING_MSI)
12321 return !!memcmp(&old->msi, &new->msi, sizeof(new->msi));
12324 bool kvm_vector_hashing_enabled(void)
12326 return vector_hashing;
12329 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
12331 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
12333 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
12336 int kvm_spec_ctrl_test_value(u64 value)
12339 * test that setting IA32_SPEC_CTRL to given value
12340 * is allowed by the host processor
12344 unsigned long flags;
12347 local_irq_save(flags);
12349 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
12351 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
12354 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
12356 local_irq_restore(flags);
12360 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
12362 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
12364 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
12365 struct x86_exception fault;
12366 u32 access = error_code &
12367 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
12369 if (!(error_code & PFERR_PRESENT_MASK) ||
12370 mmu->gva_to_gpa(vcpu, mmu, gva, access, &fault) != UNMAPPED_GVA) {
12372 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
12373 * tables probably do not match the TLB. Just proceed
12374 * with the error code that the processor gave.
12376 fault.vector = PF_VECTOR;
12377 fault.error_code_valid = true;
12378 fault.error_code = error_code;
12379 fault.nested_page_fault = false;
12380 fault.address = gva;
12382 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
12384 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
12387 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
12388 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
12389 * indicates whether exit to userspace is needed.
12391 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
12392 struct x86_exception *e)
12394 if (r == X86EMUL_PROPAGATE_FAULT) {
12395 kvm_inject_emulated_page_fault(vcpu, e);
12400 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
12401 * while handling a VMX instruction KVM could've handled the request
12402 * correctly by exiting to userspace and performing I/O but there
12403 * doesn't seem to be a real use-case behind such requests, just return
12404 * KVM_EXIT_INTERNAL_ERROR for now.
12406 kvm_prepare_emulation_failure_exit(vcpu);
12410 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
12412 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
12415 struct x86_exception e;
12422 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
12423 if (r != X86EMUL_CONTINUE)
12424 return kvm_handle_memory_failure(vcpu, r, &e);
12426 if (operand.pcid >> 12 != 0) {
12427 kvm_inject_gp(vcpu, 0);
12431 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
12434 case INVPCID_TYPE_INDIV_ADDR:
12435 if ((!pcid_enabled && (operand.pcid != 0)) ||
12436 is_noncanonical_address(operand.gla, vcpu)) {
12437 kvm_inject_gp(vcpu, 0);
12440 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
12441 return kvm_skip_emulated_instruction(vcpu);
12443 case INVPCID_TYPE_SINGLE_CTXT:
12444 if (!pcid_enabled && (operand.pcid != 0)) {
12445 kvm_inject_gp(vcpu, 0);
12449 kvm_invalidate_pcid(vcpu, operand.pcid);
12450 return kvm_skip_emulated_instruction(vcpu);
12452 case INVPCID_TYPE_ALL_NON_GLOBAL:
12454 * Currently, KVM doesn't mark global entries in the shadow
12455 * page tables, so a non-global flush just degenerates to a
12456 * global flush. If needed, we could optimize this later by
12457 * keeping track of global entries in shadow page tables.
12461 case INVPCID_TYPE_ALL_INCL_GLOBAL:
12462 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
12463 return kvm_skip_emulated_instruction(vcpu);
12466 kvm_inject_gp(vcpu, 0);
12470 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
12472 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
12474 struct kvm_run *run = vcpu->run;
12475 struct kvm_mmio_fragment *frag;
12478 BUG_ON(!vcpu->mmio_needed);
12480 /* Complete previous fragment */
12481 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
12482 len = min(8u, frag->len);
12483 if (!vcpu->mmio_is_write)
12484 memcpy(frag->data, run->mmio.data, len);
12486 if (frag->len <= 8) {
12487 /* Switch to the next fragment. */
12489 vcpu->mmio_cur_fragment++;
12491 /* Go forward to the next mmio piece. */
12497 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
12498 vcpu->mmio_needed = 0;
12500 // VMG change, at this point, we're always done
12501 // RIP has already been advanced
12505 // More MMIO is needed
12506 run->mmio.phys_addr = frag->gpa;
12507 run->mmio.len = min(8u, frag->len);
12508 run->mmio.is_write = vcpu->mmio_is_write;
12509 if (run->mmio.is_write)
12510 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
12511 run->exit_reason = KVM_EXIT_MMIO;
12513 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12518 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12522 struct kvm_mmio_fragment *frag;
12527 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12528 if (handled == bytes)
12535 /*TODO: Check if need to increment number of frags */
12536 frag = vcpu->mmio_fragments;
12537 vcpu->mmio_nr_fragments = 1;
12542 vcpu->mmio_needed = 1;
12543 vcpu->mmio_cur_fragment = 0;
12545 vcpu->run->mmio.phys_addr = gpa;
12546 vcpu->run->mmio.len = min(8u, frag->len);
12547 vcpu->run->mmio.is_write = 1;
12548 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
12549 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12551 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12555 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
12557 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
12561 struct kvm_mmio_fragment *frag;
12566 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
12567 if (handled == bytes)
12574 /*TODO: Check if need to increment number of frags */
12575 frag = vcpu->mmio_fragments;
12576 vcpu->mmio_nr_fragments = 1;
12581 vcpu->mmio_needed = 1;
12582 vcpu->mmio_cur_fragment = 0;
12584 vcpu->run->mmio.phys_addr = gpa;
12585 vcpu->run->mmio.len = min(8u, frag->len);
12586 vcpu->run->mmio.is_write = 0;
12587 vcpu->run->exit_reason = KVM_EXIT_MMIO;
12589 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
12593 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
12595 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12596 unsigned int port);
12598 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
12600 int size = vcpu->arch.pio.size;
12601 int port = vcpu->arch.pio.port;
12603 vcpu->arch.pio.count = 0;
12604 if (vcpu->arch.sev_pio_count)
12605 return kvm_sev_es_outs(vcpu, size, port);
12609 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
12613 unsigned int count =
12614 min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12615 int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
12617 /* memcpy done already by emulator_pio_out. */
12618 vcpu->arch.sev_pio_count -= count;
12619 vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12623 /* Emulation done by the kernel. */
12624 if (!vcpu->arch.sev_pio_count)
12628 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
12632 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12633 unsigned int port);
12635 static void advance_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12637 unsigned count = vcpu->arch.pio.count;
12638 complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
12639 vcpu->arch.sev_pio_count -= count;
12640 vcpu->arch.sev_pio_data += count * vcpu->arch.pio.size;
12643 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
12645 int size = vcpu->arch.pio.size;
12646 int port = vcpu->arch.pio.port;
12648 advance_sev_es_emulated_ins(vcpu);
12649 if (vcpu->arch.sev_pio_count)
12650 return kvm_sev_es_ins(vcpu, size, port);
12654 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
12658 unsigned int count =
12659 min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
12660 if (!__emulator_pio_in(vcpu, size, port, count))
12663 /* Emulation done by the kernel. */
12664 advance_sev_es_emulated_ins(vcpu);
12665 if (!vcpu->arch.sev_pio_count)
12669 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
12673 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
12674 unsigned int port, void *data, unsigned int count,
12677 vcpu->arch.sev_pio_data = data;
12678 vcpu->arch.sev_pio_count = count;
12679 return in ? kvm_sev_es_ins(vcpu, size, port)
12680 : kvm_sev_es_outs(vcpu, size, port);
12682 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
12684 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
12685 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
12686 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
12687 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
12688 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
12689 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
12690 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
12691 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
12692 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
12693 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
12694 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
12695 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
12696 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
12697 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
12698 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
12699 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
12700 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
12701 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
12702 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
12703 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
12704 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
12705 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
12706 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
12707 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_accept_irq);
12708 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
12709 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
12710 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
12711 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);