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>
18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
20 #include <linux/kvm_host.h>
26 #include "kvm_cache_regs.h"
27 #include "kvm_emulate.h"
36 #include <linux/clocksource.h>
37 #include <linux/interrupt.h>
38 #include <linux/kvm.h>
40 #include <linux/vmalloc.h>
41 #include <linux/export.h>
42 #include <linux/moduleparam.h>
43 #include <linux/mman.h>
44 #include <linux/highmem.h>
45 #include <linux/iommu.h>
46 #include <linux/cpufreq.h>
47 #include <linux/user-return-notifier.h>
48 #include <linux/srcu.h>
49 #include <linux/slab.h>
50 #include <linux/perf_event.h>
51 #include <linux/uaccess.h>
52 #include <linux/hash.h>
53 #include <linux/pci.h>
54 #include <linux/timekeeper_internal.h>
55 #include <linux/pvclock_gtod.h>
56 #include <linux/kvm_irqfd.h>
57 #include <linux/irqbypass.h>
58 #include <linux/sched/stat.h>
59 #include <linux/sched/isolation.h>
60 #include <linux/mem_encrypt.h>
61 #include <linux/entry-kvm.h>
62 #include <linux/suspend.h>
64 #include <trace/events/kvm.h>
66 #include <asm/debugreg.h>
71 #include <linux/kernel_stat.h>
72 #include <asm/fpu/api.h>
73 #include <asm/fpu/xcr.h>
74 #include <asm/fpu/xstate.h>
75 #include <asm/pvclock.h>
76 #include <asm/div64.h>
77 #include <asm/irq_remapping.h>
78 #include <asm/mshyperv.h>
79 #include <asm/hypervisor.h>
80 #include <asm/tlbflush.h>
81 #include <asm/intel_pt.h>
82 #include <asm/emulate_prefix.h>
84 #include <clocksource/hyperv_timer.h>
86 #define CREATE_TRACE_POINTS
89 #define MAX_IO_MSRS 256
90 #define KVM_MAX_MCE_BANKS 32
92 struct kvm_caps kvm_caps __read_mostly = {
93 .supported_mce_cap = MCG_CTL_P | MCG_SER_P,
95 EXPORT_SYMBOL_GPL(kvm_caps);
97 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
99 #define emul_to_vcpu(ctxt) \
100 ((struct kvm_vcpu *)(ctxt)->vcpu)
103 * - enable syscall per default because its emulated by KVM
104 * - enable LME and LMA per default on 64 bit KVM
108 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
110 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
113 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
115 #define KVM_EXIT_HYPERCALL_VALID_MASK (1 << KVM_HC_MAP_GPA_RANGE)
117 #define KVM_CAP_PMU_VALID_MASK KVM_PMU_CAP_DISABLE
119 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
120 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
122 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
123 static void process_nmi(struct kvm_vcpu *vcpu);
124 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
125 static void store_regs(struct kvm_vcpu *vcpu);
126 static int sync_regs(struct kvm_vcpu *vcpu);
127 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu);
129 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
130 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2);
132 static DEFINE_MUTEX(vendor_module_lock);
133 struct kvm_x86_ops kvm_x86_ops __read_mostly;
135 #define KVM_X86_OP(func) \
136 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
137 *(((struct kvm_x86_ops *)0)->func));
138 #define KVM_X86_OP_OPTIONAL KVM_X86_OP
139 #define KVM_X86_OP_OPTIONAL_RET0 KVM_X86_OP
140 #include <asm/kvm-x86-ops.h>
141 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
142 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
144 static bool __read_mostly ignore_msrs = 0;
145 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
147 bool __read_mostly report_ignored_msrs = true;
148 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
149 EXPORT_SYMBOL_GPL(report_ignored_msrs);
151 unsigned int min_timer_period_us = 200;
152 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
154 static bool __read_mostly kvmclock_periodic_sync = true;
155 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
157 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
158 static u32 __read_mostly tsc_tolerance_ppm = 250;
159 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
162 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
163 * adaptive tuning starting from default advancement of 1000ns. '0' disables
164 * advancement entirely. Any other value is used as-is and disables adaptive
165 * tuning, i.e. allows privileged userspace to set an exact advancement time.
167 static int __read_mostly lapic_timer_advance_ns = -1;
168 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
170 static bool __read_mostly vector_hashing = true;
171 module_param(vector_hashing, bool, S_IRUGO);
173 bool __read_mostly enable_vmware_backdoor = false;
174 module_param(enable_vmware_backdoor, bool, S_IRUGO);
175 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
178 * Flags to manipulate forced emulation behavior (any non-zero value will
179 * enable forced emulation).
181 #define KVM_FEP_CLEAR_RFLAGS_RF BIT(1)
182 static int __read_mostly force_emulation_prefix;
183 module_param(force_emulation_prefix, int, 0644);
185 int __read_mostly pi_inject_timer = -1;
186 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
188 /* Enable/disable PMU virtualization */
189 bool __read_mostly enable_pmu = true;
190 EXPORT_SYMBOL_GPL(enable_pmu);
191 module_param(enable_pmu, bool, 0444);
193 bool __read_mostly eager_page_split = true;
194 module_param(eager_page_split, bool, 0644);
197 * Restoring the host value for MSRs that are only consumed when running in
198 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
199 * returns to userspace, i.e. the kernel can run with the guest's value.
201 #define KVM_MAX_NR_USER_RETURN_MSRS 16
203 struct kvm_user_return_msrs {
204 struct user_return_notifier urn;
206 struct kvm_user_return_msr_values {
209 } values[KVM_MAX_NR_USER_RETURN_MSRS];
212 u32 __read_mostly kvm_nr_uret_msrs;
213 EXPORT_SYMBOL_GPL(kvm_nr_uret_msrs);
214 static u32 __read_mostly kvm_uret_msrs_list[KVM_MAX_NR_USER_RETURN_MSRS];
215 static struct kvm_user_return_msrs __percpu *user_return_msrs;
217 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
218 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
219 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
220 | XFEATURE_MASK_PKRU | XFEATURE_MASK_XTILE)
222 u64 __read_mostly host_efer;
223 EXPORT_SYMBOL_GPL(host_efer);
225 bool __read_mostly allow_smaller_maxphyaddr = 0;
226 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
228 bool __read_mostly enable_apicv = true;
229 EXPORT_SYMBOL_GPL(enable_apicv);
231 u64 __read_mostly host_xss;
232 EXPORT_SYMBOL_GPL(host_xss);
234 const struct _kvm_stats_desc kvm_vm_stats_desc[] = {
235 KVM_GENERIC_VM_STATS(),
236 STATS_DESC_COUNTER(VM, mmu_shadow_zapped),
237 STATS_DESC_COUNTER(VM, mmu_pte_write),
238 STATS_DESC_COUNTER(VM, mmu_pde_zapped),
239 STATS_DESC_COUNTER(VM, mmu_flooded),
240 STATS_DESC_COUNTER(VM, mmu_recycled),
241 STATS_DESC_COUNTER(VM, mmu_cache_miss),
242 STATS_DESC_ICOUNTER(VM, mmu_unsync),
243 STATS_DESC_ICOUNTER(VM, pages_4k),
244 STATS_DESC_ICOUNTER(VM, pages_2m),
245 STATS_DESC_ICOUNTER(VM, pages_1g),
246 STATS_DESC_ICOUNTER(VM, nx_lpage_splits),
247 STATS_DESC_PCOUNTER(VM, max_mmu_rmap_size),
248 STATS_DESC_PCOUNTER(VM, max_mmu_page_hash_collisions)
251 const struct kvm_stats_header kvm_vm_stats_header = {
252 .name_size = KVM_STATS_NAME_SIZE,
253 .num_desc = ARRAY_SIZE(kvm_vm_stats_desc),
254 .id_offset = sizeof(struct kvm_stats_header),
255 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
256 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
257 sizeof(kvm_vm_stats_desc),
260 const struct _kvm_stats_desc kvm_vcpu_stats_desc[] = {
261 KVM_GENERIC_VCPU_STATS(),
262 STATS_DESC_COUNTER(VCPU, pf_taken),
263 STATS_DESC_COUNTER(VCPU, pf_fixed),
264 STATS_DESC_COUNTER(VCPU, pf_emulate),
265 STATS_DESC_COUNTER(VCPU, pf_spurious),
266 STATS_DESC_COUNTER(VCPU, pf_fast),
267 STATS_DESC_COUNTER(VCPU, pf_mmio_spte_created),
268 STATS_DESC_COUNTER(VCPU, pf_guest),
269 STATS_DESC_COUNTER(VCPU, tlb_flush),
270 STATS_DESC_COUNTER(VCPU, invlpg),
271 STATS_DESC_COUNTER(VCPU, exits),
272 STATS_DESC_COUNTER(VCPU, io_exits),
273 STATS_DESC_COUNTER(VCPU, mmio_exits),
274 STATS_DESC_COUNTER(VCPU, signal_exits),
275 STATS_DESC_COUNTER(VCPU, irq_window_exits),
276 STATS_DESC_COUNTER(VCPU, nmi_window_exits),
277 STATS_DESC_COUNTER(VCPU, l1d_flush),
278 STATS_DESC_COUNTER(VCPU, halt_exits),
279 STATS_DESC_COUNTER(VCPU, request_irq_exits),
280 STATS_DESC_COUNTER(VCPU, irq_exits),
281 STATS_DESC_COUNTER(VCPU, host_state_reload),
282 STATS_DESC_COUNTER(VCPU, fpu_reload),
283 STATS_DESC_COUNTER(VCPU, insn_emulation),
284 STATS_DESC_COUNTER(VCPU, insn_emulation_fail),
285 STATS_DESC_COUNTER(VCPU, hypercalls),
286 STATS_DESC_COUNTER(VCPU, irq_injections),
287 STATS_DESC_COUNTER(VCPU, nmi_injections),
288 STATS_DESC_COUNTER(VCPU, req_event),
289 STATS_DESC_COUNTER(VCPU, nested_run),
290 STATS_DESC_COUNTER(VCPU, directed_yield_attempted),
291 STATS_DESC_COUNTER(VCPU, directed_yield_successful),
292 STATS_DESC_COUNTER(VCPU, preemption_reported),
293 STATS_DESC_COUNTER(VCPU, preemption_other),
294 STATS_DESC_IBOOLEAN(VCPU, guest_mode),
295 STATS_DESC_COUNTER(VCPU, notify_window_exits),
298 const struct kvm_stats_header kvm_vcpu_stats_header = {
299 .name_size = KVM_STATS_NAME_SIZE,
300 .num_desc = ARRAY_SIZE(kvm_vcpu_stats_desc),
301 .id_offset = sizeof(struct kvm_stats_header),
302 .desc_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE,
303 .data_offset = sizeof(struct kvm_stats_header) + KVM_STATS_NAME_SIZE +
304 sizeof(kvm_vcpu_stats_desc),
307 u64 __read_mostly host_xcr0;
309 static struct kmem_cache *x86_emulator_cache;
312 * When called, it means the previous get/set msr reached an invalid msr.
313 * Return true if we want to ignore/silent this failed msr access.
315 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
317 const char *op = write ? "wrmsr" : "rdmsr";
320 if (report_ignored_msrs)
321 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
326 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
332 static struct kmem_cache *kvm_alloc_emulator_cache(void)
334 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
335 unsigned int size = sizeof(struct x86_emulate_ctxt);
337 return kmem_cache_create_usercopy("x86_emulator", size,
338 __alignof__(struct x86_emulate_ctxt),
339 SLAB_ACCOUNT, useroffset,
340 size - useroffset, NULL);
343 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
345 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
348 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
349 vcpu->arch.apf.gfns[i] = ~0;
352 static void kvm_on_user_return(struct user_return_notifier *urn)
355 struct kvm_user_return_msrs *msrs
356 = container_of(urn, struct kvm_user_return_msrs, urn);
357 struct kvm_user_return_msr_values *values;
361 * Disabling irqs at this point since the following code could be
362 * interrupted and executed through kvm_arch_hardware_disable()
364 local_irq_save(flags);
365 if (msrs->registered) {
366 msrs->registered = false;
367 user_return_notifier_unregister(urn);
369 local_irq_restore(flags);
370 for (slot = 0; slot < kvm_nr_uret_msrs; ++slot) {
371 values = &msrs->values[slot];
372 if (values->host != values->curr) {
373 wrmsrl(kvm_uret_msrs_list[slot], values->host);
374 values->curr = values->host;
379 static int kvm_probe_user_return_msr(u32 msr)
385 ret = rdmsrl_safe(msr, &val);
388 ret = wrmsrl_safe(msr, val);
394 int kvm_add_user_return_msr(u32 msr)
396 BUG_ON(kvm_nr_uret_msrs >= KVM_MAX_NR_USER_RETURN_MSRS);
398 if (kvm_probe_user_return_msr(msr))
401 kvm_uret_msrs_list[kvm_nr_uret_msrs] = msr;
402 return kvm_nr_uret_msrs++;
404 EXPORT_SYMBOL_GPL(kvm_add_user_return_msr);
406 int kvm_find_user_return_msr(u32 msr)
410 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
411 if (kvm_uret_msrs_list[i] == msr)
416 EXPORT_SYMBOL_GPL(kvm_find_user_return_msr);
418 static void kvm_user_return_msr_cpu_online(void)
420 unsigned int cpu = smp_processor_id();
421 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
425 for (i = 0; i < kvm_nr_uret_msrs; ++i) {
426 rdmsrl_safe(kvm_uret_msrs_list[i], &value);
427 msrs->values[i].host = value;
428 msrs->values[i].curr = value;
432 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
434 unsigned int cpu = smp_processor_id();
435 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
438 value = (value & mask) | (msrs->values[slot].host & ~mask);
439 if (value == msrs->values[slot].curr)
441 err = wrmsrl_safe(kvm_uret_msrs_list[slot], value);
445 msrs->values[slot].curr = value;
446 if (!msrs->registered) {
447 msrs->urn.on_user_return = kvm_on_user_return;
448 user_return_notifier_register(&msrs->urn);
449 msrs->registered = true;
453 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
455 static void drop_user_return_notifiers(void)
457 unsigned int cpu = smp_processor_id();
458 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
460 if (msrs->registered)
461 kvm_on_user_return(&msrs->urn);
464 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
466 return vcpu->arch.apic_base;
469 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
471 return kvm_apic_mode(kvm_get_apic_base(vcpu));
473 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
475 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
477 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
478 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
479 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
480 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
482 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
484 if (!msr_info->host_initiated) {
485 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
487 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
491 kvm_lapic_set_base(vcpu, msr_info->data);
492 kvm_recalculate_apic_map(vcpu->kvm);
497 * Handle a fault on a hardware virtualization (VMX or SVM) instruction.
499 * Hardware virtualization extension instructions may fault if a reboot turns
500 * off virtualization while processes are running. Usually after catching the
501 * fault we just panic; during reboot instead the instruction is ignored.
503 noinstr void kvm_spurious_fault(void)
505 /* Fault while not rebooting. We want the trace. */
506 BUG_ON(!kvm_rebooting);
508 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
510 #define EXCPT_BENIGN 0
511 #define EXCPT_CONTRIBUTORY 1
514 static int exception_class(int vector)
524 return EXCPT_CONTRIBUTORY;
531 #define EXCPT_FAULT 0
533 #define EXCPT_ABORT 2
534 #define EXCPT_INTERRUPT 3
537 static int exception_type(int vector)
541 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
542 return EXCPT_INTERRUPT;
547 * #DBs can be trap-like or fault-like, the caller must check other CPU
548 * state, e.g. DR6, to determine whether a #DB is a trap or fault.
550 if (mask & (1 << DB_VECTOR))
553 if (mask & ((1 << BP_VECTOR) | (1 << OF_VECTOR)))
556 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
559 /* Reserved exceptions will result in fault */
563 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu,
564 struct kvm_queued_exception *ex)
566 if (!ex->has_payload)
569 switch (ex->vector) {
572 * "Certain debug exceptions may clear bit 0-3. The
573 * remaining contents of the DR6 register are never
574 * cleared by the processor".
576 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
578 * In order to reflect the #DB exception payload in guest
579 * dr6, three components need to be considered: active low
580 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
582 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
583 * In the target guest dr6:
584 * FIXED_1 bits should always be set.
585 * Active low bits should be cleared if 1-setting in payload.
586 * Active high bits should be set if 1-setting in payload.
588 * Note, the payload is compatible with the pending debug
589 * exceptions/exit qualification under VMX, that active_low bits
590 * are active high in payload.
591 * So they need to be flipped for DR6.
593 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
594 vcpu->arch.dr6 |= ex->payload;
595 vcpu->arch.dr6 ^= ex->payload & DR6_ACTIVE_LOW;
598 * The #DB payload is defined as compatible with the 'pending
599 * debug exceptions' field under VMX, not DR6. While bit 12 is
600 * defined in the 'pending debug exceptions' field (enabled
601 * breakpoint), it is reserved and must be zero in DR6.
603 vcpu->arch.dr6 &= ~BIT(12);
606 vcpu->arch.cr2 = ex->payload;
610 ex->has_payload = false;
613 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
615 static void kvm_queue_exception_vmexit(struct kvm_vcpu *vcpu, unsigned int vector,
616 bool has_error_code, u32 error_code,
617 bool has_payload, unsigned long payload)
619 struct kvm_queued_exception *ex = &vcpu->arch.exception_vmexit;
622 ex->injected = false;
624 ex->has_error_code = has_error_code;
625 ex->error_code = error_code;
626 ex->has_payload = has_payload;
627 ex->payload = payload;
630 /* Forcibly leave the nested mode in cases like a vCPU reset */
631 static void kvm_leave_nested(struct kvm_vcpu *vcpu)
633 kvm_x86_ops.nested_ops->leave_nested(vcpu);
636 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
637 unsigned nr, bool has_error, u32 error_code,
638 bool has_payload, unsigned long payload, bool reinject)
643 kvm_make_request(KVM_REQ_EVENT, vcpu);
646 * If the exception is destined for L2 and isn't being reinjected,
647 * morph it to a VM-Exit if L1 wants to intercept the exception. A
648 * previously injected exception is not checked because it was checked
649 * when it was original queued, and re-checking is incorrect if _L1_
650 * injected the exception, in which case it's exempt from interception.
652 if (!reinject && is_guest_mode(vcpu) &&
653 kvm_x86_ops.nested_ops->is_exception_vmexit(vcpu, nr, error_code)) {
654 kvm_queue_exception_vmexit(vcpu, nr, has_error, error_code,
655 has_payload, payload);
659 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
663 * On VM-Entry, an exception can be pending if and only
664 * if event injection was blocked by nested_run_pending.
665 * In that case, however, vcpu_enter_guest() requests an
666 * immediate exit, and the guest shouldn't proceed far
667 * enough to need reinjection.
669 WARN_ON_ONCE(kvm_is_exception_pending(vcpu));
670 vcpu->arch.exception.injected = true;
671 if (WARN_ON_ONCE(has_payload)) {
673 * A reinjected event has already
674 * delivered its payload.
680 vcpu->arch.exception.pending = true;
681 vcpu->arch.exception.injected = false;
683 vcpu->arch.exception.has_error_code = has_error;
684 vcpu->arch.exception.vector = nr;
685 vcpu->arch.exception.error_code = error_code;
686 vcpu->arch.exception.has_payload = has_payload;
687 vcpu->arch.exception.payload = payload;
688 if (!is_guest_mode(vcpu))
689 kvm_deliver_exception_payload(vcpu,
690 &vcpu->arch.exception);
694 /* to check exception */
695 prev_nr = vcpu->arch.exception.vector;
696 if (prev_nr == DF_VECTOR) {
697 /* triple fault -> shutdown */
698 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
701 class1 = exception_class(prev_nr);
702 class2 = exception_class(nr);
703 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY) ||
704 (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
706 * Synthesize #DF. Clear the previously injected or pending
707 * exception so as not to incorrectly trigger shutdown.
709 vcpu->arch.exception.injected = false;
710 vcpu->arch.exception.pending = false;
712 kvm_queue_exception_e(vcpu, DF_VECTOR, 0);
714 /* replace previous exception with a new one in a hope
715 that instruction re-execution will regenerate lost
721 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
723 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
725 EXPORT_SYMBOL_GPL(kvm_queue_exception);
727 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
729 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
731 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
733 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
734 unsigned long payload)
736 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
738 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
740 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
741 u32 error_code, unsigned long payload)
743 kvm_multiple_exception(vcpu, nr, true, error_code,
744 true, payload, false);
747 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
750 kvm_inject_gp(vcpu, 0);
752 return kvm_skip_emulated_instruction(vcpu);
756 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
758 static int complete_emulated_insn_gp(struct kvm_vcpu *vcpu, int err)
761 kvm_inject_gp(vcpu, 0);
765 return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE | EMULTYPE_SKIP |
766 EMULTYPE_COMPLETE_USER_EXIT);
769 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
771 ++vcpu->stat.pf_guest;
774 * Async #PF in L2 is always forwarded to L1 as a VM-Exit regardless of
775 * whether or not L1 wants to intercept "regular" #PF.
777 if (is_guest_mode(vcpu) && fault->async_page_fault)
778 kvm_queue_exception_vmexit(vcpu, PF_VECTOR,
779 true, fault->error_code,
780 true, fault->address);
782 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
786 void kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
787 struct x86_exception *fault)
789 struct kvm_mmu *fault_mmu;
790 WARN_ON_ONCE(fault->vector != PF_VECTOR);
792 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
796 * Invalidate the TLB entry for the faulting address, if it exists,
797 * else the access will fault indefinitely (and to emulate hardware).
799 if ((fault->error_code & PFERR_PRESENT_MASK) &&
800 !(fault->error_code & PFERR_RSVD_MASK))
801 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
802 fault_mmu->root.hpa);
804 fault_mmu->inject_page_fault(vcpu, fault);
806 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
808 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
810 atomic_inc(&vcpu->arch.nmi_queued);
811 kvm_make_request(KVM_REQ_NMI, vcpu);
814 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
816 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
818 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
820 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
822 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
824 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
827 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
828 * a #GP and return false.
830 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
832 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
834 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
838 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
840 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
843 kvm_queue_exception(vcpu, UD_VECTOR);
846 EXPORT_SYMBOL_GPL(kvm_require_dr);
848 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
850 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
854 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
856 int load_pdptrs(struct kvm_vcpu *vcpu, unsigned long cr3)
858 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
859 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
863 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
866 * If the MMU is nested, CR3 holds an L2 GPA and needs to be translated
869 real_gpa = kvm_translate_gpa(vcpu, mmu, gfn_to_gpa(pdpt_gfn),
870 PFERR_USER_MASK | PFERR_WRITE_MASK, NULL);
871 if (real_gpa == INVALID_GPA)
874 /* Note the offset, PDPTRs are 32 byte aligned when using PAE paging. */
875 ret = kvm_vcpu_read_guest_page(vcpu, gpa_to_gfn(real_gpa), pdpte,
876 cr3 & GENMASK(11, 5), sizeof(pdpte));
880 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
881 if ((pdpte[i] & PT_PRESENT_MASK) &&
882 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
888 * Marking VCPU_EXREG_PDPTR dirty doesn't work for !tdp_enabled.
889 * Shadow page roots need to be reconstructed instead.
891 if (!tdp_enabled && memcmp(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs)))
892 kvm_mmu_free_roots(vcpu->kvm, mmu, KVM_MMU_ROOT_CURRENT);
894 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
895 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
896 kvm_make_request(KVM_REQ_LOAD_MMU_PGD, vcpu);
897 vcpu->arch.pdptrs_from_userspace = false;
901 EXPORT_SYMBOL_GPL(load_pdptrs);
903 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
905 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
906 kvm_clear_async_pf_completion_queue(vcpu);
907 kvm_async_pf_hash_reset(vcpu);
910 * Clearing CR0.PG is defined to flush the TLB from the guest's
913 if (!(cr0 & X86_CR0_PG))
914 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
917 if ((cr0 ^ old_cr0) & KVM_MMU_CR0_ROLE_BITS)
918 kvm_mmu_reset_context(vcpu);
920 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
921 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
922 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
923 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
925 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
927 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
929 unsigned long old_cr0 = kvm_read_cr0(vcpu);
934 if (cr0 & 0xffffffff00000000UL)
938 cr0 &= ~CR0_RESERVED_BITS;
940 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
943 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
947 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
948 (cr0 & X86_CR0_PG)) {
953 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
958 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
959 is_pae(vcpu) && ((cr0 ^ old_cr0) & X86_CR0_PDPTR_BITS) &&
960 !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
963 if (!(cr0 & X86_CR0_PG) &&
964 (is_64_bit_mode(vcpu) || kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE)))
967 static_call(kvm_x86_set_cr0)(vcpu, cr0);
969 kvm_post_set_cr0(vcpu, old_cr0, cr0);
973 EXPORT_SYMBOL_GPL(kvm_set_cr0);
975 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
977 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
979 EXPORT_SYMBOL_GPL(kvm_lmsw);
981 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
983 if (vcpu->arch.guest_state_protected)
986 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
988 if (vcpu->arch.xcr0 != host_xcr0)
989 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
991 if (vcpu->arch.xsaves_enabled &&
992 vcpu->arch.ia32_xss != host_xss)
993 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
996 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
997 if (static_cpu_has(X86_FEATURE_PKU) &&
998 vcpu->arch.pkru != vcpu->arch.host_pkru &&
999 ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
1000 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)))
1001 write_pkru(vcpu->arch.pkru);
1002 #endif /* CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS */
1004 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
1006 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
1008 if (vcpu->arch.guest_state_protected)
1011 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
1012 if (static_cpu_has(X86_FEATURE_PKU) &&
1013 ((vcpu->arch.xcr0 & XFEATURE_MASK_PKRU) ||
1014 kvm_read_cr4_bits(vcpu, X86_CR4_PKE))) {
1015 vcpu->arch.pkru = rdpkru();
1016 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
1017 write_pkru(vcpu->arch.host_pkru);
1019 #endif /* CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS */
1021 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
1023 if (vcpu->arch.xcr0 != host_xcr0)
1024 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
1026 if (vcpu->arch.xsaves_enabled &&
1027 vcpu->arch.ia32_xss != host_xss)
1028 wrmsrl(MSR_IA32_XSS, host_xss);
1032 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
1034 #ifdef CONFIG_X86_64
1035 static inline u64 kvm_guest_supported_xfd(struct kvm_vcpu *vcpu)
1037 return vcpu->arch.guest_supported_xcr0 & XFEATURE_MASK_USER_DYNAMIC;
1041 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
1044 u64 old_xcr0 = vcpu->arch.xcr0;
1047 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
1048 if (index != XCR_XFEATURE_ENABLED_MASK)
1050 if (!(xcr0 & XFEATURE_MASK_FP))
1052 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
1056 * Do not allow the guest to set bits that we do not support
1057 * saving. However, xcr0 bit 0 is always set, even if the
1058 * emulated CPU does not support XSAVE (see kvm_vcpu_reset()).
1060 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
1061 if (xcr0 & ~valid_bits)
1064 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
1065 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
1068 if (xcr0 & XFEATURE_MASK_AVX512) {
1069 if (!(xcr0 & XFEATURE_MASK_YMM))
1071 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
1075 if ((xcr0 & XFEATURE_MASK_XTILE) &&
1076 ((xcr0 & XFEATURE_MASK_XTILE) != XFEATURE_MASK_XTILE))
1079 vcpu->arch.xcr0 = xcr0;
1081 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
1082 kvm_update_cpuid_runtime(vcpu);
1086 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
1088 /* Note, #UD due to CR4.OSXSAVE=0 has priority over the intercept. */
1089 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
1090 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
1091 kvm_inject_gp(vcpu, 0);
1095 return kvm_skip_emulated_instruction(vcpu);
1097 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1099 bool __kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1101 if (cr4 & cr4_reserved_bits)
1104 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1109 EXPORT_SYMBOL_GPL(__kvm_is_valid_cr4);
1111 static bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1113 return __kvm_is_valid_cr4(vcpu, cr4) &&
1114 static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1117 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1119 if ((cr4 ^ old_cr4) & KVM_MMU_CR4_ROLE_BITS)
1120 kvm_mmu_reset_context(vcpu);
1123 * If CR4.PCIDE is changed 0 -> 1, there is no need to flush the TLB
1124 * according to the SDM; however, stale prev_roots could be reused
1125 * incorrectly in the future after a MOV to CR3 with NOFLUSH=1, so we
1126 * free them all. This is *not* a superset of KVM_REQ_TLB_FLUSH_GUEST
1127 * or KVM_REQ_TLB_FLUSH_CURRENT, because the hardware TLB is not flushed,
1131 (cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE))
1132 kvm_mmu_unload(vcpu);
1135 * The TLB has to be flushed for all PCIDs if any of the following
1136 * (architecturally required) changes happen:
1137 * - CR4.PCIDE is changed from 1 to 0
1138 * - CR4.PGE is toggled
1140 * This is a superset of KVM_REQ_TLB_FLUSH_CURRENT.
1142 if (((cr4 ^ old_cr4) & X86_CR4_PGE) ||
1143 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1144 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1147 * The TLB has to be flushed for the current PCID if any of the
1148 * following (architecturally required) changes happen:
1149 * - CR4.SMEP is changed from 0 to 1
1150 * - CR4.PAE is toggled
1152 else if (((cr4 ^ old_cr4) & X86_CR4_PAE) ||
1153 ((cr4 & X86_CR4_SMEP) && !(old_cr4 & X86_CR4_SMEP)))
1154 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1157 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1159 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1161 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1163 if (!kvm_is_valid_cr4(vcpu, cr4))
1166 if (is_long_mode(vcpu)) {
1167 if (!(cr4 & X86_CR4_PAE))
1169 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1171 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1172 && ((cr4 ^ old_cr4) & X86_CR4_PDPTR_BITS)
1173 && !load_pdptrs(vcpu, kvm_read_cr3(vcpu)))
1176 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1177 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1180 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1181 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1185 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1187 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1191 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1193 static void kvm_invalidate_pcid(struct kvm_vcpu *vcpu, unsigned long pcid)
1195 struct kvm_mmu *mmu = vcpu->arch.mmu;
1196 unsigned long roots_to_free = 0;
1200 * MOV CR3 and INVPCID are usually not intercepted when using TDP, but
1201 * this is reachable when running EPT=1 and unrestricted_guest=0, and
1202 * also via the emulator. KVM's TDP page tables are not in the scope of
1203 * the invalidation, but the guest's TLB entries need to be flushed as
1204 * the CPU may have cached entries in its TLB for the target PCID.
1206 if (unlikely(tdp_enabled)) {
1207 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
1212 * If neither the current CR3 nor any of the prev_roots use the given
1213 * PCID, then nothing needs to be done here because a resync will
1214 * happen anyway before switching to any other CR3.
1216 if (kvm_get_active_pcid(vcpu) == pcid) {
1217 kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
1218 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1222 * If PCID is disabled, there is no need to free prev_roots even if the
1223 * PCIDs for them are also 0, because MOV to CR3 always flushes the TLB
1226 if (!kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
1229 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
1230 if (kvm_get_pcid(vcpu, mmu->prev_roots[i].pgd) == pcid)
1231 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
1233 kvm_mmu_free_roots(vcpu->kvm, mmu, roots_to_free);
1236 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1238 bool skip_tlb_flush = false;
1239 unsigned long pcid = 0;
1240 #ifdef CONFIG_X86_64
1241 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1244 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1245 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1246 pcid = cr3 & X86_CR3_PCID_MASK;
1250 /* PDPTRs are always reloaded for PAE paging. */
1251 if (cr3 == kvm_read_cr3(vcpu) && !is_pae_paging(vcpu))
1252 goto handle_tlb_flush;
1255 * Do not condition the GPA check on long mode, this helper is used to
1256 * stuff CR3, e.g. for RSM emulation, and there is no guarantee that
1257 * the current vCPU mode is accurate.
1259 if (kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1262 if (is_pae_paging(vcpu) && !load_pdptrs(vcpu, cr3))
1265 if (cr3 != kvm_read_cr3(vcpu))
1266 kvm_mmu_new_pgd(vcpu, cr3);
1268 vcpu->arch.cr3 = cr3;
1269 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
1270 /* Do not call post_set_cr3, we do not get here for confidential guests. */
1274 * A load of CR3 that flushes the TLB flushes only the current PCID,
1275 * even if PCID is disabled, in which case PCID=0 is flushed. It's a
1276 * moot point in the end because _disabling_ PCID will flush all PCIDs,
1277 * and it's impossible to use a non-zero PCID when PCID is disabled,
1278 * i.e. only PCID=0 can be relevant.
1280 if (!skip_tlb_flush)
1281 kvm_invalidate_pcid(vcpu, pcid);
1285 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1287 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1289 if (cr8 & CR8_RESERVED_BITS)
1291 if (lapic_in_kernel(vcpu))
1292 kvm_lapic_set_tpr(vcpu, cr8);
1294 vcpu->arch.cr8 = cr8;
1297 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1299 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1301 if (lapic_in_kernel(vcpu))
1302 return kvm_lapic_get_cr8(vcpu);
1304 return vcpu->arch.cr8;
1306 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1308 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1312 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1313 for (i = 0; i < KVM_NR_DB_REGS; i++)
1314 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1318 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1322 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1323 dr7 = vcpu->arch.guest_debug_dr7;
1325 dr7 = vcpu->arch.dr7;
1326 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1327 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1328 if (dr7 & DR7_BP_EN_MASK)
1329 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1331 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1333 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1335 u64 fixed = DR6_FIXED_1;
1337 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1340 if (!guest_cpuid_has(vcpu, X86_FEATURE_BUS_LOCK_DETECT))
1341 fixed |= DR6_BUS_LOCK;
1345 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1347 size_t size = ARRAY_SIZE(vcpu->arch.db);
1351 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1352 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1353 vcpu->arch.eff_db[dr] = val;
1357 if (!kvm_dr6_valid(val))
1359 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1363 if (!kvm_dr7_valid(val))
1365 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1366 kvm_update_dr7(vcpu);
1372 EXPORT_SYMBOL_GPL(kvm_set_dr);
1374 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1376 size_t size = ARRAY_SIZE(vcpu->arch.db);
1380 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1384 *val = vcpu->arch.dr6;
1388 *val = vcpu->arch.dr7;
1392 EXPORT_SYMBOL_GPL(kvm_get_dr);
1394 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1396 u32 ecx = kvm_rcx_read(vcpu);
1399 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1400 kvm_inject_gp(vcpu, 0);
1404 kvm_rax_write(vcpu, (u32)data);
1405 kvm_rdx_write(vcpu, data >> 32);
1406 return kvm_skip_emulated_instruction(vcpu);
1408 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1411 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1412 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1414 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1415 * extract the supported MSRs from the related const lists.
1416 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1417 * capabilities of the host cpu. This capabilities test skips MSRs that are
1418 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1419 * may depend on host virtualization features rather than host cpu features.
1422 static const u32 msrs_to_save_all[] = {
1423 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1425 #ifdef CONFIG_X86_64
1426 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1428 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1429 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1431 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1432 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1433 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1434 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1435 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1436 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1437 MSR_IA32_UMWAIT_CONTROL,
1439 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1440 MSR_ARCH_PERFMON_FIXED_CTR0 + 2,
1441 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1442 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1443 MSR_IA32_PEBS_ENABLE, MSR_IA32_DS_AREA, MSR_PEBS_DATA_CFG,
1445 /* This part of MSRs should match KVM_INTEL_PMC_MAX_GENERIC. */
1446 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1447 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1448 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1449 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1450 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1451 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1452 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1453 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1455 MSR_K7_EVNTSEL0, MSR_K7_EVNTSEL1, MSR_K7_EVNTSEL2, MSR_K7_EVNTSEL3,
1456 MSR_K7_PERFCTR0, MSR_K7_PERFCTR1, MSR_K7_PERFCTR2, MSR_K7_PERFCTR3,
1458 /* This part of MSRs should match KVM_AMD_PMC_MAX_GENERIC. */
1459 MSR_F15H_PERF_CTL0, MSR_F15H_PERF_CTL1, MSR_F15H_PERF_CTL2,
1460 MSR_F15H_PERF_CTL3, MSR_F15H_PERF_CTL4, MSR_F15H_PERF_CTL5,
1461 MSR_F15H_PERF_CTR0, MSR_F15H_PERF_CTR1, MSR_F15H_PERF_CTR2,
1462 MSR_F15H_PERF_CTR3, MSR_F15H_PERF_CTR4, MSR_F15H_PERF_CTR5,
1464 MSR_IA32_XFD, MSR_IA32_XFD_ERR,
1467 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1468 static unsigned num_msrs_to_save;
1470 static const u32 emulated_msrs_all[] = {
1471 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1472 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1473 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1474 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1475 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1476 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1477 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1479 HV_X64_MSR_VP_INDEX,
1480 HV_X64_MSR_VP_RUNTIME,
1481 HV_X64_MSR_SCONTROL,
1482 HV_X64_MSR_STIMER0_CONFIG,
1483 HV_X64_MSR_VP_ASSIST_PAGE,
1484 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1485 HV_X64_MSR_TSC_EMULATION_STATUS, HV_X64_MSR_TSC_INVARIANT_CONTROL,
1486 HV_X64_MSR_SYNDBG_OPTIONS,
1487 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1488 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1489 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1491 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1492 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1494 MSR_IA32_TSC_ADJUST,
1495 MSR_IA32_TSC_DEADLINE,
1496 MSR_IA32_ARCH_CAPABILITIES,
1497 MSR_IA32_PERF_CAPABILITIES,
1498 MSR_IA32_MISC_ENABLE,
1499 MSR_IA32_MCG_STATUS,
1501 MSR_IA32_MCG_EXT_CTL,
1505 MSR_MISC_FEATURES_ENABLES,
1506 MSR_AMD64_VIRT_SPEC_CTRL,
1507 MSR_AMD64_TSC_RATIO,
1512 * The following list leaves out MSRs whose values are determined
1513 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1514 * We always support the "true" VMX control MSRs, even if the host
1515 * processor does not, so I am putting these registers here rather
1516 * than in msrs_to_save_all.
1519 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1520 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1521 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1522 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1524 MSR_IA32_VMX_CR0_FIXED0,
1525 MSR_IA32_VMX_CR4_FIXED0,
1526 MSR_IA32_VMX_VMCS_ENUM,
1527 MSR_IA32_VMX_PROCBASED_CTLS2,
1528 MSR_IA32_VMX_EPT_VPID_CAP,
1529 MSR_IA32_VMX_VMFUNC,
1532 MSR_KVM_POLL_CONTROL,
1535 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1536 static unsigned num_emulated_msrs;
1539 * List of msr numbers which are used to expose MSR-based features that
1540 * can be used by a hypervisor to validate requested CPU features.
1542 static const u32 msr_based_features_all[] = {
1544 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1545 MSR_IA32_VMX_PINBASED_CTLS,
1546 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1547 MSR_IA32_VMX_PROCBASED_CTLS,
1548 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1549 MSR_IA32_VMX_EXIT_CTLS,
1550 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1551 MSR_IA32_VMX_ENTRY_CTLS,
1553 MSR_IA32_VMX_CR0_FIXED0,
1554 MSR_IA32_VMX_CR0_FIXED1,
1555 MSR_IA32_VMX_CR4_FIXED0,
1556 MSR_IA32_VMX_CR4_FIXED1,
1557 MSR_IA32_VMX_VMCS_ENUM,
1558 MSR_IA32_VMX_PROCBASED_CTLS2,
1559 MSR_IA32_VMX_EPT_VPID_CAP,
1560 MSR_IA32_VMX_VMFUNC,
1564 MSR_IA32_ARCH_CAPABILITIES,
1565 MSR_IA32_PERF_CAPABILITIES,
1568 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1569 static unsigned int num_msr_based_features;
1572 * Some IA32_ARCH_CAPABILITIES bits have dependencies on MSRs that KVM
1573 * does not yet virtualize. These include:
1574 * 10 - MISC_PACKAGE_CTRLS
1575 * 11 - ENERGY_FILTERING_CTL
1577 * 18 - FB_CLEAR_CTRL
1578 * 21 - XAPIC_DISABLE_STATUS
1579 * 23 - OVERCLOCKING_STATUS
1582 #define KVM_SUPPORTED_ARCH_CAP \
1583 (ARCH_CAP_RDCL_NO | ARCH_CAP_IBRS_ALL | ARCH_CAP_RSBA | \
1584 ARCH_CAP_SKIP_VMENTRY_L1DFLUSH | ARCH_CAP_SSB_NO | ARCH_CAP_MDS_NO | \
1585 ARCH_CAP_PSCHANGE_MC_NO | ARCH_CAP_TSX_CTRL_MSR | ARCH_CAP_TAA_NO | \
1586 ARCH_CAP_SBDR_SSDP_NO | ARCH_CAP_FBSDP_NO | ARCH_CAP_PSDP_NO | \
1587 ARCH_CAP_FB_CLEAR | ARCH_CAP_RRSBA | ARCH_CAP_PBRSB_NO)
1589 static u64 kvm_get_arch_capabilities(void)
1593 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES)) {
1594 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1595 data &= KVM_SUPPORTED_ARCH_CAP;
1599 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1600 * the nested hypervisor runs with NX huge pages. If it is not,
1601 * L1 is anyway vulnerable to ITLB_MULTIHIT exploits from other
1602 * L1 guests, so it need not worry about its own (L2) guests.
1604 data |= ARCH_CAP_PSCHANGE_MC_NO;
1607 * If we're doing cache flushes (either "always" or "cond")
1608 * we will do one whenever the guest does a vmlaunch/vmresume.
1609 * If an outer hypervisor is doing the cache flush for us
1610 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1611 * capability to the guest too, and if EPT is disabled we're not
1612 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1613 * require a nested hypervisor to do a flush of its own.
1615 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1616 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1618 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1619 data |= ARCH_CAP_RDCL_NO;
1620 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1621 data |= ARCH_CAP_SSB_NO;
1622 if (!boot_cpu_has_bug(X86_BUG_MDS))
1623 data |= ARCH_CAP_MDS_NO;
1625 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1627 * If RTM=0 because the kernel has disabled TSX, the host might
1628 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1629 * and therefore knows that there cannot be TAA) but keep
1630 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1631 * and we want to allow migrating those guests to tsx=off hosts.
1633 data &= ~ARCH_CAP_TAA_NO;
1634 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1635 data |= ARCH_CAP_TAA_NO;
1638 * Nothing to do here; we emulate TSX_CTRL if present on the
1639 * host so the guest can choose between disabling TSX or
1640 * using VERW to clear CPU buffers.
1647 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1649 switch (msr->index) {
1650 case MSR_IA32_ARCH_CAPABILITIES:
1651 msr->data = kvm_get_arch_capabilities();
1653 case MSR_IA32_PERF_CAPABILITIES:
1654 msr->data = kvm_caps.supported_perf_cap;
1656 case MSR_IA32_UCODE_REV:
1657 rdmsrl_safe(msr->index, &msr->data);
1660 return static_call(kvm_x86_get_msr_feature)(msr);
1665 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1667 struct kvm_msr_entry msr;
1671 r = kvm_get_msr_feature(&msr);
1673 if (r == KVM_MSR_RET_INVALID) {
1674 /* Unconditionally clear the output for simplicity */
1676 if (kvm_msr_ignored_check(index, 0, false))
1688 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1690 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1693 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1696 if (efer & (EFER_LME | EFER_LMA) &&
1697 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1700 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1706 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1708 if (efer & efer_reserved_bits)
1711 return __kvm_valid_efer(vcpu, efer);
1713 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1715 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1717 u64 old_efer = vcpu->arch.efer;
1718 u64 efer = msr_info->data;
1721 if (efer & efer_reserved_bits)
1724 if (!msr_info->host_initiated) {
1725 if (!__kvm_valid_efer(vcpu, efer))
1728 if (is_paging(vcpu) &&
1729 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1734 efer |= vcpu->arch.efer & EFER_LMA;
1736 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1742 if ((efer ^ old_efer) & KVM_MMU_EFER_ROLE_BITS)
1743 kvm_mmu_reset_context(vcpu);
1748 void kvm_enable_efer_bits(u64 mask)
1750 efer_reserved_bits &= ~mask;
1752 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1754 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1756 struct kvm_x86_msr_filter *msr_filter;
1757 struct msr_bitmap_range *ranges;
1758 struct kvm *kvm = vcpu->kvm;
1763 /* x2APIC MSRs do not support filtering. */
1764 if (index >= 0x800 && index <= 0x8ff)
1767 idx = srcu_read_lock(&kvm->srcu);
1769 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1775 allowed = msr_filter->default_allow;
1776 ranges = msr_filter->ranges;
1778 for (i = 0; i < msr_filter->count; i++) {
1779 u32 start = ranges[i].base;
1780 u32 end = start + ranges[i].nmsrs;
1781 u32 flags = ranges[i].flags;
1782 unsigned long *bitmap = ranges[i].bitmap;
1784 if ((index >= start) && (index < end) && (flags & type)) {
1785 allowed = !!test_bit(index - start, bitmap);
1791 srcu_read_unlock(&kvm->srcu, idx);
1795 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1798 * Write @data into the MSR specified by @index. Select MSR specific fault
1799 * checks are bypassed if @host_initiated is %true.
1800 * Returns 0 on success, non-0 otherwise.
1801 * Assumes vcpu_load() was already called.
1803 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1804 bool host_initiated)
1806 struct msr_data msr;
1811 case MSR_KERNEL_GS_BASE:
1814 if (is_noncanonical_address(data, vcpu))
1817 case MSR_IA32_SYSENTER_EIP:
1818 case MSR_IA32_SYSENTER_ESP:
1820 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1821 * non-canonical address is written on Intel but not on
1822 * AMD (which ignores the top 32-bits, because it does
1823 * not implement 64-bit SYSENTER).
1825 * 64-bit code should hence be able to write a non-canonical
1826 * value on AMD. Making the address canonical ensures that
1827 * vmentry does not fail on Intel after writing a non-canonical
1828 * value, and that something deterministic happens if the guest
1829 * invokes 64-bit SYSENTER.
1831 data = __canonical_address(data, vcpu_virt_addr_bits(vcpu));
1834 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1837 if (!host_initiated &&
1838 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1839 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1843 * Per Intel's SDM, bits 63:32 are reserved, but AMD's APM has
1844 * incomplete and conflicting architectural behavior. Current
1845 * AMD CPUs completely ignore bits 63:32, i.e. they aren't
1846 * reserved and always read as zeros. Enforce Intel's reserved
1847 * bits check if and only if the guest CPU is Intel, and clear
1848 * the bits in all other cases. This ensures cross-vendor
1849 * migration will provide consistent behavior for the guest.
1851 if (guest_cpuid_is_intel(vcpu) && (data >> 32) != 0)
1860 msr.host_initiated = host_initiated;
1862 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1865 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1866 u32 index, u64 data, bool host_initiated)
1868 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1870 if (ret == KVM_MSR_RET_INVALID)
1871 if (kvm_msr_ignored_check(index, data, true))
1878 * Read the MSR specified by @index into @data. Select MSR specific fault
1879 * checks are bypassed if @host_initiated is %true.
1880 * Returns 0 on success, non-0 otherwise.
1881 * Assumes vcpu_load() was already called.
1883 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1884 bool host_initiated)
1886 struct msr_data msr;
1891 if (!kvm_is_supported_user_return_msr(MSR_TSC_AUX))
1894 if (!host_initiated &&
1895 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) &&
1896 !guest_cpuid_has(vcpu, X86_FEATURE_RDPID))
1902 msr.host_initiated = host_initiated;
1904 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1910 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1911 u32 index, u64 *data, bool host_initiated)
1913 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1915 if (ret == KVM_MSR_RET_INVALID) {
1916 /* Unconditionally clear *data for simplicity */
1918 if (kvm_msr_ignored_check(index, 0, false))
1925 static int kvm_get_msr_with_filter(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1927 if (!kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1928 return KVM_MSR_RET_FILTERED;
1929 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1932 static int kvm_set_msr_with_filter(struct kvm_vcpu *vcpu, u32 index, u64 data)
1934 if (!kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1935 return KVM_MSR_RET_FILTERED;
1936 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1939 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1941 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1943 EXPORT_SYMBOL_GPL(kvm_get_msr);
1945 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1947 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1949 EXPORT_SYMBOL_GPL(kvm_set_msr);
1951 static void complete_userspace_rdmsr(struct kvm_vcpu *vcpu)
1953 if (!vcpu->run->msr.error) {
1954 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1955 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1959 static int complete_emulated_msr_access(struct kvm_vcpu *vcpu)
1961 return complete_emulated_insn_gp(vcpu, vcpu->run->msr.error);
1964 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1966 complete_userspace_rdmsr(vcpu);
1967 return complete_emulated_msr_access(vcpu);
1970 static int complete_fast_msr_access(struct kvm_vcpu *vcpu)
1972 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1975 static int complete_fast_rdmsr(struct kvm_vcpu *vcpu)
1977 complete_userspace_rdmsr(vcpu);
1978 return complete_fast_msr_access(vcpu);
1981 static u64 kvm_msr_reason(int r)
1984 case KVM_MSR_RET_INVALID:
1985 return KVM_MSR_EXIT_REASON_UNKNOWN;
1986 case KVM_MSR_RET_FILTERED:
1987 return KVM_MSR_EXIT_REASON_FILTER;
1989 return KVM_MSR_EXIT_REASON_INVAL;
1993 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1994 u32 exit_reason, u64 data,
1995 int (*completion)(struct kvm_vcpu *vcpu),
1998 u64 msr_reason = kvm_msr_reason(r);
2000 /* Check if the user wanted to know about this MSR fault */
2001 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
2004 vcpu->run->exit_reason = exit_reason;
2005 vcpu->run->msr.error = 0;
2006 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
2007 vcpu->run->msr.reason = msr_reason;
2008 vcpu->run->msr.index = index;
2009 vcpu->run->msr.data = data;
2010 vcpu->arch.complete_userspace_io = completion;
2015 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
2017 u32 ecx = kvm_rcx_read(vcpu);
2021 r = kvm_get_msr_with_filter(vcpu, ecx, &data);
2024 trace_kvm_msr_read(ecx, data);
2026 kvm_rax_write(vcpu, data & -1u);
2027 kvm_rdx_write(vcpu, (data >> 32) & -1u);
2029 /* MSR read failed? See if we should ask user space */
2030 if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_RDMSR, 0,
2031 complete_fast_rdmsr, r))
2033 trace_kvm_msr_read_ex(ecx);
2036 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
2038 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
2040 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
2042 u32 ecx = kvm_rcx_read(vcpu);
2043 u64 data = kvm_read_edx_eax(vcpu);
2046 r = kvm_set_msr_with_filter(vcpu, ecx, data);
2049 trace_kvm_msr_write(ecx, data);
2051 /* MSR write failed? See if we should ask user space */
2052 if (kvm_msr_user_space(vcpu, ecx, KVM_EXIT_X86_WRMSR, data,
2053 complete_fast_msr_access, r))
2055 /* Signal all other negative errors to userspace */
2058 trace_kvm_msr_write_ex(ecx, data);
2061 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
2063 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
2065 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
2067 return kvm_skip_emulated_instruction(vcpu);
2070 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
2072 /* Treat an INVD instruction as a NOP and just skip it. */
2073 return kvm_emulate_as_nop(vcpu);
2075 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
2077 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
2079 kvm_queue_exception(vcpu, UD_VECTOR);
2082 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
2085 static int kvm_emulate_monitor_mwait(struct kvm_vcpu *vcpu, const char *insn)
2087 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MWAIT_NEVER_UD_FAULTS) &&
2088 !guest_cpuid_has(vcpu, X86_FEATURE_MWAIT))
2089 return kvm_handle_invalid_op(vcpu);
2091 pr_warn_once("%s instruction emulated as NOP!\n", insn);
2092 return kvm_emulate_as_nop(vcpu);
2094 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
2096 return kvm_emulate_monitor_mwait(vcpu, "MWAIT");
2098 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
2100 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
2102 return kvm_emulate_monitor_mwait(vcpu, "MONITOR");
2104 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
2106 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
2108 xfer_to_guest_mode_prepare();
2109 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
2110 xfer_to_guest_mode_work_pending();
2114 * The fast path for frequent and performance sensitive wrmsr emulation,
2115 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
2116 * the latency of virtual IPI by avoiding the expensive bits of transitioning
2117 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
2118 * other cases which must be called after interrupts are enabled on the host.
2120 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
2122 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
2125 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
2126 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
2127 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
2128 ((u32)(data >> 32) != X2APIC_BROADCAST))
2129 return kvm_x2apic_icr_write(vcpu->arch.apic, data);
2134 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
2136 if (!kvm_can_use_hv_timer(vcpu))
2139 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2143 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
2145 u32 msr = kvm_rcx_read(vcpu);
2147 fastpath_t ret = EXIT_FASTPATH_NONE;
2150 case APIC_BASE_MSR + (APIC_ICR >> 4):
2151 data = kvm_read_edx_eax(vcpu);
2152 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
2153 kvm_skip_emulated_instruction(vcpu);
2154 ret = EXIT_FASTPATH_EXIT_HANDLED;
2157 case MSR_IA32_TSC_DEADLINE:
2158 data = kvm_read_edx_eax(vcpu);
2159 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
2160 kvm_skip_emulated_instruction(vcpu);
2161 ret = EXIT_FASTPATH_REENTER_GUEST;
2168 if (ret != EXIT_FASTPATH_NONE)
2169 trace_kvm_msr_write(msr, data);
2173 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
2176 * Adapt set_msr() to msr_io()'s calling convention
2178 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2180 return kvm_get_msr_ignored_check(vcpu, index, data, true);
2183 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
2185 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
2188 #ifdef CONFIG_X86_64
2189 struct pvclock_clock {
2199 struct pvclock_gtod_data {
2202 struct pvclock_clock clock; /* extract of a clocksource struct */
2203 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
2209 static struct pvclock_gtod_data pvclock_gtod_data;
2211 static void update_pvclock_gtod(struct timekeeper *tk)
2213 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
2215 write_seqcount_begin(&vdata->seq);
2217 /* copy pvclock gtod data */
2218 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
2219 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
2220 vdata->clock.mask = tk->tkr_mono.mask;
2221 vdata->clock.mult = tk->tkr_mono.mult;
2222 vdata->clock.shift = tk->tkr_mono.shift;
2223 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
2224 vdata->clock.offset = tk->tkr_mono.base;
2226 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
2227 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
2228 vdata->raw_clock.mask = tk->tkr_raw.mask;
2229 vdata->raw_clock.mult = tk->tkr_raw.mult;
2230 vdata->raw_clock.shift = tk->tkr_raw.shift;
2231 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
2232 vdata->raw_clock.offset = tk->tkr_raw.base;
2234 vdata->wall_time_sec = tk->xtime_sec;
2236 vdata->offs_boot = tk->offs_boot;
2238 write_seqcount_end(&vdata->seq);
2241 static s64 get_kvmclock_base_ns(void)
2243 /* Count up from boot time, but with the frequency of the raw clock. */
2244 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
2247 static s64 get_kvmclock_base_ns(void)
2249 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
2250 return ktime_get_boottime_ns();
2254 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
2258 struct pvclock_wall_clock wc;
2265 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2270 ++version; /* first time write, random junk */
2274 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2278 * The guest calculates current wall clock time by adding
2279 * system time (updated by kvm_guest_time_update below) to the
2280 * wall clock specified here. We do the reverse here.
2282 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2284 wc.nsec = do_div(wall_nsec, 1000000000);
2285 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2286 wc.version = version;
2288 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2291 wc_sec_hi = wall_nsec >> 32;
2292 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2293 &wc_sec_hi, sizeof(wc_sec_hi));
2297 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2300 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2301 bool old_msr, bool host_initiated)
2303 struct kvm_arch *ka = &vcpu->kvm->arch;
2305 if (vcpu->vcpu_id == 0 && !host_initiated) {
2306 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2307 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2309 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2312 vcpu->arch.time = system_time;
2313 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2315 /* we verify if the enable bit is set... */
2316 if (system_time & 1)
2317 kvm_gpc_activate(&vcpu->arch.pv_time, system_time & ~1ULL,
2318 sizeof(struct pvclock_vcpu_time_info));
2320 kvm_gpc_deactivate(&vcpu->arch.pv_time);
2325 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2327 do_shl32_div32(dividend, divisor);
2331 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2332 s8 *pshift, u32 *pmultiplier)
2340 scaled64 = scaled_hz;
2341 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2346 tps32 = (uint32_t)tps64;
2347 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2348 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2356 *pmultiplier = div_frac(scaled64, tps32);
2359 #ifdef CONFIG_X86_64
2360 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2363 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2364 static unsigned long max_tsc_khz;
2366 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2368 u64 v = (u64)khz * (1000000 + ppm);
2373 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier);
2375 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2379 /* Guest TSC same frequency as host TSC? */
2381 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_caps.default_tsc_scaling_ratio);
2385 /* TSC scaling supported? */
2386 if (!kvm_caps.has_tsc_control) {
2387 if (user_tsc_khz > tsc_khz) {
2388 vcpu->arch.tsc_catchup = 1;
2389 vcpu->arch.tsc_always_catchup = 1;
2392 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2397 /* TSC scaling required - calculate ratio */
2398 ratio = mul_u64_u32_div(1ULL << kvm_caps.tsc_scaling_ratio_frac_bits,
2399 user_tsc_khz, tsc_khz);
2401 if (ratio == 0 || ratio >= kvm_caps.max_tsc_scaling_ratio) {
2402 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2407 kvm_vcpu_write_tsc_multiplier(vcpu, ratio);
2411 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2413 u32 thresh_lo, thresh_hi;
2414 int use_scaling = 0;
2416 /* tsc_khz can be zero if TSC calibration fails */
2417 if (user_tsc_khz == 0) {
2418 /* set tsc_scaling_ratio to a safe value */
2419 kvm_vcpu_write_tsc_multiplier(vcpu, kvm_caps.default_tsc_scaling_ratio);
2423 /* Compute a scale to convert nanoseconds in TSC cycles */
2424 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2425 &vcpu->arch.virtual_tsc_shift,
2426 &vcpu->arch.virtual_tsc_mult);
2427 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2430 * Compute the variation in TSC rate which is acceptable
2431 * within the range of tolerance and decide if the
2432 * rate being applied is within that bounds of the hardware
2433 * rate. If so, no scaling or compensation need be done.
2435 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2436 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2437 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2438 pr_debug("requested TSC rate %u falls outside tolerance [%u,%u]\n",
2439 user_tsc_khz, thresh_lo, thresh_hi);
2442 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2445 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2447 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2448 vcpu->arch.virtual_tsc_mult,
2449 vcpu->arch.virtual_tsc_shift);
2450 tsc += vcpu->arch.this_tsc_write;
2454 #ifdef CONFIG_X86_64
2455 static inline int gtod_is_based_on_tsc(int mode)
2457 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2461 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2463 #ifdef CONFIG_X86_64
2465 struct kvm_arch *ka = &vcpu->kvm->arch;
2466 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2468 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2469 atomic_read(&vcpu->kvm->online_vcpus));
2472 * Once the masterclock is enabled, always perform request in
2473 * order to update it.
2475 * In order to enable masterclock, the host clocksource must be TSC
2476 * and the vcpus need to have matched TSCs. When that happens,
2477 * perform request to enable masterclock.
2479 if (ka->use_master_clock ||
2480 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2481 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2483 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2484 atomic_read(&vcpu->kvm->online_vcpus),
2485 ka->use_master_clock, gtod->clock.vclock_mode);
2490 * Multiply tsc by a fixed point number represented by ratio.
2492 * The most significant 64-N bits (mult) of ratio represent the
2493 * integral part of the fixed point number; the remaining N bits
2494 * (frac) represent the fractional part, ie. ratio represents a fixed
2495 * point number (mult + frac * 2^(-N)).
2497 * N equals to kvm_caps.tsc_scaling_ratio_frac_bits.
2499 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2501 return mul_u64_u64_shr(tsc, ratio, kvm_caps.tsc_scaling_ratio_frac_bits);
2504 u64 kvm_scale_tsc(u64 tsc, u64 ratio)
2508 if (ratio != kvm_caps.default_tsc_scaling_ratio)
2509 _tsc = __scale_tsc(ratio, tsc);
2514 static u64 kvm_compute_l1_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2518 tsc = kvm_scale_tsc(rdtsc(), vcpu->arch.l1_tsc_scaling_ratio);
2520 return target_tsc - tsc;
2523 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2525 return vcpu->arch.l1_tsc_offset +
2526 kvm_scale_tsc(host_tsc, vcpu->arch.l1_tsc_scaling_ratio);
2528 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2530 u64 kvm_calc_nested_tsc_offset(u64 l1_offset, u64 l2_offset, u64 l2_multiplier)
2534 if (l2_multiplier == kvm_caps.default_tsc_scaling_ratio)
2535 nested_offset = l1_offset;
2537 nested_offset = mul_s64_u64_shr((s64) l1_offset, l2_multiplier,
2538 kvm_caps.tsc_scaling_ratio_frac_bits);
2540 nested_offset += l2_offset;
2541 return nested_offset;
2543 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_offset);
2545 u64 kvm_calc_nested_tsc_multiplier(u64 l1_multiplier, u64 l2_multiplier)
2547 if (l2_multiplier != kvm_caps.default_tsc_scaling_ratio)
2548 return mul_u64_u64_shr(l1_multiplier, l2_multiplier,
2549 kvm_caps.tsc_scaling_ratio_frac_bits);
2551 return l1_multiplier;
2553 EXPORT_SYMBOL_GPL(kvm_calc_nested_tsc_multiplier);
2555 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 l1_offset)
2557 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2558 vcpu->arch.l1_tsc_offset,
2561 vcpu->arch.l1_tsc_offset = l1_offset;
2564 * If we are here because L1 chose not to trap WRMSR to TSC then
2565 * according to the spec this should set L1's TSC (as opposed to
2566 * setting L1's offset for L2).
2568 if (is_guest_mode(vcpu))
2569 vcpu->arch.tsc_offset = kvm_calc_nested_tsc_offset(
2571 static_call(kvm_x86_get_l2_tsc_offset)(vcpu),
2572 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2574 vcpu->arch.tsc_offset = l1_offset;
2576 static_call(kvm_x86_write_tsc_offset)(vcpu, vcpu->arch.tsc_offset);
2579 static void kvm_vcpu_write_tsc_multiplier(struct kvm_vcpu *vcpu, u64 l1_multiplier)
2581 vcpu->arch.l1_tsc_scaling_ratio = l1_multiplier;
2583 /* Userspace is changing the multiplier while L2 is active */
2584 if (is_guest_mode(vcpu))
2585 vcpu->arch.tsc_scaling_ratio = kvm_calc_nested_tsc_multiplier(
2587 static_call(kvm_x86_get_l2_tsc_multiplier)(vcpu));
2589 vcpu->arch.tsc_scaling_ratio = l1_multiplier;
2591 if (kvm_caps.has_tsc_control)
2592 static_call(kvm_x86_write_tsc_multiplier)(
2593 vcpu, vcpu->arch.tsc_scaling_ratio);
2596 static inline bool kvm_check_tsc_unstable(void)
2598 #ifdef CONFIG_X86_64
2600 * TSC is marked unstable when we're running on Hyper-V,
2601 * 'TSC page' clocksource is good.
2603 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2606 return check_tsc_unstable();
2610 * Infers attempts to synchronize the guest's tsc from host writes. Sets the
2611 * offset for the vcpu and tracks the TSC matching generation that the vcpu
2614 static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
2615 u64 ns, bool matched)
2617 struct kvm *kvm = vcpu->kvm;
2619 lockdep_assert_held(&kvm->arch.tsc_write_lock);
2622 * We also track th most recent recorded KHZ, write and time to
2623 * allow the matching interval to be extended at each write.
2625 kvm->arch.last_tsc_nsec = ns;
2626 kvm->arch.last_tsc_write = tsc;
2627 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2628 kvm->arch.last_tsc_offset = offset;
2630 vcpu->arch.last_guest_tsc = tsc;
2632 kvm_vcpu_write_tsc_offset(vcpu, offset);
2636 * We split periods of matched TSC writes into generations.
2637 * For each generation, we track the original measured
2638 * nanosecond time, offset, and write, so if TSCs are in
2639 * sync, we can match exact offset, and if not, we can match
2640 * exact software computation in compute_guest_tsc()
2642 * These values are tracked in kvm->arch.cur_xxx variables.
2644 kvm->arch.cur_tsc_generation++;
2645 kvm->arch.cur_tsc_nsec = ns;
2646 kvm->arch.cur_tsc_write = tsc;
2647 kvm->arch.cur_tsc_offset = offset;
2648 kvm->arch.nr_vcpus_matched_tsc = 0;
2649 } else if (vcpu->arch.this_tsc_generation != kvm->arch.cur_tsc_generation) {
2650 kvm->arch.nr_vcpus_matched_tsc++;
2653 /* Keep track of which generation this VCPU has synchronized to */
2654 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2655 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2656 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2658 kvm_track_tsc_matching(vcpu);
2661 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2663 struct kvm *kvm = vcpu->kvm;
2664 u64 offset, ns, elapsed;
2665 unsigned long flags;
2666 bool matched = false;
2667 bool synchronizing = false;
2669 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2670 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2671 ns = get_kvmclock_base_ns();
2672 elapsed = ns - kvm->arch.last_tsc_nsec;
2674 if (vcpu->arch.virtual_tsc_khz) {
2677 * detection of vcpu initialization -- need to sync
2678 * with other vCPUs. This particularly helps to keep
2679 * kvm_clock stable after CPU hotplug
2681 synchronizing = true;
2683 u64 tsc_exp = kvm->arch.last_tsc_write +
2684 nsec_to_cycles(vcpu, elapsed);
2685 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2687 * Special case: TSC write with a small delta (1 second)
2688 * of virtual cycle time against real time is
2689 * interpreted as an attempt to synchronize the CPU.
2691 synchronizing = data < tsc_exp + tsc_hz &&
2692 data + tsc_hz > tsc_exp;
2697 * For a reliable TSC, we can match TSC offsets, and for an unstable
2698 * TSC, we add elapsed time in this computation. We could let the
2699 * compensation code attempt to catch up if we fall behind, but
2700 * it's better to try to match offsets from the beginning.
2702 if (synchronizing &&
2703 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2704 if (!kvm_check_tsc_unstable()) {
2705 offset = kvm->arch.cur_tsc_offset;
2707 u64 delta = nsec_to_cycles(vcpu, elapsed);
2709 offset = kvm_compute_l1_tsc_offset(vcpu, data);
2714 __kvm_synchronize_tsc(vcpu, offset, data, ns, matched);
2715 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2718 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2721 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2722 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2725 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2727 if (vcpu->arch.l1_tsc_scaling_ratio != kvm_caps.default_tsc_scaling_ratio)
2728 WARN_ON(adjustment < 0);
2729 adjustment = kvm_scale_tsc((u64) adjustment,
2730 vcpu->arch.l1_tsc_scaling_ratio);
2731 adjust_tsc_offset_guest(vcpu, adjustment);
2734 #ifdef CONFIG_X86_64
2736 static u64 read_tsc(void)
2738 u64 ret = (u64)rdtsc_ordered();
2739 u64 last = pvclock_gtod_data.clock.cycle_last;
2741 if (likely(ret >= last))
2745 * GCC likes to generate cmov here, but this branch is extremely
2746 * predictable (it's just a function of time and the likely is
2747 * very likely) and there's a data dependence, so force GCC
2748 * to generate a branch instead. I don't barrier() because
2749 * we don't actually need a barrier, and if this function
2750 * ever gets inlined it will generate worse code.
2756 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2762 switch (clock->vclock_mode) {
2763 case VDSO_CLOCKMODE_HVCLOCK:
2764 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2766 if (tsc_pg_val != U64_MAX) {
2767 /* TSC page valid */
2768 *mode = VDSO_CLOCKMODE_HVCLOCK;
2769 v = (tsc_pg_val - clock->cycle_last) &
2772 /* TSC page invalid */
2773 *mode = VDSO_CLOCKMODE_NONE;
2776 case VDSO_CLOCKMODE_TSC:
2777 *mode = VDSO_CLOCKMODE_TSC;
2778 *tsc_timestamp = read_tsc();
2779 v = (*tsc_timestamp - clock->cycle_last) &
2783 *mode = VDSO_CLOCKMODE_NONE;
2786 if (*mode == VDSO_CLOCKMODE_NONE)
2787 *tsc_timestamp = v = 0;
2789 return v * clock->mult;
2792 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2794 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2800 seq = read_seqcount_begin(>od->seq);
2801 ns = gtod->raw_clock.base_cycles;
2802 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2803 ns >>= gtod->raw_clock.shift;
2804 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2805 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2811 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2813 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2819 seq = read_seqcount_begin(>od->seq);
2820 ts->tv_sec = gtod->wall_time_sec;
2821 ns = gtod->clock.base_cycles;
2822 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2823 ns >>= gtod->clock.shift;
2824 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2826 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2832 /* returns true if host is using TSC based clocksource */
2833 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2835 /* checked again under seqlock below */
2836 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2839 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2843 /* returns true if host is using TSC based clocksource */
2844 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2847 /* checked again under seqlock below */
2848 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2851 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2857 * Assuming a stable TSC across physical CPUS, and a stable TSC
2858 * across virtual CPUs, the following condition is possible.
2859 * Each numbered line represents an event visible to both
2860 * CPUs at the next numbered event.
2862 * "timespecX" represents host monotonic time. "tscX" represents
2865 * VCPU0 on CPU0 | VCPU1 on CPU1
2867 * 1. read timespec0,tsc0
2868 * 2. | timespec1 = timespec0 + N
2870 * 3. transition to guest | transition to guest
2871 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2872 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2873 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2875 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2878 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2880 * - 0 < N - M => M < N
2882 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2883 * always the case (the difference between two distinct xtime instances
2884 * might be smaller then the difference between corresponding TSC reads,
2885 * when updating guest vcpus pvclock areas).
2887 * To avoid that problem, do not allow visibility of distinct
2888 * system_timestamp/tsc_timestamp values simultaneously: use a master
2889 * copy of host monotonic time values. Update that master copy
2892 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2896 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2898 #ifdef CONFIG_X86_64
2899 struct kvm_arch *ka = &kvm->arch;
2901 bool host_tsc_clocksource, vcpus_matched;
2903 lockdep_assert_held(&kvm->arch.tsc_write_lock);
2904 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2905 atomic_read(&kvm->online_vcpus));
2908 * If the host uses TSC clock, then passthrough TSC as stable
2911 host_tsc_clocksource = kvm_get_time_and_clockread(
2912 &ka->master_kernel_ns,
2913 &ka->master_cycle_now);
2915 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2916 && !ka->backwards_tsc_observed
2917 && !ka->boot_vcpu_runs_old_kvmclock;
2919 if (ka->use_master_clock)
2920 atomic_set(&kvm_guest_has_master_clock, 1);
2922 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2923 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2928 static void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2930 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2933 static void __kvm_start_pvclock_update(struct kvm *kvm)
2935 raw_spin_lock_irq(&kvm->arch.tsc_write_lock);
2936 write_seqcount_begin(&kvm->arch.pvclock_sc);
2939 static void kvm_start_pvclock_update(struct kvm *kvm)
2941 kvm_make_mclock_inprogress_request(kvm);
2943 /* no guest entries from this point */
2944 __kvm_start_pvclock_update(kvm);
2947 static void kvm_end_pvclock_update(struct kvm *kvm)
2949 struct kvm_arch *ka = &kvm->arch;
2950 struct kvm_vcpu *vcpu;
2953 write_seqcount_end(&ka->pvclock_sc);
2954 raw_spin_unlock_irq(&ka->tsc_write_lock);
2955 kvm_for_each_vcpu(i, vcpu, kvm)
2956 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2958 /* guest entries allowed */
2959 kvm_for_each_vcpu(i, vcpu, kvm)
2960 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2963 static void kvm_update_masterclock(struct kvm *kvm)
2965 kvm_hv_request_tsc_page_update(kvm);
2966 kvm_start_pvclock_update(kvm);
2967 pvclock_update_vm_gtod_copy(kvm);
2968 kvm_end_pvclock_update(kvm);
2972 * Use the kernel's tsc_khz directly if the TSC is constant, otherwise use KVM's
2973 * per-CPU value (which may be zero if a CPU is going offline). Note, tsc_khz
2974 * can change during boot even if the TSC is constant, as it's possible for KVM
2975 * to be loaded before TSC calibration completes. Ideally, KVM would get a
2976 * notification when calibration completes, but practically speaking calibration
2977 * will complete before userspace is alive enough to create VMs.
2979 static unsigned long get_cpu_tsc_khz(void)
2981 if (static_cpu_has(X86_FEATURE_CONSTANT_TSC))
2984 return __this_cpu_read(cpu_tsc_khz);
2987 /* Called within read_seqcount_begin/retry for kvm->pvclock_sc. */
2988 static void __get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
2990 struct kvm_arch *ka = &kvm->arch;
2991 struct pvclock_vcpu_time_info hv_clock;
2993 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2997 if (ka->use_master_clock &&
2998 (static_cpu_has(X86_FEATURE_CONSTANT_TSC) || __this_cpu_read(cpu_tsc_khz))) {
2999 #ifdef CONFIG_X86_64
3000 struct timespec64 ts;
3002 if (kvm_get_walltime_and_clockread(&ts, &data->host_tsc)) {
3003 data->realtime = ts.tv_nsec + NSEC_PER_SEC * ts.tv_sec;
3004 data->flags |= KVM_CLOCK_REALTIME | KVM_CLOCK_HOST_TSC;
3007 data->host_tsc = rdtsc();
3009 data->flags |= KVM_CLOCK_TSC_STABLE;
3010 hv_clock.tsc_timestamp = ka->master_cycle_now;
3011 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
3012 kvm_get_time_scale(NSEC_PER_SEC, get_cpu_tsc_khz() * 1000LL,
3013 &hv_clock.tsc_shift,
3014 &hv_clock.tsc_to_system_mul);
3015 data->clock = __pvclock_read_cycles(&hv_clock, data->host_tsc);
3017 data->clock = get_kvmclock_base_ns() + ka->kvmclock_offset;
3023 static void get_kvmclock(struct kvm *kvm, struct kvm_clock_data *data)
3025 struct kvm_arch *ka = &kvm->arch;
3029 seq = read_seqcount_begin(&ka->pvclock_sc);
3030 __get_kvmclock(kvm, data);
3031 } while (read_seqcount_retry(&ka->pvclock_sc, seq));
3034 u64 get_kvmclock_ns(struct kvm *kvm)
3036 struct kvm_clock_data data;
3038 get_kvmclock(kvm, &data);
3042 static void kvm_setup_guest_pvclock(struct kvm_vcpu *v,
3043 struct gfn_to_pfn_cache *gpc,
3044 unsigned int offset)
3046 struct kvm_vcpu_arch *vcpu = &v->arch;
3047 struct pvclock_vcpu_time_info *guest_hv_clock;
3048 unsigned long flags;
3050 read_lock_irqsave(&gpc->lock, flags);
3051 while (!kvm_gpc_check(gpc, offset + sizeof(*guest_hv_clock))) {
3052 read_unlock_irqrestore(&gpc->lock, flags);
3054 if (kvm_gpc_refresh(gpc, offset + sizeof(*guest_hv_clock)))
3057 read_lock_irqsave(&gpc->lock, flags);
3060 guest_hv_clock = (void *)(gpc->khva + offset);
3063 * This VCPU is paused, but it's legal for a guest to read another
3064 * VCPU's kvmclock, so we really have to follow the specification where
3065 * it says that version is odd if data is being modified, and even after
3069 guest_hv_clock->version = vcpu->hv_clock.version = (guest_hv_clock->version + 1) | 1;
3072 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
3073 vcpu->hv_clock.flags |= (guest_hv_clock->flags & PVCLOCK_GUEST_STOPPED);
3075 if (vcpu->pvclock_set_guest_stopped_request) {
3076 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
3077 vcpu->pvclock_set_guest_stopped_request = false;
3080 memcpy(guest_hv_clock, &vcpu->hv_clock, sizeof(*guest_hv_clock));
3083 guest_hv_clock->version = ++vcpu->hv_clock.version;
3085 mark_page_dirty_in_slot(v->kvm, gpc->memslot, gpc->gpa >> PAGE_SHIFT);
3086 read_unlock_irqrestore(&gpc->lock, flags);
3088 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
3091 static int kvm_guest_time_update(struct kvm_vcpu *v)
3093 unsigned long flags, tgt_tsc_khz;
3095 struct kvm_vcpu_arch *vcpu = &v->arch;
3096 struct kvm_arch *ka = &v->kvm->arch;
3098 u64 tsc_timestamp, host_tsc;
3100 bool use_master_clock;
3106 * If the host uses TSC clock, then passthrough TSC as stable
3110 seq = read_seqcount_begin(&ka->pvclock_sc);
3111 use_master_clock = ka->use_master_clock;
3112 if (use_master_clock) {
3113 host_tsc = ka->master_cycle_now;
3114 kernel_ns = ka->master_kernel_ns;
3116 } while (read_seqcount_retry(&ka->pvclock_sc, seq));
3118 /* Keep irq disabled to prevent changes to the clock */
3119 local_irq_save(flags);
3120 tgt_tsc_khz = get_cpu_tsc_khz();
3121 if (unlikely(tgt_tsc_khz == 0)) {
3122 local_irq_restore(flags);
3123 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3126 if (!use_master_clock) {
3128 kernel_ns = get_kvmclock_base_ns();
3131 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
3134 * We may have to catch up the TSC to match elapsed wall clock
3135 * time for two reasons, even if kvmclock is used.
3136 * 1) CPU could have been running below the maximum TSC rate
3137 * 2) Broken TSC compensation resets the base at each VCPU
3138 * entry to avoid unknown leaps of TSC even when running
3139 * again on the same CPU. This may cause apparent elapsed
3140 * time to disappear, and the guest to stand still or run
3143 if (vcpu->tsc_catchup) {
3144 u64 tsc = compute_guest_tsc(v, kernel_ns);
3145 if (tsc > tsc_timestamp) {
3146 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
3147 tsc_timestamp = tsc;
3151 local_irq_restore(flags);
3153 /* With all the info we got, fill in the values */
3155 if (kvm_caps.has_tsc_control)
3156 tgt_tsc_khz = kvm_scale_tsc(tgt_tsc_khz,
3157 v->arch.l1_tsc_scaling_ratio);
3159 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
3160 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
3161 &vcpu->hv_clock.tsc_shift,
3162 &vcpu->hv_clock.tsc_to_system_mul);
3163 vcpu->hw_tsc_khz = tgt_tsc_khz;
3166 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
3167 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
3168 vcpu->last_guest_tsc = tsc_timestamp;
3170 /* If the host uses TSC clocksource, then it is stable */
3172 if (use_master_clock)
3173 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
3175 vcpu->hv_clock.flags = pvclock_flags;
3177 if (vcpu->pv_time.active)
3178 kvm_setup_guest_pvclock(v, &vcpu->pv_time, 0);
3179 if (vcpu->xen.vcpu_info_cache.active)
3180 kvm_setup_guest_pvclock(v, &vcpu->xen.vcpu_info_cache,
3181 offsetof(struct compat_vcpu_info, time));
3182 if (vcpu->xen.vcpu_time_info_cache.active)
3183 kvm_setup_guest_pvclock(v, &vcpu->xen.vcpu_time_info_cache, 0);
3184 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
3189 * kvmclock updates which are isolated to a given vcpu, such as
3190 * vcpu->cpu migration, should not allow system_timestamp from
3191 * the rest of the vcpus to remain static. Otherwise ntp frequency
3192 * correction applies to one vcpu's system_timestamp but not
3195 * So in those cases, request a kvmclock update for all vcpus.
3196 * We need to rate-limit these requests though, as they can
3197 * considerably slow guests that have a large number of vcpus.
3198 * The time for a remote vcpu to update its kvmclock is bound
3199 * by the delay we use to rate-limit the updates.
3202 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
3204 static void kvmclock_update_fn(struct work_struct *work)
3207 struct delayed_work *dwork = to_delayed_work(work);
3208 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3209 kvmclock_update_work);
3210 struct kvm *kvm = container_of(ka, struct kvm, arch);
3211 struct kvm_vcpu *vcpu;
3213 kvm_for_each_vcpu(i, vcpu, kvm) {
3214 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3215 kvm_vcpu_kick(vcpu);
3219 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
3221 struct kvm *kvm = v->kvm;
3223 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
3224 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
3225 KVMCLOCK_UPDATE_DELAY);
3228 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
3230 static void kvmclock_sync_fn(struct work_struct *work)
3232 struct delayed_work *dwork = to_delayed_work(work);
3233 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
3234 kvmclock_sync_work);
3235 struct kvm *kvm = container_of(ka, struct kvm, arch);
3237 if (!kvmclock_periodic_sync)
3240 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
3241 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
3242 KVMCLOCK_SYNC_PERIOD);
3245 /* These helpers are safe iff @msr is known to be an MCx bank MSR. */
3246 static bool is_mci_control_msr(u32 msr)
3248 return (msr & 3) == 0;
3250 static bool is_mci_status_msr(u32 msr)
3252 return (msr & 3) == 1;
3256 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
3258 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
3260 /* McStatusWrEn enabled? */
3261 if (guest_cpuid_is_amd_or_hygon(vcpu))
3262 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
3267 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3269 u64 mcg_cap = vcpu->arch.mcg_cap;
3270 unsigned bank_num = mcg_cap & 0xff;
3271 u32 msr = msr_info->index;
3272 u64 data = msr_info->data;
3273 u32 offset, last_msr;
3276 case MSR_IA32_MCG_STATUS:
3277 vcpu->arch.mcg_status = data;
3279 case MSR_IA32_MCG_CTL:
3280 if (!(mcg_cap & MCG_CTL_P) &&
3281 (data || !msr_info->host_initiated))
3283 if (data != 0 && data != ~(u64)0)
3285 vcpu->arch.mcg_ctl = data;
3287 case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
3288 last_msr = MSR_IA32_MCx_CTL2(bank_num) - 1;
3292 if (!(mcg_cap & MCG_CMCI_P) && (data || !msr_info->host_initiated))
3294 /* An attempt to write a 1 to a reserved bit raises #GP */
3295 if (data & ~(MCI_CTL2_CMCI_EN | MCI_CTL2_CMCI_THRESHOLD_MASK))
3297 offset = array_index_nospec(msr - MSR_IA32_MC0_CTL2,
3298 last_msr + 1 - MSR_IA32_MC0_CTL2);
3299 vcpu->arch.mci_ctl2_banks[offset] = data;
3301 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3302 last_msr = MSR_IA32_MCx_CTL(bank_num) - 1;
3307 * Only 0 or all 1s can be written to IA32_MCi_CTL, all other
3308 * values are architecturally undefined. But, some Linux
3309 * kernels clear bit 10 in bank 4 to workaround a BIOS/GART TLB
3310 * issue on AMD K8s, allow bit 10 to be clear when setting all
3311 * other bits in order to avoid an uncaught #GP in the guest.
3313 * UNIXWARE clears bit 0 of MC1_CTL to ignore correctable,
3314 * single-bit ECC data errors.
3316 if (is_mci_control_msr(msr) &&
3317 data != 0 && (data | (1 << 10) | 1) != ~(u64)0)
3321 * All CPUs allow writing 0 to MCi_STATUS MSRs to clear the MSR.
3322 * AMD-based CPUs allow non-zero values, but if and only if
3323 * HWCR[McStatusWrEn] is set.
3325 if (!msr_info->host_initiated && is_mci_status_msr(msr) &&
3326 data != 0 && !can_set_mci_status(vcpu))
3329 offset = array_index_nospec(msr - MSR_IA32_MC0_CTL,
3330 last_msr + 1 - MSR_IA32_MC0_CTL);
3331 vcpu->arch.mce_banks[offset] = data;
3339 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
3341 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
3343 return (vcpu->arch.apf.msr_en_val & mask) == mask;
3346 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
3348 gpa_t gpa = data & ~0x3f;
3350 /* Bits 4:5 are reserved, Should be zero */
3354 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
3355 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
3358 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
3359 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
3362 if (!lapic_in_kernel(vcpu))
3363 return data ? 1 : 0;
3365 vcpu->arch.apf.msr_en_val = data;
3367 if (!kvm_pv_async_pf_enabled(vcpu)) {
3368 kvm_clear_async_pf_completion_queue(vcpu);
3369 kvm_async_pf_hash_reset(vcpu);
3373 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
3377 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
3378 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
3380 kvm_async_pf_wakeup_all(vcpu);
3385 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
3387 /* Bits 8-63 are reserved */
3391 if (!lapic_in_kernel(vcpu))
3394 vcpu->arch.apf.msr_int_val = data;
3396 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
3401 static void kvmclock_reset(struct kvm_vcpu *vcpu)
3403 kvm_gpc_deactivate(&vcpu->arch.pv_time);
3404 vcpu->arch.time = 0;
3407 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
3409 ++vcpu->stat.tlb_flush;
3410 static_call(kvm_x86_flush_tlb_all)(vcpu);
3412 /* Flushing all ASIDs flushes the current ASID... */
3413 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
3416 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3418 ++vcpu->stat.tlb_flush;
3422 * A TLB flush on behalf of the guest is equivalent to
3423 * INVPCID(all), toggling CR4.PGE, etc., which requires
3424 * a forced sync of the shadow page tables. Ensure all the
3425 * roots are synced and the guest TLB in hardware is clean.
3427 kvm_mmu_sync_roots(vcpu);
3428 kvm_mmu_sync_prev_roots(vcpu);
3431 static_call(kvm_x86_flush_tlb_guest)(vcpu);
3434 * Flushing all "guest" TLB is always a superset of Hyper-V's fine
3437 kvm_hv_vcpu_purge_flush_tlb(vcpu);
3441 static inline void kvm_vcpu_flush_tlb_current(struct kvm_vcpu *vcpu)
3443 ++vcpu->stat.tlb_flush;
3444 static_call(kvm_x86_flush_tlb_current)(vcpu);
3448 * Service "local" TLB flush requests, which are specific to the current MMU
3449 * context. In addition to the generic event handling in vcpu_enter_guest(),
3450 * TLB flushes that are targeted at an MMU context also need to be serviced
3451 * prior before nested VM-Enter/VM-Exit.
3453 void kvm_service_local_tlb_flush_requests(struct kvm_vcpu *vcpu)
3455 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
3456 kvm_vcpu_flush_tlb_current(vcpu);
3458 if (kvm_check_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu))
3459 kvm_vcpu_flush_tlb_guest(vcpu);
3461 EXPORT_SYMBOL_GPL(kvm_service_local_tlb_flush_requests);
3463 static void record_steal_time(struct kvm_vcpu *vcpu)
3465 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
3466 struct kvm_steal_time __user *st;
3467 struct kvm_memslots *slots;
3468 gpa_t gpa = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
3472 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3473 kvm_xen_runstate_set_running(vcpu);
3477 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3480 if (WARN_ON_ONCE(current->mm != vcpu->kvm->mm))
3483 slots = kvm_memslots(vcpu->kvm);
3485 if (unlikely(slots->generation != ghc->generation ||
3487 kvm_is_error_hva(ghc->hva) || !ghc->memslot)) {
3488 /* We rely on the fact that it fits in a single page. */
3489 BUILD_BUG_ON((sizeof(*st) - 1) & KVM_STEAL_VALID_BITS);
3491 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, ghc, gpa, sizeof(*st)) ||
3492 kvm_is_error_hva(ghc->hva) || !ghc->memslot)
3496 st = (struct kvm_steal_time __user *)ghc->hva;
3498 * Doing a TLB flush here, on the guest's behalf, can avoid
3501 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3502 u8 st_preempted = 0;
3505 if (!user_access_begin(st, sizeof(*st)))
3508 asm volatile("1: xchgb %0, %2\n"
3511 _ASM_EXTABLE_UA(1b, 2b)
3512 : "+q" (st_preempted),
3514 "+m" (st->preempted));
3520 vcpu->arch.st.preempted = 0;
3522 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3523 st_preempted & KVM_VCPU_FLUSH_TLB);
3524 if (st_preempted & KVM_VCPU_FLUSH_TLB)
3525 kvm_vcpu_flush_tlb_guest(vcpu);
3527 if (!user_access_begin(st, sizeof(*st)))
3530 if (!user_access_begin(st, sizeof(*st)))
3533 unsafe_put_user(0, &st->preempted, out);
3534 vcpu->arch.st.preempted = 0;
3537 unsafe_get_user(version, &st->version, out);
3539 version += 1; /* first time write, random junk */
3542 unsafe_put_user(version, &st->version, out);
3546 unsafe_get_user(steal, &st->steal, out);
3547 steal += current->sched_info.run_delay -
3548 vcpu->arch.st.last_steal;
3549 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3550 unsafe_put_user(steal, &st->steal, out);
3553 unsafe_put_user(version, &st->version, out);
3558 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
3561 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3564 u32 msr = msr_info->index;
3565 u64 data = msr_info->data;
3567 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3568 return kvm_xen_write_hypercall_page(vcpu, data);
3571 case MSR_AMD64_NB_CFG:
3572 case MSR_IA32_UCODE_WRITE:
3573 case MSR_VM_HSAVE_PA:
3574 case MSR_AMD64_PATCH_LOADER:
3575 case MSR_AMD64_BU_CFG2:
3576 case MSR_AMD64_DC_CFG:
3577 case MSR_F15H_EX_CFG:
3580 case MSR_IA32_UCODE_REV:
3581 if (msr_info->host_initiated)
3582 vcpu->arch.microcode_version = data;
3584 case MSR_IA32_ARCH_CAPABILITIES:
3585 if (!msr_info->host_initiated)
3587 vcpu->arch.arch_capabilities = data;
3589 case MSR_IA32_PERF_CAPABILITIES:
3590 if (!msr_info->host_initiated)
3592 if (data & ~kvm_caps.supported_perf_cap)
3595 vcpu->arch.perf_capabilities = data;
3596 kvm_pmu_refresh(vcpu);
3599 return set_efer(vcpu, msr_info);
3601 data &= ~(u64)0x40; /* ignore flush filter disable */
3602 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3603 data &= ~(u64)0x8; /* ignore TLB cache disable */
3605 /* Handle McStatusWrEn */
3606 if (data == BIT_ULL(18)) {
3607 vcpu->arch.msr_hwcr = data;
3608 } else if (data != 0) {
3609 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3614 case MSR_FAM10H_MMIO_CONF_BASE:
3616 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3621 case 0x200 ... MSR_IA32_MC0_CTL2 - 1:
3622 case MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) ... 0x2ff:
3623 return kvm_mtrr_set_msr(vcpu, msr, data);
3624 case MSR_IA32_APICBASE:
3625 return kvm_set_apic_base(vcpu, msr_info);
3626 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3627 return kvm_x2apic_msr_write(vcpu, msr, data);
3628 case MSR_IA32_TSC_DEADLINE:
3629 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3631 case MSR_IA32_TSC_ADJUST:
3632 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3633 if (!msr_info->host_initiated) {
3634 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3635 adjust_tsc_offset_guest(vcpu, adj);
3636 /* Before back to guest, tsc_timestamp must be adjusted
3637 * as well, otherwise guest's percpu pvclock time could jump.
3639 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3641 vcpu->arch.ia32_tsc_adjust_msr = data;
3644 case MSR_IA32_MISC_ENABLE: {
3645 u64 old_val = vcpu->arch.ia32_misc_enable_msr;
3647 if (!msr_info->host_initiated) {
3649 if ((old_val ^ data) & MSR_IA32_MISC_ENABLE_PMU_RO_MASK)
3652 /* R bits, i.e. writes are ignored, but don't fault. */
3653 data = data & ~MSR_IA32_MISC_ENABLE_EMON;
3654 data |= old_val & MSR_IA32_MISC_ENABLE_EMON;
3657 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3658 ((old_val ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3659 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3661 vcpu->arch.ia32_misc_enable_msr = data;
3662 kvm_update_cpuid_runtime(vcpu);
3664 vcpu->arch.ia32_misc_enable_msr = data;
3668 case MSR_IA32_SMBASE:
3669 if (!IS_ENABLED(CONFIG_KVM_SMM) || !msr_info->host_initiated)
3671 vcpu->arch.smbase = data;
3673 case MSR_IA32_POWER_CTL:
3674 vcpu->arch.msr_ia32_power_ctl = data;
3677 if (msr_info->host_initiated) {
3678 kvm_synchronize_tsc(vcpu, data);
3680 u64 adj = kvm_compute_l1_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3681 adjust_tsc_offset_guest(vcpu, adj);
3682 vcpu->arch.ia32_tsc_adjust_msr += adj;
3686 if (!msr_info->host_initiated &&
3687 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3690 * KVM supports exposing PT to the guest, but does not support
3691 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3692 * XSAVES/XRSTORS to save/restore PT MSRs.
3694 if (data & ~kvm_caps.supported_xss)
3696 vcpu->arch.ia32_xss = data;
3697 kvm_update_cpuid_runtime(vcpu);
3700 if (!msr_info->host_initiated)
3702 vcpu->arch.smi_count = data;
3704 case MSR_KVM_WALL_CLOCK_NEW:
3705 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3708 vcpu->kvm->arch.wall_clock = data;
3709 kvm_write_wall_clock(vcpu->kvm, data, 0);
3711 case MSR_KVM_WALL_CLOCK:
3712 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3715 vcpu->kvm->arch.wall_clock = data;
3716 kvm_write_wall_clock(vcpu->kvm, data, 0);
3718 case MSR_KVM_SYSTEM_TIME_NEW:
3719 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3722 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3724 case MSR_KVM_SYSTEM_TIME:
3725 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3728 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3730 case MSR_KVM_ASYNC_PF_EN:
3731 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3734 if (kvm_pv_enable_async_pf(vcpu, data))
3737 case MSR_KVM_ASYNC_PF_INT:
3738 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3741 if (kvm_pv_enable_async_pf_int(vcpu, data))
3744 case MSR_KVM_ASYNC_PF_ACK:
3745 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3748 vcpu->arch.apf.pageready_pending = false;
3749 kvm_check_async_pf_completion(vcpu);
3752 case MSR_KVM_STEAL_TIME:
3753 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3756 if (unlikely(!sched_info_on()))
3759 if (data & KVM_STEAL_RESERVED_MASK)
3762 vcpu->arch.st.msr_val = data;
3764 if (!(data & KVM_MSR_ENABLED))
3767 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3770 case MSR_KVM_PV_EOI_EN:
3771 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3774 if (kvm_lapic_set_pv_eoi(vcpu, data, sizeof(u8)))
3778 case MSR_KVM_POLL_CONTROL:
3779 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3782 /* only enable bit supported */
3783 if (data & (-1ULL << 1))
3786 vcpu->arch.msr_kvm_poll_control = data;
3789 case MSR_IA32_MCG_CTL:
3790 case MSR_IA32_MCG_STATUS:
3791 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3792 case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
3793 return set_msr_mce(vcpu, msr_info);
3795 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3796 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3799 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3800 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3801 if (kvm_pmu_is_valid_msr(vcpu, msr))
3802 return kvm_pmu_set_msr(vcpu, msr_info);
3804 if (pr || data != 0)
3805 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3806 "0x%x data 0x%llx\n", msr, data);
3808 case MSR_K7_CLK_CTL:
3810 * Ignore all writes to this no longer documented MSR.
3811 * Writes are only relevant for old K7 processors,
3812 * all pre-dating SVM, but a recommended workaround from
3813 * AMD for these chips. It is possible to specify the
3814 * affected processor models on the command line, hence
3815 * the need to ignore the workaround.
3818 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3819 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3820 case HV_X64_MSR_SYNDBG_OPTIONS:
3821 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3822 case HV_X64_MSR_CRASH_CTL:
3823 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3824 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3825 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3826 case HV_X64_MSR_TSC_EMULATION_STATUS:
3827 case HV_X64_MSR_TSC_INVARIANT_CONTROL:
3828 return kvm_hv_set_msr_common(vcpu, msr, data,
3829 msr_info->host_initiated);
3830 case MSR_IA32_BBL_CR_CTL3:
3831 /* Drop writes to this legacy MSR -- see rdmsr
3832 * counterpart for further detail.
3834 if (report_ignored_msrs)
3835 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3838 case MSR_AMD64_OSVW_ID_LENGTH:
3839 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3841 vcpu->arch.osvw.length = data;
3843 case MSR_AMD64_OSVW_STATUS:
3844 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3846 vcpu->arch.osvw.status = data;
3848 case MSR_PLATFORM_INFO:
3849 if (!msr_info->host_initiated ||
3850 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3851 cpuid_fault_enabled(vcpu)))
3853 vcpu->arch.msr_platform_info = data;
3855 case MSR_MISC_FEATURES_ENABLES:
3856 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3857 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3858 !supports_cpuid_fault(vcpu)))
3860 vcpu->arch.msr_misc_features_enables = data;
3862 #ifdef CONFIG_X86_64
3864 if (!msr_info->host_initiated &&
3865 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3868 if (data & ~kvm_guest_supported_xfd(vcpu))
3871 fpu_update_guest_xfd(&vcpu->arch.guest_fpu, data);
3873 case MSR_IA32_XFD_ERR:
3874 if (!msr_info->host_initiated &&
3875 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
3878 if (data & ~kvm_guest_supported_xfd(vcpu))
3881 vcpu->arch.guest_fpu.xfd_err = data;
3884 case MSR_IA32_PEBS_ENABLE:
3885 case MSR_IA32_DS_AREA:
3886 case MSR_PEBS_DATA_CFG:
3887 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3888 if (kvm_pmu_is_valid_msr(vcpu, msr))
3889 return kvm_pmu_set_msr(vcpu, msr_info);
3891 * Userspace is allowed to write '0' to MSRs that KVM reports
3892 * as to-be-saved, even if an MSRs isn't fully supported.
3894 return !msr_info->host_initiated || data;
3896 if (kvm_pmu_is_valid_msr(vcpu, msr))
3897 return kvm_pmu_set_msr(vcpu, msr_info);
3898 return KVM_MSR_RET_INVALID;
3902 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3904 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3907 u64 mcg_cap = vcpu->arch.mcg_cap;
3908 unsigned bank_num = mcg_cap & 0xff;
3909 u32 offset, last_msr;
3912 case MSR_IA32_P5_MC_ADDR:
3913 case MSR_IA32_P5_MC_TYPE:
3916 case MSR_IA32_MCG_CAP:
3917 data = vcpu->arch.mcg_cap;
3919 case MSR_IA32_MCG_CTL:
3920 if (!(mcg_cap & MCG_CTL_P) && !host)
3922 data = vcpu->arch.mcg_ctl;
3924 case MSR_IA32_MCG_STATUS:
3925 data = vcpu->arch.mcg_status;
3927 case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
3928 last_msr = MSR_IA32_MCx_CTL2(bank_num) - 1;
3932 if (!(mcg_cap & MCG_CMCI_P) && !host)
3934 offset = array_index_nospec(msr - MSR_IA32_MC0_CTL2,
3935 last_msr + 1 - MSR_IA32_MC0_CTL2);
3936 data = vcpu->arch.mci_ctl2_banks[offset];
3938 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3939 last_msr = MSR_IA32_MCx_CTL(bank_num) - 1;
3943 offset = array_index_nospec(msr - MSR_IA32_MC0_CTL,
3944 last_msr + 1 - MSR_IA32_MC0_CTL);
3945 data = vcpu->arch.mce_banks[offset];
3954 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3956 switch (msr_info->index) {
3957 case MSR_IA32_PLATFORM_ID:
3958 case MSR_IA32_EBL_CR_POWERON:
3959 case MSR_IA32_LASTBRANCHFROMIP:
3960 case MSR_IA32_LASTBRANCHTOIP:
3961 case MSR_IA32_LASTINTFROMIP:
3962 case MSR_IA32_LASTINTTOIP:
3963 case MSR_AMD64_SYSCFG:
3964 case MSR_K8_TSEG_ADDR:
3965 case MSR_K8_TSEG_MASK:
3966 case MSR_VM_HSAVE_PA:
3967 case MSR_K8_INT_PENDING_MSG:
3968 case MSR_AMD64_NB_CFG:
3969 case MSR_FAM10H_MMIO_CONF_BASE:
3970 case MSR_AMD64_BU_CFG2:
3971 case MSR_IA32_PERF_CTL:
3972 case MSR_AMD64_DC_CFG:
3973 case MSR_F15H_EX_CFG:
3975 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3976 * limit) MSRs. Just return 0, as we do not want to expose the host
3977 * data here. Do not conditionalize this on CPUID, as KVM does not do
3978 * so for existing CPU-specific MSRs.
3980 case MSR_RAPL_POWER_UNIT:
3981 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3982 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3983 case MSR_PKG_ENERGY_STATUS: /* Total package */
3984 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3987 case MSR_IA32_PEBS_ENABLE:
3988 case MSR_IA32_DS_AREA:
3989 case MSR_PEBS_DATA_CFG:
3990 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3991 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3992 return kvm_pmu_get_msr(vcpu, msr_info);
3994 * Userspace is allowed to read MSRs that KVM reports as
3995 * to-be-saved, even if an MSR isn't fully supported.
3997 if (!msr_info->host_initiated)
4001 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
4002 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
4003 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
4004 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
4005 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
4006 return kvm_pmu_get_msr(vcpu, msr_info);
4009 case MSR_IA32_UCODE_REV:
4010 msr_info->data = vcpu->arch.microcode_version;
4012 case MSR_IA32_ARCH_CAPABILITIES:
4013 if (!msr_info->host_initiated &&
4014 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
4016 msr_info->data = vcpu->arch.arch_capabilities;
4018 case MSR_IA32_PERF_CAPABILITIES:
4019 if (!msr_info->host_initiated &&
4020 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
4022 msr_info->data = vcpu->arch.perf_capabilities;
4024 case MSR_IA32_POWER_CTL:
4025 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
4027 case MSR_IA32_TSC: {
4029 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
4030 * even when not intercepted. AMD manual doesn't explicitly
4031 * state this but appears to behave the same.
4033 * On userspace reads and writes, however, we unconditionally
4034 * return L1's TSC value to ensure backwards-compatible
4035 * behavior for migration.
4039 if (msr_info->host_initiated) {
4040 offset = vcpu->arch.l1_tsc_offset;
4041 ratio = vcpu->arch.l1_tsc_scaling_ratio;
4043 offset = vcpu->arch.tsc_offset;
4044 ratio = vcpu->arch.tsc_scaling_ratio;
4047 msr_info->data = kvm_scale_tsc(rdtsc(), ratio) + offset;
4051 case 0x200 ... MSR_IA32_MC0_CTL2 - 1:
4052 case MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) ... 0x2ff:
4053 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
4054 case 0xcd: /* fsb frequency */
4058 * MSR_EBC_FREQUENCY_ID
4059 * Conservative value valid for even the basic CPU models.
4060 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
4061 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
4062 * and 266MHz for model 3, or 4. Set Core Clock
4063 * Frequency to System Bus Frequency Ratio to 1 (bits
4064 * 31:24) even though these are only valid for CPU
4065 * models > 2, however guests may end up dividing or
4066 * multiplying by zero otherwise.
4068 case MSR_EBC_FREQUENCY_ID:
4069 msr_info->data = 1 << 24;
4071 case MSR_IA32_APICBASE:
4072 msr_info->data = kvm_get_apic_base(vcpu);
4074 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
4075 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
4076 case MSR_IA32_TSC_DEADLINE:
4077 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
4079 case MSR_IA32_TSC_ADJUST:
4080 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
4082 case MSR_IA32_MISC_ENABLE:
4083 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
4085 case MSR_IA32_SMBASE:
4086 if (!IS_ENABLED(CONFIG_KVM_SMM) || !msr_info->host_initiated)
4088 msr_info->data = vcpu->arch.smbase;
4091 msr_info->data = vcpu->arch.smi_count;
4093 case MSR_IA32_PERF_STATUS:
4094 /* TSC increment by tick */
4095 msr_info->data = 1000ULL;
4096 /* CPU multiplier */
4097 msr_info->data |= (((uint64_t)4ULL) << 40);
4100 msr_info->data = vcpu->arch.efer;
4102 case MSR_KVM_WALL_CLOCK:
4103 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
4106 msr_info->data = vcpu->kvm->arch.wall_clock;
4108 case MSR_KVM_WALL_CLOCK_NEW:
4109 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
4112 msr_info->data = vcpu->kvm->arch.wall_clock;
4114 case MSR_KVM_SYSTEM_TIME:
4115 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
4118 msr_info->data = vcpu->arch.time;
4120 case MSR_KVM_SYSTEM_TIME_NEW:
4121 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
4124 msr_info->data = vcpu->arch.time;
4126 case MSR_KVM_ASYNC_PF_EN:
4127 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
4130 msr_info->data = vcpu->arch.apf.msr_en_val;
4132 case MSR_KVM_ASYNC_PF_INT:
4133 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
4136 msr_info->data = vcpu->arch.apf.msr_int_val;
4138 case MSR_KVM_ASYNC_PF_ACK:
4139 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
4144 case MSR_KVM_STEAL_TIME:
4145 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
4148 msr_info->data = vcpu->arch.st.msr_val;
4150 case MSR_KVM_PV_EOI_EN:
4151 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
4154 msr_info->data = vcpu->arch.pv_eoi.msr_val;
4156 case MSR_KVM_POLL_CONTROL:
4157 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
4160 msr_info->data = vcpu->arch.msr_kvm_poll_control;
4162 case MSR_IA32_P5_MC_ADDR:
4163 case MSR_IA32_P5_MC_TYPE:
4164 case MSR_IA32_MCG_CAP:
4165 case MSR_IA32_MCG_CTL:
4166 case MSR_IA32_MCG_STATUS:
4167 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
4168 case MSR_IA32_MC0_CTL2 ... MSR_IA32_MCx_CTL2(KVM_MAX_MCE_BANKS) - 1:
4169 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
4170 msr_info->host_initiated);
4172 if (!msr_info->host_initiated &&
4173 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
4175 msr_info->data = vcpu->arch.ia32_xss;
4177 case MSR_K7_CLK_CTL:
4179 * Provide expected ramp-up count for K7. All other
4180 * are set to zero, indicating minimum divisors for
4183 * This prevents guest kernels on AMD host with CPU
4184 * type 6, model 8 and higher from exploding due to
4185 * the rdmsr failing.
4187 msr_info->data = 0x20000000;
4189 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
4190 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
4191 case HV_X64_MSR_SYNDBG_OPTIONS:
4192 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
4193 case HV_X64_MSR_CRASH_CTL:
4194 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
4195 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
4196 case HV_X64_MSR_TSC_EMULATION_CONTROL:
4197 case HV_X64_MSR_TSC_EMULATION_STATUS:
4198 case HV_X64_MSR_TSC_INVARIANT_CONTROL:
4199 return kvm_hv_get_msr_common(vcpu,
4200 msr_info->index, &msr_info->data,
4201 msr_info->host_initiated);
4202 case MSR_IA32_BBL_CR_CTL3:
4203 /* This legacy MSR exists but isn't fully documented in current
4204 * silicon. It is however accessed by winxp in very narrow
4205 * scenarios where it sets bit #19, itself documented as
4206 * a "reserved" bit. Best effort attempt to source coherent
4207 * read data here should the balance of the register be
4208 * interpreted by the guest:
4210 * L2 cache control register 3: 64GB range, 256KB size,
4211 * enabled, latency 0x1, configured
4213 msr_info->data = 0xbe702111;
4215 case MSR_AMD64_OSVW_ID_LENGTH:
4216 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4218 msr_info->data = vcpu->arch.osvw.length;
4220 case MSR_AMD64_OSVW_STATUS:
4221 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
4223 msr_info->data = vcpu->arch.osvw.status;
4225 case MSR_PLATFORM_INFO:
4226 if (!msr_info->host_initiated &&
4227 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
4229 msr_info->data = vcpu->arch.msr_platform_info;
4231 case MSR_MISC_FEATURES_ENABLES:
4232 msr_info->data = vcpu->arch.msr_misc_features_enables;
4235 msr_info->data = vcpu->arch.msr_hwcr;
4237 #ifdef CONFIG_X86_64
4239 if (!msr_info->host_initiated &&
4240 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4243 msr_info->data = vcpu->arch.guest_fpu.fpstate->xfd;
4245 case MSR_IA32_XFD_ERR:
4246 if (!msr_info->host_initiated &&
4247 !guest_cpuid_has(vcpu, X86_FEATURE_XFD))
4250 msr_info->data = vcpu->arch.guest_fpu.xfd_err;
4254 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
4255 return kvm_pmu_get_msr(vcpu, msr_info);
4256 return KVM_MSR_RET_INVALID;
4260 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
4263 * Read or write a bunch of msrs. All parameters are kernel addresses.
4265 * @return number of msrs set successfully.
4267 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
4268 struct kvm_msr_entry *entries,
4269 int (*do_msr)(struct kvm_vcpu *vcpu,
4270 unsigned index, u64 *data))
4274 for (i = 0; i < msrs->nmsrs; ++i)
4275 if (do_msr(vcpu, entries[i].index, &entries[i].data))
4282 * Read or write a bunch of msrs. Parameters are user addresses.
4284 * @return number of msrs set successfully.
4286 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
4287 int (*do_msr)(struct kvm_vcpu *vcpu,
4288 unsigned index, u64 *data),
4291 struct kvm_msrs msrs;
4292 struct kvm_msr_entry *entries;
4297 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
4301 if (msrs.nmsrs >= MAX_IO_MSRS)
4304 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
4305 entries = memdup_user(user_msrs->entries, size);
4306 if (IS_ERR(entries)) {
4307 r = PTR_ERR(entries);
4311 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
4316 if (writeback && copy_to_user(user_msrs->entries, entries, size))
4327 static inline bool kvm_can_mwait_in_guest(void)
4329 return boot_cpu_has(X86_FEATURE_MWAIT) &&
4330 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
4331 boot_cpu_has(X86_FEATURE_ARAT);
4334 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
4335 struct kvm_cpuid2 __user *cpuid_arg)
4337 struct kvm_cpuid2 cpuid;
4341 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4344 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4349 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4355 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
4360 case KVM_CAP_IRQCHIP:
4362 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
4363 case KVM_CAP_SET_TSS_ADDR:
4364 case KVM_CAP_EXT_CPUID:
4365 case KVM_CAP_EXT_EMUL_CPUID:
4366 case KVM_CAP_CLOCKSOURCE:
4368 case KVM_CAP_NOP_IO_DELAY:
4369 case KVM_CAP_MP_STATE:
4370 case KVM_CAP_SYNC_MMU:
4371 case KVM_CAP_USER_NMI:
4372 case KVM_CAP_REINJECT_CONTROL:
4373 case KVM_CAP_IRQ_INJECT_STATUS:
4374 case KVM_CAP_IOEVENTFD:
4375 case KVM_CAP_IOEVENTFD_NO_LENGTH:
4377 case KVM_CAP_PIT_STATE2:
4378 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
4379 case KVM_CAP_VCPU_EVENTS:
4380 case KVM_CAP_HYPERV:
4381 case KVM_CAP_HYPERV_VAPIC:
4382 case KVM_CAP_HYPERV_SPIN:
4383 case KVM_CAP_HYPERV_SYNIC:
4384 case KVM_CAP_HYPERV_SYNIC2:
4385 case KVM_CAP_HYPERV_VP_INDEX:
4386 case KVM_CAP_HYPERV_EVENTFD:
4387 case KVM_CAP_HYPERV_TLBFLUSH:
4388 case KVM_CAP_HYPERV_SEND_IPI:
4389 case KVM_CAP_HYPERV_CPUID:
4390 case KVM_CAP_HYPERV_ENFORCE_CPUID:
4391 case KVM_CAP_SYS_HYPERV_CPUID:
4392 case KVM_CAP_PCI_SEGMENT:
4393 case KVM_CAP_DEBUGREGS:
4394 case KVM_CAP_X86_ROBUST_SINGLESTEP:
4396 case KVM_CAP_ASYNC_PF:
4397 case KVM_CAP_ASYNC_PF_INT:
4398 case KVM_CAP_GET_TSC_KHZ:
4399 case KVM_CAP_KVMCLOCK_CTRL:
4400 case KVM_CAP_READONLY_MEM:
4401 case KVM_CAP_HYPERV_TIME:
4402 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
4403 case KVM_CAP_TSC_DEADLINE_TIMER:
4404 case KVM_CAP_DISABLE_QUIRKS:
4405 case KVM_CAP_SET_BOOT_CPU_ID:
4406 case KVM_CAP_SPLIT_IRQCHIP:
4407 case KVM_CAP_IMMEDIATE_EXIT:
4408 case KVM_CAP_PMU_EVENT_FILTER:
4409 case KVM_CAP_GET_MSR_FEATURES:
4410 case KVM_CAP_MSR_PLATFORM_INFO:
4411 case KVM_CAP_EXCEPTION_PAYLOAD:
4412 case KVM_CAP_X86_TRIPLE_FAULT_EVENT:
4413 case KVM_CAP_SET_GUEST_DEBUG:
4414 case KVM_CAP_LAST_CPU:
4415 case KVM_CAP_X86_USER_SPACE_MSR:
4416 case KVM_CAP_X86_MSR_FILTER:
4417 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4418 #ifdef CONFIG_X86_SGX_KVM
4419 case KVM_CAP_SGX_ATTRIBUTE:
4421 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
4422 case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
4423 case KVM_CAP_SREGS2:
4424 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
4425 case KVM_CAP_VCPU_ATTRIBUTES:
4426 case KVM_CAP_SYS_ATTRIBUTES:
4428 case KVM_CAP_ENABLE_CAP:
4429 case KVM_CAP_VM_DISABLE_NX_HUGE_PAGES:
4432 case KVM_CAP_EXIT_HYPERCALL:
4433 r = KVM_EXIT_HYPERCALL_VALID_MASK;
4435 case KVM_CAP_SET_GUEST_DEBUG2:
4436 return KVM_GUESTDBG_VALID_MASK;
4437 #ifdef CONFIG_KVM_XEN
4438 case KVM_CAP_XEN_HVM:
4439 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
4440 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
4441 KVM_XEN_HVM_CONFIG_SHARED_INFO |
4442 KVM_XEN_HVM_CONFIG_EVTCHN_2LEVEL |
4443 KVM_XEN_HVM_CONFIG_EVTCHN_SEND;
4444 if (sched_info_on())
4445 r |= KVM_XEN_HVM_CONFIG_RUNSTATE |
4446 KVM_XEN_HVM_CONFIG_RUNSTATE_UPDATE_FLAG;
4449 case KVM_CAP_SYNC_REGS:
4450 r = KVM_SYNC_X86_VALID_FIELDS;
4452 case KVM_CAP_ADJUST_CLOCK:
4453 r = KVM_CLOCK_VALID_FLAGS;
4455 case KVM_CAP_X86_DISABLE_EXITS:
4456 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
4457 KVM_X86_DISABLE_EXITS_CSTATE;
4458 if(kvm_can_mwait_in_guest())
4459 r |= KVM_X86_DISABLE_EXITS_MWAIT;
4461 case KVM_CAP_X86_SMM:
4462 if (!IS_ENABLED(CONFIG_KVM_SMM))
4465 /* SMBASE is usually relocated above 1M on modern chipsets,
4466 * and SMM handlers might indeed rely on 4G segment limits,
4467 * so do not report SMM to be available if real mode is
4468 * emulated via vm86 mode. Still, do not go to great lengths
4469 * to avoid userspace's usage of the feature, because it is a
4470 * fringe case that is not enabled except via specific settings
4471 * of the module parameters.
4473 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
4475 case KVM_CAP_NR_VCPUS:
4476 r = min_t(unsigned int, num_online_cpus(), KVM_MAX_VCPUS);
4478 case KVM_CAP_MAX_VCPUS:
4481 case KVM_CAP_MAX_VCPU_ID:
4482 r = KVM_MAX_VCPU_IDS;
4484 case KVM_CAP_PV_MMU: /* obsolete */
4488 r = KVM_MAX_MCE_BANKS;
4491 r = boot_cpu_has(X86_FEATURE_XSAVE);
4493 case KVM_CAP_TSC_CONTROL:
4494 case KVM_CAP_VM_TSC_CONTROL:
4495 r = kvm_caps.has_tsc_control;
4497 case KVM_CAP_X2APIC_API:
4498 r = KVM_X2APIC_API_VALID_FLAGS;
4500 case KVM_CAP_NESTED_STATE:
4501 r = kvm_x86_ops.nested_ops->get_state ?
4502 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
4504 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4505 r = kvm_x86_ops.enable_l2_tlb_flush != NULL;
4507 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4508 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
4510 case KVM_CAP_SMALLER_MAXPHYADDR:
4511 r = (int) allow_smaller_maxphyaddr;
4513 case KVM_CAP_STEAL_TIME:
4514 r = sched_info_on();
4516 case KVM_CAP_X86_BUS_LOCK_EXIT:
4517 if (kvm_caps.has_bus_lock_exit)
4518 r = KVM_BUS_LOCK_DETECTION_OFF |
4519 KVM_BUS_LOCK_DETECTION_EXIT;
4523 case KVM_CAP_XSAVE2: {
4524 u64 guest_perm = xstate_get_guest_group_perm();
4526 r = xstate_required_size(kvm_caps.supported_xcr0 & guest_perm, false);
4527 if (r < sizeof(struct kvm_xsave))
4528 r = sizeof(struct kvm_xsave);
4531 case KVM_CAP_PMU_CAPABILITY:
4532 r = enable_pmu ? KVM_CAP_PMU_VALID_MASK : 0;
4534 case KVM_CAP_DISABLE_QUIRKS2:
4535 r = KVM_X86_VALID_QUIRKS;
4537 case KVM_CAP_X86_NOTIFY_VMEXIT:
4538 r = kvm_caps.has_notify_vmexit;
4546 static inline void __user *kvm_get_attr_addr(struct kvm_device_attr *attr)
4548 void __user *uaddr = (void __user*)(unsigned long)attr->addr;
4550 if ((u64)(unsigned long)uaddr != attr->addr)
4551 return ERR_PTR_USR(-EFAULT);
4555 static int kvm_x86_dev_get_attr(struct kvm_device_attr *attr)
4557 u64 __user *uaddr = kvm_get_attr_addr(attr);
4563 return PTR_ERR(uaddr);
4565 switch (attr->attr) {
4566 case KVM_X86_XCOMP_GUEST_SUPP:
4567 if (put_user(kvm_caps.supported_xcr0, uaddr))
4576 static int kvm_x86_dev_has_attr(struct kvm_device_attr *attr)
4581 switch (attr->attr) {
4582 case KVM_X86_XCOMP_GUEST_SUPP:
4589 long kvm_arch_dev_ioctl(struct file *filp,
4590 unsigned int ioctl, unsigned long arg)
4592 void __user *argp = (void __user *)arg;
4596 case KVM_GET_MSR_INDEX_LIST: {
4597 struct kvm_msr_list __user *user_msr_list = argp;
4598 struct kvm_msr_list msr_list;
4602 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4605 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
4606 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4609 if (n < msr_list.nmsrs)
4612 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
4613 num_msrs_to_save * sizeof(u32)))
4615 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
4617 num_emulated_msrs * sizeof(u32)))
4622 case KVM_GET_SUPPORTED_CPUID:
4623 case KVM_GET_EMULATED_CPUID: {
4624 struct kvm_cpuid2 __user *cpuid_arg = argp;
4625 struct kvm_cpuid2 cpuid;
4628 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4631 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
4637 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4642 case KVM_X86_GET_MCE_CAP_SUPPORTED:
4644 if (copy_to_user(argp, &kvm_caps.supported_mce_cap,
4645 sizeof(kvm_caps.supported_mce_cap)))
4649 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
4650 struct kvm_msr_list __user *user_msr_list = argp;
4651 struct kvm_msr_list msr_list;
4655 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
4658 msr_list.nmsrs = num_msr_based_features;
4659 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
4662 if (n < msr_list.nmsrs)
4665 if (copy_to_user(user_msr_list->indices, &msr_based_features,
4666 num_msr_based_features * sizeof(u32)))
4672 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4674 case KVM_GET_SUPPORTED_HV_CPUID:
4675 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4677 case KVM_GET_DEVICE_ATTR: {
4678 struct kvm_device_attr attr;
4680 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4682 r = kvm_x86_dev_get_attr(&attr);
4685 case KVM_HAS_DEVICE_ATTR: {
4686 struct kvm_device_attr attr;
4688 if (copy_from_user(&attr, (void __user *)arg, sizeof(attr)))
4690 r = kvm_x86_dev_has_attr(&attr);
4701 static void wbinvd_ipi(void *garbage)
4706 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4708 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4711 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4713 /* Address WBINVD may be executed by guest */
4714 if (need_emulate_wbinvd(vcpu)) {
4715 if (static_call(kvm_x86_has_wbinvd_exit)())
4716 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4717 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4718 smp_call_function_single(vcpu->cpu,
4719 wbinvd_ipi, NULL, 1);
4722 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4724 /* Save host pkru register if supported */
4725 vcpu->arch.host_pkru = read_pkru();
4727 /* Apply any externally detected TSC adjustments (due to suspend) */
4728 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4729 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4730 vcpu->arch.tsc_offset_adjustment = 0;
4731 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4734 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4735 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4736 rdtsc() - vcpu->arch.last_host_tsc;
4738 mark_tsc_unstable("KVM discovered backwards TSC");
4740 if (kvm_check_tsc_unstable()) {
4741 u64 offset = kvm_compute_l1_tsc_offset(vcpu,
4742 vcpu->arch.last_guest_tsc);
4743 kvm_vcpu_write_tsc_offset(vcpu, offset);
4744 vcpu->arch.tsc_catchup = 1;
4747 if (kvm_lapic_hv_timer_in_use(vcpu))
4748 kvm_lapic_restart_hv_timer(vcpu);
4751 * On a host with synchronized TSC, there is no need to update
4752 * kvmclock on vcpu->cpu migration
4754 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4755 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4756 if (vcpu->cpu != cpu)
4757 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4761 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4764 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4766 struct gfn_to_hva_cache *ghc = &vcpu->arch.st.cache;
4767 struct kvm_steal_time __user *st;
4768 struct kvm_memslots *slots;
4769 static const u8 preempted = KVM_VCPU_PREEMPTED;
4770 gpa_t gpa = vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS;
4773 * The vCPU can be marked preempted if and only if the VM-Exit was on
4774 * an instruction boundary and will not trigger guest emulation of any
4775 * kind (see vcpu_run). Vendor specific code controls (conservatively)
4776 * when this is true, for example allowing the vCPU to be marked
4777 * preempted if and only if the VM-Exit was due to a host interrupt.
4779 if (!vcpu->arch.at_instruction_boundary) {
4780 vcpu->stat.preemption_other++;
4784 vcpu->stat.preemption_reported++;
4785 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4788 if (vcpu->arch.st.preempted)
4791 /* This happens on process exit */
4792 if (unlikely(current->mm != vcpu->kvm->mm))
4795 slots = kvm_memslots(vcpu->kvm);
4797 if (unlikely(slots->generation != ghc->generation ||
4799 kvm_is_error_hva(ghc->hva) || !ghc->memslot))
4802 st = (struct kvm_steal_time __user *)ghc->hva;
4803 BUILD_BUG_ON(sizeof(st->preempted) != sizeof(preempted));
4805 if (!copy_to_user_nofault(&st->preempted, &preempted, sizeof(preempted)))
4806 vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4808 mark_page_dirty_in_slot(vcpu->kvm, ghc->memslot, gpa_to_gfn(ghc->gpa));
4811 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4815 if (vcpu->preempted) {
4816 if (!vcpu->arch.guest_state_protected)
4817 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4820 * Take the srcu lock as memslots will be accessed to check the gfn
4821 * cache generation against the memslots generation.
4823 idx = srcu_read_lock(&vcpu->kvm->srcu);
4824 if (kvm_xen_msr_enabled(vcpu->kvm))
4825 kvm_xen_runstate_set_preempted(vcpu);
4827 kvm_steal_time_set_preempted(vcpu);
4828 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4831 static_call(kvm_x86_vcpu_put)(vcpu);
4832 vcpu->arch.last_host_tsc = rdtsc();
4835 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4836 struct kvm_lapic_state *s)
4838 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
4840 return kvm_apic_get_state(vcpu, s);
4843 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4844 struct kvm_lapic_state *s)
4848 r = kvm_apic_set_state(vcpu, s);
4851 update_cr8_intercept(vcpu);
4856 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4859 * We can accept userspace's request for interrupt injection
4860 * as long as we have a place to store the interrupt number.
4861 * The actual injection will happen when the CPU is able to
4862 * deliver the interrupt.
4864 if (kvm_cpu_has_extint(vcpu))
4867 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4868 return (!lapic_in_kernel(vcpu) ||
4869 kvm_apic_accept_pic_intr(vcpu));
4872 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4875 * Do not cause an interrupt window exit if an exception
4876 * is pending or an event needs reinjection; userspace
4877 * might want to inject the interrupt manually using KVM_SET_REGS
4878 * or KVM_SET_SREGS. For that to work, we must be at an
4879 * instruction boundary and with no events half-injected.
4881 return (kvm_arch_interrupt_allowed(vcpu) &&
4882 kvm_cpu_accept_dm_intr(vcpu) &&
4883 !kvm_event_needs_reinjection(vcpu) &&
4884 !kvm_is_exception_pending(vcpu));
4887 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4888 struct kvm_interrupt *irq)
4890 if (irq->irq >= KVM_NR_INTERRUPTS)
4893 if (!irqchip_in_kernel(vcpu->kvm)) {
4894 kvm_queue_interrupt(vcpu, irq->irq, false);
4895 kvm_make_request(KVM_REQ_EVENT, vcpu);
4900 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4901 * fail for in-kernel 8259.
4903 if (pic_in_kernel(vcpu->kvm))
4906 if (vcpu->arch.pending_external_vector != -1)
4909 vcpu->arch.pending_external_vector = irq->irq;
4910 kvm_make_request(KVM_REQ_EVENT, vcpu);
4914 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4916 kvm_inject_nmi(vcpu);
4921 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4922 struct kvm_tpr_access_ctl *tac)
4926 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4930 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4934 unsigned bank_num = mcg_cap & 0xff, bank;
4937 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4939 if (mcg_cap & ~(kvm_caps.supported_mce_cap | 0xff | 0xff0000))
4942 vcpu->arch.mcg_cap = mcg_cap;
4943 /* Init IA32_MCG_CTL to all 1s */
4944 if (mcg_cap & MCG_CTL_P)
4945 vcpu->arch.mcg_ctl = ~(u64)0;
4946 /* Init IA32_MCi_CTL to all 1s, IA32_MCi_CTL2 to all 0s */
4947 for (bank = 0; bank < bank_num; bank++) {
4948 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4949 if (mcg_cap & MCG_CMCI_P)
4950 vcpu->arch.mci_ctl2_banks[bank] = 0;
4953 kvm_apic_after_set_mcg_cap(vcpu);
4955 static_call(kvm_x86_setup_mce)(vcpu);
4961 * Validate this is an UCNA (uncorrectable no action) error by checking the
4962 * MCG_STATUS and MCi_STATUS registers:
4963 * - none of the bits for Machine Check Exceptions are set
4964 * - both the VAL (valid) and UC (uncorrectable) bits are set
4965 * MCI_STATUS_PCC - Processor Context Corrupted
4966 * MCI_STATUS_S - Signaled as a Machine Check Exception
4967 * MCI_STATUS_AR - Software recoverable Action Required
4969 static bool is_ucna(struct kvm_x86_mce *mce)
4971 return !mce->mcg_status &&
4972 !(mce->status & (MCI_STATUS_PCC | MCI_STATUS_S | MCI_STATUS_AR)) &&
4973 (mce->status & MCI_STATUS_VAL) &&
4974 (mce->status & MCI_STATUS_UC);
4977 static int kvm_vcpu_x86_set_ucna(struct kvm_vcpu *vcpu, struct kvm_x86_mce *mce, u64* banks)
4979 u64 mcg_cap = vcpu->arch.mcg_cap;
4981 banks[1] = mce->status;
4982 banks[2] = mce->addr;
4983 banks[3] = mce->misc;
4984 vcpu->arch.mcg_status = mce->mcg_status;
4986 if (!(mcg_cap & MCG_CMCI_P) ||
4987 !(vcpu->arch.mci_ctl2_banks[mce->bank] & MCI_CTL2_CMCI_EN))
4990 if (lapic_in_kernel(vcpu))
4991 kvm_apic_local_deliver(vcpu->arch.apic, APIC_LVTCMCI);
4996 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4997 struct kvm_x86_mce *mce)
4999 u64 mcg_cap = vcpu->arch.mcg_cap;
5000 unsigned bank_num = mcg_cap & 0xff;
5001 u64 *banks = vcpu->arch.mce_banks;
5003 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
5006 banks += array_index_nospec(4 * mce->bank, 4 * bank_num);
5009 return kvm_vcpu_x86_set_ucna(vcpu, mce, banks);
5012 * if IA32_MCG_CTL is not all 1s, the uncorrected error
5013 * reporting is disabled
5015 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
5016 vcpu->arch.mcg_ctl != ~(u64)0)
5019 * if IA32_MCi_CTL is not all 1s, the uncorrected error
5020 * reporting is disabled for the bank
5022 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
5024 if (mce->status & MCI_STATUS_UC) {
5025 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
5026 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
5027 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5030 if (banks[1] & MCI_STATUS_VAL)
5031 mce->status |= MCI_STATUS_OVER;
5032 banks[2] = mce->addr;
5033 banks[3] = mce->misc;
5034 vcpu->arch.mcg_status = mce->mcg_status;
5035 banks[1] = mce->status;
5036 kvm_queue_exception(vcpu, MC_VECTOR);
5037 } else if (!(banks[1] & MCI_STATUS_VAL)
5038 || !(banks[1] & MCI_STATUS_UC)) {
5039 if (banks[1] & MCI_STATUS_VAL)
5040 mce->status |= MCI_STATUS_OVER;
5041 banks[2] = mce->addr;
5042 banks[3] = mce->misc;
5043 banks[1] = mce->status;
5045 banks[1] |= MCI_STATUS_OVER;
5049 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
5050 struct kvm_vcpu_events *events)
5052 struct kvm_queued_exception *ex;
5056 #ifdef CONFIG_KVM_SMM
5057 if (kvm_check_request(KVM_REQ_SMI, vcpu))
5062 * KVM's ABI only allows for one exception to be migrated. Luckily,
5063 * the only time there can be two queued exceptions is if there's a
5064 * non-exiting _injected_ exception, and a pending exiting exception.
5065 * In that case, ignore the VM-Exiting exception as it's an extension
5066 * of the injected exception.
5068 if (vcpu->arch.exception_vmexit.pending &&
5069 !vcpu->arch.exception.pending &&
5070 !vcpu->arch.exception.injected)
5071 ex = &vcpu->arch.exception_vmexit;
5073 ex = &vcpu->arch.exception;
5076 * In guest mode, payload delivery should be deferred if the exception
5077 * will be intercepted by L1, e.g. KVM should not modifying CR2 if L1
5078 * intercepts #PF, ditto for DR6 and #DBs. If the per-VM capability,
5079 * KVM_CAP_EXCEPTION_PAYLOAD, is not set, userspace may or may not
5080 * propagate the payload and so it cannot be safely deferred. Deliver
5081 * the payload if the capability hasn't been requested.
5083 if (!vcpu->kvm->arch.exception_payload_enabled &&
5084 ex->pending && ex->has_payload)
5085 kvm_deliver_exception_payload(vcpu, ex);
5087 memset(events, 0, sizeof(*events));
5090 * The API doesn't provide the instruction length for software
5091 * exceptions, so don't report them. As long as the guest RIP
5092 * isn't advanced, we should expect to encounter the exception
5095 if (!kvm_exception_is_soft(ex->vector)) {
5096 events->exception.injected = ex->injected;
5097 events->exception.pending = ex->pending;
5099 * For ABI compatibility, deliberately conflate
5100 * pending and injected exceptions when
5101 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
5103 if (!vcpu->kvm->arch.exception_payload_enabled)
5104 events->exception.injected |= ex->pending;
5106 events->exception.nr = ex->vector;
5107 events->exception.has_error_code = ex->has_error_code;
5108 events->exception.error_code = ex->error_code;
5109 events->exception_has_payload = ex->has_payload;
5110 events->exception_payload = ex->payload;
5112 events->interrupt.injected =
5113 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
5114 events->interrupt.nr = vcpu->arch.interrupt.nr;
5115 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
5117 events->nmi.injected = vcpu->arch.nmi_injected;
5118 events->nmi.pending = vcpu->arch.nmi_pending != 0;
5119 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
5121 /* events->sipi_vector is never valid when reporting to user space */
5123 #ifdef CONFIG_KVM_SMM
5124 events->smi.smm = is_smm(vcpu);
5125 events->smi.pending = vcpu->arch.smi_pending;
5126 events->smi.smm_inside_nmi =
5127 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
5129 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
5131 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
5132 | KVM_VCPUEVENT_VALID_SHADOW
5133 | KVM_VCPUEVENT_VALID_SMM);
5134 if (vcpu->kvm->arch.exception_payload_enabled)
5135 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
5136 if (vcpu->kvm->arch.triple_fault_event) {
5137 events->triple_fault.pending = kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5138 events->flags |= KVM_VCPUEVENT_VALID_TRIPLE_FAULT;
5142 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
5143 struct kvm_vcpu_events *events)
5145 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
5146 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
5147 | KVM_VCPUEVENT_VALID_SHADOW
5148 | KVM_VCPUEVENT_VALID_SMM
5149 | KVM_VCPUEVENT_VALID_PAYLOAD
5150 | KVM_VCPUEVENT_VALID_TRIPLE_FAULT))
5153 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
5154 if (!vcpu->kvm->arch.exception_payload_enabled)
5156 if (events->exception.pending)
5157 events->exception.injected = 0;
5159 events->exception_has_payload = 0;
5161 events->exception.pending = 0;
5162 events->exception_has_payload = 0;
5165 if ((events->exception.injected || events->exception.pending) &&
5166 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
5169 /* INITs are latched while in SMM */
5170 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
5171 (events->smi.smm || events->smi.pending) &&
5172 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
5178 * Flag that userspace is stuffing an exception, the next KVM_RUN will
5179 * morph the exception to a VM-Exit if appropriate. Do this only for
5180 * pending exceptions, already-injected exceptions are not subject to
5181 * intercpetion. Note, userspace that conflates pending and injected
5182 * is hosed, and will incorrectly convert an injected exception into a
5183 * pending exception, which in turn may cause a spurious VM-Exit.
5185 vcpu->arch.exception_from_userspace = events->exception.pending;
5187 vcpu->arch.exception_vmexit.pending = false;
5189 vcpu->arch.exception.injected = events->exception.injected;
5190 vcpu->arch.exception.pending = events->exception.pending;
5191 vcpu->arch.exception.vector = events->exception.nr;
5192 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
5193 vcpu->arch.exception.error_code = events->exception.error_code;
5194 vcpu->arch.exception.has_payload = events->exception_has_payload;
5195 vcpu->arch.exception.payload = events->exception_payload;
5197 vcpu->arch.interrupt.injected = events->interrupt.injected;
5198 vcpu->arch.interrupt.nr = events->interrupt.nr;
5199 vcpu->arch.interrupt.soft = events->interrupt.soft;
5200 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
5201 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
5202 events->interrupt.shadow);
5204 vcpu->arch.nmi_injected = events->nmi.injected;
5205 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
5206 vcpu->arch.nmi_pending = events->nmi.pending;
5207 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
5209 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
5210 lapic_in_kernel(vcpu))
5211 vcpu->arch.apic->sipi_vector = events->sipi_vector;
5213 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
5214 #ifdef CONFIG_KVM_SMM
5215 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
5216 kvm_leave_nested(vcpu);
5217 kvm_smm_changed(vcpu, events->smi.smm);
5220 vcpu->arch.smi_pending = events->smi.pending;
5222 if (events->smi.smm) {
5223 if (events->smi.smm_inside_nmi)
5224 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
5226 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
5230 if (events->smi.smm || events->smi.pending ||
5231 events->smi.smm_inside_nmi)
5235 if (lapic_in_kernel(vcpu)) {
5236 if (events->smi.latched_init)
5237 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
5239 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
5243 if (events->flags & KVM_VCPUEVENT_VALID_TRIPLE_FAULT) {
5244 if (!vcpu->kvm->arch.triple_fault_event)
5246 if (events->triple_fault.pending)
5247 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5249 kvm_clear_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5252 kvm_make_request(KVM_REQ_EVENT, vcpu);
5257 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
5258 struct kvm_debugregs *dbgregs)
5262 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
5263 kvm_get_dr(vcpu, 6, &val);
5265 dbgregs->dr7 = vcpu->arch.dr7;
5267 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
5270 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
5271 struct kvm_debugregs *dbgregs)
5276 if (!kvm_dr6_valid(dbgregs->dr6))
5278 if (!kvm_dr7_valid(dbgregs->dr7))
5281 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
5282 kvm_update_dr0123(vcpu);
5283 vcpu->arch.dr6 = dbgregs->dr6;
5284 vcpu->arch.dr7 = dbgregs->dr7;
5285 kvm_update_dr7(vcpu);
5290 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
5291 struct kvm_xsave *guest_xsave)
5293 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5296 fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
5297 guest_xsave->region,
5298 sizeof(guest_xsave->region),
5302 static void kvm_vcpu_ioctl_x86_get_xsave2(struct kvm_vcpu *vcpu,
5303 u8 *state, unsigned int size)
5305 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5308 fpu_copy_guest_fpstate_to_uabi(&vcpu->arch.guest_fpu,
5309 state, size, vcpu->arch.pkru);
5312 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
5313 struct kvm_xsave *guest_xsave)
5315 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
5318 return fpu_copy_uabi_to_guest_fpstate(&vcpu->arch.guest_fpu,
5319 guest_xsave->region,
5320 kvm_caps.supported_xcr0,
5324 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
5325 struct kvm_xcrs *guest_xcrs)
5327 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
5328 guest_xcrs->nr_xcrs = 0;
5332 guest_xcrs->nr_xcrs = 1;
5333 guest_xcrs->flags = 0;
5334 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
5335 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
5338 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
5339 struct kvm_xcrs *guest_xcrs)
5343 if (!boot_cpu_has(X86_FEATURE_XSAVE))
5346 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
5349 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
5350 /* Only support XCR0 currently */
5351 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
5352 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
5353 guest_xcrs->xcrs[i].value);
5362 * kvm_set_guest_paused() indicates to the guest kernel that it has been
5363 * stopped by the hypervisor. This function will be called from the host only.
5364 * EINVAL is returned when the host attempts to set the flag for a guest that
5365 * does not support pv clocks.
5367 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
5369 if (!vcpu->arch.pv_time.active)
5371 vcpu->arch.pvclock_set_guest_stopped_request = true;
5372 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
5376 static int kvm_arch_tsc_has_attr(struct kvm_vcpu *vcpu,
5377 struct kvm_device_attr *attr)
5381 switch (attr->attr) {
5382 case KVM_VCPU_TSC_OFFSET:
5392 static int kvm_arch_tsc_get_attr(struct kvm_vcpu *vcpu,
5393 struct kvm_device_attr *attr)
5395 u64 __user *uaddr = kvm_get_attr_addr(attr);
5399 return PTR_ERR(uaddr);
5401 switch (attr->attr) {
5402 case KVM_VCPU_TSC_OFFSET:
5404 if (put_user(vcpu->arch.l1_tsc_offset, uaddr))
5415 static int kvm_arch_tsc_set_attr(struct kvm_vcpu *vcpu,
5416 struct kvm_device_attr *attr)
5418 u64 __user *uaddr = kvm_get_attr_addr(attr);
5419 struct kvm *kvm = vcpu->kvm;
5423 return PTR_ERR(uaddr);
5425 switch (attr->attr) {
5426 case KVM_VCPU_TSC_OFFSET: {
5427 u64 offset, tsc, ns;
5428 unsigned long flags;
5432 if (get_user(offset, uaddr))
5435 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
5437 matched = (vcpu->arch.virtual_tsc_khz &&
5438 kvm->arch.last_tsc_khz == vcpu->arch.virtual_tsc_khz &&
5439 kvm->arch.last_tsc_offset == offset);
5441 tsc = kvm_scale_tsc(rdtsc(), vcpu->arch.l1_tsc_scaling_ratio) + offset;
5442 ns = get_kvmclock_base_ns();
5444 __kvm_synchronize_tsc(vcpu, offset, tsc, ns, matched);
5445 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
5457 static int kvm_vcpu_ioctl_device_attr(struct kvm_vcpu *vcpu,
5461 struct kvm_device_attr attr;
5464 if (copy_from_user(&attr, argp, sizeof(attr)))
5467 if (attr.group != KVM_VCPU_TSC_CTRL)
5471 case KVM_HAS_DEVICE_ATTR:
5472 r = kvm_arch_tsc_has_attr(vcpu, &attr);
5474 case KVM_GET_DEVICE_ATTR:
5475 r = kvm_arch_tsc_get_attr(vcpu, &attr);
5477 case KVM_SET_DEVICE_ATTR:
5478 r = kvm_arch_tsc_set_attr(vcpu, &attr);
5485 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
5486 struct kvm_enable_cap *cap)
5489 uint16_t vmcs_version;
5490 void __user *user_ptr;
5496 case KVM_CAP_HYPERV_SYNIC2:
5501 case KVM_CAP_HYPERV_SYNIC:
5502 if (!irqchip_in_kernel(vcpu->kvm))
5504 return kvm_hv_activate_synic(vcpu, cap->cap ==
5505 KVM_CAP_HYPERV_SYNIC2);
5506 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
5507 if (!kvm_x86_ops.nested_ops->enable_evmcs)
5509 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
5511 user_ptr = (void __user *)(uintptr_t)cap->args[0];
5512 if (copy_to_user(user_ptr, &vmcs_version,
5513 sizeof(vmcs_version)))
5517 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
5518 if (!kvm_x86_ops.enable_l2_tlb_flush)
5521 return static_call(kvm_x86_enable_l2_tlb_flush)(vcpu);
5523 case KVM_CAP_HYPERV_ENFORCE_CPUID:
5524 return kvm_hv_set_enforce_cpuid(vcpu, cap->args[0]);
5526 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
5527 vcpu->arch.pv_cpuid.enforce = cap->args[0];
5528 if (vcpu->arch.pv_cpuid.enforce)
5529 kvm_update_pv_runtime(vcpu);
5537 long kvm_arch_vcpu_ioctl(struct file *filp,
5538 unsigned int ioctl, unsigned long arg)
5540 struct kvm_vcpu *vcpu = filp->private_data;
5541 void __user *argp = (void __user *)arg;
5544 struct kvm_sregs2 *sregs2;
5545 struct kvm_lapic_state *lapic;
5546 struct kvm_xsave *xsave;
5547 struct kvm_xcrs *xcrs;
5555 case KVM_GET_LAPIC: {
5557 if (!lapic_in_kernel(vcpu))
5559 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
5560 GFP_KERNEL_ACCOUNT);
5565 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
5569 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
5574 case KVM_SET_LAPIC: {
5576 if (!lapic_in_kernel(vcpu))
5578 u.lapic = memdup_user(argp, sizeof(*u.lapic));
5579 if (IS_ERR(u.lapic)) {
5580 r = PTR_ERR(u.lapic);
5584 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
5587 case KVM_INTERRUPT: {
5588 struct kvm_interrupt irq;
5591 if (copy_from_user(&irq, argp, sizeof(irq)))
5593 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
5597 r = kvm_vcpu_ioctl_nmi(vcpu);
5601 r = kvm_inject_smi(vcpu);
5604 case KVM_SET_CPUID: {
5605 struct kvm_cpuid __user *cpuid_arg = argp;
5606 struct kvm_cpuid cpuid;
5609 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5611 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
5614 case KVM_SET_CPUID2: {
5615 struct kvm_cpuid2 __user *cpuid_arg = argp;
5616 struct kvm_cpuid2 cpuid;
5619 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5621 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
5622 cpuid_arg->entries);
5625 case KVM_GET_CPUID2: {
5626 struct kvm_cpuid2 __user *cpuid_arg = argp;
5627 struct kvm_cpuid2 cpuid;
5630 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
5632 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
5633 cpuid_arg->entries);
5637 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
5642 case KVM_GET_MSRS: {
5643 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5644 r = msr_io(vcpu, argp, do_get_msr, 1);
5645 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5648 case KVM_SET_MSRS: {
5649 int idx = srcu_read_lock(&vcpu->kvm->srcu);
5650 r = msr_io(vcpu, argp, do_set_msr, 0);
5651 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5654 case KVM_TPR_ACCESS_REPORTING: {
5655 struct kvm_tpr_access_ctl tac;
5658 if (copy_from_user(&tac, argp, sizeof(tac)))
5660 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
5664 if (copy_to_user(argp, &tac, sizeof(tac)))
5669 case KVM_SET_VAPIC_ADDR: {
5670 struct kvm_vapic_addr va;
5674 if (!lapic_in_kernel(vcpu))
5677 if (copy_from_user(&va, argp, sizeof(va)))
5679 idx = srcu_read_lock(&vcpu->kvm->srcu);
5680 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
5681 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5684 case KVM_X86_SETUP_MCE: {
5688 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
5690 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
5693 case KVM_X86_SET_MCE: {
5694 struct kvm_x86_mce mce;
5697 if (copy_from_user(&mce, argp, sizeof(mce)))
5699 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
5702 case KVM_GET_VCPU_EVENTS: {
5703 struct kvm_vcpu_events events;
5705 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
5708 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
5713 case KVM_SET_VCPU_EVENTS: {
5714 struct kvm_vcpu_events events;
5717 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
5720 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
5723 case KVM_GET_DEBUGREGS: {
5724 struct kvm_debugregs dbgregs;
5726 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
5729 if (copy_to_user(argp, &dbgregs,
5730 sizeof(struct kvm_debugregs)))
5735 case KVM_SET_DEBUGREGS: {
5736 struct kvm_debugregs dbgregs;
5739 if (copy_from_user(&dbgregs, argp,
5740 sizeof(struct kvm_debugregs)))
5743 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
5746 case KVM_GET_XSAVE: {
5748 if (vcpu->arch.guest_fpu.uabi_size > sizeof(struct kvm_xsave))
5751 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
5756 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
5759 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
5764 case KVM_SET_XSAVE: {
5765 int size = vcpu->arch.guest_fpu.uabi_size;
5767 u.xsave = memdup_user(argp, size);
5768 if (IS_ERR(u.xsave)) {
5769 r = PTR_ERR(u.xsave);
5773 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
5777 case KVM_GET_XSAVE2: {
5778 int size = vcpu->arch.guest_fpu.uabi_size;
5780 u.xsave = kzalloc(size, GFP_KERNEL_ACCOUNT);
5785 kvm_vcpu_ioctl_x86_get_xsave2(vcpu, u.buffer, size);
5788 if (copy_to_user(argp, u.xsave, size))
5795 case KVM_GET_XCRS: {
5796 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
5801 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
5804 if (copy_to_user(argp, u.xcrs,
5805 sizeof(struct kvm_xcrs)))
5810 case KVM_SET_XCRS: {
5811 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
5812 if (IS_ERR(u.xcrs)) {
5813 r = PTR_ERR(u.xcrs);
5817 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
5820 case KVM_SET_TSC_KHZ: {
5824 user_tsc_khz = (u32)arg;
5826 if (kvm_caps.has_tsc_control &&
5827 user_tsc_khz >= kvm_caps.max_guest_tsc_khz)
5830 if (user_tsc_khz == 0)
5831 user_tsc_khz = tsc_khz;
5833 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5838 case KVM_GET_TSC_KHZ: {
5839 r = vcpu->arch.virtual_tsc_khz;
5842 case KVM_KVMCLOCK_CTRL: {
5843 r = kvm_set_guest_paused(vcpu);
5846 case KVM_ENABLE_CAP: {
5847 struct kvm_enable_cap cap;
5850 if (copy_from_user(&cap, argp, sizeof(cap)))
5852 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5855 case KVM_GET_NESTED_STATE: {
5856 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5860 if (!kvm_x86_ops.nested_ops->get_state)
5863 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5865 if (get_user(user_data_size, &user_kvm_nested_state->size))
5868 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5873 if (r > user_data_size) {
5874 if (put_user(r, &user_kvm_nested_state->size))
5884 case KVM_SET_NESTED_STATE: {
5885 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5886 struct kvm_nested_state kvm_state;
5890 if (!kvm_x86_ops.nested_ops->set_state)
5894 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5898 if (kvm_state.size < sizeof(kvm_state))
5901 if (kvm_state.flags &
5902 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5903 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5904 | KVM_STATE_NESTED_GIF_SET))
5907 /* nested_run_pending implies guest_mode. */
5908 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5909 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5912 idx = srcu_read_lock(&vcpu->kvm->srcu);
5913 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5914 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5917 case KVM_GET_SUPPORTED_HV_CPUID:
5918 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5920 #ifdef CONFIG_KVM_XEN
5921 case KVM_XEN_VCPU_GET_ATTR: {
5922 struct kvm_xen_vcpu_attr xva;
5925 if (copy_from_user(&xva, argp, sizeof(xva)))
5927 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5928 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5932 case KVM_XEN_VCPU_SET_ATTR: {
5933 struct kvm_xen_vcpu_attr xva;
5936 if (copy_from_user(&xva, argp, sizeof(xva)))
5938 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5942 case KVM_GET_SREGS2: {
5943 u.sregs2 = kzalloc(sizeof(struct kvm_sregs2), GFP_KERNEL);
5947 __get_sregs2(vcpu, u.sregs2);
5949 if (copy_to_user(argp, u.sregs2, sizeof(struct kvm_sregs2)))
5954 case KVM_SET_SREGS2: {
5955 u.sregs2 = memdup_user(argp, sizeof(struct kvm_sregs2));
5956 if (IS_ERR(u.sregs2)) {
5957 r = PTR_ERR(u.sregs2);
5961 r = __set_sregs2(vcpu, u.sregs2);
5964 case KVM_HAS_DEVICE_ATTR:
5965 case KVM_GET_DEVICE_ATTR:
5966 case KVM_SET_DEVICE_ATTR:
5967 r = kvm_vcpu_ioctl_device_attr(vcpu, ioctl, argp);
5979 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5981 return VM_FAULT_SIGBUS;
5984 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5988 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5990 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5994 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5997 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
6000 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
6001 unsigned long kvm_nr_mmu_pages)
6003 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
6006 mutex_lock(&kvm->slots_lock);
6008 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
6009 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
6011 mutex_unlock(&kvm->slots_lock);
6015 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
6017 return kvm->arch.n_max_mmu_pages;
6020 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
6022 struct kvm_pic *pic = kvm->arch.vpic;
6026 switch (chip->chip_id) {
6027 case KVM_IRQCHIP_PIC_MASTER:
6028 memcpy(&chip->chip.pic, &pic->pics[0],
6029 sizeof(struct kvm_pic_state));
6031 case KVM_IRQCHIP_PIC_SLAVE:
6032 memcpy(&chip->chip.pic, &pic->pics[1],
6033 sizeof(struct kvm_pic_state));
6035 case KVM_IRQCHIP_IOAPIC:
6036 kvm_get_ioapic(kvm, &chip->chip.ioapic);
6045 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
6047 struct kvm_pic *pic = kvm->arch.vpic;
6051 switch (chip->chip_id) {
6052 case KVM_IRQCHIP_PIC_MASTER:
6053 spin_lock(&pic->lock);
6054 memcpy(&pic->pics[0], &chip->chip.pic,
6055 sizeof(struct kvm_pic_state));
6056 spin_unlock(&pic->lock);
6058 case KVM_IRQCHIP_PIC_SLAVE:
6059 spin_lock(&pic->lock);
6060 memcpy(&pic->pics[1], &chip->chip.pic,
6061 sizeof(struct kvm_pic_state));
6062 spin_unlock(&pic->lock);
6064 case KVM_IRQCHIP_IOAPIC:
6065 kvm_set_ioapic(kvm, &chip->chip.ioapic);
6071 kvm_pic_update_irq(pic);
6075 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
6077 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
6079 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
6081 mutex_lock(&kps->lock);
6082 memcpy(ps, &kps->channels, sizeof(*ps));
6083 mutex_unlock(&kps->lock);
6087 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
6090 struct kvm_pit *pit = kvm->arch.vpit;
6092 mutex_lock(&pit->pit_state.lock);
6093 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
6094 for (i = 0; i < 3; i++)
6095 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
6096 mutex_unlock(&pit->pit_state.lock);
6100 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
6102 mutex_lock(&kvm->arch.vpit->pit_state.lock);
6103 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
6104 sizeof(ps->channels));
6105 ps->flags = kvm->arch.vpit->pit_state.flags;
6106 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
6107 memset(&ps->reserved, 0, sizeof(ps->reserved));
6111 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
6115 u32 prev_legacy, cur_legacy;
6116 struct kvm_pit *pit = kvm->arch.vpit;
6118 mutex_lock(&pit->pit_state.lock);
6119 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
6120 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
6121 if (!prev_legacy && cur_legacy)
6123 memcpy(&pit->pit_state.channels, &ps->channels,
6124 sizeof(pit->pit_state.channels));
6125 pit->pit_state.flags = ps->flags;
6126 for (i = 0; i < 3; i++)
6127 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
6129 mutex_unlock(&pit->pit_state.lock);
6133 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
6134 struct kvm_reinject_control *control)
6136 struct kvm_pit *pit = kvm->arch.vpit;
6138 /* pit->pit_state.lock was overloaded to prevent userspace from getting
6139 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
6140 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
6142 mutex_lock(&pit->pit_state.lock);
6143 kvm_pit_set_reinject(pit, control->pit_reinject);
6144 mutex_unlock(&pit->pit_state.lock);
6149 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
6153 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
6154 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
6155 * on all VM-Exits, thus we only need to kick running vCPUs to force a
6158 struct kvm_vcpu *vcpu;
6161 kvm_for_each_vcpu(i, vcpu, kvm)
6162 kvm_vcpu_kick(vcpu);
6165 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
6168 if (!irqchip_in_kernel(kvm))
6171 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
6172 irq_event->irq, irq_event->level,
6177 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
6178 struct kvm_enable_cap *cap)
6186 case KVM_CAP_DISABLE_QUIRKS2:
6188 if (cap->args[0] & ~KVM_X86_VALID_QUIRKS)
6191 case KVM_CAP_DISABLE_QUIRKS:
6192 kvm->arch.disabled_quirks = cap->args[0];
6195 case KVM_CAP_SPLIT_IRQCHIP: {
6196 mutex_lock(&kvm->lock);
6198 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
6199 goto split_irqchip_unlock;
6201 if (irqchip_in_kernel(kvm))
6202 goto split_irqchip_unlock;
6203 if (kvm->created_vcpus)
6204 goto split_irqchip_unlock;
6205 r = kvm_setup_empty_irq_routing(kvm);
6207 goto split_irqchip_unlock;
6208 /* Pairs with irqchip_in_kernel. */
6210 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
6211 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
6212 kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_ABSENT);
6214 split_irqchip_unlock:
6215 mutex_unlock(&kvm->lock);
6218 case KVM_CAP_X2APIC_API:
6220 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
6223 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
6224 kvm->arch.x2apic_format = true;
6225 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
6226 kvm->arch.x2apic_broadcast_quirk_disabled = true;
6230 case KVM_CAP_X86_DISABLE_EXITS:
6232 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
6235 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
6236 kvm_can_mwait_in_guest())
6237 kvm->arch.mwait_in_guest = true;
6238 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
6239 kvm->arch.hlt_in_guest = true;
6240 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
6241 kvm->arch.pause_in_guest = true;
6242 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
6243 kvm->arch.cstate_in_guest = true;
6246 case KVM_CAP_MSR_PLATFORM_INFO:
6247 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
6250 case KVM_CAP_EXCEPTION_PAYLOAD:
6251 kvm->arch.exception_payload_enabled = cap->args[0];
6254 case KVM_CAP_X86_TRIPLE_FAULT_EVENT:
6255 kvm->arch.triple_fault_event = cap->args[0];
6258 case KVM_CAP_X86_USER_SPACE_MSR:
6260 if (cap->args[0] & ~KVM_MSR_EXIT_REASON_VALID_MASK)
6262 kvm->arch.user_space_msr_mask = cap->args[0];
6265 case KVM_CAP_X86_BUS_LOCK_EXIT:
6267 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
6270 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
6271 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
6274 if (kvm_caps.has_bus_lock_exit &&
6275 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
6276 kvm->arch.bus_lock_detection_enabled = true;
6279 #ifdef CONFIG_X86_SGX_KVM
6280 case KVM_CAP_SGX_ATTRIBUTE: {
6281 unsigned long allowed_attributes = 0;
6283 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
6287 /* KVM only supports the PROVISIONKEY privileged attribute. */
6288 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
6289 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
6290 kvm->arch.sgx_provisioning_allowed = true;
6296 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
6298 if (!kvm_x86_ops.vm_copy_enc_context_from)
6301 r = static_call(kvm_x86_vm_copy_enc_context_from)(kvm, cap->args[0]);
6303 case KVM_CAP_VM_MOVE_ENC_CONTEXT_FROM:
6305 if (!kvm_x86_ops.vm_move_enc_context_from)
6308 r = static_call(kvm_x86_vm_move_enc_context_from)(kvm, cap->args[0]);
6310 case KVM_CAP_EXIT_HYPERCALL:
6311 if (cap->args[0] & ~KVM_EXIT_HYPERCALL_VALID_MASK) {
6315 kvm->arch.hypercall_exit_enabled = cap->args[0];
6318 case KVM_CAP_EXIT_ON_EMULATION_FAILURE:
6320 if (cap->args[0] & ~1)
6322 kvm->arch.exit_on_emulation_error = cap->args[0];
6325 case KVM_CAP_PMU_CAPABILITY:
6327 if (!enable_pmu || (cap->args[0] & ~KVM_CAP_PMU_VALID_MASK))
6330 mutex_lock(&kvm->lock);
6331 if (!kvm->created_vcpus) {
6332 kvm->arch.enable_pmu = !(cap->args[0] & KVM_PMU_CAP_DISABLE);
6335 mutex_unlock(&kvm->lock);
6337 case KVM_CAP_MAX_VCPU_ID:
6339 if (cap->args[0] > KVM_MAX_VCPU_IDS)
6342 mutex_lock(&kvm->lock);
6343 if (kvm->arch.max_vcpu_ids == cap->args[0]) {
6345 } else if (!kvm->arch.max_vcpu_ids) {
6346 kvm->arch.max_vcpu_ids = cap->args[0];
6349 mutex_unlock(&kvm->lock);
6351 case KVM_CAP_X86_NOTIFY_VMEXIT:
6353 if ((u32)cap->args[0] & ~KVM_X86_NOTIFY_VMEXIT_VALID_BITS)
6355 if (!kvm_caps.has_notify_vmexit)
6357 if (!((u32)cap->args[0] & KVM_X86_NOTIFY_VMEXIT_ENABLED))
6359 mutex_lock(&kvm->lock);
6360 if (!kvm->created_vcpus) {
6361 kvm->arch.notify_window = cap->args[0] >> 32;
6362 kvm->arch.notify_vmexit_flags = (u32)cap->args[0];
6365 mutex_unlock(&kvm->lock);
6367 case KVM_CAP_VM_DISABLE_NX_HUGE_PAGES:
6371 * Since the risk of disabling NX hugepages is a guest crashing
6372 * the system, ensure the userspace process has permission to
6373 * reboot the system.
6375 * Note that unlike the reboot() syscall, the process must have
6376 * this capability in the root namespace because exposing
6377 * /dev/kvm into a container does not limit the scope of the
6378 * iTLB multihit bug to that container. In other words,
6379 * this must use capable(), not ns_capable().
6381 if (!capable(CAP_SYS_BOOT)) {
6389 mutex_lock(&kvm->lock);
6390 if (!kvm->created_vcpus) {
6391 kvm->arch.disable_nx_huge_pages = true;
6394 mutex_unlock(&kvm->lock);
6403 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
6405 struct kvm_x86_msr_filter *msr_filter;
6407 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
6411 msr_filter->default_allow = default_allow;
6415 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
6422 for (i = 0; i < msr_filter->count; i++)
6423 kfree(msr_filter->ranges[i].bitmap);
6428 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
6429 struct kvm_msr_filter_range *user_range)
6431 unsigned long *bitmap = NULL;
6434 if (!user_range->nmsrs)
6437 if (user_range->flags & ~KVM_MSR_FILTER_RANGE_VALID_MASK)
6440 if (!user_range->flags)
6443 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
6444 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
6447 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
6449 return PTR_ERR(bitmap);
6451 msr_filter->ranges[msr_filter->count] = (struct msr_bitmap_range) {
6452 .flags = user_range->flags,
6453 .base = user_range->base,
6454 .nmsrs = user_range->nmsrs,
6458 msr_filter->count++;
6462 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm,
6463 struct kvm_msr_filter *filter)
6465 struct kvm_x86_msr_filter *new_filter, *old_filter;
6471 if (filter->flags & ~KVM_MSR_FILTER_VALID_MASK)
6474 for (i = 0; i < ARRAY_SIZE(filter->ranges); i++)
6475 empty &= !filter->ranges[i].nmsrs;
6477 default_allow = !(filter->flags & KVM_MSR_FILTER_DEFAULT_DENY);
6478 if (empty && !default_allow)
6481 new_filter = kvm_alloc_msr_filter(default_allow);
6485 for (i = 0; i < ARRAY_SIZE(filter->ranges); i++) {
6486 r = kvm_add_msr_filter(new_filter, &filter->ranges[i]);
6488 kvm_free_msr_filter(new_filter);
6493 mutex_lock(&kvm->lock);
6495 /* The per-VM filter is protected by kvm->lock... */
6496 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
6498 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
6499 synchronize_srcu(&kvm->srcu);
6501 kvm_free_msr_filter(old_filter);
6503 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
6504 mutex_unlock(&kvm->lock);
6509 #ifdef CONFIG_KVM_COMPAT
6510 /* for KVM_X86_SET_MSR_FILTER */
6511 struct kvm_msr_filter_range_compat {
6518 struct kvm_msr_filter_compat {
6520 struct kvm_msr_filter_range_compat ranges[KVM_MSR_FILTER_MAX_RANGES];
6523 #define KVM_X86_SET_MSR_FILTER_COMPAT _IOW(KVMIO, 0xc6, struct kvm_msr_filter_compat)
6525 long kvm_arch_vm_compat_ioctl(struct file *filp, unsigned int ioctl,
6528 void __user *argp = (void __user *)arg;
6529 struct kvm *kvm = filp->private_data;
6533 case KVM_X86_SET_MSR_FILTER_COMPAT: {
6534 struct kvm_msr_filter __user *user_msr_filter = argp;
6535 struct kvm_msr_filter_compat filter_compat;
6536 struct kvm_msr_filter filter;
6539 if (copy_from_user(&filter_compat, user_msr_filter,
6540 sizeof(filter_compat)))
6543 filter.flags = filter_compat.flags;
6544 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
6545 struct kvm_msr_filter_range_compat *cr;
6547 cr = &filter_compat.ranges[i];
6548 filter.ranges[i] = (struct kvm_msr_filter_range) {
6552 .bitmap = (__u8 *)(ulong)cr->bitmap,
6556 r = kvm_vm_ioctl_set_msr_filter(kvm, &filter);
6565 #ifdef CONFIG_HAVE_KVM_PM_NOTIFIER
6566 static int kvm_arch_suspend_notifier(struct kvm *kvm)
6568 struct kvm_vcpu *vcpu;
6572 mutex_lock(&kvm->lock);
6573 kvm_for_each_vcpu(i, vcpu, kvm) {
6574 if (!vcpu->arch.pv_time.active)
6577 ret = kvm_set_guest_paused(vcpu);
6579 kvm_err("Failed to pause guest VCPU%d: %d\n",
6580 vcpu->vcpu_id, ret);
6584 mutex_unlock(&kvm->lock);
6586 return ret ? NOTIFY_BAD : NOTIFY_DONE;
6589 int kvm_arch_pm_notifier(struct kvm *kvm, unsigned long state)
6592 case PM_HIBERNATION_PREPARE:
6593 case PM_SUSPEND_PREPARE:
6594 return kvm_arch_suspend_notifier(kvm);
6599 #endif /* CONFIG_HAVE_KVM_PM_NOTIFIER */
6601 static int kvm_vm_ioctl_get_clock(struct kvm *kvm, void __user *argp)
6603 struct kvm_clock_data data = { 0 };
6605 get_kvmclock(kvm, &data);
6606 if (copy_to_user(argp, &data, sizeof(data)))
6612 static int kvm_vm_ioctl_set_clock(struct kvm *kvm, void __user *argp)
6614 struct kvm_arch *ka = &kvm->arch;
6615 struct kvm_clock_data data;
6618 if (copy_from_user(&data, argp, sizeof(data)))
6622 * Only KVM_CLOCK_REALTIME is used, but allow passing the
6623 * result of KVM_GET_CLOCK back to KVM_SET_CLOCK.
6625 if (data.flags & ~KVM_CLOCK_VALID_FLAGS)
6628 kvm_hv_request_tsc_page_update(kvm);
6629 kvm_start_pvclock_update(kvm);
6630 pvclock_update_vm_gtod_copy(kvm);
6633 * This pairs with kvm_guest_time_update(): when masterclock is
6634 * in use, we use master_kernel_ns + kvmclock_offset to set
6635 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
6636 * is slightly ahead) here we risk going negative on unsigned
6637 * 'system_time' when 'data.clock' is very small.
6639 if (data.flags & KVM_CLOCK_REALTIME) {
6640 u64 now_real_ns = ktime_get_real_ns();
6643 * Avoid stepping the kvmclock backwards.
6645 if (now_real_ns > data.realtime)
6646 data.clock += now_real_ns - data.realtime;
6649 if (ka->use_master_clock)
6650 now_raw_ns = ka->master_kernel_ns;
6652 now_raw_ns = get_kvmclock_base_ns();
6653 ka->kvmclock_offset = data.clock - now_raw_ns;
6654 kvm_end_pvclock_update(kvm);
6658 long kvm_arch_vm_ioctl(struct file *filp,
6659 unsigned int ioctl, unsigned long arg)
6661 struct kvm *kvm = filp->private_data;
6662 void __user *argp = (void __user *)arg;
6665 * This union makes it completely explicit to gcc-3.x
6666 * that these two variables' stack usage should be
6667 * combined, not added together.
6670 struct kvm_pit_state ps;
6671 struct kvm_pit_state2 ps2;
6672 struct kvm_pit_config pit_config;
6676 case KVM_SET_TSS_ADDR:
6677 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
6679 case KVM_SET_IDENTITY_MAP_ADDR: {
6682 mutex_lock(&kvm->lock);
6684 if (kvm->created_vcpus)
6685 goto set_identity_unlock;
6687 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
6688 goto set_identity_unlock;
6689 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
6690 set_identity_unlock:
6691 mutex_unlock(&kvm->lock);
6694 case KVM_SET_NR_MMU_PAGES:
6695 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
6697 case KVM_GET_NR_MMU_PAGES:
6698 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
6700 case KVM_CREATE_IRQCHIP: {
6701 mutex_lock(&kvm->lock);
6704 if (irqchip_in_kernel(kvm))
6705 goto create_irqchip_unlock;
6708 if (kvm->created_vcpus)
6709 goto create_irqchip_unlock;
6711 r = kvm_pic_init(kvm);
6713 goto create_irqchip_unlock;
6715 r = kvm_ioapic_init(kvm);
6717 kvm_pic_destroy(kvm);
6718 goto create_irqchip_unlock;
6721 r = kvm_setup_default_irq_routing(kvm);
6723 kvm_ioapic_destroy(kvm);
6724 kvm_pic_destroy(kvm);
6725 goto create_irqchip_unlock;
6727 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
6729 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
6730 kvm_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_ABSENT);
6731 create_irqchip_unlock:
6732 mutex_unlock(&kvm->lock);
6735 case KVM_CREATE_PIT:
6736 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
6738 case KVM_CREATE_PIT2:
6740 if (copy_from_user(&u.pit_config, argp,
6741 sizeof(struct kvm_pit_config)))
6744 mutex_lock(&kvm->lock);
6747 goto create_pit_unlock;
6749 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
6753 mutex_unlock(&kvm->lock);
6755 case KVM_GET_IRQCHIP: {
6756 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6757 struct kvm_irqchip *chip;
6759 chip = memdup_user(argp, sizeof(*chip));
6766 if (!irqchip_kernel(kvm))
6767 goto get_irqchip_out;
6768 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
6770 goto get_irqchip_out;
6772 if (copy_to_user(argp, chip, sizeof(*chip)))
6773 goto get_irqchip_out;
6779 case KVM_SET_IRQCHIP: {
6780 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
6781 struct kvm_irqchip *chip;
6783 chip = memdup_user(argp, sizeof(*chip));
6790 if (!irqchip_kernel(kvm))
6791 goto set_irqchip_out;
6792 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
6799 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
6802 if (!kvm->arch.vpit)
6804 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
6808 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
6815 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
6817 mutex_lock(&kvm->lock);
6819 if (!kvm->arch.vpit)
6821 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
6823 mutex_unlock(&kvm->lock);
6826 case KVM_GET_PIT2: {
6828 if (!kvm->arch.vpit)
6830 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
6834 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
6839 case KVM_SET_PIT2: {
6841 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
6843 mutex_lock(&kvm->lock);
6845 if (!kvm->arch.vpit)
6847 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
6849 mutex_unlock(&kvm->lock);
6852 case KVM_REINJECT_CONTROL: {
6853 struct kvm_reinject_control control;
6855 if (copy_from_user(&control, argp, sizeof(control)))
6858 if (!kvm->arch.vpit)
6860 r = kvm_vm_ioctl_reinject(kvm, &control);
6863 case KVM_SET_BOOT_CPU_ID:
6865 mutex_lock(&kvm->lock);
6866 if (kvm->created_vcpus)
6869 kvm->arch.bsp_vcpu_id = arg;
6870 mutex_unlock(&kvm->lock);
6872 #ifdef CONFIG_KVM_XEN
6873 case KVM_XEN_HVM_CONFIG: {
6874 struct kvm_xen_hvm_config xhc;
6876 if (copy_from_user(&xhc, argp, sizeof(xhc)))
6878 r = kvm_xen_hvm_config(kvm, &xhc);
6881 case KVM_XEN_HVM_GET_ATTR: {
6882 struct kvm_xen_hvm_attr xha;
6885 if (copy_from_user(&xha, argp, sizeof(xha)))
6887 r = kvm_xen_hvm_get_attr(kvm, &xha);
6888 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
6892 case KVM_XEN_HVM_SET_ATTR: {
6893 struct kvm_xen_hvm_attr xha;
6896 if (copy_from_user(&xha, argp, sizeof(xha)))
6898 r = kvm_xen_hvm_set_attr(kvm, &xha);
6901 case KVM_XEN_HVM_EVTCHN_SEND: {
6902 struct kvm_irq_routing_xen_evtchn uxe;
6905 if (copy_from_user(&uxe, argp, sizeof(uxe)))
6907 r = kvm_xen_hvm_evtchn_send(kvm, &uxe);
6912 r = kvm_vm_ioctl_set_clock(kvm, argp);
6915 r = kvm_vm_ioctl_get_clock(kvm, argp);
6917 case KVM_SET_TSC_KHZ: {
6921 user_tsc_khz = (u32)arg;
6923 if (kvm_caps.has_tsc_control &&
6924 user_tsc_khz >= kvm_caps.max_guest_tsc_khz)
6927 if (user_tsc_khz == 0)
6928 user_tsc_khz = tsc_khz;
6930 WRITE_ONCE(kvm->arch.default_tsc_khz, user_tsc_khz);
6935 case KVM_GET_TSC_KHZ: {
6936 r = READ_ONCE(kvm->arch.default_tsc_khz);
6939 case KVM_MEMORY_ENCRYPT_OP: {
6941 if (!kvm_x86_ops.mem_enc_ioctl)
6944 r = static_call(kvm_x86_mem_enc_ioctl)(kvm, argp);
6947 case KVM_MEMORY_ENCRYPT_REG_REGION: {
6948 struct kvm_enc_region region;
6951 if (copy_from_user(®ion, argp, sizeof(region)))
6955 if (!kvm_x86_ops.mem_enc_register_region)
6958 r = static_call(kvm_x86_mem_enc_register_region)(kvm, ®ion);
6961 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
6962 struct kvm_enc_region region;
6965 if (copy_from_user(®ion, argp, sizeof(region)))
6969 if (!kvm_x86_ops.mem_enc_unregister_region)
6972 r = static_call(kvm_x86_mem_enc_unregister_region)(kvm, ®ion);
6975 case KVM_HYPERV_EVENTFD: {
6976 struct kvm_hyperv_eventfd hvevfd;
6979 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
6981 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
6984 case KVM_SET_PMU_EVENT_FILTER:
6985 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
6987 case KVM_X86_SET_MSR_FILTER: {
6988 struct kvm_msr_filter __user *user_msr_filter = argp;
6989 struct kvm_msr_filter filter;
6991 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
6994 r = kvm_vm_ioctl_set_msr_filter(kvm, &filter);
7004 static void kvm_init_msr_list(void)
7009 BUILD_BUG_ON_MSG(KVM_PMC_MAX_FIXED != 3,
7010 "Please update the fixed PMCs in msrs_to_saved_all[]");
7012 num_msrs_to_save = 0;
7013 num_emulated_msrs = 0;
7014 num_msr_based_features = 0;
7016 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
7017 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
7021 * Even MSRs that are valid in the host may not be exposed
7022 * to the guests in some cases.
7024 switch (msrs_to_save_all[i]) {
7025 case MSR_IA32_BNDCFGS:
7026 if (!kvm_mpx_supported())
7030 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP) &&
7031 !kvm_cpu_cap_has(X86_FEATURE_RDPID))
7034 case MSR_IA32_UMWAIT_CONTROL:
7035 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
7038 case MSR_IA32_RTIT_CTL:
7039 case MSR_IA32_RTIT_STATUS:
7040 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
7043 case MSR_IA32_RTIT_CR3_MATCH:
7044 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
7045 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
7048 case MSR_IA32_RTIT_OUTPUT_BASE:
7049 case MSR_IA32_RTIT_OUTPUT_MASK:
7050 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
7051 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
7052 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
7055 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
7056 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
7057 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
7058 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
7061 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR_MAX:
7062 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
7063 min(KVM_INTEL_PMC_MAX_GENERIC, kvm_pmu_cap.num_counters_gp))
7066 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL_MAX:
7067 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
7068 min(KVM_INTEL_PMC_MAX_GENERIC, kvm_pmu_cap.num_counters_gp))
7072 case MSR_IA32_XFD_ERR:
7073 if (!kvm_cpu_cap_has(X86_FEATURE_XFD))
7080 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
7083 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
7084 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
7087 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
7090 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
7091 struct kvm_msr_entry msr;
7093 msr.index = msr_based_features_all[i];
7094 if (kvm_get_msr_feature(&msr))
7097 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
7101 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
7109 if (!(lapic_in_kernel(vcpu) &&
7110 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
7111 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
7122 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
7129 if (!(lapic_in_kernel(vcpu) &&
7130 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
7132 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
7134 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
7144 void kvm_set_segment(struct kvm_vcpu *vcpu,
7145 struct kvm_segment *var, int seg)
7147 static_call(kvm_x86_set_segment)(vcpu, var, seg);
7150 void kvm_get_segment(struct kvm_vcpu *vcpu,
7151 struct kvm_segment *var, int seg)
7153 static_call(kvm_x86_get_segment)(vcpu, var, seg);
7156 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u64 access,
7157 struct x86_exception *exception)
7159 struct kvm_mmu *mmu = vcpu->arch.mmu;
7162 BUG_ON(!mmu_is_nested(vcpu));
7164 /* NPT walks are always user-walks */
7165 access |= PFERR_USER_MASK;
7166 t_gpa = mmu->gva_to_gpa(vcpu, mmu, gpa, access, exception);
7171 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
7172 struct x86_exception *exception)
7174 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7176 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7177 return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7179 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
7181 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
7182 struct x86_exception *exception)
7184 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7186 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7187 access |= PFERR_WRITE_MASK;
7188 return mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7190 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
7192 /* uses this to access any guest's mapped memory without checking CPL */
7193 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
7194 struct x86_exception *exception)
7196 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7198 return mmu->gva_to_gpa(vcpu, mmu, gva, 0, exception);
7201 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
7202 struct kvm_vcpu *vcpu, u64 access,
7203 struct x86_exception *exception)
7205 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7207 int r = X86EMUL_CONTINUE;
7210 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
7211 unsigned offset = addr & (PAGE_SIZE-1);
7212 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
7215 if (gpa == INVALID_GPA)
7216 return X86EMUL_PROPAGATE_FAULT;
7217 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
7220 r = X86EMUL_IO_NEEDED;
7232 /* used for instruction fetching */
7233 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
7234 gva_t addr, void *val, unsigned int bytes,
7235 struct x86_exception *exception)
7237 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7238 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7239 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7243 /* Inline kvm_read_guest_virt_helper for speed. */
7244 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access|PFERR_FETCH_MASK,
7246 if (unlikely(gpa == INVALID_GPA))
7247 return X86EMUL_PROPAGATE_FAULT;
7249 offset = addr & (PAGE_SIZE-1);
7250 if (WARN_ON(offset + bytes > PAGE_SIZE))
7251 bytes = (unsigned)PAGE_SIZE - offset;
7252 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
7254 if (unlikely(ret < 0))
7255 return X86EMUL_IO_NEEDED;
7257 return X86EMUL_CONTINUE;
7260 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
7261 gva_t addr, void *val, unsigned int bytes,
7262 struct x86_exception *exception)
7264 u64 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
7267 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
7268 * is returned, but our callers are not ready for that and they blindly
7269 * call kvm_inject_page_fault. Ensure that they at least do not leak
7270 * uninitialized kernel stack memory into cr2 and error code.
7272 memset(exception, 0, sizeof(*exception));
7273 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
7276 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
7278 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
7279 gva_t addr, void *val, unsigned int bytes,
7280 struct x86_exception *exception, bool system)
7282 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7286 access |= PFERR_IMPLICIT_ACCESS;
7287 else if (static_call(kvm_x86_get_cpl)(vcpu) == 3)
7288 access |= PFERR_USER_MASK;
7290 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
7293 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
7294 struct kvm_vcpu *vcpu, u64 access,
7295 struct x86_exception *exception)
7297 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7299 int r = X86EMUL_CONTINUE;
7302 gpa_t gpa = mmu->gva_to_gpa(vcpu, mmu, addr, access, exception);
7303 unsigned offset = addr & (PAGE_SIZE-1);
7304 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
7307 if (gpa == INVALID_GPA)
7308 return X86EMUL_PROPAGATE_FAULT;
7309 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
7311 r = X86EMUL_IO_NEEDED;
7323 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
7324 unsigned int bytes, struct x86_exception *exception,
7327 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7328 u64 access = PFERR_WRITE_MASK;
7331 access |= PFERR_IMPLICIT_ACCESS;
7332 else if (static_call(kvm_x86_get_cpl)(vcpu) == 3)
7333 access |= PFERR_USER_MASK;
7335 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
7339 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
7340 unsigned int bytes, struct x86_exception *exception)
7342 /* kvm_write_guest_virt_system can pull in tons of pages. */
7343 vcpu->arch.l1tf_flush_l1d = true;
7345 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
7346 PFERR_WRITE_MASK, exception);
7348 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
7350 static int kvm_can_emulate_insn(struct kvm_vcpu *vcpu, int emul_type,
7351 void *insn, int insn_len)
7353 return static_call(kvm_x86_can_emulate_instruction)(vcpu, emul_type,
7357 int handle_ud(struct kvm_vcpu *vcpu)
7359 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
7360 int fep_flags = READ_ONCE(force_emulation_prefix);
7361 int emul_type = EMULTYPE_TRAP_UD;
7362 char sig[5]; /* ud2; .ascii "kvm" */
7363 struct x86_exception e;
7365 if (unlikely(!kvm_can_emulate_insn(vcpu, emul_type, NULL, 0)))
7369 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
7370 sig, sizeof(sig), &e) == 0 &&
7371 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
7372 if (fep_flags & KVM_FEP_CLEAR_RFLAGS_RF)
7373 kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) & ~X86_EFLAGS_RF);
7374 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
7375 emul_type = EMULTYPE_TRAP_UD_FORCED;
7378 return kvm_emulate_instruction(vcpu, emul_type);
7380 EXPORT_SYMBOL_GPL(handle_ud);
7382 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
7383 gpa_t gpa, bool write)
7385 /* For APIC access vmexit */
7386 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7389 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
7390 trace_vcpu_match_mmio(gva, gpa, write, true);
7397 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
7398 gpa_t *gpa, struct x86_exception *exception,
7401 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
7402 u64 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
7403 | (write ? PFERR_WRITE_MASK : 0);
7406 * currently PKRU is only applied to ept enabled guest so
7407 * there is no pkey in EPT page table for L1 guest or EPT
7408 * shadow page table for L2 guest.
7410 if (vcpu_match_mmio_gva(vcpu, gva) && (!is_paging(vcpu) ||
7411 !permission_fault(vcpu, vcpu->arch.walk_mmu,
7412 vcpu->arch.mmio_access, 0, access))) {
7413 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
7414 (gva & (PAGE_SIZE - 1));
7415 trace_vcpu_match_mmio(gva, *gpa, write, false);
7419 *gpa = mmu->gva_to_gpa(vcpu, mmu, gva, access, exception);
7421 if (*gpa == INVALID_GPA)
7424 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
7427 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
7428 const void *val, int bytes)
7432 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
7435 kvm_page_track_write(vcpu, gpa, val, bytes);
7439 struct read_write_emulator_ops {
7440 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
7442 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
7443 void *val, int bytes);
7444 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
7445 int bytes, void *val);
7446 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
7447 void *val, int bytes);
7451 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
7453 if (vcpu->mmio_read_completed) {
7454 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
7455 vcpu->mmio_fragments[0].gpa, val);
7456 vcpu->mmio_read_completed = 0;
7463 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
7464 void *val, int bytes)
7466 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
7469 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
7470 void *val, int bytes)
7472 return emulator_write_phys(vcpu, gpa, val, bytes);
7475 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
7477 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
7478 return vcpu_mmio_write(vcpu, gpa, bytes, val);
7481 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
7482 void *val, int bytes)
7484 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
7485 return X86EMUL_IO_NEEDED;
7488 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
7489 void *val, int bytes)
7491 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
7493 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
7494 return X86EMUL_CONTINUE;
7497 static const struct read_write_emulator_ops read_emultor = {
7498 .read_write_prepare = read_prepare,
7499 .read_write_emulate = read_emulate,
7500 .read_write_mmio = vcpu_mmio_read,
7501 .read_write_exit_mmio = read_exit_mmio,
7504 static const struct read_write_emulator_ops write_emultor = {
7505 .read_write_emulate = write_emulate,
7506 .read_write_mmio = write_mmio,
7507 .read_write_exit_mmio = write_exit_mmio,
7511 static int emulator_read_write_onepage(unsigned long addr, void *val,
7513 struct x86_exception *exception,
7514 struct kvm_vcpu *vcpu,
7515 const struct read_write_emulator_ops *ops)
7519 bool write = ops->write;
7520 struct kvm_mmio_fragment *frag;
7521 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7524 * If the exit was due to a NPF we may already have a GPA.
7525 * If the GPA is present, use it to avoid the GVA to GPA table walk.
7526 * Note, this cannot be used on string operations since string
7527 * operation using rep will only have the initial GPA from the NPF
7530 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
7531 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
7532 gpa = ctxt->gpa_val;
7533 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
7535 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
7537 return X86EMUL_PROPAGATE_FAULT;
7540 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
7541 return X86EMUL_CONTINUE;
7544 * Is this MMIO handled locally?
7546 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
7547 if (handled == bytes)
7548 return X86EMUL_CONTINUE;
7554 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
7555 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
7559 return X86EMUL_CONTINUE;
7562 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
7564 void *val, unsigned int bytes,
7565 struct x86_exception *exception,
7566 const struct read_write_emulator_ops *ops)
7568 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7572 if (ops->read_write_prepare &&
7573 ops->read_write_prepare(vcpu, val, bytes))
7574 return X86EMUL_CONTINUE;
7576 vcpu->mmio_nr_fragments = 0;
7578 /* Crossing a page boundary? */
7579 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
7582 now = -addr & ~PAGE_MASK;
7583 rc = emulator_read_write_onepage(addr, val, now, exception,
7586 if (rc != X86EMUL_CONTINUE)
7589 if (ctxt->mode != X86EMUL_MODE_PROT64)
7595 rc = emulator_read_write_onepage(addr, val, bytes, exception,
7597 if (rc != X86EMUL_CONTINUE)
7600 if (!vcpu->mmio_nr_fragments)
7603 gpa = vcpu->mmio_fragments[0].gpa;
7605 vcpu->mmio_needed = 1;
7606 vcpu->mmio_cur_fragment = 0;
7608 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
7609 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
7610 vcpu->run->exit_reason = KVM_EXIT_MMIO;
7611 vcpu->run->mmio.phys_addr = gpa;
7613 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
7616 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
7620 struct x86_exception *exception)
7622 return emulator_read_write(ctxt, addr, val, bytes,
7623 exception, &read_emultor);
7626 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
7630 struct x86_exception *exception)
7632 return emulator_read_write(ctxt, addr, (void *)val, bytes,
7633 exception, &write_emultor);
7636 #define emulator_try_cmpxchg_user(t, ptr, old, new) \
7637 (__try_cmpxchg_user((t __user *)(ptr), (t *)(old), *(t *)(new), efault ## t))
7639 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
7644 struct x86_exception *exception)
7646 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7652 /* guests cmpxchg8b have to be emulated atomically */
7653 if (bytes > 8 || (bytes & (bytes - 1)))
7656 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
7658 if (gpa == INVALID_GPA ||
7659 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
7663 * Emulate the atomic as a straight write to avoid #AC if SLD is
7664 * enabled in the host and the access splits a cache line.
7666 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
7667 page_line_mask = ~(cache_line_size() - 1);
7669 page_line_mask = PAGE_MASK;
7671 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
7674 hva = kvm_vcpu_gfn_to_hva(vcpu, gpa_to_gfn(gpa));
7675 if (kvm_is_error_hva(hva))
7678 hva += offset_in_page(gpa);
7682 r = emulator_try_cmpxchg_user(u8, hva, old, new);
7685 r = emulator_try_cmpxchg_user(u16, hva, old, new);
7688 r = emulator_try_cmpxchg_user(u32, hva, old, new);
7691 r = emulator_try_cmpxchg_user(u64, hva, old, new);
7698 return X86EMUL_UNHANDLEABLE;
7700 return X86EMUL_CMPXCHG_FAILED;
7702 kvm_page_track_write(vcpu, gpa, new, bytes);
7704 return X86EMUL_CONTINUE;
7707 pr_warn_once("emulating exchange as write\n");
7709 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
7712 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
7713 unsigned short port, void *data,
7714 unsigned int count, bool in)
7719 WARN_ON_ONCE(vcpu->arch.pio.count);
7720 for (i = 0; i < count; i++) {
7722 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, port, size, data);
7724 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS, port, size, data);
7731 * Userspace must have unregistered the device while PIO
7732 * was running. Drop writes / read as 0.
7735 memset(data, 0, size * (count - i));
7744 vcpu->arch.pio.port = port;
7745 vcpu->arch.pio.in = in;
7746 vcpu->arch.pio.count = count;
7747 vcpu->arch.pio.size = size;
7750 memset(vcpu->arch.pio_data, 0, size * count);
7752 memcpy(vcpu->arch.pio_data, data, size * count);
7754 vcpu->run->exit_reason = KVM_EXIT_IO;
7755 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
7756 vcpu->run->io.size = size;
7757 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
7758 vcpu->run->io.count = count;
7759 vcpu->run->io.port = port;
7763 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
7764 unsigned short port, void *val, unsigned int count)
7766 int r = emulator_pio_in_out(vcpu, size, port, val, count, true);
7768 trace_kvm_pio(KVM_PIO_IN, port, size, count, val);
7773 static void complete_emulator_pio_in(struct kvm_vcpu *vcpu, void *val)
7775 int size = vcpu->arch.pio.size;
7776 unsigned int count = vcpu->arch.pio.count;
7777 memcpy(val, vcpu->arch.pio_data, size * count);
7778 trace_kvm_pio(KVM_PIO_IN, vcpu->arch.pio.port, size, count, vcpu->arch.pio_data);
7779 vcpu->arch.pio.count = 0;
7782 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
7783 int size, unsigned short port, void *val,
7786 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7787 if (vcpu->arch.pio.count) {
7789 * Complete a previous iteration that required userspace I/O.
7790 * Note, @count isn't guaranteed to match pio.count as userspace
7791 * can modify ECX before rerunning the vCPU. Ignore any such
7792 * shenanigans as KVM doesn't support modifying the rep count,
7793 * and the emulator ensures @count doesn't overflow the buffer.
7795 complete_emulator_pio_in(vcpu, val);
7799 return emulator_pio_in(vcpu, size, port, val, count);
7802 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
7803 unsigned short port, const void *val,
7806 trace_kvm_pio(KVM_PIO_OUT, port, size, count, val);
7807 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
7810 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
7811 int size, unsigned short port,
7812 const void *val, unsigned int count)
7814 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
7817 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
7819 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
7822 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
7824 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
7827 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
7829 if (!need_emulate_wbinvd(vcpu))
7830 return X86EMUL_CONTINUE;
7832 if (static_call(kvm_x86_has_wbinvd_exit)()) {
7833 int cpu = get_cpu();
7835 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
7836 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
7837 wbinvd_ipi, NULL, 1);
7839 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
7842 return X86EMUL_CONTINUE;
7845 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
7847 kvm_emulate_wbinvd_noskip(vcpu);
7848 return kvm_skip_emulated_instruction(vcpu);
7850 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
7854 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
7856 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
7859 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
7860 unsigned long *dest)
7862 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
7865 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
7866 unsigned long value)
7869 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
7872 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
7874 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
7877 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
7879 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7880 unsigned long value;
7884 value = kvm_read_cr0(vcpu);
7887 value = vcpu->arch.cr2;
7890 value = kvm_read_cr3(vcpu);
7893 value = kvm_read_cr4(vcpu);
7896 value = kvm_get_cr8(vcpu);
7899 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7906 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
7908 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7913 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
7916 vcpu->arch.cr2 = val;
7919 res = kvm_set_cr3(vcpu, val);
7922 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
7925 res = kvm_set_cr8(vcpu, val);
7928 kvm_err("%s: unexpected cr %u\n", __func__, cr);
7935 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
7937 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
7940 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7942 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
7945 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7947 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
7950 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7952 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
7955 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
7957 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
7960 static unsigned long emulator_get_cached_segment_base(
7961 struct x86_emulate_ctxt *ctxt, int seg)
7963 return get_segment_base(emul_to_vcpu(ctxt), seg);
7966 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
7967 struct desc_struct *desc, u32 *base3,
7970 struct kvm_segment var;
7972 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
7973 *selector = var.selector;
7976 memset(desc, 0, sizeof(*desc));
7984 set_desc_limit(desc, var.limit);
7985 set_desc_base(desc, (unsigned long)var.base);
7986 #ifdef CONFIG_X86_64
7988 *base3 = var.base >> 32;
7990 desc->type = var.type;
7992 desc->dpl = var.dpl;
7993 desc->p = var.present;
7994 desc->avl = var.avl;
8002 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
8003 struct desc_struct *desc, u32 base3,
8006 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8007 struct kvm_segment var;
8009 var.selector = selector;
8010 var.base = get_desc_base(desc);
8011 #ifdef CONFIG_X86_64
8012 var.base |= ((u64)base3) << 32;
8014 var.limit = get_desc_limit(desc);
8016 var.limit = (var.limit << 12) | 0xfff;
8017 var.type = desc->type;
8018 var.dpl = desc->dpl;
8023 var.avl = desc->avl;
8024 var.present = desc->p;
8025 var.unusable = !var.present;
8028 kvm_set_segment(vcpu, &var, seg);
8032 static int emulator_get_msr_with_filter(struct x86_emulate_ctxt *ctxt,
8033 u32 msr_index, u64 *pdata)
8035 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8038 r = kvm_get_msr_with_filter(vcpu, msr_index, pdata);
8040 return X86EMUL_UNHANDLEABLE;
8043 if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_RDMSR, 0,
8044 complete_emulated_rdmsr, r))
8045 return X86EMUL_IO_NEEDED;
8047 trace_kvm_msr_read_ex(msr_index);
8048 return X86EMUL_PROPAGATE_FAULT;
8051 trace_kvm_msr_read(msr_index, *pdata);
8052 return X86EMUL_CONTINUE;
8055 static int emulator_set_msr_with_filter(struct x86_emulate_ctxt *ctxt,
8056 u32 msr_index, u64 data)
8058 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8061 r = kvm_set_msr_with_filter(vcpu, msr_index, data);
8063 return X86EMUL_UNHANDLEABLE;
8066 if (kvm_msr_user_space(vcpu, msr_index, KVM_EXIT_X86_WRMSR, data,
8067 complete_emulated_msr_access, r))
8068 return X86EMUL_IO_NEEDED;
8070 trace_kvm_msr_write_ex(msr_index, data);
8071 return X86EMUL_PROPAGATE_FAULT;
8074 trace_kvm_msr_write(msr_index, data);
8075 return X86EMUL_CONTINUE;
8078 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
8079 u32 msr_index, u64 *pdata)
8081 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
8084 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
8087 if (kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc))
8092 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
8093 u32 pmc, u64 *pdata)
8095 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
8098 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
8100 emul_to_vcpu(ctxt)->arch.halt_request = 1;
8103 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
8104 struct x86_instruction_info *info,
8105 enum x86_intercept_stage stage)
8107 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
8111 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
8112 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
8115 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
8118 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
8120 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
8123 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
8125 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
8128 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
8130 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
8133 static bool emulator_guest_has_rdpid(struct x86_emulate_ctxt *ctxt)
8135 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_RDPID);
8138 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
8140 return kvm_register_read_raw(emul_to_vcpu(ctxt), reg);
8143 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
8145 kvm_register_write_raw(emul_to_vcpu(ctxt), reg, val);
8148 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
8150 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
8153 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
8155 return emul_to_vcpu(ctxt)->arch.hflags;
8158 #ifndef CONFIG_KVM_SMM
8159 static int emulator_leave_smm(struct x86_emulate_ctxt *ctxt)
8162 return X86EMUL_UNHANDLEABLE;
8166 static void emulator_triple_fault(struct x86_emulate_ctxt *ctxt)
8168 kvm_make_request(KVM_REQ_TRIPLE_FAULT, emul_to_vcpu(ctxt));
8171 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
8173 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
8176 static void emulator_vm_bugged(struct x86_emulate_ctxt *ctxt)
8178 struct kvm *kvm = emul_to_vcpu(ctxt)->kvm;
8180 if (!kvm->vm_bugged)
8184 static const struct x86_emulate_ops emulate_ops = {
8185 .vm_bugged = emulator_vm_bugged,
8186 .read_gpr = emulator_read_gpr,
8187 .write_gpr = emulator_write_gpr,
8188 .read_std = emulator_read_std,
8189 .write_std = emulator_write_std,
8190 .fetch = kvm_fetch_guest_virt,
8191 .read_emulated = emulator_read_emulated,
8192 .write_emulated = emulator_write_emulated,
8193 .cmpxchg_emulated = emulator_cmpxchg_emulated,
8194 .invlpg = emulator_invlpg,
8195 .pio_in_emulated = emulator_pio_in_emulated,
8196 .pio_out_emulated = emulator_pio_out_emulated,
8197 .get_segment = emulator_get_segment,
8198 .set_segment = emulator_set_segment,
8199 .get_cached_segment_base = emulator_get_cached_segment_base,
8200 .get_gdt = emulator_get_gdt,
8201 .get_idt = emulator_get_idt,
8202 .set_gdt = emulator_set_gdt,
8203 .set_idt = emulator_set_idt,
8204 .get_cr = emulator_get_cr,
8205 .set_cr = emulator_set_cr,
8206 .cpl = emulator_get_cpl,
8207 .get_dr = emulator_get_dr,
8208 .set_dr = emulator_set_dr,
8209 .set_msr_with_filter = emulator_set_msr_with_filter,
8210 .get_msr_with_filter = emulator_get_msr_with_filter,
8211 .get_msr = emulator_get_msr,
8212 .check_pmc = emulator_check_pmc,
8213 .read_pmc = emulator_read_pmc,
8214 .halt = emulator_halt,
8215 .wbinvd = emulator_wbinvd,
8216 .fix_hypercall = emulator_fix_hypercall,
8217 .intercept = emulator_intercept,
8218 .get_cpuid = emulator_get_cpuid,
8219 .guest_has_long_mode = emulator_guest_has_long_mode,
8220 .guest_has_movbe = emulator_guest_has_movbe,
8221 .guest_has_fxsr = emulator_guest_has_fxsr,
8222 .guest_has_rdpid = emulator_guest_has_rdpid,
8223 .set_nmi_mask = emulator_set_nmi_mask,
8224 .get_hflags = emulator_get_hflags,
8225 .leave_smm = emulator_leave_smm,
8226 .triple_fault = emulator_triple_fault,
8227 .set_xcr = emulator_set_xcr,
8230 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
8232 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
8234 * an sti; sti; sequence only disable interrupts for the first
8235 * instruction. So, if the last instruction, be it emulated or
8236 * not, left the system with the INT_STI flag enabled, it
8237 * means that the last instruction is an sti. We should not
8238 * leave the flag on in this case. The same goes for mov ss
8240 if (int_shadow & mask)
8242 if (unlikely(int_shadow || mask)) {
8243 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
8245 kvm_make_request(KVM_REQ_EVENT, vcpu);
8249 static void inject_emulated_exception(struct kvm_vcpu *vcpu)
8251 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8253 if (ctxt->exception.vector == PF_VECTOR)
8254 kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
8255 else if (ctxt->exception.error_code_valid)
8256 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
8257 ctxt->exception.error_code);
8259 kvm_queue_exception(vcpu, ctxt->exception.vector);
8262 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
8264 struct x86_emulate_ctxt *ctxt;
8266 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
8268 pr_err("failed to allocate vcpu's emulator\n");
8273 ctxt->ops = &emulate_ops;
8274 vcpu->arch.emulate_ctxt = ctxt;
8279 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
8281 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8284 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
8286 ctxt->gpa_available = false;
8287 ctxt->eflags = kvm_get_rflags(vcpu);
8288 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
8290 ctxt->eip = kvm_rip_read(vcpu);
8291 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
8292 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
8293 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
8294 cs_db ? X86EMUL_MODE_PROT32 :
8295 X86EMUL_MODE_PROT16;
8296 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
8298 ctxt->interruptibility = 0;
8299 ctxt->have_exception = false;
8300 ctxt->exception.vector = -1;
8301 ctxt->perm_ok = false;
8303 init_decode_cache(ctxt);
8304 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8307 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
8309 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8312 init_emulate_ctxt(vcpu);
8316 ctxt->_eip = ctxt->eip + inc_eip;
8317 ret = emulate_int_real(ctxt, irq);
8319 if (ret != X86EMUL_CONTINUE) {
8320 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
8322 ctxt->eip = ctxt->_eip;
8323 kvm_rip_write(vcpu, ctxt->eip);
8324 kvm_set_rflags(vcpu, ctxt->eflags);
8327 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
8329 static void prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
8330 u8 ndata, u8 *insn_bytes, u8 insn_size)
8332 struct kvm_run *run = vcpu->run;
8337 * Zero the whole array used to retrieve the exit info, as casting to
8338 * u32 for select entries will leave some chunks uninitialized.
8340 memset(&info, 0, sizeof(info));
8342 static_call(kvm_x86_get_exit_info)(vcpu, (u32 *)&info[0], &info[1],
8343 &info[2], (u32 *)&info[3],
8346 run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8347 run->emulation_failure.suberror = KVM_INTERNAL_ERROR_EMULATION;
8350 * There's currently space for 13 entries, but 5 are used for the exit
8351 * reason and info. Restrict to 4 to reduce the maintenance burden
8352 * when expanding kvm_run.emulation_failure in the future.
8354 if (WARN_ON_ONCE(ndata > 4))
8357 /* Always include the flags as a 'data' entry. */
8359 run->emulation_failure.flags = 0;
8362 BUILD_BUG_ON((sizeof(run->emulation_failure.insn_size) +
8363 sizeof(run->emulation_failure.insn_bytes) != 16));
8365 run->emulation_failure.flags |=
8366 KVM_INTERNAL_ERROR_EMULATION_FLAG_INSTRUCTION_BYTES;
8367 run->emulation_failure.insn_size = insn_size;
8368 memset(run->emulation_failure.insn_bytes, 0x90,
8369 sizeof(run->emulation_failure.insn_bytes));
8370 memcpy(run->emulation_failure.insn_bytes, insn_bytes, insn_size);
8373 memcpy(&run->internal.data[info_start], info, sizeof(info));
8374 memcpy(&run->internal.data[info_start + ARRAY_SIZE(info)], data,
8375 ndata * sizeof(data[0]));
8377 run->emulation_failure.ndata = info_start + ARRAY_SIZE(info) + ndata;
8380 static void prepare_emulation_ctxt_failure_exit(struct kvm_vcpu *vcpu)
8382 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8384 prepare_emulation_failure_exit(vcpu, NULL, 0, ctxt->fetch.data,
8385 ctxt->fetch.end - ctxt->fetch.data);
8388 void __kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu, u64 *data,
8391 prepare_emulation_failure_exit(vcpu, data, ndata, NULL, 0);
8393 EXPORT_SYMBOL_GPL(__kvm_prepare_emulation_failure_exit);
8395 void kvm_prepare_emulation_failure_exit(struct kvm_vcpu *vcpu)
8397 __kvm_prepare_emulation_failure_exit(vcpu, NULL, 0);
8399 EXPORT_SYMBOL_GPL(kvm_prepare_emulation_failure_exit);
8401 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
8403 struct kvm *kvm = vcpu->kvm;
8405 ++vcpu->stat.insn_emulation_fail;
8406 trace_kvm_emulate_insn_failed(vcpu);
8408 if (emulation_type & EMULTYPE_VMWARE_GP) {
8409 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8413 if (kvm->arch.exit_on_emulation_error ||
8414 (emulation_type & EMULTYPE_SKIP)) {
8415 prepare_emulation_ctxt_failure_exit(vcpu);
8419 kvm_queue_exception(vcpu, UD_VECTOR);
8421 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
8422 prepare_emulation_ctxt_failure_exit(vcpu);
8429 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8430 bool write_fault_to_shadow_pgtable,
8433 gpa_t gpa = cr2_or_gpa;
8436 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
8439 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
8440 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
8443 if (!vcpu->arch.mmu->root_role.direct) {
8445 * Write permission should be allowed since only
8446 * write access need to be emulated.
8448 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
8451 * If the mapping is invalid in guest, let cpu retry
8452 * it to generate fault.
8454 if (gpa == INVALID_GPA)
8459 * Do not retry the unhandleable instruction if it faults on the
8460 * readonly host memory, otherwise it will goto a infinite loop:
8461 * retry instruction -> write #PF -> emulation fail -> retry
8462 * instruction -> ...
8464 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
8467 * If the instruction failed on the error pfn, it can not be fixed,
8468 * report the error to userspace.
8470 if (is_error_noslot_pfn(pfn))
8473 kvm_release_pfn_clean(pfn);
8475 /* The instructions are well-emulated on direct mmu. */
8476 if (vcpu->arch.mmu->root_role.direct) {
8477 unsigned int indirect_shadow_pages;
8479 write_lock(&vcpu->kvm->mmu_lock);
8480 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
8481 write_unlock(&vcpu->kvm->mmu_lock);
8483 if (indirect_shadow_pages)
8484 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8490 * if emulation was due to access to shadowed page table
8491 * and it failed try to unshadow page and re-enter the
8492 * guest to let CPU execute the instruction.
8494 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8497 * If the access faults on its page table, it can not
8498 * be fixed by unprotecting shadow page and it should
8499 * be reported to userspace.
8501 return !write_fault_to_shadow_pgtable;
8504 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
8505 gpa_t cr2_or_gpa, int emulation_type)
8507 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8508 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
8510 last_retry_eip = vcpu->arch.last_retry_eip;
8511 last_retry_addr = vcpu->arch.last_retry_addr;
8514 * If the emulation is caused by #PF and it is non-page_table
8515 * writing instruction, it means the VM-EXIT is caused by shadow
8516 * page protected, we can zap the shadow page and retry this
8517 * instruction directly.
8519 * Note: if the guest uses a non-page-table modifying instruction
8520 * on the PDE that points to the instruction, then we will unmap
8521 * the instruction and go to an infinite loop. So, we cache the
8522 * last retried eip and the last fault address, if we meet the eip
8523 * and the address again, we can break out of the potential infinite
8526 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
8528 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
8531 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
8532 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
8535 if (x86_page_table_writing_insn(ctxt))
8538 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
8541 vcpu->arch.last_retry_eip = ctxt->eip;
8542 vcpu->arch.last_retry_addr = cr2_or_gpa;
8544 if (!vcpu->arch.mmu->root_role.direct)
8545 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
8547 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
8552 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
8553 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
8555 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
8564 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
8565 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
8570 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
8572 struct kvm_run *kvm_run = vcpu->run;
8574 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
8575 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
8576 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
8577 kvm_run->debug.arch.exception = DB_VECTOR;
8578 kvm_run->exit_reason = KVM_EXIT_DEBUG;
8581 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
8585 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
8587 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8590 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
8594 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8597 * rflags is the old, "raw" value of the flags. The new value has
8598 * not been saved yet.
8600 * This is correct even for TF set by the guest, because "the
8601 * processor will not generate this exception after the instruction
8602 * that sets the TF flag".
8604 if (unlikely(rflags & X86_EFLAGS_TF))
8605 r = kvm_vcpu_do_singlestep(vcpu);
8608 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
8610 static bool kvm_is_code_breakpoint_inhibited(struct kvm_vcpu *vcpu)
8614 if (kvm_get_rflags(vcpu) & X86_EFLAGS_RF)
8618 * Intel CPUs inhibit code #DBs when MOV/POP SS blocking is active,
8619 * but AMD CPUs do not. MOV/POP SS blocking is rare, check that first
8620 * to avoid the relatively expensive CPUID lookup.
8622 shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
8623 return (shadow & KVM_X86_SHADOW_INT_MOV_SS) &&
8624 guest_cpuid_is_intel(vcpu);
8627 static bool kvm_vcpu_check_code_breakpoint(struct kvm_vcpu *vcpu,
8628 int emulation_type, int *r)
8630 WARN_ON_ONCE(emulation_type & EMULTYPE_NO_DECODE);
8633 * Do not check for code breakpoints if hardware has already done the
8634 * checks, as inferred from the emulation type. On NO_DECODE and SKIP,
8635 * the instruction has passed all exception checks, and all intercepted
8636 * exceptions that trigger emulation have lower priority than code
8637 * breakpoints, i.e. the fact that the intercepted exception occurred
8638 * means any code breakpoints have already been serviced.
8640 * Note, KVM needs to check for code #DBs on EMULTYPE_TRAP_UD_FORCED as
8641 * hardware has checked the RIP of the magic prefix, but not the RIP of
8642 * the instruction being emulated. The intent of forced emulation is
8643 * to behave as if KVM intercepted the instruction without an exception
8644 * and without a prefix.
8646 if (emulation_type & (EMULTYPE_NO_DECODE | EMULTYPE_SKIP |
8647 EMULTYPE_TRAP_UD | EMULTYPE_VMWARE_GP | EMULTYPE_PF))
8650 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
8651 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
8652 struct kvm_run *kvm_run = vcpu->run;
8653 unsigned long eip = kvm_get_linear_rip(vcpu);
8654 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8655 vcpu->arch.guest_debug_dr7,
8659 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
8660 kvm_run->debug.arch.pc = eip;
8661 kvm_run->debug.arch.exception = DB_VECTOR;
8662 kvm_run->exit_reason = KVM_EXIT_DEBUG;
8668 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
8669 !kvm_is_code_breakpoint_inhibited(vcpu)) {
8670 unsigned long eip = kvm_get_linear_rip(vcpu);
8671 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
8676 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
8685 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
8687 switch (ctxt->opcode_len) {
8694 case 0xe6: /* OUT */
8698 case 0x6c: /* INS */
8700 case 0x6e: /* OUTS */
8707 case 0x33: /* RDPMC */
8717 * Decode an instruction for emulation. The caller is responsible for handling
8718 * code breakpoints. Note, manually detecting code breakpoints is unnecessary
8719 * (and wrong) when emulating on an intercepted fault-like exception[*], as
8720 * code breakpoints have higher priority and thus have already been done by
8723 * [*] Except #MC, which is higher priority, but KVM should never emulate in
8724 * response to a machine check.
8726 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
8727 void *insn, int insn_len)
8729 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8732 init_emulate_ctxt(vcpu);
8734 r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
8736 trace_kvm_emulate_insn_start(vcpu);
8737 ++vcpu->stat.insn_emulation;
8741 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
8743 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
8744 int emulation_type, void *insn, int insn_len)
8747 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
8748 bool writeback = true;
8749 bool write_fault_to_spt;
8751 if (unlikely(!kvm_can_emulate_insn(vcpu, emulation_type, insn, insn_len)))
8754 vcpu->arch.l1tf_flush_l1d = true;
8757 * Clear write_fault_to_shadow_pgtable here to ensure it is
8760 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
8761 vcpu->arch.write_fault_to_shadow_pgtable = false;
8763 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
8764 kvm_clear_exception_queue(vcpu);
8767 * Return immediately if RIP hits a code breakpoint, such #DBs
8768 * are fault-like and are higher priority than any faults on
8769 * the code fetch itself.
8771 if (kvm_vcpu_check_code_breakpoint(vcpu, emulation_type, &r))
8774 r = x86_decode_emulated_instruction(vcpu, emulation_type,
8776 if (r != EMULATION_OK) {
8777 if ((emulation_type & EMULTYPE_TRAP_UD) ||
8778 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
8779 kvm_queue_exception(vcpu, UD_VECTOR);
8782 if (reexecute_instruction(vcpu, cr2_or_gpa,
8787 if (ctxt->have_exception &&
8788 !(emulation_type & EMULTYPE_SKIP)) {
8790 * #UD should result in just EMULATION_FAILED, and trap-like
8791 * exception should not be encountered during decode.
8793 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
8794 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
8795 inject_emulated_exception(vcpu);
8798 return handle_emulation_failure(vcpu, emulation_type);
8802 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
8803 !is_vmware_backdoor_opcode(ctxt)) {
8804 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
8809 * EMULTYPE_SKIP without EMULTYPE_COMPLETE_USER_EXIT is intended for
8810 * use *only* by vendor callbacks for kvm_skip_emulated_instruction().
8811 * The caller is responsible for updating interruptibility state and
8812 * injecting single-step #DBs.
8814 if (emulation_type & EMULTYPE_SKIP) {
8815 if (ctxt->mode != X86EMUL_MODE_PROT64)
8816 ctxt->eip = (u32)ctxt->_eip;
8818 ctxt->eip = ctxt->_eip;
8820 if (emulation_type & EMULTYPE_COMPLETE_USER_EXIT) {
8825 kvm_rip_write(vcpu, ctxt->eip);
8826 if (ctxt->eflags & X86_EFLAGS_RF)
8827 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
8831 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
8834 /* this is needed for vmware backdoor interface to work since it
8835 changes registers values during IO operation */
8836 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
8837 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
8838 emulator_invalidate_register_cache(ctxt);
8842 if (emulation_type & EMULTYPE_PF) {
8843 /* Save the faulting GPA (cr2) in the address field */
8844 ctxt->exception.address = cr2_or_gpa;
8846 /* With shadow page tables, cr2 contains a GVA or nGPA. */
8847 if (vcpu->arch.mmu->root_role.direct) {
8848 ctxt->gpa_available = true;
8849 ctxt->gpa_val = cr2_or_gpa;
8852 /* Sanitize the address out of an abundance of paranoia. */
8853 ctxt->exception.address = 0;
8856 r = x86_emulate_insn(ctxt);
8858 if (r == EMULATION_INTERCEPTED)
8861 if (r == EMULATION_FAILED) {
8862 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
8866 return handle_emulation_failure(vcpu, emulation_type);
8869 if (ctxt->have_exception) {
8871 inject_emulated_exception(vcpu);
8872 } else if (vcpu->arch.pio.count) {
8873 if (!vcpu->arch.pio.in) {
8874 /* FIXME: return into emulator if single-stepping. */
8875 vcpu->arch.pio.count = 0;
8878 vcpu->arch.complete_userspace_io = complete_emulated_pio;
8881 } else if (vcpu->mmio_needed) {
8882 ++vcpu->stat.mmio_exits;
8884 if (!vcpu->mmio_is_write)
8887 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8888 } else if (vcpu->arch.complete_userspace_io) {
8891 } else if (r == EMULATION_RESTART)
8898 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
8899 toggle_interruptibility(vcpu, ctxt->interruptibility);
8900 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8903 * Note, EXCPT_DB is assumed to be fault-like as the emulator
8904 * only supports code breakpoints and general detect #DB, both
8905 * of which are fault-like.
8907 if (!ctxt->have_exception ||
8908 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
8909 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_INSTRUCTIONS);
8910 if (ctxt->is_branch)
8911 kvm_pmu_trigger_event(vcpu, PERF_COUNT_HW_BRANCH_INSTRUCTIONS);
8912 kvm_rip_write(vcpu, ctxt->eip);
8913 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
8914 r = kvm_vcpu_do_singlestep(vcpu);
8915 static_call_cond(kvm_x86_update_emulated_instruction)(vcpu);
8916 __kvm_set_rflags(vcpu, ctxt->eflags);
8920 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
8921 * do nothing, and it will be requested again as soon as
8922 * the shadow expires. But we still need to check here,
8923 * because POPF has no interrupt shadow.
8925 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
8926 kvm_make_request(KVM_REQ_EVENT, vcpu);
8928 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
8933 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
8935 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
8937 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
8939 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
8940 void *insn, int insn_len)
8942 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
8944 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
8946 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
8948 vcpu->arch.pio.count = 0;
8952 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
8954 vcpu->arch.pio.count = 0;
8956 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
8959 return kvm_skip_emulated_instruction(vcpu);
8962 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
8963 unsigned short port)
8965 unsigned long val = kvm_rax_read(vcpu);
8966 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
8972 * Workaround userspace that relies on old KVM behavior of %rip being
8973 * incremented prior to exiting to userspace to handle "OUT 0x7e".
8976 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
8977 vcpu->arch.complete_userspace_io =
8978 complete_fast_pio_out_port_0x7e;
8979 kvm_skip_emulated_instruction(vcpu);
8981 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
8982 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
8987 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
8991 /* We should only ever be called with arch.pio.count equal to 1 */
8992 BUG_ON(vcpu->arch.pio.count != 1);
8994 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
8995 vcpu->arch.pio.count = 0;
8999 /* For size less than 4 we merge, else we zero extend */
9000 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
9002 complete_emulator_pio_in(vcpu, &val);
9003 kvm_rax_write(vcpu, val);
9005 return kvm_skip_emulated_instruction(vcpu);
9008 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
9009 unsigned short port)
9014 /* For size less than 4 we merge, else we zero extend */
9015 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
9017 ret = emulator_pio_in(vcpu, size, port, &val, 1);
9019 kvm_rax_write(vcpu, val);
9023 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
9024 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
9029 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
9034 ret = kvm_fast_pio_in(vcpu, size, port);
9036 ret = kvm_fast_pio_out(vcpu, size, port);
9037 return ret && kvm_skip_emulated_instruction(vcpu);
9039 EXPORT_SYMBOL_GPL(kvm_fast_pio);
9041 static int kvmclock_cpu_down_prep(unsigned int cpu)
9043 __this_cpu_write(cpu_tsc_khz, 0);
9047 static void tsc_khz_changed(void *data)
9049 struct cpufreq_freqs *freq = data;
9050 unsigned long khz = 0;
9052 WARN_ON_ONCE(boot_cpu_has(X86_FEATURE_CONSTANT_TSC));
9057 khz = cpufreq_quick_get(raw_smp_processor_id());
9060 __this_cpu_write(cpu_tsc_khz, khz);
9063 #ifdef CONFIG_X86_64
9064 static void kvm_hyperv_tsc_notifier(void)
9069 mutex_lock(&kvm_lock);
9070 list_for_each_entry(kvm, &vm_list, vm_list)
9071 kvm_make_mclock_inprogress_request(kvm);
9073 /* no guest entries from this point */
9074 hyperv_stop_tsc_emulation();
9076 /* TSC frequency always matches when on Hyper-V */
9077 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9078 for_each_present_cpu(cpu)
9079 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
9081 kvm_caps.max_guest_tsc_khz = tsc_khz;
9083 list_for_each_entry(kvm, &vm_list, vm_list) {
9084 __kvm_start_pvclock_update(kvm);
9085 pvclock_update_vm_gtod_copy(kvm);
9086 kvm_end_pvclock_update(kvm);
9089 mutex_unlock(&kvm_lock);
9093 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
9096 struct kvm_vcpu *vcpu;
9101 * We allow guests to temporarily run on slowing clocks,
9102 * provided we notify them after, or to run on accelerating
9103 * clocks, provided we notify them before. Thus time never
9106 * However, we have a problem. We can't atomically update
9107 * the frequency of a given CPU from this function; it is
9108 * merely a notifier, which can be called from any CPU.
9109 * Changing the TSC frequency at arbitrary points in time
9110 * requires a recomputation of local variables related to
9111 * the TSC for each VCPU. We must flag these local variables
9112 * to be updated and be sure the update takes place with the
9113 * new frequency before any guests proceed.
9115 * Unfortunately, the combination of hotplug CPU and frequency
9116 * change creates an intractable locking scenario; the order
9117 * of when these callouts happen is undefined with respect to
9118 * CPU hotplug, and they can race with each other. As such,
9119 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
9120 * undefined; you can actually have a CPU frequency change take
9121 * place in between the computation of X and the setting of the
9122 * variable. To protect against this problem, all updates of
9123 * the per_cpu tsc_khz variable are done in an interrupt
9124 * protected IPI, and all callers wishing to update the value
9125 * must wait for a synchronous IPI to complete (which is trivial
9126 * if the caller is on the CPU already). This establishes the
9127 * necessary total order on variable updates.
9129 * Note that because a guest time update may take place
9130 * anytime after the setting of the VCPU's request bit, the
9131 * correct TSC value must be set before the request. However,
9132 * to ensure the update actually makes it to any guest which
9133 * starts running in hardware virtualization between the set
9134 * and the acquisition of the spinlock, we must also ping the
9135 * CPU after setting the request bit.
9139 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
9141 mutex_lock(&kvm_lock);
9142 list_for_each_entry(kvm, &vm_list, vm_list) {
9143 kvm_for_each_vcpu(i, vcpu, kvm) {
9144 if (vcpu->cpu != cpu)
9146 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9147 if (vcpu->cpu != raw_smp_processor_id())
9151 mutex_unlock(&kvm_lock);
9153 if (freq->old < freq->new && send_ipi) {
9155 * We upscale the frequency. Must make the guest
9156 * doesn't see old kvmclock values while running with
9157 * the new frequency, otherwise we risk the guest sees
9158 * time go backwards.
9160 * In case we update the frequency for another cpu
9161 * (which might be in guest context) send an interrupt
9162 * to kick the cpu out of guest context. Next time
9163 * guest context is entered kvmclock will be updated,
9164 * so the guest will not see stale values.
9166 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
9170 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
9173 struct cpufreq_freqs *freq = data;
9176 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
9178 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
9181 for_each_cpu(cpu, freq->policy->cpus)
9182 __kvmclock_cpufreq_notifier(freq, cpu);
9187 static struct notifier_block kvmclock_cpufreq_notifier_block = {
9188 .notifier_call = kvmclock_cpufreq_notifier
9191 static int kvmclock_cpu_online(unsigned int cpu)
9193 tsc_khz_changed(NULL);
9197 static void kvm_timer_init(void)
9199 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9200 max_tsc_khz = tsc_khz;
9202 if (IS_ENABLED(CONFIG_CPU_FREQ)) {
9203 struct cpufreq_policy *policy;
9207 policy = cpufreq_cpu_get(cpu);
9209 if (policy->cpuinfo.max_freq)
9210 max_tsc_khz = policy->cpuinfo.max_freq;
9211 cpufreq_cpu_put(policy);
9215 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
9216 CPUFREQ_TRANSITION_NOTIFIER);
9218 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
9219 kvmclock_cpu_online, kvmclock_cpu_down_prep);
9223 #ifdef CONFIG_X86_64
9224 static void pvclock_gtod_update_fn(struct work_struct *work)
9227 struct kvm_vcpu *vcpu;
9230 mutex_lock(&kvm_lock);
9231 list_for_each_entry(kvm, &vm_list, vm_list)
9232 kvm_for_each_vcpu(i, vcpu, kvm)
9233 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9234 atomic_set(&kvm_guest_has_master_clock, 0);
9235 mutex_unlock(&kvm_lock);
9238 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
9241 * Indirection to move queue_work() out of the tk_core.seq write held
9242 * region to prevent possible deadlocks against time accessors which
9243 * are invoked with work related locks held.
9245 static void pvclock_irq_work_fn(struct irq_work *w)
9247 queue_work(system_long_wq, &pvclock_gtod_work);
9250 static DEFINE_IRQ_WORK(pvclock_irq_work, pvclock_irq_work_fn);
9253 * Notification about pvclock gtod data update.
9255 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
9258 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
9259 struct timekeeper *tk = priv;
9261 update_pvclock_gtod(tk);
9264 * Disable master clock if host does not trust, or does not use,
9265 * TSC based clocksource. Delegate queue_work() to irq_work as
9266 * this is invoked with tk_core.seq write held.
9268 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
9269 atomic_read(&kvm_guest_has_master_clock) != 0)
9270 irq_work_queue(&pvclock_irq_work);
9274 static struct notifier_block pvclock_gtod_notifier = {
9275 .notifier_call = pvclock_gtod_notify,
9279 static inline void kvm_ops_update(struct kvm_x86_init_ops *ops)
9281 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
9283 #define __KVM_X86_OP(func) \
9284 static_call_update(kvm_x86_##func, kvm_x86_ops.func);
9285 #define KVM_X86_OP(func) \
9286 WARN_ON(!kvm_x86_ops.func); __KVM_X86_OP(func)
9287 #define KVM_X86_OP_OPTIONAL __KVM_X86_OP
9288 #define KVM_X86_OP_OPTIONAL_RET0(func) \
9289 static_call_update(kvm_x86_##func, (void *)kvm_x86_ops.func ? : \
9290 (void *)__static_call_return0);
9291 #include <asm/kvm-x86-ops.h>
9294 kvm_pmu_ops_update(ops->pmu_ops);
9297 static int kvm_x86_check_processor_compatibility(void)
9299 int cpu = smp_processor_id();
9300 struct cpuinfo_x86 *c = &cpu_data(cpu);
9303 * Compatibility checks are done when loading KVM and when enabling
9304 * hardware, e.g. during CPU hotplug, to ensure all online CPUs are
9305 * compatible, i.e. KVM should never perform a compatibility check on
9308 WARN_ON(!cpu_online(cpu));
9310 if (__cr4_reserved_bits(cpu_has, c) !=
9311 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
9314 return static_call(kvm_x86_check_processor_compatibility)();
9317 static void kvm_x86_check_cpu_compat(void *ret)
9319 *(int *)ret = kvm_x86_check_processor_compatibility();
9322 static int __kvm_x86_vendor_init(struct kvm_x86_init_ops *ops)
9327 if (kvm_x86_ops.hardware_enable) {
9328 pr_err("already loaded vendor module '%s'\n", kvm_x86_ops.name);
9333 * KVM explicitly assumes that the guest has an FPU and
9334 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
9335 * vCPU's FPU state as a fxregs_state struct.
9337 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
9338 pr_err("inadequate fpu\n");
9342 if (IS_ENABLED(CONFIG_PREEMPT_RT) && !boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9343 pr_err("RT requires X86_FEATURE_CONSTANT_TSC\n");
9348 * KVM assumes that PAT entry '0' encodes WB memtype and simply zeroes
9349 * the PAT bits in SPTEs. Bail if PAT[0] is programmed to something
9350 * other than WB. Note, EPT doesn't utilize the PAT, but don't bother
9351 * with an exception. PAT[0] is set to WB on RESET and also by the
9352 * kernel, i.e. failure indicates a kernel bug or broken firmware.
9354 if (rdmsrl_safe(MSR_IA32_CR_PAT, &host_pat) ||
9355 (host_pat & GENMASK(2, 0)) != 6) {
9356 pr_err("host PAT[0] is not WB\n");
9360 x86_emulator_cache = kvm_alloc_emulator_cache();
9361 if (!x86_emulator_cache) {
9362 pr_err("failed to allocate cache for x86 emulator\n");
9366 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
9367 if (!user_return_msrs) {
9368 pr_err("failed to allocate percpu kvm_user_return_msrs\n");
9370 goto out_free_x86_emulator_cache;
9372 kvm_nr_uret_msrs = 0;
9374 r = kvm_mmu_vendor_module_init();
9376 goto out_free_percpu;
9378 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
9379 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
9380 kvm_caps.supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
9383 rdmsrl_safe(MSR_EFER, &host_efer);
9385 if (boot_cpu_has(X86_FEATURE_XSAVES))
9386 rdmsrl(MSR_IA32_XSS, host_xss);
9388 kvm_init_pmu_capability();
9390 r = ops->hardware_setup();
9394 kvm_ops_update(ops);
9396 for_each_online_cpu(cpu) {
9397 smp_call_function_single(cpu, kvm_x86_check_cpu_compat, &r, 1);
9399 goto out_unwind_ops;
9403 * Point of no return! DO NOT add error paths below this point unless
9404 * absolutely necessary, as most operations from this point forward
9405 * require unwinding.
9409 if (pi_inject_timer == -1)
9410 pi_inject_timer = housekeeping_enabled(HK_TYPE_TIMER);
9411 #ifdef CONFIG_X86_64
9412 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
9414 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
9415 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
9418 kvm_register_perf_callbacks(ops->handle_intel_pt_intr);
9420 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
9421 kvm_caps.supported_xss = 0;
9423 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
9424 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
9425 #undef __kvm_cpu_cap_has
9427 if (kvm_caps.has_tsc_control) {
9429 * Make sure the user can only configure tsc_khz values that
9430 * fit into a signed integer.
9431 * A min value is not calculated because it will always
9432 * be 1 on all machines.
9434 u64 max = min(0x7fffffffULL,
9435 __scale_tsc(kvm_caps.max_tsc_scaling_ratio, tsc_khz));
9436 kvm_caps.max_guest_tsc_khz = max;
9438 kvm_caps.default_tsc_scaling_ratio = 1ULL << kvm_caps.tsc_scaling_ratio_frac_bits;
9439 kvm_init_msr_list();
9443 kvm_x86_ops.hardware_enable = NULL;
9444 static_call(kvm_x86_hardware_unsetup)();
9446 kvm_mmu_vendor_module_exit();
9448 free_percpu(user_return_msrs);
9449 out_free_x86_emulator_cache:
9450 kmem_cache_destroy(x86_emulator_cache);
9454 int kvm_x86_vendor_init(struct kvm_x86_init_ops *ops)
9458 mutex_lock(&vendor_module_lock);
9459 r = __kvm_x86_vendor_init(ops);
9460 mutex_unlock(&vendor_module_lock);
9464 EXPORT_SYMBOL_GPL(kvm_x86_vendor_init);
9466 void kvm_x86_vendor_exit(void)
9468 kvm_unregister_perf_callbacks();
9470 #ifdef CONFIG_X86_64
9471 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
9472 clear_hv_tscchange_cb();
9476 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
9477 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
9478 CPUFREQ_TRANSITION_NOTIFIER);
9479 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
9481 #ifdef CONFIG_X86_64
9482 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
9483 irq_work_sync(&pvclock_irq_work);
9484 cancel_work_sync(&pvclock_gtod_work);
9486 static_call(kvm_x86_hardware_unsetup)();
9487 kvm_mmu_vendor_module_exit();
9488 free_percpu(user_return_msrs);
9489 kmem_cache_destroy(x86_emulator_cache);
9490 #ifdef CONFIG_KVM_XEN
9491 static_key_deferred_flush(&kvm_xen_enabled);
9492 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
9494 mutex_lock(&vendor_module_lock);
9495 kvm_x86_ops.hardware_enable = NULL;
9496 mutex_unlock(&vendor_module_lock);
9498 EXPORT_SYMBOL_GPL(kvm_x86_vendor_exit);
9500 static int __kvm_emulate_halt(struct kvm_vcpu *vcpu, int state, int reason)
9503 * The vCPU has halted, e.g. executed HLT. Update the run state if the
9504 * local APIC is in-kernel, the run loop will detect the non-runnable
9505 * state and halt the vCPU. Exit to userspace if the local APIC is
9506 * managed by userspace, in which case userspace is responsible for
9507 * handling wake events.
9509 ++vcpu->stat.halt_exits;
9510 if (lapic_in_kernel(vcpu)) {
9511 vcpu->arch.mp_state = state;
9514 vcpu->run->exit_reason = reason;
9519 int kvm_emulate_halt_noskip(struct kvm_vcpu *vcpu)
9521 return __kvm_emulate_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
9523 EXPORT_SYMBOL_GPL(kvm_emulate_halt_noskip);
9525 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
9527 int ret = kvm_skip_emulated_instruction(vcpu);
9529 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
9530 * KVM_EXIT_DEBUG here.
9532 return kvm_emulate_halt_noskip(vcpu) && ret;
9534 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
9536 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
9538 int ret = kvm_skip_emulated_instruction(vcpu);
9540 return __kvm_emulate_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD,
9541 KVM_EXIT_AP_RESET_HOLD) && ret;
9543 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
9545 #ifdef CONFIG_X86_64
9546 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
9547 unsigned long clock_type)
9549 struct kvm_clock_pairing clock_pairing;
9550 struct timespec64 ts;
9554 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
9555 return -KVM_EOPNOTSUPP;
9558 * When tsc is in permanent catchup mode guests won't be able to use
9559 * pvclock_read_retry loop to get consistent view of pvclock
9561 if (vcpu->arch.tsc_always_catchup)
9562 return -KVM_EOPNOTSUPP;
9564 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
9565 return -KVM_EOPNOTSUPP;
9567 clock_pairing.sec = ts.tv_sec;
9568 clock_pairing.nsec = ts.tv_nsec;
9569 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
9570 clock_pairing.flags = 0;
9571 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
9574 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
9575 sizeof(struct kvm_clock_pairing)))
9583 * kvm_pv_kick_cpu_op: Kick a vcpu.
9585 * @apicid - apicid of vcpu to be kicked.
9587 static void kvm_pv_kick_cpu_op(struct kvm *kvm, int apicid)
9590 * All other fields are unused for APIC_DM_REMRD, but may be consumed by
9591 * common code, e.g. for tracing. Defer initialization to the compiler.
9593 struct kvm_lapic_irq lapic_irq = {
9594 .delivery_mode = APIC_DM_REMRD,
9595 .dest_mode = APIC_DEST_PHYSICAL,
9596 .shorthand = APIC_DEST_NOSHORT,
9600 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
9603 bool kvm_apicv_activated(struct kvm *kvm)
9605 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
9607 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
9609 bool kvm_vcpu_apicv_activated(struct kvm_vcpu *vcpu)
9611 ulong vm_reasons = READ_ONCE(vcpu->kvm->arch.apicv_inhibit_reasons);
9612 ulong vcpu_reasons = static_call(kvm_x86_vcpu_get_apicv_inhibit_reasons)(vcpu);
9614 return (vm_reasons | vcpu_reasons) == 0;
9616 EXPORT_SYMBOL_GPL(kvm_vcpu_apicv_activated);
9618 static void set_or_clear_apicv_inhibit(unsigned long *inhibits,
9619 enum kvm_apicv_inhibit reason, bool set)
9622 __set_bit(reason, inhibits);
9624 __clear_bit(reason, inhibits);
9626 trace_kvm_apicv_inhibit_changed(reason, set, *inhibits);
9629 static void kvm_apicv_init(struct kvm *kvm)
9631 unsigned long *inhibits = &kvm->arch.apicv_inhibit_reasons;
9633 init_rwsem(&kvm->arch.apicv_update_lock);
9635 set_or_clear_apicv_inhibit(inhibits, APICV_INHIBIT_REASON_ABSENT, true);
9638 set_or_clear_apicv_inhibit(inhibits,
9639 APICV_INHIBIT_REASON_DISABLE, true);
9642 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
9644 struct kvm_vcpu *target = NULL;
9645 struct kvm_apic_map *map;
9647 vcpu->stat.directed_yield_attempted++;
9649 if (single_task_running())
9653 map = rcu_dereference(vcpu->kvm->arch.apic_map);
9655 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
9656 target = map->phys_map[dest_id]->vcpu;
9660 if (!target || !READ_ONCE(target->ready))
9663 /* Ignore requests to yield to self */
9667 if (kvm_vcpu_yield_to(target) <= 0)
9670 vcpu->stat.directed_yield_successful++;
9676 static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
9678 u64 ret = vcpu->run->hypercall.ret;
9680 if (!is_64_bit_mode(vcpu))
9682 kvm_rax_write(vcpu, ret);
9683 ++vcpu->stat.hypercalls;
9684 return kvm_skip_emulated_instruction(vcpu);
9687 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
9689 unsigned long nr, a0, a1, a2, a3, ret;
9692 if (kvm_xen_hypercall_enabled(vcpu->kvm))
9693 return kvm_xen_hypercall(vcpu);
9695 if (kvm_hv_hypercall_enabled(vcpu))
9696 return kvm_hv_hypercall(vcpu);
9698 nr = kvm_rax_read(vcpu);
9699 a0 = kvm_rbx_read(vcpu);
9700 a1 = kvm_rcx_read(vcpu);
9701 a2 = kvm_rdx_read(vcpu);
9702 a3 = kvm_rsi_read(vcpu);
9704 trace_kvm_hypercall(nr, a0, a1, a2, a3);
9706 op_64_bit = is_64_bit_hypercall(vcpu);
9715 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
9723 case KVM_HC_VAPIC_POLL_IRQ:
9726 case KVM_HC_KICK_CPU:
9727 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
9730 kvm_pv_kick_cpu_op(vcpu->kvm, a1);
9731 kvm_sched_yield(vcpu, a1);
9734 #ifdef CONFIG_X86_64
9735 case KVM_HC_CLOCK_PAIRING:
9736 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
9739 case KVM_HC_SEND_IPI:
9740 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
9743 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
9745 case KVM_HC_SCHED_YIELD:
9746 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
9749 kvm_sched_yield(vcpu, a0);
9752 case KVM_HC_MAP_GPA_RANGE: {
9753 u64 gpa = a0, npages = a1, attrs = a2;
9756 if (!(vcpu->kvm->arch.hypercall_exit_enabled & (1 << KVM_HC_MAP_GPA_RANGE)))
9759 if (!PAGE_ALIGNED(gpa) || !npages ||
9760 gpa_to_gfn(gpa) + npages <= gpa_to_gfn(gpa)) {
9765 vcpu->run->exit_reason = KVM_EXIT_HYPERCALL;
9766 vcpu->run->hypercall.nr = KVM_HC_MAP_GPA_RANGE;
9767 vcpu->run->hypercall.args[0] = gpa;
9768 vcpu->run->hypercall.args[1] = npages;
9769 vcpu->run->hypercall.args[2] = attrs;
9770 vcpu->run->hypercall.longmode = op_64_bit;
9771 vcpu->arch.complete_userspace_io = complete_hypercall_exit;
9781 kvm_rax_write(vcpu, ret);
9783 ++vcpu->stat.hypercalls;
9784 return kvm_skip_emulated_instruction(vcpu);
9786 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
9788 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
9790 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
9791 char instruction[3];
9792 unsigned long rip = kvm_rip_read(vcpu);
9795 * If the quirk is disabled, synthesize a #UD and let the guest pick up
9798 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_FIX_HYPERCALL_INSN)) {
9799 ctxt->exception.error_code_valid = false;
9800 ctxt->exception.vector = UD_VECTOR;
9801 ctxt->have_exception = true;
9802 return X86EMUL_PROPAGATE_FAULT;
9805 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
9807 return emulator_write_emulated(ctxt, rip, instruction, 3,
9811 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
9813 return vcpu->run->request_interrupt_window &&
9814 likely(!pic_in_kernel(vcpu->kvm));
9817 /* Called within kvm->srcu read side. */
9818 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
9820 struct kvm_run *kvm_run = vcpu->run;
9822 kvm_run->if_flag = static_call(kvm_x86_get_if_flag)(vcpu);
9823 kvm_run->cr8 = kvm_get_cr8(vcpu);
9824 kvm_run->apic_base = kvm_get_apic_base(vcpu);
9826 kvm_run->ready_for_interrupt_injection =
9827 pic_in_kernel(vcpu->kvm) ||
9828 kvm_vcpu_ready_for_interrupt_injection(vcpu);
9831 kvm_run->flags |= KVM_RUN_X86_SMM;
9834 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
9838 if (!kvm_x86_ops.update_cr8_intercept)
9841 if (!lapic_in_kernel(vcpu))
9844 if (vcpu->arch.apic->apicv_active)
9847 if (!vcpu->arch.apic->vapic_addr)
9848 max_irr = kvm_lapic_find_highest_irr(vcpu);
9855 tpr = kvm_lapic_get_cr8(vcpu);
9857 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
9861 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
9863 if (kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9864 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9868 return kvm_x86_ops.nested_ops->check_events(vcpu);
9871 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
9873 trace_kvm_inj_exception(vcpu->arch.exception.vector,
9874 vcpu->arch.exception.has_error_code,
9875 vcpu->arch.exception.error_code,
9876 vcpu->arch.exception.injected);
9878 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
9879 vcpu->arch.exception.error_code = false;
9880 static_call(kvm_x86_inject_exception)(vcpu);
9884 * Check for any event (interrupt or exception) that is ready to be injected,
9885 * and if there is at least one event, inject the event with the highest
9886 * priority. This handles both "pending" events, i.e. events that have never
9887 * been injected into the guest, and "injected" events, i.e. events that were
9888 * injected as part of a previous VM-Enter, but weren't successfully delivered
9889 * and need to be re-injected.
9891 * Note, this is not guaranteed to be invoked on a guest instruction boundary,
9892 * i.e. doesn't guarantee that there's an event window in the guest. KVM must
9893 * be able to inject exceptions in the "middle" of an instruction, and so must
9894 * also be able to re-inject NMIs and IRQs in the middle of an instruction.
9895 * I.e. for exceptions and re-injected events, NOT invoking this on instruction
9896 * boundaries is necessary and correct.
9898 * For simplicity, KVM uses a single path to inject all events (except events
9899 * that are injected directly from L1 to L2) and doesn't explicitly track
9900 * instruction boundaries for asynchronous events. However, because VM-Exits
9901 * that can occur during instruction execution typically result in KVM skipping
9902 * the instruction or injecting an exception, e.g. instruction and exception
9903 * intercepts, and because pending exceptions have higher priority than pending
9904 * interrupts, KVM still honors instruction boundaries in most scenarios.
9906 * But, if a VM-Exit occurs during instruction execution, and KVM does NOT skip
9907 * the instruction or inject an exception, then KVM can incorrecty inject a new
9908 * asynchrounous event if the event became pending after the CPU fetched the
9909 * instruction (in the guest). E.g. if a page fault (#PF, #NPF, EPT violation)
9910 * occurs and is resolved by KVM, a coincident NMI, SMI, IRQ, etc... can be
9911 * injected on the restarted instruction instead of being deferred until the
9912 * instruction completes.
9914 * In practice, this virtualization hole is unlikely to be observed by the
9915 * guest, and even less likely to cause functional problems. To detect the
9916 * hole, the guest would have to trigger an event on a side effect of an early
9917 * phase of instruction execution, e.g. on the instruction fetch from memory.
9918 * And for it to be a functional problem, the guest would need to depend on the
9919 * ordering between that side effect, the instruction completing, _and_ the
9920 * delivery of the asynchronous event.
9922 static int kvm_check_and_inject_events(struct kvm_vcpu *vcpu,
9923 bool *req_immediate_exit)
9929 * Process nested events first, as nested VM-Exit supercedes event
9930 * re-injection. If there's an event queued for re-injection, it will
9931 * be saved into the appropriate vmc{b,s}12 fields on nested VM-Exit.
9933 if (is_guest_mode(vcpu))
9934 r = kvm_check_nested_events(vcpu);
9939 * Re-inject exceptions and events *especially* if immediate entry+exit
9940 * to/from L2 is needed, as any event that has already been injected
9941 * into L2 needs to complete its lifecycle before injecting a new event.
9943 * Don't re-inject an NMI or interrupt if there is a pending exception.
9944 * This collision arises if an exception occurred while vectoring the
9945 * injected event, KVM intercepted said exception, and KVM ultimately
9946 * determined the fault belongs to the guest and queues the exception
9947 * for injection back into the guest.
9949 * "Injected" interrupts can also collide with pending exceptions if
9950 * userspace ignores the "ready for injection" flag and blindly queues
9951 * an interrupt. In that case, prioritizing the exception is correct,
9952 * as the exception "occurred" before the exit to userspace. Trap-like
9953 * exceptions, e.g. most #DBs, have higher priority than interrupts.
9954 * And while fault-like exceptions, e.g. #GP and #PF, are the lowest
9955 * priority, they're only generated (pended) during instruction
9956 * execution, and interrupts are recognized at instruction boundaries.
9957 * Thus a pending fault-like exception means the fault occurred on the
9958 * *previous* instruction and must be serviced prior to recognizing any
9959 * new events in order to fully complete the previous instruction.
9961 if (vcpu->arch.exception.injected)
9962 kvm_inject_exception(vcpu);
9963 else if (kvm_is_exception_pending(vcpu))
9965 else if (vcpu->arch.nmi_injected)
9966 static_call(kvm_x86_inject_nmi)(vcpu);
9967 else if (vcpu->arch.interrupt.injected)
9968 static_call(kvm_x86_inject_irq)(vcpu, true);
9971 * Exceptions that morph to VM-Exits are handled above, and pending
9972 * exceptions on top of injected exceptions that do not VM-Exit should
9973 * either morph to #DF or, sadly, override the injected exception.
9975 WARN_ON_ONCE(vcpu->arch.exception.injected &&
9976 vcpu->arch.exception.pending);
9979 * Bail if immediate entry+exit to/from the guest is needed to complete
9980 * nested VM-Enter or event re-injection so that a different pending
9981 * event can be serviced (or if KVM needs to exit to userspace).
9983 * Otherwise, continue processing events even if VM-Exit occurred. The
9984 * VM-Exit will have cleared exceptions that were meant for L2, but
9985 * there may now be events that can be injected into L1.
9991 * A pending exception VM-Exit should either result in nested VM-Exit
9992 * or force an immediate re-entry and exit to/from L2, and exception
9993 * VM-Exits cannot be injected (flag should _never_ be set).
9995 WARN_ON_ONCE(vcpu->arch.exception_vmexit.injected ||
9996 vcpu->arch.exception_vmexit.pending);
9999 * New events, other than exceptions, cannot be injected if KVM needs
10000 * to re-inject a previous event. See above comments on re-injecting
10001 * for why pending exceptions get priority.
10003 can_inject = !kvm_event_needs_reinjection(vcpu);
10005 if (vcpu->arch.exception.pending) {
10007 * Fault-class exceptions, except #DBs, set RF=1 in the RFLAGS
10008 * value pushed on the stack. Trap-like exception and all #DBs
10009 * leave RF as-is (KVM follows Intel's behavior in this regard;
10010 * AMD states that code breakpoint #DBs excplitly clear RF=0).
10012 * Note, most versions of Intel's SDM and AMD's APM incorrectly
10013 * describe the behavior of General Detect #DBs, which are
10014 * fault-like. They do _not_ set RF, a la code breakpoints.
10016 if (exception_type(vcpu->arch.exception.vector) == EXCPT_FAULT)
10017 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
10020 if (vcpu->arch.exception.vector == DB_VECTOR) {
10021 kvm_deliver_exception_payload(vcpu, &vcpu->arch.exception);
10022 if (vcpu->arch.dr7 & DR7_GD) {
10023 vcpu->arch.dr7 &= ~DR7_GD;
10024 kvm_update_dr7(vcpu);
10028 kvm_inject_exception(vcpu);
10030 vcpu->arch.exception.pending = false;
10031 vcpu->arch.exception.injected = true;
10033 can_inject = false;
10036 /* Don't inject interrupts if the user asked to avoid doing so */
10037 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ)
10041 * Finally, inject interrupt events. If an event cannot be injected
10042 * due to architectural conditions (e.g. IF=0) a window-open exit
10043 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
10044 * and can architecturally be injected, but we cannot do it right now:
10045 * an interrupt could have arrived just now and we have to inject it
10046 * as a vmexit, or there could already an event in the queue, which is
10047 * indicated by can_inject. In that case we request an immediate exit
10048 * in order to make progress and get back here for another iteration.
10049 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
10051 #ifdef CONFIG_KVM_SMM
10052 if (vcpu->arch.smi_pending) {
10053 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
10057 vcpu->arch.smi_pending = false;
10058 ++vcpu->arch.smi_count;
10060 can_inject = false;
10062 static_call(kvm_x86_enable_smi_window)(vcpu);
10066 if (vcpu->arch.nmi_pending) {
10067 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
10071 --vcpu->arch.nmi_pending;
10072 vcpu->arch.nmi_injected = true;
10073 static_call(kvm_x86_inject_nmi)(vcpu);
10074 can_inject = false;
10075 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
10077 if (vcpu->arch.nmi_pending)
10078 static_call(kvm_x86_enable_nmi_window)(vcpu);
10081 if (kvm_cpu_has_injectable_intr(vcpu)) {
10082 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
10086 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
10087 static_call(kvm_x86_inject_irq)(vcpu, false);
10088 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
10090 if (kvm_cpu_has_injectable_intr(vcpu))
10091 static_call(kvm_x86_enable_irq_window)(vcpu);
10094 if (is_guest_mode(vcpu) &&
10095 kvm_x86_ops.nested_ops->has_events &&
10096 kvm_x86_ops.nested_ops->has_events(vcpu))
10097 *req_immediate_exit = true;
10100 * KVM must never queue a new exception while injecting an event; KVM
10101 * is done emulating and should only propagate the to-be-injected event
10102 * to the VMCS/VMCB. Queueing a new exception can put the vCPU into an
10103 * infinite loop as KVM will bail from VM-Enter to inject the pending
10104 * exception and start the cycle all over.
10106 * Exempt triple faults as they have special handling and won't put the
10107 * vCPU into an infinite loop. Triple fault can be queued when running
10108 * VMX without unrestricted guest, as that requires KVM to emulate Real
10109 * Mode events (see kvm_inject_realmode_interrupt()).
10111 WARN_ON_ONCE(vcpu->arch.exception.pending ||
10112 vcpu->arch.exception_vmexit.pending);
10117 *req_immediate_exit = true;
10123 static void process_nmi(struct kvm_vcpu *vcpu)
10125 unsigned limit = 2;
10128 * x86 is limited to one NMI running, and one NMI pending after it.
10129 * If an NMI is already in progress, limit further NMIs to just one.
10130 * Otherwise, allow two (and we'll inject the first one immediately).
10132 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
10135 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
10136 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
10137 kvm_make_request(KVM_REQ_EVENT, vcpu);
10140 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
10141 unsigned long *vcpu_bitmap)
10143 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC, vcpu_bitmap);
10146 void kvm_make_scan_ioapic_request(struct kvm *kvm)
10148 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
10151 void __kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
10153 struct kvm_lapic *apic = vcpu->arch.apic;
10156 if (!lapic_in_kernel(vcpu))
10159 down_read(&vcpu->kvm->arch.apicv_update_lock);
10162 /* Do not activate APICV when APIC is disabled */
10163 activate = kvm_vcpu_apicv_activated(vcpu) &&
10164 (kvm_get_apic_mode(vcpu) != LAPIC_MODE_DISABLED);
10166 if (apic->apicv_active == activate)
10169 apic->apicv_active = activate;
10170 kvm_apic_update_apicv(vcpu);
10171 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
10174 * When APICv gets disabled, we may still have injected interrupts
10175 * pending. At the same time, KVM_REQ_EVENT may not be set as APICv was
10176 * still active when the interrupt got accepted. Make sure
10177 * kvm_check_and_inject_events() is called to check for that.
10179 if (!apic->apicv_active)
10180 kvm_make_request(KVM_REQ_EVENT, vcpu);
10184 up_read(&vcpu->kvm->arch.apicv_update_lock);
10186 EXPORT_SYMBOL_GPL(__kvm_vcpu_update_apicv);
10188 static void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
10190 if (!lapic_in_kernel(vcpu))
10194 * Due to sharing page tables across vCPUs, the xAPIC memslot must be
10195 * deleted if any vCPU has xAPIC virtualization and x2APIC enabled, but
10196 * and hardware doesn't support x2APIC virtualization. E.g. some AMD
10197 * CPUs support AVIC but not x2APIC. KVM still allows enabling AVIC in
10198 * this case so that KVM can the AVIC doorbell to inject interrupts to
10199 * running vCPUs, but KVM must not create SPTEs for the APIC base as
10200 * the vCPU would incorrectly be able to access the vAPIC page via MMIO
10201 * despite being in x2APIC mode. For simplicity, inhibiting the APIC
10202 * access page is sticky.
10204 if (apic_x2apic_mode(vcpu->arch.apic) &&
10205 kvm_x86_ops.allow_apicv_in_x2apic_without_x2apic_virtualization)
10206 kvm_inhibit_apic_access_page(vcpu);
10208 __kvm_vcpu_update_apicv(vcpu);
10211 void __kvm_set_or_clear_apicv_inhibit(struct kvm *kvm,
10212 enum kvm_apicv_inhibit reason, bool set)
10214 unsigned long old, new;
10216 lockdep_assert_held_write(&kvm->arch.apicv_update_lock);
10218 if (!(kvm_x86_ops.required_apicv_inhibits & BIT(reason)))
10221 old = new = kvm->arch.apicv_inhibit_reasons;
10223 set_or_clear_apicv_inhibit(&new, reason, set);
10225 if (!!old != !!new) {
10227 * Kick all vCPUs before setting apicv_inhibit_reasons to avoid
10228 * false positives in the sanity check WARN in svm_vcpu_run().
10229 * This task will wait for all vCPUs to ack the kick IRQ before
10230 * updating apicv_inhibit_reasons, and all other vCPUs will
10231 * block on acquiring apicv_update_lock so that vCPUs can't
10232 * redo svm_vcpu_run() without seeing the new inhibit state.
10234 * Note, holding apicv_update_lock and taking it in the read
10235 * side (handling the request) also prevents other vCPUs from
10236 * servicing the request with a stale apicv_inhibit_reasons.
10238 kvm_make_all_cpus_request(kvm, KVM_REQ_APICV_UPDATE);
10239 kvm->arch.apicv_inhibit_reasons = new;
10241 unsigned long gfn = gpa_to_gfn(APIC_DEFAULT_PHYS_BASE);
10242 int idx = srcu_read_lock(&kvm->srcu);
10244 kvm_zap_gfn_range(kvm, gfn, gfn+1);
10245 srcu_read_unlock(&kvm->srcu, idx);
10248 kvm->arch.apicv_inhibit_reasons = new;
10252 void kvm_set_or_clear_apicv_inhibit(struct kvm *kvm,
10253 enum kvm_apicv_inhibit reason, bool set)
10258 down_write(&kvm->arch.apicv_update_lock);
10259 __kvm_set_or_clear_apicv_inhibit(kvm, reason, set);
10260 up_write(&kvm->arch.apicv_update_lock);
10262 EXPORT_SYMBOL_GPL(kvm_set_or_clear_apicv_inhibit);
10264 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
10266 if (!kvm_apic_present(vcpu))
10269 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
10271 if (irqchip_split(vcpu->kvm))
10272 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
10274 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10275 if (ioapic_in_kernel(vcpu->kvm))
10276 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
10279 if (is_guest_mode(vcpu))
10280 vcpu->arch.load_eoi_exitmap_pending = true;
10282 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
10285 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
10287 u64 eoi_exit_bitmap[4];
10289 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
10292 if (to_hv_vcpu(vcpu)) {
10293 bitmap_or((ulong *)eoi_exit_bitmap,
10294 vcpu->arch.ioapic_handled_vectors,
10295 to_hv_synic(vcpu)->vec_bitmap, 256);
10296 static_call_cond(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
10300 static_call_cond(kvm_x86_load_eoi_exitmap)(
10301 vcpu, (u64 *)vcpu->arch.ioapic_handled_vectors);
10304 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
10305 unsigned long start, unsigned long end)
10307 unsigned long apic_address;
10310 * The physical address of apic access page is stored in the VMCS.
10311 * Update it when it becomes invalid.
10313 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
10314 if (start <= apic_address && apic_address < end)
10315 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
10318 void kvm_arch_guest_memory_reclaimed(struct kvm *kvm)
10320 static_call_cond(kvm_x86_guest_memory_reclaimed)(kvm);
10323 static void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
10325 if (!lapic_in_kernel(vcpu))
10328 static_call_cond(kvm_x86_set_apic_access_page_addr)(vcpu);
10331 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
10333 smp_send_reschedule(vcpu->cpu);
10335 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
10338 * Called within kvm->srcu read side.
10339 * Returns 1 to let vcpu_run() continue the guest execution loop without
10340 * exiting to the userspace. Otherwise, the value will be returned to the
10343 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
10347 dm_request_for_irq_injection(vcpu) &&
10348 kvm_cpu_accept_dm_intr(vcpu);
10349 fastpath_t exit_fastpath;
10351 bool req_immediate_exit = false;
10353 if (kvm_request_pending(vcpu)) {
10354 if (kvm_check_request(KVM_REQ_VM_DEAD, vcpu)) {
10359 if (kvm_dirty_ring_check_request(vcpu)) {
10364 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
10365 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
10370 if (kvm_check_request(KVM_REQ_MMU_FREE_OBSOLETE_ROOTS, vcpu))
10371 kvm_mmu_free_obsolete_roots(vcpu);
10372 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
10373 __kvm_migrate_timers(vcpu);
10374 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
10375 kvm_update_masterclock(vcpu->kvm);
10376 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
10377 kvm_gen_kvmclock_update(vcpu);
10378 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
10379 r = kvm_guest_time_update(vcpu);
10383 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
10384 kvm_mmu_sync_roots(vcpu);
10385 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
10386 kvm_mmu_load_pgd(vcpu);
10389 * Note, the order matters here, as flushing "all" TLB entries
10390 * also flushes the "current" TLB entries, i.e. servicing the
10391 * flush "all" will clear any request to flush "current".
10393 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
10394 kvm_vcpu_flush_tlb_all(vcpu);
10396 kvm_service_local_tlb_flush_requests(vcpu);
10399 * Fall back to a "full" guest flush if Hyper-V's precise
10400 * flushing fails. Note, Hyper-V's flushing is per-vCPU, but
10401 * the flushes are considered "remote" and not "local" because
10402 * the requests can be initiated from other vCPUs.
10404 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu) &&
10405 kvm_hv_vcpu_flush_tlb(vcpu))
10406 kvm_vcpu_flush_tlb_guest(vcpu);
10408 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
10409 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
10413 if (kvm_test_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
10414 if (is_guest_mode(vcpu))
10415 kvm_x86_ops.nested_ops->triple_fault(vcpu);
10417 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
10418 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
10419 vcpu->mmio_needed = 0;
10424 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
10425 /* Page is swapped out. Do synthetic halt */
10426 vcpu->arch.apf.halted = true;
10430 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
10431 record_steal_time(vcpu);
10432 #ifdef CONFIG_KVM_SMM
10433 if (kvm_check_request(KVM_REQ_SMI, vcpu))
10436 if (kvm_check_request(KVM_REQ_NMI, vcpu))
10438 if (kvm_check_request(KVM_REQ_PMU, vcpu))
10439 kvm_pmu_handle_event(vcpu);
10440 if (kvm_check_request(KVM_REQ_PMI, vcpu))
10441 kvm_pmu_deliver_pmi(vcpu);
10442 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
10443 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
10444 if (test_bit(vcpu->arch.pending_ioapic_eoi,
10445 vcpu->arch.ioapic_handled_vectors)) {
10446 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
10447 vcpu->run->eoi.vector =
10448 vcpu->arch.pending_ioapic_eoi;
10453 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
10454 vcpu_scan_ioapic(vcpu);
10455 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
10456 vcpu_load_eoi_exitmap(vcpu);
10457 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
10458 kvm_vcpu_reload_apic_access_page(vcpu);
10459 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
10460 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
10461 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
10462 vcpu->run->system_event.ndata = 0;
10466 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
10467 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
10468 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
10469 vcpu->run->system_event.ndata = 0;
10473 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
10474 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
10476 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
10477 vcpu->run->hyperv = hv_vcpu->exit;
10483 * KVM_REQ_HV_STIMER has to be processed after
10484 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
10485 * depend on the guest clock being up-to-date
10487 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
10488 kvm_hv_process_stimers(vcpu);
10489 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
10490 kvm_vcpu_update_apicv(vcpu);
10491 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
10492 kvm_check_async_pf_completion(vcpu);
10493 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
10494 static_call(kvm_x86_msr_filter_changed)(vcpu);
10496 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
10497 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
10500 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
10501 kvm_xen_has_interrupt(vcpu)) {
10502 ++vcpu->stat.req_event;
10503 r = kvm_apic_accept_events(vcpu);
10508 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
10513 r = kvm_check_and_inject_events(vcpu, &req_immediate_exit);
10519 static_call(kvm_x86_enable_irq_window)(vcpu);
10521 if (kvm_lapic_enabled(vcpu)) {
10522 update_cr8_intercept(vcpu);
10523 kvm_lapic_sync_to_vapic(vcpu);
10527 r = kvm_mmu_reload(vcpu);
10529 goto cancel_injection;
10534 static_call(kvm_x86_prepare_switch_to_guest)(vcpu);
10537 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
10538 * IPI are then delayed after guest entry, which ensures that they
10539 * result in virtual interrupt delivery.
10541 local_irq_disable();
10543 /* Store vcpu->apicv_active before vcpu->mode. */
10544 smp_store_release(&vcpu->mode, IN_GUEST_MODE);
10546 kvm_vcpu_srcu_read_unlock(vcpu);
10549 * 1) We should set ->mode before checking ->requests. Please see
10550 * the comment in kvm_vcpu_exiting_guest_mode().
10552 * 2) For APICv, we should set ->mode before checking PID.ON. This
10553 * pairs with the memory barrier implicit in pi_test_and_set_on
10554 * (see vmx_deliver_posted_interrupt).
10556 * 3) This also orders the write to mode from any reads to the page
10557 * tables done while the VCPU is running. Please see the comment
10558 * in kvm_flush_remote_tlbs.
10560 smp_mb__after_srcu_read_unlock();
10563 * Process pending posted interrupts to handle the case where the
10564 * notification IRQ arrived in the host, or was never sent (because the
10565 * target vCPU wasn't running). Do this regardless of the vCPU's APICv
10566 * status, KVM doesn't update assigned devices when APICv is inhibited,
10567 * i.e. they can post interrupts even if APICv is temporarily disabled.
10569 if (kvm_lapic_enabled(vcpu))
10570 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10572 if (kvm_vcpu_exit_request(vcpu)) {
10573 vcpu->mode = OUTSIDE_GUEST_MODE;
10575 local_irq_enable();
10577 kvm_vcpu_srcu_read_lock(vcpu);
10579 goto cancel_injection;
10582 if (req_immediate_exit) {
10583 kvm_make_request(KVM_REQ_EVENT, vcpu);
10584 static_call(kvm_x86_request_immediate_exit)(vcpu);
10587 fpregs_assert_state_consistent();
10588 if (test_thread_flag(TIF_NEED_FPU_LOAD))
10589 switch_fpu_return();
10591 if (vcpu->arch.guest_fpu.xfd_err)
10592 wrmsrl(MSR_IA32_XFD_ERR, vcpu->arch.guest_fpu.xfd_err);
10594 if (unlikely(vcpu->arch.switch_db_regs)) {
10595 set_debugreg(0, 7);
10596 set_debugreg(vcpu->arch.eff_db[0], 0);
10597 set_debugreg(vcpu->arch.eff_db[1], 1);
10598 set_debugreg(vcpu->arch.eff_db[2], 2);
10599 set_debugreg(vcpu->arch.eff_db[3], 3);
10600 } else if (unlikely(hw_breakpoint_active())) {
10601 set_debugreg(0, 7);
10604 guest_timing_enter_irqoff();
10608 * Assert that vCPU vs. VM APICv state is consistent. An APICv
10609 * update must kick and wait for all vCPUs before toggling the
10610 * per-VM state, and responsing vCPUs must wait for the update
10611 * to complete before servicing KVM_REQ_APICV_UPDATE.
10613 WARN_ON_ONCE((kvm_vcpu_apicv_activated(vcpu) != kvm_vcpu_apicv_active(vcpu)) &&
10614 (kvm_get_apic_mode(vcpu) != LAPIC_MODE_DISABLED));
10616 exit_fastpath = static_call(kvm_x86_vcpu_run)(vcpu);
10617 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
10620 if (kvm_lapic_enabled(vcpu))
10621 static_call_cond(kvm_x86_sync_pir_to_irr)(vcpu);
10623 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
10624 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
10630 * Do this here before restoring debug registers on the host. And
10631 * since we do this before handling the vmexit, a DR access vmexit
10632 * can (a) read the correct value of the debug registers, (b) set
10633 * KVM_DEBUGREG_WONT_EXIT again.
10635 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
10636 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
10637 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
10638 kvm_update_dr0123(vcpu);
10639 kvm_update_dr7(vcpu);
10643 * If the guest has used debug registers, at least dr7
10644 * will be disabled while returning to the host.
10645 * If we don't have active breakpoints in the host, we don't
10646 * care about the messed up debug address registers. But if
10647 * we have some of them active, restore the old state.
10649 if (hw_breakpoint_active())
10650 hw_breakpoint_restore();
10652 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
10653 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
10655 vcpu->mode = OUTSIDE_GUEST_MODE;
10659 * Sync xfd before calling handle_exit_irqoff() which may
10660 * rely on the fact that guest_fpu::xfd is up-to-date (e.g.
10661 * in #NM irqoff handler).
10663 if (vcpu->arch.xfd_no_write_intercept)
10664 fpu_sync_guest_vmexit_xfd_state();
10666 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
10668 if (vcpu->arch.guest_fpu.xfd_err)
10669 wrmsrl(MSR_IA32_XFD_ERR, 0);
10672 * Consume any pending interrupts, including the possible source of
10673 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
10674 * An instruction is required after local_irq_enable() to fully unblock
10675 * interrupts on processors that implement an interrupt shadow, the
10676 * stat.exits increment will do nicely.
10678 kvm_before_interrupt(vcpu, KVM_HANDLING_IRQ);
10679 local_irq_enable();
10680 ++vcpu->stat.exits;
10681 local_irq_disable();
10682 kvm_after_interrupt(vcpu);
10685 * Wait until after servicing IRQs to account guest time so that any
10686 * ticks that occurred while running the guest are properly accounted
10687 * to the guest. Waiting until IRQs are enabled degrades the accuracy
10688 * of accounting via context tracking, but the loss of accuracy is
10689 * acceptable for all known use cases.
10691 guest_timing_exit_irqoff();
10693 local_irq_enable();
10696 kvm_vcpu_srcu_read_lock(vcpu);
10699 * Profile KVM exit RIPs:
10701 if (unlikely(prof_on == KVM_PROFILING)) {
10702 unsigned long rip = kvm_rip_read(vcpu);
10703 profile_hit(KVM_PROFILING, (void *)rip);
10706 if (unlikely(vcpu->arch.tsc_always_catchup))
10707 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10709 if (vcpu->arch.apic_attention)
10710 kvm_lapic_sync_from_vapic(vcpu);
10712 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
10716 if (req_immediate_exit)
10717 kvm_make_request(KVM_REQ_EVENT, vcpu);
10718 static_call(kvm_x86_cancel_injection)(vcpu);
10719 if (unlikely(vcpu->arch.apic_attention))
10720 kvm_lapic_sync_from_vapic(vcpu);
10725 /* Called within kvm->srcu read side. */
10726 static inline int vcpu_block(struct kvm_vcpu *vcpu)
10730 if (!kvm_arch_vcpu_runnable(vcpu)) {
10732 * Switch to the software timer before halt-polling/blocking as
10733 * the guest's timer may be a break event for the vCPU, and the
10734 * hypervisor timer runs only when the CPU is in guest mode.
10735 * Switch before halt-polling so that KVM recognizes an expired
10736 * timer before blocking.
10738 hv_timer = kvm_lapic_hv_timer_in_use(vcpu);
10740 kvm_lapic_switch_to_sw_timer(vcpu);
10742 kvm_vcpu_srcu_read_unlock(vcpu);
10743 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10744 kvm_vcpu_halt(vcpu);
10746 kvm_vcpu_block(vcpu);
10747 kvm_vcpu_srcu_read_lock(vcpu);
10750 kvm_lapic_switch_to_hv_timer(vcpu);
10753 * If the vCPU is not runnable, a signal or another host event
10754 * of some kind is pending; service it without changing the
10755 * vCPU's activity state.
10757 if (!kvm_arch_vcpu_runnable(vcpu))
10762 * Evaluate nested events before exiting the halted state. This allows
10763 * the halt state to be recorded properly in the VMCS12's activity
10764 * state field (AMD does not have a similar field and a VM-Exit always
10765 * causes a spurious wakeup from HLT).
10767 if (is_guest_mode(vcpu)) {
10768 if (kvm_check_nested_events(vcpu) < 0)
10772 if (kvm_apic_accept_events(vcpu) < 0)
10774 switch(vcpu->arch.mp_state) {
10775 case KVM_MP_STATE_HALTED:
10776 case KVM_MP_STATE_AP_RESET_HOLD:
10777 vcpu->arch.pv.pv_unhalted = false;
10778 vcpu->arch.mp_state =
10779 KVM_MP_STATE_RUNNABLE;
10781 case KVM_MP_STATE_RUNNABLE:
10782 vcpu->arch.apf.halted = false;
10784 case KVM_MP_STATE_INIT_RECEIVED:
10793 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
10795 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
10796 !vcpu->arch.apf.halted);
10799 /* Called within kvm->srcu read side. */
10800 static int vcpu_run(struct kvm_vcpu *vcpu)
10804 vcpu->arch.l1tf_flush_l1d = true;
10808 * If another guest vCPU requests a PV TLB flush in the middle
10809 * of instruction emulation, the rest of the emulation could
10810 * use a stale page translation. Assume that any code after
10811 * this point can start executing an instruction.
10813 vcpu->arch.at_instruction_boundary = false;
10814 if (kvm_vcpu_running(vcpu)) {
10815 r = vcpu_enter_guest(vcpu);
10817 r = vcpu_block(vcpu);
10823 kvm_clear_request(KVM_REQ_UNBLOCK, vcpu);
10824 if (kvm_xen_has_pending_events(vcpu))
10825 kvm_xen_inject_pending_events(vcpu);
10827 if (kvm_cpu_has_pending_timer(vcpu))
10828 kvm_inject_pending_timer_irqs(vcpu);
10830 if (dm_request_for_irq_injection(vcpu) &&
10831 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
10833 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
10834 ++vcpu->stat.request_irq_exits;
10838 if (__xfer_to_guest_mode_work_pending()) {
10839 kvm_vcpu_srcu_read_unlock(vcpu);
10840 r = xfer_to_guest_mode_handle_work(vcpu);
10841 kvm_vcpu_srcu_read_lock(vcpu);
10850 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
10852 return kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
10855 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
10857 BUG_ON(!vcpu->arch.pio.count);
10859 return complete_emulated_io(vcpu);
10863 * Implements the following, as a state machine:
10866 * for each fragment
10867 * for each mmio piece in the fragment
10874 * for each fragment
10875 * for each mmio piece in the fragment
10880 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
10882 struct kvm_run *run = vcpu->run;
10883 struct kvm_mmio_fragment *frag;
10886 BUG_ON(!vcpu->mmio_needed);
10888 /* Complete previous fragment */
10889 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
10890 len = min(8u, frag->len);
10891 if (!vcpu->mmio_is_write)
10892 memcpy(frag->data, run->mmio.data, len);
10894 if (frag->len <= 8) {
10895 /* Switch to the next fragment. */
10897 vcpu->mmio_cur_fragment++;
10899 /* Go forward to the next mmio piece. */
10905 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
10906 vcpu->mmio_needed = 0;
10908 /* FIXME: return into emulator if single-stepping. */
10909 if (vcpu->mmio_is_write)
10911 vcpu->mmio_read_completed = 1;
10912 return complete_emulated_io(vcpu);
10915 run->exit_reason = KVM_EXIT_MMIO;
10916 run->mmio.phys_addr = frag->gpa;
10917 if (vcpu->mmio_is_write)
10918 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
10919 run->mmio.len = min(8u, frag->len);
10920 run->mmio.is_write = vcpu->mmio_is_write;
10921 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
10925 /* Swap (qemu) user FPU context for the guest FPU context. */
10926 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
10928 /* Exclude PKRU, it's restored separately immediately after VM-Exit. */
10929 fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, true);
10933 /* When vcpu_run ends, restore user space FPU context. */
10934 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
10936 fpu_swap_kvm_fpstate(&vcpu->arch.guest_fpu, false);
10937 ++vcpu->stat.fpu_reload;
10941 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
10943 struct kvm_queued_exception *ex = &vcpu->arch.exception;
10944 struct kvm_run *kvm_run = vcpu->run;
10948 kvm_sigset_activate(vcpu);
10949 kvm_run->flags = 0;
10950 kvm_load_guest_fpu(vcpu);
10952 kvm_vcpu_srcu_read_lock(vcpu);
10953 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
10954 if (kvm_run->immediate_exit) {
10959 * It should be impossible for the hypervisor timer to be in
10960 * use before KVM has ever run the vCPU.
10962 WARN_ON_ONCE(kvm_lapic_hv_timer_in_use(vcpu));
10964 kvm_vcpu_srcu_read_unlock(vcpu);
10965 kvm_vcpu_block(vcpu);
10966 kvm_vcpu_srcu_read_lock(vcpu);
10968 if (kvm_apic_accept_events(vcpu) < 0) {
10973 if (signal_pending(current)) {
10975 kvm_run->exit_reason = KVM_EXIT_INTR;
10976 ++vcpu->stat.signal_exits;
10981 if ((kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) ||
10982 (kvm_run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)) {
10987 if (kvm_run->kvm_dirty_regs) {
10988 r = sync_regs(vcpu);
10993 /* re-sync apic's tpr */
10994 if (!lapic_in_kernel(vcpu)) {
10995 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
11002 * If userspace set a pending exception and L2 is active, convert it to
11003 * a pending VM-Exit if L1 wants to intercept the exception.
11005 if (vcpu->arch.exception_from_userspace && is_guest_mode(vcpu) &&
11006 kvm_x86_ops.nested_ops->is_exception_vmexit(vcpu, ex->vector,
11008 kvm_queue_exception_vmexit(vcpu, ex->vector,
11009 ex->has_error_code, ex->error_code,
11010 ex->has_payload, ex->payload);
11011 ex->injected = false;
11012 ex->pending = false;
11014 vcpu->arch.exception_from_userspace = false;
11016 if (unlikely(vcpu->arch.complete_userspace_io)) {
11017 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
11018 vcpu->arch.complete_userspace_io = NULL;
11023 WARN_ON_ONCE(vcpu->arch.pio.count);
11024 WARN_ON_ONCE(vcpu->mmio_needed);
11027 if (kvm_run->immediate_exit) {
11032 r = static_call(kvm_x86_vcpu_pre_run)(vcpu);
11036 r = vcpu_run(vcpu);
11039 kvm_put_guest_fpu(vcpu);
11040 if (kvm_run->kvm_valid_regs)
11042 post_kvm_run_save(vcpu);
11043 kvm_vcpu_srcu_read_unlock(vcpu);
11045 kvm_sigset_deactivate(vcpu);
11050 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11052 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
11054 * We are here if userspace calls get_regs() in the middle of
11055 * instruction emulation. Registers state needs to be copied
11056 * back from emulation context to vcpu. Userspace shouldn't do
11057 * that usually, but some bad designed PV devices (vmware
11058 * backdoor interface) need this to work
11060 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
11061 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
11063 regs->rax = kvm_rax_read(vcpu);
11064 regs->rbx = kvm_rbx_read(vcpu);
11065 regs->rcx = kvm_rcx_read(vcpu);
11066 regs->rdx = kvm_rdx_read(vcpu);
11067 regs->rsi = kvm_rsi_read(vcpu);
11068 regs->rdi = kvm_rdi_read(vcpu);
11069 regs->rsp = kvm_rsp_read(vcpu);
11070 regs->rbp = kvm_rbp_read(vcpu);
11071 #ifdef CONFIG_X86_64
11072 regs->r8 = kvm_r8_read(vcpu);
11073 regs->r9 = kvm_r9_read(vcpu);
11074 regs->r10 = kvm_r10_read(vcpu);
11075 regs->r11 = kvm_r11_read(vcpu);
11076 regs->r12 = kvm_r12_read(vcpu);
11077 regs->r13 = kvm_r13_read(vcpu);
11078 regs->r14 = kvm_r14_read(vcpu);
11079 regs->r15 = kvm_r15_read(vcpu);
11082 regs->rip = kvm_rip_read(vcpu);
11083 regs->rflags = kvm_get_rflags(vcpu);
11086 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11089 __get_regs(vcpu, regs);
11094 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11096 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
11097 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
11099 kvm_rax_write(vcpu, regs->rax);
11100 kvm_rbx_write(vcpu, regs->rbx);
11101 kvm_rcx_write(vcpu, regs->rcx);
11102 kvm_rdx_write(vcpu, regs->rdx);
11103 kvm_rsi_write(vcpu, regs->rsi);
11104 kvm_rdi_write(vcpu, regs->rdi);
11105 kvm_rsp_write(vcpu, regs->rsp);
11106 kvm_rbp_write(vcpu, regs->rbp);
11107 #ifdef CONFIG_X86_64
11108 kvm_r8_write(vcpu, regs->r8);
11109 kvm_r9_write(vcpu, regs->r9);
11110 kvm_r10_write(vcpu, regs->r10);
11111 kvm_r11_write(vcpu, regs->r11);
11112 kvm_r12_write(vcpu, regs->r12);
11113 kvm_r13_write(vcpu, regs->r13);
11114 kvm_r14_write(vcpu, regs->r14);
11115 kvm_r15_write(vcpu, regs->r15);
11118 kvm_rip_write(vcpu, regs->rip);
11119 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
11121 vcpu->arch.exception.pending = false;
11122 vcpu->arch.exception_vmexit.pending = false;
11124 kvm_make_request(KVM_REQ_EVENT, vcpu);
11127 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
11130 __set_regs(vcpu, regs);
11135 static void __get_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11137 struct desc_ptr dt;
11139 if (vcpu->arch.guest_state_protected)
11140 goto skip_protected_regs;
11142 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
11143 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
11144 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
11145 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
11146 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
11147 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
11149 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
11150 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
11152 static_call(kvm_x86_get_idt)(vcpu, &dt);
11153 sregs->idt.limit = dt.size;
11154 sregs->idt.base = dt.address;
11155 static_call(kvm_x86_get_gdt)(vcpu, &dt);
11156 sregs->gdt.limit = dt.size;
11157 sregs->gdt.base = dt.address;
11159 sregs->cr2 = vcpu->arch.cr2;
11160 sregs->cr3 = kvm_read_cr3(vcpu);
11162 skip_protected_regs:
11163 sregs->cr0 = kvm_read_cr0(vcpu);
11164 sregs->cr4 = kvm_read_cr4(vcpu);
11165 sregs->cr8 = kvm_get_cr8(vcpu);
11166 sregs->efer = vcpu->arch.efer;
11167 sregs->apic_base = kvm_get_apic_base(vcpu);
11170 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11172 __get_sregs_common(vcpu, sregs);
11174 if (vcpu->arch.guest_state_protected)
11177 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
11178 set_bit(vcpu->arch.interrupt.nr,
11179 (unsigned long *)sregs->interrupt_bitmap);
11182 static void __get_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
11186 __get_sregs_common(vcpu, (struct kvm_sregs *)sregs2);
11188 if (vcpu->arch.guest_state_protected)
11191 if (is_pae_paging(vcpu)) {
11192 for (i = 0 ; i < 4 ; i++)
11193 sregs2->pdptrs[i] = kvm_pdptr_read(vcpu, i);
11194 sregs2->flags |= KVM_SREGS2_FLAGS_PDPTRS_VALID;
11198 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
11199 struct kvm_sregs *sregs)
11202 __get_sregs(vcpu, sregs);
11207 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
11208 struct kvm_mp_state *mp_state)
11213 if (kvm_mpx_supported())
11214 kvm_load_guest_fpu(vcpu);
11216 r = kvm_apic_accept_events(vcpu);
11221 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
11222 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
11223 vcpu->arch.pv.pv_unhalted)
11224 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
11226 mp_state->mp_state = vcpu->arch.mp_state;
11229 if (kvm_mpx_supported())
11230 kvm_put_guest_fpu(vcpu);
11235 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
11236 struct kvm_mp_state *mp_state)
11242 switch (mp_state->mp_state) {
11243 case KVM_MP_STATE_UNINITIALIZED:
11244 case KVM_MP_STATE_HALTED:
11245 case KVM_MP_STATE_AP_RESET_HOLD:
11246 case KVM_MP_STATE_INIT_RECEIVED:
11247 case KVM_MP_STATE_SIPI_RECEIVED:
11248 if (!lapic_in_kernel(vcpu))
11252 case KVM_MP_STATE_RUNNABLE:
11260 * Pending INITs are reported using KVM_SET_VCPU_EVENTS, disallow
11261 * forcing the guest into INIT/SIPI if those events are supposed to be
11262 * blocked. KVM prioritizes SMI over INIT, so reject INIT/SIPI state
11263 * if an SMI is pending as well.
11265 if ((!kvm_apic_init_sipi_allowed(vcpu) || vcpu->arch.smi_pending) &&
11266 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
11267 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
11270 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
11271 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
11272 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
11274 vcpu->arch.mp_state = mp_state->mp_state;
11275 kvm_make_request(KVM_REQ_EVENT, vcpu);
11283 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
11284 int reason, bool has_error_code, u32 error_code)
11286 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
11289 init_emulate_ctxt(vcpu);
11291 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
11292 has_error_code, error_code);
11294 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11295 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11296 vcpu->run->internal.ndata = 0;
11300 kvm_rip_write(vcpu, ctxt->eip);
11301 kvm_set_rflags(vcpu, ctxt->eflags);
11304 EXPORT_SYMBOL_GPL(kvm_task_switch);
11306 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11308 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
11310 * When EFER.LME and CR0.PG are set, the processor is in
11311 * 64-bit mode (though maybe in a 32-bit code segment).
11312 * CR4.PAE and EFER.LMA must be set.
11314 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
11316 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
11320 * Not in 64-bit mode: EFER.LMA is clear and the code
11321 * segment cannot be 64-bit.
11323 if (sregs->efer & EFER_LMA || sregs->cs.l)
11327 return kvm_is_valid_cr4(vcpu, sregs->cr4);
11330 static int __set_sregs_common(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs,
11331 int *mmu_reset_needed, bool update_pdptrs)
11333 struct msr_data apic_base_msr;
11335 struct desc_ptr dt;
11337 if (!kvm_is_valid_sregs(vcpu, sregs))
11340 apic_base_msr.data = sregs->apic_base;
11341 apic_base_msr.host_initiated = true;
11342 if (kvm_set_apic_base(vcpu, &apic_base_msr))
11345 if (vcpu->arch.guest_state_protected)
11348 dt.size = sregs->idt.limit;
11349 dt.address = sregs->idt.base;
11350 static_call(kvm_x86_set_idt)(vcpu, &dt);
11351 dt.size = sregs->gdt.limit;
11352 dt.address = sregs->gdt.base;
11353 static_call(kvm_x86_set_gdt)(vcpu, &dt);
11355 vcpu->arch.cr2 = sregs->cr2;
11356 *mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
11357 vcpu->arch.cr3 = sregs->cr3;
11358 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11359 static_call_cond(kvm_x86_post_set_cr3)(vcpu, sregs->cr3);
11361 kvm_set_cr8(vcpu, sregs->cr8);
11363 *mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
11364 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
11366 *mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
11367 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
11368 vcpu->arch.cr0 = sregs->cr0;
11370 *mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
11371 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
11373 if (update_pdptrs) {
11374 idx = srcu_read_lock(&vcpu->kvm->srcu);
11375 if (is_pae_paging(vcpu)) {
11376 load_pdptrs(vcpu, kvm_read_cr3(vcpu));
11377 *mmu_reset_needed = 1;
11379 srcu_read_unlock(&vcpu->kvm->srcu, idx);
11382 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
11383 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
11384 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
11385 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
11386 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
11387 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
11389 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
11390 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
11392 update_cr8_intercept(vcpu);
11394 /* Older userspace won't unhalt the vcpu on reset. */
11395 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
11396 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
11397 !is_protmode(vcpu))
11398 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11403 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
11405 int pending_vec, max_bits;
11406 int mmu_reset_needed = 0;
11407 int ret = __set_sregs_common(vcpu, sregs, &mmu_reset_needed, true);
11412 if (mmu_reset_needed)
11413 kvm_mmu_reset_context(vcpu);
11415 max_bits = KVM_NR_INTERRUPTS;
11416 pending_vec = find_first_bit(
11417 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
11419 if (pending_vec < max_bits) {
11420 kvm_queue_interrupt(vcpu, pending_vec, false);
11421 pr_debug("Set back pending irq %d\n", pending_vec);
11422 kvm_make_request(KVM_REQ_EVENT, vcpu);
11427 static int __set_sregs2(struct kvm_vcpu *vcpu, struct kvm_sregs2 *sregs2)
11429 int mmu_reset_needed = 0;
11430 bool valid_pdptrs = sregs2->flags & KVM_SREGS2_FLAGS_PDPTRS_VALID;
11431 bool pae = (sregs2->cr0 & X86_CR0_PG) && (sregs2->cr4 & X86_CR4_PAE) &&
11432 !(sregs2->efer & EFER_LMA);
11435 if (sregs2->flags & ~KVM_SREGS2_FLAGS_PDPTRS_VALID)
11438 if (valid_pdptrs && (!pae || vcpu->arch.guest_state_protected))
11441 ret = __set_sregs_common(vcpu, (struct kvm_sregs *)sregs2,
11442 &mmu_reset_needed, !valid_pdptrs);
11446 if (valid_pdptrs) {
11447 for (i = 0; i < 4 ; i++)
11448 kvm_pdptr_write(vcpu, i, sregs2->pdptrs[i]);
11450 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
11451 mmu_reset_needed = 1;
11452 vcpu->arch.pdptrs_from_userspace = true;
11454 if (mmu_reset_needed)
11455 kvm_mmu_reset_context(vcpu);
11459 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
11460 struct kvm_sregs *sregs)
11465 ret = __set_sregs(vcpu, sregs);
11470 static void kvm_arch_vcpu_guestdbg_update_apicv_inhibit(struct kvm *kvm)
11473 struct kvm_vcpu *vcpu;
11479 down_write(&kvm->arch.apicv_update_lock);
11481 kvm_for_each_vcpu(i, vcpu, kvm) {
11482 if (vcpu->guest_debug & KVM_GUESTDBG_BLOCKIRQ) {
11487 __kvm_set_or_clear_apicv_inhibit(kvm, APICV_INHIBIT_REASON_BLOCKIRQ, set);
11488 up_write(&kvm->arch.apicv_update_lock);
11491 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
11492 struct kvm_guest_debug *dbg)
11494 unsigned long rflags;
11497 if (vcpu->arch.guest_state_protected)
11502 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
11504 if (kvm_is_exception_pending(vcpu))
11506 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
11507 kvm_queue_exception(vcpu, DB_VECTOR);
11509 kvm_queue_exception(vcpu, BP_VECTOR);
11513 * Read rflags as long as potentially injected trace flags are still
11516 rflags = kvm_get_rflags(vcpu);
11518 vcpu->guest_debug = dbg->control;
11519 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
11520 vcpu->guest_debug = 0;
11522 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
11523 for (i = 0; i < KVM_NR_DB_REGS; ++i)
11524 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
11525 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
11527 for (i = 0; i < KVM_NR_DB_REGS; i++)
11528 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
11530 kvm_update_dr7(vcpu);
11532 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11533 vcpu->arch.singlestep_rip = kvm_get_linear_rip(vcpu);
11536 * Trigger an rflags update that will inject or remove the trace
11539 kvm_set_rflags(vcpu, rflags);
11541 static_call(kvm_x86_update_exception_bitmap)(vcpu);
11543 kvm_arch_vcpu_guestdbg_update_apicv_inhibit(vcpu->kvm);
11553 * Translate a guest virtual address to a guest physical address.
11555 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
11556 struct kvm_translation *tr)
11558 unsigned long vaddr = tr->linear_address;
11564 idx = srcu_read_lock(&vcpu->kvm->srcu);
11565 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
11566 srcu_read_unlock(&vcpu->kvm->srcu, idx);
11567 tr->physical_address = gpa;
11568 tr->valid = gpa != INVALID_GPA;
11576 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
11578 struct fxregs_state *fxsave;
11580 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
11585 fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
11586 memcpy(fpu->fpr, fxsave->st_space, 128);
11587 fpu->fcw = fxsave->cwd;
11588 fpu->fsw = fxsave->swd;
11589 fpu->ftwx = fxsave->twd;
11590 fpu->last_opcode = fxsave->fop;
11591 fpu->last_ip = fxsave->rip;
11592 fpu->last_dp = fxsave->rdp;
11593 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
11599 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
11601 struct fxregs_state *fxsave;
11603 if (fpstate_is_confidential(&vcpu->arch.guest_fpu))
11608 fxsave = &vcpu->arch.guest_fpu.fpstate->regs.fxsave;
11610 memcpy(fxsave->st_space, fpu->fpr, 128);
11611 fxsave->cwd = fpu->fcw;
11612 fxsave->swd = fpu->fsw;
11613 fxsave->twd = fpu->ftwx;
11614 fxsave->fop = fpu->last_opcode;
11615 fxsave->rip = fpu->last_ip;
11616 fxsave->rdp = fpu->last_dp;
11617 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
11623 static void store_regs(struct kvm_vcpu *vcpu)
11625 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
11627 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
11628 __get_regs(vcpu, &vcpu->run->s.regs.regs);
11630 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
11631 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
11633 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
11634 kvm_vcpu_ioctl_x86_get_vcpu_events(
11635 vcpu, &vcpu->run->s.regs.events);
11638 static int sync_regs(struct kvm_vcpu *vcpu)
11640 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
11641 __set_regs(vcpu, &vcpu->run->s.regs.regs);
11642 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
11644 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
11645 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
11647 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
11649 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
11650 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
11651 vcpu, &vcpu->run->s.regs.events))
11653 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
11659 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
11661 if (kvm_check_tsc_unstable() && kvm->created_vcpus)
11662 pr_warn_once("SMP vm created on host with unstable TSC; "
11663 "guest TSC will not be reliable\n");
11665 if (!kvm->arch.max_vcpu_ids)
11666 kvm->arch.max_vcpu_ids = KVM_MAX_VCPU_IDS;
11668 if (id >= kvm->arch.max_vcpu_ids)
11671 return static_call(kvm_x86_vcpu_precreate)(kvm);
11674 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
11679 vcpu->arch.last_vmentry_cpu = -1;
11680 vcpu->arch.regs_avail = ~0;
11681 vcpu->arch.regs_dirty = ~0;
11683 kvm_gpc_init(&vcpu->arch.pv_time, vcpu->kvm, vcpu, KVM_HOST_USES_PFN);
11685 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
11686 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11688 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
11690 r = kvm_mmu_create(vcpu);
11694 if (irqchip_in_kernel(vcpu->kvm)) {
11695 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
11697 goto fail_mmu_destroy;
11700 * Defer evaluating inhibits until the vCPU is first run, as
11701 * this vCPU will not get notified of any changes until this
11702 * vCPU is visible to other vCPUs (marked online and added to
11703 * the set of vCPUs). Opportunistically mark APICv active as
11704 * VMX in particularly is highly unlikely to have inhibits.
11705 * Ignore the current per-VM APICv state so that vCPU creation
11706 * is guaranteed to run with a deterministic value, the request
11707 * will ensure the vCPU gets the correct state before VM-Entry.
11709 if (enable_apicv) {
11710 vcpu->arch.apic->apicv_active = true;
11711 kvm_make_request(KVM_REQ_APICV_UPDATE, vcpu);
11714 static_branch_inc(&kvm_has_noapic_vcpu);
11718 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
11720 goto fail_free_lapic;
11721 vcpu->arch.pio_data = page_address(page);
11723 vcpu->arch.mce_banks = kcalloc(KVM_MAX_MCE_BANKS * 4, sizeof(u64),
11724 GFP_KERNEL_ACCOUNT);
11725 vcpu->arch.mci_ctl2_banks = kcalloc(KVM_MAX_MCE_BANKS, sizeof(u64),
11726 GFP_KERNEL_ACCOUNT);
11727 if (!vcpu->arch.mce_banks || !vcpu->arch.mci_ctl2_banks)
11728 goto fail_free_mce_banks;
11729 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
11731 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
11732 GFP_KERNEL_ACCOUNT))
11733 goto fail_free_mce_banks;
11735 if (!alloc_emulate_ctxt(vcpu))
11736 goto free_wbinvd_dirty_mask;
11738 if (!fpu_alloc_guest_fpstate(&vcpu->arch.guest_fpu)) {
11739 pr_err("failed to allocate vcpu's fpu\n");
11740 goto free_emulate_ctxt;
11743 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
11744 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
11746 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
11748 kvm_async_pf_hash_reset(vcpu);
11750 vcpu->arch.perf_capabilities = kvm_caps.supported_perf_cap;
11751 kvm_pmu_init(vcpu);
11753 vcpu->arch.pending_external_vector = -1;
11754 vcpu->arch.preempted_in_kernel = false;
11756 #if IS_ENABLED(CONFIG_HYPERV)
11757 vcpu->arch.hv_root_tdp = INVALID_PAGE;
11760 r = static_call(kvm_x86_vcpu_create)(vcpu);
11762 goto free_guest_fpu;
11764 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
11765 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
11766 kvm_xen_init_vcpu(vcpu);
11767 kvm_vcpu_mtrr_init(vcpu);
11769 kvm_set_tsc_khz(vcpu, vcpu->kvm->arch.default_tsc_khz);
11770 kvm_vcpu_reset(vcpu, false);
11771 kvm_init_mmu(vcpu);
11776 fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11778 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11779 free_wbinvd_dirty_mask:
11780 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11781 fail_free_mce_banks:
11782 kfree(vcpu->arch.mce_banks);
11783 kfree(vcpu->arch.mci_ctl2_banks);
11784 free_page((unsigned long)vcpu->arch.pio_data);
11786 kvm_free_lapic(vcpu);
11788 kvm_mmu_destroy(vcpu);
11792 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
11794 struct kvm *kvm = vcpu->kvm;
11796 if (mutex_lock_killable(&vcpu->mutex))
11799 kvm_synchronize_tsc(vcpu, 0);
11802 /* poll control enabled by default */
11803 vcpu->arch.msr_kvm_poll_control = 1;
11805 mutex_unlock(&vcpu->mutex);
11807 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
11808 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
11809 KVMCLOCK_SYNC_PERIOD);
11812 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
11816 kvmclock_reset(vcpu);
11818 static_call(kvm_x86_vcpu_free)(vcpu);
11820 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
11821 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
11822 fpu_free_guest_fpstate(&vcpu->arch.guest_fpu);
11824 kvm_xen_destroy_vcpu(vcpu);
11825 kvm_hv_vcpu_uninit(vcpu);
11826 kvm_pmu_destroy(vcpu);
11827 kfree(vcpu->arch.mce_banks);
11828 kfree(vcpu->arch.mci_ctl2_banks);
11829 kvm_free_lapic(vcpu);
11830 idx = srcu_read_lock(&vcpu->kvm->srcu);
11831 kvm_mmu_destroy(vcpu);
11832 srcu_read_unlock(&vcpu->kvm->srcu, idx);
11833 free_page((unsigned long)vcpu->arch.pio_data);
11834 kvfree(vcpu->arch.cpuid_entries);
11835 if (!lapic_in_kernel(vcpu))
11836 static_branch_dec(&kvm_has_noapic_vcpu);
11839 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
11841 struct kvm_cpuid_entry2 *cpuid_0x1;
11842 unsigned long old_cr0 = kvm_read_cr0(vcpu);
11843 unsigned long new_cr0;
11846 * Several of the "set" flows, e.g. ->set_cr0(), read other registers
11847 * to handle side effects. RESET emulation hits those flows and relies
11848 * on emulated/virtualized registers, including those that are loaded
11849 * into hardware, to be zeroed at vCPU creation. Use CRs as a sentinel
11850 * to detect improper or missing initialization.
11852 WARN_ON_ONCE(!init_event &&
11853 (old_cr0 || kvm_read_cr3(vcpu) || kvm_read_cr4(vcpu)));
11856 * SVM doesn't unconditionally VM-Exit on INIT and SHUTDOWN, thus it's
11857 * possible to INIT the vCPU while L2 is active. Force the vCPU back
11858 * into L1 as EFER.SVME is cleared on INIT (along with all other EFER
11859 * bits), i.e. virtualization is disabled.
11861 if (is_guest_mode(vcpu))
11862 kvm_leave_nested(vcpu);
11864 kvm_lapic_reset(vcpu, init_event);
11866 WARN_ON_ONCE(is_guest_mode(vcpu) || is_smm(vcpu));
11867 vcpu->arch.hflags = 0;
11869 vcpu->arch.smi_pending = 0;
11870 vcpu->arch.smi_count = 0;
11871 atomic_set(&vcpu->arch.nmi_queued, 0);
11872 vcpu->arch.nmi_pending = 0;
11873 vcpu->arch.nmi_injected = false;
11874 kvm_clear_interrupt_queue(vcpu);
11875 kvm_clear_exception_queue(vcpu);
11877 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
11878 kvm_update_dr0123(vcpu);
11879 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
11880 vcpu->arch.dr7 = DR7_FIXED_1;
11881 kvm_update_dr7(vcpu);
11883 vcpu->arch.cr2 = 0;
11885 kvm_make_request(KVM_REQ_EVENT, vcpu);
11886 vcpu->arch.apf.msr_en_val = 0;
11887 vcpu->arch.apf.msr_int_val = 0;
11888 vcpu->arch.st.msr_val = 0;
11890 kvmclock_reset(vcpu);
11892 kvm_clear_async_pf_completion_queue(vcpu);
11893 kvm_async_pf_hash_reset(vcpu);
11894 vcpu->arch.apf.halted = false;
11896 if (vcpu->arch.guest_fpu.fpstate && kvm_mpx_supported()) {
11897 struct fpstate *fpstate = vcpu->arch.guest_fpu.fpstate;
11900 * All paths that lead to INIT are required to load the guest's
11901 * FPU state (because most paths are buried in KVM_RUN).
11904 kvm_put_guest_fpu(vcpu);
11906 fpstate_clear_xstate_component(fpstate, XFEATURE_BNDREGS);
11907 fpstate_clear_xstate_component(fpstate, XFEATURE_BNDCSR);
11910 kvm_load_guest_fpu(vcpu);
11914 kvm_pmu_reset(vcpu);
11915 vcpu->arch.smbase = 0x30000;
11917 vcpu->arch.msr_misc_features_enables = 0;
11918 vcpu->arch.ia32_misc_enable_msr = MSR_IA32_MISC_ENABLE_PEBS_UNAVAIL |
11919 MSR_IA32_MISC_ENABLE_BTS_UNAVAIL;
11921 __kvm_set_xcr(vcpu, 0, XFEATURE_MASK_FP);
11922 __kvm_set_msr(vcpu, MSR_IA32_XSS, 0, true);
11925 /* All GPRs except RDX (handled below) are zeroed on RESET/INIT. */
11926 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
11927 kvm_register_mark_dirty(vcpu, VCPU_REGS_RSP);
11930 * Fall back to KVM's default Family/Model/Stepping of 0x600 (P6/Athlon)
11931 * if no CPUID match is found. Note, it's impossible to get a match at
11932 * RESET since KVM emulates RESET before exposing the vCPU to userspace,
11933 * i.e. it's impossible for kvm_find_cpuid_entry() to find a valid entry
11934 * on RESET. But, go through the motions in case that's ever remedied.
11936 cpuid_0x1 = kvm_find_cpuid_entry(vcpu, 1);
11937 kvm_rdx_write(vcpu, cpuid_0x1 ? cpuid_0x1->eax : 0x600);
11939 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
11941 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
11942 kvm_rip_write(vcpu, 0xfff0);
11944 vcpu->arch.cr3 = 0;
11945 kvm_register_mark_dirty(vcpu, VCPU_EXREG_CR3);
11948 * CR0.CD/NW are set on RESET, preserved on INIT. Note, some versions
11949 * of Intel's SDM list CD/NW as being set on INIT, but they contradict
11950 * (or qualify) that with a footnote stating that CD/NW are preserved.
11952 new_cr0 = X86_CR0_ET;
11954 new_cr0 |= (old_cr0 & (X86_CR0_NW | X86_CR0_CD));
11956 new_cr0 |= X86_CR0_NW | X86_CR0_CD;
11958 static_call(kvm_x86_set_cr0)(vcpu, new_cr0);
11959 static_call(kvm_x86_set_cr4)(vcpu, 0);
11960 static_call(kvm_x86_set_efer)(vcpu, 0);
11961 static_call(kvm_x86_update_exception_bitmap)(vcpu);
11964 * On the standard CR0/CR4/EFER modification paths, there are several
11965 * complex conditions determining whether the MMU has to be reset and/or
11966 * which PCIDs have to be flushed. However, CR0.WP and the paging-related
11967 * bits in CR4 and EFER are irrelevant if CR0.PG was '0'; and a reset+flush
11968 * is needed anyway if CR0.PG was '1' (which can only happen for INIT, as
11969 * CR0 will be '0' prior to RESET). So we only need to check CR0.PG here.
11971 if (old_cr0 & X86_CR0_PG) {
11972 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11973 kvm_mmu_reset_context(vcpu);
11977 * Intel's SDM states that all TLB entries are flushed on INIT. AMD's
11978 * APM states the TLBs are untouched by INIT, but it also states that
11979 * the TLBs are flushed on "External initialization of the processor."
11980 * Flush the guest TLB regardless of vendor, there is no meaningful
11981 * benefit in relying on the guest to flush the TLB immediately after
11982 * INIT. A spurious TLB flush is benign and likely negligible from a
11983 * performance perspective.
11986 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
11988 EXPORT_SYMBOL_GPL(kvm_vcpu_reset);
11990 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
11992 struct kvm_segment cs;
11994 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
11995 cs.selector = vector << 8;
11996 cs.base = vector << 12;
11997 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
11998 kvm_rip_write(vcpu, 0);
12000 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
12002 int kvm_arch_hardware_enable(void)
12005 struct kvm_vcpu *vcpu;
12010 bool stable, backwards_tsc = false;
12012 kvm_user_return_msr_cpu_online();
12014 ret = kvm_x86_check_processor_compatibility();
12018 ret = static_call(kvm_x86_hardware_enable)();
12022 local_tsc = rdtsc();
12023 stable = !kvm_check_tsc_unstable();
12024 list_for_each_entry(kvm, &vm_list, vm_list) {
12025 kvm_for_each_vcpu(i, vcpu, kvm) {
12026 if (!stable && vcpu->cpu == smp_processor_id())
12027 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
12028 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
12029 backwards_tsc = true;
12030 if (vcpu->arch.last_host_tsc > max_tsc)
12031 max_tsc = vcpu->arch.last_host_tsc;
12037 * Sometimes, even reliable TSCs go backwards. This happens on
12038 * platforms that reset TSC during suspend or hibernate actions, but
12039 * maintain synchronization. We must compensate. Fortunately, we can
12040 * detect that condition here, which happens early in CPU bringup,
12041 * before any KVM threads can be running. Unfortunately, we can't
12042 * bring the TSCs fully up to date with real time, as we aren't yet far
12043 * enough into CPU bringup that we know how much real time has actually
12044 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
12045 * variables that haven't been updated yet.
12047 * So we simply find the maximum observed TSC above, then record the
12048 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
12049 * the adjustment will be applied. Note that we accumulate
12050 * adjustments, in case multiple suspend cycles happen before some VCPU
12051 * gets a chance to run again. In the event that no KVM threads get a
12052 * chance to run, we will miss the entire elapsed period, as we'll have
12053 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
12054 * loose cycle time. This isn't too big a deal, since the loss will be
12055 * uniform across all VCPUs (not to mention the scenario is extremely
12056 * unlikely). It is possible that a second hibernate recovery happens
12057 * much faster than a first, causing the observed TSC here to be
12058 * smaller; this would require additional padding adjustment, which is
12059 * why we set last_host_tsc to the local tsc observed here.
12061 * N.B. - this code below runs only on platforms with reliable TSC,
12062 * as that is the only way backwards_tsc is set above. Also note
12063 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
12064 * have the same delta_cyc adjustment applied if backwards_tsc
12065 * is detected. Note further, this adjustment is only done once,
12066 * as we reset last_host_tsc on all VCPUs to stop this from being
12067 * called multiple times (one for each physical CPU bringup).
12069 * Platforms with unreliable TSCs don't have to deal with this, they
12070 * will be compensated by the logic in vcpu_load, which sets the TSC to
12071 * catchup mode. This will catchup all VCPUs to real time, but cannot
12072 * guarantee that they stay in perfect synchronization.
12074 if (backwards_tsc) {
12075 u64 delta_cyc = max_tsc - local_tsc;
12076 list_for_each_entry(kvm, &vm_list, vm_list) {
12077 kvm->arch.backwards_tsc_observed = true;
12078 kvm_for_each_vcpu(i, vcpu, kvm) {
12079 vcpu->arch.tsc_offset_adjustment += delta_cyc;
12080 vcpu->arch.last_host_tsc = local_tsc;
12081 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
12085 * We have to disable TSC offset matching.. if you were
12086 * booting a VM while issuing an S4 host suspend....
12087 * you may have some problem. Solving this issue is
12088 * left as an exercise to the reader.
12090 kvm->arch.last_tsc_nsec = 0;
12091 kvm->arch.last_tsc_write = 0;
12098 void kvm_arch_hardware_disable(void)
12100 static_call(kvm_x86_hardware_disable)();
12101 drop_user_return_notifiers();
12104 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
12106 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
12109 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
12111 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
12114 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
12115 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
12117 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
12119 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
12121 vcpu->arch.l1tf_flush_l1d = true;
12122 if (pmu->version && unlikely(pmu->event_count)) {
12123 pmu->need_cleanup = true;
12124 kvm_make_request(KVM_REQ_PMU, vcpu);
12126 static_call(kvm_x86_sched_in)(vcpu, cpu);
12129 void kvm_arch_free_vm(struct kvm *kvm)
12131 kfree(to_kvm_hv(kvm)->hv_pa_pg);
12132 __kvm_arch_free_vm(kvm);
12136 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
12139 unsigned long flags;
12144 ret = kvm_page_track_init(kvm);
12148 ret = kvm_mmu_init_vm(kvm);
12150 goto out_page_track;
12152 ret = static_call(kvm_x86_vm_init)(kvm);
12154 goto out_uninit_mmu;
12156 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
12157 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
12158 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
12160 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
12161 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
12162 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
12163 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
12164 &kvm->arch.irq_sources_bitmap);
12166 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
12167 mutex_init(&kvm->arch.apic_map_lock);
12168 seqcount_raw_spinlock_init(&kvm->arch.pvclock_sc, &kvm->arch.tsc_write_lock);
12169 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
12171 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
12172 pvclock_update_vm_gtod_copy(kvm);
12173 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
12175 kvm->arch.default_tsc_khz = max_tsc_khz ? : tsc_khz;
12176 kvm->arch.guest_can_read_msr_platform_info = true;
12177 kvm->arch.enable_pmu = enable_pmu;
12179 #if IS_ENABLED(CONFIG_HYPERV)
12180 spin_lock_init(&kvm->arch.hv_root_tdp_lock);
12181 kvm->arch.hv_root_tdp = INVALID_PAGE;
12184 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
12185 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
12187 kvm_apicv_init(kvm);
12188 kvm_hv_init_vm(kvm);
12189 kvm_xen_init_vm(kvm);
12194 kvm_mmu_uninit_vm(kvm);
12196 kvm_page_track_cleanup(kvm);
12201 int kvm_arch_post_init_vm(struct kvm *kvm)
12203 return kvm_mmu_post_init_vm(kvm);
12206 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
12209 kvm_mmu_unload(vcpu);
12213 static void kvm_unload_vcpu_mmus(struct kvm *kvm)
12216 struct kvm_vcpu *vcpu;
12218 kvm_for_each_vcpu(i, vcpu, kvm) {
12219 kvm_clear_async_pf_completion_queue(vcpu);
12220 kvm_unload_vcpu_mmu(vcpu);
12224 void kvm_arch_sync_events(struct kvm *kvm)
12226 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
12227 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
12232 * __x86_set_memory_region: Setup KVM internal memory slot
12234 * @kvm: the kvm pointer to the VM.
12235 * @id: the slot ID to setup.
12236 * @gpa: the GPA to install the slot (unused when @size == 0).
12237 * @size: the size of the slot. Set to zero to uninstall a slot.
12239 * This function helps to setup a KVM internal memory slot. Specify
12240 * @size > 0 to install a new slot, while @size == 0 to uninstall a
12241 * slot. The return code can be one of the following:
12243 * HVA: on success (uninstall will return a bogus HVA)
12246 * The caller should always use IS_ERR() to check the return value
12247 * before use. Note, the KVM internal memory slots are guaranteed to
12248 * remain valid and unchanged until the VM is destroyed, i.e., the
12249 * GPA->HVA translation will not change. However, the HVA is a user
12250 * address, i.e. its accessibility is not guaranteed, and must be
12251 * accessed via __copy_{to,from}_user().
12253 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
12257 unsigned long hva, old_npages;
12258 struct kvm_memslots *slots = kvm_memslots(kvm);
12259 struct kvm_memory_slot *slot;
12261 /* Called with kvm->slots_lock held. */
12262 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
12263 return ERR_PTR_USR(-EINVAL);
12265 slot = id_to_memslot(slots, id);
12267 if (slot && slot->npages)
12268 return ERR_PTR_USR(-EEXIST);
12271 * MAP_SHARED to prevent internal slot pages from being moved
12274 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
12275 MAP_SHARED | MAP_ANONYMOUS, 0);
12276 if (IS_ERR((void *)hva))
12277 return (void __user *)hva;
12279 if (!slot || !slot->npages)
12282 old_npages = slot->npages;
12283 hva = slot->userspace_addr;
12286 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
12287 struct kvm_userspace_memory_region m;
12289 m.slot = id | (i << 16);
12291 m.guest_phys_addr = gpa;
12292 m.userspace_addr = hva;
12293 m.memory_size = size;
12294 r = __kvm_set_memory_region(kvm, &m);
12296 return ERR_PTR_USR(r);
12300 vm_munmap(hva, old_npages * PAGE_SIZE);
12302 return (void __user *)hva;
12304 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
12306 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
12308 kvm_mmu_pre_destroy_vm(kvm);
12311 void kvm_arch_destroy_vm(struct kvm *kvm)
12313 if (current->mm == kvm->mm) {
12315 * Free memory regions allocated on behalf of userspace,
12316 * unless the memory map has changed due to process exit
12319 mutex_lock(&kvm->slots_lock);
12320 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
12322 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
12324 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
12325 mutex_unlock(&kvm->slots_lock);
12327 kvm_unload_vcpu_mmus(kvm);
12328 static_call_cond(kvm_x86_vm_destroy)(kvm);
12329 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
12330 kvm_pic_destroy(kvm);
12331 kvm_ioapic_destroy(kvm);
12332 kvm_destroy_vcpus(kvm);
12333 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
12334 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
12335 kvm_mmu_uninit_vm(kvm);
12336 kvm_page_track_cleanup(kvm);
12337 kvm_xen_destroy_vm(kvm);
12338 kvm_hv_destroy_vm(kvm);
12341 static void memslot_rmap_free(struct kvm_memory_slot *slot)
12345 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
12346 kvfree(slot->arch.rmap[i]);
12347 slot->arch.rmap[i] = NULL;
12351 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
12355 memslot_rmap_free(slot);
12357 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12358 kvfree(slot->arch.lpage_info[i - 1]);
12359 slot->arch.lpage_info[i - 1] = NULL;
12362 kvm_page_track_free_memslot(slot);
12365 int memslot_rmap_alloc(struct kvm_memory_slot *slot, unsigned long npages)
12367 const int sz = sizeof(*slot->arch.rmap[0]);
12370 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
12372 int lpages = __kvm_mmu_slot_lpages(slot, npages, level);
12374 if (slot->arch.rmap[i])
12377 slot->arch.rmap[i] = __vcalloc(lpages, sz, GFP_KERNEL_ACCOUNT);
12378 if (!slot->arch.rmap[i]) {
12379 memslot_rmap_free(slot);
12387 static int kvm_alloc_memslot_metadata(struct kvm *kvm,
12388 struct kvm_memory_slot *slot)
12390 unsigned long npages = slot->npages;
12394 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
12395 * old arrays will be freed by __kvm_set_memory_region() if installing
12396 * the new memslot is successful.
12398 memset(&slot->arch, 0, sizeof(slot->arch));
12400 if (kvm_memslots_have_rmaps(kvm)) {
12401 r = memslot_rmap_alloc(slot, npages);
12406 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12407 struct kvm_lpage_info *linfo;
12408 unsigned long ugfn;
12412 lpages = __kvm_mmu_slot_lpages(slot, npages, level);
12414 linfo = __vcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
12418 slot->arch.lpage_info[i - 1] = linfo;
12420 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
12421 linfo[0].disallow_lpage = 1;
12422 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
12423 linfo[lpages - 1].disallow_lpage = 1;
12424 ugfn = slot->userspace_addr >> PAGE_SHIFT;
12426 * If the gfn and userspace address are not aligned wrt each
12427 * other, disable large page support for this slot.
12429 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
12432 for (j = 0; j < lpages; ++j)
12433 linfo[j].disallow_lpage = 1;
12437 if (kvm_page_track_create_memslot(kvm, slot, npages))
12443 memslot_rmap_free(slot);
12445 for (i = 1; i < KVM_NR_PAGE_SIZES; ++i) {
12446 kvfree(slot->arch.lpage_info[i - 1]);
12447 slot->arch.lpage_info[i - 1] = NULL;
12452 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
12454 struct kvm_vcpu *vcpu;
12458 * memslots->generation has been incremented.
12459 * mmio generation may have reached its maximum value.
12461 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
12463 /* Force re-initialization of steal_time cache */
12464 kvm_for_each_vcpu(i, vcpu, kvm)
12465 kvm_vcpu_kick(vcpu);
12468 int kvm_arch_prepare_memory_region(struct kvm *kvm,
12469 const struct kvm_memory_slot *old,
12470 struct kvm_memory_slot *new,
12471 enum kvm_mr_change change)
12473 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE) {
12474 if ((new->base_gfn + new->npages - 1) > kvm_mmu_max_gfn())
12477 return kvm_alloc_memslot_metadata(kvm, new);
12480 if (change == KVM_MR_FLAGS_ONLY)
12481 memcpy(&new->arch, &old->arch, sizeof(old->arch));
12482 else if (WARN_ON_ONCE(change != KVM_MR_DELETE))
12489 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
12491 struct kvm_arch *ka = &kvm->arch;
12493 if (!kvm_x86_ops.cpu_dirty_log_size)
12496 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
12497 (!enable && --ka->cpu_dirty_logging_count == 0))
12498 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
12500 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
12503 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
12504 struct kvm_memory_slot *old,
12505 const struct kvm_memory_slot *new,
12506 enum kvm_mr_change change)
12508 u32 old_flags = old ? old->flags : 0;
12509 u32 new_flags = new ? new->flags : 0;
12510 bool log_dirty_pages = new_flags & KVM_MEM_LOG_DIRTY_PAGES;
12513 * Update CPU dirty logging if dirty logging is being toggled. This
12514 * applies to all operations.
12516 if ((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)
12517 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
12520 * Nothing more to do for RO slots (which can't be dirtied and can't be
12521 * made writable) or CREATE/MOVE/DELETE of a slot.
12523 * For a memslot with dirty logging disabled:
12524 * CREATE: No dirty mappings will already exist.
12525 * MOVE/DELETE: The old mappings will already have been cleaned up by
12526 * kvm_arch_flush_shadow_memslot()
12528 * For a memslot with dirty logging enabled:
12529 * CREATE: No shadow pages exist, thus nothing to write-protect
12530 * and no dirty bits to clear.
12531 * MOVE/DELETE: The old mappings will already have been cleaned up by
12532 * kvm_arch_flush_shadow_memslot().
12534 if ((change != KVM_MR_FLAGS_ONLY) || (new_flags & KVM_MEM_READONLY))
12538 * READONLY and non-flags changes were filtered out above, and the only
12539 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
12540 * logging isn't being toggled on or off.
12542 if (WARN_ON_ONCE(!((old_flags ^ new_flags) & KVM_MEM_LOG_DIRTY_PAGES)))
12545 if (!log_dirty_pages) {
12547 * Dirty logging tracks sptes in 4k granularity, meaning that
12548 * large sptes have to be split. If live migration succeeds,
12549 * the guest in the source machine will be destroyed and large
12550 * sptes will be created in the destination. However, if the
12551 * guest continues to run in the source machine (for example if
12552 * live migration fails), small sptes will remain around and
12553 * cause bad performance.
12555 * Scan sptes if dirty logging has been stopped, dropping those
12556 * which can be collapsed into a single large-page spte. Later
12557 * page faults will create the large-page sptes.
12559 kvm_mmu_zap_collapsible_sptes(kvm, new);
12562 * Initially-all-set does not require write protecting any page,
12563 * because they're all assumed to be dirty.
12565 if (kvm_dirty_log_manual_protect_and_init_set(kvm))
12568 if (READ_ONCE(eager_page_split))
12569 kvm_mmu_slot_try_split_huge_pages(kvm, new, PG_LEVEL_4K);
12571 if (kvm_x86_ops.cpu_dirty_log_size) {
12572 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
12573 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_2M);
12575 kvm_mmu_slot_remove_write_access(kvm, new, PG_LEVEL_4K);
12579 * Unconditionally flush the TLBs after enabling dirty logging.
12580 * A flush is almost always going to be necessary (see below),
12581 * and unconditionally flushing allows the helpers to omit
12582 * the subtly complex checks when removing write access.
12584 * Do the flush outside of mmu_lock to reduce the amount of
12585 * time mmu_lock is held. Flushing after dropping mmu_lock is
12586 * safe as KVM only needs to guarantee the slot is fully
12587 * write-protected before returning to userspace, i.e. before
12588 * userspace can consume the dirty status.
12590 * Flushing outside of mmu_lock requires KVM to be careful when
12591 * making decisions based on writable status of an SPTE, e.g. a
12592 * !writable SPTE doesn't guarantee a CPU can't perform writes.
12594 * Specifically, KVM also write-protects guest page tables to
12595 * monitor changes when using shadow paging, and must guarantee
12596 * no CPUs can write to those page before mmu_lock is dropped.
12597 * Because CPUs may have stale TLB entries at this point, a
12598 * !writable SPTE doesn't guarantee CPUs can't perform writes.
12600 * KVM also allows making SPTES writable outside of mmu_lock,
12601 * e.g. to allow dirty logging without taking mmu_lock.
12603 * To handle these scenarios, KVM uses a separate software-only
12604 * bit (MMU-writable) to track if a SPTE is !writable due to
12605 * a guest page table being write-protected (KVM clears the
12606 * MMU-writable flag when write-protecting for shadow paging).
12608 * The use of MMU-writable is also the primary motivation for
12609 * the unconditional flush. Because KVM must guarantee that a
12610 * CPU doesn't contain stale, writable TLB entries for a
12611 * !MMU-writable SPTE, KVM must flush if it encounters any
12612 * MMU-writable SPTE regardless of whether the actual hardware
12613 * writable bit was set. I.e. KVM is almost guaranteed to need
12614 * to flush, while unconditionally flushing allows the "remove
12615 * write access" helpers to ignore MMU-writable entirely.
12617 * See is_writable_pte() for more details (the case involving
12618 * access-tracked SPTEs is particularly relevant).
12620 kvm_arch_flush_remote_tlbs_memslot(kvm, new);
12624 void kvm_arch_commit_memory_region(struct kvm *kvm,
12625 struct kvm_memory_slot *old,
12626 const struct kvm_memory_slot *new,
12627 enum kvm_mr_change change)
12629 if (!kvm->arch.n_requested_mmu_pages &&
12630 (change == KVM_MR_CREATE || change == KVM_MR_DELETE)) {
12631 unsigned long nr_mmu_pages;
12633 nr_mmu_pages = kvm->nr_memslot_pages / KVM_MEMSLOT_PAGES_TO_MMU_PAGES_RATIO;
12634 nr_mmu_pages = max(nr_mmu_pages, KVM_MIN_ALLOC_MMU_PAGES);
12635 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
12638 kvm_mmu_slot_apply_flags(kvm, old, new, change);
12640 /* Free the arrays associated with the old memslot. */
12641 if (change == KVM_MR_MOVE)
12642 kvm_arch_free_memslot(kvm, old);
12645 void kvm_arch_flush_shadow_all(struct kvm *kvm)
12647 kvm_mmu_zap_all(kvm);
12650 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
12651 struct kvm_memory_slot *slot)
12653 kvm_page_track_flush_slot(kvm, slot);
12656 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
12658 return (is_guest_mode(vcpu) &&
12659 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
12662 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
12664 if (!list_empty_careful(&vcpu->async_pf.done))
12667 if (kvm_apic_has_pending_init_or_sipi(vcpu) &&
12668 kvm_apic_init_sipi_allowed(vcpu))
12671 if (vcpu->arch.pv.pv_unhalted)
12674 if (kvm_is_exception_pending(vcpu))
12677 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12678 (vcpu->arch.nmi_pending &&
12679 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
12682 #ifdef CONFIG_KVM_SMM
12683 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
12684 (vcpu->arch.smi_pending &&
12685 static_call(kvm_x86_smi_allowed)(vcpu, false)))
12689 if (kvm_arch_interrupt_allowed(vcpu) &&
12690 (kvm_cpu_has_interrupt(vcpu) ||
12691 kvm_guest_apic_has_interrupt(vcpu)))
12694 if (kvm_hv_has_stimer_pending(vcpu))
12697 if (is_guest_mode(vcpu) &&
12698 kvm_x86_ops.nested_ops->has_events &&
12699 kvm_x86_ops.nested_ops->has_events(vcpu))
12702 if (kvm_xen_has_pending_events(vcpu))
12708 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
12710 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
12713 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
12715 if (kvm_vcpu_apicv_active(vcpu) &&
12716 static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
12722 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
12724 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
12727 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
12728 #ifdef CONFIG_KVM_SMM
12729 kvm_test_request(KVM_REQ_SMI, vcpu) ||
12731 kvm_test_request(KVM_REQ_EVENT, vcpu))
12734 return kvm_arch_dy_has_pending_interrupt(vcpu);
12737 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
12739 if (vcpu->arch.guest_state_protected)
12742 return vcpu->arch.preempted_in_kernel;
12745 unsigned long kvm_arch_vcpu_get_ip(struct kvm_vcpu *vcpu)
12747 return kvm_rip_read(vcpu);
12750 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
12752 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
12755 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
12757 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
12760 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
12762 /* Can't read the RIP when guest state is protected, just return 0 */
12763 if (vcpu->arch.guest_state_protected)
12766 if (is_64_bit_mode(vcpu))
12767 return kvm_rip_read(vcpu);
12768 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
12769 kvm_rip_read(vcpu));
12771 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
12773 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
12775 return kvm_get_linear_rip(vcpu) == linear_rip;
12777 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
12779 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
12781 unsigned long rflags;
12783 rflags = static_call(kvm_x86_get_rflags)(vcpu);
12784 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
12785 rflags &= ~X86_EFLAGS_TF;
12788 EXPORT_SYMBOL_GPL(kvm_get_rflags);
12790 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12792 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
12793 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
12794 rflags |= X86_EFLAGS_TF;
12795 static_call(kvm_x86_set_rflags)(vcpu, rflags);
12798 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
12800 __kvm_set_rflags(vcpu, rflags);
12801 kvm_make_request(KVM_REQ_EVENT, vcpu);
12803 EXPORT_SYMBOL_GPL(kvm_set_rflags);
12805 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
12807 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
12809 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
12812 static inline u32 kvm_async_pf_next_probe(u32 key)
12814 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
12817 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12819 u32 key = kvm_async_pf_hash_fn(gfn);
12821 while (vcpu->arch.apf.gfns[key] != ~0)
12822 key = kvm_async_pf_next_probe(key);
12824 vcpu->arch.apf.gfns[key] = gfn;
12827 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
12830 u32 key = kvm_async_pf_hash_fn(gfn);
12832 for (i = 0; i < ASYNC_PF_PER_VCPU &&
12833 (vcpu->arch.apf.gfns[key] != gfn &&
12834 vcpu->arch.apf.gfns[key] != ~0); i++)
12835 key = kvm_async_pf_next_probe(key);
12840 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12842 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
12845 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
12849 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
12851 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
12855 vcpu->arch.apf.gfns[i] = ~0;
12857 j = kvm_async_pf_next_probe(j);
12858 if (vcpu->arch.apf.gfns[j] == ~0)
12860 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
12862 * k lies cyclically in ]i,j]
12864 * |....j i.k.| or |.k..j i...|
12866 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
12867 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
12872 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
12874 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
12876 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
12880 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
12882 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12884 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12885 &token, offset, sizeof(token));
12888 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
12890 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
12893 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
12894 &val, offset, sizeof(val)))
12900 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
12903 if (!kvm_pv_async_pf_enabled(vcpu))
12906 if (vcpu->arch.apf.send_user_only &&
12907 static_call(kvm_x86_get_cpl)(vcpu) == 0)
12910 if (is_guest_mode(vcpu)) {
12912 * L1 needs to opt into the special #PF vmexits that are
12913 * used to deliver async page faults.
12915 return vcpu->arch.apf.delivery_as_pf_vmexit;
12918 * Play it safe in case the guest temporarily disables paging.
12919 * The real mode IDT in particular is unlikely to have a #PF
12922 return is_paging(vcpu);
12926 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
12928 if (unlikely(!lapic_in_kernel(vcpu) ||
12929 kvm_event_needs_reinjection(vcpu) ||
12930 kvm_is_exception_pending(vcpu)))
12933 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
12937 * If interrupts are off we cannot even use an artificial
12940 return kvm_arch_interrupt_allowed(vcpu);
12943 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
12944 struct kvm_async_pf *work)
12946 struct x86_exception fault;
12948 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
12949 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
12951 if (kvm_can_deliver_async_pf(vcpu) &&
12952 !apf_put_user_notpresent(vcpu)) {
12953 fault.vector = PF_VECTOR;
12954 fault.error_code_valid = true;
12955 fault.error_code = 0;
12956 fault.nested_page_fault = false;
12957 fault.address = work->arch.token;
12958 fault.async_page_fault = true;
12959 kvm_inject_page_fault(vcpu, &fault);
12963 * It is not possible to deliver a paravirtualized asynchronous
12964 * page fault, but putting the guest in an artificial halt state
12965 * can be beneficial nevertheless: if an interrupt arrives, we
12966 * can deliver it timely and perhaps the guest will schedule
12967 * another process. When the instruction that triggered a page
12968 * fault is retried, hopefully the page will be ready in the host.
12970 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
12975 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
12976 struct kvm_async_pf *work)
12978 struct kvm_lapic_irq irq = {
12979 .delivery_mode = APIC_DM_FIXED,
12980 .vector = vcpu->arch.apf.vec
12983 if (work->wakeup_all)
12984 work->arch.token = ~0; /* broadcast wakeup */
12986 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
12987 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
12989 if ((work->wakeup_all || work->notpresent_injected) &&
12990 kvm_pv_async_pf_enabled(vcpu) &&
12991 !apf_put_user_ready(vcpu, work->arch.token)) {
12992 vcpu->arch.apf.pageready_pending = true;
12993 kvm_apic_set_irq(vcpu, &irq, NULL);
12996 vcpu->arch.apf.halted = false;
12997 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
13000 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
13002 kvm_make_request(KVM_REQ_APF_READY, vcpu);
13003 if (!vcpu->arch.apf.pageready_pending)
13004 kvm_vcpu_kick(vcpu);
13007 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
13009 if (!kvm_pv_async_pf_enabled(vcpu))
13012 return kvm_lapic_enabled(vcpu) && apf_pageready_slot_free(vcpu);
13015 void kvm_arch_start_assignment(struct kvm *kvm)
13017 if (atomic_inc_return(&kvm->arch.assigned_device_count) == 1)
13018 static_call_cond(kvm_x86_pi_start_assignment)(kvm);
13020 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
13022 void kvm_arch_end_assignment(struct kvm *kvm)
13024 atomic_dec(&kvm->arch.assigned_device_count);
13026 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
13028 bool noinstr kvm_arch_has_assigned_device(struct kvm *kvm)
13030 return arch_atomic_read(&kvm->arch.assigned_device_count);
13032 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
13034 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
13036 atomic_inc(&kvm->arch.noncoherent_dma_count);
13038 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
13040 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
13042 atomic_dec(&kvm->arch.noncoherent_dma_count);
13044 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
13046 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
13048 return atomic_read(&kvm->arch.noncoherent_dma_count);
13050 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
13052 bool kvm_arch_has_irq_bypass(void)
13057 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
13058 struct irq_bypass_producer *prod)
13060 struct kvm_kernel_irqfd *irqfd =
13061 container_of(cons, struct kvm_kernel_irqfd, consumer);
13064 irqfd->producer = prod;
13065 kvm_arch_start_assignment(irqfd->kvm);
13066 ret = static_call(kvm_x86_pi_update_irte)(irqfd->kvm,
13067 prod->irq, irqfd->gsi, 1);
13070 kvm_arch_end_assignment(irqfd->kvm);
13075 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
13076 struct irq_bypass_producer *prod)
13079 struct kvm_kernel_irqfd *irqfd =
13080 container_of(cons, struct kvm_kernel_irqfd, consumer);
13082 WARN_ON(irqfd->producer != prod);
13083 irqfd->producer = NULL;
13086 * When producer of consumer is unregistered, we change back to
13087 * remapped mode, so we can re-use the current implementation
13088 * when the irq is masked/disabled or the consumer side (KVM
13089 * int this case doesn't want to receive the interrupts.
13091 ret = static_call(kvm_x86_pi_update_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
13093 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
13094 " fails: %d\n", irqfd->consumer.token, ret);
13096 kvm_arch_end_assignment(irqfd->kvm);
13099 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
13100 uint32_t guest_irq, bool set)
13102 return static_call(kvm_x86_pi_update_irte)(kvm, host_irq, guest_irq, set);
13105 bool kvm_arch_irqfd_route_changed(struct kvm_kernel_irq_routing_entry *old,
13106 struct kvm_kernel_irq_routing_entry *new)
13108 if (new->type != KVM_IRQ_ROUTING_MSI)
13111 return !!memcmp(&old->msi, &new->msi, sizeof(new->msi));
13114 bool kvm_vector_hashing_enabled(void)
13116 return vector_hashing;
13119 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
13121 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
13123 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
13126 int kvm_spec_ctrl_test_value(u64 value)
13129 * test that setting IA32_SPEC_CTRL to given value
13130 * is allowed by the host processor
13134 unsigned long flags;
13137 local_irq_save(flags);
13139 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
13141 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
13144 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
13146 local_irq_restore(flags);
13150 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
13152 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
13154 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
13155 struct x86_exception fault;
13156 u64 access = error_code &
13157 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
13159 if (!(error_code & PFERR_PRESENT_MASK) ||
13160 mmu->gva_to_gpa(vcpu, mmu, gva, access, &fault) != INVALID_GPA) {
13162 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
13163 * tables probably do not match the TLB. Just proceed
13164 * with the error code that the processor gave.
13166 fault.vector = PF_VECTOR;
13167 fault.error_code_valid = true;
13168 fault.error_code = error_code;
13169 fault.nested_page_fault = false;
13170 fault.address = gva;
13171 fault.async_page_fault = false;
13173 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
13175 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
13178 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
13179 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
13180 * indicates whether exit to userspace is needed.
13182 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
13183 struct x86_exception *e)
13185 if (r == X86EMUL_PROPAGATE_FAULT) {
13186 if (KVM_BUG_ON(!e, vcpu->kvm))
13189 kvm_inject_emulated_page_fault(vcpu, e);
13194 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
13195 * while handling a VMX instruction KVM could've handled the request
13196 * correctly by exiting to userspace and performing I/O but there
13197 * doesn't seem to be a real use-case behind such requests, just return
13198 * KVM_EXIT_INTERNAL_ERROR for now.
13200 kvm_prepare_emulation_failure_exit(vcpu);
13204 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
13206 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
13209 struct x86_exception e;
13216 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
13217 if (r != X86EMUL_CONTINUE)
13218 return kvm_handle_memory_failure(vcpu, r, &e);
13220 if (operand.pcid >> 12 != 0) {
13221 kvm_inject_gp(vcpu, 0);
13225 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
13228 case INVPCID_TYPE_INDIV_ADDR:
13229 if ((!pcid_enabled && (operand.pcid != 0)) ||
13230 is_noncanonical_address(operand.gla, vcpu)) {
13231 kvm_inject_gp(vcpu, 0);
13234 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
13235 return kvm_skip_emulated_instruction(vcpu);
13237 case INVPCID_TYPE_SINGLE_CTXT:
13238 if (!pcid_enabled && (operand.pcid != 0)) {
13239 kvm_inject_gp(vcpu, 0);
13243 kvm_invalidate_pcid(vcpu, operand.pcid);
13244 return kvm_skip_emulated_instruction(vcpu);
13246 case INVPCID_TYPE_ALL_NON_GLOBAL:
13248 * Currently, KVM doesn't mark global entries in the shadow
13249 * page tables, so a non-global flush just degenerates to a
13250 * global flush. If needed, we could optimize this later by
13251 * keeping track of global entries in shadow page tables.
13255 case INVPCID_TYPE_ALL_INCL_GLOBAL:
13256 kvm_make_request(KVM_REQ_TLB_FLUSH_GUEST, vcpu);
13257 return kvm_skip_emulated_instruction(vcpu);
13260 kvm_inject_gp(vcpu, 0);
13264 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
13266 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
13268 struct kvm_run *run = vcpu->run;
13269 struct kvm_mmio_fragment *frag;
13272 BUG_ON(!vcpu->mmio_needed);
13274 /* Complete previous fragment */
13275 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
13276 len = min(8u, frag->len);
13277 if (!vcpu->mmio_is_write)
13278 memcpy(frag->data, run->mmio.data, len);
13280 if (frag->len <= 8) {
13281 /* Switch to the next fragment. */
13283 vcpu->mmio_cur_fragment++;
13285 /* Go forward to the next mmio piece. */
13291 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
13292 vcpu->mmio_needed = 0;
13294 // VMG change, at this point, we're always done
13295 // RIP has already been advanced
13299 // More MMIO is needed
13300 run->mmio.phys_addr = frag->gpa;
13301 run->mmio.len = min(8u, frag->len);
13302 run->mmio.is_write = vcpu->mmio_is_write;
13303 if (run->mmio.is_write)
13304 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
13305 run->exit_reason = KVM_EXIT_MMIO;
13307 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13312 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
13316 struct kvm_mmio_fragment *frag;
13321 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
13322 if (handled == bytes)
13329 /*TODO: Check if need to increment number of frags */
13330 frag = vcpu->mmio_fragments;
13331 vcpu->mmio_nr_fragments = 1;
13336 vcpu->mmio_needed = 1;
13337 vcpu->mmio_cur_fragment = 0;
13339 vcpu->run->mmio.phys_addr = gpa;
13340 vcpu->run->mmio.len = min(8u, frag->len);
13341 vcpu->run->mmio.is_write = 1;
13342 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
13343 vcpu->run->exit_reason = KVM_EXIT_MMIO;
13345 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13349 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
13351 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
13355 struct kvm_mmio_fragment *frag;
13360 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
13361 if (handled == bytes)
13368 /*TODO: Check if need to increment number of frags */
13369 frag = vcpu->mmio_fragments;
13370 vcpu->mmio_nr_fragments = 1;
13375 vcpu->mmio_needed = 1;
13376 vcpu->mmio_cur_fragment = 0;
13378 vcpu->run->mmio.phys_addr = gpa;
13379 vcpu->run->mmio.len = min(8u, frag->len);
13380 vcpu->run->mmio.is_write = 0;
13381 vcpu->run->exit_reason = KVM_EXIT_MMIO;
13383 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
13387 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
13389 static void advance_sev_es_emulated_pio(struct kvm_vcpu *vcpu, unsigned count, int size)
13391 vcpu->arch.sev_pio_count -= count;
13392 vcpu->arch.sev_pio_data += count * size;
13395 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
13396 unsigned int port);
13398 static int complete_sev_es_emulated_outs(struct kvm_vcpu *vcpu)
13400 int size = vcpu->arch.pio.size;
13401 int port = vcpu->arch.pio.port;
13403 vcpu->arch.pio.count = 0;
13404 if (vcpu->arch.sev_pio_count)
13405 return kvm_sev_es_outs(vcpu, size, port);
13409 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
13413 unsigned int count =
13414 min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
13415 int ret = emulator_pio_out(vcpu, size, port, vcpu->arch.sev_pio_data, count);
13417 /* memcpy done already by emulator_pio_out. */
13418 advance_sev_es_emulated_pio(vcpu, count, size);
13422 /* Emulation done by the kernel. */
13423 if (!vcpu->arch.sev_pio_count)
13427 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_outs;
13431 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
13432 unsigned int port);
13434 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
13436 unsigned count = vcpu->arch.pio.count;
13437 int size = vcpu->arch.pio.size;
13438 int port = vcpu->arch.pio.port;
13440 complete_emulator_pio_in(vcpu, vcpu->arch.sev_pio_data);
13441 advance_sev_es_emulated_pio(vcpu, count, size);
13442 if (vcpu->arch.sev_pio_count)
13443 return kvm_sev_es_ins(vcpu, size, port);
13447 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
13451 unsigned int count =
13452 min_t(unsigned int, PAGE_SIZE / size, vcpu->arch.sev_pio_count);
13453 if (!emulator_pio_in(vcpu, size, port, vcpu->arch.sev_pio_data, count))
13456 /* Emulation done by the kernel. */
13457 advance_sev_es_emulated_pio(vcpu, count, size);
13458 if (!vcpu->arch.sev_pio_count)
13462 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
13466 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
13467 unsigned int port, void *data, unsigned int count,
13470 vcpu->arch.sev_pio_data = data;
13471 vcpu->arch.sev_pio_count = count;
13472 return in ? kvm_sev_es_ins(vcpu, size, port)
13473 : kvm_sev_es_outs(vcpu, size, port);
13475 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
13477 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
13478 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
13479 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
13480 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
13481 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
13482 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
13483 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
13484 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter);
13485 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
13486 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
13487 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
13488 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
13489 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
13490 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
13491 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
13492 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
13493 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
13494 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
13495 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
13496 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
13497 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
13498 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
13499 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_kick_vcpu_slowpath);
13500 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_doorbell);
13501 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_accept_irq);
13502 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
13503 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
13504 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
13505 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);
13507 static int __init kvm_x86_init(void)
13509 kvm_mmu_x86_module_init();
13512 module_init(kvm_x86_init);
13514 static void __exit kvm_x86_exit(void)
13517 * If module_init() is implemented, module_exit() must also be
13518 * implemented to allow module unload.
13521 module_exit(kvm_x86_exit);