1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * derived from drivers/kvm/kvm_main.c
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
19 #include <linux/kvm_host.h>
25 #include "kvm_cache_regs.h"
26 #include "kvm_emulate.h"
34 #include <linux/clocksource.h>
35 #include <linux/interrupt.h>
36 #include <linux/kvm.h>
38 #include <linux/vmalloc.h>
39 #include <linux/export.h>
40 #include <linux/moduleparam.h>
41 #include <linux/mman.h>
42 #include <linux/highmem.h>
43 #include <linux/iommu.h>
44 #include <linux/intel-iommu.h>
45 #include <linux/cpufreq.h>
46 #include <linux/user-return-notifier.h>
47 #include <linux/srcu.h>
48 #include <linux/slab.h>
49 #include <linux/perf_event.h>
50 #include <linux/uaccess.h>
51 #include <linux/hash.h>
52 #include <linux/pci.h>
53 #include <linux/timekeeper_internal.h>
54 #include <linux/pvclock_gtod.h>
55 #include <linux/kvm_irqfd.h>
56 #include <linux/irqbypass.h>
57 #include <linux/sched/stat.h>
58 #include <linux/sched/isolation.h>
59 #include <linux/mem_encrypt.h>
60 #include <linux/entry-kvm.h>
62 #include <trace/events/kvm.h>
64 #include <asm/debugreg.h>
68 #include <linux/kernel_stat.h>
69 #include <asm/fpu/internal.h> /* Ugh! */
70 #include <asm/pvclock.h>
71 #include <asm/div64.h>
72 #include <asm/irq_remapping.h>
73 #include <asm/mshyperv.h>
74 #include <asm/hypervisor.h>
75 #include <asm/tlbflush.h>
76 #include <asm/intel_pt.h>
77 #include <asm/emulate_prefix.h>
79 #include <clocksource/hyperv_timer.h>
81 #define CREATE_TRACE_POINTS
84 #define MAX_IO_MSRS 256
85 #define KVM_MAX_MCE_BANKS 32
86 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
87 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
89 #define emul_to_vcpu(ctxt) \
90 ((struct kvm_vcpu *)(ctxt)->vcpu)
93 * - enable syscall per default because its emulated by KVM
94 * - enable LME and LMA per default on 64 bit KVM
98 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
100 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
103 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
105 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
106 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
108 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
109 static void process_nmi(struct kvm_vcpu *vcpu);
110 static void process_smi(struct kvm_vcpu *vcpu);
111 static void enter_smm(struct kvm_vcpu *vcpu);
112 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
113 static void store_regs(struct kvm_vcpu *vcpu);
114 static int sync_regs(struct kvm_vcpu *vcpu);
116 struct kvm_x86_ops kvm_x86_ops __read_mostly;
117 EXPORT_SYMBOL_GPL(kvm_x86_ops);
119 #define KVM_X86_OP(func) \
120 DEFINE_STATIC_CALL_NULL(kvm_x86_##func, \
121 *(((struct kvm_x86_ops *)0)->func));
122 #define KVM_X86_OP_NULL KVM_X86_OP
123 #include <asm/kvm-x86-ops.h>
124 EXPORT_STATIC_CALL_GPL(kvm_x86_get_cs_db_l_bits);
125 EXPORT_STATIC_CALL_GPL(kvm_x86_cache_reg);
126 EXPORT_STATIC_CALL_GPL(kvm_x86_tlb_flush_current);
128 static bool __read_mostly ignore_msrs = 0;
129 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
131 bool __read_mostly report_ignored_msrs = true;
132 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
133 EXPORT_SYMBOL_GPL(report_ignored_msrs);
135 unsigned int min_timer_period_us = 200;
136 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
138 static bool __read_mostly kvmclock_periodic_sync = true;
139 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
141 bool __read_mostly kvm_has_tsc_control;
142 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
143 u32 __read_mostly kvm_max_guest_tsc_khz;
144 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
145 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
146 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
147 u64 __read_mostly kvm_max_tsc_scaling_ratio;
148 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
149 u64 __read_mostly kvm_default_tsc_scaling_ratio;
150 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
151 bool __read_mostly kvm_has_bus_lock_exit;
152 EXPORT_SYMBOL_GPL(kvm_has_bus_lock_exit);
154 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
155 static u32 __read_mostly tsc_tolerance_ppm = 250;
156 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
159 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
160 * adaptive tuning starting from default advancment of 1000ns. '0' disables
161 * advancement entirely. Any other value is used as-is and disables adaptive
162 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
164 static int __read_mostly lapic_timer_advance_ns = -1;
165 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
167 static bool __read_mostly vector_hashing = true;
168 module_param(vector_hashing, bool, S_IRUGO);
170 bool __read_mostly enable_vmware_backdoor = false;
171 module_param(enable_vmware_backdoor, bool, S_IRUGO);
172 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
174 static bool __read_mostly force_emulation_prefix = false;
175 module_param(force_emulation_prefix, bool, S_IRUGO);
177 int __read_mostly pi_inject_timer = -1;
178 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
181 * Restoring the host value for MSRs that are only consumed when running in
182 * usermode, e.g. SYSCALL MSRs and TSC_AUX, can be deferred until the CPU
183 * returns to userspace, i.e. the kernel can run with the guest's value.
185 #define KVM_MAX_NR_USER_RETURN_MSRS 16
187 struct kvm_user_return_msrs_global {
189 u32 msrs[KVM_MAX_NR_USER_RETURN_MSRS];
192 struct kvm_user_return_msrs {
193 struct user_return_notifier urn;
195 struct kvm_user_return_msr_values {
198 } values[KVM_MAX_NR_USER_RETURN_MSRS];
201 static struct kvm_user_return_msrs_global __read_mostly user_return_msrs_global;
202 static struct kvm_user_return_msrs __percpu *user_return_msrs;
204 #define KVM_SUPPORTED_XCR0 (XFEATURE_MASK_FP | XFEATURE_MASK_SSE \
205 | XFEATURE_MASK_YMM | XFEATURE_MASK_BNDREGS \
206 | XFEATURE_MASK_BNDCSR | XFEATURE_MASK_AVX512 \
207 | XFEATURE_MASK_PKRU)
209 u64 __read_mostly host_efer;
210 EXPORT_SYMBOL_GPL(host_efer);
212 bool __read_mostly allow_smaller_maxphyaddr = 0;
213 EXPORT_SYMBOL_GPL(allow_smaller_maxphyaddr);
215 u64 __read_mostly host_xss;
216 EXPORT_SYMBOL_GPL(host_xss);
217 u64 __read_mostly supported_xss;
218 EXPORT_SYMBOL_GPL(supported_xss);
220 struct kvm_stats_debugfs_item debugfs_entries[] = {
221 VCPU_STAT("pf_fixed", pf_fixed),
222 VCPU_STAT("pf_guest", pf_guest),
223 VCPU_STAT("tlb_flush", tlb_flush),
224 VCPU_STAT("invlpg", invlpg),
225 VCPU_STAT("exits", exits),
226 VCPU_STAT("io_exits", io_exits),
227 VCPU_STAT("mmio_exits", mmio_exits),
228 VCPU_STAT("signal_exits", signal_exits),
229 VCPU_STAT("irq_window", irq_window_exits),
230 VCPU_STAT("nmi_window", nmi_window_exits),
231 VCPU_STAT("halt_exits", halt_exits),
232 VCPU_STAT("halt_successful_poll", halt_successful_poll),
233 VCPU_STAT("halt_attempted_poll", halt_attempted_poll),
234 VCPU_STAT("halt_poll_invalid", halt_poll_invalid),
235 VCPU_STAT("halt_wakeup", halt_wakeup),
236 VCPU_STAT("hypercalls", hypercalls),
237 VCPU_STAT("request_irq", request_irq_exits),
238 VCPU_STAT("irq_exits", irq_exits),
239 VCPU_STAT("host_state_reload", host_state_reload),
240 VCPU_STAT("fpu_reload", fpu_reload),
241 VCPU_STAT("insn_emulation", insn_emulation),
242 VCPU_STAT("insn_emulation_fail", insn_emulation_fail),
243 VCPU_STAT("irq_injections", irq_injections),
244 VCPU_STAT("nmi_injections", nmi_injections),
245 VCPU_STAT("req_event", req_event),
246 VCPU_STAT("l1d_flush", l1d_flush),
247 VCPU_STAT("halt_poll_success_ns", halt_poll_success_ns),
248 VCPU_STAT("halt_poll_fail_ns", halt_poll_fail_ns),
249 VCPU_STAT("nested_run", nested_run),
250 VCPU_STAT("directed_yield_attempted", directed_yield_attempted),
251 VCPU_STAT("directed_yield_successful", directed_yield_successful),
252 VM_STAT("mmu_shadow_zapped", mmu_shadow_zapped),
253 VM_STAT("mmu_pte_write", mmu_pte_write),
254 VM_STAT("mmu_pde_zapped", mmu_pde_zapped),
255 VM_STAT("mmu_flooded", mmu_flooded),
256 VM_STAT("mmu_recycled", mmu_recycled),
257 VM_STAT("mmu_cache_miss", mmu_cache_miss),
258 VM_STAT("mmu_unsync", mmu_unsync),
259 VM_STAT("remote_tlb_flush", remote_tlb_flush),
260 VM_STAT("largepages", lpages, .mode = 0444),
261 VM_STAT("nx_largepages_splitted", nx_lpage_splits, .mode = 0444),
262 VM_STAT("max_mmu_page_hash_collisions", max_mmu_page_hash_collisions),
266 u64 __read_mostly host_xcr0;
267 u64 __read_mostly supported_xcr0;
268 EXPORT_SYMBOL_GPL(supported_xcr0);
270 static struct kmem_cache *x86_fpu_cache;
272 static struct kmem_cache *x86_emulator_cache;
275 * When called, it means the previous get/set msr reached an invalid msr.
276 * Return true if we want to ignore/silent this failed msr access.
278 static bool kvm_msr_ignored_check(u32 msr, u64 data, bool write)
280 const char *op = write ? "wrmsr" : "rdmsr";
283 if (report_ignored_msrs)
284 kvm_pr_unimpl("ignored %s: 0x%x data 0x%llx\n",
289 kvm_debug_ratelimited("unhandled %s: 0x%x data 0x%llx\n",
295 static struct kmem_cache *kvm_alloc_emulator_cache(void)
297 unsigned int useroffset = offsetof(struct x86_emulate_ctxt, src);
298 unsigned int size = sizeof(struct x86_emulate_ctxt);
300 return kmem_cache_create_usercopy("x86_emulator", size,
301 __alignof__(struct x86_emulate_ctxt),
302 SLAB_ACCOUNT, useroffset,
303 size - useroffset, NULL);
306 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
308 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
311 for (i = 0; i < ASYNC_PF_PER_VCPU; i++)
312 vcpu->arch.apf.gfns[i] = ~0;
315 static void kvm_on_user_return(struct user_return_notifier *urn)
318 struct kvm_user_return_msrs *msrs
319 = container_of(urn, struct kvm_user_return_msrs, urn);
320 struct kvm_user_return_msr_values *values;
324 * Disabling irqs at this point since the following code could be
325 * interrupted and executed through kvm_arch_hardware_disable()
327 local_irq_save(flags);
328 if (msrs->registered) {
329 msrs->registered = false;
330 user_return_notifier_unregister(urn);
332 local_irq_restore(flags);
333 for (slot = 0; slot < user_return_msrs_global.nr; ++slot) {
334 values = &msrs->values[slot];
335 if (values->host != values->curr) {
336 wrmsrl(user_return_msrs_global.msrs[slot], values->host);
337 values->curr = values->host;
342 void kvm_define_user_return_msr(unsigned slot, u32 msr)
344 BUG_ON(slot >= KVM_MAX_NR_USER_RETURN_MSRS);
345 user_return_msrs_global.msrs[slot] = msr;
346 if (slot >= user_return_msrs_global.nr)
347 user_return_msrs_global.nr = slot + 1;
349 EXPORT_SYMBOL_GPL(kvm_define_user_return_msr);
351 static void kvm_user_return_msr_cpu_online(void)
353 unsigned int cpu = smp_processor_id();
354 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
358 for (i = 0; i < user_return_msrs_global.nr; ++i) {
359 rdmsrl_safe(user_return_msrs_global.msrs[i], &value);
360 msrs->values[i].host = value;
361 msrs->values[i].curr = value;
365 int kvm_set_user_return_msr(unsigned slot, u64 value, u64 mask)
367 unsigned int cpu = smp_processor_id();
368 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
371 value = (value & mask) | (msrs->values[slot].host & ~mask);
372 if (value == msrs->values[slot].curr)
374 err = wrmsrl_safe(user_return_msrs_global.msrs[slot], value);
378 msrs->values[slot].curr = value;
379 if (!msrs->registered) {
380 msrs->urn.on_user_return = kvm_on_user_return;
381 user_return_notifier_register(&msrs->urn);
382 msrs->registered = true;
386 EXPORT_SYMBOL_GPL(kvm_set_user_return_msr);
388 static void drop_user_return_notifiers(void)
390 unsigned int cpu = smp_processor_id();
391 struct kvm_user_return_msrs *msrs = per_cpu_ptr(user_return_msrs, cpu);
393 if (msrs->registered)
394 kvm_on_user_return(&msrs->urn);
397 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
399 return vcpu->arch.apic_base;
401 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
403 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
405 return kvm_apic_mode(kvm_get_apic_base(vcpu));
407 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
409 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
411 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
412 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
413 u64 reserved_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu) | 0x2ff |
414 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
416 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
418 if (!msr_info->host_initiated) {
419 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
421 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
425 kvm_lapic_set_base(vcpu, msr_info->data);
426 kvm_recalculate_apic_map(vcpu->kvm);
429 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
431 asmlinkage __visible noinstr void kvm_spurious_fault(void)
433 /* Fault while not rebooting. We want the trace. */
434 BUG_ON(!kvm_rebooting);
436 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
438 #define EXCPT_BENIGN 0
439 #define EXCPT_CONTRIBUTORY 1
442 static int exception_class(int vector)
452 return EXCPT_CONTRIBUTORY;
459 #define EXCPT_FAULT 0
461 #define EXCPT_ABORT 2
462 #define EXCPT_INTERRUPT 3
464 static int exception_type(int vector)
468 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
469 return EXCPT_INTERRUPT;
473 /* #DB is trap, as instruction watchpoints are handled elsewhere */
474 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
477 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
480 /* Reserved exceptions will result in fault */
484 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
486 unsigned nr = vcpu->arch.exception.nr;
487 bool has_payload = vcpu->arch.exception.has_payload;
488 unsigned long payload = vcpu->arch.exception.payload;
496 * "Certain debug exceptions may clear bit 0-3. The
497 * remaining contents of the DR6 register are never
498 * cleared by the processor".
500 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
502 * In order to reflect the #DB exception payload in guest
503 * dr6, three components need to be considered: active low
504 * bit, FIXED_1 bits and active high bits (e.g. DR6_BD,
506 * DR6_ACTIVE_LOW contains the FIXED_1 and active low bits.
507 * In the target guest dr6:
508 * FIXED_1 bits should always be set.
509 * Active low bits should be cleared if 1-setting in payload.
510 * Active high bits should be set if 1-setting in payload.
512 * Note, the payload is compatible with the pending debug
513 * exceptions/exit qualification under VMX, that active_low bits
514 * are active high in payload.
515 * So they need to be flipped for DR6.
517 vcpu->arch.dr6 |= DR6_ACTIVE_LOW;
518 vcpu->arch.dr6 |= payload;
519 vcpu->arch.dr6 ^= payload & DR6_ACTIVE_LOW;
522 * The #DB payload is defined as compatible with the 'pending
523 * debug exceptions' field under VMX, not DR6. While bit 12 is
524 * defined in the 'pending debug exceptions' field (enabled
525 * breakpoint), it is reserved and must be zero in DR6.
527 vcpu->arch.dr6 &= ~BIT(12);
530 vcpu->arch.cr2 = payload;
534 vcpu->arch.exception.has_payload = false;
535 vcpu->arch.exception.payload = 0;
537 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
539 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
540 unsigned nr, bool has_error, u32 error_code,
541 bool has_payload, unsigned long payload, bool reinject)
546 kvm_make_request(KVM_REQ_EVENT, vcpu);
548 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
552 * On vmentry, vcpu->arch.exception.pending is only
553 * true if an event injection was blocked by
554 * nested_run_pending. In that case, however,
555 * vcpu_enter_guest requests an immediate exit,
556 * and the guest shouldn't proceed far enough to
559 WARN_ON_ONCE(vcpu->arch.exception.pending);
560 vcpu->arch.exception.injected = true;
561 if (WARN_ON_ONCE(has_payload)) {
563 * A reinjected event has already
564 * delivered its payload.
570 vcpu->arch.exception.pending = true;
571 vcpu->arch.exception.injected = false;
573 vcpu->arch.exception.has_error_code = has_error;
574 vcpu->arch.exception.nr = nr;
575 vcpu->arch.exception.error_code = error_code;
576 vcpu->arch.exception.has_payload = has_payload;
577 vcpu->arch.exception.payload = payload;
578 if (!is_guest_mode(vcpu))
579 kvm_deliver_exception_payload(vcpu);
583 /* to check exception */
584 prev_nr = vcpu->arch.exception.nr;
585 if (prev_nr == DF_VECTOR) {
586 /* triple fault -> shutdown */
587 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
590 class1 = exception_class(prev_nr);
591 class2 = exception_class(nr);
592 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
593 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
595 * Generate double fault per SDM Table 5-5. Set
596 * exception.pending = true so that the double fault
597 * can trigger a nested vmexit.
599 vcpu->arch.exception.pending = true;
600 vcpu->arch.exception.injected = false;
601 vcpu->arch.exception.has_error_code = true;
602 vcpu->arch.exception.nr = DF_VECTOR;
603 vcpu->arch.exception.error_code = 0;
604 vcpu->arch.exception.has_payload = false;
605 vcpu->arch.exception.payload = 0;
607 /* replace previous exception with a new one in a hope
608 that instruction re-execution will regenerate lost
613 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
615 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
617 EXPORT_SYMBOL_GPL(kvm_queue_exception);
619 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
621 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
623 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
625 void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
626 unsigned long payload)
628 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
630 EXPORT_SYMBOL_GPL(kvm_queue_exception_p);
632 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
633 u32 error_code, unsigned long payload)
635 kvm_multiple_exception(vcpu, nr, true, error_code,
636 true, payload, false);
639 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
642 kvm_inject_gp(vcpu, 0);
644 return kvm_skip_emulated_instruction(vcpu);
648 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
650 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
652 ++vcpu->stat.pf_guest;
653 vcpu->arch.exception.nested_apf =
654 is_guest_mode(vcpu) && fault->async_page_fault;
655 if (vcpu->arch.exception.nested_apf) {
656 vcpu->arch.apf.nested_apf_token = fault->address;
657 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
659 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
663 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
665 bool kvm_inject_emulated_page_fault(struct kvm_vcpu *vcpu,
666 struct x86_exception *fault)
668 struct kvm_mmu *fault_mmu;
669 WARN_ON_ONCE(fault->vector != PF_VECTOR);
671 fault_mmu = fault->nested_page_fault ? vcpu->arch.mmu :
675 * Invalidate the TLB entry for the faulting address, if it exists,
676 * else the access will fault indefinitely (and to emulate hardware).
678 if ((fault->error_code & PFERR_PRESENT_MASK) &&
679 !(fault->error_code & PFERR_RSVD_MASK))
680 kvm_mmu_invalidate_gva(vcpu, fault_mmu, fault->address,
681 fault_mmu->root_hpa);
683 fault_mmu->inject_page_fault(vcpu, fault);
684 return fault->nested_page_fault;
686 EXPORT_SYMBOL_GPL(kvm_inject_emulated_page_fault);
688 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
690 atomic_inc(&vcpu->arch.nmi_queued);
691 kvm_make_request(KVM_REQ_NMI, vcpu);
693 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
695 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
697 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
699 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
701 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
703 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
705 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
708 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
709 * a #GP and return false.
711 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
713 if (static_call(kvm_x86_get_cpl)(vcpu) <= required_cpl)
715 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
718 EXPORT_SYMBOL_GPL(kvm_require_cpl);
720 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
722 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
725 kvm_queue_exception(vcpu, UD_VECTOR);
728 EXPORT_SYMBOL_GPL(kvm_require_dr);
731 * This function will be used to read from the physical memory of the currently
732 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
733 * can read from guest physical or from the guest's guest physical memory.
735 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
736 gfn_t ngfn, void *data, int offset, int len,
739 struct x86_exception exception;
743 ngpa = gfn_to_gpa(ngfn);
744 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
745 if (real_gfn == UNMAPPED_GVA)
748 real_gfn = gpa_to_gfn(real_gfn);
750 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
752 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
754 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
755 void *data, int offset, int len, u32 access)
757 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
758 data, offset, len, access);
761 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
763 return vcpu->arch.reserved_gpa_bits | rsvd_bits(5, 8) | rsvd_bits(1, 2);
767 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
769 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
771 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
772 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
775 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
777 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
778 offset * sizeof(u64), sizeof(pdpte),
779 PFERR_USER_MASK|PFERR_WRITE_MASK);
784 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
785 if ((pdpte[i] & PT_PRESENT_MASK) &&
786 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
793 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
794 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
800 EXPORT_SYMBOL_GPL(load_pdptrs);
802 bool pdptrs_changed(struct kvm_vcpu *vcpu)
804 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
809 if (!is_pae_paging(vcpu))
812 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
815 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
816 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
817 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
818 PFERR_USER_MASK | PFERR_WRITE_MASK);
822 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
824 EXPORT_SYMBOL_GPL(pdptrs_changed);
826 void kvm_post_set_cr0(struct kvm_vcpu *vcpu, unsigned long old_cr0, unsigned long cr0)
828 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
830 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
831 kvm_clear_async_pf_completion_queue(vcpu);
832 kvm_async_pf_hash_reset(vcpu);
835 if ((cr0 ^ old_cr0) & update_bits)
836 kvm_mmu_reset_context(vcpu);
838 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
839 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
840 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
841 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
843 EXPORT_SYMBOL_GPL(kvm_post_set_cr0);
845 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
847 unsigned long old_cr0 = kvm_read_cr0(vcpu);
848 unsigned long pdptr_bits = X86_CR0_CD | X86_CR0_NW | X86_CR0_PG;
853 if (cr0 & 0xffffffff00000000UL)
857 cr0 &= ~CR0_RESERVED_BITS;
859 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
862 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
866 if ((vcpu->arch.efer & EFER_LME) && !is_paging(vcpu) &&
867 (cr0 & X86_CR0_PG)) {
872 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
877 if (!(vcpu->arch.efer & EFER_LME) && (cr0 & X86_CR0_PG) &&
878 is_pae(vcpu) && ((cr0 ^ old_cr0) & pdptr_bits) &&
879 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu)))
882 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
885 static_call(kvm_x86_set_cr0)(vcpu, cr0);
887 kvm_post_set_cr0(vcpu, old_cr0, cr0);
891 EXPORT_SYMBOL_GPL(kvm_set_cr0);
893 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
895 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
897 EXPORT_SYMBOL_GPL(kvm_lmsw);
899 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
901 if (vcpu->arch.guest_state_protected)
904 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
906 if (vcpu->arch.xcr0 != host_xcr0)
907 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
909 if (vcpu->arch.xsaves_enabled &&
910 vcpu->arch.ia32_xss != host_xss)
911 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
914 if (static_cpu_has(X86_FEATURE_PKU) &&
915 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
916 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU)) &&
917 vcpu->arch.pkru != vcpu->arch.host_pkru)
918 __write_pkru(vcpu->arch.pkru);
920 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
922 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
924 if (vcpu->arch.guest_state_protected)
927 if (static_cpu_has(X86_FEATURE_PKU) &&
928 (kvm_read_cr4_bits(vcpu, X86_CR4_PKE) ||
929 (vcpu->arch.xcr0 & XFEATURE_MASK_PKRU))) {
930 vcpu->arch.pkru = rdpkru();
931 if (vcpu->arch.pkru != vcpu->arch.host_pkru)
932 __write_pkru(vcpu->arch.host_pkru);
935 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
937 if (vcpu->arch.xcr0 != host_xcr0)
938 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
940 if (vcpu->arch.xsaves_enabled &&
941 vcpu->arch.ia32_xss != host_xss)
942 wrmsrl(MSR_IA32_XSS, host_xss);
946 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
948 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
951 u64 old_xcr0 = vcpu->arch.xcr0;
954 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
955 if (index != XCR_XFEATURE_ENABLED_MASK)
957 if (!(xcr0 & XFEATURE_MASK_FP))
959 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
963 * Do not allow the guest to set bits that we do not support
964 * saving. However, xcr0 bit 0 is always set, even if the
965 * emulated CPU does not support XSAVE (see fx_init).
967 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
968 if (xcr0 & ~valid_bits)
971 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
972 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
975 if (xcr0 & XFEATURE_MASK_AVX512) {
976 if (!(xcr0 & XFEATURE_MASK_YMM))
978 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
981 vcpu->arch.xcr0 = xcr0;
983 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
984 kvm_update_cpuid_runtime(vcpu);
988 int kvm_emulate_xsetbv(struct kvm_vcpu *vcpu)
990 if (static_call(kvm_x86_get_cpl)(vcpu) != 0 ||
991 __kvm_set_xcr(vcpu, kvm_rcx_read(vcpu), kvm_read_edx_eax(vcpu))) {
992 kvm_inject_gp(vcpu, 0);
996 return kvm_skip_emulated_instruction(vcpu);
998 EXPORT_SYMBOL_GPL(kvm_emulate_xsetbv);
1000 bool kvm_is_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1002 if (cr4 & cr4_reserved_bits)
1005 if (cr4 & vcpu->arch.cr4_guest_rsvd_bits)
1008 return static_call(kvm_x86_is_valid_cr4)(vcpu, cr4);
1010 EXPORT_SYMBOL_GPL(kvm_is_valid_cr4);
1012 void kvm_post_set_cr4(struct kvm_vcpu *vcpu, unsigned long old_cr4, unsigned long cr4)
1014 unsigned long mmu_role_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1015 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
1017 if (((cr4 ^ old_cr4) & mmu_role_bits) ||
1018 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
1019 kvm_mmu_reset_context(vcpu);
1021 EXPORT_SYMBOL_GPL(kvm_post_set_cr4);
1023 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1025 unsigned long old_cr4 = kvm_read_cr4(vcpu);
1026 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
1029 if (!kvm_is_valid_cr4(vcpu, cr4))
1032 if (is_long_mode(vcpu)) {
1033 if (!(cr4 & X86_CR4_PAE))
1035 if ((cr4 ^ old_cr4) & X86_CR4_LA57)
1037 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
1038 && ((cr4 ^ old_cr4) & pdptr_bits)
1039 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
1040 kvm_read_cr3(vcpu)))
1043 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
1044 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
1047 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
1048 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
1052 static_call(kvm_x86_set_cr4)(vcpu, cr4);
1054 kvm_post_set_cr4(vcpu, old_cr4, cr4);
1058 EXPORT_SYMBOL_GPL(kvm_set_cr4);
1060 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1062 bool skip_tlb_flush = false;
1063 #ifdef CONFIG_X86_64
1064 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
1067 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
1068 cr3 &= ~X86_CR3_PCID_NOFLUSH;
1072 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
1073 if (!skip_tlb_flush) {
1074 kvm_mmu_sync_roots(vcpu);
1075 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
1080 if (is_long_mode(vcpu) && kvm_vcpu_is_illegal_gpa(vcpu, cr3))
1082 else if (is_pae_paging(vcpu) &&
1083 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
1086 kvm_mmu_new_pgd(vcpu, cr3, skip_tlb_flush, skip_tlb_flush);
1087 vcpu->arch.cr3 = cr3;
1088 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1092 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1094 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1096 if (cr8 & CR8_RESERVED_BITS)
1098 if (lapic_in_kernel(vcpu))
1099 kvm_lapic_set_tpr(vcpu, cr8);
1101 vcpu->arch.cr8 = cr8;
1104 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1106 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1108 if (lapic_in_kernel(vcpu))
1109 return kvm_lapic_get_cr8(vcpu);
1111 return vcpu->arch.cr8;
1113 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1115 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1119 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1120 for (i = 0; i < KVM_NR_DB_REGS; i++)
1121 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1122 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1126 void kvm_update_dr7(struct kvm_vcpu *vcpu)
1130 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1131 dr7 = vcpu->arch.guest_debug_dr7;
1133 dr7 = vcpu->arch.dr7;
1134 static_call(kvm_x86_set_dr7)(vcpu, dr7);
1135 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1136 if (dr7 & DR7_BP_EN_MASK)
1137 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1139 EXPORT_SYMBOL_GPL(kvm_update_dr7);
1141 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1143 u64 fixed = DR6_FIXED_1;
1145 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1150 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1152 size_t size = ARRAY_SIZE(vcpu->arch.db);
1156 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1157 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1158 vcpu->arch.eff_db[dr] = val;
1162 if (!kvm_dr6_valid(val))
1164 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1168 if (!kvm_dr7_valid(val))
1170 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1171 kvm_update_dr7(vcpu);
1177 EXPORT_SYMBOL_GPL(kvm_set_dr);
1179 void kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1181 size_t size = ARRAY_SIZE(vcpu->arch.db);
1185 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1189 *val = vcpu->arch.dr6;
1193 *val = vcpu->arch.dr7;
1197 EXPORT_SYMBOL_GPL(kvm_get_dr);
1199 int kvm_emulate_rdpmc(struct kvm_vcpu *vcpu)
1201 u32 ecx = kvm_rcx_read(vcpu);
1204 if (kvm_pmu_rdpmc(vcpu, ecx, &data)) {
1205 kvm_inject_gp(vcpu, 0);
1209 kvm_rax_write(vcpu, (u32)data);
1210 kvm_rdx_write(vcpu, data >> 32);
1211 return kvm_skip_emulated_instruction(vcpu);
1213 EXPORT_SYMBOL_GPL(kvm_emulate_rdpmc);
1216 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1217 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1219 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1220 * extract the supported MSRs from the related const lists.
1221 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1222 * capabilities of the host cpu. This capabilities test skips MSRs that are
1223 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1224 * may depend on host virtualization features rather than host cpu features.
1227 static const u32 msrs_to_save_all[] = {
1228 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1230 #ifdef CONFIG_X86_64
1231 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1233 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1234 MSR_IA32_FEAT_CTL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1236 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1237 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1238 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1239 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1240 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1241 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1242 MSR_IA32_UMWAIT_CONTROL,
1244 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1245 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1246 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1247 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1248 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1249 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1250 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1251 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1252 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1253 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1254 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1255 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1256 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1257 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1258 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1259 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1260 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1261 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1262 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1263 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1264 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1265 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1268 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1269 static unsigned num_msrs_to_save;
1271 static const u32 emulated_msrs_all[] = {
1272 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1273 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1274 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1275 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1276 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1277 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1278 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1280 HV_X64_MSR_VP_INDEX,
1281 HV_X64_MSR_VP_RUNTIME,
1282 HV_X64_MSR_SCONTROL,
1283 HV_X64_MSR_STIMER0_CONFIG,
1284 HV_X64_MSR_VP_ASSIST_PAGE,
1285 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1286 HV_X64_MSR_TSC_EMULATION_STATUS,
1287 HV_X64_MSR_SYNDBG_OPTIONS,
1288 HV_X64_MSR_SYNDBG_CONTROL, HV_X64_MSR_SYNDBG_STATUS,
1289 HV_X64_MSR_SYNDBG_SEND_BUFFER, HV_X64_MSR_SYNDBG_RECV_BUFFER,
1290 HV_X64_MSR_SYNDBG_PENDING_BUFFER,
1292 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1293 MSR_KVM_PV_EOI_EN, MSR_KVM_ASYNC_PF_INT, MSR_KVM_ASYNC_PF_ACK,
1295 MSR_IA32_TSC_ADJUST,
1296 MSR_IA32_TSCDEADLINE,
1297 MSR_IA32_ARCH_CAPABILITIES,
1298 MSR_IA32_PERF_CAPABILITIES,
1299 MSR_IA32_MISC_ENABLE,
1300 MSR_IA32_MCG_STATUS,
1302 MSR_IA32_MCG_EXT_CTL,
1306 MSR_MISC_FEATURES_ENABLES,
1307 MSR_AMD64_VIRT_SPEC_CTRL,
1312 * The following list leaves out MSRs whose values are determined
1313 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1314 * We always support the "true" VMX control MSRs, even if the host
1315 * processor does not, so I am putting these registers here rather
1316 * than in msrs_to_save_all.
1319 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1320 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1321 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1322 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1324 MSR_IA32_VMX_CR0_FIXED0,
1325 MSR_IA32_VMX_CR4_FIXED0,
1326 MSR_IA32_VMX_VMCS_ENUM,
1327 MSR_IA32_VMX_PROCBASED_CTLS2,
1328 MSR_IA32_VMX_EPT_VPID_CAP,
1329 MSR_IA32_VMX_VMFUNC,
1332 MSR_KVM_POLL_CONTROL,
1335 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1336 static unsigned num_emulated_msrs;
1339 * List of msr numbers which are used to expose MSR-based features that
1340 * can be used by a hypervisor to validate requested CPU features.
1342 static const u32 msr_based_features_all[] = {
1344 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1345 MSR_IA32_VMX_PINBASED_CTLS,
1346 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1347 MSR_IA32_VMX_PROCBASED_CTLS,
1348 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1349 MSR_IA32_VMX_EXIT_CTLS,
1350 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1351 MSR_IA32_VMX_ENTRY_CTLS,
1353 MSR_IA32_VMX_CR0_FIXED0,
1354 MSR_IA32_VMX_CR0_FIXED1,
1355 MSR_IA32_VMX_CR4_FIXED0,
1356 MSR_IA32_VMX_CR4_FIXED1,
1357 MSR_IA32_VMX_VMCS_ENUM,
1358 MSR_IA32_VMX_PROCBASED_CTLS2,
1359 MSR_IA32_VMX_EPT_VPID_CAP,
1360 MSR_IA32_VMX_VMFUNC,
1364 MSR_IA32_ARCH_CAPABILITIES,
1365 MSR_IA32_PERF_CAPABILITIES,
1368 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1369 static unsigned int num_msr_based_features;
1371 static u64 kvm_get_arch_capabilities(void)
1375 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1376 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1379 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1380 * the nested hypervisor runs with NX huge pages. If it is not,
1381 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1382 * L1 guests, so it need not worry about its own (L2) guests.
1384 data |= ARCH_CAP_PSCHANGE_MC_NO;
1387 * If we're doing cache flushes (either "always" or "cond")
1388 * we will do one whenever the guest does a vmlaunch/vmresume.
1389 * If an outer hypervisor is doing the cache flush for us
1390 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1391 * capability to the guest too, and if EPT is disabled we're not
1392 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1393 * require a nested hypervisor to do a flush of its own.
1395 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1396 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1398 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1399 data |= ARCH_CAP_RDCL_NO;
1400 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1401 data |= ARCH_CAP_SSB_NO;
1402 if (!boot_cpu_has_bug(X86_BUG_MDS))
1403 data |= ARCH_CAP_MDS_NO;
1405 if (!boot_cpu_has(X86_FEATURE_RTM)) {
1407 * If RTM=0 because the kernel has disabled TSX, the host might
1408 * have TAA_NO or TSX_CTRL. Clear TAA_NO (the guest sees RTM=0
1409 * and therefore knows that there cannot be TAA) but keep
1410 * TSX_CTRL: some buggy userspaces leave it set on tsx=on hosts,
1411 * and we want to allow migrating those guests to tsx=off hosts.
1413 data &= ~ARCH_CAP_TAA_NO;
1414 } else if (!boot_cpu_has_bug(X86_BUG_TAA)) {
1415 data |= ARCH_CAP_TAA_NO;
1418 * Nothing to do here; we emulate TSX_CTRL if present on the
1419 * host so the guest can choose between disabling TSX or
1420 * using VERW to clear CPU buffers.
1427 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1429 switch (msr->index) {
1430 case MSR_IA32_ARCH_CAPABILITIES:
1431 msr->data = kvm_get_arch_capabilities();
1433 case MSR_IA32_UCODE_REV:
1434 rdmsrl_safe(msr->index, &msr->data);
1437 return static_call(kvm_x86_get_msr_feature)(msr);
1442 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1444 struct kvm_msr_entry msr;
1448 r = kvm_get_msr_feature(&msr);
1450 if (r == KVM_MSR_RET_INVALID) {
1451 /* Unconditionally clear the output for simplicity */
1453 if (kvm_msr_ignored_check(index, 0, false))
1465 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1467 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1470 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1473 if (efer & (EFER_LME | EFER_LMA) &&
1474 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1477 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1483 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1485 if (efer & efer_reserved_bits)
1488 return __kvm_valid_efer(vcpu, efer);
1490 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1492 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1494 u64 old_efer = vcpu->arch.efer;
1495 u64 efer = msr_info->data;
1498 if (efer & efer_reserved_bits)
1501 if (!msr_info->host_initiated) {
1502 if (!__kvm_valid_efer(vcpu, efer))
1505 if (is_paging(vcpu) &&
1506 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1511 efer |= vcpu->arch.efer & EFER_LMA;
1513 r = static_call(kvm_x86_set_efer)(vcpu, efer);
1519 /* Update reserved bits */
1520 if ((efer ^ old_efer) & EFER_NX)
1521 kvm_mmu_reset_context(vcpu);
1526 void kvm_enable_efer_bits(u64 mask)
1528 efer_reserved_bits &= ~mask;
1530 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1532 bool kvm_msr_allowed(struct kvm_vcpu *vcpu, u32 index, u32 type)
1534 struct kvm_x86_msr_filter *msr_filter;
1535 struct msr_bitmap_range *ranges;
1536 struct kvm *kvm = vcpu->kvm;
1541 /* x2APIC MSRs do not support filtering. */
1542 if (index >= 0x800 && index <= 0x8ff)
1545 idx = srcu_read_lock(&kvm->srcu);
1547 msr_filter = srcu_dereference(kvm->arch.msr_filter, &kvm->srcu);
1553 allowed = msr_filter->default_allow;
1554 ranges = msr_filter->ranges;
1556 for (i = 0; i < msr_filter->count; i++) {
1557 u32 start = ranges[i].base;
1558 u32 end = start + ranges[i].nmsrs;
1559 u32 flags = ranges[i].flags;
1560 unsigned long *bitmap = ranges[i].bitmap;
1562 if ((index >= start) && (index < end) && (flags & type)) {
1563 allowed = !!test_bit(index - start, bitmap);
1569 srcu_read_unlock(&kvm->srcu, idx);
1573 EXPORT_SYMBOL_GPL(kvm_msr_allowed);
1576 * Write @data into the MSR specified by @index. Select MSR specific fault
1577 * checks are bypassed if @host_initiated is %true.
1578 * Returns 0 on success, non-0 otherwise.
1579 * Assumes vcpu_load() was already called.
1581 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1582 bool host_initiated)
1584 struct msr_data msr;
1586 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_WRITE))
1587 return KVM_MSR_RET_FILTERED;
1592 case MSR_KERNEL_GS_BASE:
1595 if (is_noncanonical_address(data, vcpu))
1598 case MSR_IA32_SYSENTER_EIP:
1599 case MSR_IA32_SYSENTER_ESP:
1601 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1602 * non-canonical address is written on Intel but not on
1603 * AMD (which ignores the top 32-bits, because it does
1604 * not implement 64-bit SYSENTER).
1606 * 64-bit code should hence be able to write a non-canonical
1607 * value on AMD. Making the address canonical ensures that
1608 * vmentry does not fail on Intel after writing a non-canonical
1609 * value, and that something deterministic happens if the guest
1610 * invokes 64-bit SYSENTER.
1612 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1617 msr.host_initiated = host_initiated;
1619 return static_call(kvm_x86_set_msr)(vcpu, &msr);
1622 static int kvm_set_msr_ignored_check(struct kvm_vcpu *vcpu,
1623 u32 index, u64 data, bool host_initiated)
1625 int ret = __kvm_set_msr(vcpu, index, data, host_initiated);
1627 if (ret == KVM_MSR_RET_INVALID)
1628 if (kvm_msr_ignored_check(index, data, true))
1635 * Read the MSR specified by @index into @data. Select MSR specific fault
1636 * checks are bypassed if @host_initiated is %true.
1637 * Returns 0 on success, non-0 otherwise.
1638 * Assumes vcpu_load() was already called.
1640 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1641 bool host_initiated)
1643 struct msr_data msr;
1646 if (!host_initiated && !kvm_msr_allowed(vcpu, index, KVM_MSR_FILTER_READ))
1647 return KVM_MSR_RET_FILTERED;
1650 msr.host_initiated = host_initiated;
1652 ret = static_call(kvm_x86_get_msr)(vcpu, &msr);
1658 static int kvm_get_msr_ignored_check(struct kvm_vcpu *vcpu,
1659 u32 index, u64 *data, bool host_initiated)
1661 int ret = __kvm_get_msr(vcpu, index, data, host_initiated);
1663 if (ret == KVM_MSR_RET_INVALID) {
1664 /* Unconditionally clear *data for simplicity */
1666 if (kvm_msr_ignored_check(index, 0, false))
1673 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1675 return kvm_get_msr_ignored_check(vcpu, index, data, false);
1677 EXPORT_SYMBOL_GPL(kvm_get_msr);
1679 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1681 return kvm_set_msr_ignored_check(vcpu, index, data, false);
1683 EXPORT_SYMBOL_GPL(kvm_set_msr);
1685 static int complete_emulated_rdmsr(struct kvm_vcpu *vcpu)
1687 int err = vcpu->run->msr.error;
1689 kvm_rax_write(vcpu, (u32)vcpu->run->msr.data);
1690 kvm_rdx_write(vcpu, vcpu->run->msr.data >> 32);
1693 return static_call(kvm_x86_complete_emulated_msr)(vcpu, err);
1696 static int complete_emulated_wrmsr(struct kvm_vcpu *vcpu)
1698 return static_call(kvm_x86_complete_emulated_msr)(vcpu, vcpu->run->msr.error);
1701 static u64 kvm_msr_reason(int r)
1704 case KVM_MSR_RET_INVALID:
1705 return KVM_MSR_EXIT_REASON_UNKNOWN;
1706 case KVM_MSR_RET_FILTERED:
1707 return KVM_MSR_EXIT_REASON_FILTER;
1709 return KVM_MSR_EXIT_REASON_INVAL;
1713 static int kvm_msr_user_space(struct kvm_vcpu *vcpu, u32 index,
1714 u32 exit_reason, u64 data,
1715 int (*completion)(struct kvm_vcpu *vcpu),
1718 u64 msr_reason = kvm_msr_reason(r);
1720 /* Check if the user wanted to know about this MSR fault */
1721 if (!(vcpu->kvm->arch.user_space_msr_mask & msr_reason))
1724 vcpu->run->exit_reason = exit_reason;
1725 vcpu->run->msr.error = 0;
1726 memset(vcpu->run->msr.pad, 0, sizeof(vcpu->run->msr.pad));
1727 vcpu->run->msr.reason = msr_reason;
1728 vcpu->run->msr.index = index;
1729 vcpu->run->msr.data = data;
1730 vcpu->arch.complete_userspace_io = completion;
1735 static int kvm_get_msr_user_space(struct kvm_vcpu *vcpu, u32 index, int r)
1737 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_RDMSR, 0,
1738 complete_emulated_rdmsr, r);
1741 static int kvm_set_msr_user_space(struct kvm_vcpu *vcpu, u32 index, u64 data, int r)
1743 return kvm_msr_user_space(vcpu, index, KVM_EXIT_X86_WRMSR, data,
1744 complete_emulated_wrmsr, r);
1747 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1749 u32 ecx = kvm_rcx_read(vcpu);
1753 r = kvm_get_msr(vcpu, ecx, &data);
1755 /* MSR read failed? See if we should ask user space */
1756 if (r && kvm_get_msr_user_space(vcpu, ecx, r)) {
1757 /* Bounce to user space */
1762 trace_kvm_msr_read(ecx, data);
1764 kvm_rax_write(vcpu, data & -1u);
1765 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1767 trace_kvm_msr_read_ex(ecx);
1770 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1772 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1774 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1776 u32 ecx = kvm_rcx_read(vcpu);
1777 u64 data = kvm_read_edx_eax(vcpu);
1780 r = kvm_set_msr(vcpu, ecx, data);
1782 /* MSR write failed? See if we should ask user space */
1783 if (r && kvm_set_msr_user_space(vcpu, ecx, data, r))
1784 /* Bounce to user space */
1787 /* Signal all other negative errors to userspace */
1792 trace_kvm_msr_write(ecx, data);
1794 trace_kvm_msr_write_ex(ecx, data);
1796 return static_call(kvm_x86_complete_emulated_msr)(vcpu, r);
1798 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1800 int kvm_emulate_as_nop(struct kvm_vcpu *vcpu)
1802 return kvm_skip_emulated_instruction(vcpu);
1804 EXPORT_SYMBOL_GPL(kvm_emulate_as_nop);
1806 int kvm_emulate_invd(struct kvm_vcpu *vcpu)
1808 /* Treat an INVD instruction as a NOP and just skip it. */
1809 return kvm_emulate_as_nop(vcpu);
1811 EXPORT_SYMBOL_GPL(kvm_emulate_invd);
1813 int kvm_emulate_mwait(struct kvm_vcpu *vcpu)
1815 pr_warn_once("kvm: MWAIT instruction emulated as NOP!\n");
1816 return kvm_emulate_as_nop(vcpu);
1818 EXPORT_SYMBOL_GPL(kvm_emulate_mwait);
1820 int kvm_handle_invalid_op(struct kvm_vcpu *vcpu)
1822 kvm_queue_exception(vcpu, UD_VECTOR);
1825 EXPORT_SYMBOL_GPL(kvm_handle_invalid_op);
1827 int kvm_emulate_monitor(struct kvm_vcpu *vcpu)
1829 pr_warn_once("kvm: MONITOR instruction emulated as NOP!\n");
1830 return kvm_emulate_as_nop(vcpu);
1832 EXPORT_SYMBOL_GPL(kvm_emulate_monitor);
1834 static inline bool kvm_vcpu_exit_request(struct kvm_vcpu *vcpu)
1836 xfer_to_guest_mode_prepare();
1837 return vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu) ||
1838 xfer_to_guest_mode_work_pending();
1842 * The fast path for frequent and performance sensitive wrmsr emulation,
1843 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1844 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1845 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1846 * other cases which must be called after interrupts are enabled on the host.
1848 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1850 if (!lapic_in_kernel(vcpu) || !apic_x2apic_mode(vcpu->arch.apic))
1853 if (((data & APIC_SHORT_MASK) == APIC_DEST_NOSHORT) &&
1854 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1855 ((data & APIC_MODE_MASK) == APIC_DM_FIXED) &&
1856 ((u32)(data >> 32) != X2APIC_BROADCAST)) {
1859 kvm_apic_send_ipi(vcpu->arch.apic, (u32)data, (u32)(data >> 32));
1860 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1861 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR, (u32)data);
1862 trace_kvm_apic_write(APIC_ICR, (u32)data);
1869 static int handle_fastpath_set_tscdeadline(struct kvm_vcpu *vcpu, u64 data)
1871 if (!kvm_can_use_hv_timer(vcpu))
1874 kvm_set_lapic_tscdeadline_msr(vcpu, data);
1878 fastpath_t handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1880 u32 msr = kvm_rcx_read(vcpu);
1882 fastpath_t ret = EXIT_FASTPATH_NONE;
1885 case APIC_BASE_MSR + (APIC_ICR >> 4):
1886 data = kvm_read_edx_eax(vcpu);
1887 if (!handle_fastpath_set_x2apic_icr_irqoff(vcpu, data)) {
1888 kvm_skip_emulated_instruction(vcpu);
1889 ret = EXIT_FASTPATH_EXIT_HANDLED;
1892 case MSR_IA32_TSCDEADLINE:
1893 data = kvm_read_edx_eax(vcpu);
1894 if (!handle_fastpath_set_tscdeadline(vcpu, data)) {
1895 kvm_skip_emulated_instruction(vcpu);
1896 ret = EXIT_FASTPATH_REENTER_GUEST;
1903 if (ret != EXIT_FASTPATH_NONE)
1904 trace_kvm_msr_write(msr, data);
1908 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1911 * Adapt set_msr() to msr_io()'s calling convention
1913 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1915 return kvm_get_msr_ignored_check(vcpu, index, data, true);
1918 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1920 return kvm_set_msr_ignored_check(vcpu, index, *data, true);
1923 #ifdef CONFIG_X86_64
1924 struct pvclock_clock {
1934 struct pvclock_gtod_data {
1937 struct pvclock_clock clock; /* extract of a clocksource struct */
1938 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1944 static struct pvclock_gtod_data pvclock_gtod_data;
1946 static void update_pvclock_gtod(struct timekeeper *tk)
1948 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1950 write_seqcount_begin(&vdata->seq);
1952 /* copy pvclock gtod data */
1953 vdata->clock.vclock_mode = tk->tkr_mono.clock->vdso_clock_mode;
1954 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1955 vdata->clock.mask = tk->tkr_mono.mask;
1956 vdata->clock.mult = tk->tkr_mono.mult;
1957 vdata->clock.shift = tk->tkr_mono.shift;
1958 vdata->clock.base_cycles = tk->tkr_mono.xtime_nsec;
1959 vdata->clock.offset = tk->tkr_mono.base;
1961 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->vdso_clock_mode;
1962 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1963 vdata->raw_clock.mask = tk->tkr_raw.mask;
1964 vdata->raw_clock.mult = tk->tkr_raw.mult;
1965 vdata->raw_clock.shift = tk->tkr_raw.shift;
1966 vdata->raw_clock.base_cycles = tk->tkr_raw.xtime_nsec;
1967 vdata->raw_clock.offset = tk->tkr_raw.base;
1969 vdata->wall_time_sec = tk->xtime_sec;
1971 vdata->offs_boot = tk->offs_boot;
1973 write_seqcount_end(&vdata->seq);
1976 static s64 get_kvmclock_base_ns(void)
1978 /* Count up from boot time, but with the frequency of the raw clock. */
1979 return ktime_to_ns(ktime_add(ktime_get_raw(), pvclock_gtod_data.offs_boot));
1982 static s64 get_kvmclock_base_ns(void)
1984 /* Master clock not used, so we can just use CLOCK_BOOTTIME. */
1985 return ktime_get_boottime_ns();
1989 void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock, int sec_hi_ofs)
1993 struct pvclock_wall_clock wc;
2000 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
2005 ++version; /* first time write, random junk */
2009 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
2013 * The guest calculates current wall clock time by adding
2014 * system time (updated by kvm_guest_time_update below) to the
2015 * wall clock specified here. We do the reverse here.
2017 wall_nsec = ktime_get_real_ns() - get_kvmclock_ns(kvm);
2019 wc.nsec = do_div(wall_nsec, 1000000000);
2020 wc.sec = (u32)wall_nsec; /* overflow in 2106 guest time */
2021 wc.version = version;
2023 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
2026 wc_sec_hi = wall_nsec >> 32;
2027 kvm_write_guest(kvm, wall_clock + sec_hi_ofs,
2028 &wc_sec_hi, sizeof(wc_sec_hi));
2032 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
2035 static void kvm_write_system_time(struct kvm_vcpu *vcpu, gpa_t system_time,
2036 bool old_msr, bool host_initiated)
2038 struct kvm_arch *ka = &vcpu->kvm->arch;
2040 if (vcpu->vcpu_id == 0 && !host_initiated) {
2041 if (ka->boot_vcpu_runs_old_kvmclock != old_msr)
2042 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2044 ka->boot_vcpu_runs_old_kvmclock = old_msr;
2047 vcpu->arch.time = system_time;
2048 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2050 /* we verify if the enable bit is set... */
2051 vcpu->arch.pv_time_enabled = false;
2052 if (!(system_time & 1))
2055 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2056 &vcpu->arch.pv_time, system_time & ~1ULL,
2057 sizeof(struct pvclock_vcpu_time_info)))
2058 vcpu->arch.pv_time_enabled = true;
2063 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
2065 do_shl32_div32(dividend, divisor);
2069 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
2070 s8 *pshift, u32 *pmultiplier)
2078 scaled64 = scaled_hz;
2079 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
2084 tps32 = (uint32_t)tps64;
2085 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
2086 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
2094 *pmultiplier = div_frac(scaled64, tps32);
2097 #ifdef CONFIG_X86_64
2098 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
2101 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
2102 static unsigned long max_tsc_khz;
2104 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
2106 u64 v = (u64)khz * (1000000 + ppm);
2111 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2115 /* Guest TSC same frequency as host TSC? */
2117 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2121 /* TSC scaling supported? */
2122 if (!kvm_has_tsc_control) {
2123 if (user_tsc_khz > tsc_khz) {
2124 vcpu->arch.tsc_catchup = 1;
2125 vcpu->arch.tsc_always_catchup = 1;
2128 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
2133 /* TSC scaling required - calculate ratio */
2134 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
2135 user_tsc_khz, tsc_khz);
2137 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
2138 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
2143 vcpu->arch.tsc_scaling_ratio = ratio;
2147 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
2149 u32 thresh_lo, thresh_hi;
2150 int use_scaling = 0;
2152 /* tsc_khz can be zero if TSC calibration fails */
2153 if (user_tsc_khz == 0) {
2154 /* set tsc_scaling_ratio to a safe value */
2155 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
2159 /* Compute a scale to convert nanoseconds in TSC cycles */
2160 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
2161 &vcpu->arch.virtual_tsc_shift,
2162 &vcpu->arch.virtual_tsc_mult);
2163 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
2166 * Compute the variation in TSC rate which is acceptable
2167 * within the range of tolerance and decide if the
2168 * rate being applied is within that bounds of the hardware
2169 * rate. If so, no scaling or compensation need be done.
2171 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
2172 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
2173 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
2174 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
2177 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
2180 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
2182 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
2183 vcpu->arch.virtual_tsc_mult,
2184 vcpu->arch.virtual_tsc_shift);
2185 tsc += vcpu->arch.this_tsc_write;
2189 static inline int gtod_is_based_on_tsc(int mode)
2191 return mode == VDSO_CLOCKMODE_TSC || mode == VDSO_CLOCKMODE_HVCLOCK;
2194 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
2196 #ifdef CONFIG_X86_64
2198 struct kvm_arch *ka = &vcpu->kvm->arch;
2199 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2201 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2202 atomic_read(&vcpu->kvm->online_vcpus));
2205 * Once the masterclock is enabled, always perform request in
2206 * order to update it.
2208 * In order to enable masterclock, the host clocksource must be TSC
2209 * and the vcpus need to have matched TSCs. When that happens,
2210 * perform request to enable masterclock.
2212 if (ka->use_master_clock ||
2213 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
2214 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2216 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
2217 atomic_read(&vcpu->kvm->online_vcpus),
2218 ka->use_master_clock, gtod->clock.vclock_mode);
2223 * Multiply tsc by a fixed point number represented by ratio.
2225 * The most significant 64-N bits (mult) of ratio represent the
2226 * integral part of the fixed point number; the remaining N bits
2227 * (frac) represent the fractional part, ie. ratio represents a fixed
2228 * point number (mult + frac * 2^(-N)).
2230 * N equals to kvm_tsc_scaling_ratio_frac_bits.
2232 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
2234 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
2237 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
2240 u64 ratio = vcpu->arch.tsc_scaling_ratio;
2242 if (ratio != kvm_default_tsc_scaling_ratio)
2243 _tsc = __scale_tsc(ratio, tsc);
2247 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
2249 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2253 tsc = kvm_scale_tsc(vcpu, rdtsc());
2255 return target_tsc - tsc;
2258 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2260 return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
2262 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
2264 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2266 vcpu->arch.l1_tsc_offset = offset;
2267 vcpu->arch.tsc_offset = static_call(kvm_x86_write_l1_tsc_offset)(vcpu, offset);
2270 static inline bool kvm_check_tsc_unstable(void)
2272 #ifdef CONFIG_X86_64
2274 * TSC is marked unstable when we're running on Hyper-V,
2275 * 'TSC page' clocksource is good.
2277 if (pvclock_gtod_data.clock.vclock_mode == VDSO_CLOCKMODE_HVCLOCK)
2280 return check_tsc_unstable();
2283 static void kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 data)
2285 struct kvm *kvm = vcpu->kvm;
2286 u64 offset, ns, elapsed;
2287 unsigned long flags;
2289 bool already_matched;
2290 bool synchronizing = false;
2292 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
2293 offset = kvm_compute_tsc_offset(vcpu, data);
2294 ns = get_kvmclock_base_ns();
2295 elapsed = ns - kvm->arch.last_tsc_nsec;
2297 if (vcpu->arch.virtual_tsc_khz) {
2300 * detection of vcpu initialization -- need to sync
2301 * with other vCPUs. This particularly helps to keep
2302 * kvm_clock stable after CPU hotplug
2304 synchronizing = true;
2306 u64 tsc_exp = kvm->arch.last_tsc_write +
2307 nsec_to_cycles(vcpu, elapsed);
2308 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
2310 * Special case: TSC write with a small delta (1 second)
2311 * of virtual cycle time against real time is
2312 * interpreted as an attempt to synchronize the CPU.
2314 synchronizing = data < tsc_exp + tsc_hz &&
2315 data + tsc_hz > tsc_exp;
2320 * For a reliable TSC, we can match TSC offsets, and for an unstable
2321 * TSC, we add elapsed time in this computation. We could let the
2322 * compensation code attempt to catch up if we fall behind, but
2323 * it's better to try to match offsets from the beginning.
2325 if (synchronizing &&
2326 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
2327 if (!kvm_check_tsc_unstable()) {
2328 offset = kvm->arch.cur_tsc_offset;
2330 u64 delta = nsec_to_cycles(vcpu, elapsed);
2332 offset = kvm_compute_tsc_offset(vcpu, data);
2335 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
2338 * We split periods of matched TSC writes into generations.
2339 * For each generation, we track the original measured
2340 * nanosecond time, offset, and write, so if TSCs are in
2341 * sync, we can match exact offset, and if not, we can match
2342 * exact software computation in compute_guest_tsc()
2344 * These values are tracked in kvm->arch.cur_xxx variables.
2346 kvm->arch.cur_tsc_generation++;
2347 kvm->arch.cur_tsc_nsec = ns;
2348 kvm->arch.cur_tsc_write = data;
2349 kvm->arch.cur_tsc_offset = offset;
2354 * We also track th most recent recorded KHZ, write and time to
2355 * allow the matching interval to be extended at each write.
2357 kvm->arch.last_tsc_nsec = ns;
2358 kvm->arch.last_tsc_write = data;
2359 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2361 vcpu->arch.last_guest_tsc = data;
2363 /* Keep track of which generation this VCPU has synchronized to */
2364 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2365 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2366 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2368 kvm_vcpu_write_tsc_offset(vcpu, offset);
2369 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2371 spin_lock_irqsave(&kvm->arch.pvclock_gtod_sync_lock, flags);
2373 kvm->arch.nr_vcpus_matched_tsc = 0;
2374 } else if (!already_matched) {
2375 kvm->arch.nr_vcpus_matched_tsc++;
2378 kvm_track_tsc_matching(vcpu);
2379 spin_unlock_irqrestore(&kvm->arch.pvclock_gtod_sync_lock, flags);
2382 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2385 u64 tsc_offset = vcpu->arch.l1_tsc_offset;
2386 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2389 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2391 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2392 WARN_ON(adjustment < 0);
2393 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2394 adjust_tsc_offset_guest(vcpu, adjustment);
2397 #ifdef CONFIG_X86_64
2399 static u64 read_tsc(void)
2401 u64 ret = (u64)rdtsc_ordered();
2402 u64 last = pvclock_gtod_data.clock.cycle_last;
2404 if (likely(ret >= last))
2408 * GCC likes to generate cmov here, but this branch is extremely
2409 * predictable (it's just a function of time and the likely is
2410 * very likely) and there's a data dependence, so force GCC
2411 * to generate a branch instead. I don't barrier() because
2412 * we don't actually need a barrier, and if this function
2413 * ever gets inlined it will generate worse code.
2419 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2425 switch (clock->vclock_mode) {
2426 case VDSO_CLOCKMODE_HVCLOCK:
2427 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2429 if (tsc_pg_val != U64_MAX) {
2430 /* TSC page valid */
2431 *mode = VDSO_CLOCKMODE_HVCLOCK;
2432 v = (tsc_pg_val - clock->cycle_last) &
2435 /* TSC page invalid */
2436 *mode = VDSO_CLOCKMODE_NONE;
2439 case VDSO_CLOCKMODE_TSC:
2440 *mode = VDSO_CLOCKMODE_TSC;
2441 *tsc_timestamp = read_tsc();
2442 v = (*tsc_timestamp - clock->cycle_last) &
2446 *mode = VDSO_CLOCKMODE_NONE;
2449 if (*mode == VDSO_CLOCKMODE_NONE)
2450 *tsc_timestamp = v = 0;
2452 return v * clock->mult;
2455 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2457 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2463 seq = read_seqcount_begin(>od->seq);
2464 ns = gtod->raw_clock.base_cycles;
2465 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2466 ns >>= gtod->raw_clock.shift;
2467 ns += ktime_to_ns(ktime_add(gtod->raw_clock.offset, gtod->offs_boot));
2468 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2474 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2476 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2482 seq = read_seqcount_begin(>od->seq);
2483 ts->tv_sec = gtod->wall_time_sec;
2484 ns = gtod->clock.base_cycles;
2485 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2486 ns >>= gtod->clock.shift;
2487 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2489 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2495 /* returns true if host is using TSC based clocksource */
2496 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2498 /* checked again under seqlock below */
2499 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2502 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2506 /* returns true if host is using TSC based clocksource */
2507 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2510 /* checked again under seqlock below */
2511 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2514 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2520 * Assuming a stable TSC across physical CPUS, and a stable TSC
2521 * across virtual CPUs, the following condition is possible.
2522 * Each numbered line represents an event visible to both
2523 * CPUs at the next numbered event.
2525 * "timespecX" represents host monotonic time. "tscX" represents
2528 * VCPU0 on CPU0 | VCPU1 on CPU1
2530 * 1. read timespec0,tsc0
2531 * 2. | timespec1 = timespec0 + N
2533 * 3. transition to guest | transition to guest
2534 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2535 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2536 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2538 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2541 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2543 * - 0 < N - M => M < N
2545 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2546 * always the case (the difference between two distinct xtime instances
2547 * might be smaller then the difference between corresponding TSC reads,
2548 * when updating guest vcpus pvclock areas).
2550 * To avoid that problem, do not allow visibility of distinct
2551 * system_timestamp/tsc_timestamp values simultaneously: use a master
2552 * copy of host monotonic time values. Update that master copy
2555 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2559 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2561 #ifdef CONFIG_X86_64
2562 struct kvm_arch *ka = &kvm->arch;
2564 bool host_tsc_clocksource, vcpus_matched;
2566 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2567 atomic_read(&kvm->online_vcpus));
2570 * If the host uses TSC clock, then passthrough TSC as stable
2573 host_tsc_clocksource = kvm_get_time_and_clockread(
2574 &ka->master_kernel_ns,
2575 &ka->master_cycle_now);
2577 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2578 && !ka->backwards_tsc_observed
2579 && !ka->boot_vcpu_runs_old_kvmclock;
2581 if (ka->use_master_clock)
2582 atomic_set(&kvm_guest_has_master_clock, 1);
2584 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2585 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2590 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2592 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2595 static void kvm_gen_update_masterclock(struct kvm *kvm)
2597 #ifdef CONFIG_X86_64
2599 struct kvm_vcpu *vcpu;
2600 struct kvm_arch *ka = &kvm->arch;
2601 unsigned long flags;
2603 kvm_hv_invalidate_tsc_page(kvm);
2605 kvm_make_mclock_inprogress_request(kvm);
2607 /* no guest entries from this point */
2608 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2609 pvclock_update_vm_gtod_copy(kvm);
2610 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2612 kvm_for_each_vcpu(i, vcpu, kvm)
2613 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2615 /* guest entries allowed */
2616 kvm_for_each_vcpu(i, vcpu, kvm)
2617 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2621 u64 get_kvmclock_ns(struct kvm *kvm)
2623 struct kvm_arch *ka = &kvm->arch;
2624 struct pvclock_vcpu_time_info hv_clock;
2625 unsigned long flags;
2628 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2629 if (!ka->use_master_clock) {
2630 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2631 return get_kvmclock_base_ns() + ka->kvmclock_offset;
2634 hv_clock.tsc_timestamp = ka->master_cycle_now;
2635 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2636 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2638 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2641 if (__this_cpu_read(cpu_tsc_khz)) {
2642 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2643 &hv_clock.tsc_shift,
2644 &hv_clock.tsc_to_system_mul);
2645 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2647 ret = get_kvmclock_base_ns() + ka->kvmclock_offset;
2654 static void kvm_setup_pvclock_page(struct kvm_vcpu *v,
2655 struct gfn_to_hva_cache *cache,
2656 unsigned int offset)
2658 struct kvm_vcpu_arch *vcpu = &v->arch;
2659 struct pvclock_vcpu_time_info guest_hv_clock;
2661 if (unlikely(kvm_read_guest_offset_cached(v->kvm, cache,
2662 &guest_hv_clock, offset, sizeof(guest_hv_clock))))
2665 /* This VCPU is paused, but it's legal for a guest to read another
2666 * VCPU's kvmclock, so we really have to follow the specification where
2667 * it says that version is odd if data is being modified, and even after
2670 * Version field updates must be kept separate. This is because
2671 * kvm_write_guest_cached might use a "rep movs" instruction, and
2672 * writes within a string instruction are weakly ordered. So there
2673 * are three writes overall.
2675 * As a small optimization, only write the version field in the first
2676 * and third write. The vcpu->pv_time cache is still valid, because the
2677 * version field is the first in the struct.
2679 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2681 if (guest_hv_clock.version & 1)
2682 ++guest_hv_clock.version; /* first time write, random junk */
2684 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2685 kvm_write_guest_offset_cached(v->kvm, cache,
2686 &vcpu->hv_clock, offset,
2687 sizeof(vcpu->hv_clock.version));
2691 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2692 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2694 if (vcpu->pvclock_set_guest_stopped_request) {
2695 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2696 vcpu->pvclock_set_guest_stopped_request = false;
2699 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2701 kvm_write_guest_offset_cached(v->kvm, cache,
2702 &vcpu->hv_clock, offset,
2703 sizeof(vcpu->hv_clock));
2707 vcpu->hv_clock.version++;
2708 kvm_write_guest_offset_cached(v->kvm, cache,
2709 &vcpu->hv_clock, offset,
2710 sizeof(vcpu->hv_clock.version));
2713 static int kvm_guest_time_update(struct kvm_vcpu *v)
2715 unsigned long flags, tgt_tsc_khz;
2716 struct kvm_vcpu_arch *vcpu = &v->arch;
2717 struct kvm_arch *ka = &v->kvm->arch;
2719 u64 tsc_timestamp, host_tsc;
2721 bool use_master_clock;
2727 * If the host uses TSC clock, then passthrough TSC as stable
2730 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
2731 use_master_clock = ka->use_master_clock;
2732 if (use_master_clock) {
2733 host_tsc = ka->master_cycle_now;
2734 kernel_ns = ka->master_kernel_ns;
2736 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
2738 /* Keep irq disabled to prevent changes to the clock */
2739 local_irq_save(flags);
2740 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2741 if (unlikely(tgt_tsc_khz == 0)) {
2742 local_irq_restore(flags);
2743 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2746 if (!use_master_clock) {
2748 kernel_ns = get_kvmclock_base_ns();
2751 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2754 * We may have to catch up the TSC to match elapsed wall clock
2755 * time for two reasons, even if kvmclock is used.
2756 * 1) CPU could have been running below the maximum TSC rate
2757 * 2) Broken TSC compensation resets the base at each VCPU
2758 * entry to avoid unknown leaps of TSC even when running
2759 * again on the same CPU. This may cause apparent elapsed
2760 * time to disappear, and the guest to stand still or run
2763 if (vcpu->tsc_catchup) {
2764 u64 tsc = compute_guest_tsc(v, kernel_ns);
2765 if (tsc > tsc_timestamp) {
2766 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2767 tsc_timestamp = tsc;
2771 local_irq_restore(flags);
2773 /* With all the info we got, fill in the values */
2775 if (kvm_has_tsc_control)
2776 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2778 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2779 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2780 &vcpu->hv_clock.tsc_shift,
2781 &vcpu->hv_clock.tsc_to_system_mul);
2782 vcpu->hw_tsc_khz = tgt_tsc_khz;
2785 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2786 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2787 vcpu->last_guest_tsc = tsc_timestamp;
2789 /* If the host uses TSC clocksource, then it is stable */
2791 if (use_master_clock)
2792 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2794 vcpu->hv_clock.flags = pvclock_flags;
2796 if (vcpu->pv_time_enabled)
2797 kvm_setup_pvclock_page(v, &vcpu->pv_time, 0);
2798 if (vcpu->xen.vcpu_info_set)
2799 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_info_cache,
2800 offsetof(struct compat_vcpu_info, time));
2801 if (vcpu->xen.vcpu_time_info_set)
2802 kvm_setup_pvclock_page(v, &vcpu->xen.vcpu_time_info_cache, 0);
2803 if (v == kvm_get_vcpu(v->kvm, 0))
2804 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2809 * kvmclock updates which are isolated to a given vcpu, such as
2810 * vcpu->cpu migration, should not allow system_timestamp from
2811 * the rest of the vcpus to remain static. Otherwise ntp frequency
2812 * correction applies to one vcpu's system_timestamp but not
2815 * So in those cases, request a kvmclock update for all vcpus.
2816 * We need to rate-limit these requests though, as they can
2817 * considerably slow guests that have a large number of vcpus.
2818 * The time for a remote vcpu to update its kvmclock is bound
2819 * by the delay we use to rate-limit the updates.
2822 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2824 static void kvmclock_update_fn(struct work_struct *work)
2827 struct delayed_work *dwork = to_delayed_work(work);
2828 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2829 kvmclock_update_work);
2830 struct kvm *kvm = container_of(ka, struct kvm, arch);
2831 struct kvm_vcpu *vcpu;
2833 kvm_for_each_vcpu(i, vcpu, kvm) {
2834 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2835 kvm_vcpu_kick(vcpu);
2839 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2841 struct kvm *kvm = v->kvm;
2843 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2844 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2845 KVMCLOCK_UPDATE_DELAY);
2848 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2850 static void kvmclock_sync_fn(struct work_struct *work)
2852 struct delayed_work *dwork = to_delayed_work(work);
2853 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2854 kvmclock_sync_work);
2855 struct kvm *kvm = container_of(ka, struct kvm, arch);
2857 if (!kvmclock_periodic_sync)
2860 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2861 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2862 KVMCLOCK_SYNC_PERIOD);
2866 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2868 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2870 /* McStatusWrEn enabled? */
2871 if (guest_cpuid_is_amd_or_hygon(vcpu))
2872 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2877 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2879 u64 mcg_cap = vcpu->arch.mcg_cap;
2880 unsigned bank_num = mcg_cap & 0xff;
2881 u32 msr = msr_info->index;
2882 u64 data = msr_info->data;
2885 case MSR_IA32_MCG_STATUS:
2886 vcpu->arch.mcg_status = data;
2888 case MSR_IA32_MCG_CTL:
2889 if (!(mcg_cap & MCG_CTL_P) &&
2890 (data || !msr_info->host_initiated))
2892 if (data != 0 && data != ~(u64)0)
2894 vcpu->arch.mcg_ctl = data;
2897 if (msr >= MSR_IA32_MC0_CTL &&
2898 msr < MSR_IA32_MCx_CTL(bank_num)) {
2899 u32 offset = array_index_nospec(
2900 msr - MSR_IA32_MC0_CTL,
2901 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2903 /* only 0 or all 1s can be written to IA32_MCi_CTL
2904 * some Linux kernels though clear bit 10 in bank 4 to
2905 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2906 * this to avoid an uncatched #GP in the guest
2908 if ((offset & 0x3) == 0 &&
2909 data != 0 && (data | (1 << 10)) != ~(u64)0)
2913 if (!msr_info->host_initiated &&
2914 (offset & 0x3) == 1 && data != 0) {
2915 if (!can_set_mci_status(vcpu))
2919 vcpu->arch.mce_banks[offset] = data;
2927 static inline bool kvm_pv_async_pf_enabled(struct kvm_vcpu *vcpu)
2929 u64 mask = KVM_ASYNC_PF_ENABLED | KVM_ASYNC_PF_DELIVERY_AS_INT;
2931 return (vcpu->arch.apf.msr_en_val & mask) == mask;
2934 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2936 gpa_t gpa = data & ~0x3f;
2938 /* Bits 4:5 are reserved, Should be zero */
2942 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_VMEXIT) &&
2943 (data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT))
2946 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT) &&
2947 (data & KVM_ASYNC_PF_DELIVERY_AS_INT))
2950 if (!lapic_in_kernel(vcpu))
2951 return data ? 1 : 0;
2953 vcpu->arch.apf.msr_en_val = data;
2955 if (!kvm_pv_async_pf_enabled(vcpu)) {
2956 kvm_clear_async_pf_completion_queue(vcpu);
2957 kvm_async_pf_hash_reset(vcpu);
2961 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2965 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2966 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2968 kvm_async_pf_wakeup_all(vcpu);
2973 static int kvm_pv_enable_async_pf_int(struct kvm_vcpu *vcpu, u64 data)
2975 /* Bits 8-63 are reserved */
2979 if (!lapic_in_kernel(vcpu))
2982 vcpu->arch.apf.msr_int_val = data;
2984 vcpu->arch.apf.vec = data & KVM_ASYNC_PF_VEC_MASK;
2989 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2991 vcpu->arch.pv_time_enabled = false;
2992 vcpu->arch.time = 0;
2995 static void kvm_vcpu_flush_tlb_all(struct kvm_vcpu *vcpu)
2997 ++vcpu->stat.tlb_flush;
2998 static_call(kvm_x86_tlb_flush_all)(vcpu);
3001 static void kvm_vcpu_flush_tlb_guest(struct kvm_vcpu *vcpu)
3003 ++vcpu->stat.tlb_flush;
3004 static_call(kvm_x86_tlb_flush_guest)(vcpu);
3007 static void record_steal_time(struct kvm_vcpu *vcpu)
3009 struct kvm_host_map map;
3010 struct kvm_steal_time *st;
3012 if (kvm_xen_msr_enabled(vcpu->kvm)) {
3013 kvm_xen_runstate_set_running(vcpu);
3017 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3020 /* -EAGAIN is returned in atomic context so we can just return. */
3021 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT,
3022 &map, &vcpu->arch.st.cache, false))
3026 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
3029 * Doing a TLB flush here, on the guest's behalf, can avoid
3032 if (guest_pv_has(vcpu, KVM_FEATURE_PV_TLB_FLUSH)) {
3033 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
3034 st->preempted & KVM_VCPU_FLUSH_TLB);
3035 if (xchg(&st->preempted, 0) & KVM_VCPU_FLUSH_TLB)
3036 kvm_vcpu_flush_tlb_guest(vcpu);
3039 vcpu->arch.st.preempted = 0;
3041 if (st->version & 1)
3042 st->version += 1; /* first time write, random junk */
3048 st->steal += current->sched_info.run_delay -
3049 vcpu->arch.st.last_steal;
3050 vcpu->arch.st.last_steal = current->sched_info.run_delay;
3056 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, false);
3059 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3062 u32 msr = msr_info->index;
3063 u64 data = msr_info->data;
3065 if (msr && msr == vcpu->kvm->arch.xen_hvm_config.msr)
3066 return kvm_xen_write_hypercall_page(vcpu, data);
3069 case MSR_AMD64_NB_CFG:
3070 case MSR_IA32_UCODE_WRITE:
3071 case MSR_VM_HSAVE_PA:
3072 case MSR_AMD64_PATCH_LOADER:
3073 case MSR_AMD64_BU_CFG2:
3074 case MSR_AMD64_DC_CFG:
3075 case MSR_F15H_EX_CFG:
3078 case MSR_IA32_UCODE_REV:
3079 if (msr_info->host_initiated)
3080 vcpu->arch.microcode_version = data;
3082 case MSR_IA32_ARCH_CAPABILITIES:
3083 if (!msr_info->host_initiated)
3085 vcpu->arch.arch_capabilities = data;
3087 case MSR_IA32_PERF_CAPABILITIES: {
3088 struct kvm_msr_entry msr_ent = {.index = msr, .data = 0};
3090 if (!msr_info->host_initiated)
3092 if (guest_cpuid_has(vcpu, X86_FEATURE_PDCM) && kvm_get_msr_feature(&msr_ent))
3094 if (data & ~msr_ent.data)
3097 vcpu->arch.perf_capabilities = data;
3102 return set_efer(vcpu, msr_info);
3104 data &= ~(u64)0x40; /* ignore flush filter disable */
3105 data &= ~(u64)0x100; /* ignore ignne emulation enable */
3106 data &= ~(u64)0x8; /* ignore TLB cache disable */
3108 /* Handle McStatusWrEn */
3109 if (data == BIT_ULL(18)) {
3110 vcpu->arch.msr_hwcr = data;
3111 } else if (data != 0) {
3112 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
3117 case MSR_FAM10H_MMIO_CONF_BASE:
3119 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
3124 case 0x200 ... 0x2ff:
3125 return kvm_mtrr_set_msr(vcpu, msr, data);
3126 case MSR_IA32_APICBASE:
3127 return kvm_set_apic_base(vcpu, msr_info);
3128 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3129 return kvm_x2apic_msr_write(vcpu, msr, data);
3130 case MSR_IA32_TSCDEADLINE:
3131 kvm_set_lapic_tscdeadline_msr(vcpu, data);
3133 case MSR_IA32_TSC_ADJUST:
3134 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
3135 if (!msr_info->host_initiated) {
3136 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
3137 adjust_tsc_offset_guest(vcpu, adj);
3139 vcpu->arch.ia32_tsc_adjust_msr = data;
3142 case MSR_IA32_MISC_ENABLE:
3143 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
3144 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
3145 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
3147 vcpu->arch.ia32_misc_enable_msr = data;
3148 kvm_update_cpuid_runtime(vcpu);
3150 vcpu->arch.ia32_misc_enable_msr = data;
3153 case MSR_IA32_SMBASE:
3154 if (!msr_info->host_initiated)
3156 vcpu->arch.smbase = data;
3158 case MSR_IA32_POWER_CTL:
3159 vcpu->arch.msr_ia32_power_ctl = data;
3162 if (msr_info->host_initiated) {
3163 kvm_synchronize_tsc(vcpu, data);
3165 u64 adj = kvm_compute_tsc_offset(vcpu, data) - vcpu->arch.l1_tsc_offset;
3166 adjust_tsc_offset_guest(vcpu, adj);
3167 vcpu->arch.ia32_tsc_adjust_msr += adj;
3171 if (!msr_info->host_initiated &&
3172 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3175 * KVM supports exposing PT to the guest, but does not support
3176 * IA32_XSS[bit 8]. Guests have to use RDMSR/WRMSR rather than
3177 * XSAVES/XRSTORS to save/restore PT MSRs.
3179 if (data & ~supported_xss)
3181 vcpu->arch.ia32_xss = data;
3184 if (!msr_info->host_initiated)
3186 vcpu->arch.smi_count = data;
3188 case MSR_KVM_WALL_CLOCK_NEW:
3189 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3192 vcpu->kvm->arch.wall_clock = data;
3193 kvm_write_wall_clock(vcpu->kvm, data, 0);
3195 case MSR_KVM_WALL_CLOCK:
3196 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3199 vcpu->kvm->arch.wall_clock = data;
3200 kvm_write_wall_clock(vcpu->kvm, data, 0);
3202 case MSR_KVM_SYSTEM_TIME_NEW:
3203 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3206 kvm_write_system_time(vcpu, data, false, msr_info->host_initiated);
3208 case MSR_KVM_SYSTEM_TIME:
3209 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3212 kvm_write_system_time(vcpu, data, true, msr_info->host_initiated);
3214 case MSR_KVM_ASYNC_PF_EN:
3215 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3218 if (kvm_pv_enable_async_pf(vcpu, data))
3221 case MSR_KVM_ASYNC_PF_INT:
3222 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3225 if (kvm_pv_enable_async_pf_int(vcpu, data))
3228 case MSR_KVM_ASYNC_PF_ACK:
3229 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3232 vcpu->arch.apf.pageready_pending = false;
3233 kvm_check_async_pf_completion(vcpu);
3236 case MSR_KVM_STEAL_TIME:
3237 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3240 if (unlikely(!sched_info_on()))
3243 if (data & KVM_STEAL_RESERVED_MASK)
3246 vcpu->arch.st.msr_val = data;
3248 if (!(data & KVM_MSR_ENABLED))
3251 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3254 case MSR_KVM_PV_EOI_EN:
3255 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3258 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
3262 case MSR_KVM_POLL_CONTROL:
3263 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3266 /* only enable bit supported */
3267 if (data & (-1ULL << 1))
3270 vcpu->arch.msr_kvm_poll_control = data;
3273 case MSR_IA32_MCG_CTL:
3274 case MSR_IA32_MCG_STATUS:
3275 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3276 return set_msr_mce(vcpu, msr_info);
3278 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3279 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3282 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3283 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3284 if (kvm_pmu_is_valid_msr(vcpu, msr))
3285 return kvm_pmu_set_msr(vcpu, msr_info);
3287 if (pr || data != 0)
3288 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
3289 "0x%x data 0x%llx\n", msr, data);
3291 case MSR_K7_CLK_CTL:
3293 * Ignore all writes to this no longer documented MSR.
3294 * Writes are only relevant for old K7 processors,
3295 * all pre-dating SVM, but a recommended workaround from
3296 * AMD for these chips. It is possible to specify the
3297 * affected processor models on the command line, hence
3298 * the need to ignore the workaround.
3301 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3302 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3303 case HV_X64_MSR_SYNDBG_OPTIONS:
3304 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3305 case HV_X64_MSR_CRASH_CTL:
3306 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3307 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3308 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3309 case HV_X64_MSR_TSC_EMULATION_STATUS:
3310 return kvm_hv_set_msr_common(vcpu, msr, data,
3311 msr_info->host_initiated);
3312 case MSR_IA32_BBL_CR_CTL3:
3313 /* Drop writes to this legacy MSR -- see rdmsr
3314 * counterpart for further detail.
3316 if (report_ignored_msrs)
3317 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
3320 case MSR_AMD64_OSVW_ID_LENGTH:
3321 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3323 vcpu->arch.osvw.length = data;
3325 case MSR_AMD64_OSVW_STATUS:
3326 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3328 vcpu->arch.osvw.status = data;
3330 case MSR_PLATFORM_INFO:
3331 if (!msr_info->host_initiated ||
3332 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
3333 cpuid_fault_enabled(vcpu)))
3335 vcpu->arch.msr_platform_info = data;
3337 case MSR_MISC_FEATURES_ENABLES:
3338 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
3339 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
3340 !supports_cpuid_fault(vcpu)))
3342 vcpu->arch.msr_misc_features_enables = data;
3345 if (kvm_pmu_is_valid_msr(vcpu, msr))
3346 return kvm_pmu_set_msr(vcpu, msr_info);
3347 return KVM_MSR_RET_INVALID;
3351 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
3353 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
3356 u64 mcg_cap = vcpu->arch.mcg_cap;
3357 unsigned bank_num = mcg_cap & 0xff;
3360 case MSR_IA32_P5_MC_ADDR:
3361 case MSR_IA32_P5_MC_TYPE:
3364 case MSR_IA32_MCG_CAP:
3365 data = vcpu->arch.mcg_cap;
3367 case MSR_IA32_MCG_CTL:
3368 if (!(mcg_cap & MCG_CTL_P) && !host)
3370 data = vcpu->arch.mcg_ctl;
3372 case MSR_IA32_MCG_STATUS:
3373 data = vcpu->arch.mcg_status;
3376 if (msr >= MSR_IA32_MC0_CTL &&
3377 msr < MSR_IA32_MCx_CTL(bank_num)) {
3378 u32 offset = array_index_nospec(
3379 msr - MSR_IA32_MC0_CTL,
3380 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3382 data = vcpu->arch.mce_banks[offset];
3391 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3393 switch (msr_info->index) {
3394 case MSR_IA32_PLATFORM_ID:
3395 case MSR_IA32_EBL_CR_POWERON:
3396 case MSR_IA32_LASTBRANCHFROMIP:
3397 case MSR_IA32_LASTBRANCHTOIP:
3398 case MSR_IA32_LASTINTFROMIP:
3399 case MSR_IA32_LASTINTTOIP:
3401 case MSR_K8_TSEG_ADDR:
3402 case MSR_K8_TSEG_MASK:
3403 case MSR_VM_HSAVE_PA:
3404 case MSR_K8_INT_PENDING_MSG:
3405 case MSR_AMD64_NB_CFG:
3406 case MSR_FAM10H_MMIO_CONF_BASE:
3407 case MSR_AMD64_BU_CFG2:
3408 case MSR_IA32_PERF_CTL:
3409 case MSR_AMD64_DC_CFG:
3410 case MSR_F15H_EX_CFG:
3412 * Intel Sandy Bridge CPUs must support the RAPL (running average power
3413 * limit) MSRs. Just return 0, as we do not want to expose the host
3414 * data here. Do not conditionalize this on CPUID, as KVM does not do
3415 * so for existing CPU-specific MSRs.
3417 case MSR_RAPL_POWER_UNIT:
3418 case MSR_PP0_ENERGY_STATUS: /* Power plane 0 (core) */
3419 case MSR_PP1_ENERGY_STATUS: /* Power plane 1 (graphics uncore) */
3420 case MSR_PKG_ENERGY_STATUS: /* Total package */
3421 case MSR_DRAM_ENERGY_STATUS: /* DRAM controller */
3424 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3425 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3426 return kvm_pmu_get_msr(vcpu, msr_info);
3427 if (!msr_info->host_initiated)
3431 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3432 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3433 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3434 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3435 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3436 return kvm_pmu_get_msr(vcpu, msr_info);
3439 case MSR_IA32_UCODE_REV:
3440 msr_info->data = vcpu->arch.microcode_version;
3442 case MSR_IA32_ARCH_CAPABILITIES:
3443 if (!msr_info->host_initiated &&
3444 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3446 msr_info->data = vcpu->arch.arch_capabilities;
3448 case MSR_IA32_PERF_CAPABILITIES:
3449 if (!msr_info->host_initiated &&
3450 !guest_cpuid_has(vcpu, X86_FEATURE_PDCM))
3452 msr_info->data = vcpu->arch.perf_capabilities;
3454 case MSR_IA32_POWER_CTL:
3455 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3457 case MSR_IA32_TSC: {
3459 * Intel SDM states that MSR_IA32_TSC read adds the TSC offset
3460 * even when not intercepted. AMD manual doesn't explicitly
3461 * state this but appears to behave the same.
3463 * On userspace reads and writes, however, we unconditionally
3464 * return L1's TSC value to ensure backwards-compatible
3465 * behavior for migration.
3467 u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
3468 vcpu->arch.tsc_offset;
3470 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
3474 case 0x200 ... 0x2ff:
3475 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3476 case 0xcd: /* fsb frequency */
3480 * MSR_EBC_FREQUENCY_ID
3481 * Conservative value valid for even the basic CPU models.
3482 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3483 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3484 * and 266MHz for model 3, or 4. Set Core Clock
3485 * Frequency to System Bus Frequency Ratio to 1 (bits
3486 * 31:24) even though these are only valid for CPU
3487 * models > 2, however guests may end up dividing or
3488 * multiplying by zero otherwise.
3490 case MSR_EBC_FREQUENCY_ID:
3491 msr_info->data = 1 << 24;
3493 case MSR_IA32_APICBASE:
3494 msr_info->data = kvm_get_apic_base(vcpu);
3496 case APIC_BASE_MSR ... APIC_BASE_MSR + 0xff:
3497 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3498 case MSR_IA32_TSCDEADLINE:
3499 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3501 case MSR_IA32_TSC_ADJUST:
3502 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3504 case MSR_IA32_MISC_ENABLE:
3505 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3507 case MSR_IA32_SMBASE:
3508 if (!msr_info->host_initiated)
3510 msr_info->data = vcpu->arch.smbase;
3513 msr_info->data = vcpu->arch.smi_count;
3515 case MSR_IA32_PERF_STATUS:
3516 /* TSC increment by tick */
3517 msr_info->data = 1000ULL;
3518 /* CPU multiplier */
3519 msr_info->data |= (((uint64_t)4ULL) << 40);
3522 msr_info->data = vcpu->arch.efer;
3524 case MSR_KVM_WALL_CLOCK:
3525 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3528 msr_info->data = vcpu->kvm->arch.wall_clock;
3530 case MSR_KVM_WALL_CLOCK_NEW:
3531 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3534 msr_info->data = vcpu->kvm->arch.wall_clock;
3536 case MSR_KVM_SYSTEM_TIME:
3537 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE))
3540 msr_info->data = vcpu->arch.time;
3542 case MSR_KVM_SYSTEM_TIME_NEW:
3543 if (!guest_pv_has(vcpu, KVM_FEATURE_CLOCKSOURCE2))
3546 msr_info->data = vcpu->arch.time;
3548 case MSR_KVM_ASYNC_PF_EN:
3549 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3552 msr_info->data = vcpu->arch.apf.msr_en_val;
3554 case MSR_KVM_ASYNC_PF_INT:
3555 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF_INT))
3558 msr_info->data = vcpu->arch.apf.msr_int_val;
3560 case MSR_KVM_ASYNC_PF_ACK:
3561 if (!guest_pv_has(vcpu, KVM_FEATURE_ASYNC_PF))
3566 case MSR_KVM_STEAL_TIME:
3567 if (!guest_pv_has(vcpu, KVM_FEATURE_STEAL_TIME))
3570 msr_info->data = vcpu->arch.st.msr_val;
3572 case MSR_KVM_PV_EOI_EN:
3573 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_EOI))
3576 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3578 case MSR_KVM_POLL_CONTROL:
3579 if (!guest_pv_has(vcpu, KVM_FEATURE_POLL_CONTROL))
3582 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3584 case MSR_IA32_P5_MC_ADDR:
3585 case MSR_IA32_P5_MC_TYPE:
3586 case MSR_IA32_MCG_CAP:
3587 case MSR_IA32_MCG_CTL:
3588 case MSR_IA32_MCG_STATUS:
3589 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3590 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3591 msr_info->host_initiated);
3593 if (!msr_info->host_initiated &&
3594 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3596 msr_info->data = vcpu->arch.ia32_xss;
3598 case MSR_K7_CLK_CTL:
3600 * Provide expected ramp-up count for K7. All other
3601 * are set to zero, indicating minimum divisors for
3604 * This prevents guest kernels on AMD host with CPU
3605 * type 6, model 8 and higher from exploding due to
3606 * the rdmsr failing.
3608 msr_info->data = 0x20000000;
3610 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3611 case HV_X64_MSR_SYNDBG_CONTROL ... HV_X64_MSR_SYNDBG_PENDING_BUFFER:
3612 case HV_X64_MSR_SYNDBG_OPTIONS:
3613 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3614 case HV_X64_MSR_CRASH_CTL:
3615 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3616 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3617 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3618 case HV_X64_MSR_TSC_EMULATION_STATUS:
3619 return kvm_hv_get_msr_common(vcpu,
3620 msr_info->index, &msr_info->data,
3621 msr_info->host_initiated);
3622 case MSR_IA32_BBL_CR_CTL3:
3623 /* This legacy MSR exists but isn't fully documented in current
3624 * silicon. It is however accessed by winxp in very narrow
3625 * scenarios where it sets bit #19, itself documented as
3626 * a "reserved" bit. Best effort attempt to source coherent
3627 * read data here should the balance of the register be
3628 * interpreted by the guest:
3630 * L2 cache control register 3: 64GB range, 256KB size,
3631 * enabled, latency 0x1, configured
3633 msr_info->data = 0xbe702111;
3635 case MSR_AMD64_OSVW_ID_LENGTH:
3636 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3638 msr_info->data = vcpu->arch.osvw.length;
3640 case MSR_AMD64_OSVW_STATUS:
3641 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3643 msr_info->data = vcpu->arch.osvw.status;
3645 case MSR_PLATFORM_INFO:
3646 if (!msr_info->host_initiated &&
3647 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3649 msr_info->data = vcpu->arch.msr_platform_info;
3651 case MSR_MISC_FEATURES_ENABLES:
3652 msr_info->data = vcpu->arch.msr_misc_features_enables;
3655 msr_info->data = vcpu->arch.msr_hwcr;
3658 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3659 return kvm_pmu_get_msr(vcpu, msr_info);
3660 return KVM_MSR_RET_INVALID;
3664 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3667 * Read or write a bunch of msrs. All parameters are kernel addresses.
3669 * @return number of msrs set successfully.
3671 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3672 struct kvm_msr_entry *entries,
3673 int (*do_msr)(struct kvm_vcpu *vcpu,
3674 unsigned index, u64 *data))
3678 for (i = 0; i < msrs->nmsrs; ++i)
3679 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3686 * Read or write a bunch of msrs. Parameters are user addresses.
3688 * @return number of msrs set successfully.
3690 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3691 int (*do_msr)(struct kvm_vcpu *vcpu,
3692 unsigned index, u64 *data),
3695 struct kvm_msrs msrs;
3696 struct kvm_msr_entry *entries;
3701 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3705 if (msrs.nmsrs >= MAX_IO_MSRS)
3708 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3709 entries = memdup_user(user_msrs->entries, size);
3710 if (IS_ERR(entries)) {
3711 r = PTR_ERR(entries);
3715 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3720 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3731 static inline bool kvm_can_mwait_in_guest(void)
3733 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3734 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3735 boot_cpu_has(X86_FEATURE_ARAT);
3738 static int kvm_ioctl_get_supported_hv_cpuid(struct kvm_vcpu *vcpu,
3739 struct kvm_cpuid2 __user *cpuid_arg)
3741 struct kvm_cpuid2 cpuid;
3745 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3748 r = kvm_get_hv_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3753 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3759 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3764 case KVM_CAP_IRQCHIP:
3766 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3767 case KVM_CAP_SET_TSS_ADDR:
3768 case KVM_CAP_EXT_CPUID:
3769 case KVM_CAP_EXT_EMUL_CPUID:
3770 case KVM_CAP_CLOCKSOURCE:
3772 case KVM_CAP_NOP_IO_DELAY:
3773 case KVM_CAP_MP_STATE:
3774 case KVM_CAP_SYNC_MMU:
3775 case KVM_CAP_USER_NMI:
3776 case KVM_CAP_REINJECT_CONTROL:
3777 case KVM_CAP_IRQ_INJECT_STATUS:
3778 case KVM_CAP_IOEVENTFD:
3779 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3781 case KVM_CAP_PIT_STATE2:
3782 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3783 case KVM_CAP_VCPU_EVENTS:
3784 case KVM_CAP_HYPERV:
3785 case KVM_CAP_HYPERV_VAPIC:
3786 case KVM_CAP_HYPERV_SPIN:
3787 case KVM_CAP_HYPERV_SYNIC:
3788 case KVM_CAP_HYPERV_SYNIC2:
3789 case KVM_CAP_HYPERV_VP_INDEX:
3790 case KVM_CAP_HYPERV_EVENTFD:
3791 case KVM_CAP_HYPERV_TLBFLUSH:
3792 case KVM_CAP_HYPERV_SEND_IPI:
3793 case KVM_CAP_HYPERV_CPUID:
3794 case KVM_CAP_SYS_HYPERV_CPUID:
3795 case KVM_CAP_PCI_SEGMENT:
3796 case KVM_CAP_DEBUGREGS:
3797 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3799 case KVM_CAP_ASYNC_PF:
3800 case KVM_CAP_ASYNC_PF_INT:
3801 case KVM_CAP_GET_TSC_KHZ:
3802 case KVM_CAP_KVMCLOCK_CTRL:
3803 case KVM_CAP_READONLY_MEM:
3804 case KVM_CAP_HYPERV_TIME:
3805 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3806 case KVM_CAP_TSC_DEADLINE_TIMER:
3807 case KVM_CAP_DISABLE_QUIRKS:
3808 case KVM_CAP_SET_BOOT_CPU_ID:
3809 case KVM_CAP_SPLIT_IRQCHIP:
3810 case KVM_CAP_IMMEDIATE_EXIT:
3811 case KVM_CAP_PMU_EVENT_FILTER:
3812 case KVM_CAP_GET_MSR_FEATURES:
3813 case KVM_CAP_MSR_PLATFORM_INFO:
3814 case KVM_CAP_EXCEPTION_PAYLOAD:
3815 case KVM_CAP_SET_GUEST_DEBUG:
3816 case KVM_CAP_LAST_CPU:
3817 case KVM_CAP_X86_USER_SPACE_MSR:
3818 case KVM_CAP_X86_MSR_FILTER:
3819 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
3820 #ifdef CONFIG_X86_SGX_KVM
3821 case KVM_CAP_SGX_ATTRIBUTE:
3823 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
3826 case KVM_CAP_SET_GUEST_DEBUG2:
3827 return KVM_GUESTDBG_VALID_MASK;
3828 #ifdef CONFIG_KVM_XEN
3829 case KVM_CAP_XEN_HVM:
3830 r = KVM_XEN_HVM_CONFIG_HYPERCALL_MSR |
3831 KVM_XEN_HVM_CONFIG_INTERCEPT_HCALL |
3832 KVM_XEN_HVM_CONFIG_SHARED_INFO;
3833 if (sched_info_on())
3834 r |= KVM_XEN_HVM_CONFIG_RUNSTATE;
3837 case KVM_CAP_SYNC_REGS:
3838 r = KVM_SYNC_X86_VALID_FIELDS;
3840 case KVM_CAP_ADJUST_CLOCK:
3841 r = KVM_CLOCK_TSC_STABLE;
3843 case KVM_CAP_X86_DISABLE_EXITS:
3844 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3845 KVM_X86_DISABLE_EXITS_CSTATE;
3846 if(kvm_can_mwait_in_guest())
3847 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3849 case KVM_CAP_X86_SMM:
3850 /* SMBASE is usually relocated above 1M on modern chipsets,
3851 * and SMM handlers might indeed rely on 4G segment limits,
3852 * so do not report SMM to be available if real mode is
3853 * emulated via vm86 mode. Still, do not go to great lengths
3854 * to avoid userspace's usage of the feature, because it is a
3855 * fringe case that is not enabled except via specific settings
3856 * of the module parameters.
3858 r = static_call(kvm_x86_has_emulated_msr)(kvm, MSR_IA32_SMBASE);
3861 r = !static_call(kvm_x86_cpu_has_accelerated_tpr)();
3863 case KVM_CAP_NR_VCPUS:
3864 r = KVM_SOFT_MAX_VCPUS;
3866 case KVM_CAP_MAX_VCPUS:
3869 case KVM_CAP_MAX_VCPU_ID:
3870 r = KVM_MAX_VCPU_ID;
3872 case KVM_CAP_PV_MMU: /* obsolete */
3876 r = KVM_MAX_MCE_BANKS;
3879 r = boot_cpu_has(X86_FEATURE_XSAVE);
3881 case KVM_CAP_TSC_CONTROL:
3882 r = kvm_has_tsc_control;
3884 case KVM_CAP_X2APIC_API:
3885 r = KVM_X2APIC_API_VALID_FLAGS;
3887 case KVM_CAP_NESTED_STATE:
3888 r = kvm_x86_ops.nested_ops->get_state ?
3889 kvm_x86_ops.nested_ops->get_state(NULL, NULL, 0) : 0;
3891 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3892 r = kvm_x86_ops.enable_direct_tlbflush != NULL;
3894 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3895 r = kvm_x86_ops.nested_ops->enable_evmcs != NULL;
3897 case KVM_CAP_SMALLER_MAXPHYADDR:
3898 r = (int) allow_smaller_maxphyaddr;
3900 case KVM_CAP_STEAL_TIME:
3901 r = sched_info_on();
3903 case KVM_CAP_X86_BUS_LOCK_EXIT:
3904 if (kvm_has_bus_lock_exit)
3905 r = KVM_BUS_LOCK_DETECTION_OFF |
3906 KVM_BUS_LOCK_DETECTION_EXIT;
3917 long kvm_arch_dev_ioctl(struct file *filp,
3918 unsigned int ioctl, unsigned long arg)
3920 void __user *argp = (void __user *)arg;
3924 case KVM_GET_MSR_INDEX_LIST: {
3925 struct kvm_msr_list __user *user_msr_list = argp;
3926 struct kvm_msr_list msr_list;
3930 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3933 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3934 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3937 if (n < msr_list.nmsrs)
3940 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3941 num_msrs_to_save * sizeof(u32)))
3943 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3945 num_emulated_msrs * sizeof(u32)))
3950 case KVM_GET_SUPPORTED_CPUID:
3951 case KVM_GET_EMULATED_CPUID: {
3952 struct kvm_cpuid2 __user *cpuid_arg = argp;
3953 struct kvm_cpuid2 cpuid;
3956 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3959 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3965 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3970 case KVM_X86_GET_MCE_CAP_SUPPORTED:
3972 if (copy_to_user(argp, &kvm_mce_cap_supported,
3973 sizeof(kvm_mce_cap_supported)))
3977 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3978 struct kvm_msr_list __user *user_msr_list = argp;
3979 struct kvm_msr_list msr_list;
3983 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3986 msr_list.nmsrs = num_msr_based_features;
3987 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3990 if (n < msr_list.nmsrs)
3993 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3994 num_msr_based_features * sizeof(u32)))
4000 r = msr_io(NULL, argp, do_get_msr_feature, 1);
4002 case KVM_GET_SUPPORTED_HV_CPUID:
4003 r = kvm_ioctl_get_supported_hv_cpuid(NULL, argp);
4013 static void wbinvd_ipi(void *garbage)
4018 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
4020 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
4023 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
4025 /* Address WBINVD may be executed by guest */
4026 if (need_emulate_wbinvd(vcpu)) {
4027 if (static_call(kvm_x86_has_wbinvd_exit)())
4028 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
4029 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
4030 smp_call_function_single(vcpu->cpu,
4031 wbinvd_ipi, NULL, 1);
4034 static_call(kvm_x86_vcpu_load)(vcpu, cpu);
4036 /* Save host pkru register if supported */
4037 vcpu->arch.host_pkru = read_pkru();
4039 /* Apply any externally detected TSC adjustments (due to suspend) */
4040 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
4041 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
4042 vcpu->arch.tsc_offset_adjustment = 0;
4043 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4046 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
4047 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
4048 rdtsc() - vcpu->arch.last_host_tsc;
4050 mark_tsc_unstable("KVM discovered backwards TSC");
4052 if (kvm_check_tsc_unstable()) {
4053 u64 offset = kvm_compute_tsc_offset(vcpu,
4054 vcpu->arch.last_guest_tsc);
4055 kvm_vcpu_write_tsc_offset(vcpu, offset);
4056 vcpu->arch.tsc_catchup = 1;
4059 if (kvm_lapic_hv_timer_in_use(vcpu))
4060 kvm_lapic_restart_hv_timer(vcpu);
4063 * On a host with synchronized TSC, there is no need to update
4064 * kvmclock on vcpu->cpu migration
4066 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
4067 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
4068 if (vcpu->cpu != cpu)
4069 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
4073 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
4076 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
4078 struct kvm_host_map map;
4079 struct kvm_steal_time *st;
4082 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
4085 if (vcpu->arch.st.preempted)
4089 * Take the srcu lock as memslots will be accessed to check the gfn
4090 * cache generation against the memslots generation.
4092 idx = srcu_read_lock(&vcpu->kvm->srcu);
4094 if (kvm_map_gfn(vcpu, vcpu->arch.st.msr_val >> PAGE_SHIFT, &map,
4095 &vcpu->arch.st.cache, true))
4099 offset_in_page(vcpu->arch.st.msr_val & KVM_STEAL_VALID_BITS);
4101 st->preempted = vcpu->arch.st.preempted = KVM_VCPU_PREEMPTED;
4103 kvm_unmap_gfn(vcpu, &map, &vcpu->arch.st.cache, true, true);
4106 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4109 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
4111 if (vcpu->preempted && !vcpu->arch.guest_state_protected)
4112 vcpu->arch.preempted_in_kernel = !static_call(kvm_x86_get_cpl)(vcpu);
4114 if (kvm_xen_msr_enabled(vcpu->kvm))
4115 kvm_xen_runstate_set_preempted(vcpu);
4117 kvm_steal_time_set_preempted(vcpu);
4119 static_call(kvm_x86_vcpu_put)(vcpu);
4120 vcpu->arch.last_host_tsc = rdtsc();
4122 * If userspace has set any breakpoints or watchpoints, dr6 is restored
4123 * on every vmexit, but if not, we might have a stale dr6 from the
4124 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
4129 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
4130 struct kvm_lapic_state *s)
4132 if (vcpu->arch.apicv_active)
4133 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
4135 return kvm_apic_get_state(vcpu, s);
4138 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
4139 struct kvm_lapic_state *s)
4143 r = kvm_apic_set_state(vcpu, s);
4146 update_cr8_intercept(vcpu);
4151 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
4154 * We can accept userspace's request for interrupt injection
4155 * as long as we have a place to store the interrupt number.
4156 * The actual injection will happen when the CPU is able to
4157 * deliver the interrupt.
4159 if (kvm_cpu_has_extint(vcpu))
4162 /* Acknowledging ExtINT does not happen if LINT0 is masked. */
4163 return (!lapic_in_kernel(vcpu) ||
4164 kvm_apic_accept_pic_intr(vcpu));
4167 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
4169 return kvm_arch_interrupt_allowed(vcpu) &&
4170 kvm_cpu_accept_dm_intr(vcpu);
4173 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
4174 struct kvm_interrupt *irq)
4176 if (irq->irq >= KVM_NR_INTERRUPTS)
4179 if (!irqchip_in_kernel(vcpu->kvm)) {
4180 kvm_queue_interrupt(vcpu, irq->irq, false);
4181 kvm_make_request(KVM_REQ_EVENT, vcpu);
4186 * With in-kernel LAPIC, we only use this to inject EXTINT, so
4187 * fail for in-kernel 8259.
4189 if (pic_in_kernel(vcpu->kvm))
4192 if (vcpu->arch.pending_external_vector != -1)
4195 vcpu->arch.pending_external_vector = irq->irq;
4196 kvm_make_request(KVM_REQ_EVENT, vcpu);
4200 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
4202 kvm_inject_nmi(vcpu);
4207 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
4209 kvm_make_request(KVM_REQ_SMI, vcpu);
4214 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
4215 struct kvm_tpr_access_ctl *tac)
4219 vcpu->arch.tpr_access_reporting = !!tac->enabled;
4223 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
4227 unsigned bank_num = mcg_cap & 0xff, bank;
4230 if (!bank_num || bank_num > KVM_MAX_MCE_BANKS)
4232 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
4235 vcpu->arch.mcg_cap = mcg_cap;
4236 /* Init IA32_MCG_CTL to all 1s */
4237 if (mcg_cap & MCG_CTL_P)
4238 vcpu->arch.mcg_ctl = ~(u64)0;
4239 /* Init IA32_MCi_CTL to all 1s */
4240 for (bank = 0; bank < bank_num; bank++)
4241 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
4243 static_call(kvm_x86_setup_mce)(vcpu);
4248 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
4249 struct kvm_x86_mce *mce)
4251 u64 mcg_cap = vcpu->arch.mcg_cap;
4252 unsigned bank_num = mcg_cap & 0xff;
4253 u64 *banks = vcpu->arch.mce_banks;
4255 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
4258 * if IA32_MCG_CTL is not all 1s, the uncorrected error
4259 * reporting is disabled
4261 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
4262 vcpu->arch.mcg_ctl != ~(u64)0)
4264 banks += 4 * mce->bank;
4266 * if IA32_MCi_CTL is not all 1s, the uncorrected error
4267 * reporting is disabled for the bank
4269 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
4271 if (mce->status & MCI_STATUS_UC) {
4272 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
4273 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
4274 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4277 if (banks[1] & MCI_STATUS_VAL)
4278 mce->status |= MCI_STATUS_OVER;
4279 banks[2] = mce->addr;
4280 banks[3] = mce->misc;
4281 vcpu->arch.mcg_status = mce->mcg_status;
4282 banks[1] = mce->status;
4283 kvm_queue_exception(vcpu, MC_VECTOR);
4284 } else if (!(banks[1] & MCI_STATUS_VAL)
4285 || !(banks[1] & MCI_STATUS_UC)) {
4286 if (banks[1] & MCI_STATUS_VAL)
4287 mce->status |= MCI_STATUS_OVER;
4288 banks[2] = mce->addr;
4289 banks[3] = mce->misc;
4290 banks[1] = mce->status;
4292 banks[1] |= MCI_STATUS_OVER;
4296 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
4297 struct kvm_vcpu_events *events)
4301 if (kvm_check_request(KVM_REQ_SMI, vcpu))
4305 * In guest mode, payload delivery should be deferred,
4306 * so that the L1 hypervisor can intercept #PF before
4307 * CR2 is modified (or intercept #DB before DR6 is
4308 * modified under nVMX). Unless the per-VM capability,
4309 * KVM_CAP_EXCEPTION_PAYLOAD, is set, we may not defer the delivery of
4310 * an exception payload and handle after a KVM_GET_VCPU_EVENTS. Since we
4311 * opportunistically defer the exception payload, deliver it if the
4312 * capability hasn't been requested before processing a
4313 * KVM_GET_VCPU_EVENTS.
4315 if (!vcpu->kvm->arch.exception_payload_enabled &&
4316 vcpu->arch.exception.pending && vcpu->arch.exception.has_payload)
4317 kvm_deliver_exception_payload(vcpu);
4320 * The API doesn't provide the instruction length for software
4321 * exceptions, so don't report them. As long as the guest RIP
4322 * isn't advanced, we should expect to encounter the exception
4325 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
4326 events->exception.injected = 0;
4327 events->exception.pending = 0;
4329 events->exception.injected = vcpu->arch.exception.injected;
4330 events->exception.pending = vcpu->arch.exception.pending;
4332 * For ABI compatibility, deliberately conflate
4333 * pending and injected exceptions when
4334 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
4336 if (!vcpu->kvm->arch.exception_payload_enabled)
4337 events->exception.injected |=
4338 vcpu->arch.exception.pending;
4340 events->exception.nr = vcpu->arch.exception.nr;
4341 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
4342 events->exception.error_code = vcpu->arch.exception.error_code;
4343 events->exception_has_payload = vcpu->arch.exception.has_payload;
4344 events->exception_payload = vcpu->arch.exception.payload;
4346 events->interrupt.injected =
4347 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
4348 events->interrupt.nr = vcpu->arch.interrupt.nr;
4349 events->interrupt.soft = 0;
4350 events->interrupt.shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
4352 events->nmi.injected = vcpu->arch.nmi_injected;
4353 events->nmi.pending = vcpu->arch.nmi_pending != 0;
4354 events->nmi.masked = static_call(kvm_x86_get_nmi_mask)(vcpu);
4355 events->nmi.pad = 0;
4357 events->sipi_vector = 0; /* never valid when reporting to user space */
4359 events->smi.smm = is_smm(vcpu);
4360 events->smi.pending = vcpu->arch.smi_pending;
4361 events->smi.smm_inside_nmi =
4362 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
4363 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
4365 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
4366 | KVM_VCPUEVENT_VALID_SHADOW
4367 | KVM_VCPUEVENT_VALID_SMM);
4368 if (vcpu->kvm->arch.exception_payload_enabled)
4369 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
4371 memset(&events->reserved, 0, sizeof(events->reserved));
4374 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
4376 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
4377 struct kvm_vcpu_events *events)
4379 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
4380 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
4381 | KVM_VCPUEVENT_VALID_SHADOW
4382 | KVM_VCPUEVENT_VALID_SMM
4383 | KVM_VCPUEVENT_VALID_PAYLOAD))
4386 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
4387 if (!vcpu->kvm->arch.exception_payload_enabled)
4389 if (events->exception.pending)
4390 events->exception.injected = 0;
4392 events->exception_has_payload = 0;
4394 events->exception.pending = 0;
4395 events->exception_has_payload = 0;
4398 if ((events->exception.injected || events->exception.pending) &&
4399 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
4402 /* INITs are latched while in SMM */
4403 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
4404 (events->smi.smm || events->smi.pending) &&
4405 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
4409 vcpu->arch.exception.injected = events->exception.injected;
4410 vcpu->arch.exception.pending = events->exception.pending;
4411 vcpu->arch.exception.nr = events->exception.nr;
4412 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
4413 vcpu->arch.exception.error_code = events->exception.error_code;
4414 vcpu->arch.exception.has_payload = events->exception_has_payload;
4415 vcpu->arch.exception.payload = events->exception_payload;
4417 vcpu->arch.interrupt.injected = events->interrupt.injected;
4418 vcpu->arch.interrupt.nr = events->interrupt.nr;
4419 vcpu->arch.interrupt.soft = events->interrupt.soft;
4420 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
4421 static_call(kvm_x86_set_interrupt_shadow)(vcpu,
4422 events->interrupt.shadow);
4424 vcpu->arch.nmi_injected = events->nmi.injected;
4425 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
4426 vcpu->arch.nmi_pending = events->nmi.pending;
4427 static_call(kvm_x86_set_nmi_mask)(vcpu, events->nmi.masked);
4429 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
4430 lapic_in_kernel(vcpu))
4431 vcpu->arch.apic->sipi_vector = events->sipi_vector;
4433 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
4434 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
4435 if (events->smi.smm)
4436 vcpu->arch.hflags |= HF_SMM_MASK;
4438 vcpu->arch.hflags &= ~HF_SMM_MASK;
4439 kvm_smm_changed(vcpu);
4442 vcpu->arch.smi_pending = events->smi.pending;
4444 if (events->smi.smm) {
4445 if (events->smi.smm_inside_nmi)
4446 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
4448 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
4451 if (lapic_in_kernel(vcpu)) {
4452 if (events->smi.latched_init)
4453 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4455 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
4459 kvm_make_request(KVM_REQ_EVENT, vcpu);
4464 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
4465 struct kvm_debugregs *dbgregs)
4469 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
4470 kvm_get_dr(vcpu, 6, &val);
4472 dbgregs->dr7 = vcpu->arch.dr7;
4474 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
4477 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
4478 struct kvm_debugregs *dbgregs)
4483 if (!kvm_dr6_valid(dbgregs->dr6))
4485 if (!kvm_dr7_valid(dbgregs->dr7))
4488 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
4489 kvm_update_dr0123(vcpu);
4490 vcpu->arch.dr6 = dbgregs->dr6;
4491 vcpu->arch.dr7 = dbgregs->dr7;
4492 kvm_update_dr7(vcpu);
4497 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
4499 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
4501 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4502 u64 xstate_bv = xsave->header.xfeatures;
4506 * Copy legacy XSAVE area, to avoid complications with CPUID
4507 * leaves 0 and 1 in the loop below.
4509 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
4512 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
4513 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
4516 * Copy each region from the possibly compacted offset to the
4517 * non-compacted offset.
4519 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4521 u64 xfeature_mask = valid & -valid;
4522 int xfeature_nr = fls64(xfeature_mask) - 1;
4523 void *src = get_xsave_addr(xsave, xfeature_nr);
4526 u32 size, offset, ecx, edx;
4527 cpuid_count(XSTATE_CPUID, xfeature_nr,
4528 &size, &offset, &ecx, &edx);
4529 if (xfeature_nr == XFEATURE_PKRU)
4530 memcpy(dest + offset, &vcpu->arch.pkru,
4531 sizeof(vcpu->arch.pkru));
4533 memcpy(dest + offset, src, size);
4537 valid -= xfeature_mask;
4541 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4543 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4544 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4548 * Copy legacy XSAVE area, to avoid complications with CPUID
4549 * leaves 0 and 1 in the loop below.
4551 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4553 /* Set XSTATE_BV and possibly XCOMP_BV. */
4554 xsave->header.xfeatures = xstate_bv;
4555 if (boot_cpu_has(X86_FEATURE_XSAVES))
4556 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4559 * Copy each region from the non-compacted offset to the
4560 * possibly compacted offset.
4562 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4564 u64 xfeature_mask = valid & -valid;
4565 int xfeature_nr = fls64(xfeature_mask) - 1;
4566 void *dest = get_xsave_addr(xsave, xfeature_nr);
4569 u32 size, offset, ecx, edx;
4570 cpuid_count(XSTATE_CPUID, xfeature_nr,
4571 &size, &offset, &ecx, &edx);
4572 if (xfeature_nr == XFEATURE_PKRU)
4573 memcpy(&vcpu->arch.pkru, src + offset,
4574 sizeof(vcpu->arch.pkru));
4576 memcpy(dest, src + offset, size);
4579 valid -= xfeature_mask;
4583 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4584 struct kvm_xsave *guest_xsave)
4586 if (!vcpu->arch.guest_fpu)
4589 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4590 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4591 fill_xsave((u8 *) guest_xsave->region, vcpu);
4593 memcpy(guest_xsave->region,
4594 &vcpu->arch.guest_fpu->state.fxsave,
4595 sizeof(struct fxregs_state));
4596 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4597 XFEATURE_MASK_FPSSE;
4601 #define XSAVE_MXCSR_OFFSET 24
4603 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4604 struct kvm_xsave *guest_xsave)
4609 if (!vcpu->arch.guest_fpu)
4612 xstate_bv = *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4613 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4615 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4617 * Here we allow setting states that are not present in
4618 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4619 * with old userspace.
4621 if (xstate_bv & ~supported_xcr0 || mxcsr & ~mxcsr_feature_mask)
4623 load_xsave(vcpu, (u8 *)guest_xsave->region);
4625 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4626 mxcsr & ~mxcsr_feature_mask)
4628 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4629 guest_xsave->region, sizeof(struct fxregs_state));
4634 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4635 struct kvm_xcrs *guest_xcrs)
4637 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4638 guest_xcrs->nr_xcrs = 0;
4642 guest_xcrs->nr_xcrs = 1;
4643 guest_xcrs->flags = 0;
4644 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4645 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4648 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4649 struct kvm_xcrs *guest_xcrs)
4653 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4656 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4659 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4660 /* Only support XCR0 currently */
4661 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4662 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4663 guest_xcrs->xcrs[i].value);
4672 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4673 * stopped by the hypervisor. This function will be called from the host only.
4674 * EINVAL is returned when the host attempts to set the flag for a guest that
4675 * does not support pv clocks.
4677 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4679 if (!vcpu->arch.pv_time_enabled)
4681 vcpu->arch.pvclock_set_guest_stopped_request = true;
4682 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4686 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4687 struct kvm_enable_cap *cap)
4690 uint16_t vmcs_version;
4691 void __user *user_ptr;
4697 case KVM_CAP_HYPERV_SYNIC2:
4702 case KVM_CAP_HYPERV_SYNIC:
4703 if (!irqchip_in_kernel(vcpu->kvm))
4705 return kvm_hv_activate_synic(vcpu, cap->cap ==
4706 KVM_CAP_HYPERV_SYNIC2);
4707 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4708 if (!kvm_x86_ops.nested_ops->enable_evmcs)
4710 r = kvm_x86_ops.nested_ops->enable_evmcs(vcpu, &vmcs_version);
4712 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4713 if (copy_to_user(user_ptr, &vmcs_version,
4714 sizeof(vmcs_version)))
4718 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4719 if (!kvm_x86_ops.enable_direct_tlbflush)
4722 return static_call(kvm_x86_enable_direct_tlbflush)(vcpu);
4724 case KVM_CAP_ENFORCE_PV_FEATURE_CPUID:
4725 vcpu->arch.pv_cpuid.enforce = cap->args[0];
4726 if (vcpu->arch.pv_cpuid.enforce)
4727 kvm_update_pv_runtime(vcpu);
4735 long kvm_arch_vcpu_ioctl(struct file *filp,
4736 unsigned int ioctl, unsigned long arg)
4738 struct kvm_vcpu *vcpu = filp->private_data;
4739 void __user *argp = (void __user *)arg;
4742 struct kvm_lapic_state *lapic;
4743 struct kvm_xsave *xsave;
4744 struct kvm_xcrs *xcrs;
4752 case KVM_GET_LAPIC: {
4754 if (!lapic_in_kernel(vcpu))
4756 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4757 GFP_KERNEL_ACCOUNT);
4762 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4766 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4771 case KVM_SET_LAPIC: {
4773 if (!lapic_in_kernel(vcpu))
4775 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4776 if (IS_ERR(u.lapic)) {
4777 r = PTR_ERR(u.lapic);
4781 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4784 case KVM_INTERRUPT: {
4785 struct kvm_interrupt irq;
4788 if (copy_from_user(&irq, argp, sizeof(irq)))
4790 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4794 r = kvm_vcpu_ioctl_nmi(vcpu);
4798 r = kvm_vcpu_ioctl_smi(vcpu);
4801 case KVM_SET_CPUID: {
4802 struct kvm_cpuid __user *cpuid_arg = argp;
4803 struct kvm_cpuid cpuid;
4806 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4808 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4811 case KVM_SET_CPUID2: {
4812 struct kvm_cpuid2 __user *cpuid_arg = argp;
4813 struct kvm_cpuid2 cpuid;
4816 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4818 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4819 cpuid_arg->entries);
4822 case KVM_GET_CPUID2: {
4823 struct kvm_cpuid2 __user *cpuid_arg = argp;
4824 struct kvm_cpuid2 cpuid;
4827 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4829 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4830 cpuid_arg->entries);
4834 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4839 case KVM_GET_MSRS: {
4840 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4841 r = msr_io(vcpu, argp, do_get_msr, 1);
4842 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4845 case KVM_SET_MSRS: {
4846 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4847 r = msr_io(vcpu, argp, do_set_msr, 0);
4848 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4851 case KVM_TPR_ACCESS_REPORTING: {
4852 struct kvm_tpr_access_ctl tac;
4855 if (copy_from_user(&tac, argp, sizeof(tac)))
4857 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4861 if (copy_to_user(argp, &tac, sizeof(tac)))
4866 case KVM_SET_VAPIC_ADDR: {
4867 struct kvm_vapic_addr va;
4871 if (!lapic_in_kernel(vcpu))
4874 if (copy_from_user(&va, argp, sizeof(va)))
4876 idx = srcu_read_lock(&vcpu->kvm->srcu);
4877 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4878 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4881 case KVM_X86_SETUP_MCE: {
4885 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4887 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4890 case KVM_X86_SET_MCE: {
4891 struct kvm_x86_mce mce;
4894 if (copy_from_user(&mce, argp, sizeof(mce)))
4896 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4899 case KVM_GET_VCPU_EVENTS: {
4900 struct kvm_vcpu_events events;
4902 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4905 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4910 case KVM_SET_VCPU_EVENTS: {
4911 struct kvm_vcpu_events events;
4914 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4917 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4920 case KVM_GET_DEBUGREGS: {
4921 struct kvm_debugregs dbgregs;
4923 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4926 if (copy_to_user(argp, &dbgregs,
4927 sizeof(struct kvm_debugregs)))
4932 case KVM_SET_DEBUGREGS: {
4933 struct kvm_debugregs dbgregs;
4936 if (copy_from_user(&dbgregs, argp,
4937 sizeof(struct kvm_debugregs)))
4940 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4943 case KVM_GET_XSAVE: {
4944 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4949 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4952 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4957 case KVM_SET_XSAVE: {
4958 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4959 if (IS_ERR(u.xsave)) {
4960 r = PTR_ERR(u.xsave);
4964 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4967 case KVM_GET_XCRS: {
4968 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4973 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4976 if (copy_to_user(argp, u.xcrs,
4977 sizeof(struct kvm_xcrs)))
4982 case KVM_SET_XCRS: {
4983 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4984 if (IS_ERR(u.xcrs)) {
4985 r = PTR_ERR(u.xcrs);
4989 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4992 case KVM_SET_TSC_KHZ: {
4996 user_tsc_khz = (u32)arg;
4998 if (kvm_has_tsc_control &&
4999 user_tsc_khz >= kvm_max_guest_tsc_khz)
5002 if (user_tsc_khz == 0)
5003 user_tsc_khz = tsc_khz;
5005 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
5010 case KVM_GET_TSC_KHZ: {
5011 r = vcpu->arch.virtual_tsc_khz;
5014 case KVM_KVMCLOCK_CTRL: {
5015 r = kvm_set_guest_paused(vcpu);
5018 case KVM_ENABLE_CAP: {
5019 struct kvm_enable_cap cap;
5022 if (copy_from_user(&cap, argp, sizeof(cap)))
5024 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
5027 case KVM_GET_NESTED_STATE: {
5028 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5032 if (!kvm_x86_ops.nested_ops->get_state)
5035 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
5037 if (get_user(user_data_size, &user_kvm_nested_state->size))
5040 r = kvm_x86_ops.nested_ops->get_state(vcpu, user_kvm_nested_state,
5045 if (r > user_data_size) {
5046 if (put_user(r, &user_kvm_nested_state->size))
5056 case KVM_SET_NESTED_STATE: {
5057 struct kvm_nested_state __user *user_kvm_nested_state = argp;
5058 struct kvm_nested_state kvm_state;
5062 if (!kvm_x86_ops.nested_ops->set_state)
5066 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
5070 if (kvm_state.size < sizeof(kvm_state))
5073 if (kvm_state.flags &
5074 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
5075 | KVM_STATE_NESTED_EVMCS | KVM_STATE_NESTED_MTF_PENDING
5076 | KVM_STATE_NESTED_GIF_SET))
5079 /* nested_run_pending implies guest_mode. */
5080 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
5081 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
5084 idx = srcu_read_lock(&vcpu->kvm->srcu);
5085 r = kvm_x86_ops.nested_ops->set_state(vcpu, user_kvm_nested_state, &kvm_state);
5086 srcu_read_unlock(&vcpu->kvm->srcu, idx);
5089 case KVM_GET_SUPPORTED_HV_CPUID:
5090 r = kvm_ioctl_get_supported_hv_cpuid(vcpu, argp);
5092 #ifdef CONFIG_KVM_XEN
5093 case KVM_XEN_VCPU_GET_ATTR: {
5094 struct kvm_xen_vcpu_attr xva;
5097 if (copy_from_user(&xva, argp, sizeof(xva)))
5099 r = kvm_xen_vcpu_get_attr(vcpu, &xva);
5100 if (!r && copy_to_user(argp, &xva, sizeof(xva)))
5104 case KVM_XEN_VCPU_SET_ATTR: {
5105 struct kvm_xen_vcpu_attr xva;
5108 if (copy_from_user(&xva, argp, sizeof(xva)))
5110 r = kvm_xen_vcpu_set_attr(vcpu, &xva);
5124 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
5126 return VM_FAULT_SIGBUS;
5129 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
5133 if (addr > (unsigned int)(-3 * PAGE_SIZE))
5135 ret = static_call(kvm_x86_set_tss_addr)(kvm, addr);
5139 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
5142 return static_call(kvm_x86_set_identity_map_addr)(kvm, ident_addr);
5145 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
5146 unsigned long kvm_nr_mmu_pages)
5148 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
5151 mutex_lock(&kvm->slots_lock);
5153 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
5154 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
5156 mutex_unlock(&kvm->slots_lock);
5160 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
5162 return kvm->arch.n_max_mmu_pages;
5165 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5167 struct kvm_pic *pic = kvm->arch.vpic;
5171 switch (chip->chip_id) {
5172 case KVM_IRQCHIP_PIC_MASTER:
5173 memcpy(&chip->chip.pic, &pic->pics[0],
5174 sizeof(struct kvm_pic_state));
5176 case KVM_IRQCHIP_PIC_SLAVE:
5177 memcpy(&chip->chip.pic, &pic->pics[1],
5178 sizeof(struct kvm_pic_state));
5180 case KVM_IRQCHIP_IOAPIC:
5181 kvm_get_ioapic(kvm, &chip->chip.ioapic);
5190 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
5192 struct kvm_pic *pic = kvm->arch.vpic;
5196 switch (chip->chip_id) {
5197 case KVM_IRQCHIP_PIC_MASTER:
5198 spin_lock(&pic->lock);
5199 memcpy(&pic->pics[0], &chip->chip.pic,
5200 sizeof(struct kvm_pic_state));
5201 spin_unlock(&pic->lock);
5203 case KVM_IRQCHIP_PIC_SLAVE:
5204 spin_lock(&pic->lock);
5205 memcpy(&pic->pics[1], &chip->chip.pic,
5206 sizeof(struct kvm_pic_state));
5207 spin_unlock(&pic->lock);
5209 case KVM_IRQCHIP_IOAPIC:
5210 kvm_set_ioapic(kvm, &chip->chip.ioapic);
5216 kvm_pic_update_irq(pic);
5220 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5222 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
5224 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
5226 mutex_lock(&kps->lock);
5227 memcpy(ps, &kps->channels, sizeof(*ps));
5228 mutex_unlock(&kps->lock);
5232 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
5235 struct kvm_pit *pit = kvm->arch.vpit;
5237 mutex_lock(&pit->pit_state.lock);
5238 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
5239 for (i = 0; i < 3; i++)
5240 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
5241 mutex_unlock(&pit->pit_state.lock);
5245 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5247 mutex_lock(&kvm->arch.vpit->pit_state.lock);
5248 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
5249 sizeof(ps->channels));
5250 ps->flags = kvm->arch.vpit->pit_state.flags;
5251 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
5252 memset(&ps->reserved, 0, sizeof(ps->reserved));
5256 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
5260 u32 prev_legacy, cur_legacy;
5261 struct kvm_pit *pit = kvm->arch.vpit;
5263 mutex_lock(&pit->pit_state.lock);
5264 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
5265 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
5266 if (!prev_legacy && cur_legacy)
5268 memcpy(&pit->pit_state.channels, &ps->channels,
5269 sizeof(pit->pit_state.channels));
5270 pit->pit_state.flags = ps->flags;
5271 for (i = 0; i < 3; i++)
5272 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
5274 mutex_unlock(&pit->pit_state.lock);
5278 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
5279 struct kvm_reinject_control *control)
5281 struct kvm_pit *pit = kvm->arch.vpit;
5283 /* pit->pit_state.lock was overloaded to prevent userspace from getting
5284 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
5285 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
5287 mutex_lock(&pit->pit_state.lock);
5288 kvm_pit_set_reinject(pit, control->pit_reinject);
5289 mutex_unlock(&pit->pit_state.lock);
5294 void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
5298 * Flush all CPUs' dirty log buffers to the dirty_bitmap. Called
5299 * before reporting dirty_bitmap to userspace. KVM flushes the buffers
5300 * on all VM-Exits, thus we only need to kick running vCPUs to force a
5303 struct kvm_vcpu *vcpu;
5306 kvm_for_each_vcpu(i, vcpu, kvm)
5307 kvm_vcpu_kick(vcpu);
5310 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
5313 if (!irqchip_in_kernel(kvm))
5316 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
5317 irq_event->irq, irq_event->level,
5322 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
5323 struct kvm_enable_cap *cap)
5331 case KVM_CAP_DISABLE_QUIRKS:
5332 kvm->arch.disabled_quirks = cap->args[0];
5335 case KVM_CAP_SPLIT_IRQCHIP: {
5336 mutex_lock(&kvm->lock);
5338 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
5339 goto split_irqchip_unlock;
5341 if (irqchip_in_kernel(kvm))
5342 goto split_irqchip_unlock;
5343 if (kvm->created_vcpus)
5344 goto split_irqchip_unlock;
5345 r = kvm_setup_empty_irq_routing(kvm);
5347 goto split_irqchip_unlock;
5348 /* Pairs with irqchip_in_kernel. */
5350 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
5351 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
5353 split_irqchip_unlock:
5354 mutex_unlock(&kvm->lock);
5357 case KVM_CAP_X2APIC_API:
5359 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
5362 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
5363 kvm->arch.x2apic_format = true;
5364 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
5365 kvm->arch.x2apic_broadcast_quirk_disabled = true;
5369 case KVM_CAP_X86_DISABLE_EXITS:
5371 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
5374 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
5375 kvm_can_mwait_in_guest())
5376 kvm->arch.mwait_in_guest = true;
5377 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
5378 kvm->arch.hlt_in_guest = true;
5379 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
5380 kvm->arch.pause_in_guest = true;
5381 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
5382 kvm->arch.cstate_in_guest = true;
5385 case KVM_CAP_MSR_PLATFORM_INFO:
5386 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
5389 case KVM_CAP_EXCEPTION_PAYLOAD:
5390 kvm->arch.exception_payload_enabled = cap->args[0];
5393 case KVM_CAP_X86_USER_SPACE_MSR:
5394 kvm->arch.user_space_msr_mask = cap->args[0];
5397 case KVM_CAP_X86_BUS_LOCK_EXIT:
5399 if (cap->args[0] & ~KVM_BUS_LOCK_DETECTION_VALID_MODE)
5402 if ((cap->args[0] & KVM_BUS_LOCK_DETECTION_OFF) &&
5403 (cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT))
5406 if (kvm_has_bus_lock_exit &&
5407 cap->args[0] & KVM_BUS_LOCK_DETECTION_EXIT)
5408 kvm->arch.bus_lock_detection_enabled = true;
5411 #ifdef CONFIG_X86_SGX_KVM
5412 case KVM_CAP_SGX_ATTRIBUTE: {
5413 unsigned long allowed_attributes = 0;
5415 r = sgx_set_attribute(&allowed_attributes, cap->args[0]);
5419 /* KVM only supports the PROVISIONKEY privileged attribute. */
5420 if ((allowed_attributes & SGX_ATTR_PROVISIONKEY) &&
5421 !(allowed_attributes & ~SGX_ATTR_PROVISIONKEY))
5422 kvm->arch.sgx_provisioning_allowed = true;
5428 case KVM_CAP_VM_COPY_ENC_CONTEXT_FROM:
5430 if (kvm_x86_ops.vm_copy_enc_context_from)
5431 r = kvm_x86_ops.vm_copy_enc_context_from(kvm, cap->args[0]);
5440 static struct kvm_x86_msr_filter *kvm_alloc_msr_filter(bool default_allow)
5442 struct kvm_x86_msr_filter *msr_filter;
5444 msr_filter = kzalloc(sizeof(*msr_filter), GFP_KERNEL_ACCOUNT);
5448 msr_filter->default_allow = default_allow;
5452 static void kvm_free_msr_filter(struct kvm_x86_msr_filter *msr_filter)
5459 for (i = 0; i < msr_filter->count; i++)
5460 kfree(msr_filter->ranges[i].bitmap);
5465 static int kvm_add_msr_filter(struct kvm_x86_msr_filter *msr_filter,
5466 struct kvm_msr_filter_range *user_range)
5468 struct msr_bitmap_range range;
5469 unsigned long *bitmap = NULL;
5473 if (!user_range->nmsrs)
5476 bitmap_size = BITS_TO_LONGS(user_range->nmsrs) * sizeof(long);
5477 if (!bitmap_size || bitmap_size > KVM_MSR_FILTER_MAX_BITMAP_SIZE)
5480 bitmap = memdup_user((__user u8*)user_range->bitmap, bitmap_size);
5482 return PTR_ERR(bitmap);
5484 range = (struct msr_bitmap_range) {
5485 .flags = user_range->flags,
5486 .base = user_range->base,
5487 .nmsrs = user_range->nmsrs,
5491 if (range.flags & ~(KVM_MSR_FILTER_READ | KVM_MSR_FILTER_WRITE)) {
5501 /* Everything ok, add this range identifier. */
5502 msr_filter->ranges[msr_filter->count] = range;
5503 msr_filter->count++;
5511 static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
5513 struct kvm_msr_filter __user *user_msr_filter = argp;
5514 struct kvm_x86_msr_filter *new_filter, *old_filter;
5515 struct kvm_msr_filter filter;
5521 if (copy_from_user(&filter, user_msr_filter, sizeof(filter)))
5524 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++)
5525 empty &= !filter.ranges[i].nmsrs;
5527 default_allow = !(filter.flags & KVM_MSR_FILTER_DEFAULT_DENY);
5528 if (empty && !default_allow)
5531 new_filter = kvm_alloc_msr_filter(default_allow);
5535 for (i = 0; i < ARRAY_SIZE(filter.ranges); i++) {
5536 r = kvm_add_msr_filter(new_filter, &filter.ranges[i]);
5538 kvm_free_msr_filter(new_filter);
5543 mutex_lock(&kvm->lock);
5545 /* The per-VM filter is protected by kvm->lock... */
5546 old_filter = srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1);
5548 rcu_assign_pointer(kvm->arch.msr_filter, new_filter);
5549 synchronize_srcu(&kvm->srcu);
5551 kvm_free_msr_filter(old_filter);
5553 kvm_make_all_cpus_request(kvm, KVM_REQ_MSR_FILTER_CHANGED);
5554 mutex_unlock(&kvm->lock);
5559 long kvm_arch_vm_ioctl(struct file *filp,
5560 unsigned int ioctl, unsigned long arg)
5562 struct kvm *kvm = filp->private_data;
5563 void __user *argp = (void __user *)arg;
5566 * This union makes it completely explicit to gcc-3.x
5567 * that these two variables' stack usage should be
5568 * combined, not added together.
5571 struct kvm_pit_state ps;
5572 struct kvm_pit_state2 ps2;
5573 struct kvm_pit_config pit_config;
5577 case KVM_SET_TSS_ADDR:
5578 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
5580 case KVM_SET_IDENTITY_MAP_ADDR: {
5583 mutex_lock(&kvm->lock);
5585 if (kvm->created_vcpus)
5586 goto set_identity_unlock;
5588 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
5589 goto set_identity_unlock;
5590 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
5591 set_identity_unlock:
5592 mutex_unlock(&kvm->lock);
5595 case KVM_SET_NR_MMU_PAGES:
5596 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
5598 case KVM_GET_NR_MMU_PAGES:
5599 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
5601 case KVM_CREATE_IRQCHIP: {
5602 mutex_lock(&kvm->lock);
5605 if (irqchip_in_kernel(kvm))
5606 goto create_irqchip_unlock;
5609 if (kvm->created_vcpus)
5610 goto create_irqchip_unlock;
5612 r = kvm_pic_init(kvm);
5614 goto create_irqchip_unlock;
5616 r = kvm_ioapic_init(kvm);
5618 kvm_pic_destroy(kvm);
5619 goto create_irqchip_unlock;
5622 r = kvm_setup_default_irq_routing(kvm);
5624 kvm_ioapic_destroy(kvm);
5625 kvm_pic_destroy(kvm);
5626 goto create_irqchip_unlock;
5628 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
5630 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
5631 create_irqchip_unlock:
5632 mutex_unlock(&kvm->lock);
5635 case KVM_CREATE_PIT:
5636 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
5638 case KVM_CREATE_PIT2:
5640 if (copy_from_user(&u.pit_config, argp,
5641 sizeof(struct kvm_pit_config)))
5644 mutex_lock(&kvm->lock);
5647 goto create_pit_unlock;
5649 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5653 mutex_unlock(&kvm->lock);
5655 case KVM_GET_IRQCHIP: {
5656 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5657 struct kvm_irqchip *chip;
5659 chip = memdup_user(argp, sizeof(*chip));
5666 if (!irqchip_kernel(kvm))
5667 goto get_irqchip_out;
5668 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5670 goto get_irqchip_out;
5672 if (copy_to_user(argp, chip, sizeof(*chip)))
5673 goto get_irqchip_out;
5679 case KVM_SET_IRQCHIP: {
5680 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5681 struct kvm_irqchip *chip;
5683 chip = memdup_user(argp, sizeof(*chip));
5690 if (!irqchip_kernel(kvm))
5691 goto set_irqchip_out;
5692 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5699 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5702 if (!kvm->arch.vpit)
5704 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5708 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5715 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5717 mutex_lock(&kvm->lock);
5719 if (!kvm->arch.vpit)
5721 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5723 mutex_unlock(&kvm->lock);
5726 case KVM_GET_PIT2: {
5728 if (!kvm->arch.vpit)
5730 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5734 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5739 case KVM_SET_PIT2: {
5741 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5743 mutex_lock(&kvm->lock);
5745 if (!kvm->arch.vpit)
5747 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5749 mutex_unlock(&kvm->lock);
5752 case KVM_REINJECT_CONTROL: {
5753 struct kvm_reinject_control control;
5755 if (copy_from_user(&control, argp, sizeof(control)))
5758 if (!kvm->arch.vpit)
5760 r = kvm_vm_ioctl_reinject(kvm, &control);
5763 case KVM_SET_BOOT_CPU_ID:
5765 mutex_lock(&kvm->lock);
5766 if (kvm->created_vcpus)
5769 kvm->arch.bsp_vcpu_id = arg;
5770 mutex_unlock(&kvm->lock);
5772 #ifdef CONFIG_KVM_XEN
5773 case KVM_XEN_HVM_CONFIG: {
5774 struct kvm_xen_hvm_config xhc;
5776 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5778 r = kvm_xen_hvm_config(kvm, &xhc);
5781 case KVM_XEN_HVM_GET_ATTR: {
5782 struct kvm_xen_hvm_attr xha;
5785 if (copy_from_user(&xha, argp, sizeof(xha)))
5787 r = kvm_xen_hvm_get_attr(kvm, &xha);
5788 if (!r && copy_to_user(argp, &xha, sizeof(xha)))
5792 case KVM_XEN_HVM_SET_ATTR: {
5793 struct kvm_xen_hvm_attr xha;
5796 if (copy_from_user(&xha, argp, sizeof(xha)))
5798 r = kvm_xen_hvm_set_attr(kvm, &xha);
5802 case KVM_SET_CLOCK: {
5803 struct kvm_arch *ka = &kvm->arch;
5804 struct kvm_clock_data user_ns;
5808 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5817 * TODO: userspace has to take care of races with VCPU_RUN, so
5818 * kvm_gen_update_masterclock() can be cut down to locked
5819 * pvclock_update_vm_gtod_copy().
5821 kvm_gen_update_masterclock(kvm);
5824 * This pairs with kvm_guest_time_update(): when masterclock is
5825 * in use, we use master_kernel_ns + kvmclock_offset to set
5826 * unsigned 'system_time' so if we use get_kvmclock_ns() (which
5827 * is slightly ahead) here we risk going negative on unsigned
5828 * 'system_time' when 'user_ns.clock' is very small.
5830 spin_lock_irq(&ka->pvclock_gtod_sync_lock);
5831 if (kvm->arch.use_master_clock)
5832 now_ns = ka->master_kernel_ns;
5834 now_ns = get_kvmclock_base_ns();
5835 ka->kvmclock_offset = user_ns.clock - now_ns;
5836 spin_unlock_irq(&ka->pvclock_gtod_sync_lock);
5838 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5841 case KVM_GET_CLOCK: {
5842 struct kvm_clock_data user_ns;
5845 now_ns = get_kvmclock_ns(kvm);
5846 user_ns.clock = now_ns;
5847 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5848 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5851 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5856 case KVM_MEMORY_ENCRYPT_OP: {
5858 if (kvm_x86_ops.mem_enc_op)
5859 r = static_call(kvm_x86_mem_enc_op)(kvm, argp);
5862 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5863 struct kvm_enc_region region;
5866 if (copy_from_user(®ion, argp, sizeof(region)))
5870 if (kvm_x86_ops.mem_enc_reg_region)
5871 r = static_call(kvm_x86_mem_enc_reg_region)(kvm, ®ion);
5874 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5875 struct kvm_enc_region region;
5878 if (copy_from_user(®ion, argp, sizeof(region)))
5882 if (kvm_x86_ops.mem_enc_unreg_region)
5883 r = static_call(kvm_x86_mem_enc_unreg_region)(kvm, ®ion);
5886 case KVM_HYPERV_EVENTFD: {
5887 struct kvm_hyperv_eventfd hvevfd;
5890 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5892 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5895 case KVM_SET_PMU_EVENT_FILTER:
5896 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5898 case KVM_X86_SET_MSR_FILTER:
5899 r = kvm_vm_ioctl_set_msr_filter(kvm, argp);
5908 static void kvm_init_msr_list(void)
5910 struct x86_pmu_capability x86_pmu;
5914 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5915 "Please update the fixed PMCs in msrs_to_saved_all[]");
5917 perf_get_x86_pmu_capability(&x86_pmu);
5919 num_msrs_to_save = 0;
5920 num_emulated_msrs = 0;
5921 num_msr_based_features = 0;
5923 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5924 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5928 * Even MSRs that are valid in the host may not be exposed
5929 * to the guests in some cases.
5931 switch (msrs_to_save_all[i]) {
5932 case MSR_IA32_BNDCFGS:
5933 if (!kvm_mpx_supported())
5937 if (!kvm_cpu_cap_has(X86_FEATURE_RDTSCP))
5940 case MSR_IA32_UMWAIT_CONTROL:
5941 if (!kvm_cpu_cap_has(X86_FEATURE_WAITPKG))
5944 case MSR_IA32_RTIT_CTL:
5945 case MSR_IA32_RTIT_STATUS:
5946 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT))
5949 case MSR_IA32_RTIT_CR3_MATCH:
5950 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5951 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5954 case MSR_IA32_RTIT_OUTPUT_BASE:
5955 case MSR_IA32_RTIT_OUTPUT_MASK:
5956 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5957 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5958 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5961 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B:
5962 if (!kvm_cpu_cap_has(X86_FEATURE_INTEL_PT) ||
5963 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5964 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5967 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5968 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5969 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5972 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5973 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5974 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5981 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5984 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5985 if (!static_call(kvm_x86_has_emulated_msr)(NULL, emulated_msrs_all[i]))
5988 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5991 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5992 struct kvm_msr_entry msr;
5994 msr.index = msr_based_features_all[i];
5995 if (kvm_get_msr_feature(&msr))
5998 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
6002 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
6010 if (!(lapic_in_kernel(vcpu) &&
6011 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
6012 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
6023 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
6030 if (!(lapic_in_kernel(vcpu) &&
6031 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
6033 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
6035 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
6045 static void kvm_set_segment(struct kvm_vcpu *vcpu,
6046 struct kvm_segment *var, int seg)
6048 static_call(kvm_x86_set_segment)(vcpu, var, seg);
6051 void kvm_get_segment(struct kvm_vcpu *vcpu,
6052 struct kvm_segment *var, int seg)
6054 static_call(kvm_x86_get_segment)(vcpu, var, seg);
6057 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
6058 struct x86_exception *exception)
6062 BUG_ON(!mmu_is_nested(vcpu));
6064 /* NPT walks are always user-walks */
6065 access |= PFERR_USER_MASK;
6066 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
6071 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
6072 struct x86_exception *exception)
6074 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6075 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6077 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_read);
6079 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
6080 struct x86_exception *exception)
6082 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6083 access |= PFERR_FETCH_MASK;
6084 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6087 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
6088 struct x86_exception *exception)
6090 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6091 access |= PFERR_WRITE_MASK;
6092 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6094 EXPORT_SYMBOL_GPL(kvm_mmu_gva_to_gpa_write);
6096 /* uses this to access any guest's mapped memory without checking CPL */
6097 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
6098 struct x86_exception *exception)
6100 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
6103 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6104 struct kvm_vcpu *vcpu, u32 access,
6105 struct x86_exception *exception)
6108 int r = X86EMUL_CONTINUE;
6111 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
6113 unsigned offset = addr & (PAGE_SIZE-1);
6114 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
6117 if (gpa == UNMAPPED_GVA)
6118 return X86EMUL_PROPAGATE_FAULT;
6119 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
6122 r = X86EMUL_IO_NEEDED;
6134 /* used for instruction fetching */
6135 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
6136 gva_t addr, void *val, unsigned int bytes,
6137 struct x86_exception *exception)
6139 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6140 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6144 /* Inline kvm_read_guest_virt_helper for speed. */
6145 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
6147 if (unlikely(gpa == UNMAPPED_GVA))
6148 return X86EMUL_PROPAGATE_FAULT;
6150 offset = addr & (PAGE_SIZE-1);
6151 if (WARN_ON(offset + bytes > PAGE_SIZE))
6152 bytes = (unsigned)PAGE_SIZE - offset;
6153 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
6155 if (unlikely(ret < 0))
6156 return X86EMUL_IO_NEEDED;
6158 return X86EMUL_CONTINUE;
6161 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
6162 gva_t addr, void *val, unsigned int bytes,
6163 struct x86_exception *exception)
6165 u32 access = (static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0;
6168 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
6169 * is returned, but our callers are not ready for that and they blindly
6170 * call kvm_inject_page_fault. Ensure that they at least do not leak
6171 * uninitialized kernel stack memory into cr2 and error code.
6173 memset(exception, 0, sizeof(*exception));
6174 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
6177 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
6179 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
6180 gva_t addr, void *val, unsigned int bytes,
6181 struct x86_exception *exception, bool system)
6183 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6186 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6187 access |= PFERR_USER_MASK;
6189 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
6192 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
6193 unsigned long addr, void *val, unsigned int bytes)
6195 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6196 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
6198 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
6201 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
6202 struct kvm_vcpu *vcpu, u32 access,
6203 struct x86_exception *exception)
6206 int r = X86EMUL_CONTINUE;
6209 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
6212 unsigned offset = addr & (PAGE_SIZE-1);
6213 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
6216 if (gpa == UNMAPPED_GVA)
6217 return X86EMUL_PROPAGATE_FAULT;
6218 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
6220 r = X86EMUL_IO_NEEDED;
6232 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
6233 unsigned int bytes, struct x86_exception *exception,
6236 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6237 u32 access = PFERR_WRITE_MASK;
6239 if (!system && static_call(kvm_x86_get_cpl)(vcpu) == 3)
6240 access |= PFERR_USER_MASK;
6242 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6246 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
6247 unsigned int bytes, struct x86_exception *exception)
6249 /* kvm_write_guest_virt_system can pull in tons of pages. */
6250 vcpu->arch.l1tf_flush_l1d = true;
6252 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
6253 PFERR_WRITE_MASK, exception);
6255 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
6257 int handle_ud(struct kvm_vcpu *vcpu)
6259 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
6260 int emul_type = EMULTYPE_TRAP_UD;
6261 char sig[5]; /* ud2; .ascii "kvm" */
6262 struct x86_exception e;
6264 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, NULL, 0)))
6267 if (force_emulation_prefix &&
6268 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
6269 sig, sizeof(sig), &e) == 0 &&
6270 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
6271 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
6272 emul_type = EMULTYPE_TRAP_UD_FORCED;
6275 return kvm_emulate_instruction(vcpu, emul_type);
6277 EXPORT_SYMBOL_GPL(handle_ud);
6279 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6280 gpa_t gpa, bool write)
6282 /* For APIC access vmexit */
6283 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6286 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
6287 trace_vcpu_match_mmio(gva, gpa, write, true);
6294 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
6295 gpa_t *gpa, struct x86_exception *exception,
6298 u32 access = ((static_call(kvm_x86_get_cpl)(vcpu) == 3) ? PFERR_USER_MASK : 0)
6299 | (write ? PFERR_WRITE_MASK : 0);
6302 * currently PKRU is only applied to ept enabled guest so
6303 * there is no pkey in EPT page table for L1 guest or EPT
6304 * shadow page table for L2 guest.
6306 if (vcpu_match_mmio_gva(vcpu, gva)
6307 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
6308 vcpu->arch.mmio_access, 0, access)) {
6309 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
6310 (gva & (PAGE_SIZE - 1));
6311 trace_vcpu_match_mmio(gva, *gpa, write, false);
6315 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
6317 if (*gpa == UNMAPPED_GVA)
6320 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
6323 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
6324 const void *val, int bytes)
6328 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
6331 kvm_page_track_write(vcpu, gpa, val, bytes);
6335 struct read_write_emulator_ops {
6336 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
6338 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
6339 void *val, int bytes);
6340 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6341 int bytes, void *val);
6342 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
6343 void *val, int bytes);
6347 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
6349 if (vcpu->mmio_read_completed) {
6350 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
6351 vcpu->mmio_fragments[0].gpa, val);
6352 vcpu->mmio_read_completed = 0;
6359 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6360 void *val, int bytes)
6362 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
6365 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
6366 void *val, int bytes)
6368 return emulator_write_phys(vcpu, gpa, val, bytes);
6371 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
6373 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
6374 return vcpu_mmio_write(vcpu, gpa, bytes, val);
6377 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6378 void *val, int bytes)
6380 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
6381 return X86EMUL_IO_NEEDED;
6384 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
6385 void *val, int bytes)
6387 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
6389 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
6390 return X86EMUL_CONTINUE;
6393 static const struct read_write_emulator_ops read_emultor = {
6394 .read_write_prepare = read_prepare,
6395 .read_write_emulate = read_emulate,
6396 .read_write_mmio = vcpu_mmio_read,
6397 .read_write_exit_mmio = read_exit_mmio,
6400 static const struct read_write_emulator_ops write_emultor = {
6401 .read_write_emulate = write_emulate,
6402 .read_write_mmio = write_mmio,
6403 .read_write_exit_mmio = write_exit_mmio,
6407 static int emulator_read_write_onepage(unsigned long addr, void *val,
6409 struct x86_exception *exception,
6410 struct kvm_vcpu *vcpu,
6411 const struct read_write_emulator_ops *ops)
6415 bool write = ops->write;
6416 struct kvm_mmio_fragment *frag;
6417 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
6420 * If the exit was due to a NPF we may already have a GPA.
6421 * If the GPA is present, use it to avoid the GVA to GPA table walk.
6422 * Note, this cannot be used on string operations since string
6423 * operation using rep will only have the initial GPA from the NPF
6426 if (ctxt->gpa_available && emulator_can_use_gpa(ctxt) &&
6427 (addr & ~PAGE_MASK) == (ctxt->gpa_val & ~PAGE_MASK)) {
6428 gpa = ctxt->gpa_val;
6429 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
6431 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
6433 return X86EMUL_PROPAGATE_FAULT;
6436 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
6437 return X86EMUL_CONTINUE;
6440 * Is this MMIO handled locally?
6442 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
6443 if (handled == bytes)
6444 return X86EMUL_CONTINUE;
6450 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
6451 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
6455 return X86EMUL_CONTINUE;
6458 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
6460 void *val, unsigned int bytes,
6461 struct x86_exception *exception,
6462 const struct read_write_emulator_ops *ops)
6464 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6468 if (ops->read_write_prepare &&
6469 ops->read_write_prepare(vcpu, val, bytes))
6470 return X86EMUL_CONTINUE;
6472 vcpu->mmio_nr_fragments = 0;
6474 /* Crossing a page boundary? */
6475 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
6478 now = -addr & ~PAGE_MASK;
6479 rc = emulator_read_write_onepage(addr, val, now, exception,
6482 if (rc != X86EMUL_CONTINUE)
6485 if (ctxt->mode != X86EMUL_MODE_PROT64)
6491 rc = emulator_read_write_onepage(addr, val, bytes, exception,
6493 if (rc != X86EMUL_CONTINUE)
6496 if (!vcpu->mmio_nr_fragments)
6499 gpa = vcpu->mmio_fragments[0].gpa;
6501 vcpu->mmio_needed = 1;
6502 vcpu->mmio_cur_fragment = 0;
6504 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
6505 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
6506 vcpu->run->exit_reason = KVM_EXIT_MMIO;
6507 vcpu->run->mmio.phys_addr = gpa;
6509 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
6512 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
6516 struct x86_exception *exception)
6518 return emulator_read_write(ctxt, addr, val, bytes,
6519 exception, &read_emultor);
6522 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
6526 struct x86_exception *exception)
6528 return emulator_read_write(ctxt, addr, (void *)val, bytes,
6529 exception, &write_emultor);
6532 #define CMPXCHG_TYPE(t, ptr, old, new) \
6533 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
6535 #ifdef CONFIG_X86_64
6536 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
6538 # define CMPXCHG64(ptr, old, new) \
6539 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
6542 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
6547 struct x86_exception *exception)
6549 struct kvm_host_map map;
6550 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6556 /* guests cmpxchg8b have to be emulated atomically */
6557 if (bytes > 8 || (bytes & (bytes - 1)))
6560 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
6562 if (gpa == UNMAPPED_GVA ||
6563 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
6567 * Emulate the atomic as a straight write to avoid #AC if SLD is
6568 * enabled in the host and the access splits a cache line.
6570 if (boot_cpu_has(X86_FEATURE_SPLIT_LOCK_DETECT))
6571 page_line_mask = ~(cache_line_size() - 1);
6573 page_line_mask = PAGE_MASK;
6575 if (((gpa + bytes - 1) & page_line_mask) != (gpa & page_line_mask))
6578 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
6581 kaddr = map.hva + offset_in_page(gpa);
6585 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
6588 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
6591 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
6594 exchanged = CMPXCHG64(kaddr, old, new);
6600 kvm_vcpu_unmap(vcpu, &map, true);
6603 return X86EMUL_CMPXCHG_FAILED;
6605 kvm_page_track_write(vcpu, gpa, new, bytes);
6607 return X86EMUL_CONTINUE;
6610 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
6612 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
6615 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
6619 for (i = 0; i < vcpu->arch.pio.count; i++) {
6620 if (vcpu->arch.pio.in)
6621 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
6622 vcpu->arch.pio.size, pd);
6624 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
6625 vcpu->arch.pio.port, vcpu->arch.pio.size,
6629 pd += vcpu->arch.pio.size;
6634 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
6635 unsigned short port, void *val,
6636 unsigned int count, bool in)
6638 vcpu->arch.pio.port = port;
6639 vcpu->arch.pio.in = in;
6640 vcpu->arch.pio.count = count;
6641 vcpu->arch.pio.size = size;
6643 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
6644 vcpu->arch.pio.count = 0;
6648 vcpu->run->exit_reason = KVM_EXIT_IO;
6649 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
6650 vcpu->run->io.size = size;
6651 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
6652 vcpu->run->io.count = count;
6653 vcpu->run->io.port = port;
6658 static int emulator_pio_in(struct kvm_vcpu *vcpu, int size,
6659 unsigned short port, void *val, unsigned int count)
6663 if (vcpu->arch.pio.count)
6666 memset(vcpu->arch.pio_data, 0, size * count);
6668 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
6671 memcpy(val, vcpu->arch.pio_data, size * count);
6672 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
6673 vcpu->arch.pio.count = 0;
6680 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
6681 int size, unsigned short port, void *val,
6684 return emulator_pio_in(emul_to_vcpu(ctxt), size, port, val, count);
6688 static int emulator_pio_out(struct kvm_vcpu *vcpu, int size,
6689 unsigned short port, const void *val,
6692 memcpy(vcpu->arch.pio_data, val, size * count);
6693 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
6694 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
6697 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
6698 int size, unsigned short port,
6699 const void *val, unsigned int count)
6701 return emulator_pio_out(emul_to_vcpu(ctxt), size, port, val, count);
6704 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
6706 return static_call(kvm_x86_get_segment_base)(vcpu, seg);
6709 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
6711 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
6714 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6716 if (!need_emulate_wbinvd(vcpu))
6717 return X86EMUL_CONTINUE;
6719 if (static_call(kvm_x86_has_wbinvd_exit)()) {
6720 int cpu = get_cpu();
6722 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6723 on_each_cpu_mask(vcpu->arch.wbinvd_dirty_mask,
6724 wbinvd_ipi, NULL, 1);
6726 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6729 return X86EMUL_CONTINUE;
6732 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6734 kvm_emulate_wbinvd_noskip(vcpu);
6735 return kvm_skip_emulated_instruction(vcpu);
6737 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6741 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6743 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6746 static void emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6747 unsigned long *dest)
6749 kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6752 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6753 unsigned long value)
6756 return kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6759 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6761 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6764 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6766 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6767 unsigned long value;
6771 value = kvm_read_cr0(vcpu);
6774 value = vcpu->arch.cr2;
6777 value = kvm_read_cr3(vcpu);
6780 value = kvm_read_cr4(vcpu);
6783 value = kvm_get_cr8(vcpu);
6786 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6793 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6795 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6800 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6803 vcpu->arch.cr2 = val;
6806 res = kvm_set_cr3(vcpu, val);
6809 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6812 res = kvm_set_cr8(vcpu, val);
6815 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6822 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6824 return static_call(kvm_x86_get_cpl)(emul_to_vcpu(ctxt));
6827 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6829 static_call(kvm_x86_get_gdt)(emul_to_vcpu(ctxt), dt);
6832 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6834 static_call(kvm_x86_get_idt)(emul_to_vcpu(ctxt), dt);
6837 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6839 static_call(kvm_x86_set_gdt)(emul_to_vcpu(ctxt), dt);
6842 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6844 static_call(kvm_x86_set_idt)(emul_to_vcpu(ctxt), dt);
6847 static unsigned long emulator_get_cached_segment_base(
6848 struct x86_emulate_ctxt *ctxt, int seg)
6850 return get_segment_base(emul_to_vcpu(ctxt), seg);
6853 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6854 struct desc_struct *desc, u32 *base3,
6857 struct kvm_segment var;
6859 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6860 *selector = var.selector;
6863 memset(desc, 0, sizeof(*desc));
6871 set_desc_limit(desc, var.limit);
6872 set_desc_base(desc, (unsigned long)var.base);
6873 #ifdef CONFIG_X86_64
6875 *base3 = var.base >> 32;
6877 desc->type = var.type;
6879 desc->dpl = var.dpl;
6880 desc->p = var.present;
6881 desc->avl = var.avl;
6889 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6890 struct desc_struct *desc, u32 base3,
6893 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6894 struct kvm_segment var;
6896 var.selector = selector;
6897 var.base = get_desc_base(desc);
6898 #ifdef CONFIG_X86_64
6899 var.base |= ((u64)base3) << 32;
6901 var.limit = get_desc_limit(desc);
6903 var.limit = (var.limit << 12) | 0xfff;
6904 var.type = desc->type;
6905 var.dpl = desc->dpl;
6910 var.avl = desc->avl;
6911 var.present = desc->p;
6912 var.unusable = !var.present;
6915 kvm_set_segment(vcpu, &var, seg);
6919 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6920 u32 msr_index, u64 *pdata)
6922 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6925 r = kvm_get_msr(vcpu, msr_index, pdata);
6927 if (r && kvm_get_msr_user_space(vcpu, msr_index, r)) {
6928 /* Bounce to user space */
6929 return X86EMUL_IO_NEEDED;
6935 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6936 u32 msr_index, u64 data)
6938 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6941 r = kvm_set_msr(vcpu, msr_index, data);
6943 if (r && kvm_set_msr_user_space(vcpu, msr_index, data, r)) {
6944 /* Bounce to user space */
6945 return X86EMUL_IO_NEEDED;
6951 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6953 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6955 return vcpu->arch.smbase;
6958 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6960 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6962 vcpu->arch.smbase = smbase;
6965 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6968 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6971 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6972 u32 pmc, u64 *pdata)
6974 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6977 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6979 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6982 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6983 struct x86_instruction_info *info,
6984 enum x86_intercept_stage stage)
6986 return static_call(kvm_x86_check_intercept)(emul_to_vcpu(ctxt), info, stage,
6990 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6991 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx,
6994 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, exact_only);
6997 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6999 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
7002 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
7004 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
7007 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
7009 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
7012 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
7014 return kvm_register_read(emul_to_vcpu(ctxt), reg);
7017 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
7019 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
7022 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
7024 static_call(kvm_x86_set_nmi_mask)(emul_to_vcpu(ctxt), masked);
7027 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
7029 return emul_to_vcpu(ctxt)->arch.hflags;
7032 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
7034 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
7037 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
7038 const char *smstate)
7040 return static_call(kvm_x86_pre_leave_smm)(emul_to_vcpu(ctxt), smstate);
7043 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
7045 kvm_smm_changed(emul_to_vcpu(ctxt));
7048 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
7050 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
7053 static const struct x86_emulate_ops emulate_ops = {
7054 .read_gpr = emulator_read_gpr,
7055 .write_gpr = emulator_write_gpr,
7056 .read_std = emulator_read_std,
7057 .write_std = emulator_write_std,
7058 .read_phys = kvm_read_guest_phys_system,
7059 .fetch = kvm_fetch_guest_virt,
7060 .read_emulated = emulator_read_emulated,
7061 .write_emulated = emulator_write_emulated,
7062 .cmpxchg_emulated = emulator_cmpxchg_emulated,
7063 .invlpg = emulator_invlpg,
7064 .pio_in_emulated = emulator_pio_in_emulated,
7065 .pio_out_emulated = emulator_pio_out_emulated,
7066 .get_segment = emulator_get_segment,
7067 .set_segment = emulator_set_segment,
7068 .get_cached_segment_base = emulator_get_cached_segment_base,
7069 .get_gdt = emulator_get_gdt,
7070 .get_idt = emulator_get_idt,
7071 .set_gdt = emulator_set_gdt,
7072 .set_idt = emulator_set_idt,
7073 .get_cr = emulator_get_cr,
7074 .set_cr = emulator_set_cr,
7075 .cpl = emulator_get_cpl,
7076 .get_dr = emulator_get_dr,
7077 .set_dr = emulator_set_dr,
7078 .get_smbase = emulator_get_smbase,
7079 .set_smbase = emulator_set_smbase,
7080 .set_msr = emulator_set_msr,
7081 .get_msr = emulator_get_msr,
7082 .check_pmc = emulator_check_pmc,
7083 .read_pmc = emulator_read_pmc,
7084 .halt = emulator_halt,
7085 .wbinvd = emulator_wbinvd,
7086 .fix_hypercall = emulator_fix_hypercall,
7087 .intercept = emulator_intercept,
7088 .get_cpuid = emulator_get_cpuid,
7089 .guest_has_long_mode = emulator_guest_has_long_mode,
7090 .guest_has_movbe = emulator_guest_has_movbe,
7091 .guest_has_fxsr = emulator_guest_has_fxsr,
7092 .set_nmi_mask = emulator_set_nmi_mask,
7093 .get_hflags = emulator_get_hflags,
7094 .set_hflags = emulator_set_hflags,
7095 .pre_leave_smm = emulator_pre_leave_smm,
7096 .post_leave_smm = emulator_post_leave_smm,
7097 .set_xcr = emulator_set_xcr,
7100 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
7102 u32 int_shadow = static_call(kvm_x86_get_interrupt_shadow)(vcpu);
7104 * an sti; sti; sequence only disable interrupts for the first
7105 * instruction. So, if the last instruction, be it emulated or
7106 * not, left the system with the INT_STI flag enabled, it
7107 * means that the last instruction is an sti. We should not
7108 * leave the flag on in this case. The same goes for mov ss
7110 if (int_shadow & mask)
7112 if (unlikely(int_shadow || mask)) {
7113 static_call(kvm_x86_set_interrupt_shadow)(vcpu, mask);
7115 kvm_make_request(KVM_REQ_EVENT, vcpu);
7119 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
7121 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7122 if (ctxt->exception.vector == PF_VECTOR)
7123 return kvm_inject_emulated_page_fault(vcpu, &ctxt->exception);
7125 if (ctxt->exception.error_code_valid)
7126 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
7127 ctxt->exception.error_code);
7129 kvm_queue_exception(vcpu, ctxt->exception.vector);
7133 static struct x86_emulate_ctxt *alloc_emulate_ctxt(struct kvm_vcpu *vcpu)
7135 struct x86_emulate_ctxt *ctxt;
7137 ctxt = kmem_cache_zalloc(x86_emulator_cache, GFP_KERNEL_ACCOUNT);
7139 pr_err("kvm: failed to allocate vcpu's emulator\n");
7144 ctxt->ops = &emulate_ops;
7145 vcpu->arch.emulate_ctxt = ctxt;
7150 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
7152 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7155 static_call(kvm_x86_get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
7157 ctxt->gpa_available = false;
7158 ctxt->eflags = kvm_get_rflags(vcpu);
7159 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
7161 ctxt->eip = kvm_rip_read(vcpu);
7162 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
7163 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
7164 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
7165 cs_db ? X86EMUL_MODE_PROT32 :
7166 X86EMUL_MODE_PROT16;
7167 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
7168 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
7169 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
7171 init_decode_cache(ctxt);
7172 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7175 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
7177 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7180 init_emulate_ctxt(vcpu);
7184 ctxt->_eip = ctxt->eip + inc_eip;
7185 ret = emulate_int_real(ctxt, irq);
7187 if (ret != X86EMUL_CONTINUE) {
7188 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7190 ctxt->eip = ctxt->_eip;
7191 kvm_rip_write(vcpu, ctxt->eip);
7192 kvm_set_rflags(vcpu, ctxt->eflags);
7195 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
7197 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
7199 ++vcpu->stat.insn_emulation_fail;
7200 trace_kvm_emulate_insn_failed(vcpu);
7202 if (emulation_type & EMULTYPE_VMWARE_GP) {
7203 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7207 if (emulation_type & EMULTYPE_SKIP) {
7208 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7209 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7210 vcpu->run->internal.ndata = 0;
7214 kvm_queue_exception(vcpu, UD_VECTOR);
7216 if (!is_guest_mode(vcpu) && static_call(kvm_x86_get_cpl)(vcpu) == 0) {
7217 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
7218 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
7219 vcpu->run->internal.ndata = 0;
7226 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7227 bool write_fault_to_shadow_pgtable,
7230 gpa_t gpa = cr2_or_gpa;
7233 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7236 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7237 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7240 if (!vcpu->arch.mmu->direct_map) {
7242 * Write permission should be allowed since only
7243 * write access need to be emulated.
7245 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7248 * If the mapping is invalid in guest, let cpu retry
7249 * it to generate fault.
7251 if (gpa == UNMAPPED_GVA)
7256 * Do not retry the unhandleable instruction if it faults on the
7257 * readonly host memory, otherwise it will goto a infinite loop:
7258 * retry instruction -> write #PF -> emulation fail -> retry
7259 * instruction -> ...
7261 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
7264 * If the instruction failed on the error pfn, it can not be fixed,
7265 * report the error to userspace.
7267 if (is_error_noslot_pfn(pfn))
7270 kvm_release_pfn_clean(pfn);
7272 /* The instructions are well-emulated on direct mmu. */
7273 if (vcpu->arch.mmu->direct_map) {
7274 unsigned int indirect_shadow_pages;
7276 write_lock(&vcpu->kvm->mmu_lock);
7277 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
7278 write_unlock(&vcpu->kvm->mmu_lock);
7280 if (indirect_shadow_pages)
7281 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7287 * if emulation was due to access to shadowed page table
7288 * and it failed try to unshadow page and re-enter the
7289 * guest to let CPU execute the instruction.
7291 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7294 * If the access faults on its page table, it can not
7295 * be fixed by unprotecting shadow page and it should
7296 * be reported to userspace.
7298 return !write_fault_to_shadow_pgtable;
7301 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
7302 gpa_t cr2_or_gpa, int emulation_type)
7304 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7305 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
7307 last_retry_eip = vcpu->arch.last_retry_eip;
7308 last_retry_addr = vcpu->arch.last_retry_addr;
7311 * If the emulation is caused by #PF and it is non-page_table
7312 * writing instruction, it means the VM-EXIT is caused by shadow
7313 * page protected, we can zap the shadow page and retry this
7314 * instruction directly.
7316 * Note: if the guest uses a non-page-table modifying instruction
7317 * on the PDE that points to the instruction, then we will unmap
7318 * the instruction and go to an infinite loop. So, we cache the
7319 * last retried eip and the last fault address, if we meet the eip
7320 * and the address again, we can break out of the potential infinite
7323 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
7325 if (!(emulation_type & EMULTYPE_ALLOW_RETRY_PF))
7328 if (WARN_ON_ONCE(is_guest_mode(vcpu)) ||
7329 WARN_ON_ONCE(!(emulation_type & EMULTYPE_PF)))
7332 if (x86_page_table_writing_insn(ctxt))
7335 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
7338 vcpu->arch.last_retry_eip = ctxt->eip;
7339 vcpu->arch.last_retry_addr = cr2_or_gpa;
7341 if (!vcpu->arch.mmu->direct_map)
7342 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
7344 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
7349 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
7350 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
7352 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
7354 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
7355 /* This is a good place to trace that we are exiting SMM. */
7356 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
7358 /* Process a latched INIT or SMI, if any. */
7359 kvm_make_request(KVM_REQ_EVENT, vcpu);
7362 kvm_mmu_reset_context(vcpu);
7365 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
7374 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
7375 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
7380 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
7382 struct kvm_run *kvm_run = vcpu->run;
7384 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
7385 kvm_run->debug.arch.dr6 = DR6_BS | DR6_ACTIVE_LOW;
7386 kvm_run->debug.arch.pc = kvm_get_linear_rip(vcpu);
7387 kvm_run->debug.arch.exception = DB_VECTOR;
7388 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7391 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
7395 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
7397 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7400 r = static_call(kvm_x86_skip_emulated_instruction)(vcpu);
7405 * rflags is the old, "raw" value of the flags. The new value has
7406 * not been saved yet.
7408 * This is correct even for TF set by the guest, because "the
7409 * processor will not generate this exception after the instruction
7410 * that sets the TF flag".
7412 if (unlikely(rflags & X86_EFLAGS_TF))
7413 r = kvm_vcpu_do_singlestep(vcpu);
7416 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
7418 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
7420 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
7421 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
7422 struct kvm_run *kvm_run = vcpu->run;
7423 unsigned long eip = kvm_get_linear_rip(vcpu);
7424 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7425 vcpu->arch.guest_debug_dr7,
7429 kvm_run->debug.arch.dr6 = dr6 | DR6_ACTIVE_LOW;
7430 kvm_run->debug.arch.pc = eip;
7431 kvm_run->debug.arch.exception = DB_VECTOR;
7432 kvm_run->exit_reason = KVM_EXIT_DEBUG;
7438 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
7439 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
7440 unsigned long eip = kvm_get_linear_rip(vcpu);
7441 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
7446 kvm_queue_exception_p(vcpu, DB_VECTOR, dr6);
7455 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
7457 switch (ctxt->opcode_len) {
7464 case 0xe6: /* OUT */
7468 case 0x6c: /* INS */
7470 case 0x6e: /* OUTS */
7477 case 0x33: /* RDPMC */
7487 * Decode to be emulated instruction. Return EMULATION_OK if success.
7489 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
7490 void *insn, int insn_len)
7492 int r = EMULATION_OK;
7493 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7495 init_emulate_ctxt(vcpu);
7498 * We will reenter on the same instruction since we do not set
7499 * complete_userspace_io. This does not handle watchpoints yet,
7500 * those would be handled in the emulate_ops.
7502 if (!(emulation_type & EMULTYPE_SKIP) &&
7503 kvm_vcpu_check_breakpoint(vcpu, &r))
7506 ctxt->interruptibility = 0;
7507 ctxt->have_exception = false;
7508 ctxt->exception.vector = -1;
7509 ctxt->perm_ok = false;
7511 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
7513 r = x86_decode_insn(ctxt, insn, insn_len);
7515 trace_kvm_emulate_insn_start(vcpu);
7516 ++vcpu->stat.insn_emulation;
7520 EXPORT_SYMBOL_GPL(x86_decode_emulated_instruction);
7522 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
7523 int emulation_type, void *insn, int insn_len)
7526 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
7527 bool writeback = true;
7528 bool write_fault_to_spt;
7530 if (unlikely(!static_call(kvm_x86_can_emulate_instruction)(vcpu, insn, insn_len)))
7533 vcpu->arch.l1tf_flush_l1d = true;
7536 * Clear write_fault_to_shadow_pgtable here to ensure it is
7539 write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
7540 vcpu->arch.write_fault_to_shadow_pgtable = false;
7542 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
7543 kvm_clear_exception_queue(vcpu);
7545 r = x86_decode_emulated_instruction(vcpu, emulation_type,
7547 if (r != EMULATION_OK) {
7548 if ((emulation_type & EMULTYPE_TRAP_UD) ||
7549 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
7550 kvm_queue_exception(vcpu, UD_VECTOR);
7553 if (reexecute_instruction(vcpu, cr2_or_gpa,
7557 if (ctxt->have_exception) {
7559 * #UD should result in just EMULATION_FAILED, and trap-like
7560 * exception should not be encountered during decode.
7562 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
7563 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
7564 inject_emulated_exception(vcpu);
7567 return handle_emulation_failure(vcpu, emulation_type);
7571 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
7572 !is_vmware_backdoor_opcode(ctxt)) {
7573 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7578 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
7579 * for kvm_skip_emulated_instruction(). The caller is responsible for
7580 * updating interruptibility state and injecting single-step #DBs.
7582 if (emulation_type & EMULTYPE_SKIP) {
7583 kvm_rip_write(vcpu, ctxt->_eip);
7584 if (ctxt->eflags & X86_EFLAGS_RF)
7585 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
7589 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
7592 /* this is needed for vmware backdoor interface to work since it
7593 changes registers values during IO operation */
7594 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
7595 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
7596 emulator_invalidate_register_cache(ctxt);
7600 if (emulation_type & EMULTYPE_PF) {
7601 /* Save the faulting GPA (cr2) in the address field */
7602 ctxt->exception.address = cr2_or_gpa;
7604 /* With shadow page tables, cr2 contains a GVA or nGPA. */
7605 if (vcpu->arch.mmu->direct_map) {
7606 ctxt->gpa_available = true;
7607 ctxt->gpa_val = cr2_or_gpa;
7610 /* Sanitize the address out of an abundance of paranoia. */
7611 ctxt->exception.address = 0;
7614 r = x86_emulate_insn(ctxt);
7616 if (r == EMULATION_INTERCEPTED)
7619 if (r == EMULATION_FAILED) {
7620 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
7624 return handle_emulation_failure(vcpu, emulation_type);
7627 if (ctxt->have_exception) {
7629 if (inject_emulated_exception(vcpu))
7631 } else if (vcpu->arch.pio.count) {
7632 if (!vcpu->arch.pio.in) {
7633 /* FIXME: return into emulator if single-stepping. */
7634 vcpu->arch.pio.count = 0;
7637 vcpu->arch.complete_userspace_io = complete_emulated_pio;
7640 } else if (vcpu->mmio_needed) {
7641 ++vcpu->stat.mmio_exits;
7643 if (!vcpu->mmio_is_write)
7646 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
7647 } else if (r == EMULATION_RESTART)
7653 unsigned long rflags = static_call(kvm_x86_get_rflags)(vcpu);
7654 toggle_interruptibility(vcpu, ctxt->interruptibility);
7655 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
7656 if (!ctxt->have_exception ||
7657 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
7658 kvm_rip_write(vcpu, ctxt->eip);
7659 if (r && (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
7660 r = kvm_vcpu_do_singlestep(vcpu);
7661 if (kvm_x86_ops.update_emulated_instruction)
7662 static_call(kvm_x86_update_emulated_instruction)(vcpu);
7663 __kvm_set_rflags(vcpu, ctxt->eflags);
7667 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
7668 * do nothing, and it will be requested again as soon as
7669 * the shadow expires. But we still need to check here,
7670 * because POPF has no interrupt shadow.
7672 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
7673 kvm_make_request(KVM_REQ_EVENT, vcpu);
7675 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
7680 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
7682 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
7684 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
7686 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
7687 void *insn, int insn_len)
7689 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
7691 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
7693 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
7695 vcpu->arch.pio.count = 0;
7699 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
7701 vcpu->arch.pio.count = 0;
7703 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
7706 return kvm_skip_emulated_instruction(vcpu);
7709 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
7710 unsigned short port)
7712 unsigned long val = kvm_rax_read(vcpu);
7713 int ret = emulator_pio_out(vcpu, size, port, &val, 1);
7719 * Workaround userspace that relies on old KVM behavior of %rip being
7720 * incremented prior to exiting to userspace to handle "OUT 0x7e".
7723 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
7724 vcpu->arch.complete_userspace_io =
7725 complete_fast_pio_out_port_0x7e;
7726 kvm_skip_emulated_instruction(vcpu);
7728 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7729 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
7734 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
7738 /* We should only ever be called with arch.pio.count equal to 1 */
7739 BUG_ON(vcpu->arch.pio.count != 1);
7741 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
7742 vcpu->arch.pio.count = 0;
7746 /* For size less than 4 we merge, else we zero extend */
7747 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
7750 * Since vcpu->arch.pio.count == 1 let emulator_pio_in perform
7751 * the copy and tracing
7753 emulator_pio_in(vcpu, vcpu->arch.pio.size, vcpu->arch.pio.port, &val, 1);
7754 kvm_rax_write(vcpu, val);
7756 return kvm_skip_emulated_instruction(vcpu);
7759 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
7760 unsigned short port)
7765 /* For size less than 4 we merge, else we zero extend */
7766 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
7768 ret = emulator_pio_in(vcpu, size, port, &val, 1);
7770 kvm_rax_write(vcpu, val);
7774 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
7775 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
7780 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
7785 ret = kvm_fast_pio_in(vcpu, size, port);
7787 ret = kvm_fast_pio_out(vcpu, size, port);
7788 return ret && kvm_skip_emulated_instruction(vcpu);
7790 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7792 static int kvmclock_cpu_down_prep(unsigned int cpu)
7794 __this_cpu_write(cpu_tsc_khz, 0);
7798 static void tsc_khz_changed(void *data)
7800 struct cpufreq_freqs *freq = data;
7801 unsigned long khz = 0;
7805 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7806 khz = cpufreq_quick_get(raw_smp_processor_id());
7809 __this_cpu_write(cpu_tsc_khz, khz);
7812 #ifdef CONFIG_X86_64
7813 static void kvm_hyperv_tsc_notifier(void)
7816 struct kvm_vcpu *vcpu;
7818 unsigned long flags;
7820 mutex_lock(&kvm_lock);
7821 list_for_each_entry(kvm, &vm_list, vm_list)
7822 kvm_make_mclock_inprogress_request(kvm);
7824 hyperv_stop_tsc_emulation();
7826 /* TSC frequency always matches when on Hyper-V */
7827 for_each_present_cpu(cpu)
7828 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7829 kvm_max_guest_tsc_khz = tsc_khz;
7831 list_for_each_entry(kvm, &vm_list, vm_list) {
7832 struct kvm_arch *ka = &kvm->arch;
7834 spin_lock_irqsave(&ka->pvclock_gtod_sync_lock, flags);
7835 pvclock_update_vm_gtod_copy(kvm);
7836 spin_unlock_irqrestore(&ka->pvclock_gtod_sync_lock, flags);
7838 kvm_for_each_vcpu(cpu, vcpu, kvm)
7839 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7841 kvm_for_each_vcpu(cpu, vcpu, kvm)
7842 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7844 mutex_unlock(&kvm_lock);
7848 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7851 struct kvm_vcpu *vcpu;
7852 int i, send_ipi = 0;
7855 * We allow guests to temporarily run on slowing clocks,
7856 * provided we notify them after, or to run on accelerating
7857 * clocks, provided we notify them before. Thus time never
7860 * However, we have a problem. We can't atomically update
7861 * the frequency of a given CPU from this function; it is
7862 * merely a notifier, which can be called from any CPU.
7863 * Changing the TSC frequency at arbitrary points in time
7864 * requires a recomputation of local variables related to
7865 * the TSC for each VCPU. We must flag these local variables
7866 * to be updated and be sure the update takes place with the
7867 * new frequency before any guests proceed.
7869 * Unfortunately, the combination of hotplug CPU and frequency
7870 * change creates an intractable locking scenario; the order
7871 * of when these callouts happen is undefined with respect to
7872 * CPU hotplug, and they can race with each other. As such,
7873 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7874 * undefined; you can actually have a CPU frequency change take
7875 * place in between the computation of X and the setting of the
7876 * variable. To protect against this problem, all updates of
7877 * the per_cpu tsc_khz variable are done in an interrupt
7878 * protected IPI, and all callers wishing to update the value
7879 * must wait for a synchronous IPI to complete (which is trivial
7880 * if the caller is on the CPU already). This establishes the
7881 * necessary total order on variable updates.
7883 * Note that because a guest time update may take place
7884 * anytime after the setting of the VCPU's request bit, the
7885 * correct TSC value must be set before the request. However,
7886 * to ensure the update actually makes it to any guest which
7887 * starts running in hardware virtualization between the set
7888 * and the acquisition of the spinlock, we must also ping the
7889 * CPU after setting the request bit.
7893 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7895 mutex_lock(&kvm_lock);
7896 list_for_each_entry(kvm, &vm_list, vm_list) {
7897 kvm_for_each_vcpu(i, vcpu, kvm) {
7898 if (vcpu->cpu != cpu)
7900 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7901 if (vcpu->cpu != raw_smp_processor_id())
7905 mutex_unlock(&kvm_lock);
7907 if (freq->old < freq->new && send_ipi) {
7909 * We upscale the frequency. Must make the guest
7910 * doesn't see old kvmclock values while running with
7911 * the new frequency, otherwise we risk the guest sees
7912 * time go backwards.
7914 * In case we update the frequency for another cpu
7915 * (which might be in guest context) send an interrupt
7916 * to kick the cpu out of guest context. Next time
7917 * guest context is entered kvmclock will be updated,
7918 * so the guest will not see stale values.
7920 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7924 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7927 struct cpufreq_freqs *freq = data;
7930 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7932 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7935 for_each_cpu(cpu, freq->policy->cpus)
7936 __kvmclock_cpufreq_notifier(freq, cpu);
7941 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7942 .notifier_call = kvmclock_cpufreq_notifier
7945 static int kvmclock_cpu_online(unsigned int cpu)
7947 tsc_khz_changed(NULL);
7951 static void kvm_timer_init(void)
7953 max_tsc_khz = tsc_khz;
7955 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7956 #ifdef CONFIG_CPU_FREQ
7957 struct cpufreq_policy *policy;
7961 policy = cpufreq_cpu_get(cpu);
7963 if (policy->cpuinfo.max_freq)
7964 max_tsc_khz = policy->cpuinfo.max_freq;
7965 cpufreq_cpu_put(policy);
7969 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7970 CPUFREQ_TRANSITION_NOTIFIER);
7973 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7974 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7977 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7978 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7980 int kvm_is_in_guest(void)
7982 return __this_cpu_read(current_vcpu) != NULL;
7985 static int kvm_is_user_mode(void)
7989 if (__this_cpu_read(current_vcpu))
7990 user_mode = static_call(kvm_x86_get_cpl)(__this_cpu_read(current_vcpu));
7992 return user_mode != 0;
7995 static unsigned long kvm_get_guest_ip(void)
7997 unsigned long ip = 0;
7999 if (__this_cpu_read(current_vcpu))
8000 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
8005 static void kvm_handle_intel_pt_intr(void)
8007 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
8009 kvm_make_request(KVM_REQ_PMI, vcpu);
8010 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
8011 (unsigned long *)&vcpu->arch.pmu.global_status);
8014 static struct perf_guest_info_callbacks kvm_guest_cbs = {
8015 .is_in_guest = kvm_is_in_guest,
8016 .is_user_mode = kvm_is_user_mode,
8017 .get_guest_ip = kvm_get_guest_ip,
8018 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
8021 #ifdef CONFIG_X86_64
8022 static void pvclock_gtod_update_fn(struct work_struct *work)
8026 struct kvm_vcpu *vcpu;
8029 mutex_lock(&kvm_lock);
8030 list_for_each_entry(kvm, &vm_list, vm_list)
8031 kvm_for_each_vcpu(i, vcpu, kvm)
8032 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8033 atomic_set(&kvm_guest_has_master_clock, 0);
8034 mutex_unlock(&kvm_lock);
8037 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
8040 * Notification about pvclock gtod data update.
8042 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
8045 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
8046 struct timekeeper *tk = priv;
8048 update_pvclock_gtod(tk);
8050 /* disable master clock if host does not trust, or does not
8051 * use, TSC based clocksource.
8053 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
8054 atomic_read(&kvm_guest_has_master_clock) != 0)
8055 queue_work(system_long_wq, &pvclock_gtod_work);
8060 static struct notifier_block pvclock_gtod_notifier = {
8061 .notifier_call = pvclock_gtod_notify,
8065 int kvm_arch_init(void *opaque)
8067 struct kvm_x86_init_ops *ops = opaque;
8070 if (kvm_x86_ops.hardware_enable) {
8071 printk(KERN_ERR "kvm: already loaded the other module\n");
8076 if (!ops->cpu_has_kvm_support()) {
8077 pr_err_ratelimited("kvm: no hardware support\n");
8081 if (ops->disabled_by_bios()) {
8082 pr_err_ratelimited("kvm: disabled by bios\n");
8088 * KVM explicitly assumes that the guest has an FPU and
8089 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
8090 * vCPU's FPU state as a fxregs_state struct.
8092 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
8093 printk(KERN_ERR "kvm: inadequate fpu\n");
8099 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
8100 __alignof__(struct fpu), SLAB_ACCOUNT,
8102 if (!x86_fpu_cache) {
8103 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
8107 x86_emulator_cache = kvm_alloc_emulator_cache();
8108 if (!x86_emulator_cache) {
8109 pr_err("kvm: failed to allocate cache for x86 emulator\n");
8110 goto out_free_x86_fpu_cache;
8113 user_return_msrs = alloc_percpu(struct kvm_user_return_msrs);
8114 if (!user_return_msrs) {
8115 printk(KERN_ERR "kvm: failed to allocate percpu kvm_user_return_msrs\n");
8116 goto out_free_x86_emulator_cache;
8119 r = kvm_mmu_module_init();
8121 goto out_free_percpu;
8125 perf_register_guest_info_callbacks(&kvm_guest_cbs);
8127 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
8128 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
8129 supported_xcr0 = host_xcr0 & KVM_SUPPORTED_XCR0;
8132 if (pi_inject_timer == -1)
8133 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
8134 #ifdef CONFIG_X86_64
8135 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
8137 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8138 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
8144 free_percpu(user_return_msrs);
8145 out_free_x86_emulator_cache:
8146 kmem_cache_destroy(x86_emulator_cache);
8147 out_free_x86_fpu_cache:
8148 kmem_cache_destroy(x86_fpu_cache);
8153 void kvm_arch_exit(void)
8155 #ifdef CONFIG_X86_64
8156 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
8157 clear_hv_tscchange_cb();
8160 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
8162 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
8163 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
8164 CPUFREQ_TRANSITION_NOTIFIER);
8165 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
8166 #ifdef CONFIG_X86_64
8167 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
8169 kvm_x86_ops.hardware_enable = NULL;
8170 kvm_mmu_module_exit();
8171 free_percpu(user_return_msrs);
8172 kmem_cache_destroy(x86_fpu_cache);
8173 #ifdef CONFIG_KVM_XEN
8174 static_key_deferred_flush(&kvm_xen_enabled);
8175 WARN_ON(static_branch_unlikely(&kvm_xen_enabled.key));
8179 static int __kvm_vcpu_halt(struct kvm_vcpu *vcpu, int state, int reason)
8181 ++vcpu->stat.halt_exits;
8182 if (lapic_in_kernel(vcpu)) {
8183 vcpu->arch.mp_state = state;
8186 vcpu->run->exit_reason = reason;
8191 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
8193 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_HALTED, KVM_EXIT_HLT);
8195 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
8197 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
8199 int ret = kvm_skip_emulated_instruction(vcpu);
8201 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
8202 * KVM_EXIT_DEBUG here.
8204 return kvm_vcpu_halt(vcpu) && ret;
8206 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
8208 int kvm_emulate_ap_reset_hold(struct kvm_vcpu *vcpu)
8210 int ret = kvm_skip_emulated_instruction(vcpu);
8212 return __kvm_vcpu_halt(vcpu, KVM_MP_STATE_AP_RESET_HOLD, KVM_EXIT_AP_RESET_HOLD) && ret;
8214 EXPORT_SYMBOL_GPL(kvm_emulate_ap_reset_hold);
8216 #ifdef CONFIG_X86_64
8217 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
8218 unsigned long clock_type)
8220 struct kvm_clock_pairing clock_pairing;
8221 struct timespec64 ts;
8225 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
8226 return -KVM_EOPNOTSUPP;
8228 if (!kvm_get_walltime_and_clockread(&ts, &cycle))
8229 return -KVM_EOPNOTSUPP;
8231 clock_pairing.sec = ts.tv_sec;
8232 clock_pairing.nsec = ts.tv_nsec;
8233 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
8234 clock_pairing.flags = 0;
8235 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
8238 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
8239 sizeof(struct kvm_clock_pairing)))
8247 * kvm_pv_kick_cpu_op: Kick a vcpu.
8249 * @apicid - apicid of vcpu to be kicked.
8251 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
8253 struct kvm_lapic_irq lapic_irq;
8255 lapic_irq.shorthand = APIC_DEST_NOSHORT;
8256 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
8257 lapic_irq.level = 0;
8258 lapic_irq.dest_id = apicid;
8259 lapic_irq.msi_redir_hint = false;
8261 lapic_irq.delivery_mode = APIC_DM_REMRD;
8262 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
8265 bool kvm_apicv_activated(struct kvm *kvm)
8267 return (READ_ONCE(kvm->arch.apicv_inhibit_reasons) == 0);
8269 EXPORT_SYMBOL_GPL(kvm_apicv_activated);
8271 void kvm_apicv_init(struct kvm *kvm, bool enable)
8274 clear_bit(APICV_INHIBIT_REASON_DISABLE,
8275 &kvm->arch.apicv_inhibit_reasons);
8277 set_bit(APICV_INHIBIT_REASON_DISABLE,
8278 &kvm->arch.apicv_inhibit_reasons);
8280 EXPORT_SYMBOL_GPL(kvm_apicv_init);
8282 static void kvm_sched_yield(struct kvm_vcpu *vcpu, unsigned long dest_id)
8284 struct kvm_vcpu *target = NULL;
8285 struct kvm_apic_map *map;
8287 vcpu->stat.directed_yield_attempted++;
8290 map = rcu_dereference(vcpu->kvm->arch.apic_map);
8292 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
8293 target = map->phys_map[dest_id]->vcpu;
8297 if (!target || !READ_ONCE(target->ready))
8300 /* Ignore requests to yield to self */
8304 if (kvm_vcpu_yield_to(target) <= 0)
8307 vcpu->stat.directed_yield_successful++;
8313 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
8315 unsigned long nr, a0, a1, a2, a3, ret;
8318 if (kvm_xen_hypercall_enabled(vcpu->kvm))
8319 return kvm_xen_hypercall(vcpu);
8321 if (kvm_hv_hypercall_enabled(vcpu))
8322 return kvm_hv_hypercall(vcpu);
8324 nr = kvm_rax_read(vcpu);
8325 a0 = kvm_rbx_read(vcpu);
8326 a1 = kvm_rcx_read(vcpu);
8327 a2 = kvm_rdx_read(vcpu);
8328 a3 = kvm_rsi_read(vcpu);
8330 trace_kvm_hypercall(nr, a0, a1, a2, a3);
8332 op_64_bit = is_64_bit_mode(vcpu);
8341 if (static_call(kvm_x86_get_cpl)(vcpu) != 0) {
8349 case KVM_HC_VAPIC_POLL_IRQ:
8352 case KVM_HC_KICK_CPU:
8353 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_UNHALT))
8356 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
8357 kvm_sched_yield(vcpu, a1);
8360 #ifdef CONFIG_X86_64
8361 case KVM_HC_CLOCK_PAIRING:
8362 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
8365 case KVM_HC_SEND_IPI:
8366 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SEND_IPI))
8369 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
8371 case KVM_HC_SCHED_YIELD:
8372 if (!guest_pv_has(vcpu, KVM_FEATURE_PV_SCHED_YIELD))
8375 kvm_sched_yield(vcpu, a0);
8385 kvm_rax_write(vcpu, ret);
8387 ++vcpu->stat.hypercalls;
8388 return kvm_skip_emulated_instruction(vcpu);
8390 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
8392 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
8394 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
8395 char instruction[3];
8396 unsigned long rip = kvm_rip_read(vcpu);
8398 static_call(kvm_x86_patch_hypercall)(vcpu, instruction);
8400 return emulator_write_emulated(ctxt, rip, instruction, 3,
8404 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
8406 return vcpu->run->request_interrupt_window &&
8407 likely(!pic_in_kernel(vcpu->kvm));
8410 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
8412 struct kvm_run *kvm_run = vcpu->run;
8415 * if_flag is obsolete and useless, so do not bother
8416 * setting it for SEV-ES guests. Userspace can just
8417 * use kvm_run->ready_for_interrupt_injection.
8419 kvm_run->if_flag = !vcpu->arch.guest_state_protected
8420 && (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
8422 kvm_run->cr8 = kvm_get_cr8(vcpu);
8423 kvm_run->apic_base = kvm_get_apic_base(vcpu);
8424 kvm_run->ready_for_interrupt_injection =
8425 pic_in_kernel(vcpu->kvm) ||
8426 kvm_vcpu_ready_for_interrupt_injection(vcpu);
8429 kvm_run->flags |= KVM_RUN_X86_SMM;
8432 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
8436 if (!kvm_x86_ops.update_cr8_intercept)
8439 if (!lapic_in_kernel(vcpu))
8442 if (vcpu->arch.apicv_active)
8445 if (!vcpu->arch.apic->vapic_addr)
8446 max_irr = kvm_lapic_find_highest_irr(vcpu);
8453 tpr = kvm_lapic_get_cr8(vcpu);
8455 static_call(kvm_x86_update_cr8_intercept)(vcpu, tpr, max_irr);
8459 int kvm_check_nested_events(struct kvm_vcpu *vcpu)
8461 if (WARN_ON_ONCE(!is_guest_mode(vcpu)))
8464 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8465 kvm_x86_ops.nested_ops->triple_fault(vcpu);
8469 return kvm_x86_ops.nested_ops->check_events(vcpu);
8472 static void kvm_inject_exception(struct kvm_vcpu *vcpu)
8474 if (vcpu->arch.exception.error_code && !is_protmode(vcpu))
8475 vcpu->arch.exception.error_code = false;
8476 static_call(kvm_x86_queue_exception)(vcpu);
8479 static void inject_pending_event(struct kvm_vcpu *vcpu, bool *req_immediate_exit)
8482 bool can_inject = true;
8484 /* try to reinject previous events if any */
8486 if (vcpu->arch.exception.injected) {
8487 kvm_inject_exception(vcpu);
8491 * Do not inject an NMI or interrupt if there is a pending
8492 * exception. Exceptions and interrupts are recognized at
8493 * instruction boundaries, i.e. the start of an instruction.
8494 * Trap-like exceptions, e.g. #DB, have higher priority than
8495 * NMIs and interrupts, i.e. traps are recognized before an
8496 * NMI/interrupt that's pending on the same instruction.
8497 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
8498 * priority, but are only generated (pended) during instruction
8499 * execution, i.e. a pending fault-like exception means the
8500 * fault occurred on the *previous* instruction and must be
8501 * serviced prior to recognizing any new events in order to
8502 * fully complete the previous instruction.
8504 else if (!vcpu->arch.exception.pending) {
8505 if (vcpu->arch.nmi_injected) {
8506 static_call(kvm_x86_set_nmi)(vcpu);
8508 } else if (vcpu->arch.interrupt.injected) {
8509 static_call(kvm_x86_set_irq)(vcpu);
8514 WARN_ON_ONCE(vcpu->arch.exception.injected &&
8515 vcpu->arch.exception.pending);
8518 * Call check_nested_events() even if we reinjected a previous event
8519 * in order for caller to determine if it should require immediate-exit
8520 * from L2 to L1 due to pending L1 events which require exit
8523 if (is_guest_mode(vcpu)) {
8524 r = kvm_check_nested_events(vcpu);
8529 /* try to inject new event if pending */
8530 if (vcpu->arch.exception.pending) {
8531 trace_kvm_inj_exception(vcpu->arch.exception.nr,
8532 vcpu->arch.exception.has_error_code,
8533 vcpu->arch.exception.error_code);
8535 vcpu->arch.exception.pending = false;
8536 vcpu->arch.exception.injected = true;
8538 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
8539 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
8542 if (vcpu->arch.exception.nr == DB_VECTOR) {
8543 kvm_deliver_exception_payload(vcpu);
8544 if (vcpu->arch.dr7 & DR7_GD) {
8545 vcpu->arch.dr7 &= ~DR7_GD;
8546 kvm_update_dr7(vcpu);
8550 kvm_inject_exception(vcpu);
8555 * Finally, inject interrupt events. If an event cannot be injected
8556 * due to architectural conditions (e.g. IF=0) a window-open exit
8557 * will re-request KVM_REQ_EVENT. Sometimes however an event is pending
8558 * and can architecturally be injected, but we cannot do it right now:
8559 * an interrupt could have arrived just now and we have to inject it
8560 * as a vmexit, or there could already an event in the queue, which is
8561 * indicated by can_inject. In that case we request an immediate exit
8562 * in order to make progress and get back here for another iteration.
8563 * The kvm_x86_ops hooks communicate this by returning -EBUSY.
8565 if (vcpu->arch.smi_pending) {
8566 r = can_inject ? static_call(kvm_x86_smi_allowed)(vcpu, true) : -EBUSY;
8570 vcpu->arch.smi_pending = false;
8571 ++vcpu->arch.smi_count;
8575 static_call(kvm_x86_enable_smi_window)(vcpu);
8578 if (vcpu->arch.nmi_pending) {
8579 r = can_inject ? static_call(kvm_x86_nmi_allowed)(vcpu, true) : -EBUSY;
8583 --vcpu->arch.nmi_pending;
8584 vcpu->arch.nmi_injected = true;
8585 static_call(kvm_x86_set_nmi)(vcpu);
8587 WARN_ON(static_call(kvm_x86_nmi_allowed)(vcpu, true) < 0);
8589 if (vcpu->arch.nmi_pending)
8590 static_call(kvm_x86_enable_nmi_window)(vcpu);
8593 if (kvm_cpu_has_injectable_intr(vcpu)) {
8594 r = can_inject ? static_call(kvm_x86_interrupt_allowed)(vcpu, true) : -EBUSY;
8598 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu), false);
8599 static_call(kvm_x86_set_irq)(vcpu);
8600 WARN_ON(static_call(kvm_x86_interrupt_allowed)(vcpu, true) < 0);
8602 if (kvm_cpu_has_injectable_intr(vcpu))
8603 static_call(kvm_x86_enable_irq_window)(vcpu);
8606 if (is_guest_mode(vcpu) &&
8607 kvm_x86_ops.nested_ops->hv_timer_pending &&
8608 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
8609 *req_immediate_exit = true;
8611 WARN_ON(vcpu->arch.exception.pending);
8615 *req_immediate_exit = true;
8619 static void process_nmi(struct kvm_vcpu *vcpu)
8624 * x86 is limited to one NMI running, and one NMI pending after it.
8625 * If an NMI is already in progress, limit further NMIs to just one.
8626 * Otherwise, allow two (and we'll inject the first one immediately).
8628 if (static_call(kvm_x86_get_nmi_mask)(vcpu) || vcpu->arch.nmi_injected)
8631 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
8632 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
8633 kvm_make_request(KVM_REQ_EVENT, vcpu);
8636 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
8639 flags |= seg->g << 23;
8640 flags |= seg->db << 22;
8641 flags |= seg->l << 21;
8642 flags |= seg->avl << 20;
8643 flags |= seg->present << 15;
8644 flags |= seg->dpl << 13;
8645 flags |= seg->s << 12;
8646 flags |= seg->type << 8;
8650 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
8652 struct kvm_segment seg;
8655 kvm_get_segment(vcpu, &seg, n);
8656 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
8659 offset = 0x7f84 + n * 12;
8661 offset = 0x7f2c + (n - 3) * 12;
8663 put_smstate(u32, buf, offset + 8, seg.base);
8664 put_smstate(u32, buf, offset + 4, seg.limit);
8665 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
8668 #ifdef CONFIG_X86_64
8669 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
8671 struct kvm_segment seg;
8675 kvm_get_segment(vcpu, &seg, n);
8676 offset = 0x7e00 + n * 16;
8678 flags = enter_smm_get_segment_flags(&seg) >> 8;
8679 put_smstate(u16, buf, offset, seg.selector);
8680 put_smstate(u16, buf, offset + 2, flags);
8681 put_smstate(u32, buf, offset + 4, seg.limit);
8682 put_smstate(u64, buf, offset + 8, seg.base);
8686 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
8689 struct kvm_segment seg;
8693 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
8694 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
8695 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
8696 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
8698 for (i = 0; i < 8; i++)
8699 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
8701 kvm_get_dr(vcpu, 6, &val);
8702 put_smstate(u32, buf, 0x7fcc, (u32)val);
8703 kvm_get_dr(vcpu, 7, &val);
8704 put_smstate(u32, buf, 0x7fc8, (u32)val);
8706 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8707 put_smstate(u32, buf, 0x7fc4, seg.selector);
8708 put_smstate(u32, buf, 0x7f64, seg.base);
8709 put_smstate(u32, buf, 0x7f60, seg.limit);
8710 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
8712 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8713 put_smstate(u32, buf, 0x7fc0, seg.selector);
8714 put_smstate(u32, buf, 0x7f80, seg.base);
8715 put_smstate(u32, buf, 0x7f7c, seg.limit);
8716 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
8718 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8719 put_smstate(u32, buf, 0x7f74, dt.address);
8720 put_smstate(u32, buf, 0x7f70, dt.size);
8722 static_call(kvm_x86_get_idt)(vcpu, &dt);
8723 put_smstate(u32, buf, 0x7f58, dt.address);
8724 put_smstate(u32, buf, 0x7f54, dt.size);
8726 for (i = 0; i < 6; i++)
8727 enter_smm_save_seg_32(vcpu, buf, i);
8729 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
8732 put_smstate(u32, buf, 0x7efc, 0x00020000);
8733 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
8736 #ifdef CONFIG_X86_64
8737 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
8740 struct kvm_segment seg;
8744 for (i = 0; i < 16; i++)
8745 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
8747 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
8748 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
8750 kvm_get_dr(vcpu, 6, &val);
8751 put_smstate(u64, buf, 0x7f68, val);
8752 kvm_get_dr(vcpu, 7, &val);
8753 put_smstate(u64, buf, 0x7f60, val);
8755 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
8756 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
8757 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
8759 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
8762 put_smstate(u32, buf, 0x7efc, 0x00020064);
8764 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
8766 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
8767 put_smstate(u16, buf, 0x7e90, seg.selector);
8768 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
8769 put_smstate(u32, buf, 0x7e94, seg.limit);
8770 put_smstate(u64, buf, 0x7e98, seg.base);
8772 static_call(kvm_x86_get_idt)(vcpu, &dt);
8773 put_smstate(u32, buf, 0x7e84, dt.size);
8774 put_smstate(u64, buf, 0x7e88, dt.address);
8776 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
8777 put_smstate(u16, buf, 0x7e70, seg.selector);
8778 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
8779 put_smstate(u32, buf, 0x7e74, seg.limit);
8780 put_smstate(u64, buf, 0x7e78, seg.base);
8782 static_call(kvm_x86_get_gdt)(vcpu, &dt);
8783 put_smstate(u32, buf, 0x7e64, dt.size);
8784 put_smstate(u64, buf, 0x7e68, dt.address);
8786 for (i = 0; i < 6; i++)
8787 enter_smm_save_seg_64(vcpu, buf, i);
8791 static void enter_smm(struct kvm_vcpu *vcpu)
8793 struct kvm_segment cs, ds;
8798 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
8799 memset(buf, 0, 512);
8800 #ifdef CONFIG_X86_64
8801 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8802 enter_smm_save_state_64(vcpu, buf);
8805 enter_smm_save_state_32(vcpu, buf);
8808 * Give pre_enter_smm() a chance to make ISA-specific changes to the
8809 * vCPU state (e.g. leave guest mode) after we've saved the state into
8810 * the SMM state-save area.
8812 static_call(kvm_x86_pre_enter_smm)(vcpu, buf);
8814 vcpu->arch.hflags |= HF_SMM_MASK;
8815 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
8817 if (static_call(kvm_x86_get_nmi_mask)(vcpu))
8818 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
8820 static_call(kvm_x86_set_nmi_mask)(vcpu, true);
8822 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
8823 kvm_rip_write(vcpu, 0x8000);
8825 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
8826 static_call(kvm_x86_set_cr0)(vcpu, cr0);
8827 vcpu->arch.cr0 = cr0;
8829 static_call(kvm_x86_set_cr4)(vcpu, 0);
8831 /* Undocumented: IDT limit is set to zero on entry to SMM. */
8832 dt.address = dt.size = 0;
8833 static_call(kvm_x86_set_idt)(vcpu, &dt);
8835 kvm_set_dr(vcpu, 7, DR7_FIXED_1);
8837 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
8838 cs.base = vcpu->arch.smbase;
8843 cs.limit = ds.limit = 0xffffffff;
8844 cs.type = ds.type = 0x3;
8845 cs.dpl = ds.dpl = 0;
8850 cs.avl = ds.avl = 0;
8851 cs.present = ds.present = 1;
8852 cs.unusable = ds.unusable = 0;
8853 cs.padding = ds.padding = 0;
8855 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8856 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
8857 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
8858 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
8859 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
8860 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
8862 #ifdef CONFIG_X86_64
8863 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
8864 static_call(kvm_x86_set_efer)(vcpu, 0);
8867 kvm_update_cpuid_runtime(vcpu);
8868 kvm_mmu_reset_context(vcpu);
8871 static void process_smi(struct kvm_vcpu *vcpu)
8873 vcpu->arch.smi_pending = true;
8874 kvm_make_request(KVM_REQ_EVENT, vcpu);
8877 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
8878 unsigned long *vcpu_bitmap)
8882 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
8884 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
8885 NULL, vcpu_bitmap, cpus);
8887 free_cpumask_var(cpus);
8890 void kvm_make_scan_ioapic_request(struct kvm *kvm)
8892 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
8895 void kvm_vcpu_update_apicv(struct kvm_vcpu *vcpu)
8897 if (!lapic_in_kernel(vcpu))
8900 vcpu->arch.apicv_active = kvm_apicv_activated(vcpu->kvm);
8901 kvm_apic_update_apicv(vcpu);
8902 static_call(kvm_x86_refresh_apicv_exec_ctrl)(vcpu);
8904 EXPORT_SYMBOL_GPL(kvm_vcpu_update_apicv);
8907 * NOTE: Do not hold any lock prior to calling this.
8909 * In particular, kvm_request_apicv_update() expects kvm->srcu not to be
8910 * locked, because it calls __x86_set_memory_region() which does
8911 * synchronize_srcu(&kvm->srcu).
8913 void kvm_request_apicv_update(struct kvm *kvm, bool activate, ulong bit)
8915 struct kvm_vcpu *except;
8916 unsigned long old, new, expected;
8918 if (!kvm_x86_ops.check_apicv_inhibit_reasons ||
8919 !static_call(kvm_x86_check_apicv_inhibit_reasons)(bit))
8922 old = READ_ONCE(kvm->arch.apicv_inhibit_reasons);
8924 expected = new = old;
8926 __clear_bit(bit, &new);
8928 __set_bit(bit, &new);
8931 old = cmpxchg(&kvm->arch.apicv_inhibit_reasons, expected, new);
8932 } while (old != expected);
8937 trace_kvm_apicv_update_request(activate, bit);
8938 if (kvm_x86_ops.pre_update_apicv_exec_ctrl)
8939 static_call(kvm_x86_pre_update_apicv_exec_ctrl)(kvm, activate);
8942 * Sending request to update APICV for all other vcpus,
8943 * while update the calling vcpu immediately instead of
8944 * waiting for another #VMEXIT to handle the request.
8946 except = kvm_get_running_vcpu();
8947 kvm_make_all_cpus_request_except(kvm, KVM_REQ_APICV_UPDATE,
8950 kvm_vcpu_update_apicv(except);
8952 EXPORT_SYMBOL_GPL(kvm_request_apicv_update);
8954 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
8956 if (!kvm_apic_present(vcpu))
8959 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8961 if (irqchip_split(vcpu->kvm))
8962 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8964 if (vcpu->arch.apicv_active)
8965 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
8966 if (ioapic_in_kernel(vcpu->kvm))
8967 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8970 if (is_guest_mode(vcpu))
8971 vcpu->arch.load_eoi_exitmap_pending = true;
8973 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8976 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8978 u64 eoi_exit_bitmap[4];
8980 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8983 if (to_hv_vcpu(vcpu))
8984 bitmap_or((ulong *)eoi_exit_bitmap,
8985 vcpu->arch.ioapic_handled_vectors,
8986 to_hv_synic(vcpu)->vec_bitmap, 256);
8988 static_call(kvm_x86_load_eoi_exitmap)(vcpu, eoi_exit_bitmap);
8991 void kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8992 unsigned long start, unsigned long end)
8994 unsigned long apic_address;
8997 * The physical address of apic access page is stored in the VMCS.
8998 * Update it when it becomes invalid.
9000 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
9001 if (start <= apic_address && apic_address < end)
9002 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
9005 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
9007 if (!lapic_in_kernel(vcpu))
9010 if (!kvm_x86_ops.set_apic_access_page_addr)
9013 static_call(kvm_x86_set_apic_access_page_addr)(vcpu);
9016 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
9018 smp_send_reschedule(vcpu->cpu);
9020 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
9023 * Returns 1 to let vcpu_run() continue the guest execution loop without
9024 * exiting to the userspace. Otherwise, the value will be returned to the
9027 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
9031 dm_request_for_irq_injection(vcpu) &&
9032 kvm_cpu_accept_dm_intr(vcpu);
9033 fastpath_t exit_fastpath;
9035 bool req_immediate_exit = false;
9037 /* Forbid vmenter if vcpu dirty ring is soft-full */
9038 if (unlikely(vcpu->kvm->dirty_ring_size &&
9039 kvm_dirty_ring_soft_full(&vcpu->dirty_ring))) {
9040 vcpu->run->exit_reason = KVM_EXIT_DIRTY_RING_FULL;
9041 trace_kvm_dirty_ring_exit(vcpu);
9046 if (kvm_request_pending(vcpu)) {
9047 if (kvm_check_request(KVM_REQ_GET_NESTED_STATE_PAGES, vcpu)) {
9048 if (unlikely(!kvm_x86_ops.nested_ops->get_nested_state_pages(vcpu))) {
9053 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
9054 kvm_mmu_unload(vcpu);
9055 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
9056 __kvm_migrate_timers(vcpu);
9057 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
9058 kvm_gen_update_masterclock(vcpu->kvm);
9059 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
9060 kvm_gen_kvmclock_update(vcpu);
9061 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
9062 r = kvm_guest_time_update(vcpu);
9066 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
9067 kvm_mmu_sync_roots(vcpu);
9068 if (kvm_check_request(KVM_REQ_LOAD_MMU_PGD, vcpu))
9069 kvm_mmu_load_pgd(vcpu);
9070 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu)) {
9071 kvm_vcpu_flush_tlb_all(vcpu);
9073 /* Flushing all ASIDs flushes the current ASID... */
9074 kvm_clear_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
9076 if (kvm_check_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu))
9077 kvm_vcpu_flush_tlb_current(vcpu);
9078 if (kvm_check_request(KVM_REQ_HV_TLB_FLUSH, vcpu))
9079 kvm_vcpu_flush_tlb_guest(vcpu);
9081 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
9082 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
9086 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
9087 if (is_guest_mode(vcpu)) {
9088 kvm_x86_ops.nested_ops->triple_fault(vcpu);
9090 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
9091 vcpu->mmio_needed = 0;
9096 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
9097 /* Page is swapped out. Do synthetic halt */
9098 vcpu->arch.apf.halted = true;
9102 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
9103 record_steal_time(vcpu);
9104 if (kvm_check_request(KVM_REQ_SMI, vcpu))
9106 if (kvm_check_request(KVM_REQ_NMI, vcpu))
9108 if (kvm_check_request(KVM_REQ_PMU, vcpu))
9109 kvm_pmu_handle_event(vcpu);
9110 if (kvm_check_request(KVM_REQ_PMI, vcpu))
9111 kvm_pmu_deliver_pmi(vcpu);
9112 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
9113 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
9114 if (test_bit(vcpu->arch.pending_ioapic_eoi,
9115 vcpu->arch.ioapic_handled_vectors)) {
9116 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
9117 vcpu->run->eoi.vector =
9118 vcpu->arch.pending_ioapic_eoi;
9123 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
9124 vcpu_scan_ioapic(vcpu);
9125 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
9126 vcpu_load_eoi_exitmap(vcpu);
9127 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
9128 kvm_vcpu_reload_apic_access_page(vcpu);
9129 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
9130 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9131 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
9135 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
9136 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
9137 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
9141 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
9142 struct kvm_vcpu_hv *hv_vcpu = to_hv_vcpu(vcpu);
9144 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
9145 vcpu->run->hyperv = hv_vcpu->exit;
9151 * KVM_REQ_HV_STIMER has to be processed after
9152 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
9153 * depend on the guest clock being up-to-date
9155 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
9156 kvm_hv_process_stimers(vcpu);
9157 if (kvm_check_request(KVM_REQ_APICV_UPDATE, vcpu))
9158 kvm_vcpu_update_apicv(vcpu);
9159 if (kvm_check_request(KVM_REQ_APF_READY, vcpu))
9160 kvm_check_async_pf_completion(vcpu);
9161 if (kvm_check_request(KVM_REQ_MSR_FILTER_CHANGED, vcpu))
9162 static_call(kvm_x86_msr_filter_changed)(vcpu);
9164 if (kvm_check_request(KVM_REQ_UPDATE_CPU_DIRTY_LOGGING, vcpu))
9165 static_call(kvm_x86_update_cpu_dirty_logging)(vcpu);
9168 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win ||
9169 kvm_xen_has_interrupt(vcpu)) {
9170 ++vcpu->stat.req_event;
9171 kvm_apic_accept_events(vcpu);
9172 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
9177 inject_pending_event(vcpu, &req_immediate_exit);
9179 static_call(kvm_x86_enable_irq_window)(vcpu);
9181 if (kvm_lapic_enabled(vcpu)) {
9182 update_cr8_intercept(vcpu);
9183 kvm_lapic_sync_to_vapic(vcpu);
9187 r = kvm_mmu_reload(vcpu);
9189 goto cancel_injection;
9194 static_call(kvm_x86_prepare_guest_switch)(vcpu);
9197 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
9198 * IPI are then delayed after guest entry, which ensures that they
9199 * result in virtual interrupt delivery.
9201 local_irq_disable();
9202 vcpu->mode = IN_GUEST_MODE;
9204 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9207 * 1) We should set ->mode before checking ->requests. Please see
9208 * the comment in kvm_vcpu_exiting_guest_mode().
9210 * 2) For APICv, we should set ->mode before checking PID.ON. This
9211 * pairs with the memory barrier implicit in pi_test_and_set_on
9212 * (see vmx_deliver_posted_interrupt).
9214 * 3) This also orders the write to mode from any reads to the page
9215 * tables done while the VCPU is running. Please see the comment
9216 * in kvm_flush_remote_tlbs.
9218 smp_mb__after_srcu_read_unlock();
9221 * This handles the case where a posted interrupt was
9222 * notified with kvm_vcpu_kick.
9224 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
9225 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9227 if (kvm_vcpu_exit_request(vcpu)) {
9228 vcpu->mode = OUTSIDE_GUEST_MODE;
9232 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9234 goto cancel_injection;
9237 if (req_immediate_exit) {
9238 kvm_make_request(KVM_REQ_EVENT, vcpu);
9239 static_call(kvm_x86_request_immediate_exit)(vcpu);
9242 fpregs_assert_state_consistent();
9243 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9244 switch_fpu_return();
9246 if (unlikely(vcpu->arch.switch_db_regs)) {
9248 set_debugreg(vcpu->arch.eff_db[0], 0);
9249 set_debugreg(vcpu->arch.eff_db[1], 1);
9250 set_debugreg(vcpu->arch.eff_db[2], 2);
9251 set_debugreg(vcpu->arch.eff_db[3], 3);
9252 set_debugreg(vcpu->arch.dr6, 6);
9253 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9257 exit_fastpath = static_call(kvm_x86_run)(vcpu);
9258 if (likely(exit_fastpath != EXIT_FASTPATH_REENTER_GUEST))
9261 if (unlikely(kvm_vcpu_exit_request(vcpu))) {
9262 exit_fastpath = EXIT_FASTPATH_EXIT_HANDLED;
9266 if (vcpu->arch.apicv_active)
9267 static_call(kvm_x86_sync_pir_to_irr)(vcpu);
9271 * Do this here before restoring debug registers on the host. And
9272 * since we do this before handling the vmexit, a DR access vmexit
9273 * can (a) read the correct value of the debug registers, (b) set
9274 * KVM_DEBUGREG_WONT_EXIT again.
9276 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
9277 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
9278 static_call(kvm_x86_sync_dirty_debug_regs)(vcpu);
9279 kvm_update_dr0123(vcpu);
9280 kvm_update_dr7(vcpu);
9281 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
9285 * If the guest has used debug registers, at least dr7
9286 * will be disabled while returning to the host.
9287 * If we don't have active breakpoints in the host, we don't
9288 * care about the messed up debug address registers. But if
9289 * we have some of them active, restore the old state.
9291 if (hw_breakpoint_active())
9292 hw_breakpoint_restore();
9294 vcpu->arch.last_vmentry_cpu = vcpu->cpu;
9295 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
9297 vcpu->mode = OUTSIDE_GUEST_MODE;
9300 static_call(kvm_x86_handle_exit_irqoff)(vcpu);
9303 * Consume any pending interrupts, including the possible source of
9304 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
9305 * An instruction is required after local_irq_enable() to fully unblock
9306 * interrupts on processors that implement an interrupt shadow, the
9307 * stat.exits increment will do nicely.
9309 kvm_before_interrupt(vcpu);
9312 local_irq_disable();
9313 kvm_after_interrupt(vcpu);
9315 if (lapic_in_kernel(vcpu)) {
9316 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
9317 if (delta != S64_MIN) {
9318 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
9319 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
9326 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9329 * Profile KVM exit RIPs:
9331 if (unlikely(prof_on == KVM_PROFILING)) {
9332 unsigned long rip = kvm_rip_read(vcpu);
9333 profile_hit(KVM_PROFILING, (void *)rip);
9336 if (unlikely(vcpu->arch.tsc_always_catchup))
9337 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9339 if (vcpu->arch.apic_attention)
9340 kvm_lapic_sync_from_vapic(vcpu);
9342 r = static_call(kvm_x86_handle_exit)(vcpu, exit_fastpath);
9346 if (req_immediate_exit)
9347 kvm_make_request(KVM_REQ_EVENT, vcpu);
9348 static_call(kvm_x86_cancel_injection)(vcpu);
9349 if (unlikely(vcpu->arch.apic_attention))
9350 kvm_lapic_sync_from_vapic(vcpu);
9355 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
9357 if (!kvm_arch_vcpu_runnable(vcpu) &&
9358 (!kvm_x86_ops.pre_block || static_call(kvm_x86_pre_block)(vcpu) == 0)) {
9359 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9360 kvm_vcpu_block(vcpu);
9361 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9363 if (kvm_x86_ops.post_block)
9364 static_call(kvm_x86_post_block)(vcpu);
9366 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
9370 kvm_apic_accept_events(vcpu);
9371 switch(vcpu->arch.mp_state) {
9372 case KVM_MP_STATE_HALTED:
9373 case KVM_MP_STATE_AP_RESET_HOLD:
9374 vcpu->arch.pv.pv_unhalted = false;
9375 vcpu->arch.mp_state =
9376 KVM_MP_STATE_RUNNABLE;
9378 case KVM_MP_STATE_RUNNABLE:
9379 vcpu->arch.apf.halted = false;
9381 case KVM_MP_STATE_INIT_RECEIVED:
9389 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
9391 if (is_guest_mode(vcpu))
9392 kvm_check_nested_events(vcpu);
9394 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
9395 !vcpu->arch.apf.halted);
9398 static int vcpu_run(struct kvm_vcpu *vcpu)
9401 struct kvm *kvm = vcpu->kvm;
9403 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9404 vcpu->arch.l1tf_flush_l1d = true;
9407 if (kvm_vcpu_running(vcpu)) {
9408 r = vcpu_enter_guest(vcpu);
9410 r = vcpu_block(kvm, vcpu);
9416 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
9417 if (kvm_cpu_has_pending_timer(vcpu))
9418 kvm_inject_pending_timer_irqs(vcpu);
9420 if (dm_request_for_irq_injection(vcpu) &&
9421 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
9423 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
9424 ++vcpu->stat.request_irq_exits;
9428 if (__xfer_to_guest_mode_work_pending()) {
9429 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9430 r = xfer_to_guest_mode_handle_work(vcpu);
9433 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
9437 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
9442 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
9446 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
9447 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
9448 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
9452 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
9454 BUG_ON(!vcpu->arch.pio.count);
9456 return complete_emulated_io(vcpu);
9460 * Implements the following, as a state machine:
9464 * for each mmio piece in the fragment
9472 * for each mmio piece in the fragment
9477 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
9479 struct kvm_run *run = vcpu->run;
9480 struct kvm_mmio_fragment *frag;
9483 BUG_ON(!vcpu->mmio_needed);
9485 /* Complete previous fragment */
9486 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
9487 len = min(8u, frag->len);
9488 if (!vcpu->mmio_is_write)
9489 memcpy(frag->data, run->mmio.data, len);
9491 if (frag->len <= 8) {
9492 /* Switch to the next fragment. */
9494 vcpu->mmio_cur_fragment++;
9496 /* Go forward to the next mmio piece. */
9502 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
9503 vcpu->mmio_needed = 0;
9505 /* FIXME: return into emulator if single-stepping. */
9506 if (vcpu->mmio_is_write)
9508 vcpu->mmio_read_completed = 1;
9509 return complete_emulated_io(vcpu);
9512 run->exit_reason = KVM_EXIT_MMIO;
9513 run->mmio.phys_addr = frag->gpa;
9514 if (vcpu->mmio_is_write)
9515 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
9516 run->mmio.len = min(8u, frag->len);
9517 run->mmio.is_write = vcpu->mmio_is_write;
9518 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
9522 static void kvm_save_current_fpu(struct fpu *fpu)
9525 * If the target FPU state is not resident in the CPU registers, just
9526 * memcpy() from current, else save CPU state directly to the target.
9528 if (test_thread_flag(TIF_NEED_FPU_LOAD))
9529 memcpy(&fpu->state, ¤t->thread.fpu.state,
9530 fpu_kernel_xstate_size);
9532 copy_fpregs_to_fpstate(fpu);
9535 /* Swap (qemu) user FPU context for the guest FPU context. */
9536 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
9540 kvm_save_current_fpu(vcpu->arch.user_fpu);
9543 * Guests with protected state can't have it set by the hypervisor,
9544 * so skip trying to set it.
9546 if (vcpu->arch.guest_fpu)
9547 /* PKRU is separately restored in kvm_x86_ops.run. */
9548 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
9549 ~XFEATURE_MASK_PKRU);
9551 fpregs_mark_activate();
9557 /* When vcpu_run ends, restore user space FPU context. */
9558 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
9563 * Guests with protected state can't have it read by the hypervisor,
9564 * so skip trying to save it.
9566 if (vcpu->arch.guest_fpu)
9567 kvm_save_current_fpu(vcpu->arch.guest_fpu);
9569 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
9571 fpregs_mark_activate();
9574 ++vcpu->stat.fpu_reload;
9578 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu)
9580 struct kvm_run *kvm_run = vcpu->run;
9584 kvm_sigset_activate(vcpu);
9586 kvm_load_guest_fpu(vcpu);
9588 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
9589 if (kvm_run->immediate_exit) {
9593 kvm_vcpu_block(vcpu);
9594 kvm_apic_accept_events(vcpu);
9595 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
9597 if (signal_pending(current)) {
9599 kvm_run->exit_reason = KVM_EXIT_INTR;
9600 ++vcpu->stat.signal_exits;
9605 if (kvm_run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
9610 if (kvm_run->kvm_dirty_regs) {
9611 r = sync_regs(vcpu);
9616 /* re-sync apic's tpr */
9617 if (!lapic_in_kernel(vcpu)) {
9618 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
9624 if (unlikely(vcpu->arch.complete_userspace_io)) {
9625 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
9626 vcpu->arch.complete_userspace_io = NULL;
9631 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
9633 if (kvm_run->immediate_exit)
9639 kvm_put_guest_fpu(vcpu);
9640 if (kvm_run->kvm_valid_regs)
9642 post_kvm_run_save(vcpu);
9643 kvm_sigset_deactivate(vcpu);
9649 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9651 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
9653 * We are here if userspace calls get_regs() in the middle of
9654 * instruction emulation. Registers state needs to be copied
9655 * back from emulation context to vcpu. Userspace shouldn't do
9656 * that usually, but some bad designed PV devices (vmware
9657 * backdoor interface) need this to work
9659 emulator_writeback_register_cache(vcpu->arch.emulate_ctxt);
9660 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9662 regs->rax = kvm_rax_read(vcpu);
9663 regs->rbx = kvm_rbx_read(vcpu);
9664 regs->rcx = kvm_rcx_read(vcpu);
9665 regs->rdx = kvm_rdx_read(vcpu);
9666 regs->rsi = kvm_rsi_read(vcpu);
9667 regs->rdi = kvm_rdi_read(vcpu);
9668 regs->rsp = kvm_rsp_read(vcpu);
9669 regs->rbp = kvm_rbp_read(vcpu);
9670 #ifdef CONFIG_X86_64
9671 regs->r8 = kvm_r8_read(vcpu);
9672 regs->r9 = kvm_r9_read(vcpu);
9673 regs->r10 = kvm_r10_read(vcpu);
9674 regs->r11 = kvm_r11_read(vcpu);
9675 regs->r12 = kvm_r12_read(vcpu);
9676 regs->r13 = kvm_r13_read(vcpu);
9677 regs->r14 = kvm_r14_read(vcpu);
9678 regs->r15 = kvm_r15_read(vcpu);
9681 regs->rip = kvm_rip_read(vcpu);
9682 regs->rflags = kvm_get_rflags(vcpu);
9685 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9688 __get_regs(vcpu, regs);
9693 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9695 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
9696 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
9698 kvm_rax_write(vcpu, regs->rax);
9699 kvm_rbx_write(vcpu, regs->rbx);
9700 kvm_rcx_write(vcpu, regs->rcx);
9701 kvm_rdx_write(vcpu, regs->rdx);
9702 kvm_rsi_write(vcpu, regs->rsi);
9703 kvm_rdi_write(vcpu, regs->rdi);
9704 kvm_rsp_write(vcpu, regs->rsp);
9705 kvm_rbp_write(vcpu, regs->rbp);
9706 #ifdef CONFIG_X86_64
9707 kvm_r8_write(vcpu, regs->r8);
9708 kvm_r9_write(vcpu, regs->r9);
9709 kvm_r10_write(vcpu, regs->r10);
9710 kvm_r11_write(vcpu, regs->r11);
9711 kvm_r12_write(vcpu, regs->r12);
9712 kvm_r13_write(vcpu, regs->r13);
9713 kvm_r14_write(vcpu, regs->r14);
9714 kvm_r15_write(vcpu, regs->r15);
9717 kvm_rip_write(vcpu, regs->rip);
9718 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
9720 vcpu->arch.exception.pending = false;
9722 kvm_make_request(KVM_REQ_EVENT, vcpu);
9725 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
9728 __set_regs(vcpu, regs);
9733 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
9735 struct kvm_segment cs;
9737 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9741 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
9743 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9747 if (vcpu->arch.guest_state_protected)
9748 goto skip_protected_regs;
9750 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9751 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9752 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9753 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9754 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9755 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9757 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9758 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9760 static_call(kvm_x86_get_idt)(vcpu, &dt);
9761 sregs->idt.limit = dt.size;
9762 sregs->idt.base = dt.address;
9763 static_call(kvm_x86_get_gdt)(vcpu, &dt);
9764 sregs->gdt.limit = dt.size;
9765 sregs->gdt.base = dt.address;
9767 sregs->cr2 = vcpu->arch.cr2;
9768 sregs->cr3 = kvm_read_cr3(vcpu);
9770 skip_protected_regs:
9771 sregs->cr0 = kvm_read_cr0(vcpu);
9772 sregs->cr4 = kvm_read_cr4(vcpu);
9773 sregs->cr8 = kvm_get_cr8(vcpu);
9774 sregs->efer = vcpu->arch.efer;
9775 sregs->apic_base = kvm_get_apic_base(vcpu);
9777 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
9779 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
9780 set_bit(vcpu->arch.interrupt.nr,
9781 (unsigned long *)sregs->interrupt_bitmap);
9784 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
9785 struct kvm_sregs *sregs)
9788 __get_sregs(vcpu, sregs);
9793 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
9794 struct kvm_mp_state *mp_state)
9797 if (kvm_mpx_supported())
9798 kvm_load_guest_fpu(vcpu);
9800 kvm_apic_accept_events(vcpu);
9801 if ((vcpu->arch.mp_state == KVM_MP_STATE_HALTED ||
9802 vcpu->arch.mp_state == KVM_MP_STATE_AP_RESET_HOLD) &&
9803 vcpu->arch.pv.pv_unhalted)
9804 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
9806 mp_state->mp_state = vcpu->arch.mp_state;
9808 if (kvm_mpx_supported())
9809 kvm_put_guest_fpu(vcpu);
9814 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
9815 struct kvm_mp_state *mp_state)
9821 if (!lapic_in_kernel(vcpu) &&
9822 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
9826 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
9827 * INIT state; latched init should be reported using
9828 * KVM_SET_VCPU_EVENTS, so reject it here.
9830 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
9831 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
9832 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
9835 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
9836 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
9837 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
9839 vcpu->arch.mp_state = mp_state->mp_state;
9840 kvm_make_request(KVM_REQ_EVENT, vcpu);
9848 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
9849 int reason, bool has_error_code, u32 error_code)
9851 struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
9854 init_emulate_ctxt(vcpu);
9856 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
9857 has_error_code, error_code);
9859 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
9860 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
9861 vcpu->run->internal.ndata = 0;
9865 kvm_rip_write(vcpu, ctxt->eip);
9866 kvm_set_rflags(vcpu, ctxt->eflags);
9869 EXPORT_SYMBOL_GPL(kvm_task_switch);
9871 static bool kvm_is_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9873 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
9875 * When EFER.LME and CR0.PG are set, the processor is in
9876 * 64-bit mode (though maybe in a 32-bit code segment).
9877 * CR4.PAE and EFER.LMA must be set.
9879 if (!(sregs->cr4 & X86_CR4_PAE) || !(sregs->efer & EFER_LMA))
9881 if (kvm_vcpu_is_illegal_gpa(vcpu, sregs->cr3))
9885 * Not in 64-bit mode: EFER.LMA is clear and the code
9886 * segment cannot be 64-bit.
9888 if (sregs->efer & EFER_LMA || sregs->cs.l)
9892 return kvm_is_valid_cr4(vcpu, sregs->cr4);
9895 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
9897 struct msr_data apic_base_msr;
9898 int mmu_reset_needed = 0;
9899 int pending_vec, max_bits, idx;
9903 if (!kvm_is_valid_sregs(vcpu, sregs))
9906 apic_base_msr.data = sregs->apic_base;
9907 apic_base_msr.host_initiated = true;
9908 if (kvm_set_apic_base(vcpu, &apic_base_msr))
9911 if (vcpu->arch.guest_state_protected)
9912 goto skip_protected_regs;
9914 dt.size = sregs->idt.limit;
9915 dt.address = sregs->idt.base;
9916 static_call(kvm_x86_set_idt)(vcpu, &dt);
9917 dt.size = sregs->gdt.limit;
9918 dt.address = sregs->gdt.base;
9919 static_call(kvm_x86_set_gdt)(vcpu, &dt);
9921 vcpu->arch.cr2 = sregs->cr2;
9922 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
9923 vcpu->arch.cr3 = sregs->cr3;
9924 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
9926 kvm_set_cr8(vcpu, sregs->cr8);
9928 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
9929 static_call(kvm_x86_set_efer)(vcpu, sregs->efer);
9931 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
9932 static_call(kvm_x86_set_cr0)(vcpu, sregs->cr0);
9933 vcpu->arch.cr0 = sregs->cr0;
9935 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
9936 static_call(kvm_x86_set_cr4)(vcpu, sregs->cr4);
9938 idx = srcu_read_lock(&vcpu->kvm->srcu);
9939 if (is_pae_paging(vcpu)) {
9940 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
9941 mmu_reset_needed = 1;
9943 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9945 if (mmu_reset_needed)
9946 kvm_mmu_reset_context(vcpu);
9948 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
9949 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
9950 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
9951 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
9952 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
9953 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
9955 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
9956 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
9958 update_cr8_intercept(vcpu);
9960 /* Older userspace won't unhalt the vcpu on reset. */
9961 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
9962 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
9964 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9966 skip_protected_regs:
9967 max_bits = KVM_NR_INTERRUPTS;
9968 pending_vec = find_first_bit(
9969 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
9970 if (pending_vec < max_bits) {
9971 kvm_queue_interrupt(vcpu, pending_vec, false);
9972 pr_debug("Set back pending irq %d\n", pending_vec);
9975 kvm_make_request(KVM_REQ_EVENT, vcpu);
9982 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
9983 struct kvm_sregs *sregs)
9988 ret = __set_sregs(vcpu, sregs);
9993 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9994 struct kvm_guest_debug *dbg)
9996 unsigned long rflags;
9999 if (vcpu->arch.guest_state_protected)
10004 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
10006 if (vcpu->arch.exception.pending)
10008 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
10009 kvm_queue_exception(vcpu, DB_VECTOR);
10011 kvm_queue_exception(vcpu, BP_VECTOR);
10015 * Read rflags as long as potentially injected trace flags are still
10018 rflags = kvm_get_rflags(vcpu);
10020 vcpu->guest_debug = dbg->control;
10021 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
10022 vcpu->guest_debug = 0;
10024 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
10025 for (i = 0; i < KVM_NR_DB_REGS; ++i)
10026 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
10027 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
10029 for (i = 0; i < KVM_NR_DB_REGS; i++)
10030 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
10032 kvm_update_dr7(vcpu);
10034 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10035 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
10036 get_segment_base(vcpu, VCPU_SREG_CS);
10039 * Trigger an rflags update that will inject or remove the trace
10042 kvm_set_rflags(vcpu, rflags);
10044 static_call(kvm_x86_update_exception_bitmap)(vcpu);
10054 * Translate a guest virtual address to a guest physical address.
10056 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
10057 struct kvm_translation *tr)
10059 unsigned long vaddr = tr->linear_address;
10065 idx = srcu_read_lock(&vcpu->kvm->srcu);
10066 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
10067 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10068 tr->physical_address = gpa;
10069 tr->valid = gpa != UNMAPPED_GVA;
10077 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10079 struct fxregs_state *fxsave;
10081 if (!vcpu->arch.guest_fpu)
10086 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10087 memcpy(fpu->fpr, fxsave->st_space, 128);
10088 fpu->fcw = fxsave->cwd;
10089 fpu->fsw = fxsave->swd;
10090 fpu->ftwx = fxsave->twd;
10091 fpu->last_opcode = fxsave->fop;
10092 fpu->last_ip = fxsave->rip;
10093 fpu->last_dp = fxsave->rdp;
10094 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
10100 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
10102 struct fxregs_state *fxsave;
10104 if (!vcpu->arch.guest_fpu)
10109 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
10111 memcpy(fxsave->st_space, fpu->fpr, 128);
10112 fxsave->cwd = fpu->fcw;
10113 fxsave->swd = fpu->fsw;
10114 fxsave->twd = fpu->ftwx;
10115 fxsave->fop = fpu->last_opcode;
10116 fxsave->rip = fpu->last_ip;
10117 fxsave->rdp = fpu->last_dp;
10118 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
10124 static void store_regs(struct kvm_vcpu *vcpu)
10126 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
10128 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
10129 __get_regs(vcpu, &vcpu->run->s.regs.regs);
10131 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
10132 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
10134 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
10135 kvm_vcpu_ioctl_x86_get_vcpu_events(
10136 vcpu, &vcpu->run->s.regs.events);
10139 static int sync_regs(struct kvm_vcpu *vcpu)
10141 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
10144 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
10145 __set_regs(vcpu, &vcpu->run->s.regs.regs);
10146 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
10148 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
10149 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
10151 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
10153 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
10154 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
10155 vcpu, &vcpu->run->s.regs.events))
10157 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
10163 static void fx_init(struct kvm_vcpu *vcpu)
10165 if (!vcpu->arch.guest_fpu)
10168 fpstate_init(&vcpu->arch.guest_fpu->state);
10169 if (boot_cpu_has(X86_FEATURE_XSAVES))
10170 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
10171 host_xcr0 | XSTATE_COMPACTION_ENABLED;
10174 * Ensure guest xcr0 is valid for loading
10176 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10178 vcpu->arch.cr0 |= X86_CR0_ET;
10181 void kvm_free_guest_fpu(struct kvm_vcpu *vcpu)
10183 if (vcpu->arch.guest_fpu) {
10184 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
10185 vcpu->arch.guest_fpu = NULL;
10188 EXPORT_SYMBOL_GPL(kvm_free_guest_fpu);
10190 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
10192 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
10193 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
10194 "guest TSC will not be reliable\n");
10199 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
10204 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
10205 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10207 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
10209 kvm_set_tsc_khz(vcpu, max_tsc_khz);
10211 r = kvm_mmu_create(vcpu);
10215 if (irqchip_in_kernel(vcpu->kvm)) {
10216 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
10218 goto fail_mmu_destroy;
10219 if (kvm_apicv_activated(vcpu->kvm))
10220 vcpu->arch.apicv_active = true;
10222 static_branch_inc(&kvm_has_noapic_vcpu);
10226 page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
10228 goto fail_free_lapic;
10229 vcpu->arch.pio_data = page_address(page);
10231 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
10232 GFP_KERNEL_ACCOUNT);
10233 if (!vcpu->arch.mce_banks)
10234 goto fail_free_pio_data;
10235 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
10237 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
10238 GFP_KERNEL_ACCOUNT))
10239 goto fail_free_mce_banks;
10241 if (!alloc_emulate_ctxt(vcpu))
10242 goto free_wbinvd_dirty_mask;
10244 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
10245 GFP_KERNEL_ACCOUNT);
10246 if (!vcpu->arch.user_fpu) {
10247 pr_err("kvm: failed to allocate userspace's fpu\n");
10248 goto free_emulate_ctxt;
10251 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
10252 GFP_KERNEL_ACCOUNT);
10253 if (!vcpu->arch.guest_fpu) {
10254 pr_err("kvm: failed to allocate vcpu's fpu\n");
10255 goto free_user_fpu;
10259 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
10260 vcpu->arch.reserved_gpa_bits = kvm_vcpu_reserved_gpa_bits_raw(vcpu);
10262 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
10264 kvm_async_pf_hash_reset(vcpu);
10265 kvm_pmu_init(vcpu);
10267 vcpu->arch.pending_external_vector = -1;
10268 vcpu->arch.preempted_in_kernel = false;
10270 r = static_call(kvm_x86_vcpu_create)(vcpu);
10272 goto free_guest_fpu;
10274 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
10275 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
10276 kvm_vcpu_mtrr_init(vcpu);
10278 kvm_vcpu_reset(vcpu, false);
10279 kvm_init_mmu(vcpu, false);
10284 kvm_free_guest_fpu(vcpu);
10286 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10288 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10289 free_wbinvd_dirty_mask:
10290 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10291 fail_free_mce_banks:
10292 kfree(vcpu->arch.mce_banks);
10293 fail_free_pio_data:
10294 free_page((unsigned long)vcpu->arch.pio_data);
10296 kvm_free_lapic(vcpu);
10298 kvm_mmu_destroy(vcpu);
10302 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
10304 struct kvm *kvm = vcpu->kvm;
10306 if (mutex_lock_killable(&vcpu->mutex))
10309 kvm_synchronize_tsc(vcpu, 0);
10312 /* poll control enabled by default */
10313 vcpu->arch.msr_kvm_poll_control = 1;
10315 mutex_unlock(&vcpu->mutex);
10317 if (kvmclock_periodic_sync && vcpu->vcpu_idx == 0)
10318 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
10319 KVMCLOCK_SYNC_PERIOD);
10322 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
10324 struct gfn_to_pfn_cache *cache = &vcpu->arch.st.cache;
10327 kvm_release_pfn(cache->pfn, cache->dirty, cache);
10329 kvmclock_reset(vcpu);
10331 static_call(kvm_x86_vcpu_free)(vcpu);
10333 kmem_cache_free(x86_emulator_cache, vcpu->arch.emulate_ctxt);
10334 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
10335 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
10336 kvm_free_guest_fpu(vcpu);
10338 kvm_hv_vcpu_uninit(vcpu);
10339 kvm_pmu_destroy(vcpu);
10340 kfree(vcpu->arch.mce_banks);
10341 kvm_free_lapic(vcpu);
10342 idx = srcu_read_lock(&vcpu->kvm->srcu);
10343 kvm_mmu_destroy(vcpu);
10344 srcu_read_unlock(&vcpu->kvm->srcu, idx);
10345 free_page((unsigned long)vcpu->arch.pio_data);
10346 kvfree(vcpu->arch.cpuid_entries);
10347 if (!lapic_in_kernel(vcpu))
10348 static_branch_dec(&kvm_has_noapic_vcpu);
10351 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
10353 kvm_lapic_reset(vcpu, init_event);
10355 vcpu->arch.hflags = 0;
10357 vcpu->arch.smi_pending = 0;
10358 vcpu->arch.smi_count = 0;
10359 atomic_set(&vcpu->arch.nmi_queued, 0);
10360 vcpu->arch.nmi_pending = 0;
10361 vcpu->arch.nmi_injected = false;
10362 kvm_clear_interrupt_queue(vcpu);
10363 kvm_clear_exception_queue(vcpu);
10365 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
10366 kvm_update_dr0123(vcpu);
10367 vcpu->arch.dr6 = DR6_ACTIVE_LOW;
10368 vcpu->arch.dr7 = DR7_FIXED_1;
10369 kvm_update_dr7(vcpu);
10371 vcpu->arch.cr2 = 0;
10373 kvm_make_request(KVM_REQ_EVENT, vcpu);
10374 vcpu->arch.apf.msr_en_val = 0;
10375 vcpu->arch.apf.msr_int_val = 0;
10376 vcpu->arch.st.msr_val = 0;
10378 kvmclock_reset(vcpu);
10380 kvm_clear_async_pf_completion_queue(vcpu);
10381 kvm_async_pf_hash_reset(vcpu);
10382 vcpu->arch.apf.halted = false;
10384 if (vcpu->arch.guest_fpu && kvm_mpx_supported()) {
10385 void *mpx_state_buffer;
10388 * To avoid have the INIT path from kvm_apic_has_events() that be
10389 * called with loaded FPU and does not let userspace fix the state.
10392 kvm_put_guest_fpu(vcpu);
10393 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10395 if (mpx_state_buffer)
10396 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
10397 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
10399 if (mpx_state_buffer)
10400 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
10402 kvm_load_guest_fpu(vcpu);
10406 kvm_pmu_reset(vcpu);
10407 vcpu->arch.smbase = 0x30000;
10409 vcpu->arch.msr_misc_features_enables = 0;
10411 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
10414 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
10415 vcpu->arch.regs_avail = ~0;
10416 vcpu->arch.regs_dirty = ~0;
10418 vcpu->arch.ia32_xss = 0;
10420 static_call(kvm_x86_vcpu_reset)(vcpu, init_event);
10423 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
10425 struct kvm_segment cs;
10427 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
10428 cs.selector = vector << 8;
10429 cs.base = vector << 12;
10430 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
10431 kvm_rip_write(vcpu, 0);
10433 EXPORT_SYMBOL_GPL(kvm_vcpu_deliver_sipi_vector);
10435 int kvm_arch_hardware_enable(void)
10438 struct kvm_vcpu *vcpu;
10443 bool stable, backwards_tsc = false;
10445 kvm_user_return_msr_cpu_online();
10446 ret = static_call(kvm_x86_hardware_enable)();
10450 local_tsc = rdtsc();
10451 stable = !kvm_check_tsc_unstable();
10452 list_for_each_entry(kvm, &vm_list, vm_list) {
10453 kvm_for_each_vcpu(i, vcpu, kvm) {
10454 if (!stable && vcpu->cpu == smp_processor_id())
10455 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
10456 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
10457 backwards_tsc = true;
10458 if (vcpu->arch.last_host_tsc > max_tsc)
10459 max_tsc = vcpu->arch.last_host_tsc;
10465 * Sometimes, even reliable TSCs go backwards. This happens on
10466 * platforms that reset TSC during suspend or hibernate actions, but
10467 * maintain synchronization. We must compensate. Fortunately, we can
10468 * detect that condition here, which happens early in CPU bringup,
10469 * before any KVM threads can be running. Unfortunately, we can't
10470 * bring the TSCs fully up to date with real time, as we aren't yet far
10471 * enough into CPU bringup that we know how much real time has actually
10472 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
10473 * variables that haven't been updated yet.
10475 * So we simply find the maximum observed TSC above, then record the
10476 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
10477 * the adjustment will be applied. Note that we accumulate
10478 * adjustments, in case multiple suspend cycles happen before some VCPU
10479 * gets a chance to run again. In the event that no KVM threads get a
10480 * chance to run, we will miss the entire elapsed period, as we'll have
10481 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
10482 * loose cycle time. This isn't too big a deal, since the loss will be
10483 * uniform across all VCPUs (not to mention the scenario is extremely
10484 * unlikely). It is possible that a second hibernate recovery happens
10485 * much faster than a first, causing the observed TSC here to be
10486 * smaller; this would require additional padding adjustment, which is
10487 * why we set last_host_tsc to the local tsc observed here.
10489 * N.B. - this code below runs only on platforms with reliable TSC,
10490 * as that is the only way backwards_tsc is set above. Also note
10491 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
10492 * have the same delta_cyc adjustment applied if backwards_tsc
10493 * is detected. Note further, this adjustment is only done once,
10494 * as we reset last_host_tsc on all VCPUs to stop this from being
10495 * called multiple times (one for each physical CPU bringup).
10497 * Platforms with unreliable TSCs don't have to deal with this, they
10498 * will be compensated by the logic in vcpu_load, which sets the TSC to
10499 * catchup mode. This will catchup all VCPUs to real time, but cannot
10500 * guarantee that they stay in perfect synchronization.
10502 if (backwards_tsc) {
10503 u64 delta_cyc = max_tsc - local_tsc;
10504 list_for_each_entry(kvm, &vm_list, vm_list) {
10505 kvm->arch.backwards_tsc_observed = true;
10506 kvm_for_each_vcpu(i, vcpu, kvm) {
10507 vcpu->arch.tsc_offset_adjustment += delta_cyc;
10508 vcpu->arch.last_host_tsc = local_tsc;
10509 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
10513 * We have to disable TSC offset matching.. if you were
10514 * booting a VM while issuing an S4 host suspend....
10515 * you may have some problem. Solving this issue is
10516 * left as an exercise to the reader.
10518 kvm->arch.last_tsc_nsec = 0;
10519 kvm->arch.last_tsc_write = 0;
10526 void kvm_arch_hardware_disable(void)
10528 static_call(kvm_x86_hardware_disable)();
10529 drop_user_return_notifiers();
10532 int kvm_arch_hardware_setup(void *opaque)
10534 struct kvm_x86_init_ops *ops = opaque;
10537 rdmsrl_safe(MSR_EFER, &host_efer);
10539 if (boot_cpu_has(X86_FEATURE_XSAVES))
10540 rdmsrl(MSR_IA32_XSS, host_xss);
10542 r = ops->hardware_setup();
10546 memcpy(&kvm_x86_ops, ops->runtime_ops, sizeof(kvm_x86_ops));
10547 kvm_ops_static_call_update();
10549 if (!kvm_cpu_cap_has(X86_FEATURE_XSAVES))
10552 #define __kvm_cpu_cap_has(UNUSED_, f) kvm_cpu_cap_has(f)
10553 cr4_reserved_bits = __cr4_reserved_bits(__kvm_cpu_cap_has, UNUSED_);
10554 #undef __kvm_cpu_cap_has
10556 if (kvm_has_tsc_control) {
10558 * Make sure the user can only configure tsc_khz values that
10559 * fit into a signed integer.
10560 * A min value is not calculated because it will always
10561 * be 1 on all machines.
10563 u64 max = min(0x7fffffffULL,
10564 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
10565 kvm_max_guest_tsc_khz = max;
10567 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
10570 kvm_init_msr_list();
10574 void kvm_arch_hardware_unsetup(void)
10576 static_call(kvm_x86_hardware_unsetup)();
10579 int kvm_arch_check_processor_compat(void *opaque)
10581 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
10582 struct kvm_x86_init_ops *ops = opaque;
10584 WARN_ON(!irqs_disabled());
10586 if (__cr4_reserved_bits(cpu_has, c) !=
10587 __cr4_reserved_bits(cpu_has, &boot_cpu_data))
10590 return ops->check_processor_compatibility();
10593 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
10595 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
10597 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
10599 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
10601 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
10604 __read_mostly DEFINE_STATIC_KEY_FALSE(kvm_has_noapic_vcpu);
10605 EXPORT_SYMBOL_GPL(kvm_has_noapic_vcpu);
10607 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
10609 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
10611 vcpu->arch.l1tf_flush_l1d = true;
10612 if (pmu->version && unlikely(pmu->event_count)) {
10613 pmu->need_cleanup = true;
10614 kvm_make_request(KVM_REQ_PMU, vcpu);
10616 static_call(kvm_x86_sched_in)(vcpu, cpu);
10619 void kvm_arch_free_vm(struct kvm *kvm)
10621 kfree(to_kvm_hv(kvm)->hv_pa_pg);
10626 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
10631 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
10632 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
10633 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
10634 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
10635 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
10636 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
10638 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
10639 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
10640 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
10641 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
10642 &kvm->arch.irq_sources_bitmap);
10644 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
10645 mutex_init(&kvm->arch.apic_map_lock);
10646 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
10648 kvm->arch.kvmclock_offset = -get_kvmclock_base_ns();
10649 pvclock_update_vm_gtod_copy(kvm);
10651 kvm->arch.guest_can_read_msr_platform_info = true;
10653 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
10654 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
10656 kvm_hv_init_vm(kvm);
10657 kvm_page_track_init(kvm);
10658 kvm_mmu_init_vm(kvm);
10660 return static_call(kvm_x86_vm_init)(kvm);
10663 int kvm_arch_post_init_vm(struct kvm *kvm)
10665 return kvm_mmu_post_init_vm(kvm);
10668 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
10671 kvm_mmu_unload(vcpu);
10675 static void kvm_free_vcpus(struct kvm *kvm)
10678 struct kvm_vcpu *vcpu;
10681 * Unpin any mmu pages first.
10683 kvm_for_each_vcpu(i, vcpu, kvm) {
10684 kvm_clear_async_pf_completion_queue(vcpu);
10685 kvm_unload_vcpu_mmu(vcpu);
10687 kvm_for_each_vcpu(i, vcpu, kvm)
10688 kvm_vcpu_destroy(vcpu);
10690 mutex_lock(&kvm->lock);
10691 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
10692 kvm->vcpus[i] = NULL;
10694 atomic_set(&kvm->online_vcpus, 0);
10695 mutex_unlock(&kvm->lock);
10698 void kvm_arch_sync_events(struct kvm *kvm)
10700 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
10701 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
10705 #define ERR_PTR_USR(e) ((void __user *)ERR_PTR(e))
10708 * __x86_set_memory_region: Setup KVM internal memory slot
10710 * @kvm: the kvm pointer to the VM.
10711 * @id: the slot ID to setup.
10712 * @gpa: the GPA to install the slot (unused when @size == 0).
10713 * @size: the size of the slot. Set to zero to uninstall a slot.
10715 * This function helps to setup a KVM internal memory slot. Specify
10716 * @size > 0 to install a new slot, while @size == 0 to uninstall a
10717 * slot. The return code can be one of the following:
10719 * HVA: on success (uninstall will return a bogus HVA)
10722 * The caller should always use IS_ERR() to check the return value
10723 * before use. Note, the KVM internal memory slots are guaranteed to
10724 * remain valid and unchanged until the VM is destroyed, i.e., the
10725 * GPA->HVA translation will not change. However, the HVA is a user
10726 * address, i.e. its accessibility is not guaranteed, and must be
10727 * accessed via __copy_{to,from}_user().
10729 void __user * __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa,
10733 unsigned long hva, old_npages;
10734 struct kvm_memslots *slots = kvm_memslots(kvm);
10735 struct kvm_memory_slot *slot;
10737 /* Called with kvm->slots_lock held. */
10738 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
10739 return ERR_PTR_USR(-EINVAL);
10741 slot = id_to_memslot(slots, id);
10743 if (slot && slot->npages)
10744 return ERR_PTR_USR(-EEXIST);
10747 * MAP_SHARED to prevent internal slot pages from being moved
10750 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
10751 MAP_SHARED | MAP_ANONYMOUS, 0);
10752 if (IS_ERR((void *)hva))
10753 return (void __user *)hva;
10755 if (!slot || !slot->npages)
10758 old_npages = slot->npages;
10759 hva = slot->userspace_addr;
10762 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
10763 struct kvm_userspace_memory_region m;
10765 m.slot = id | (i << 16);
10767 m.guest_phys_addr = gpa;
10768 m.userspace_addr = hva;
10769 m.memory_size = size;
10770 r = __kvm_set_memory_region(kvm, &m);
10772 return ERR_PTR_USR(r);
10776 vm_munmap(hva, old_npages * PAGE_SIZE);
10778 return (void __user *)hva;
10780 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
10782 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
10784 kvm_mmu_pre_destroy_vm(kvm);
10787 void kvm_arch_destroy_vm(struct kvm *kvm)
10789 if (current->mm == kvm->mm) {
10791 * Free memory regions allocated on behalf of userspace,
10792 * unless the the memory map has changed due to process exit
10795 mutex_lock(&kvm->slots_lock);
10796 __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
10798 __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
10800 __x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
10801 mutex_unlock(&kvm->slots_lock);
10803 static_call_cond(kvm_x86_vm_destroy)(kvm);
10804 kvm_free_msr_filter(srcu_dereference_check(kvm->arch.msr_filter, &kvm->srcu, 1));
10805 kvm_pic_destroy(kvm);
10806 kvm_ioapic_destroy(kvm);
10807 kvm_free_vcpus(kvm);
10808 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
10809 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
10810 kvm_mmu_uninit_vm(kvm);
10811 kvm_page_track_cleanup(kvm);
10812 kvm_xen_destroy_vm(kvm);
10813 kvm_hv_destroy_vm(kvm);
10816 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *slot)
10820 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10821 kvfree(slot->arch.rmap[i]);
10822 slot->arch.rmap[i] = NULL;
10827 kvfree(slot->arch.lpage_info[i - 1]);
10828 slot->arch.lpage_info[i - 1] = NULL;
10831 kvm_page_track_free_memslot(slot);
10834 static int kvm_alloc_memslot_metadata(struct kvm_memory_slot *slot,
10835 unsigned long npages)
10840 * Clear out the previous array pointers for the KVM_MR_MOVE case. The
10841 * old arrays will be freed by __kvm_set_memory_region() if installing
10842 * the new memslot is successful.
10844 memset(&slot->arch, 0, sizeof(slot->arch));
10846 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10847 struct kvm_lpage_info *linfo;
10848 unsigned long ugfn;
10852 lpages = gfn_to_index(slot->base_gfn + npages - 1,
10853 slot->base_gfn, level) + 1;
10855 slot->arch.rmap[i] =
10856 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
10857 GFP_KERNEL_ACCOUNT);
10858 if (!slot->arch.rmap[i])
10863 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
10867 slot->arch.lpage_info[i - 1] = linfo;
10869 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
10870 linfo[0].disallow_lpage = 1;
10871 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
10872 linfo[lpages - 1].disallow_lpage = 1;
10873 ugfn = slot->userspace_addr >> PAGE_SHIFT;
10875 * If the gfn and userspace address are not aligned wrt each
10876 * other, disable large page support for this slot.
10878 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1)) {
10881 for (j = 0; j < lpages; ++j)
10882 linfo[j].disallow_lpage = 1;
10886 if (kvm_page_track_create_memslot(slot, npages))
10892 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
10893 kvfree(slot->arch.rmap[i]);
10894 slot->arch.rmap[i] = NULL;
10898 kvfree(slot->arch.lpage_info[i - 1]);
10899 slot->arch.lpage_info[i - 1] = NULL;
10904 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
10906 struct kvm_vcpu *vcpu;
10910 * memslots->generation has been incremented.
10911 * mmio generation may have reached its maximum value.
10913 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
10915 /* Force re-initialization of steal_time cache */
10916 kvm_for_each_vcpu(i, vcpu, kvm)
10917 kvm_vcpu_kick(vcpu);
10920 int kvm_arch_prepare_memory_region(struct kvm *kvm,
10921 struct kvm_memory_slot *memslot,
10922 const struct kvm_userspace_memory_region *mem,
10923 enum kvm_mr_change change)
10925 if (change == KVM_MR_CREATE || change == KVM_MR_MOVE)
10926 return kvm_alloc_memslot_metadata(memslot,
10927 mem->memory_size >> PAGE_SHIFT);
10932 static void kvm_mmu_update_cpu_dirty_logging(struct kvm *kvm, bool enable)
10934 struct kvm_arch *ka = &kvm->arch;
10936 if (!kvm_x86_ops.cpu_dirty_log_size)
10939 if ((enable && ++ka->cpu_dirty_logging_count == 1) ||
10940 (!enable && --ka->cpu_dirty_logging_count == 0))
10941 kvm_make_all_cpus_request(kvm, KVM_REQ_UPDATE_CPU_DIRTY_LOGGING);
10943 WARN_ON_ONCE(ka->cpu_dirty_logging_count < 0);
10946 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
10947 struct kvm_memory_slot *old,
10948 struct kvm_memory_slot *new,
10949 enum kvm_mr_change change)
10951 bool log_dirty_pages = new->flags & KVM_MEM_LOG_DIRTY_PAGES;
10954 * Update CPU dirty logging if dirty logging is being toggled. This
10955 * applies to all operations.
10957 if ((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)
10958 kvm_mmu_update_cpu_dirty_logging(kvm, log_dirty_pages);
10961 * Nothing more to do for RO slots (which can't be dirtied and can't be
10962 * made writable) or CREATE/MOVE/DELETE of a slot.
10964 * For a memslot with dirty logging disabled:
10965 * CREATE: No dirty mappings will already exist.
10966 * MOVE/DELETE: The old mappings will already have been cleaned up by
10967 * kvm_arch_flush_shadow_memslot()
10969 * For a memslot with dirty logging enabled:
10970 * CREATE: No shadow pages exist, thus nothing to write-protect
10971 * and no dirty bits to clear.
10972 * MOVE/DELETE: The old mappings will already have been cleaned up by
10973 * kvm_arch_flush_shadow_memslot().
10975 if ((change != KVM_MR_FLAGS_ONLY) || (new->flags & KVM_MEM_READONLY))
10979 * READONLY and non-flags changes were filtered out above, and the only
10980 * other flag is LOG_DIRTY_PAGES, i.e. something is wrong if dirty
10981 * logging isn't being toggled on or off.
10983 if (WARN_ON_ONCE(!((old->flags ^ new->flags) & KVM_MEM_LOG_DIRTY_PAGES)))
10986 if (!log_dirty_pages) {
10988 * Dirty logging tracks sptes in 4k granularity, meaning that
10989 * large sptes have to be split. If live migration succeeds,
10990 * the guest in the source machine will be destroyed and large
10991 * sptes will be created in the destination. However, if the
10992 * guest continues to run in the source machine (for example if
10993 * live migration fails), small sptes will remain around and
10994 * cause bad performance.
10996 * Scan sptes if dirty logging has been stopped, dropping those
10997 * which can be collapsed into a single large-page spte. Later
10998 * page faults will create the large-page sptes.
11000 kvm_mmu_zap_collapsible_sptes(kvm, new);
11002 /* By default, write-protect everything to log writes. */
11003 int level = PG_LEVEL_4K;
11005 if (kvm_x86_ops.cpu_dirty_log_size) {
11007 * Clear all dirty bits, unless pages are treated as
11008 * dirty from the get-go.
11010 if (!kvm_dirty_log_manual_protect_and_init_set(kvm))
11011 kvm_mmu_slot_leaf_clear_dirty(kvm, new);
11014 * Write-protect large pages on write so that dirty
11015 * logging happens at 4k granularity. No need to
11016 * write-protect small SPTEs since write accesses are
11017 * logged by the CPU via dirty bits.
11019 level = PG_LEVEL_2M;
11020 } else if (kvm_dirty_log_manual_protect_and_init_set(kvm)) {
11022 * If we're with initial-all-set, we don't need
11023 * to write protect any small page because
11024 * they're reported as dirty already. However
11025 * we still need to write-protect huge pages
11026 * so that the page split can happen lazily on
11027 * the first write to the huge page.
11029 level = PG_LEVEL_2M;
11031 kvm_mmu_slot_remove_write_access(kvm, new, level);
11035 void kvm_arch_commit_memory_region(struct kvm *kvm,
11036 const struct kvm_userspace_memory_region *mem,
11037 struct kvm_memory_slot *old,
11038 const struct kvm_memory_slot *new,
11039 enum kvm_mr_change change)
11041 if (!kvm->arch.n_requested_mmu_pages)
11042 kvm_mmu_change_mmu_pages(kvm,
11043 kvm_mmu_calculate_default_mmu_pages(kvm));
11046 * FIXME: const-ify all uses of struct kvm_memory_slot.
11048 kvm_mmu_slot_apply_flags(kvm, old, (struct kvm_memory_slot *) new, change);
11050 /* Free the arrays associated with the old memslot. */
11051 if (change == KVM_MR_MOVE)
11052 kvm_arch_free_memslot(kvm, old);
11055 void kvm_arch_flush_shadow_all(struct kvm *kvm)
11057 kvm_mmu_zap_all(kvm);
11060 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
11061 struct kvm_memory_slot *slot)
11063 kvm_page_track_flush_slot(kvm, slot);
11066 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
11068 return (is_guest_mode(vcpu) &&
11069 kvm_x86_ops.guest_apic_has_interrupt &&
11070 static_call(kvm_x86_guest_apic_has_interrupt)(vcpu));
11073 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
11075 if (!list_empty_careful(&vcpu->async_pf.done))
11078 if (kvm_apic_has_events(vcpu))
11081 if (vcpu->arch.pv.pv_unhalted)
11084 if (vcpu->arch.exception.pending)
11087 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11088 (vcpu->arch.nmi_pending &&
11089 static_call(kvm_x86_nmi_allowed)(vcpu, false)))
11092 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
11093 (vcpu->arch.smi_pending &&
11094 static_call(kvm_x86_smi_allowed)(vcpu, false)))
11097 if (kvm_arch_interrupt_allowed(vcpu) &&
11098 (kvm_cpu_has_interrupt(vcpu) ||
11099 kvm_guest_apic_has_interrupt(vcpu)))
11102 if (kvm_hv_has_stimer_pending(vcpu))
11105 if (is_guest_mode(vcpu) &&
11106 kvm_x86_ops.nested_ops->hv_timer_pending &&
11107 kvm_x86_ops.nested_ops->hv_timer_pending(vcpu))
11113 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
11115 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
11118 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
11120 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
11123 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
11124 kvm_test_request(KVM_REQ_SMI, vcpu) ||
11125 kvm_test_request(KVM_REQ_EVENT, vcpu))
11128 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11134 bool kvm_arch_dy_has_pending_interrupt(struct kvm_vcpu *vcpu)
11136 if (vcpu->arch.apicv_active && static_call(kvm_x86_dy_apicv_has_pending_interrupt)(vcpu))
11142 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
11144 return vcpu->arch.preempted_in_kernel;
11147 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
11149 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
11152 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
11154 return static_call(kvm_x86_interrupt_allowed)(vcpu, false);
11157 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
11159 /* Can't read the RIP when guest state is protected, just return 0 */
11160 if (vcpu->arch.guest_state_protected)
11163 if (is_64_bit_mode(vcpu))
11164 return kvm_rip_read(vcpu);
11165 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
11166 kvm_rip_read(vcpu));
11168 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
11170 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
11172 return kvm_get_linear_rip(vcpu) == linear_rip;
11174 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
11176 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
11178 unsigned long rflags;
11180 rflags = static_call(kvm_x86_get_rflags)(vcpu);
11181 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
11182 rflags &= ~X86_EFLAGS_TF;
11185 EXPORT_SYMBOL_GPL(kvm_get_rflags);
11187 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11189 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
11190 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
11191 rflags |= X86_EFLAGS_TF;
11192 static_call(kvm_x86_set_rflags)(vcpu, rflags);
11195 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
11197 __kvm_set_rflags(vcpu, rflags);
11198 kvm_make_request(KVM_REQ_EVENT, vcpu);
11200 EXPORT_SYMBOL_GPL(kvm_set_rflags);
11202 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
11206 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
11210 r = kvm_mmu_reload(vcpu);
11214 if (!vcpu->arch.mmu->direct_map &&
11215 work->arch.cr3 != vcpu->arch.mmu->get_guest_pgd(vcpu))
11218 kvm_mmu_do_page_fault(vcpu, work->cr2_or_gpa, 0, true);
11221 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
11223 BUILD_BUG_ON(!is_power_of_2(ASYNC_PF_PER_VCPU));
11225 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
11228 static inline u32 kvm_async_pf_next_probe(u32 key)
11230 return (key + 1) & (ASYNC_PF_PER_VCPU - 1);
11233 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11235 u32 key = kvm_async_pf_hash_fn(gfn);
11237 while (vcpu->arch.apf.gfns[key] != ~0)
11238 key = kvm_async_pf_next_probe(key);
11240 vcpu->arch.apf.gfns[key] = gfn;
11243 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
11246 u32 key = kvm_async_pf_hash_fn(gfn);
11248 for (i = 0; i < ASYNC_PF_PER_VCPU &&
11249 (vcpu->arch.apf.gfns[key] != gfn &&
11250 vcpu->arch.apf.gfns[key] != ~0); i++)
11251 key = kvm_async_pf_next_probe(key);
11256 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11258 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
11261 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
11265 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
11267 if (WARN_ON_ONCE(vcpu->arch.apf.gfns[i] != gfn))
11271 vcpu->arch.apf.gfns[i] = ~0;
11273 j = kvm_async_pf_next_probe(j);
11274 if (vcpu->arch.apf.gfns[j] == ~0)
11276 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
11278 * k lies cyclically in ]i,j]
11280 * |....j i.k.| or |.k..j i...|
11282 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
11283 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
11288 static inline int apf_put_user_notpresent(struct kvm_vcpu *vcpu)
11290 u32 reason = KVM_PV_REASON_PAGE_NOT_PRESENT;
11292 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &reason,
11296 static inline int apf_put_user_ready(struct kvm_vcpu *vcpu, u32 token)
11298 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11300 return kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11301 &token, offset, sizeof(token));
11304 static inline bool apf_pageready_slot_free(struct kvm_vcpu *vcpu)
11306 unsigned int offset = offsetof(struct kvm_vcpu_pv_apf_data, token);
11309 if (kvm_read_guest_offset_cached(vcpu->kvm, &vcpu->arch.apf.data,
11310 &val, offset, sizeof(val)))
11316 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
11318 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
11321 if (!kvm_pv_async_pf_enabled(vcpu) ||
11322 (vcpu->arch.apf.send_user_only && static_call(kvm_x86_get_cpl)(vcpu) == 0))
11328 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
11330 if (unlikely(!lapic_in_kernel(vcpu) ||
11331 kvm_event_needs_reinjection(vcpu) ||
11332 vcpu->arch.exception.pending))
11335 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
11339 * If interrupts are off we cannot even use an artificial
11342 return kvm_arch_interrupt_allowed(vcpu);
11345 bool kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
11346 struct kvm_async_pf *work)
11348 struct x86_exception fault;
11350 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
11351 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
11353 if (kvm_can_deliver_async_pf(vcpu) &&
11354 !apf_put_user_notpresent(vcpu)) {
11355 fault.vector = PF_VECTOR;
11356 fault.error_code_valid = true;
11357 fault.error_code = 0;
11358 fault.nested_page_fault = false;
11359 fault.address = work->arch.token;
11360 fault.async_page_fault = true;
11361 kvm_inject_page_fault(vcpu, &fault);
11365 * It is not possible to deliver a paravirtualized asynchronous
11366 * page fault, but putting the guest in an artificial halt state
11367 * can be beneficial nevertheless: if an interrupt arrives, we
11368 * can deliver it timely and perhaps the guest will schedule
11369 * another process. When the instruction that triggered a page
11370 * fault is retried, hopefully the page will be ready in the host.
11372 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
11377 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
11378 struct kvm_async_pf *work)
11380 struct kvm_lapic_irq irq = {
11381 .delivery_mode = APIC_DM_FIXED,
11382 .vector = vcpu->arch.apf.vec
11385 if (work->wakeup_all)
11386 work->arch.token = ~0; /* broadcast wakeup */
11388 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
11389 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
11391 if ((work->wakeup_all || work->notpresent_injected) &&
11392 kvm_pv_async_pf_enabled(vcpu) &&
11393 !apf_put_user_ready(vcpu, work->arch.token)) {
11394 vcpu->arch.apf.pageready_pending = true;
11395 kvm_apic_set_irq(vcpu, &irq, NULL);
11398 vcpu->arch.apf.halted = false;
11399 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11402 void kvm_arch_async_page_present_queued(struct kvm_vcpu *vcpu)
11404 kvm_make_request(KVM_REQ_APF_READY, vcpu);
11405 if (!vcpu->arch.apf.pageready_pending)
11406 kvm_vcpu_kick(vcpu);
11409 bool kvm_arch_can_dequeue_async_page_present(struct kvm_vcpu *vcpu)
11411 if (!kvm_pv_async_pf_enabled(vcpu))
11414 return apf_pageready_slot_free(vcpu);
11417 void kvm_arch_start_assignment(struct kvm *kvm)
11419 atomic_inc(&kvm->arch.assigned_device_count);
11421 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
11423 void kvm_arch_end_assignment(struct kvm *kvm)
11425 atomic_dec(&kvm->arch.assigned_device_count);
11427 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
11429 bool kvm_arch_has_assigned_device(struct kvm *kvm)
11431 return atomic_read(&kvm->arch.assigned_device_count);
11433 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
11435 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
11437 atomic_inc(&kvm->arch.noncoherent_dma_count);
11439 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
11441 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
11443 atomic_dec(&kvm->arch.noncoherent_dma_count);
11445 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
11447 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
11449 return atomic_read(&kvm->arch.noncoherent_dma_count);
11451 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
11453 bool kvm_arch_has_irq_bypass(void)
11458 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
11459 struct irq_bypass_producer *prod)
11461 struct kvm_kernel_irqfd *irqfd =
11462 container_of(cons, struct kvm_kernel_irqfd, consumer);
11465 irqfd->producer = prod;
11466 kvm_arch_start_assignment(irqfd->kvm);
11467 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm,
11468 prod->irq, irqfd->gsi, 1);
11471 kvm_arch_end_assignment(irqfd->kvm);
11476 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
11477 struct irq_bypass_producer *prod)
11480 struct kvm_kernel_irqfd *irqfd =
11481 container_of(cons, struct kvm_kernel_irqfd, consumer);
11483 WARN_ON(irqfd->producer != prod);
11484 irqfd->producer = NULL;
11487 * When producer of consumer is unregistered, we change back to
11488 * remapped mode, so we can re-use the current implementation
11489 * when the irq is masked/disabled or the consumer side (KVM
11490 * int this case doesn't want to receive the interrupts.
11492 ret = static_call(kvm_x86_update_pi_irte)(irqfd->kvm, prod->irq, irqfd->gsi, 0);
11494 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
11495 " fails: %d\n", irqfd->consumer.token, ret);
11497 kvm_arch_end_assignment(irqfd->kvm);
11500 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
11501 uint32_t guest_irq, bool set)
11503 return static_call(kvm_x86_update_pi_irte)(kvm, host_irq, guest_irq, set);
11506 bool kvm_vector_hashing_enabled(void)
11508 return vector_hashing;
11511 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
11513 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
11515 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
11518 int kvm_spec_ctrl_test_value(u64 value)
11521 * test that setting IA32_SPEC_CTRL to given value
11522 * is allowed by the host processor
11526 unsigned long flags;
11529 local_irq_save(flags);
11531 if (rdmsrl_safe(MSR_IA32_SPEC_CTRL, &saved_value))
11533 else if (wrmsrl_safe(MSR_IA32_SPEC_CTRL, value))
11536 wrmsrl(MSR_IA32_SPEC_CTRL, saved_value);
11538 local_irq_restore(flags);
11542 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_test_value);
11544 void kvm_fixup_and_inject_pf_error(struct kvm_vcpu *vcpu, gva_t gva, u16 error_code)
11546 struct x86_exception fault;
11547 u32 access = error_code &
11548 (PFERR_WRITE_MASK | PFERR_FETCH_MASK | PFERR_USER_MASK);
11550 if (!(error_code & PFERR_PRESENT_MASK) ||
11551 vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, &fault) != UNMAPPED_GVA) {
11553 * If vcpu->arch.walk_mmu->gva_to_gpa succeeded, the page
11554 * tables probably do not match the TLB. Just proceed
11555 * with the error code that the processor gave.
11557 fault.vector = PF_VECTOR;
11558 fault.error_code_valid = true;
11559 fault.error_code = error_code;
11560 fault.nested_page_fault = false;
11561 fault.address = gva;
11563 vcpu->arch.walk_mmu->inject_page_fault(vcpu, &fault);
11565 EXPORT_SYMBOL_GPL(kvm_fixup_and_inject_pf_error);
11568 * Handles kvm_read/write_guest_virt*() result and either injects #PF or returns
11569 * KVM_EXIT_INTERNAL_ERROR for cases not currently handled by KVM. Return value
11570 * indicates whether exit to userspace is needed.
11572 int kvm_handle_memory_failure(struct kvm_vcpu *vcpu, int r,
11573 struct x86_exception *e)
11575 if (r == X86EMUL_PROPAGATE_FAULT) {
11576 kvm_inject_emulated_page_fault(vcpu, e);
11581 * In case kvm_read/write_guest_virt*() failed with X86EMUL_IO_NEEDED
11582 * while handling a VMX instruction KVM could've handled the request
11583 * correctly by exiting to userspace and performing I/O but there
11584 * doesn't seem to be a real use-case behind such requests, just return
11585 * KVM_EXIT_INTERNAL_ERROR for now.
11587 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
11588 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
11589 vcpu->run->internal.ndata = 0;
11593 EXPORT_SYMBOL_GPL(kvm_handle_memory_failure);
11595 int kvm_handle_invpcid(struct kvm_vcpu *vcpu, unsigned long type, gva_t gva)
11598 struct x86_exception e;
11600 unsigned long roots_to_free = 0;
11607 r = kvm_read_guest_virt(vcpu, gva, &operand, sizeof(operand), &e);
11608 if (r != X86EMUL_CONTINUE)
11609 return kvm_handle_memory_failure(vcpu, r, &e);
11611 if (operand.pcid >> 12 != 0) {
11612 kvm_inject_gp(vcpu, 0);
11616 pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
11619 case INVPCID_TYPE_INDIV_ADDR:
11620 if ((!pcid_enabled && (operand.pcid != 0)) ||
11621 is_noncanonical_address(operand.gla, vcpu)) {
11622 kvm_inject_gp(vcpu, 0);
11625 kvm_mmu_invpcid_gva(vcpu, operand.gla, operand.pcid);
11626 return kvm_skip_emulated_instruction(vcpu);
11628 case INVPCID_TYPE_SINGLE_CTXT:
11629 if (!pcid_enabled && (operand.pcid != 0)) {
11630 kvm_inject_gp(vcpu, 0);
11634 if (kvm_get_active_pcid(vcpu) == operand.pcid) {
11635 kvm_mmu_sync_roots(vcpu);
11636 kvm_make_request(KVM_REQ_TLB_FLUSH_CURRENT, vcpu);
11639 for (i = 0; i < KVM_MMU_NUM_PREV_ROOTS; i++)
11640 if (kvm_get_pcid(vcpu, vcpu->arch.mmu->prev_roots[i].pgd)
11642 roots_to_free |= KVM_MMU_ROOT_PREVIOUS(i);
11644 kvm_mmu_free_roots(vcpu, vcpu->arch.mmu, roots_to_free);
11646 * If neither the current cr3 nor any of the prev_roots use the
11647 * given PCID, then nothing needs to be done here because a
11648 * resync will happen anyway before switching to any other CR3.
11651 return kvm_skip_emulated_instruction(vcpu);
11653 case INVPCID_TYPE_ALL_NON_GLOBAL:
11655 * Currently, KVM doesn't mark global entries in the shadow
11656 * page tables, so a non-global flush just degenerates to a
11657 * global flush. If needed, we could optimize this later by
11658 * keeping track of global entries in shadow page tables.
11662 case INVPCID_TYPE_ALL_INCL_GLOBAL:
11663 kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
11664 return kvm_skip_emulated_instruction(vcpu);
11667 BUG(); /* We have already checked above that type <= 3 */
11670 EXPORT_SYMBOL_GPL(kvm_handle_invpcid);
11672 static int complete_sev_es_emulated_mmio(struct kvm_vcpu *vcpu)
11674 struct kvm_run *run = vcpu->run;
11675 struct kvm_mmio_fragment *frag;
11678 BUG_ON(!vcpu->mmio_needed);
11680 /* Complete previous fragment */
11681 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
11682 len = min(8u, frag->len);
11683 if (!vcpu->mmio_is_write)
11684 memcpy(frag->data, run->mmio.data, len);
11686 if (frag->len <= 8) {
11687 /* Switch to the next fragment. */
11689 vcpu->mmio_cur_fragment++;
11691 /* Go forward to the next mmio piece. */
11697 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
11698 vcpu->mmio_needed = 0;
11700 // VMG change, at this point, we're always done
11701 // RIP has already been advanced
11705 // More MMIO is needed
11706 run->mmio.phys_addr = frag->gpa;
11707 run->mmio.len = min(8u, frag->len);
11708 run->mmio.is_write = vcpu->mmio_is_write;
11709 if (run->mmio.is_write)
11710 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
11711 run->exit_reason = KVM_EXIT_MMIO;
11713 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11718 int kvm_sev_es_mmio_write(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11722 struct kvm_mmio_fragment *frag;
11727 handled = write_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11728 if (handled == bytes)
11735 /*TODO: Check if need to increment number of frags */
11736 frag = vcpu->mmio_fragments;
11737 vcpu->mmio_nr_fragments = 1;
11742 vcpu->mmio_needed = 1;
11743 vcpu->mmio_cur_fragment = 0;
11745 vcpu->run->mmio.phys_addr = gpa;
11746 vcpu->run->mmio.len = min(8u, frag->len);
11747 vcpu->run->mmio.is_write = 1;
11748 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
11749 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11751 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11755 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_write);
11757 int kvm_sev_es_mmio_read(struct kvm_vcpu *vcpu, gpa_t gpa, unsigned int bytes,
11761 struct kvm_mmio_fragment *frag;
11766 handled = read_emultor.read_write_mmio(vcpu, gpa, bytes, data);
11767 if (handled == bytes)
11774 /*TODO: Check if need to increment number of frags */
11775 frag = vcpu->mmio_fragments;
11776 vcpu->mmio_nr_fragments = 1;
11781 vcpu->mmio_needed = 1;
11782 vcpu->mmio_cur_fragment = 0;
11784 vcpu->run->mmio.phys_addr = gpa;
11785 vcpu->run->mmio.len = min(8u, frag->len);
11786 vcpu->run->mmio.is_write = 0;
11787 vcpu->run->exit_reason = KVM_EXIT_MMIO;
11789 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_mmio;
11793 EXPORT_SYMBOL_GPL(kvm_sev_es_mmio_read);
11795 static int complete_sev_es_emulated_ins(struct kvm_vcpu *vcpu)
11797 memcpy(vcpu->arch.guest_ins_data, vcpu->arch.pio_data,
11798 vcpu->arch.pio.count * vcpu->arch.pio.size);
11799 vcpu->arch.pio.count = 0;
11804 static int kvm_sev_es_outs(struct kvm_vcpu *vcpu, unsigned int size,
11805 unsigned int port, void *data, unsigned int count)
11809 ret = emulator_pio_out_emulated(vcpu->arch.emulate_ctxt, size, port,
11814 vcpu->arch.pio.count = 0;
11819 static int kvm_sev_es_ins(struct kvm_vcpu *vcpu, unsigned int size,
11820 unsigned int port, void *data, unsigned int count)
11824 ret = emulator_pio_in_emulated(vcpu->arch.emulate_ctxt, size, port,
11827 vcpu->arch.pio.count = 0;
11829 vcpu->arch.guest_ins_data = data;
11830 vcpu->arch.complete_userspace_io = complete_sev_es_emulated_ins;
11836 int kvm_sev_es_string_io(struct kvm_vcpu *vcpu, unsigned int size,
11837 unsigned int port, void *data, unsigned int count,
11840 return in ? kvm_sev_es_ins(vcpu, size, port, data, count)
11841 : kvm_sev_es_outs(vcpu, size, port, data, count);
11843 EXPORT_SYMBOL_GPL(kvm_sev_es_string_io);
11845 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_entry);
11846 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
11847 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
11848 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
11849 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
11850 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
11851 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
11852 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
11853 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
11854 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
11855 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
11856 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
11857 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
11858 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
11859 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
11860 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
11861 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
11862 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
11863 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
11864 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
11865 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);
11866 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_ga_log);
11867 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_apicv_update_request);
11868 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_enter);
11869 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_exit);
11870 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_enter);
11871 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_vmgexit_msr_protocol_exit);