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>
24 #include "kvm_cache_regs.h"
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
34 #include <linux/vmalloc.h>
35 #include <linux/export.h>
36 #include <linux/moduleparam.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <linux/kvm_irqfd.h>
52 #include <linux/irqbypass.h>
53 #include <linux/sched/stat.h>
54 #include <linux/sched/isolation.h>
55 #include <linux/mem_encrypt.h>
57 #include <trace/events/kvm.h>
59 #include <asm/debugreg.h>
63 #include <linux/kernel_stat.h>
64 #include <asm/fpu/internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
67 #include <asm/irq_remapping.h>
68 #include <asm/mshyperv.h>
69 #include <asm/hypervisor.h>
70 #include <asm/intel_pt.h>
71 #include <asm/emulate_prefix.h>
72 #include <clocksource/hyperv_timer.h>
74 #define CREATE_TRACE_POINTS
77 #define MAX_IO_MSRS 256
78 #define KVM_MAX_MCE_BANKS 32
79 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
80 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
82 #define emul_to_vcpu(ctxt) \
83 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
86 * - enable syscall per default because its emulated by KVM
87 * - enable LME and LMA per default on 64 bit KVM
91 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
93 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
96 static u64 __read_mostly cr4_reserved_bits = CR4_RESERVED_BITS;
98 #define VM_STAT(x, ...) offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__
99 #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__
101 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
102 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
104 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
105 static void process_nmi(struct kvm_vcpu *vcpu);
106 static void enter_smm(struct kvm_vcpu *vcpu);
107 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
108 static void store_regs(struct kvm_vcpu *vcpu);
109 static int sync_regs(struct kvm_vcpu *vcpu);
111 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
112 EXPORT_SYMBOL_GPL(kvm_x86_ops);
114 static bool __read_mostly ignore_msrs = 0;
115 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
117 static bool __read_mostly report_ignored_msrs = true;
118 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
120 unsigned int min_timer_period_us = 200;
121 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
123 static bool __read_mostly kvmclock_periodic_sync = true;
124 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
126 bool __read_mostly kvm_has_tsc_control;
127 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
128 u32 __read_mostly kvm_max_guest_tsc_khz;
129 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
130 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
131 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
132 u64 __read_mostly kvm_max_tsc_scaling_ratio;
133 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
134 u64 __read_mostly kvm_default_tsc_scaling_ratio;
135 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
137 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
138 static u32 __read_mostly tsc_tolerance_ppm = 250;
139 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
142 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
143 * adaptive tuning starting from default advancment of 1000ns. '0' disables
144 * advancement entirely. Any other value is used as-is and disables adaptive
145 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
147 static int __read_mostly lapic_timer_advance_ns = -1;
148 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
150 static bool __read_mostly vector_hashing = true;
151 module_param(vector_hashing, bool, S_IRUGO);
153 bool __read_mostly enable_vmware_backdoor = false;
154 module_param(enable_vmware_backdoor, bool, S_IRUGO);
155 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
157 static bool __read_mostly force_emulation_prefix = false;
158 module_param(force_emulation_prefix, bool, S_IRUGO);
160 int __read_mostly pi_inject_timer = -1;
161 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
163 #define KVM_NR_SHARED_MSRS 16
165 struct kvm_shared_msrs_global {
167 u32 msrs[KVM_NR_SHARED_MSRS];
170 struct kvm_shared_msrs {
171 struct user_return_notifier urn;
173 struct kvm_shared_msr_values {
176 } values[KVM_NR_SHARED_MSRS];
179 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
180 static struct kvm_shared_msrs __percpu *shared_msrs;
182 static u64 __read_mostly host_xss;
184 struct kvm_stats_debugfs_item debugfs_entries[] = {
185 { "pf_fixed", VCPU_STAT(pf_fixed) },
186 { "pf_guest", VCPU_STAT(pf_guest) },
187 { "tlb_flush", VCPU_STAT(tlb_flush) },
188 { "invlpg", VCPU_STAT(invlpg) },
189 { "exits", VCPU_STAT(exits) },
190 { "io_exits", VCPU_STAT(io_exits) },
191 { "mmio_exits", VCPU_STAT(mmio_exits) },
192 { "signal_exits", VCPU_STAT(signal_exits) },
193 { "irq_window", VCPU_STAT(irq_window_exits) },
194 { "nmi_window", VCPU_STAT(nmi_window_exits) },
195 { "halt_exits", VCPU_STAT(halt_exits) },
196 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
197 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
198 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
199 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
200 { "hypercalls", VCPU_STAT(hypercalls) },
201 { "request_irq", VCPU_STAT(request_irq_exits) },
202 { "irq_exits", VCPU_STAT(irq_exits) },
203 { "host_state_reload", VCPU_STAT(host_state_reload) },
204 { "fpu_reload", VCPU_STAT(fpu_reload) },
205 { "insn_emulation", VCPU_STAT(insn_emulation) },
206 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
207 { "irq_injections", VCPU_STAT(irq_injections) },
208 { "nmi_injections", VCPU_STAT(nmi_injections) },
209 { "req_event", VCPU_STAT(req_event) },
210 { "l1d_flush", VCPU_STAT(l1d_flush) },
211 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
212 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
213 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
214 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
215 { "mmu_flooded", VM_STAT(mmu_flooded) },
216 { "mmu_recycled", VM_STAT(mmu_recycled) },
217 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
218 { "mmu_unsync", VM_STAT(mmu_unsync) },
219 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
220 { "largepages", VM_STAT(lpages, .mode = 0444) },
221 { "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) },
222 { "max_mmu_page_hash_collisions",
223 VM_STAT(max_mmu_page_hash_collisions) },
227 u64 __read_mostly host_xcr0;
229 struct kmem_cache *x86_fpu_cache;
230 EXPORT_SYMBOL_GPL(x86_fpu_cache);
232 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
234 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
237 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
238 vcpu->arch.apf.gfns[i] = ~0;
241 static void kvm_on_user_return(struct user_return_notifier *urn)
244 struct kvm_shared_msrs *locals
245 = container_of(urn, struct kvm_shared_msrs, urn);
246 struct kvm_shared_msr_values *values;
250 * Disabling irqs at this point since the following code could be
251 * interrupted and executed through kvm_arch_hardware_disable()
253 local_irq_save(flags);
254 if (locals->registered) {
255 locals->registered = false;
256 user_return_notifier_unregister(urn);
258 local_irq_restore(flags);
259 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
260 values = &locals->values[slot];
261 if (values->host != values->curr) {
262 wrmsrl(shared_msrs_global.msrs[slot], values->host);
263 values->curr = values->host;
268 void kvm_define_shared_msr(unsigned slot, u32 msr)
270 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
271 shared_msrs_global.msrs[slot] = msr;
272 if (slot >= shared_msrs_global.nr)
273 shared_msrs_global.nr = slot + 1;
275 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
277 static void kvm_shared_msr_cpu_online(void)
279 unsigned int cpu = smp_processor_id();
280 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
284 for (i = 0; i < shared_msrs_global.nr; ++i) {
285 rdmsrl_safe(shared_msrs_global.msrs[i], &value);
286 smsr->values[i].host = value;
287 smsr->values[i].curr = value;
291 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
293 unsigned int cpu = smp_processor_id();
294 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
297 value = (value & mask) | (smsr->values[slot].host & ~mask);
298 if (value == smsr->values[slot].curr)
300 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
304 smsr->values[slot].curr = value;
305 if (!smsr->registered) {
306 smsr->urn.on_user_return = kvm_on_user_return;
307 user_return_notifier_register(&smsr->urn);
308 smsr->registered = true;
312 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
314 static void drop_user_return_notifiers(void)
316 unsigned int cpu = smp_processor_id();
317 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
319 if (smsr->registered)
320 kvm_on_user_return(&smsr->urn);
323 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
325 return vcpu->arch.apic_base;
327 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
329 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
331 return kvm_apic_mode(kvm_get_apic_base(vcpu));
333 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
335 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
337 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
338 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
339 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
340 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
342 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
344 if (!msr_info->host_initiated) {
345 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
347 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
351 kvm_lapic_set_base(vcpu, msr_info->data);
354 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
356 asmlinkage __visible void kvm_spurious_fault(void)
358 /* Fault while not rebooting. We want the trace. */
359 BUG_ON(!kvm_rebooting);
361 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
363 #define EXCPT_BENIGN 0
364 #define EXCPT_CONTRIBUTORY 1
367 static int exception_class(int vector)
377 return EXCPT_CONTRIBUTORY;
384 #define EXCPT_FAULT 0
386 #define EXCPT_ABORT 2
387 #define EXCPT_INTERRUPT 3
389 static int exception_type(int vector)
393 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
394 return EXCPT_INTERRUPT;
398 /* #DB is trap, as instruction watchpoints are handled elsewhere */
399 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
402 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
405 /* Reserved exceptions will result in fault */
409 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
411 unsigned nr = vcpu->arch.exception.nr;
412 bool has_payload = vcpu->arch.exception.has_payload;
413 unsigned long payload = vcpu->arch.exception.payload;
421 * "Certain debug exceptions may clear bit 0-3. The
422 * remaining contents of the DR6 register are never
423 * cleared by the processor".
425 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
427 * DR6.RTM is set by all #DB exceptions that don't clear it.
429 vcpu->arch.dr6 |= DR6_RTM;
430 vcpu->arch.dr6 |= payload;
432 * Bit 16 should be set in the payload whenever the #DB
433 * exception should clear DR6.RTM. This makes the payload
434 * compatible with the pending debug exceptions under VMX.
435 * Though not currently documented in the SDM, this also
436 * makes the payload compatible with the exit qualification
437 * for #DB exceptions under VMX.
439 vcpu->arch.dr6 ^= payload & DR6_RTM;
442 vcpu->arch.cr2 = payload;
446 vcpu->arch.exception.has_payload = false;
447 vcpu->arch.exception.payload = 0;
449 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
451 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
452 unsigned nr, bool has_error, u32 error_code,
453 bool has_payload, unsigned long payload, bool reinject)
458 kvm_make_request(KVM_REQ_EVENT, vcpu);
460 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
462 if (has_error && !is_protmode(vcpu))
466 * On vmentry, vcpu->arch.exception.pending is only
467 * true if an event injection was blocked by
468 * nested_run_pending. In that case, however,
469 * vcpu_enter_guest requests an immediate exit,
470 * and the guest shouldn't proceed far enough to
473 WARN_ON_ONCE(vcpu->arch.exception.pending);
474 vcpu->arch.exception.injected = true;
475 if (WARN_ON_ONCE(has_payload)) {
477 * A reinjected event has already
478 * delivered its payload.
484 vcpu->arch.exception.pending = true;
485 vcpu->arch.exception.injected = false;
487 vcpu->arch.exception.has_error_code = has_error;
488 vcpu->arch.exception.nr = nr;
489 vcpu->arch.exception.error_code = error_code;
490 vcpu->arch.exception.has_payload = has_payload;
491 vcpu->arch.exception.payload = payload;
493 * In guest mode, payload delivery should be deferred,
494 * so that the L1 hypervisor can intercept #PF before
495 * CR2 is modified (or intercept #DB before DR6 is
496 * modified under nVMX). However, for ABI
497 * compatibility with KVM_GET_VCPU_EVENTS and
498 * KVM_SET_VCPU_EVENTS, we can't delay payload
499 * delivery unless userspace has enabled this
500 * functionality via the per-VM capability,
501 * KVM_CAP_EXCEPTION_PAYLOAD.
503 if (!vcpu->kvm->arch.exception_payload_enabled ||
504 !is_guest_mode(vcpu))
505 kvm_deliver_exception_payload(vcpu);
509 /* to check exception */
510 prev_nr = vcpu->arch.exception.nr;
511 if (prev_nr == DF_VECTOR) {
512 /* triple fault -> shutdown */
513 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
516 class1 = exception_class(prev_nr);
517 class2 = exception_class(nr);
518 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
519 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
521 * Generate double fault per SDM Table 5-5. Set
522 * exception.pending = true so that the double fault
523 * can trigger a nested vmexit.
525 vcpu->arch.exception.pending = true;
526 vcpu->arch.exception.injected = false;
527 vcpu->arch.exception.has_error_code = true;
528 vcpu->arch.exception.nr = DF_VECTOR;
529 vcpu->arch.exception.error_code = 0;
530 vcpu->arch.exception.has_payload = false;
531 vcpu->arch.exception.payload = 0;
533 /* replace previous exception with a new one in a hope
534 that instruction re-execution will regenerate lost
539 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
541 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
543 EXPORT_SYMBOL_GPL(kvm_queue_exception);
545 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
547 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
549 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
551 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
552 unsigned long payload)
554 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
557 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
558 u32 error_code, unsigned long payload)
560 kvm_multiple_exception(vcpu, nr, true, error_code,
561 true, payload, false);
564 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
567 kvm_inject_gp(vcpu, 0);
569 return kvm_skip_emulated_instruction(vcpu);
573 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
575 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
577 ++vcpu->stat.pf_guest;
578 vcpu->arch.exception.nested_apf =
579 is_guest_mode(vcpu) && fault->async_page_fault;
580 if (vcpu->arch.exception.nested_apf) {
581 vcpu->arch.apf.nested_apf_token = fault->address;
582 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
584 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
588 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
590 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
592 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
593 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
595 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
597 return fault->nested_page_fault;
600 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
602 atomic_inc(&vcpu->arch.nmi_queued);
603 kvm_make_request(KVM_REQ_NMI, vcpu);
605 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
607 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
609 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
611 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
613 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
615 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
617 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
620 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
621 * a #GP and return false.
623 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
625 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
627 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
630 EXPORT_SYMBOL_GPL(kvm_require_cpl);
632 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
634 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
637 kvm_queue_exception(vcpu, UD_VECTOR);
640 EXPORT_SYMBOL_GPL(kvm_require_dr);
643 * This function will be used to read from the physical memory of the currently
644 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
645 * can read from guest physical or from the guest's guest physical memory.
647 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
648 gfn_t ngfn, void *data, int offset, int len,
651 struct x86_exception exception;
655 ngpa = gfn_to_gpa(ngfn);
656 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
657 if (real_gfn == UNMAPPED_GVA)
660 real_gfn = gpa_to_gfn(real_gfn);
662 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
664 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
666 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
667 void *data, int offset, int len, u32 access)
669 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
670 data, offset, len, access);
673 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
675 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
680 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
682 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
684 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
685 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
688 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
690 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
691 offset * sizeof(u64), sizeof(pdpte),
692 PFERR_USER_MASK|PFERR_WRITE_MASK);
697 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
698 if ((pdpte[i] & PT_PRESENT_MASK) &&
699 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
706 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
707 kvm_register_mark_dirty(vcpu, VCPU_EXREG_PDPTR);
713 EXPORT_SYMBOL_GPL(load_pdptrs);
715 bool pdptrs_changed(struct kvm_vcpu *vcpu)
717 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
722 if (!is_pae_paging(vcpu))
725 if (!kvm_register_is_available(vcpu, VCPU_EXREG_PDPTR))
728 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
729 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
730 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
731 PFERR_USER_MASK | PFERR_WRITE_MASK);
735 return memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
737 EXPORT_SYMBOL_GPL(pdptrs_changed);
739 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
741 unsigned long old_cr0 = kvm_read_cr0(vcpu);
742 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
747 if (cr0 & 0xffffffff00000000UL)
751 cr0 &= ~CR0_RESERVED_BITS;
753 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
756 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
759 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
761 if ((vcpu->arch.efer & EFER_LME)) {
766 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
771 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
776 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
779 kvm_x86_ops->set_cr0(vcpu, cr0);
781 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
782 kvm_clear_async_pf_completion_queue(vcpu);
783 kvm_async_pf_hash_reset(vcpu);
786 if ((cr0 ^ old_cr0) & update_bits)
787 kvm_mmu_reset_context(vcpu);
789 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
790 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
791 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
792 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
796 EXPORT_SYMBOL_GPL(kvm_set_cr0);
798 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
800 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
802 EXPORT_SYMBOL_GPL(kvm_lmsw);
804 void kvm_load_guest_xsave_state(struct kvm_vcpu *vcpu)
806 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
808 if (vcpu->arch.xcr0 != host_xcr0)
809 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
811 if (vcpu->arch.xsaves_enabled &&
812 vcpu->arch.ia32_xss != host_xss)
813 wrmsrl(MSR_IA32_XSS, vcpu->arch.ia32_xss);
816 EXPORT_SYMBOL_GPL(kvm_load_guest_xsave_state);
818 void kvm_load_host_xsave_state(struct kvm_vcpu *vcpu)
820 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE)) {
822 if (vcpu->arch.xcr0 != host_xcr0)
823 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
825 if (vcpu->arch.xsaves_enabled &&
826 vcpu->arch.ia32_xss != host_xss)
827 wrmsrl(MSR_IA32_XSS, host_xss);
831 EXPORT_SYMBOL_GPL(kvm_load_host_xsave_state);
833 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
836 u64 old_xcr0 = vcpu->arch.xcr0;
839 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
840 if (index != XCR_XFEATURE_ENABLED_MASK)
842 if (!(xcr0 & XFEATURE_MASK_FP))
844 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
848 * Do not allow the guest to set bits that we do not support
849 * saving. However, xcr0 bit 0 is always set, even if the
850 * emulated CPU does not support XSAVE (see fx_init).
852 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
853 if (xcr0 & ~valid_bits)
856 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
857 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
860 if (xcr0 & XFEATURE_MASK_AVX512) {
861 if (!(xcr0 & XFEATURE_MASK_YMM))
863 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
866 vcpu->arch.xcr0 = xcr0;
868 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
869 kvm_update_cpuid(vcpu);
873 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
875 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
876 __kvm_set_xcr(vcpu, index, xcr)) {
877 kvm_inject_gp(vcpu, 0);
882 EXPORT_SYMBOL_GPL(kvm_set_xcr);
884 #define __cr4_reserved_bits(__cpu_has, __c) \
886 u64 __reserved_bits = CR4_RESERVED_BITS; \
888 if (!__cpu_has(__c, X86_FEATURE_XSAVE)) \
889 __reserved_bits |= X86_CR4_OSXSAVE; \
890 if (!__cpu_has(__c, X86_FEATURE_SMEP)) \
891 __reserved_bits |= X86_CR4_SMEP; \
892 if (!__cpu_has(__c, X86_FEATURE_SMAP)) \
893 __reserved_bits |= X86_CR4_SMAP; \
894 if (!__cpu_has(__c, X86_FEATURE_FSGSBASE)) \
895 __reserved_bits |= X86_CR4_FSGSBASE; \
896 if (!__cpu_has(__c, X86_FEATURE_PKU)) \
897 __reserved_bits |= X86_CR4_PKE; \
898 if (!__cpu_has(__c, X86_FEATURE_LA57)) \
899 __reserved_bits |= X86_CR4_LA57; \
903 static u64 kvm_host_cr4_reserved_bits(struct cpuinfo_x86 *c)
905 u64 reserved_bits = __cr4_reserved_bits(cpu_has, c);
907 if (cpuid_ecx(0x7) & feature_bit(LA57))
908 reserved_bits &= ~X86_CR4_LA57;
910 if (kvm_x86_ops->umip_emulated())
911 reserved_bits &= ~X86_CR4_UMIP;
913 return reserved_bits;
916 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
918 if (cr4 & cr4_reserved_bits)
921 if (cr4 & __cr4_reserved_bits(guest_cpuid_has, vcpu))
927 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
929 unsigned long old_cr4 = kvm_read_cr4(vcpu);
930 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
931 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
933 if (kvm_valid_cr4(vcpu, cr4))
936 if (is_long_mode(vcpu)) {
937 if (!(cr4 & X86_CR4_PAE))
939 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
940 && ((cr4 ^ old_cr4) & pdptr_bits)
941 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
945 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
946 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
949 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
950 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
954 if (kvm_x86_ops->set_cr4(vcpu, cr4))
957 if (((cr4 ^ old_cr4) & pdptr_bits) ||
958 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
959 kvm_mmu_reset_context(vcpu);
961 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
962 kvm_update_cpuid(vcpu);
966 EXPORT_SYMBOL_GPL(kvm_set_cr4);
968 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
970 bool skip_tlb_flush = false;
972 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
975 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
976 cr3 &= ~X86_CR3_PCID_NOFLUSH;
980 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
981 if (!skip_tlb_flush) {
982 kvm_mmu_sync_roots(vcpu);
983 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
988 if (is_long_mode(vcpu) &&
989 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
991 else if (is_pae_paging(vcpu) &&
992 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
995 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
996 vcpu->arch.cr3 = cr3;
997 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
1001 EXPORT_SYMBOL_GPL(kvm_set_cr3);
1003 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
1005 if (cr8 & CR8_RESERVED_BITS)
1007 if (lapic_in_kernel(vcpu))
1008 kvm_lapic_set_tpr(vcpu, cr8);
1010 vcpu->arch.cr8 = cr8;
1013 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1015 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1017 if (lapic_in_kernel(vcpu))
1018 return kvm_lapic_get_cr8(vcpu);
1020 return vcpu->arch.cr8;
1022 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1024 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1028 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1029 for (i = 0; i < KVM_NR_DB_REGS; i++)
1030 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1031 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1035 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1037 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1038 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1041 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1045 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1046 dr7 = vcpu->arch.guest_debug_dr7;
1048 dr7 = vcpu->arch.dr7;
1049 kvm_x86_ops->set_dr7(vcpu, dr7);
1050 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1051 if (dr7 & DR7_BP_EN_MASK)
1052 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1055 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1057 u64 fixed = DR6_FIXED_1;
1059 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1064 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1066 size_t size = ARRAY_SIZE(vcpu->arch.db);
1070 vcpu->arch.db[array_index_nospec(dr, size)] = val;
1071 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1072 vcpu->arch.eff_db[dr] = val;
1077 if (val & 0xffffffff00000000ULL)
1078 return -1; /* #GP */
1079 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1080 kvm_update_dr6(vcpu);
1085 if (val & 0xffffffff00000000ULL)
1086 return -1; /* #GP */
1087 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1088 kvm_update_dr7(vcpu);
1095 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1097 if (__kvm_set_dr(vcpu, dr, val)) {
1098 kvm_inject_gp(vcpu, 0);
1103 EXPORT_SYMBOL_GPL(kvm_set_dr);
1105 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1107 size_t size = ARRAY_SIZE(vcpu->arch.db);
1111 *val = vcpu->arch.db[array_index_nospec(dr, size)];
1116 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1117 *val = vcpu->arch.dr6;
1119 *val = kvm_x86_ops->get_dr6(vcpu);
1124 *val = vcpu->arch.dr7;
1129 EXPORT_SYMBOL_GPL(kvm_get_dr);
1131 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1133 u32 ecx = kvm_rcx_read(vcpu);
1137 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1140 kvm_rax_write(vcpu, (u32)data);
1141 kvm_rdx_write(vcpu, data >> 32);
1144 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1147 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1148 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1150 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1151 * extract the supported MSRs from the related const lists.
1152 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1153 * capabilities of the host cpu. This capabilities test skips MSRs that are
1154 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1155 * may depend on host virtualization features rather than host cpu features.
1158 static const u32 msrs_to_save_all[] = {
1159 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1161 #ifdef CONFIG_X86_64
1162 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1164 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1165 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1167 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1168 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1169 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1170 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1171 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1172 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1173 MSR_IA32_UMWAIT_CONTROL,
1175 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1176 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1177 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1178 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1179 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1180 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1181 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1182 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1183 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1184 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1185 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1186 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1187 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1188 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1189 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1190 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1191 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1192 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1193 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1194 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1195 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1196 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1199 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1200 static unsigned num_msrs_to_save;
1202 static const u32 emulated_msrs_all[] = {
1203 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1204 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1205 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1206 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1207 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1208 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1209 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1211 HV_X64_MSR_VP_INDEX,
1212 HV_X64_MSR_VP_RUNTIME,
1213 HV_X64_MSR_SCONTROL,
1214 HV_X64_MSR_STIMER0_CONFIG,
1215 HV_X64_MSR_VP_ASSIST_PAGE,
1216 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1217 HV_X64_MSR_TSC_EMULATION_STATUS,
1219 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1222 MSR_IA32_TSC_ADJUST,
1223 MSR_IA32_TSCDEADLINE,
1224 MSR_IA32_ARCH_CAPABILITIES,
1225 MSR_IA32_MISC_ENABLE,
1226 MSR_IA32_MCG_STATUS,
1228 MSR_IA32_MCG_EXT_CTL,
1232 MSR_MISC_FEATURES_ENABLES,
1233 MSR_AMD64_VIRT_SPEC_CTRL,
1238 * The following list leaves out MSRs whose values are determined
1239 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1240 * We always support the "true" VMX control MSRs, even if the host
1241 * processor does not, so I am putting these registers here rather
1242 * than in msrs_to_save_all.
1245 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1246 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1247 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1248 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1250 MSR_IA32_VMX_CR0_FIXED0,
1251 MSR_IA32_VMX_CR4_FIXED0,
1252 MSR_IA32_VMX_VMCS_ENUM,
1253 MSR_IA32_VMX_PROCBASED_CTLS2,
1254 MSR_IA32_VMX_EPT_VPID_CAP,
1255 MSR_IA32_VMX_VMFUNC,
1258 MSR_KVM_POLL_CONTROL,
1261 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1262 static unsigned num_emulated_msrs;
1265 * List of msr numbers which are used to expose MSR-based features that
1266 * can be used by a hypervisor to validate requested CPU features.
1268 static const u32 msr_based_features_all[] = {
1270 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1271 MSR_IA32_VMX_PINBASED_CTLS,
1272 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1273 MSR_IA32_VMX_PROCBASED_CTLS,
1274 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1275 MSR_IA32_VMX_EXIT_CTLS,
1276 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1277 MSR_IA32_VMX_ENTRY_CTLS,
1279 MSR_IA32_VMX_CR0_FIXED0,
1280 MSR_IA32_VMX_CR0_FIXED1,
1281 MSR_IA32_VMX_CR4_FIXED0,
1282 MSR_IA32_VMX_CR4_FIXED1,
1283 MSR_IA32_VMX_VMCS_ENUM,
1284 MSR_IA32_VMX_PROCBASED_CTLS2,
1285 MSR_IA32_VMX_EPT_VPID_CAP,
1286 MSR_IA32_VMX_VMFUNC,
1290 MSR_IA32_ARCH_CAPABILITIES,
1293 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1294 static unsigned int num_msr_based_features;
1296 static u64 kvm_get_arch_capabilities(void)
1300 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1301 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1304 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1305 * the nested hypervisor runs with NX huge pages. If it is not,
1306 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1307 * L1 guests, so it need not worry about its own (L2) guests.
1309 data |= ARCH_CAP_PSCHANGE_MC_NO;
1312 * If we're doing cache flushes (either "always" or "cond")
1313 * we will do one whenever the guest does a vmlaunch/vmresume.
1314 * If an outer hypervisor is doing the cache flush for us
1315 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1316 * capability to the guest too, and if EPT is disabled we're not
1317 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1318 * require a nested hypervisor to do a flush of its own.
1320 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1321 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1323 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1324 data |= ARCH_CAP_RDCL_NO;
1325 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1326 data |= ARCH_CAP_SSB_NO;
1327 if (!boot_cpu_has_bug(X86_BUG_MDS))
1328 data |= ARCH_CAP_MDS_NO;
1331 * On TAA affected systems:
1332 * - nothing to do if TSX is disabled on the host.
1333 * - we emulate TSX_CTRL if present on the host.
1334 * This lets the guest use VERW to clear CPU buffers.
1336 if (!boot_cpu_has(X86_FEATURE_RTM))
1337 data &= ~(ARCH_CAP_TAA_NO | ARCH_CAP_TSX_CTRL_MSR);
1338 else if (!boot_cpu_has_bug(X86_BUG_TAA))
1339 data |= ARCH_CAP_TAA_NO;
1344 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1346 switch (msr->index) {
1347 case MSR_IA32_ARCH_CAPABILITIES:
1348 msr->data = kvm_get_arch_capabilities();
1350 case MSR_IA32_UCODE_REV:
1351 rdmsrl_safe(msr->index, &msr->data);
1354 if (kvm_x86_ops->get_msr_feature(msr))
1360 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1362 struct kvm_msr_entry msr;
1366 r = kvm_get_msr_feature(&msr);
1375 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1377 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1380 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1383 if (efer & (EFER_LME | EFER_LMA) &&
1384 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1387 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1393 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1395 if (efer & efer_reserved_bits)
1398 return __kvm_valid_efer(vcpu, efer);
1400 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1402 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1404 u64 old_efer = vcpu->arch.efer;
1405 u64 efer = msr_info->data;
1407 if (efer & efer_reserved_bits)
1410 if (!msr_info->host_initiated) {
1411 if (!__kvm_valid_efer(vcpu, efer))
1414 if (is_paging(vcpu) &&
1415 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1420 efer |= vcpu->arch.efer & EFER_LMA;
1422 kvm_x86_ops->set_efer(vcpu, efer);
1424 /* Update reserved bits */
1425 if ((efer ^ old_efer) & EFER_NX)
1426 kvm_mmu_reset_context(vcpu);
1431 void kvm_enable_efer_bits(u64 mask)
1433 efer_reserved_bits &= ~mask;
1435 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1438 * Write @data into the MSR specified by @index. Select MSR specific fault
1439 * checks are bypassed if @host_initiated is %true.
1440 * Returns 0 on success, non-0 otherwise.
1441 * Assumes vcpu_load() was already called.
1443 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1444 bool host_initiated)
1446 struct msr_data msr;
1451 case MSR_KERNEL_GS_BASE:
1454 if (is_noncanonical_address(data, vcpu))
1457 case MSR_IA32_SYSENTER_EIP:
1458 case MSR_IA32_SYSENTER_ESP:
1460 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1461 * non-canonical address is written on Intel but not on
1462 * AMD (which ignores the top 32-bits, because it does
1463 * not implement 64-bit SYSENTER).
1465 * 64-bit code should hence be able to write a non-canonical
1466 * value on AMD. Making the address canonical ensures that
1467 * vmentry does not fail on Intel after writing a non-canonical
1468 * value, and that something deterministic happens if the guest
1469 * invokes 64-bit SYSENTER.
1471 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1476 msr.host_initiated = host_initiated;
1478 return kvm_x86_ops->set_msr(vcpu, &msr);
1482 * Read the MSR specified by @index into @data. Select MSR specific fault
1483 * checks are bypassed if @host_initiated is %true.
1484 * Returns 0 on success, non-0 otherwise.
1485 * Assumes vcpu_load() was already called.
1487 int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1488 bool host_initiated)
1490 struct msr_data msr;
1494 msr.host_initiated = host_initiated;
1496 ret = kvm_x86_ops->get_msr(vcpu, &msr);
1502 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1504 return __kvm_get_msr(vcpu, index, data, false);
1506 EXPORT_SYMBOL_GPL(kvm_get_msr);
1508 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1510 return __kvm_set_msr(vcpu, index, data, false);
1512 EXPORT_SYMBOL_GPL(kvm_set_msr);
1514 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1516 u32 ecx = kvm_rcx_read(vcpu);
1519 if (kvm_get_msr(vcpu, ecx, &data)) {
1520 trace_kvm_msr_read_ex(ecx);
1521 kvm_inject_gp(vcpu, 0);
1525 trace_kvm_msr_read(ecx, data);
1527 kvm_rax_write(vcpu, data & -1u);
1528 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1529 return kvm_skip_emulated_instruction(vcpu);
1531 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1533 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1535 u32 ecx = kvm_rcx_read(vcpu);
1536 u64 data = kvm_read_edx_eax(vcpu);
1538 if (kvm_set_msr(vcpu, ecx, data)) {
1539 trace_kvm_msr_write_ex(ecx, data);
1540 kvm_inject_gp(vcpu, 0);
1544 trace_kvm_msr_write(ecx, data);
1545 return kvm_skip_emulated_instruction(vcpu);
1547 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1550 * The fast path for frequent and performance sensitive wrmsr emulation,
1551 * i.e. the sending of IPI, sending IPI early in the VM-Exit flow reduces
1552 * the latency of virtual IPI by avoiding the expensive bits of transitioning
1553 * from guest to host, e.g. reacquiring KVM's SRCU lock. In contrast to the
1554 * other cases which must be called after interrupts are enabled on the host.
1556 static int handle_fastpath_set_x2apic_icr_irqoff(struct kvm_vcpu *vcpu, u64 data)
1558 if (lapic_in_kernel(vcpu) && apic_x2apic_mode(vcpu->arch.apic) &&
1559 ((data & APIC_DEST_MASK) == APIC_DEST_PHYSICAL) &&
1560 ((data & APIC_MODE_MASK) == APIC_DM_FIXED)) {
1562 kvm_lapic_set_reg(vcpu->arch.apic, APIC_ICR2, (u32)(data >> 32));
1563 return kvm_lapic_reg_write(vcpu->arch.apic, APIC_ICR, (u32)data);
1569 enum exit_fastpath_completion handle_fastpath_set_msr_irqoff(struct kvm_vcpu *vcpu)
1571 u32 msr = kvm_rcx_read(vcpu);
1572 u64 data = kvm_read_edx_eax(vcpu);
1576 case APIC_BASE_MSR + (APIC_ICR >> 4):
1577 ret = handle_fastpath_set_x2apic_icr_irqoff(vcpu, data);
1580 return EXIT_FASTPATH_NONE;
1584 trace_kvm_msr_write(msr, data);
1585 return EXIT_FASTPATH_SKIP_EMUL_INS;
1588 return EXIT_FASTPATH_NONE;
1590 EXPORT_SYMBOL_GPL(handle_fastpath_set_msr_irqoff);
1593 * Adapt set_msr() to msr_io()'s calling convention
1595 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1597 return __kvm_get_msr(vcpu, index, data, true);
1600 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1602 return __kvm_set_msr(vcpu, index, *data, true);
1605 #ifdef CONFIG_X86_64
1606 struct pvclock_clock {
1614 struct pvclock_gtod_data {
1617 struct pvclock_clock clock; /* extract of a clocksource struct */
1618 struct pvclock_clock raw_clock; /* extract of a clocksource struct */
1624 u64 monotonic_raw_nsec;
1627 static struct pvclock_gtod_data pvclock_gtod_data;
1629 static void update_pvclock_gtod(struct timekeeper *tk)
1631 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1632 u64 boot_ns, boot_ns_raw;
1634 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1635 boot_ns_raw = ktime_to_ns(ktime_add(tk->tkr_raw.base, tk->offs_boot));
1637 write_seqcount_begin(&vdata->seq);
1639 /* copy pvclock gtod data */
1640 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1641 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1642 vdata->clock.mask = tk->tkr_mono.mask;
1643 vdata->clock.mult = tk->tkr_mono.mult;
1644 vdata->clock.shift = tk->tkr_mono.shift;
1646 vdata->raw_clock.vclock_mode = tk->tkr_raw.clock->archdata.vclock_mode;
1647 vdata->raw_clock.cycle_last = tk->tkr_raw.cycle_last;
1648 vdata->raw_clock.mask = tk->tkr_raw.mask;
1649 vdata->raw_clock.mult = tk->tkr_raw.mult;
1650 vdata->raw_clock.shift = tk->tkr_raw.shift;
1652 vdata->boot_ns = boot_ns;
1653 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1655 vdata->wall_time_sec = tk->xtime_sec;
1657 vdata->boot_ns_raw = boot_ns_raw;
1658 vdata->monotonic_raw_nsec = tk->tkr_raw.xtime_nsec;
1660 write_seqcount_end(&vdata->seq);
1664 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1666 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1667 kvm_vcpu_kick(vcpu);
1670 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1674 struct pvclock_wall_clock wc;
1675 struct timespec64 boot;
1680 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1685 ++version; /* first time write, random junk */
1689 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1693 * The guest calculates current wall clock time by adding
1694 * system time (updated by kvm_guest_time_update below) to the
1695 * wall clock specified here. guest system time equals host
1696 * system time for us, thus we must fill in host boot time here.
1698 getboottime64(&boot);
1700 if (kvm->arch.kvmclock_offset) {
1701 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1702 boot = timespec64_sub(boot, ts);
1704 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1705 wc.nsec = boot.tv_nsec;
1706 wc.version = version;
1708 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1711 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1714 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1716 do_shl32_div32(dividend, divisor);
1720 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1721 s8 *pshift, u32 *pmultiplier)
1729 scaled64 = scaled_hz;
1730 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1735 tps32 = (uint32_t)tps64;
1736 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1737 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1745 *pmultiplier = div_frac(scaled64, tps32);
1748 #ifdef CONFIG_X86_64
1749 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1752 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1753 static unsigned long max_tsc_khz;
1755 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1757 u64 v = (u64)khz * (1000000 + ppm);
1762 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1766 /* Guest TSC same frequency as host TSC? */
1768 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1772 /* TSC scaling supported? */
1773 if (!kvm_has_tsc_control) {
1774 if (user_tsc_khz > tsc_khz) {
1775 vcpu->arch.tsc_catchup = 1;
1776 vcpu->arch.tsc_always_catchup = 1;
1779 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1784 /* TSC scaling required - calculate ratio */
1785 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1786 user_tsc_khz, tsc_khz);
1788 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1789 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1794 vcpu->arch.tsc_scaling_ratio = ratio;
1798 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1800 u32 thresh_lo, thresh_hi;
1801 int use_scaling = 0;
1803 /* tsc_khz can be zero if TSC calibration fails */
1804 if (user_tsc_khz == 0) {
1805 /* set tsc_scaling_ratio to a safe value */
1806 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1810 /* Compute a scale to convert nanoseconds in TSC cycles */
1811 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1812 &vcpu->arch.virtual_tsc_shift,
1813 &vcpu->arch.virtual_tsc_mult);
1814 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1817 * Compute the variation in TSC rate which is acceptable
1818 * within the range of tolerance and decide if the
1819 * rate being applied is within that bounds of the hardware
1820 * rate. If so, no scaling or compensation need be done.
1822 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1823 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1824 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1825 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1828 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1831 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1833 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1834 vcpu->arch.virtual_tsc_mult,
1835 vcpu->arch.virtual_tsc_shift);
1836 tsc += vcpu->arch.this_tsc_write;
1840 static inline int gtod_is_based_on_tsc(int mode)
1842 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1845 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1847 #ifdef CONFIG_X86_64
1849 struct kvm_arch *ka = &vcpu->kvm->arch;
1850 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1852 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1853 atomic_read(&vcpu->kvm->online_vcpus));
1856 * Once the masterclock is enabled, always perform request in
1857 * order to update it.
1859 * In order to enable masterclock, the host clocksource must be TSC
1860 * and the vcpus need to have matched TSCs. When that happens,
1861 * perform request to enable masterclock.
1863 if (ka->use_master_clock ||
1864 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1865 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1867 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1868 atomic_read(&vcpu->kvm->online_vcpus),
1869 ka->use_master_clock, gtod->clock.vclock_mode);
1873 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1875 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1876 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1880 * Multiply tsc by a fixed point number represented by ratio.
1882 * The most significant 64-N bits (mult) of ratio represent the
1883 * integral part of the fixed point number; the remaining N bits
1884 * (frac) represent the fractional part, ie. ratio represents a fixed
1885 * point number (mult + frac * 2^(-N)).
1887 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1889 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1891 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1894 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1897 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1899 if (ratio != kvm_default_tsc_scaling_ratio)
1900 _tsc = __scale_tsc(ratio, tsc);
1904 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1906 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1910 tsc = kvm_scale_tsc(vcpu, rdtsc());
1912 return target_tsc - tsc;
1915 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1917 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1919 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1921 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1923 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1925 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1928 static inline bool kvm_check_tsc_unstable(void)
1930 #ifdef CONFIG_X86_64
1932 * TSC is marked unstable when we're running on Hyper-V,
1933 * 'TSC page' clocksource is good.
1935 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1938 return check_tsc_unstable();
1941 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1943 struct kvm *kvm = vcpu->kvm;
1944 u64 offset, ns, elapsed;
1945 unsigned long flags;
1947 bool already_matched;
1948 u64 data = msr->data;
1949 bool synchronizing = false;
1951 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1952 offset = kvm_compute_tsc_offset(vcpu, data);
1953 ns = ktime_get_boottime_ns();
1954 elapsed = ns - kvm->arch.last_tsc_nsec;
1956 if (vcpu->arch.virtual_tsc_khz) {
1957 if (data == 0 && msr->host_initiated) {
1959 * detection of vcpu initialization -- need to sync
1960 * with other vCPUs. This particularly helps to keep
1961 * kvm_clock stable after CPU hotplug
1963 synchronizing = true;
1965 u64 tsc_exp = kvm->arch.last_tsc_write +
1966 nsec_to_cycles(vcpu, elapsed);
1967 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1969 * Special case: TSC write with a small delta (1 second)
1970 * of virtual cycle time against real time is
1971 * interpreted as an attempt to synchronize the CPU.
1973 synchronizing = data < tsc_exp + tsc_hz &&
1974 data + tsc_hz > tsc_exp;
1979 * For a reliable TSC, we can match TSC offsets, and for an unstable
1980 * TSC, we add elapsed time in this computation. We could let the
1981 * compensation code attempt to catch up if we fall behind, but
1982 * it's better to try to match offsets from the beginning.
1984 if (synchronizing &&
1985 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1986 if (!kvm_check_tsc_unstable()) {
1987 offset = kvm->arch.cur_tsc_offset;
1989 u64 delta = nsec_to_cycles(vcpu, elapsed);
1991 offset = kvm_compute_tsc_offset(vcpu, data);
1994 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1997 * We split periods of matched TSC writes into generations.
1998 * For each generation, we track the original measured
1999 * nanosecond time, offset, and write, so if TSCs are in
2000 * sync, we can match exact offset, and if not, we can match
2001 * exact software computation in compute_guest_tsc()
2003 * These values are tracked in kvm->arch.cur_xxx variables.
2005 kvm->arch.cur_tsc_generation++;
2006 kvm->arch.cur_tsc_nsec = ns;
2007 kvm->arch.cur_tsc_write = data;
2008 kvm->arch.cur_tsc_offset = offset;
2013 * We also track th most recent recorded KHZ, write and time to
2014 * allow the matching interval to be extended at each write.
2016 kvm->arch.last_tsc_nsec = ns;
2017 kvm->arch.last_tsc_write = data;
2018 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
2020 vcpu->arch.last_guest_tsc = data;
2022 /* Keep track of which generation this VCPU has synchronized to */
2023 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
2024 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
2025 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
2027 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
2028 update_ia32_tsc_adjust_msr(vcpu, offset);
2030 kvm_vcpu_write_tsc_offset(vcpu, offset);
2031 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
2033 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
2035 kvm->arch.nr_vcpus_matched_tsc = 0;
2036 } else if (!already_matched) {
2037 kvm->arch.nr_vcpus_matched_tsc++;
2040 kvm_track_tsc_matching(vcpu);
2041 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
2044 EXPORT_SYMBOL_GPL(kvm_write_tsc);
2046 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
2049 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
2050 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
2053 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
2055 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
2056 WARN_ON(adjustment < 0);
2057 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
2058 adjust_tsc_offset_guest(vcpu, adjustment);
2061 #ifdef CONFIG_X86_64
2063 static u64 read_tsc(void)
2065 u64 ret = (u64)rdtsc_ordered();
2066 u64 last = pvclock_gtod_data.clock.cycle_last;
2068 if (likely(ret >= last))
2072 * GCC likes to generate cmov here, but this branch is extremely
2073 * predictable (it's just a function of time and the likely is
2074 * very likely) and there's a data dependence, so force GCC
2075 * to generate a branch instead. I don't barrier() because
2076 * we don't actually need a barrier, and if this function
2077 * ever gets inlined it will generate worse code.
2083 static inline u64 vgettsc(struct pvclock_clock *clock, u64 *tsc_timestamp,
2089 switch (clock->vclock_mode) {
2090 case VCLOCK_HVCLOCK:
2091 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2093 if (tsc_pg_val != U64_MAX) {
2094 /* TSC page valid */
2095 *mode = VCLOCK_HVCLOCK;
2096 v = (tsc_pg_val - clock->cycle_last) &
2099 /* TSC page invalid */
2100 *mode = VCLOCK_NONE;
2105 *tsc_timestamp = read_tsc();
2106 v = (*tsc_timestamp - clock->cycle_last) &
2110 *mode = VCLOCK_NONE;
2113 if (*mode == VCLOCK_NONE)
2114 *tsc_timestamp = v = 0;
2116 return v * clock->mult;
2119 static int do_monotonic_raw(s64 *t, u64 *tsc_timestamp)
2121 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2127 seq = read_seqcount_begin(>od->seq);
2128 ns = gtod->monotonic_raw_nsec;
2129 ns += vgettsc(>od->raw_clock, tsc_timestamp, &mode);
2130 ns >>= gtod->clock.shift;
2131 ns += gtod->boot_ns_raw;
2132 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2138 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2140 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2146 seq = read_seqcount_begin(>od->seq);
2147 ts->tv_sec = gtod->wall_time_sec;
2148 ns = gtod->nsec_base;
2149 ns += vgettsc(>od->clock, tsc_timestamp, &mode);
2150 ns >>= gtod->clock.shift;
2151 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2153 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2159 /* returns true if host is using TSC based clocksource */
2160 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2162 /* checked again under seqlock below */
2163 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2166 return gtod_is_based_on_tsc(do_monotonic_raw(kernel_ns,
2170 /* returns true if host is using TSC based clocksource */
2171 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2174 /* checked again under seqlock below */
2175 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2178 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2184 * Assuming a stable TSC across physical CPUS, and a stable TSC
2185 * across virtual CPUs, the following condition is possible.
2186 * Each numbered line represents an event visible to both
2187 * CPUs at the next numbered event.
2189 * "timespecX" represents host monotonic time. "tscX" represents
2192 * VCPU0 on CPU0 | VCPU1 on CPU1
2194 * 1. read timespec0,tsc0
2195 * 2. | timespec1 = timespec0 + N
2197 * 3. transition to guest | transition to guest
2198 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2199 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2200 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2202 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2205 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2207 * - 0 < N - M => M < N
2209 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2210 * always the case (the difference between two distinct xtime instances
2211 * might be smaller then the difference between corresponding TSC reads,
2212 * when updating guest vcpus pvclock areas).
2214 * To avoid that problem, do not allow visibility of distinct
2215 * system_timestamp/tsc_timestamp values simultaneously: use a master
2216 * copy of host monotonic time values. Update that master copy
2219 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2223 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2225 #ifdef CONFIG_X86_64
2226 struct kvm_arch *ka = &kvm->arch;
2228 bool host_tsc_clocksource, vcpus_matched;
2230 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2231 atomic_read(&kvm->online_vcpus));
2234 * If the host uses TSC clock, then passthrough TSC as stable
2237 host_tsc_clocksource = kvm_get_time_and_clockread(
2238 &ka->master_kernel_ns,
2239 &ka->master_cycle_now);
2241 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2242 && !ka->backwards_tsc_observed
2243 && !ka->boot_vcpu_runs_old_kvmclock;
2245 if (ka->use_master_clock)
2246 atomic_set(&kvm_guest_has_master_clock, 1);
2248 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2249 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2254 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2256 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2259 static void kvm_gen_update_masterclock(struct kvm *kvm)
2261 #ifdef CONFIG_X86_64
2263 struct kvm_vcpu *vcpu;
2264 struct kvm_arch *ka = &kvm->arch;
2266 spin_lock(&ka->pvclock_gtod_sync_lock);
2267 kvm_make_mclock_inprogress_request(kvm);
2268 /* no guest entries from this point */
2269 pvclock_update_vm_gtod_copy(kvm);
2271 kvm_for_each_vcpu(i, vcpu, kvm)
2272 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2274 /* guest entries allowed */
2275 kvm_for_each_vcpu(i, vcpu, kvm)
2276 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2278 spin_unlock(&ka->pvclock_gtod_sync_lock);
2282 u64 get_kvmclock_ns(struct kvm *kvm)
2284 struct kvm_arch *ka = &kvm->arch;
2285 struct pvclock_vcpu_time_info hv_clock;
2288 spin_lock(&ka->pvclock_gtod_sync_lock);
2289 if (!ka->use_master_clock) {
2290 spin_unlock(&ka->pvclock_gtod_sync_lock);
2291 return ktime_get_boottime_ns() + ka->kvmclock_offset;
2294 hv_clock.tsc_timestamp = ka->master_cycle_now;
2295 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2296 spin_unlock(&ka->pvclock_gtod_sync_lock);
2298 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2301 if (__this_cpu_read(cpu_tsc_khz)) {
2302 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2303 &hv_clock.tsc_shift,
2304 &hv_clock.tsc_to_system_mul);
2305 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2307 ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2314 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2316 struct kvm_vcpu_arch *vcpu = &v->arch;
2317 struct pvclock_vcpu_time_info guest_hv_clock;
2319 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2320 &guest_hv_clock, sizeof(guest_hv_clock))))
2323 /* This VCPU is paused, but it's legal for a guest to read another
2324 * VCPU's kvmclock, so we really have to follow the specification where
2325 * it says that version is odd if data is being modified, and even after
2328 * Version field updates must be kept separate. This is because
2329 * kvm_write_guest_cached might use a "rep movs" instruction, and
2330 * writes within a string instruction are weakly ordered. So there
2331 * are three writes overall.
2333 * As a small optimization, only write the version field in the first
2334 * and third write. The vcpu->pv_time cache is still valid, because the
2335 * version field is the first in the struct.
2337 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2339 if (guest_hv_clock.version & 1)
2340 ++guest_hv_clock.version; /* first time write, random junk */
2342 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2343 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2345 sizeof(vcpu->hv_clock.version));
2349 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2350 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2352 if (vcpu->pvclock_set_guest_stopped_request) {
2353 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2354 vcpu->pvclock_set_guest_stopped_request = false;
2357 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2359 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2361 sizeof(vcpu->hv_clock));
2365 vcpu->hv_clock.version++;
2366 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2368 sizeof(vcpu->hv_clock.version));
2371 static int kvm_guest_time_update(struct kvm_vcpu *v)
2373 unsigned long flags, tgt_tsc_khz;
2374 struct kvm_vcpu_arch *vcpu = &v->arch;
2375 struct kvm_arch *ka = &v->kvm->arch;
2377 u64 tsc_timestamp, host_tsc;
2379 bool use_master_clock;
2385 * If the host uses TSC clock, then passthrough TSC as stable
2388 spin_lock(&ka->pvclock_gtod_sync_lock);
2389 use_master_clock = ka->use_master_clock;
2390 if (use_master_clock) {
2391 host_tsc = ka->master_cycle_now;
2392 kernel_ns = ka->master_kernel_ns;
2394 spin_unlock(&ka->pvclock_gtod_sync_lock);
2396 /* Keep irq disabled to prevent changes to the clock */
2397 local_irq_save(flags);
2398 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2399 if (unlikely(tgt_tsc_khz == 0)) {
2400 local_irq_restore(flags);
2401 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2404 if (!use_master_clock) {
2406 kernel_ns = ktime_get_boottime_ns();
2409 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2412 * We may have to catch up the TSC to match elapsed wall clock
2413 * time for two reasons, even if kvmclock is used.
2414 * 1) CPU could have been running below the maximum TSC rate
2415 * 2) Broken TSC compensation resets the base at each VCPU
2416 * entry to avoid unknown leaps of TSC even when running
2417 * again on the same CPU. This may cause apparent elapsed
2418 * time to disappear, and the guest to stand still or run
2421 if (vcpu->tsc_catchup) {
2422 u64 tsc = compute_guest_tsc(v, kernel_ns);
2423 if (tsc > tsc_timestamp) {
2424 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2425 tsc_timestamp = tsc;
2429 local_irq_restore(flags);
2431 /* With all the info we got, fill in the values */
2433 if (kvm_has_tsc_control)
2434 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2436 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2437 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2438 &vcpu->hv_clock.tsc_shift,
2439 &vcpu->hv_clock.tsc_to_system_mul);
2440 vcpu->hw_tsc_khz = tgt_tsc_khz;
2443 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2444 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2445 vcpu->last_guest_tsc = tsc_timestamp;
2447 /* If the host uses TSC clocksource, then it is stable */
2449 if (use_master_clock)
2450 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2452 vcpu->hv_clock.flags = pvclock_flags;
2454 if (vcpu->pv_time_enabled)
2455 kvm_setup_pvclock_page(v);
2456 if (v == kvm_get_vcpu(v->kvm, 0))
2457 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2462 * kvmclock updates which are isolated to a given vcpu, such as
2463 * vcpu->cpu migration, should not allow system_timestamp from
2464 * the rest of the vcpus to remain static. Otherwise ntp frequency
2465 * correction applies to one vcpu's system_timestamp but not
2468 * So in those cases, request a kvmclock update for all vcpus.
2469 * We need to rate-limit these requests though, as they can
2470 * considerably slow guests that have a large number of vcpus.
2471 * The time for a remote vcpu to update its kvmclock is bound
2472 * by the delay we use to rate-limit the updates.
2475 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2477 static void kvmclock_update_fn(struct work_struct *work)
2480 struct delayed_work *dwork = to_delayed_work(work);
2481 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2482 kvmclock_update_work);
2483 struct kvm *kvm = container_of(ka, struct kvm, arch);
2484 struct kvm_vcpu *vcpu;
2486 kvm_for_each_vcpu(i, vcpu, kvm) {
2487 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2488 kvm_vcpu_kick(vcpu);
2492 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2494 struct kvm *kvm = v->kvm;
2496 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2497 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2498 KVMCLOCK_UPDATE_DELAY);
2501 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2503 static void kvmclock_sync_fn(struct work_struct *work)
2505 struct delayed_work *dwork = to_delayed_work(work);
2506 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2507 kvmclock_sync_work);
2508 struct kvm *kvm = container_of(ka, struct kvm, arch);
2510 if (!kvmclock_periodic_sync)
2513 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2514 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2515 KVMCLOCK_SYNC_PERIOD);
2519 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2521 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2523 /* McStatusWrEn enabled? */
2524 if (guest_cpuid_is_amd(vcpu))
2525 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2530 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2532 u64 mcg_cap = vcpu->arch.mcg_cap;
2533 unsigned bank_num = mcg_cap & 0xff;
2534 u32 msr = msr_info->index;
2535 u64 data = msr_info->data;
2538 case MSR_IA32_MCG_STATUS:
2539 vcpu->arch.mcg_status = data;
2541 case MSR_IA32_MCG_CTL:
2542 if (!(mcg_cap & MCG_CTL_P) &&
2543 (data || !msr_info->host_initiated))
2545 if (data != 0 && data != ~(u64)0)
2547 vcpu->arch.mcg_ctl = data;
2550 if (msr >= MSR_IA32_MC0_CTL &&
2551 msr < MSR_IA32_MCx_CTL(bank_num)) {
2552 u32 offset = array_index_nospec(
2553 msr - MSR_IA32_MC0_CTL,
2554 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
2556 /* only 0 or all 1s can be written to IA32_MCi_CTL
2557 * some Linux kernels though clear bit 10 in bank 4 to
2558 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2559 * this to avoid an uncatched #GP in the guest
2561 if ((offset & 0x3) == 0 &&
2562 data != 0 && (data | (1 << 10)) != ~(u64)0)
2566 if (!msr_info->host_initiated &&
2567 (offset & 0x3) == 1 && data != 0) {
2568 if (!can_set_mci_status(vcpu))
2572 vcpu->arch.mce_banks[offset] = data;
2580 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2582 struct kvm *kvm = vcpu->kvm;
2583 int lm = is_long_mode(vcpu);
2584 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2585 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2586 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2587 : kvm->arch.xen_hvm_config.blob_size_32;
2588 u32 page_num = data & ~PAGE_MASK;
2589 u64 page_addr = data & PAGE_MASK;
2594 if (page_num >= blob_size)
2597 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2602 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2611 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2613 gpa_t gpa = data & ~0x3f;
2615 /* Bits 3:5 are reserved, Should be zero */
2619 vcpu->arch.apf.msr_val = data;
2621 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2622 kvm_clear_async_pf_completion_queue(vcpu);
2623 kvm_async_pf_hash_reset(vcpu);
2627 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2631 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2632 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2633 kvm_async_pf_wakeup_all(vcpu);
2637 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2639 vcpu->arch.pv_time_enabled = false;
2640 vcpu->arch.time = 0;
2643 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2645 ++vcpu->stat.tlb_flush;
2646 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2649 static void record_steal_time(struct kvm_vcpu *vcpu)
2651 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2654 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2655 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2659 * Doing a TLB flush here, on the guest's behalf, can avoid
2662 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2663 vcpu->arch.st.steal.preempted & KVM_VCPU_FLUSH_TLB);
2664 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2665 kvm_vcpu_flush_tlb(vcpu, false);
2667 if (vcpu->arch.st.steal.version & 1)
2668 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2670 vcpu->arch.st.steal.version += 1;
2672 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2673 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2677 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2678 vcpu->arch.st.last_steal;
2679 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2681 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2682 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2686 vcpu->arch.st.steal.version += 1;
2688 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2689 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2692 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2695 u32 msr = msr_info->index;
2696 u64 data = msr_info->data;
2699 case MSR_AMD64_NB_CFG:
2700 case MSR_IA32_UCODE_WRITE:
2701 case MSR_VM_HSAVE_PA:
2702 case MSR_AMD64_PATCH_LOADER:
2703 case MSR_AMD64_BU_CFG2:
2704 case MSR_AMD64_DC_CFG:
2705 case MSR_F15H_EX_CFG:
2708 case MSR_IA32_UCODE_REV:
2709 if (msr_info->host_initiated)
2710 vcpu->arch.microcode_version = data;
2712 case MSR_IA32_ARCH_CAPABILITIES:
2713 if (!msr_info->host_initiated)
2715 vcpu->arch.arch_capabilities = data;
2718 return set_efer(vcpu, msr_info);
2720 data &= ~(u64)0x40; /* ignore flush filter disable */
2721 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2722 data &= ~(u64)0x8; /* ignore TLB cache disable */
2724 /* Handle McStatusWrEn */
2725 if (data == BIT_ULL(18)) {
2726 vcpu->arch.msr_hwcr = data;
2727 } else if (data != 0) {
2728 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2733 case MSR_FAM10H_MMIO_CONF_BASE:
2735 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2740 case MSR_IA32_DEBUGCTLMSR:
2742 /* We support the non-activated case already */
2744 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2745 /* Values other than LBR and BTF are vendor-specific,
2746 thus reserved and should throw a #GP */
2749 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2752 case 0x200 ... 0x2ff:
2753 return kvm_mtrr_set_msr(vcpu, msr, data);
2754 case MSR_IA32_APICBASE:
2755 return kvm_set_apic_base(vcpu, msr_info);
2756 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2757 return kvm_x2apic_msr_write(vcpu, msr, data);
2758 case MSR_IA32_TSCDEADLINE:
2759 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2761 case MSR_IA32_TSC_ADJUST:
2762 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2763 if (!msr_info->host_initiated) {
2764 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2765 adjust_tsc_offset_guest(vcpu, adj);
2767 vcpu->arch.ia32_tsc_adjust_msr = data;
2770 case MSR_IA32_MISC_ENABLE:
2771 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2772 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2773 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2775 vcpu->arch.ia32_misc_enable_msr = data;
2776 kvm_update_cpuid(vcpu);
2778 vcpu->arch.ia32_misc_enable_msr = data;
2781 case MSR_IA32_SMBASE:
2782 if (!msr_info->host_initiated)
2784 vcpu->arch.smbase = data;
2786 case MSR_IA32_POWER_CTL:
2787 vcpu->arch.msr_ia32_power_ctl = data;
2790 kvm_write_tsc(vcpu, msr_info);
2793 if (!msr_info->host_initiated &&
2794 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
2797 * We do support PT if kvm_x86_ops->pt_supported(), but we do
2798 * not support IA32_XSS[bit 8]. Guests will have to use
2799 * RDMSR/WRMSR rather than XSAVES/XRSTORS to save/restore PT
2804 vcpu->arch.ia32_xss = data;
2807 if (!msr_info->host_initiated)
2809 vcpu->arch.smi_count = data;
2811 case MSR_KVM_WALL_CLOCK_NEW:
2812 case MSR_KVM_WALL_CLOCK:
2813 vcpu->kvm->arch.wall_clock = data;
2814 kvm_write_wall_clock(vcpu->kvm, data);
2816 case MSR_KVM_SYSTEM_TIME_NEW:
2817 case MSR_KVM_SYSTEM_TIME: {
2818 struct kvm_arch *ka = &vcpu->kvm->arch;
2820 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2821 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2823 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2824 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2826 ka->boot_vcpu_runs_old_kvmclock = tmp;
2829 vcpu->arch.time = data;
2830 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2832 /* we verify if the enable bit is set... */
2833 vcpu->arch.pv_time_enabled = false;
2837 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2838 &vcpu->arch.pv_time, data & ~1ULL,
2839 sizeof(struct pvclock_vcpu_time_info)))
2840 vcpu->arch.pv_time_enabled = true;
2844 case MSR_KVM_ASYNC_PF_EN:
2845 if (kvm_pv_enable_async_pf(vcpu, data))
2848 case MSR_KVM_STEAL_TIME:
2850 if (unlikely(!sched_info_on()))
2853 if (data & KVM_STEAL_RESERVED_MASK)
2856 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2857 data & KVM_STEAL_VALID_BITS,
2858 sizeof(struct kvm_steal_time)))
2861 vcpu->arch.st.msr_val = data;
2863 if (!(data & KVM_MSR_ENABLED))
2866 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2869 case MSR_KVM_PV_EOI_EN:
2870 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2874 case MSR_KVM_POLL_CONTROL:
2875 /* only enable bit supported */
2876 if (data & (-1ULL << 1))
2879 vcpu->arch.msr_kvm_poll_control = data;
2882 case MSR_IA32_MCG_CTL:
2883 case MSR_IA32_MCG_STATUS:
2884 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2885 return set_msr_mce(vcpu, msr_info);
2887 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2888 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2889 pr = true; /* fall through */
2890 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2891 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2892 if (kvm_pmu_is_valid_msr(vcpu, msr))
2893 return kvm_pmu_set_msr(vcpu, msr_info);
2895 if (pr || data != 0)
2896 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2897 "0x%x data 0x%llx\n", msr, data);
2899 case MSR_K7_CLK_CTL:
2901 * Ignore all writes to this no longer documented MSR.
2902 * Writes are only relevant for old K7 processors,
2903 * all pre-dating SVM, but a recommended workaround from
2904 * AMD for these chips. It is possible to specify the
2905 * affected processor models on the command line, hence
2906 * the need to ignore the workaround.
2909 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2910 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2911 case HV_X64_MSR_CRASH_CTL:
2912 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2913 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2914 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2915 case HV_X64_MSR_TSC_EMULATION_STATUS:
2916 return kvm_hv_set_msr_common(vcpu, msr, data,
2917 msr_info->host_initiated);
2918 case MSR_IA32_BBL_CR_CTL3:
2919 /* Drop writes to this legacy MSR -- see rdmsr
2920 * counterpart for further detail.
2922 if (report_ignored_msrs)
2923 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2926 case MSR_AMD64_OSVW_ID_LENGTH:
2927 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2929 vcpu->arch.osvw.length = data;
2931 case MSR_AMD64_OSVW_STATUS:
2932 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2934 vcpu->arch.osvw.status = data;
2936 case MSR_PLATFORM_INFO:
2937 if (!msr_info->host_initiated ||
2938 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2939 cpuid_fault_enabled(vcpu)))
2941 vcpu->arch.msr_platform_info = data;
2943 case MSR_MISC_FEATURES_ENABLES:
2944 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2945 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2946 !supports_cpuid_fault(vcpu)))
2948 vcpu->arch.msr_misc_features_enables = data;
2951 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2952 return xen_hvm_config(vcpu, data);
2953 if (kvm_pmu_is_valid_msr(vcpu, msr))
2954 return kvm_pmu_set_msr(vcpu, msr_info);
2956 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2960 if (report_ignored_msrs)
2962 "ignored wrmsr: 0x%x data 0x%llx\n",
2969 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2971 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2974 u64 mcg_cap = vcpu->arch.mcg_cap;
2975 unsigned bank_num = mcg_cap & 0xff;
2978 case MSR_IA32_P5_MC_ADDR:
2979 case MSR_IA32_P5_MC_TYPE:
2982 case MSR_IA32_MCG_CAP:
2983 data = vcpu->arch.mcg_cap;
2985 case MSR_IA32_MCG_CTL:
2986 if (!(mcg_cap & MCG_CTL_P) && !host)
2988 data = vcpu->arch.mcg_ctl;
2990 case MSR_IA32_MCG_STATUS:
2991 data = vcpu->arch.mcg_status;
2994 if (msr >= MSR_IA32_MC0_CTL &&
2995 msr < MSR_IA32_MCx_CTL(bank_num)) {
2996 u32 offset = array_index_nospec(
2997 msr - MSR_IA32_MC0_CTL,
2998 MSR_IA32_MCx_CTL(bank_num) - MSR_IA32_MC0_CTL);
3000 data = vcpu->arch.mce_banks[offset];
3009 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3011 switch (msr_info->index) {
3012 case MSR_IA32_PLATFORM_ID:
3013 case MSR_IA32_EBL_CR_POWERON:
3014 case MSR_IA32_DEBUGCTLMSR:
3015 case MSR_IA32_LASTBRANCHFROMIP:
3016 case MSR_IA32_LASTBRANCHTOIP:
3017 case MSR_IA32_LASTINTFROMIP:
3018 case MSR_IA32_LASTINTTOIP:
3020 case MSR_K8_TSEG_ADDR:
3021 case MSR_K8_TSEG_MASK:
3022 case MSR_VM_HSAVE_PA:
3023 case MSR_K8_INT_PENDING_MSG:
3024 case MSR_AMD64_NB_CFG:
3025 case MSR_FAM10H_MMIO_CONF_BASE:
3026 case MSR_AMD64_BU_CFG2:
3027 case MSR_IA32_PERF_CTL:
3028 case MSR_AMD64_DC_CFG:
3029 case MSR_F15H_EX_CFG:
3032 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
3033 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
3034 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
3035 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
3036 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
3037 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3038 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3041 case MSR_IA32_UCODE_REV:
3042 msr_info->data = vcpu->arch.microcode_version;
3044 case MSR_IA32_ARCH_CAPABILITIES:
3045 if (!msr_info->host_initiated &&
3046 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
3048 msr_info->data = vcpu->arch.arch_capabilities;
3050 case MSR_IA32_POWER_CTL:
3051 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
3054 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
3057 case 0x200 ... 0x2ff:
3058 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
3059 case 0xcd: /* fsb frequency */
3063 * MSR_EBC_FREQUENCY_ID
3064 * Conservative value valid for even the basic CPU models.
3065 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
3066 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
3067 * and 266MHz for model 3, or 4. Set Core Clock
3068 * Frequency to System Bus Frequency Ratio to 1 (bits
3069 * 31:24) even though these are only valid for CPU
3070 * models > 2, however guests may end up dividing or
3071 * multiplying by zero otherwise.
3073 case MSR_EBC_FREQUENCY_ID:
3074 msr_info->data = 1 << 24;
3076 case MSR_IA32_APICBASE:
3077 msr_info->data = kvm_get_apic_base(vcpu);
3079 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
3080 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
3082 case MSR_IA32_TSCDEADLINE:
3083 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3085 case MSR_IA32_TSC_ADJUST:
3086 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3088 case MSR_IA32_MISC_ENABLE:
3089 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3091 case MSR_IA32_SMBASE:
3092 if (!msr_info->host_initiated)
3094 msr_info->data = vcpu->arch.smbase;
3097 msr_info->data = vcpu->arch.smi_count;
3099 case MSR_IA32_PERF_STATUS:
3100 /* TSC increment by tick */
3101 msr_info->data = 1000ULL;
3102 /* CPU multiplier */
3103 msr_info->data |= (((uint64_t)4ULL) << 40);
3106 msr_info->data = vcpu->arch.efer;
3108 case MSR_KVM_WALL_CLOCK:
3109 case MSR_KVM_WALL_CLOCK_NEW:
3110 msr_info->data = vcpu->kvm->arch.wall_clock;
3112 case MSR_KVM_SYSTEM_TIME:
3113 case MSR_KVM_SYSTEM_TIME_NEW:
3114 msr_info->data = vcpu->arch.time;
3116 case MSR_KVM_ASYNC_PF_EN:
3117 msr_info->data = vcpu->arch.apf.msr_val;
3119 case MSR_KVM_STEAL_TIME:
3120 msr_info->data = vcpu->arch.st.msr_val;
3122 case MSR_KVM_PV_EOI_EN:
3123 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3125 case MSR_KVM_POLL_CONTROL:
3126 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3128 case MSR_IA32_P5_MC_ADDR:
3129 case MSR_IA32_P5_MC_TYPE:
3130 case MSR_IA32_MCG_CAP:
3131 case MSR_IA32_MCG_CTL:
3132 case MSR_IA32_MCG_STATUS:
3133 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3134 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3135 msr_info->host_initiated);
3137 if (!msr_info->host_initiated &&
3138 !guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3140 msr_info->data = vcpu->arch.ia32_xss;
3142 case MSR_K7_CLK_CTL:
3144 * Provide expected ramp-up count for K7. All other
3145 * are set to zero, indicating minimum divisors for
3148 * This prevents guest kernels on AMD host with CPU
3149 * type 6, model 8 and higher from exploding due to
3150 * the rdmsr failing.
3152 msr_info->data = 0x20000000;
3154 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3155 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3156 case HV_X64_MSR_CRASH_CTL:
3157 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3158 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3159 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3160 case HV_X64_MSR_TSC_EMULATION_STATUS:
3161 return kvm_hv_get_msr_common(vcpu,
3162 msr_info->index, &msr_info->data,
3163 msr_info->host_initiated);
3165 case MSR_IA32_BBL_CR_CTL3:
3166 /* This legacy MSR exists but isn't fully documented in current
3167 * silicon. It is however accessed by winxp in very narrow
3168 * scenarios where it sets bit #19, itself documented as
3169 * a "reserved" bit. Best effort attempt to source coherent
3170 * read data here should the balance of the register be
3171 * interpreted by the guest:
3173 * L2 cache control register 3: 64GB range, 256KB size,
3174 * enabled, latency 0x1, configured
3176 msr_info->data = 0xbe702111;
3178 case MSR_AMD64_OSVW_ID_LENGTH:
3179 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3181 msr_info->data = vcpu->arch.osvw.length;
3183 case MSR_AMD64_OSVW_STATUS:
3184 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3186 msr_info->data = vcpu->arch.osvw.status;
3188 case MSR_PLATFORM_INFO:
3189 if (!msr_info->host_initiated &&
3190 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3192 msr_info->data = vcpu->arch.msr_platform_info;
3194 case MSR_MISC_FEATURES_ENABLES:
3195 msr_info->data = vcpu->arch.msr_misc_features_enables;
3198 msr_info->data = vcpu->arch.msr_hwcr;
3201 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3202 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3204 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3208 if (report_ignored_msrs)
3209 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3217 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3220 * Read or write a bunch of msrs. All parameters are kernel addresses.
3222 * @return number of msrs set successfully.
3224 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3225 struct kvm_msr_entry *entries,
3226 int (*do_msr)(struct kvm_vcpu *vcpu,
3227 unsigned index, u64 *data))
3231 for (i = 0; i < msrs->nmsrs; ++i)
3232 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3239 * Read or write a bunch of msrs. Parameters are user addresses.
3241 * @return number of msrs set successfully.
3243 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3244 int (*do_msr)(struct kvm_vcpu *vcpu,
3245 unsigned index, u64 *data),
3248 struct kvm_msrs msrs;
3249 struct kvm_msr_entry *entries;
3254 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3258 if (msrs.nmsrs >= MAX_IO_MSRS)
3261 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3262 entries = memdup_user(user_msrs->entries, size);
3263 if (IS_ERR(entries)) {
3264 r = PTR_ERR(entries);
3268 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3273 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3284 static inline bool kvm_can_mwait_in_guest(void)
3286 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3287 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3288 boot_cpu_has(X86_FEATURE_ARAT);
3291 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3296 case KVM_CAP_IRQCHIP:
3298 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3299 case KVM_CAP_SET_TSS_ADDR:
3300 case KVM_CAP_EXT_CPUID:
3301 case KVM_CAP_EXT_EMUL_CPUID:
3302 case KVM_CAP_CLOCKSOURCE:
3304 case KVM_CAP_NOP_IO_DELAY:
3305 case KVM_CAP_MP_STATE:
3306 case KVM_CAP_SYNC_MMU:
3307 case KVM_CAP_USER_NMI:
3308 case KVM_CAP_REINJECT_CONTROL:
3309 case KVM_CAP_IRQ_INJECT_STATUS:
3310 case KVM_CAP_IOEVENTFD:
3311 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3313 case KVM_CAP_PIT_STATE2:
3314 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3315 case KVM_CAP_XEN_HVM:
3316 case KVM_CAP_VCPU_EVENTS:
3317 case KVM_CAP_HYPERV:
3318 case KVM_CAP_HYPERV_VAPIC:
3319 case KVM_CAP_HYPERV_SPIN:
3320 case KVM_CAP_HYPERV_SYNIC:
3321 case KVM_CAP_HYPERV_SYNIC2:
3322 case KVM_CAP_HYPERV_VP_INDEX:
3323 case KVM_CAP_HYPERV_EVENTFD:
3324 case KVM_CAP_HYPERV_TLBFLUSH:
3325 case KVM_CAP_HYPERV_SEND_IPI:
3326 case KVM_CAP_HYPERV_CPUID:
3327 case KVM_CAP_PCI_SEGMENT:
3328 case KVM_CAP_DEBUGREGS:
3329 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3331 case KVM_CAP_ASYNC_PF:
3332 case KVM_CAP_GET_TSC_KHZ:
3333 case KVM_CAP_KVMCLOCK_CTRL:
3334 case KVM_CAP_READONLY_MEM:
3335 case KVM_CAP_HYPERV_TIME:
3336 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3337 case KVM_CAP_TSC_DEADLINE_TIMER:
3338 case KVM_CAP_DISABLE_QUIRKS:
3339 case KVM_CAP_SET_BOOT_CPU_ID:
3340 case KVM_CAP_SPLIT_IRQCHIP:
3341 case KVM_CAP_IMMEDIATE_EXIT:
3342 case KVM_CAP_PMU_EVENT_FILTER:
3343 case KVM_CAP_GET_MSR_FEATURES:
3344 case KVM_CAP_MSR_PLATFORM_INFO:
3345 case KVM_CAP_EXCEPTION_PAYLOAD:
3348 case KVM_CAP_SYNC_REGS:
3349 r = KVM_SYNC_X86_VALID_FIELDS;
3351 case KVM_CAP_ADJUST_CLOCK:
3352 r = KVM_CLOCK_TSC_STABLE;
3354 case KVM_CAP_X86_DISABLE_EXITS:
3355 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3356 KVM_X86_DISABLE_EXITS_CSTATE;
3357 if(kvm_can_mwait_in_guest())
3358 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3360 case KVM_CAP_X86_SMM:
3361 /* SMBASE is usually relocated above 1M on modern chipsets,
3362 * and SMM handlers might indeed rely on 4G segment limits,
3363 * so do not report SMM to be available if real mode is
3364 * emulated via vm86 mode. Still, do not go to great lengths
3365 * to avoid userspace's usage of the feature, because it is a
3366 * fringe case that is not enabled except via specific settings
3367 * of the module parameters.
3369 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3372 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3374 case KVM_CAP_NR_VCPUS:
3375 r = KVM_SOFT_MAX_VCPUS;
3377 case KVM_CAP_MAX_VCPUS:
3380 case KVM_CAP_MAX_VCPU_ID:
3381 r = KVM_MAX_VCPU_ID;
3383 case KVM_CAP_PV_MMU: /* obsolete */
3387 r = KVM_MAX_MCE_BANKS;
3390 r = boot_cpu_has(X86_FEATURE_XSAVE);
3392 case KVM_CAP_TSC_CONTROL:
3393 r = kvm_has_tsc_control;
3395 case KVM_CAP_X2APIC_API:
3396 r = KVM_X2APIC_API_VALID_FLAGS;
3398 case KVM_CAP_NESTED_STATE:
3399 r = kvm_x86_ops->get_nested_state ?
3400 kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3402 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3403 r = kvm_x86_ops->enable_direct_tlbflush != NULL;
3405 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3406 r = kvm_x86_ops->nested_enable_evmcs != NULL;
3415 long kvm_arch_dev_ioctl(struct file *filp,
3416 unsigned int ioctl, unsigned long arg)
3418 void __user *argp = (void __user *)arg;
3422 case KVM_GET_MSR_INDEX_LIST: {
3423 struct kvm_msr_list __user *user_msr_list = argp;
3424 struct kvm_msr_list msr_list;
3428 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3431 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3432 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3435 if (n < msr_list.nmsrs)
3438 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3439 num_msrs_to_save * sizeof(u32)))
3441 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3443 num_emulated_msrs * sizeof(u32)))
3448 case KVM_GET_SUPPORTED_CPUID:
3449 case KVM_GET_EMULATED_CPUID: {
3450 struct kvm_cpuid2 __user *cpuid_arg = argp;
3451 struct kvm_cpuid2 cpuid;
3454 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3457 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3463 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3468 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3470 if (copy_to_user(argp, &kvm_mce_cap_supported,
3471 sizeof(kvm_mce_cap_supported)))
3475 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3476 struct kvm_msr_list __user *user_msr_list = argp;
3477 struct kvm_msr_list msr_list;
3481 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3484 msr_list.nmsrs = num_msr_based_features;
3485 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3488 if (n < msr_list.nmsrs)
3491 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3492 num_msr_based_features * sizeof(u32)))
3498 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3508 static void wbinvd_ipi(void *garbage)
3513 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3515 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3518 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3520 /* Address WBINVD may be executed by guest */
3521 if (need_emulate_wbinvd(vcpu)) {
3522 if (kvm_x86_ops->has_wbinvd_exit())
3523 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3524 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3525 smp_call_function_single(vcpu->cpu,
3526 wbinvd_ipi, NULL, 1);
3529 kvm_x86_ops->vcpu_load(vcpu, cpu);
3531 fpregs_assert_state_consistent();
3532 if (test_thread_flag(TIF_NEED_FPU_LOAD))
3533 switch_fpu_return();
3535 /* Apply any externally detected TSC adjustments (due to suspend) */
3536 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3537 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3538 vcpu->arch.tsc_offset_adjustment = 0;
3539 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3542 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3543 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3544 rdtsc() - vcpu->arch.last_host_tsc;
3546 mark_tsc_unstable("KVM discovered backwards TSC");
3548 if (kvm_check_tsc_unstable()) {
3549 u64 offset = kvm_compute_tsc_offset(vcpu,
3550 vcpu->arch.last_guest_tsc);
3551 kvm_vcpu_write_tsc_offset(vcpu, offset);
3552 vcpu->arch.tsc_catchup = 1;
3555 if (kvm_lapic_hv_timer_in_use(vcpu))
3556 kvm_lapic_restart_hv_timer(vcpu);
3559 * On a host with synchronized TSC, there is no need to update
3560 * kvmclock on vcpu->cpu migration
3562 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3563 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3564 if (vcpu->cpu != cpu)
3565 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3569 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3572 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3574 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3577 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3579 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3580 &vcpu->arch.st.steal.preempted,
3581 offsetof(struct kvm_steal_time, preempted),
3582 sizeof(vcpu->arch.st.steal.preempted));
3585 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3589 if (vcpu->preempted)
3590 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3593 * Disable page faults because we're in atomic context here.
3594 * kvm_write_guest_offset_cached() would call might_fault()
3595 * that relies on pagefault_disable() to tell if there's a
3596 * bug. NOTE: the write to guest memory may not go through if
3597 * during postcopy live migration or if there's heavy guest
3600 pagefault_disable();
3602 * kvm_memslots() will be called by
3603 * kvm_write_guest_offset_cached() so take the srcu lock.
3605 idx = srcu_read_lock(&vcpu->kvm->srcu);
3606 kvm_steal_time_set_preempted(vcpu);
3607 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3609 kvm_x86_ops->vcpu_put(vcpu);
3610 vcpu->arch.last_host_tsc = rdtsc();
3612 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3613 * on every vmexit, but if not, we might have a stale dr6 from the
3614 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3619 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3620 struct kvm_lapic_state *s)
3622 if (vcpu->arch.apicv_active)
3623 kvm_x86_ops->sync_pir_to_irr(vcpu);
3625 return kvm_apic_get_state(vcpu, s);
3628 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3629 struct kvm_lapic_state *s)
3633 r = kvm_apic_set_state(vcpu, s);
3636 update_cr8_intercept(vcpu);
3641 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3643 return (!lapic_in_kernel(vcpu) ||
3644 kvm_apic_accept_pic_intr(vcpu));
3648 * if userspace requested an interrupt window, check that the
3649 * interrupt window is open.
3651 * No need to exit to userspace if we already have an interrupt queued.
3653 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3655 return kvm_arch_interrupt_allowed(vcpu) &&
3656 !kvm_cpu_has_interrupt(vcpu) &&
3657 !kvm_event_needs_reinjection(vcpu) &&
3658 kvm_cpu_accept_dm_intr(vcpu);
3661 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3662 struct kvm_interrupt *irq)
3664 if (irq->irq >= KVM_NR_INTERRUPTS)
3667 if (!irqchip_in_kernel(vcpu->kvm)) {
3668 kvm_queue_interrupt(vcpu, irq->irq, false);
3669 kvm_make_request(KVM_REQ_EVENT, vcpu);
3674 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3675 * fail for in-kernel 8259.
3677 if (pic_in_kernel(vcpu->kvm))
3680 if (vcpu->arch.pending_external_vector != -1)
3683 vcpu->arch.pending_external_vector = irq->irq;
3684 kvm_make_request(KVM_REQ_EVENT, vcpu);
3688 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3690 kvm_inject_nmi(vcpu);
3695 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3697 kvm_make_request(KVM_REQ_SMI, vcpu);
3702 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3703 struct kvm_tpr_access_ctl *tac)
3707 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3711 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3715 unsigned bank_num = mcg_cap & 0xff, bank;
3718 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3720 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3723 vcpu->arch.mcg_cap = mcg_cap;
3724 /* Init IA32_MCG_CTL to all 1s */
3725 if (mcg_cap & MCG_CTL_P)
3726 vcpu->arch.mcg_ctl = ~(u64)0;
3727 /* Init IA32_MCi_CTL to all 1s */
3728 for (bank = 0; bank < bank_num; bank++)
3729 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3731 kvm_x86_ops->setup_mce(vcpu);
3736 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3737 struct kvm_x86_mce *mce)
3739 u64 mcg_cap = vcpu->arch.mcg_cap;
3740 unsigned bank_num = mcg_cap & 0xff;
3741 u64 *banks = vcpu->arch.mce_banks;
3743 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3746 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3747 * reporting is disabled
3749 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3750 vcpu->arch.mcg_ctl != ~(u64)0)
3752 banks += 4 * mce->bank;
3754 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3755 * reporting is disabled for the bank
3757 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3759 if (mce->status & MCI_STATUS_UC) {
3760 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3761 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3762 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3765 if (banks[1] & MCI_STATUS_VAL)
3766 mce->status |= MCI_STATUS_OVER;
3767 banks[2] = mce->addr;
3768 banks[3] = mce->misc;
3769 vcpu->arch.mcg_status = mce->mcg_status;
3770 banks[1] = mce->status;
3771 kvm_queue_exception(vcpu, MC_VECTOR);
3772 } else if (!(banks[1] & MCI_STATUS_VAL)
3773 || !(banks[1] & MCI_STATUS_UC)) {
3774 if (banks[1] & MCI_STATUS_VAL)
3775 mce->status |= MCI_STATUS_OVER;
3776 banks[2] = mce->addr;
3777 banks[3] = mce->misc;
3778 banks[1] = mce->status;
3780 banks[1] |= MCI_STATUS_OVER;
3784 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3785 struct kvm_vcpu_events *events)
3790 * The API doesn't provide the instruction length for software
3791 * exceptions, so don't report them. As long as the guest RIP
3792 * isn't advanced, we should expect to encounter the exception
3795 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3796 events->exception.injected = 0;
3797 events->exception.pending = 0;
3799 events->exception.injected = vcpu->arch.exception.injected;
3800 events->exception.pending = vcpu->arch.exception.pending;
3802 * For ABI compatibility, deliberately conflate
3803 * pending and injected exceptions when
3804 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3806 if (!vcpu->kvm->arch.exception_payload_enabled)
3807 events->exception.injected |=
3808 vcpu->arch.exception.pending;
3810 events->exception.nr = vcpu->arch.exception.nr;
3811 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3812 events->exception.error_code = vcpu->arch.exception.error_code;
3813 events->exception_has_payload = vcpu->arch.exception.has_payload;
3814 events->exception_payload = vcpu->arch.exception.payload;
3816 events->interrupt.injected =
3817 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3818 events->interrupt.nr = vcpu->arch.interrupt.nr;
3819 events->interrupt.soft = 0;
3820 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3822 events->nmi.injected = vcpu->arch.nmi_injected;
3823 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3824 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3825 events->nmi.pad = 0;
3827 events->sipi_vector = 0; /* never valid when reporting to user space */
3829 events->smi.smm = is_smm(vcpu);
3830 events->smi.pending = vcpu->arch.smi_pending;
3831 events->smi.smm_inside_nmi =
3832 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3833 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3835 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3836 | KVM_VCPUEVENT_VALID_SHADOW
3837 | KVM_VCPUEVENT_VALID_SMM);
3838 if (vcpu->kvm->arch.exception_payload_enabled)
3839 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3841 memset(&events->reserved, 0, sizeof(events->reserved));
3844 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3846 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3847 struct kvm_vcpu_events *events)
3849 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3850 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3851 | KVM_VCPUEVENT_VALID_SHADOW
3852 | KVM_VCPUEVENT_VALID_SMM
3853 | KVM_VCPUEVENT_VALID_PAYLOAD))
3856 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3857 if (!vcpu->kvm->arch.exception_payload_enabled)
3859 if (events->exception.pending)
3860 events->exception.injected = 0;
3862 events->exception_has_payload = 0;
3864 events->exception.pending = 0;
3865 events->exception_has_payload = 0;
3868 if ((events->exception.injected || events->exception.pending) &&
3869 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3872 /* INITs are latched while in SMM */
3873 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3874 (events->smi.smm || events->smi.pending) &&
3875 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3879 vcpu->arch.exception.injected = events->exception.injected;
3880 vcpu->arch.exception.pending = events->exception.pending;
3881 vcpu->arch.exception.nr = events->exception.nr;
3882 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3883 vcpu->arch.exception.error_code = events->exception.error_code;
3884 vcpu->arch.exception.has_payload = events->exception_has_payload;
3885 vcpu->arch.exception.payload = events->exception_payload;
3887 vcpu->arch.interrupt.injected = events->interrupt.injected;
3888 vcpu->arch.interrupt.nr = events->interrupt.nr;
3889 vcpu->arch.interrupt.soft = events->interrupt.soft;
3890 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3891 kvm_x86_ops->set_interrupt_shadow(vcpu,
3892 events->interrupt.shadow);
3894 vcpu->arch.nmi_injected = events->nmi.injected;
3895 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3896 vcpu->arch.nmi_pending = events->nmi.pending;
3897 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3899 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3900 lapic_in_kernel(vcpu))
3901 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3903 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3904 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3905 if (events->smi.smm)
3906 vcpu->arch.hflags |= HF_SMM_MASK;
3908 vcpu->arch.hflags &= ~HF_SMM_MASK;
3909 kvm_smm_changed(vcpu);
3912 vcpu->arch.smi_pending = events->smi.pending;
3914 if (events->smi.smm) {
3915 if (events->smi.smm_inside_nmi)
3916 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3918 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3921 if (lapic_in_kernel(vcpu)) {
3922 if (events->smi.latched_init)
3923 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3925 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3929 kvm_make_request(KVM_REQ_EVENT, vcpu);
3934 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3935 struct kvm_debugregs *dbgregs)
3939 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3940 kvm_get_dr(vcpu, 6, &val);
3942 dbgregs->dr7 = vcpu->arch.dr7;
3944 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3947 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3948 struct kvm_debugregs *dbgregs)
3953 if (dbgregs->dr6 & ~0xffffffffull)
3955 if (dbgregs->dr7 & ~0xffffffffull)
3958 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3959 kvm_update_dr0123(vcpu);
3960 vcpu->arch.dr6 = dbgregs->dr6;
3961 kvm_update_dr6(vcpu);
3962 vcpu->arch.dr7 = dbgregs->dr7;
3963 kvm_update_dr7(vcpu);
3968 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3970 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3972 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3973 u64 xstate_bv = xsave->header.xfeatures;
3977 * Copy legacy XSAVE area, to avoid complications with CPUID
3978 * leaves 0 and 1 in the loop below.
3980 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3983 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3984 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3987 * Copy each region from the possibly compacted offset to the
3988 * non-compacted offset.
3990 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3992 u64 xfeature_mask = valid & -valid;
3993 int xfeature_nr = fls64(xfeature_mask) - 1;
3994 void *src = get_xsave_addr(xsave, xfeature_nr);
3997 u32 size, offset, ecx, edx;
3998 cpuid_count(XSTATE_CPUID, xfeature_nr,
3999 &size, &offset, &ecx, &edx);
4000 if (xfeature_nr == XFEATURE_PKRU)
4001 memcpy(dest + offset, &vcpu->arch.pkru,
4002 sizeof(vcpu->arch.pkru));
4004 memcpy(dest + offset, src, size);
4008 valid -= xfeature_mask;
4012 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
4014 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
4015 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
4019 * Copy legacy XSAVE area, to avoid complications with CPUID
4020 * leaves 0 and 1 in the loop below.
4022 memcpy(xsave, src, XSAVE_HDR_OFFSET);
4024 /* Set XSTATE_BV and possibly XCOMP_BV. */
4025 xsave->header.xfeatures = xstate_bv;
4026 if (boot_cpu_has(X86_FEATURE_XSAVES))
4027 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
4030 * Copy each region from the non-compacted offset to the
4031 * possibly compacted offset.
4033 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
4035 u64 xfeature_mask = valid & -valid;
4036 int xfeature_nr = fls64(xfeature_mask) - 1;
4037 void *dest = get_xsave_addr(xsave, xfeature_nr);
4040 u32 size, offset, ecx, edx;
4041 cpuid_count(XSTATE_CPUID, xfeature_nr,
4042 &size, &offset, &ecx, &edx);
4043 if (xfeature_nr == XFEATURE_PKRU)
4044 memcpy(&vcpu->arch.pkru, src + offset,
4045 sizeof(vcpu->arch.pkru));
4047 memcpy(dest, src + offset, size);
4050 valid -= xfeature_mask;
4054 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
4055 struct kvm_xsave *guest_xsave)
4057 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4058 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
4059 fill_xsave((u8 *) guest_xsave->region, vcpu);
4061 memcpy(guest_xsave->region,
4062 &vcpu->arch.guest_fpu->state.fxsave,
4063 sizeof(struct fxregs_state));
4064 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
4065 XFEATURE_MASK_FPSSE;
4069 #define XSAVE_MXCSR_OFFSET 24
4071 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
4072 struct kvm_xsave *guest_xsave)
4075 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
4076 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
4078 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
4080 * Here we allow setting states that are not present in
4081 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
4082 * with old userspace.
4084 if (xstate_bv & ~kvm_supported_xcr0() ||
4085 mxcsr & ~mxcsr_feature_mask)
4087 load_xsave(vcpu, (u8 *)guest_xsave->region);
4089 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
4090 mxcsr & ~mxcsr_feature_mask)
4092 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4093 guest_xsave->region, sizeof(struct fxregs_state));
4098 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4099 struct kvm_xcrs *guest_xcrs)
4101 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4102 guest_xcrs->nr_xcrs = 0;
4106 guest_xcrs->nr_xcrs = 1;
4107 guest_xcrs->flags = 0;
4108 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4109 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4112 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4113 struct kvm_xcrs *guest_xcrs)
4117 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4120 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4123 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4124 /* Only support XCR0 currently */
4125 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4126 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4127 guest_xcrs->xcrs[i].value);
4136 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4137 * stopped by the hypervisor. This function will be called from the host only.
4138 * EINVAL is returned when the host attempts to set the flag for a guest that
4139 * does not support pv clocks.
4141 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4143 if (!vcpu->arch.pv_time_enabled)
4145 vcpu->arch.pvclock_set_guest_stopped_request = true;
4146 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4150 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4151 struct kvm_enable_cap *cap)
4154 uint16_t vmcs_version;
4155 void __user *user_ptr;
4161 case KVM_CAP_HYPERV_SYNIC2:
4166 case KVM_CAP_HYPERV_SYNIC:
4167 if (!irqchip_in_kernel(vcpu->kvm))
4169 return kvm_hv_activate_synic(vcpu, cap->cap ==
4170 KVM_CAP_HYPERV_SYNIC2);
4171 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4172 if (!kvm_x86_ops->nested_enable_evmcs)
4174 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4176 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4177 if (copy_to_user(user_ptr, &vmcs_version,
4178 sizeof(vmcs_version)))
4182 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4183 if (!kvm_x86_ops->enable_direct_tlbflush)
4186 return kvm_x86_ops->enable_direct_tlbflush(vcpu);
4193 long kvm_arch_vcpu_ioctl(struct file *filp,
4194 unsigned int ioctl, unsigned long arg)
4196 struct kvm_vcpu *vcpu = filp->private_data;
4197 void __user *argp = (void __user *)arg;
4200 struct kvm_lapic_state *lapic;
4201 struct kvm_xsave *xsave;
4202 struct kvm_xcrs *xcrs;
4210 case KVM_GET_LAPIC: {
4212 if (!lapic_in_kernel(vcpu))
4214 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4215 GFP_KERNEL_ACCOUNT);
4220 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4224 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4229 case KVM_SET_LAPIC: {
4231 if (!lapic_in_kernel(vcpu))
4233 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4234 if (IS_ERR(u.lapic)) {
4235 r = PTR_ERR(u.lapic);
4239 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4242 case KVM_INTERRUPT: {
4243 struct kvm_interrupt irq;
4246 if (copy_from_user(&irq, argp, sizeof(irq)))
4248 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4252 r = kvm_vcpu_ioctl_nmi(vcpu);
4256 r = kvm_vcpu_ioctl_smi(vcpu);
4259 case KVM_SET_CPUID: {
4260 struct kvm_cpuid __user *cpuid_arg = argp;
4261 struct kvm_cpuid cpuid;
4264 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4266 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4269 case KVM_SET_CPUID2: {
4270 struct kvm_cpuid2 __user *cpuid_arg = argp;
4271 struct kvm_cpuid2 cpuid;
4274 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4276 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4277 cpuid_arg->entries);
4280 case KVM_GET_CPUID2: {
4281 struct kvm_cpuid2 __user *cpuid_arg = argp;
4282 struct kvm_cpuid2 cpuid;
4285 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4287 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4288 cpuid_arg->entries);
4292 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4297 case KVM_GET_MSRS: {
4298 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4299 r = msr_io(vcpu, argp, do_get_msr, 1);
4300 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4303 case KVM_SET_MSRS: {
4304 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4305 r = msr_io(vcpu, argp, do_set_msr, 0);
4306 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4309 case KVM_TPR_ACCESS_REPORTING: {
4310 struct kvm_tpr_access_ctl tac;
4313 if (copy_from_user(&tac, argp, sizeof(tac)))
4315 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4319 if (copy_to_user(argp, &tac, sizeof(tac)))
4324 case KVM_SET_VAPIC_ADDR: {
4325 struct kvm_vapic_addr va;
4329 if (!lapic_in_kernel(vcpu))
4332 if (copy_from_user(&va, argp, sizeof(va)))
4334 idx = srcu_read_lock(&vcpu->kvm->srcu);
4335 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4336 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4339 case KVM_X86_SETUP_MCE: {
4343 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4345 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4348 case KVM_X86_SET_MCE: {
4349 struct kvm_x86_mce mce;
4352 if (copy_from_user(&mce, argp, sizeof(mce)))
4354 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4357 case KVM_GET_VCPU_EVENTS: {
4358 struct kvm_vcpu_events events;
4360 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4363 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4368 case KVM_SET_VCPU_EVENTS: {
4369 struct kvm_vcpu_events events;
4372 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4375 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4378 case KVM_GET_DEBUGREGS: {
4379 struct kvm_debugregs dbgregs;
4381 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4384 if (copy_to_user(argp, &dbgregs,
4385 sizeof(struct kvm_debugregs)))
4390 case KVM_SET_DEBUGREGS: {
4391 struct kvm_debugregs dbgregs;
4394 if (copy_from_user(&dbgregs, argp,
4395 sizeof(struct kvm_debugregs)))
4398 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4401 case KVM_GET_XSAVE: {
4402 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4407 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4410 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4415 case KVM_SET_XSAVE: {
4416 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4417 if (IS_ERR(u.xsave)) {
4418 r = PTR_ERR(u.xsave);
4422 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4425 case KVM_GET_XCRS: {
4426 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4431 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4434 if (copy_to_user(argp, u.xcrs,
4435 sizeof(struct kvm_xcrs)))
4440 case KVM_SET_XCRS: {
4441 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4442 if (IS_ERR(u.xcrs)) {
4443 r = PTR_ERR(u.xcrs);
4447 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4450 case KVM_SET_TSC_KHZ: {
4454 user_tsc_khz = (u32)arg;
4456 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4459 if (user_tsc_khz == 0)
4460 user_tsc_khz = tsc_khz;
4462 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4467 case KVM_GET_TSC_KHZ: {
4468 r = vcpu->arch.virtual_tsc_khz;
4471 case KVM_KVMCLOCK_CTRL: {
4472 r = kvm_set_guest_paused(vcpu);
4475 case KVM_ENABLE_CAP: {
4476 struct kvm_enable_cap cap;
4479 if (copy_from_user(&cap, argp, sizeof(cap)))
4481 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4484 case KVM_GET_NESTED_STATE: {
4485 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4489 if (!kvm_x86_ops->get_nested_state)
4492 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4494 if (get_user(user_data_size, &user_kvm_nested_state->size))
4497 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4502 if (r > user_data_size) {
4503 if (put_user(r, &user_kvm_nested_state->size))
4513 case KVM_SET_NESTED_STATE: {
4514 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4515 struct kvm_nested_state kvm_state;
4519 if (!kvm_x86_ops->set_nested_state)
4523 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4527 if (kvm_state.size < sizeof(kvm_state))
4530 if (kvm_state.flags &
4531 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4532 | KVM_STATE_NESTED_EVMCS))
4535 /* nested_run_pending implies guest_mode. */
4536 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4537 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4540 idx = srcu_read_lock(&vcpu->kvm->srcu);
4541 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4542 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4545 case KVM_GET_SUPPORTED_HV_CPUID: {
4546 struct kvm_cpuid2 __user *cpuid_arg = argp;
4547 struct kvm_cpuid2 cpuid;
4550 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4553 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4554 cpuid_arg->entries);
4559 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4574 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4576 return VM_FAULT_SIGBUS;
4579 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4583 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4585 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4589 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4592 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4595 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4596 unsigned long kvm_nr_mmu_pages)
4598 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4601 mutex_lock(&kvm->slots_lock);
4603 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4604 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4606 mutex_unlock(&kvm->slots_lock);
4610 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4612 return kvm->arch.n_max_mmu_pages;
4615 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4617 struct kvm_pic *pic = kvm->arch.vpic;
4621 switch (chip->chip_id) {
4622 case KVM_IRQCHIP_PIC_MASTER:
4623 memcpy(&chip->chip.pic, &pic->pics[0],
4624 sizeof(struct kvm_pic_state));
4626 case KVM_IRQCHIP_PIC_SLAVE:
4627 memcpy(&chip->chip.pic, &pic->pics[1],
4628 sizeof(struct kvm_pic_state));
4630 case KVM_IRQCHIP_IOAPIC:
4631 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4640 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4642 struct kvm_pic *pic = kvm->arch.vpic;
4646 switch (chip->chip_id) {
4647 case KVM_IRQCHIP_PIC_MASTER:
4648 spin_lock(&pic->lock);
4649 memcpy(&pic->pics[0], &chip->chip.pic,
4650 sizeof(struct kvm_pic_state));
4651 spin_unlock(&pic->lock);
4653 case KVM_IRQCHIP_PIC_SLAVE:
4654 spin_lock(&pic->lock);
4655 memcpy(&pic->pics[1], &chip->chip.pic,
4656 sizeof(struct kvm_pic_state));
4657 spin_unlock(&pic->lock);
4659 case KVM_IRQCHIP_IOAPIC:
4660 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4666 kvm_pic_update_irq(pic);
4670 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4672 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4674 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4676 mutex_lock(&kps->lock);
4677 memcpy(ps, &kps->channels, sizeof(*ps));
4678 mutex_unlock(&kps->lock);
4682 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4685 struct kvm_pit *pit = kvm->arch.vpit;
4687 mutex_lock(&pit->pit_state.lock);
4688 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4689 for (i = 0; i < 3; i++)
4690 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4691 mutex_unlock(&pit->pit_state.lock);
4695 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4697 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4698 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4699 sizeof(ps->channels));
4700 ps->flags = kvm->arch.vpit->pit_state.flags;
4701 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4702 memset(&ps->reserved, 0, sizeof(ps->reserved));
4706 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4710 u32 prev_legacy, cur_legacy;
4711 struct kvm_pit *pit = kvm->arch.vpit;
4713 mutex_lock(&pit->pit_state.lock);
4714 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4715 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4716 if (!prev_legacy && cur_legacy)
4718 memcpy(&pit->pit_state.channels, &ps->channels,
4719 sizeof(pit->pit_state.channels));
4720 pit->pit_state.flags = ps->flags;
4721 for (i = 0; i < 3; i++)
4722 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4724 mutex_unlock(&pit->pit_state.lock);
4728 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4729 struct kvm_reinject_control *control)
4731 struct kvm_pit *pit = kvm->arch.vpit;
4733 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4734 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4735 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4737 mutex_lock(&pit->pit_state.lock);
4738 kvm_pit_set_reinject(pit, control->pit_reinject);
4739 mutex_unlock(&pit->pit_state.lock);
4745 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4746 * @kvm: kvm instance
4747 * @log: slot id and address to which we copy the log
4749 * Steps 1-4 below provide general overview of dirty page logging. See
4750 * kvm_get_dirty_log_protect() function description for additional details.
4752 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4753 * always flush the TLB (step 4) even if previous step failed and the dirty
4754 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4755 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4756 * writes will be marked dirty for next log read.
4758 * 1. Take a snapshot of the bit and clear it if needed.
4759 * 2. Write protect the corresponding page.
4760 * 3. Copy the snapshot to the userspace.
4761 * 4. Flush TLB's if needed.
4763 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4768 mutex_lock(&kvm->slots_lock);
4771 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4773 if (kvm_x86_ops->flush_log_dirty)
4774 kvm_x86_ops->flush_log_dirty(kvm);
4776 r = kvm_get_dirty_log_protect(kvm, log, &flush);
4779 * All the TLBs can be flushed out of mmu lock, see the comments in
4780 * kvm_mmu_slot_remove_write_access().
4782 lockdep_assert_held(&kvm->slots_lock);
4784 kvm_flush_remote_tlbs(kvm);
4786 mutex_unlock(&kvm->slots_lock);
4790 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4795 mutex_lock(&kvm->slots_lock);
4798 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4800 if (kvm_x86_ops->flush_log_dirty)
4801 kvm_x86_ops->flush_log_dirty(kvm);
4803 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4806 * All the TLBs can be flushed out of mmu lock, see the comments in
4807 * kvm_mmu_slot_remove_write_access().
4809 lockdep_assert_held(&kvm->slots_lock);
4811 kvm_flush_remote_tlbs(kvm);
4813 mutex_unlock(&kvm->slots_lock);
4817 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4820 if (!irqchip_in_kernel(kvm))
4823 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4824 irq_event->irq, irq_event->level,
4829 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4830 struct kvm_enable_cap *cap)
4838 case KVM_CAP_DISABLE_QUIRKS:
4839 kvm->arch.disabled_quirks = cap->args[0];
4842 case KVM_CAP_SPLIT_IRQCHIP: {
4843 mutex_lock(&kvm->lock);
4845 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4846 goto split_irqchip_unlock;
4848 if (irqchip_in_kernel(kvm))
4849 goto split_irqchip_unlock;
4850 if (kvm->created_vcpus)
4851 goto split_irqchip_unlock;
4852 r = kvm_setup_empty_irq_routing(kvm);
4854 goto split_irqchip_unlock;
4855 /* Pairs with irqchip_in_kernel. */
4857 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4858 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4860 split_irqchip_unlock:
4861 mutex_unlock(&kvm->lock);
4864 case KVM_CAP_X2APIC_API:
4866 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4869 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4870 kvm->arch.x2apic_format = true;
4871 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4872 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4876 case KVM_CAP_X86_DISABLE_EXITS:
4878 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4881 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4882 kvm_can_mwait_in_guest())
4883 kvm->arch.mwait_in_guest = true;
4884 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4885 kvm->arch.hlt_in_guest = true;
4886 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4887 kvm->arch.pause_in_guest = true;
4888 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4889 kvm->arch.cstate_in_guest = true;
4892 case KVM_CAP_MSR_PLATFORM_INFO:
4893 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4896 case KVM_CAP_EXCEPTION_PAYLOAD:
4897 kvm->arch.exception_payload_enabled = cap->args[0];
4907 long kvm_arch_vm_ioctl(struct file *filp,
4908 unsigned int ioctl, unsigned long arg)
4910 struct kvm *kvm = filp->private_data;
4911 void __user *argp = (void __user *)arg;
4914 * This union makes it completely explicit to gcc-3.x
4915 * that these two variables' stack usage should be
4916 * combined, not added together.
4919 struct kvm_pit_state ps;
4920 struct kvm_pit_state2 ps2;
4921 struct kvm_pit_config pit_config;
4925 case KVM_SET_TSS_ADDR:
4926 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4928 case KVM_SET_IDENTITY_MAP_ADDR: {
4931 mutex_lock(&kvm->lock);
4933 if (kvm->created_vcpus)
4934 goto set_identity_unlock;
4936 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4937 goto set_identity_unlock;
4938 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4939 set_identity_unlock:
4940 mutex_unlock(&kvm->lock);
4943 case KVM_SET_NR_MMU_PAGES:
4944 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4946 case KVM_GET_NR_MMU_PAGES:
4947 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4949 case KVM_CREATE_IRQCHIP: {
4950 mutex_lock(&kvm->lock);
4953 if (irqchip_in_kernel(kvm))
4954 goto create_irqchip_unlock;
4957 if (kvm->created_vcpus)
4958 goto create_irqchip_unlock;
4960 r = kvm_pic_init(kvm);
4962 goto create_irqchip_unlock;
4964 r = kvm_ioapic_init(kvm);
4966 kvm_pic_destroy(kvm);
4967 goto create_irqchip_unlock;
4970 r = kvm_setup_default_irq_routing(kvm);
4972 kvm_ioapic_destroy(kvm);
4973 kvm_pic_destroy(kvm);
4974 goto create_irqchip_unlock;
4976 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4978 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4979 create_irqchip_unlock:
4980 mutex_unlock(&kvm->lock);
4983 case KVM_CREATE_PIT:
4984 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4986 case KVM_CREATE_PIT2:
4988 if (copy_from_user(&u.pit_config, argp,
4989 sizeof(struct kvm_pit_config)))
4992 mutex_lock(&kvm->lock);
4995 goto create_pit_unlock;
4997 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
5001 mutex_unlock(&kvm->lock);
5003 case KVM_GET_IRQCHIP: {
5004 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5005 struct kvm_irqchip *chip;
5007 chip = memdup_user(argp, sizeof(*chip));
5014 if (!irqchip_kernel(kvm))
5015 goto get_irqchip_out;
5016 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
5018 goto get_irqchip_out;
5020 if (copy_to_user(argp, chip, sizeof(*chip)))
5021 goto get_irqchip_out;
5027 case KVM_SET_IRQCHIP: {
5028 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
5029 struct kvm_irqchip *chip;
5031 chip = memdup_user(argp, sizeof(*chip));
5038 if (!irqchip_kernel(kvm))
5039 goto set_irqchip_out;
5040 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
5047 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
5050 if (!kvm->arch.vpit)
5052 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
5056 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
5063 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
5066 if (!kvm->arch.vpit)
5068 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
5071 case KVM_GET_PIT2: {
5073 if (!kvm->arch.vpit)
5075 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
5079 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
5084 case KVM_SET_PIT2: {
5086 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5089 if (!kvm->arch.vpit)
5091 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5094 case KVM_REINJECT_CONTROL: {
5095 struct kvm_reinject_control control;
5097 if (copy_from_user(&control, argp, sizeof(control)))
5100 if (!kvm->arch.vpit)
5102 r = kvm_vm_ioctl_reinject(kvm, &control);
5105 case KVM_SET_BOOT_CPU_ID:
5107 mutex_lock(&kvm->lock);
5108 if (kvm->created_vcpus)
5111 kvm->arch.bsp_vcpu_id = arg;
5112 mutex_unlock(&kvm->lock);
5114 case KVM_XEN_HVM_CONFIG: {
5115 struct kvm_xen_hvm_config xhc;
5117 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5122 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5126 case KVM_SET_CLOCK: {
5127 struct kvm_clock_data user_ns;
5131 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5140 * TODO: userspace has to take care of races with VCPU_RUN, so
5141 * kvm_gen_update_masterclock() can be cut down to locked
5142 * pvclock_update_vm_gtod_copy().
5144 kvm_gen_update_masterclock(kvm);
5145 now_ns = get_kvmclock_ns(kvm);
5146 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5147 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5150 case KVM_GET_CLOCK: {
5151 struct kvm_clock_data user_ns;
5154 now_ns = get_kvmclock_ns(kvm);
5155 user_ns.clock = now_ns;
5156 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5157 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5160 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5165 case KVM_MEMORY_ENCRYPT_OP: {
5167 if (kvm_x86_ops->mem_enc_op)
5168 r = kvm_x86_ops->mem_enc_op(kvm, argp);
5171 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5172 struct kvm_enc_region region;
5175 if (copy_from_user(®ion, argp, sizeof(region)))
5179 if (kvm_x86_ops->mem_enc_reg_region)
5180 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
5183 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5184 struct kvm_enc_region region;
5187 if (copy_from_user(®ion, argp, sizeof(region)))
5191 if (kvm_x86_ops->mem_enc_unreg_region)
5192 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
5195 case KVM_HYPERV_EVENTFD: {
5196 struct kvm_hyperv_eventfd hvevfd;
5199 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5201 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5204 case KVM_SET_PMU_EVENT_FILTER:
5205 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5214 static void kvm_init_msr_list(void)
5216 struct x86_pmu_capability x86_pmu;
5220 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5221 "Please update the fixed PMCs in msrs_to_saved_all[]");
5223 perf_get_x86_pmu_capability(&x86_pmu);
5225 num_msrs_to_save = 0;
5226 num_emulated_msrs = 0;
5227 num_msr_based_features = 0;
5229 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5230 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5234 * Even MSRs that are valid in the host may not be exposed
5235 * to the guests in some cases.
5237 switch (msrs_to_save_all[i]) {
5238 case MSR_IA32_BNDCFGS:
5239 if (!kvm_mpx_supported())
5243 if (!kvm_x86_ops->rdtscp_supported())
5246 case MSR_IA32_RTIT_CTL:
5247 case MSR_IA32_RTIT_STATUS:
5248 if (!kvm_x86_ops->pt_supported())
5251 case MSR_IA32_RTIT_CR3_MATCH:
5252 if (!kvm_x86_ops->pt_supported() ||
5253 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5256 case MSR_IA32_RTIT_OUTPUT_BASE:
5257 case MSR_IA32_RTIT_OUTPUT_MASK:
5258 if (!kvm_x86_ops->pt_supported() ||
5259 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5260 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5263 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5264 if (!kvm_x86_ops->pt_supported() ||
5265 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5266 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5269 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5270 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5271 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5274 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5275 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5276 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5283 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5286 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5287 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs_all[i]))
5290 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5293 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5294 struct kvm_msr_entry msr;
5296 msr.index = msr_based_features_all[i];
5297 if (kvm_get_msr_feature(&msr))
5300 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5304 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5312 if (!(lapic_in_kernel(vcpu) &&
5313 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5314 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5325 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5332 if (!(lapic_in_kernel(vcpu) &&
5333 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5335 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5337 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5347 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5348 struct kvm_segment *var, int seg)
5350 kvm_x86_ops->set_segment(vcpu, var, seg);
5353 void kvm_get_segment(struct kvm_vcpu *vcpu,
5354 struct kvm_segment *var, int seg)
5356 kvm_x86_ops->get_segment(vcpu, var, seg);
5359 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5360 struct x86_exception *exception)
5364 BUG_ON(!mmu_is_nested(vcpu));
5366 /* NPT walks are always user-walks */
5367 access |= PFERR_USER_MASK;
5368 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5373 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5374 struct x86_exception *exception)
5376 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5377 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5380 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5381 struct x86_exception *exception)
5383 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5384 access |= PFERR_FETCH_MASK;
5385 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5388 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5389 struct x86_exception *exception)
5391 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5392 access |= PFERR_WRITE_MASK;
5393 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5396 /* uses this to access any guest's mapped memory without checking CPL */
5397 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5398 struct x86_exception *exception)
5400 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5403 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5404 struct kvm_vcpu *vcpu, u32 access,
5405 struct x86_exception *exception)
5408 int r = X86EMUL_CONTINUE;
5411 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5413 unsigned offset = addr & (PAGE_SIZE-1);
5414 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5417 if (gpa == UNMAPPED_GVA)
5418 return X86EMUL_PROPAGATE_FAULT;
5419 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5422 r = X86EMUL_IO_NEEDED;
5434 /* used for instruction fetching */
5435 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5436 gva_t addr, void *val, unsigned int bytes,
5437 struct x86_exception *exception)
5439 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5440 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5444 /* Inline kvm_read_guest_virt_helper for speed. */
5445 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5447 if (unlikely(gpa == UNMAPPED_GVA))
5448 return X86EMUL_PROPAGATE_FAULT;
5450 offset = addr & (PAGE_SIZE-1);
5451 if (WARN_ON(offset + bytes > PAGE_SIZE))
5452 bytes = (unsigned)PAGE_SIZE - offset;
5453 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5455 if (unlikely(ret < 0))
5456 return X86EMUL_IO_NEEDED;
5458 return X86EMUL_CONTINUE;
5461 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5462 gva_t addr, void *val, unsigned int bytes,
5463 struct x86_exception *exception)
5465 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5468 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5469 * is returned, but our callers are not ready for that and they blindly
5470 * call kvm_inject_page_fault. Ensure that they at least do not leak
5471 * uninitialized kernel stack memory into cr2 and error code.
5473 memset(exception, 0, sizeof(*exception));
5474 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5477 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5479 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5480 gva_t addr, void *val, unsigned int bytes,
5481 struct x86_exception *exception, bool system)
5483 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5486 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5487 access |= PFERR_USER_MASK;
5489 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5492 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5493 unsigned long addr, void *val, unsigned int bytes)
5495 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5496 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5498 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5501 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5502 struct kvm_vcpu *vcpu, u32 access,
5503 struct x86_exception *exception)
5506 int r = X86EMUL_CONTINUE;
5509 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5512 unsigned offset = addr & (PAGE_SIZE-1);
5513 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5516 if (gpa == UNMAPPED_GVA)
5517 return X86EMUL_PROPAGATE_FAULT;
5518 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5520 r = X86EMUL_IO_NEEDED;
5532 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5533 unsigned int bytes, struct x86_exception *exception,
5536 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5537 u32 access = PFERR_WRITE_MASK;
5539 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5540 access |= PFERR_USER_MASK;
5542 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5546 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5547 unsigned int bytes, struct x86_exception *exception)
5549 /* kvm_write_guest_virt_system can pull in tons of pages. */
5550 vcpu->arch.l1tf_flush_l1d = true;
5553 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5554 * is returned, but our callers are not ready for that and they blindly
5555 * call kvm_inject_page_fault. Ensure that they at least do not leak
5556 * uninitialized kernel stack memory into cr2 and error code.
5558 memset(exception, 0, sizeof(*exception));
5559 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5560 PFERR_WRITE_MASK, exception);
5562 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5564 int handle_ud(struct kvm_vcpu *vcpu)
5566 static const char kvm_emulate_prefix[] = { __KVM_EMULATE_PREFIX };
5567 int emul_type = EMULTYPE_TRAP_UD;
5568 char sig[5]; /* ud2; .ascii "kvm" */
5569 struct x86_exception e;
5571 if (force_emulation_prefix &&
5572 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5573 sig, sizeof(sig), &e) == 0 &&
5574 memcmp(sig, kvm_emulate_prefix, sizeof(sig)) == 0) {
5575 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5576 emul_type = EMULTYPE_TRAP_UD_FORCED;
5579 return kvm_emulate_instruction(vcpu, emul_type);
5581 EXPORT_SYMBOL_GPL(handle_ud);
5583 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5584 gpa_t gpa, bool write)
5586 /* For APIC access vmexit */
5587 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5590 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5591 trace_vcpu_match_mmio(gva, gpa, write, true);
5598 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5599 gpa_t *gpa, struct x86_exception *exception,
5602 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5603 | (write ? PFERR_WRITE_MASK : 0);
5606 * currently PKRU is only applied to ept enabled guest so
5607 * there is no pkey in EPT page table for L1 guest or EPT
5608 * shadow page table for L2 guest.
5610 if (vcpu_match_mmio_gva(vcpu, gva)
5611 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5612 vcpu->arch.mmio_access, 0, access)) {
5613 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5614 (gva & (PAGE_SIZE - 1));
5615 trace_vcpu_match_mmio(gva, *gpa, write, false);
5619 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5621 if (*gpa == UNMAPPED_GVA)
5624 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5627 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5628 const void *val, int bytes)
5632 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5635 kvm_page_track_write(vcpu, gpa, val, bytes);
5639 struct read_write_emulator_ops {
5640 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5642 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5643 void *val, int bytes);
5644 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5645 int bytes, void *val);
5646 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5647 void *val, int bytes);
5651 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5653 if (vcpu->mmio_read_completed) {
5654 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5655 vcpu->mmio_fragments[0].gpa, val);
5656 vcpu->mmio_read_completed = 0;
5663 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5664 void *val, int bytes)
5666 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5669 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5670 void *val, int bytes)
5672 return emulator_write_phys(vcpu, gpa, val, bytes);
5675 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5677 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5678 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5681 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5682 void *val, int bytes)
5684 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5685 return X86EMUL_IO_NEEDED;
5688 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5689 void *val, int bytes)
5691 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5693 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5694 return X86EMUL_CONTINUE;
5697 static const struct read_write_emulator_ops read_emultor = {
5698 .read_write_prepare = read_prepare,
5699 .read_write_emulate = read_emulate,
5700 .read_write_mmio = vcpu_mmio_read,
5701 .read_write_exit_mmio = read_exit_mmio,
5704 static const struct read_write_emulator_ops write_emultor = {
5705 .read_write_emulate = write_emulate,
5706 .read_write_mmio = write_mmio,
5707 .read_write_exit_mmio = write_exit_mmio,
5711 static int emulator_read_write_onepage(unsigned long addr, void *val,
5713 struct x86_exception *exception,
5714 struct kvm_vcpu *vcpu,
5715 const struct read_write_emulator_ops *ops)
5719 bool write = ops->write;
5720 struct kvm_mmio_fragment *frag;
5721 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5724 * If the exit was due to a NPF we may already have a GPA.
5725 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5726 * Note, this cannot be used on string operations since string
5727 * operation using rep will only have the initial GPA from the NPF
5730 if (vcpu->arch.gpa_available &&
5731 emulator_can_use_gpa(ctxt) &&
5732 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5733 gpa = vcpu->arch.gpa_val;
5734 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5736 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5738 return X86EMUL_PROPAGATE_FAULT;
5741 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5742 return X86EMUL_CONTINUE;
5745 * Is this MMIO handled locally?
5747 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5748 if (handled == bytes)
5749 return X86EMUL_CONTINUE;
5755 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5756 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5760 return X86EMUL_CONTINUE;
5763 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5765 void *val, unsigned int bytes,
5766 struct x86_exception *exception,
5767 const struct read_write_emulator_ops *ops)
5769 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5773 if (ops->read_write_prepare &&
5774 ops->read_write_prepare(vcpu, val, bytes))
5775 return X86EMUL_CONTINUE;
5777 vcpu->mmio_nr_fragments = 0;
5779 /* Crossing a page boundary? */
5780 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5783 now = -addr & ~PAGE_MASK;
5784 rc = emulator_read_write_onepage(addr, val, now, exception,
5787 if (rc != X86EMUL_CONTINUE)
5790 if (ctxt->mode != X86EMUL_MODE_PROT64)
5796 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5798 if (rc != X86EMUL_CONTINUE)
5801 if (!vcpu->mmio_nr_fragments)
5804 gpa = vcpu->mmio_fragments[0].gpa;
5806 vcpu->mmio_needed = 1;
5807 vcpu->mmio_cur_fragment = 0;
5809 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5810 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5811 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5812 vcpu->run->mmio.phys_addr = gpa;
5814 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5817 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5821 struct x86_exception *exception)
5823 return emulator_read_write(ctxt, addr, val, bytes,
5824 exception, &read_emultor);
5827 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5831 struct x86_exception *exception)
5833 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5834 exception, &write_emultor);
5837 #define CMPXCHG_TYPE(t, ptr, old, new) \
5838 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5840 #ifdef CONFIG_X86_64
5841 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5843 # define CMPXCHG64(ptr, old, new) \
5844 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5847 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5852 struct x86_exception *exception)
5854 struct kvm_host_map map;
5855 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5860 /* guests cmpxchg8b have to be emulated atomically */
5861 if (bytes > 8 || (bytes & (bytes - 1)))
5864 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5866 if (gpa == UNMAPPED_GVA ||
5867 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5870 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5873 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5876 kaddr = map.hva + offset_in_page(gpa);
5880 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5883 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5886 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5889 exchanged = CMPXCHG64(kaddr, old, new);
5895 kvm_vcpu_unmap(vcpu, &map, true);
5898 return X86EMUL_CMPXCHG_FAILED;
5900 kvm_page_track_write(vcpu, gpa, new, bytes);
5902 return X86EMUL_CONTINUE;
5905 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5907 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5910 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5914 for (i = 0; i < vcpu->arch.pio.count; i++) {
5915 if (vcpu->arch.pio.in)
5916 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5917 vcpu->arch.pio.size, pd);
5919 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5920 vcpu->arch.pio.port, vcpu->arch.pio.size,
5924 pd += vcpu->arch.pio.size;
5929 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5930 unsigned short port, void *val,
5931 unsigned int count, bool in)
5933 vcpu->arch.pio.port = port;
5934 vcpu->arch.pio.in = in;
5935 vcpu->arch.pio.count = count;
5936 vcpu->arch.pio.size = size;
5938 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5939 vcpu->arch.pio.count = 0;
5943 vcpu->run->exit_reason = KVM_EXIT_IO;
5944 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5945 vcpu->run->io.size = size;
5946 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5947 vcpu->run->io.count = count;
5948 vcpu->run->io.port = port;
5953 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5954 int size, unsigned short port, void *val,
5957 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5960 if (vcpu->arch.pio.count)
5963 memset(vcpu->arch.pio_data, 0, size * count);
5965 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5968 memcpy(val, vcpu->arch.pio_data, size * count);
5969 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5970 vcpu->arch.pio.count = 0;
5977 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5978 int size, unsigned short port,
5979 const void *val, unsigned int count)
5981 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5983 memcpy(vcpu->arch.pio_data, val, size * count);
5984 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5985 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5988 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5990 return kvm_x86_ops->get_segment_base(vcpu, seg);
5993 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5995 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5998 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
6000 if (!need_emulate_wbinvd(vcpu))
6001 return X86EMUL_CONTINUE;
6003 if (kvm_x86_ops->has_wbinvd_exit()) {
6004 int cpu = get_cpu();
6006 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
6007 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
6008 wbinvd_ipi, NULL, 1);
6010 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
6013 return X86EMUL_CONTINUE;
6016 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
6018 kvm_emulate_wbinvd_noskip(vcpu);
6019 return kvm_skip_emulated_instruction(vcpu);
6021 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
6025 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
6027 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
6030 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
6031 unsigned long *dest)
6033 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
6036 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
6037 unsigned long value)
6040 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
6043 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
6045 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
6048 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
6050 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6051 unsigned long value;
6055 value = kvm_read_cr0(vcpu);
6058 value = vcpu->arch.cr2;
6061 value = kvm_read_cr3(vcpu);
6064 value = kvm_read_cr4(vcpu);
6067 value = kvm_get_cr8(vcpu);
6070 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6077 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
6079 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6084 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
6087 vcpu->arch.cr2 = val;
6090 res = kvm_set_cr3(vcpu, val);
6093 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6096 res = kvm_set_cr8(vcpu, val);
6099 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6106 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6108 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
6111 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6113 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
6116 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6118 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
6121 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6123 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
6126 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6128 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
6131 static unsigned long emulator_get_cached_segment_base(
6132 struct x86_emulate_ctxt *ctxt, int seg)
6134 return get_segment_base(emul_to_vcpu(ctxt), seg);
6137 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6138 struct desc_struct *desc, u32 *base3,
6141 struct kvm_segment var;
6143 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6144 *selector = var.selector;
6147 memset(desc, 0, sizeof(*desc));
6155 set_desc_limit(desc, var.limit);
6156 set_desc_base(desc, (unsigned long)var.base);
6157 #ifdef CONFIG_X86_64
6159 *base3 = var.base >> 32;
6161 desc->type = var.type;
6163 desc->dpl = var.dpl;
6164 desc->p = var.present;
6165 desc->avl = var.avl;
6173 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6174 struct desc_struct *desc, u32 base3,
6177 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6178 struct kvm_segment var;
6180 var.selector = selector;
6181 var.base = get_desc_base(desc);
6182 #ifdef CONFIG_X86_64
6183 var.base |= ((u64)base3) << 32;
6185 var.limit = get_desc_limit(desc);
6187 var.limit = (var.limit << 12) | 0xfff;
6188 var.type = desc->type;
6189 var.dpl = desc->dpl;
6194 var.avl = desc->avl;
6195 var.present = desc->p;
6196 var.unusable = !var.present;
6199 kvm_set_segment(vcpu, &var, seg);
6203 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6204 u32 msr_index, u64 *pdata)
6206 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6209 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6210 u32 msr_index, u64 data)
6212 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6215 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6217 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6219 return vcpu->arch.smbase;
6222 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6224 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6226 vcpu->arch.smbase = smbase;
6229 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6232 return kvm_pmu_is_valid_rdpmc_ecx(emul_to_vcpu(ctxt), pmc);
6235 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6236 u32 pmc, u64 *pdata)
6238 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6241 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6243 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6246 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6247 struct x86_instruction_info *info,
6248 enum x86_intercept_stage stage)
6250 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6253 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6254 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6256 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6259 static bool emulator_guest_has_long_mode(struct x86_emulate_ctxt *ctxt)
6261 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_LM);
6264 static bool emulator_guest_has_movbe(struct x86_emulate_ctxt *ctxt)
6266 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_MOVBE);
6269 static bool emulator_guest_has_fxsr(struct x86_emulate_ctxt *ctxt)
6271 return guest_cpuid_has(emul_to_vcpu(ctxt), X86_FEATURE_FXSR);
6274 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6276 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6279 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6281 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6284 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6286 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6289 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6291 return emul_to_vcpu(ctxt)->arch.hflags;
6294 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6296 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6299 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6300 const char *smstate)
6302 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6305 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6307 kvm_smm_changed(emul_to_vcpu(ctxt));
6310 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6312 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6315 static const struct x86_emulate_ops emulate_ops = {
6316 .read_gpr = emulator_read_gpr,
6317 .write_gpr = emulator_write_gpr,
6318 .read_std = emulator_read_std,
6319 .write_std = emulator_write_std,
6320 .read_phys = kvm_read_guest_phys_system,
6321 .fetch = kvm_fetch_guest_virt,
6322 .read_emulated = emulator_read_emulated,
6323 .write_emulated = emulator_write_emulated,
6324 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6325 .invlpg = emulator_invlpg,
6326 .pio_in_emulated = emulator_pio_in_emulated,
6327 .pio_out_emulated = emulator_pio_out_emulated,
6328 .get_segment = emulator_get_segment,
6329 .set_segment = emulator_set_segment,
6330 .get_cached_segment_base = emulator_get_cached_segment_base,
6331 .get_gdt = emulator_get_gdt,
6332 .get_idt = emulator_get_idt,
6333 .set_gdt = emulator_set_gdt,
6334 .set_idt = emulator_set_idt,
6335 .get_cr = emulator_get_cr,
6336 .set_cr = emulator_set_cr,
6337 .cpl = emulator_get_cpl,
6338 .get_dr = emulator_get_dr,
6339 .set_dr = emulator_set_dr,
6340 .get_smbase = emulator_get_smbase,
6341 .set_smbase = emulator_set_smbase,
6342 .set_msr = emulator_set_msr,
6343 .get_msr = emulator_get_msr,
6344 .check_pmc = emulator_check_pmc,
6345 .read_pmc = emulator_read_pmc,
6346 .halt = emulator_halt,
6347 .wbinvd = emulator_wbinvd,
6348 .fix_hypercall = emulator_fix_hypercall,
6349 .intercept = emulator_intercept,
6350 .get_cpuid = emulator_get_cpuid,
6351 .guest_has_long_mode = emulator_guest_has_long_mode,
6352 .guest_has_movbe = emulator_guest_has_movbe,
6353 .guest_has_fxsr = emulator_guest_has_fxsr,
6354 .set_nmi_mask = emulator_set_nmi_mask,
6355 .get_hflags = emulator_get_hflags,
6356 .set_hflags = emulator_set_hflags,
6357 .pre_leave_smm = emulator_pre_leave_smm,
6358 .post_leave_smm = emulator_post_leave_smm,
6359 .set_xcr = emulator_set_xcr,
6362 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6364 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6366 * an sti; sti; sequence only disable interrupts for the first
6367 * instruction. So, if the last instruction, be it emulated or
6368 * not, left the system with the INT_STI flag enabled, it
6369 * means that the last instruction is an sti. We should not
6370 * leave the flag on in this case. The same goes for mov ss
6372 if (int_shadow & mask)
6374 if (unlikely(int_shadow || mask)) {
6375 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6377 kvm_make_request(KVM_REQ_EVENT, vcpu);
6381 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6383 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6384 if (ctxt->exception.vector == PF_VECTOR)
6385 return kvm_propagate_fault(vcpu, &ctxt->exception);
6387 if (ctxt->exception.error_code_valid)
6388 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6389 ctxt->exception.error_code);
6391 kvm_queue_exception(vcpu, ctxt->exception.vector);
6395 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6397 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6400 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6402 ctxt->eflags = kvm_get_rflags(vcpu);
6403 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6405 ctxt->eip = kvm_rip_read(vcpu);
6406 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6407 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6408 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6409 cs_db ? X86EMUL_MODE_PROT32 :
6410 X86EMUL_MODE_PROT16;
6411 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6412 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6413 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6415 init_decode_cache(ctxt);
6416 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6419 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6421 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6424 init_emulate_ctxt(vcpu);
6428 ctxt->_eip = ctxt->eip + inc_eip;
6429 ret = emulate_int_real(ctxt, irq);
6431 if (ret != X86EMUL_CONTINUE) {
6432 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6434 ctxt->eip = ctxt->_eip;
6435 kvm_rip_write(vcpu, ctxt->eip);
6436 kvm_set_rflags(vcpu, ctxt->eflags);
6439 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6441 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6443 ++vcpu->stat.insn_emulation_fail;
6444 trace_kvm_emulate_insn_failed(vcpu);
6446 if (emulation_type & EMULTYPE_VMWARE_GP) {
6447 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6451 if (emulation_type & EMULTYPE_SKIP) {
6452 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6453 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6454 vcpu->run->internal.ndata = 0;
6458 kvm_queue_exception(vcpu, UD_VECTOR);
6460 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6461 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6462 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6463 vcpu->run->internal.ndata = 0;
6470 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6471 bool write_fault_to_shadow_pgtable,
6474 gpa_t gpa = cr2_or_gpa;
6477 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6480 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6483 if (!vcpu->arch.mmu->direct_map) {
6485 * Write permission should be allowed since only
6486 * write access need to be emulated.
6488 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6491 * If the mapping is invalid in guest, let cpu retry
6492 * it to generate fault.
6494 if (gpa == UNMAPPED_GVA)
6499 * Do not retry the unhandleable instruction if it faults on the
6500 * readonly host memory, otherwise it will goto a infinite loop:
6501 * retry instruction -> write #PF -> emulation fail -> retry
6502 * instruction -> ...
6504 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6507 * If the instruction failed on the error pfn, it can not be fixed,
6508 * report the error to userspace.
6510 if (is_error_noslot_pfn(pfn))
6513 kvm_release_pfn_clean(pfn);
6515 /* The instructions are well-emulated on direct mmu. */
6516 if (vcpu->arch.mmu->direct_map) {
6517 unsigned int indirect_shadow_pages;
6519 spin_lock(&vcpu->kvm->mmu_lock);
6520 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6521 spin_unlock(&vcpu->kvm->mmu_lock);
6523 if (indirect_shadow_pages)
6524 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6530 * if emulation was due to access to shadowed page table
6531 * and it failed try to unshadow page and re-enter the
6532 * guest to let CPU execute the instruction.
6534 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6537 * If the access faults on its page table, it can not
6538 * be fixed by unprotecting shadow page and it should
6539 * be reported to userspace.
6541 return !write_fault_to_shadow_pgtable;
6544 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6545 gpa_t cr2_or_gpa, int emulation_type)
6547 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6548 unsigned long last_retry_eip, last_retry_addr, gpa = cr2_or_gpa;
6550 last_retry_eip = vcpu->arch.last_retry_eip;
6551 last_retry_addr = vcpu->arch.last_retry_addr;
6554 * If the emulation is caused by #PF and it is non-page_table
6555 * writing instruction, it means the VM-EXIT is caused by shadow
6556 * page protected, we can zap the shadow page and retry this
6557 * instruction directly.
6559 * Note: if the guest uses a non-page-table modifying instruction
6560 * on the PDE that points to the instruction, then we will unmap
6561 * the instruction and go to an infinite loop. So, we cache the
6562 * last retried eip and the last fault address, if we meet the eip
6563 * and the address again, we can break out of the potential infinite
6566 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6568 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6571 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6574 if (x86_page_table_writing_insn(ctxt))
6577 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2_or_gpa)
6580 vcpu->arch.last_retry_eip = ctxt->eip;
6581 vcpu->arch.last_retry_addr = cr2_or_gpa;
6583 if (!vcpu->arch.mmu->direct_map)
6584 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2_or_gpa, NULL);
6586 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6591 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6592 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6594 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6596 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6597 /* This is a good place to trace that we are exiting SMM. */
6598 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6600 /* Process a latched INIT or SMI, if any. */
6601 kvm_make_request(KVM_REQ_EVENT, vcpu);
6604 kvm_mmu_reset_context(vcpu);
6607 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6616 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6617 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6622 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6624 struct kvm_run *kvm_run = vcpu->run;
6626 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6627 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6628 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6629 kvm_run->debug.arch.exception = DB_VECTOR;
6630 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6633 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6637 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6639 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6642 r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6647 * rflags is the old, "raw" value of the flags. The new value has
6648 * not been saved yet.
6650 * This is correct even for TF set by the guest, because "the
6651 * processor will not generate this exception after the instruction
6652 * that sets the TF flag".
6654 if (unlikely(rflags & X86_EFLAGS_TF))
6655 r = kvm_vcpu_do_singlestep(vcpu);
6658 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6660 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6662 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6663 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6664 struct kvm_run *kvm_run = vcpu->run;
6665 unsigned long eip = kvm_get_linear_rip(vcpu);
6666 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6667 vcpu->arch.guest_debug_dr7,
6671 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6672 kvm_run->debug.arch.pc = eip;
6673 kvm_run->debug.arch.exception = DB_VECTOR;
6674 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6680 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6681 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6682 unsigned long eip = kvm_get_linear_rip(vcpu);
6683 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6688 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6689 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6690 kvm_queue_exception(vcpu, DB_VECTOR);
6699 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6701 switch (ctxt->opcode_len) {
6708 case 0xe6: /* OUT */
6712 case 0x6c: /* INS */
6714 case 0x6e: /* OUTS */
6721 case 0x33: /* RDPMC */
6730 int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
6731 int emulation_type, void *insn, int insn_len)
6734 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6735 bool writeback = true;
6736 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6738 vcpu->arch.l1tf_flush_l1d = true;
6741 * Clear write_fault_to_shadow_pgtable here to ensure it is
6744 vcpu->arch.write_fault_to_shadow_pgtable = false;
6745 kvm_clear_exception_queue(vcpu);
6747 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6748 init_emulate_ctxt(vcpu);
6751 * We will reenter on the same instruction since
6752 * we do not set complete_userspace_io. This does not
6753 * handle watchpoints yet, those would be handled in
6756 if (!(emulation_type & EMULTYPE_SKIP) &&
6757 kvm_vcpu_check_breakpoint(vcpu, &r))
6760 ctxt->interruptibility = 0;
6761 ctxt->have_exception = false;
6762 ctxt->exception.vector = -1;
6763 ctxt->perm_ok = false;
6765 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6767 r = x86_decode_insn(ctxt, insn, insn_len);
6769 trace_kvm_emulate_insn_start(vcpu);
6770 ++vcpu->stat.insn_emulation;
6771 if (r != EMULATION_OK) {
6772 if ((emulation_type & EMULTYPE_TRAP_UD) ||
6773 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6774 kvm_queue_exception(vcpu, UD_VECTOR);
6777 if (reexecute_instruction(vcpu, cr2_or_gpa,
6781 if (ctxt->have_exception) {
6783 * #UD should result in just EMULATION_FAILED, and trap-like
6784 * exception should not be encountered during decode.
6786 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6787 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6788 inject_emulated_exception(vcpu);
6791 return handle_emulation_failure(vcpu, emulation_type);
6795 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6796 !is_vmware_backdoor_opcode(ctxt)) {
6797 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6802 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6803 * for kvm_skip_emulated_instruction(). The caller is responsible for
6804 * updating interruptibility state and injecting single-step #DBs.
6806 if (emulation_type & EMULTYPE_SKIP) {
6807 kvm_rip_write(vcpu, ctxt->_eip);
6808 if (ctxt->eflags & X86_EFLAGS_RF)
6809 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6813 if (retry_instruction(ctxt, cr2_or_gpa, emulation_type))
6816 /* this is needed for vmware backdoor interface to work since it
6817 changes registers values during IO operation */
6818 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6819 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6820 emulator_invalidate_register_cache(ctxt);
6824 /* Save the faulting GPA (cr2) in the address field */
6825 ctxt->exception.address = cr2_or_gpa;
6827 r = x86_emulate_insn(ctxt);
6829 if (r == EMULATION_INTERCEPTED)
6832 if (r == EMULATION_FAILED) {
6833 if (reexecute_instruction(vcpu, cr2_or_gpa, write_fault_to_spt,
6837 return handle_emulation_failure(vcpu, emulation_type);
6840 if (ctxt->have_exception) {
6842 if (inject_emulated_exception(vcpu))
6844 } else if (vcpu->arch.pio.count) {
6845 if (!vcpu->arch.pio.in) {
6846 /* FIXME: return into emulator if single-stepping. */
6847 vcpu->arch.pio.count = 0;
6850 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6853 } else if (vcpu->mmio_needed) {
6854 ++vcpu->stat.mmio_exits;
6856 if (!vcpu->mmio_is_write)
6859 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6860 } else if (r == EMULATION_RESTART)
6866 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6867 toggle_interruptibility(vcpu, ctxt->interruptibility);
6868 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6869 if (!ctxt->have_exception ||
6870 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6871 kvm_rip_write(vcpu, ctxt->eip);
6873 r = kvm_vcpu_do_singlestep(vcpu);
6874 __kvm_set_rflags(vcpu, ctxt->eflags);
6878 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6879 * do nothing, and it will be requested again as soon as
6880 * the shadow expires. But we still need to check here,
6881 * because POPF has no interrupt shadow.
6883 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6884 kvm_make_request(KVM_REQ_EVENT, vcpu);
6886 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6891 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6893 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6895 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6897 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6898 void *insn, int insn_len)
6900 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6902 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6904 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6906 vcpu->arch.pio.count = 0;
6910 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6912 vcpu->arch.pio.count = 0;
6914 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6917 return kvm_skip_emulated_instruction(vcpu);
6920 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6921 unsigned short port)
6923 unsigned long val = kvm_rax_read(vcpu);
6924 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6925 size, port, &val, 1);
6930 * Workaround userspace that relies on old KVM behavior of %rip being
6931 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6934 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6935 vcpu->arch.complete_userspace_io =
6936 complete_fast_pio_out_port_0x7e;
6937 kvm_skip_emulated_instruction(vcpu);
6939 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6940 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6945 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6949 /* We should only ever be called with arch.pio.count equal to 1 */
6950 BUG_ON(vcpu->arch.pio.count != 1);
6952 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6953 vcpu->arch.pio.count = 0;
6957 /* For size less than 4 we merge, else we zero extend */
6958 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6961 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6962 * the copy and tracing
6964 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6965 vcpu->arch.pio.port, &val, 1);
6966 kvm_rax_write(vcpu, val);
6968 return kvm_skip_emulated_instruction(vcpu);
6971 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6972 unsigned short port)
6977 /* For size less than 4 we merge, else we zero extend */
6978 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6980 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6983 kvm_rax_write(vcpu, val);
6987 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6988 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6993 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6998 ret = kvm_fast_pio_in(vcpu, size, port);
7000 ret = kvm_fast_pio_out(vcpu, size, port);
7001 return ret && kvm_skip_emulated_instruction(vcpu);
7003 EXPORT_SYMBOL_GPL(kvm_fast_pio);
7005 static int kvmclock_cpu_down_prep(unsigned int cpu)
7007 __this_cpu_write(cpu_tsc_khz, 0);
7011 static void tsc_khz_changed(void *data)
7013 struct cpufreq_freqs *freq = data;
7014 unsigned long khz = 0;
7018 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7019 khz = cpufreq_quick_get(raw_smp_processor_id());
7022 __this_cpu_write(cpu_tsc_khz, khz);
7025 #ifdef CONFIG_X86_64
7026 static void kvm_hyperv_tsc_notifier(void)
7029 struct kvm_vcpu *vcpu;
7032 mutex_lock(&kvm_lock);
7033 list_for_each_entry(kvm, &vm_list, vm_list)
7034 kvm_make_mclock_inprogress_request(kvm);
7036 hyperv_stop_tsc_emulation();
7038 /* TSC frequency always matches when on Hyper-V */
7039 for_each_present_cpu(cpu)
7040 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
7041 kvm_max_guest_tsc_khz = tsc_khz;
7043 list_for_each_entry(kvm, &vm_list, vm_list) {
7044 struct kvm_arch *ka = &kvm->arch;
7046 spin_lock(&ka->pvclock_gtod_sync_lock);
7048 pvclock_update_vm_gtod_copy(kvm);
7050 kvm_for_each_vcpu(cpu, vcpu, kvm)
7051 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7053 kvm_for_each_vcpu(cpu, vcpu, kvm)
7054 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
7056 spin_unlock(&ka->pvclock_gtod_sync_lock);
7058 mutex_unlock(&kvm_lock);
7062 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
7065 struct kvm_vcpu *vcpu;
7066 int i, send_ipi = 0;
7069 * We allow guests to temporarily run on slowing clocks,
7070 * provided we notify them after, or to run on accelerating
7071 * clocks, provided we notify them before. Thus time never
7074 * However, we have a problem. We can't atomically update
7075 * the frequency of a given CPU from this function; it is
7076 * merely a notifier, which can be called from any CPU.
7077 * Changing the TSC frequency at arbitrary points in time
7078 * requires a recomputation of local variables related to
7079 * the TSC for each VCPU. We must flag these local variables
7080 * to be updated and be sure the update takes place with the
7081 * new frequency before any guests proceed.
7083 * Unfortunately, the combination of hotplug CPU and frequency
7084 * change creates an intractable locking scenario; the order
7085 * of when these callouts happen is undefined with respect to
7086 * CPU hotplug, and they can race with each other. As such,
7087 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
7088 * undefined; you can actually have a CPU frequency change take
7089 * place in between the computation of X and the setting of the
7090 * variable. To protect against this problem, all updates of
7091 * the per_cpu tsc_khz variable are done in an interrupt
7092 * protected IPI, and all callers wishing to update the value
7093 * must wait for a synchronous IPI to complete (which is trivial
7094 * if the caller is on the CPU already). This establishes the
7095 * necessary total order on variable updates.
7097 * Note that because a guest time update may take place
7098 * anytime after the setting of the VCPU's request bit, the
7099 * correct TSC value must be set before the request. However,
7100 * to ensure the update actually makes it to any guest which
7101 * starts running in hardware virtualization between the set
7102 * and the acquisition of the spinlock, we must also ping the
7103 * CPU after setting the request bit.
7107 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7109 mutex_lock(&kvm_lock);
7110 list_for_each_entry(kvm, &vm_list, vm_list) {
7111 kvm_for_each_vcpu(i, vcpu, kvm) {
7112 if (vcpu->cpu != cpu)
7114 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7115 if (vcpu->cpu != raw_smp_processor_id())
7119 mutex_unlock(&kvm_lock);
7121 if (freq->old < freq->new && send_ipi) {
7123 * We upscale the frequency. Must make the guest
7124 * doesn't see old kvmclock values while running with
7125 * the new frequency, otherwise we risk the guest sees
7126 * time go backwards.
7128 * In case we update the frequency for another cpu
7129 * (which might be in guest context) send an interrupt
7130 * to kick the cpu out of guest context. Next time
7131 * guest context is entered kvmclock will be updated,
7132 * so the guest will not see stale values.
7134 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7138 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7141 struct cpufreq_freqs *freq = data;
7144 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7146 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7149 for_each_cpu(cpu, freq->policy->cpus)
7150 __kvmclock_cpufreq_notifier(freq, cpu);
7155 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7156 .notifier_call = kvmclock_cpufreq_notifier
7159 static int kvmclock_cpu_online(unsigned int cpu)
7161 tsc_khz_changed(NULL);
7165 static void kvm_timer_init(void)
7167 max_tsc_khz = tsc_khz;
7169 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7170 #ifdef CONFIG_CPU_FREQ
7171 struct cpufreq_policy policy;
7174 memset(&policy, 0, sizeof(policy));
7176 cpufreq_get_policy(&policy, cpu);
7177 if (policy.cpuinfo.max_freq)
7178 max_tsc_khz = policy.cpuinfo.max_freq;
7181 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7182 CPUFREQ_TRANSITION_NOTIFIER);
7185 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7186 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7189 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7190 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7192 int kvm_is_in_guest(void)
7194 return __this_cpu_read(current_vcpu) != NULL;
7197 static int kvm_is_user_mode(void)
7201 if (__this_cpu_read(current_vcpu))
7202 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7204 return user_mode != 0;
7207 static unsigned long kvm_get_guest_ip(void)
7209 unsigned long ip = 0;
7211 if (__this_cpu_read(current_vcpu))
7212 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7217 static void kvm_handle_intel_pt_intr(void)
7219 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7221 kvm_make_request(KVM_REQ_PMI, vcpu);
7222 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7223 (unsigned long *)&vcpu->arch.pmu.global_status);
7226 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7227 .is_in_guest = kvm_is_in_guest,
7228 .is_user_mode = kvm_is_user_mode,
7229 .get_guest_ip = kvm_get_guest_ip,
7230 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7233 #ifdef CONFIG_X86_64
7234 static void pvclock_gtod_update_fn(struct work_struct *work)
7238 struct kvm_vcpu *vcpu;
7241 mutex_lock(&kvm_lock);
7242 list_for_each_entry(kvm, &vm_list, vm_list)
7243 kvm_for_each_vcpu(i, vcpu, kvm)
7244 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7245 atomic_set(&kvm_guest_has_master_clock, 0);
7246 mutex_unlock(&kvm_lock);
7249 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7252 * Notification about pvclock gtod data update.
7254 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7257 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7258 struct timekeeper *tk = priv;
7260 update_pvclock_gtod(tk);
7262 /* disable master clock if host does not trust, or does not
7263 * use, TSC based clocksource.
7265 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7266 atomic_read(&kvm_guest_has_master_clock) != 0)
7267 queue_work(system_long_wq, &pvclock_gtod_work);
7272 static struct notifier_block pvclock_gtod_notifier = {
7273 .notifier_call = pvclock_gtod_notify,
7277 int kvm_arch_init(void *opaque)
7280 struct kvm_x86_ops *ops = opaque;
7283 printk(KERN_ERR "kvm: already loaded the other module\n");
7288 if (!ops->cpu_has_kvm_support()) {
7289 printk(KERN_ERR "kvm: no hardware support\n");
7293 if (ops->disabled_by_bios()) {
7294 printk(KERN_ERR "kvm: disabled by bios\n");
7300 * KVM explicitly assumes that the guest has an FPU and
7301 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7302 * vCPU's FPU state as a fxregs_state struct.
7304 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7305 printk(KERN_ERR "kvm: inadequate fpu\n");
7311 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7312 __alignof__(struct fpu), SLAB_ACCOUNT,
7314 if (!x86_fpu_cache) {
7315 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7319 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7321 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7322 goto out_free_x86_fpu_cache;
7325 r = kvm_mmu_module_init();
7327 goto out_free_percpu;
7331 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7332 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7333 PT_PRESENT_MASK, 0, sme_me_mask);
7336 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7338 if (boot_cpu_has(X86_FEATURE_XSAVE))
7339 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7342 if (pi_inject_timer == -1)
7343 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7344 #ifdef CONFIG_X86_64
7345 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7347 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7348 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7354 free_percpu(shared_msrs);
7355 out_free_x86_fpu_cache:
7356 kmem_cache_destroy(x86_fpu_cache);
7361 void kvm_arch_exit(void)
7363 #ifdef CONFIG_X86_64
7364 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7365 clear_hv_tscchange_cb();
7368 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7370 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7371 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7372 CPUFREQ_TRANSITION_NOTIFIER);
7373 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7374 #ifdef CONFIG_X86_64
7375 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7378 kvm_mmu_module_exit();
7379 free_percpu(shared_msrs);
7380 kmem_cache_destroy(x86_fpu_cache);
7383 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7385 ++vcpu->stat.halt_exits;
7386 if (lapic_in_kernel(vcpu)) {
7387 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7390 vcpu->run->exit_reason = KVM_EXIT_HLT;
7394 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7396 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7398 int ret = kvm_skip_emulated_instruction(vcpu);
7400 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7401 * KVM_EXIT_DEBUG here.
7403 return kvm_vcpu_halt(vcpu) && ret;
7405 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7407 #ifdef CONFIG_X86_64
7408 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7409 unsigned long clock_type)
7411 struct kvm_clock_pairing clock_pairing;
7412 struct timespec64 ts;
7416 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7417 return -KVM_EOPNOTSUPP;
7419 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7420 return -KVM_EOPNOTSUPP;
7422 clock_pairing.sec = ts.tv_sec;
7423 clock_pairing.nsec = ts.tv_nsec;
7424 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7425 clock_pairing.flags = 0;
7426 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7429 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7430 sizeof(struct kvm_clock_pairing)))
7438 * kvm_pv_kick_cpu_op: Kick a vcpu.
7440 * @apicid - apicid of vcpu to be kicked.
7442 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7444 struct kvm_lapic_irq lapic_irq;
7446 lapic_irq.shorthand = APIC_DEST_NOSHORT;
7447 lapic_irq.dest_mode = APIC_DEST_PHYSICAL;
7448 lapic_irq.level = 0;
7449 lapic_irq.dest_id = apicid;
7450 lapic_irq.msi_redir_hint = false;
7452 lapic_irq.delivery_mode = APIC_DM_REMRD;
7453 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7456 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7458 if (!lapic_in_kernel(vcpu)) {
7459 WARN_ON_ONCE(vcpu->arch.apicv_active);
7462 if (!vcpu->arch.apicv_active)
7465 vcpu->arch.apicv_active = false;
7466 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7469 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7471 struct kvm_vcpu *target = NULL;
7472 struct kvm_apic_map *map;
7475 map = rcu_dereference(kvm->arch.apic_map);
7477 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7478 target = map->phys_map[dest_id]->vcpu;
7482 if (target && READ_ONCE(target->ready))
7483 kvm_vcpu_yield_to(target);
7486 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7488 unsigned long nr, a0, a1, a2, a3, ret;
7491 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7492 return kvm_hv_hypercall(vcpu);
7494 nr = kvm_rax_read(vcpu);
7495 a0 = kvm_rbx_read(vcpu);
7496 a1 = kvm_rcx_read(vcpu);
7497 a2 = kvm_rdx_read(vcpu);
7498 a3 = kvm_rsi_read(vcpu);
7500 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7502 op_64_bit = is_64_bit_mode(vcpu);
7511 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7517 case KVM_HC_VAPIC_POLL_IRQ:
7520 case KVM_HC_KICK_CPU:
7521 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7522 kvm_sched_yield(vcpu->kvm, a1);
7525 #ifdef CONFIG_X86_64
7526 case KVM_HC_CLOCK_PAIRING:
7527 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7530 case KVM_HC_SEND_IPI:
7531 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7533 case KVM_HC_SCHED_YIELD:
7534 kvm_sched_yield(vcpu->kvm, a0);
7544 kvm_rax_write(vcpu, ret);
7546 ++vcpu->stat.hypercalls;
7547 return kvm_skip_emulated_instruction(vcpu);
7549 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7551 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7553 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7554 char instruction[3];
7555 unsigned long rip = kvm_rip_read(vcpu);
7557 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7559 return emulator_write_emulated(ctxt, rip, instruction, 3,
7563 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7565 return vcpu->run->request_interrupt_window &&
7566 likely(!pic_in_kernel(vcpu->kvm));
7569 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7571 struct kvm_run *kvm_run = vcpu->run;
7573 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7574 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7575 kvm_run->cr8 = kvm_get_cr8(vcpu);
7576 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7577 kvm_run->ready_for_interrupt_injection =
7578 pic_in_kernel(vcpu->kvm) ||
7579 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7582 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7586 if (!kvm_x86_ops->update_cr8_intercept)
7589 if (!lapic_in_kernel(vcpu))
7592 if (vcpu->arch.apicv_active)
7595 if (!vcpu->arch.apic->vapic_addr)
7596 max_irr = kvm_lapic_find_highest_irr(vcpu);
7603 tpr = kvm_lapic_get_cr8(vcpu);
7605 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7608 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7612 /* try to reinject previous events if any */
7614 if (vcpu->arch.exception.injected)
7615 kvm_x86_ops->queue_exception(vcpu);
7617 * Do not inject an NMI or interrupt if there is a pending
7618 * exception. Exceptions and interrupts are recognized at
7619 * instruction boundaries, i.e. the start of an instruction.
7620 * Trap-like exceptions, e.g. #DB, have higher priority than
7621 * NMIs and interrupts, i.e. traps are recognized before an
7622 * NMI/interrupt that's pending on the same instruction.
7623 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7624 * priority, but are only generated (pended) during instruction
7625 * execution, i.e. a pending fault-like exception means the
7626 * fault occurred on the *previous* instruction and must be
7627 * serviced prior to recognizing any new events in order to
7628 * fully complete the previous instruction.
7630 else if (!vcpu->arch.exception.pending) {
7631 if (vcpu->arch.nmi_injected)
7632 kvm_x86_ops->set_nmi(vcpu);
7633 else if (vcpu->arch.interrupt.injected)
7634 kvm_x86_ops->set_irq(vcpu);
7638 * Call check_nested_events() even if we reinjected a previous event
7639 * in order for caller to determine if it should require immediate-exit
7640 * from L2 to L1 due to pending L1 events which require exit
7643 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7644 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7649 /* try to inject new event if pending */
7650 if (vcpu->arch.exception.pending) {
7651 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7652 vcpu->arch.exception.has_error_code,
7653 vcpu->arch.exception.error_code);
7655 WARN_ON_ONCE(vcpu->arch.exception.injected);
7656 vcpu->arch.exception.pending = false;
7657 vcpu->arch.exception.injected = true;
7659 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7660 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7663 if (vcpu->arch.exception.nr == DB_VECTOR) {
7665 * This code assumes that nSVM doesn't use
7666 * check_nested_events(). If it does, the
7667 * DR6/DR7 changes should happen before L1
7668 * gets a #VMEXIT for an intercepted #DB in
7669 * L2. (Under VMX, on the other hand, the
7670 * DR6/DR7 changes should not happen in the
7671 * event of a VM-exit to L1 for an intercepted
7674 kvm_deliver_exception_payload(vcpu);
7675 if (vcpu->arch.dr7 & DR7_GD) {
7676 vcpu->arch.dr7 &= ~DR7_GD;
7677 kvm_update_dr7(vcpu);
7681 kvm_x86_ops->queue_exception(vcpu);
7684 /* Don't consider new event if we re-injected an event */
7685 if (kvm_event_needs_reinjection(vcpu))
7688 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7689 kvm_x86_ops->smi_allowed(vcpu)) {
7690 vcpu->arch.smi_pending = false;
7691 ++vcpu->arch.smi_count;
7693 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7694 --vcpu->arch.nmi_pending;
7695 vcpu->arch.nmi_injected = true;
7696 kvm_x86_ops->set_nmi(vcpu);
7697 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7699 * Because interrupts can be injected asynchronously, we are
7700 * calling check_nested_events again here to avoid a race condition.
7701 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7702 * proposal and current concerns. Perhaps we should be setting
7703 * KVM_REQ_EVENT only on certain events and not unconditionally?
7705 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7706 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7710 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7711 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7713 kvm_x86_ops->set_irq(vcpu);
7720 static void process_nmi(struct kvm_vcpu *vcpu)
7725 * x86 is limited to one NMI running, and one NMI pending after it.
7726 * If an NMI is already in progress, limit further NMIs to just one.
7727 * Otherwise, allow two (and we'll inject the first one immediately).
7729 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7732 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7733 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7734 kvm_make_request(KVM_REQ_EVENT, vcpu);
7737 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7740 flags |= seg->g << 23;
7741 flags |= seg->db << 22;
7742 flags |= seg->l << 21;
7743 flags |= seg->avl << 20;
7744 flags |= seg->present << 15;
7745 flags |= seg->dpl << 13;
7746 flags |= seg->s << 12;
7747 flags |= seg->type << 8;
7751 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7753 struct kvm_segment seg;
7756 kvm_get_segment(vcpu, &seg, n);
7757 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7760 offset = 0x7f84 + n * 12;
7762 offset = 0x7f2c + (n - 3) * 12;
7764 put_smstate(u32, buf, offset + 8, seg.base);
7765 put_smstate(u32, buf, offset + 4, seg.limit);
7766 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7769 #ifdef CONFIG_X86_64
7770 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7772 struct kvm_segment seg;
7776 kvm_get_segment(vcpu, &seg, n);
7777 offset = 0x7e00 + n * 16;
7779 flags = enter_smm_get_segment_flags(&seg) >> 8;
7780 put_smstate(u16, buf, offset, seg.selector);
7781 put_smstate(u16, buf, offset + 2, flags);
7782 put_smstate(u32, buf, offset + 4, seg.limit);
7783 put_smstate(u64, buf, offset + 8, seg.base);
7787 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7790 struct kvm_segment seg;
7794 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7795 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7796 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7797 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7799 for (i = 0; i < 8; i++)
7800 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7802 kvm_get_dr(vcpu, 6, &val);
7803 put_smstate(u32, buf, 0x7fcc, (u32)val);
7804 kvm_get_dr(vcpu, 7, &val);
7805 put_smstate(u32, buf, 0x7fc8, (u32)val);
7807 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7808 put_smstate(u32, buf, 0x7fc4, seg.selector);
7809 put_smstate(u32, buf, 0x7f64, seg.base);
7810 put_smstate(u32, buf, 0x7f60, seg.limit);
7811 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7813 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7814 put_smstate(u32, buf, 0x7fc0, seg.selector);
7815 put_smstate(u32, buf, 0x7f80, seg.base);
7816 put_smstate(u32, buf, 0x7f7c, seg.limit);
7817 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7819 kvm_x86_ops->get_gdt(vcpu, &dt);
7820 put_smstate(u32, buf, 0x7f74, dt.address);
7821 put_smstate(u32, buf, 0x7f70, dt.size);
7823 kvm_x86_ops->get_idt(vcpu, &dt);
7824 put_smstate(u32, buf, 0x7f58, dt.address);
7825 put_smstate(u32, buf, 0x7f54, dt.size);
7827 for (i = 0; i < 6; i++)
7828 enter_smm_save_seg_32(vcpu, buf, i);
7830 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7833 put_smstate(u32, buf, 0x7efc, 0x00020000);
7834 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7837 #ifdef CONFIG_X86_64
7838 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7841 struct kvm_segment seg;
7845 for (i = 0; i < 16; i++)
7846 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7848 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7849 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7851 kvm_get_dr(vcpu, 6, &val);
7852 put_smstate(u64, buf, 0x7f68, val);
7853 kvm_get_dr(vcpu, 7, &val);
7854 put_smstate(u64, buf, 0x7f60, val);
7856 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7857 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7858 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7860 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7863 put_smstate(u32, buf, 0x7efc, 0x00020064);
7865 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7867 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7868 put_smstate(u16, buf, 0x7e90, seg.selector);
7869 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7870 put_smstate(u32, buf, 0x7e94, seg.limit);
7871 put_smstate(u64, buf, 0x7e98, seg.base);
7873 kvm_x86_ops->get_idt(vcpu, &dt);
7874 put_smstate(u32, buf, 0x7e84, dt.size);
7875 put_smstate(u64, buf, 0x7e88, dt.address);
7877 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7878 put_smstate(u16, buf, 0x7e70, seg.selector);
7879 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7880 put_smstate(u32, buf, 0x7e74, seg.limit);
7881 put_smstate(u64, buf, 0x7e78, seg.base);
7883 kvm_x86_ops->get_gdt(vcpu, &dt);
7884 put_smstate(u32, buf, 0x7e64, dt.size);
7885 put_smstate(u64, buf, 0x7e68, dt.address);
7887 for (i = 0; i < 6; i++)
7888 enter_smm_save_seg_64(vcpu, buf, i);
7892 static void enter_smm(struct kvm_vcpu *vcpu)
7894 struct kvm_segment cs, ds;
7899 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7900 memset(buf, 0, 512);
7901 #ifdef CONFIG_X86_64
7902 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7903 enter_smm_save_state_64(vcpu, buf);
7906 enter_smm_save_state_32(vcpu, buf);
7909 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7910 * vCPU state (e.g. leave guest mode) after we've saved the state into
7911 * the SMM state-save area.
7913 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7915 vcpu->arch.hflags |= HF_SMM_MASK;
7916 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7918 if (kvm_x86_ops->get_nmi_mask(vcpu))
7919 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7921 kvm_x86_ops->set_nmi_mask(vcpu, true);
7923 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7924 kvm_rip_write(vcpu, 0x8000);
7926 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7927 kvm_x86_ops->set_cr0(vcpu, cr0);
7928 vcpu->arch.cr0 = cr0;
7930 kvm_x86_ops->set_cr4(vcpu, 0);
7932 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7933 dt.address = dt.size = 0;
7934 kvm_x86_ops->set_idt(vcpu, &dt);
7936 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7938 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7939 cs.base = vcpu->arch.smbase;
7944 cs.limit = ds.limit = 0xffffffff;
7945 cs.type = ds.type = 0x3;
7946 cs.dpl = ds.dpl = 0;
7951 cs.avl = ds.avl = 0;
7952 cs.present = ds.present = 1;
7953 cs.unusable = ds.unusable = 0;
7954 cs.padding = ds.padding = 0;
7956 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7957 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7958 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7959 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7960 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7961 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7963 #ifdef CONFIG_X86_64
7964 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7965 kvm_x86_ops->set_efer(vcpu, 0);
7968 kvm_update_cpuid(vcpu);
7969 kvm_mmu_reset_context(vcpu);
7972 static void process_smi(struct kvm_vcpu *vcpu)
7974 vcpu->arch.smi_pending = true;
7975 kvm_make_request(KVM_REQ_EVENT, vcpu);
7978 void kvm_make_scan_ioapic_request_mask(struct kvm *kvm,
7979 unsigned long *vcpu_bitmap)
7983 zalloc_cpumask_var(&cpus, GFP_ATOMIC);
7985 kvm_make_vcpus_request_mask(kvm, KVM_REQ_SCAN_IOAPIC,
7988 free_cpumask_var(cpus);
7991 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7993 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7996 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7998 if (!kvm_apic_present(vcpu))
8001 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
8003 if (irqchip_split(vcpu->kvm))
8004 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
8006 if (vcpu->arch.apicv_active)
8007 kvm_x86_ops->sync_pir_to_irr(vcpu);
8008 if (ioapic_in_kernel(vcpu->kvm))
8009 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
8012 if (is_guest_mode(vcpu))
8013 vcpu->arch.load_eoi_exitmap_pending = true;
8015 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
8018 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8020 u64 eoi_exit_bitmap[4];
8022 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
8025 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
8026 vcpu_to_synic(vcpu)->vec_bitmap, 256);
8027 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
8030 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
8031 unsigned long start, unsigned long end,
8034 unsigned long apic_address;
8037 * The physical address of apic access page is stored in the VMCS.
8038 * Update it when it becomes invalid.
8040 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8041 if (start <= apic_address && apic_address < end)
8042 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
8047 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
8049 struct page *page = NULL;
8051 if (!lapic_in_kernel(vcpu))
8054 if (!kvm_x86_ops->set_apic_access_page_addr)
8057 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
8058 if (is_error_page(page))
8060 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
8063 * Do not pin apic access page in memory, the MMU notifier
8064 * will call us again if it is migrated or swapped out.
8069 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
8071 smp_send_reschedule(vcpu->cpu);
8073 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
8076 * Returns 1 to let vcpu_run() continue the guest execution loop without
8077 * exiting to the userspace. Otherwise, the value will be returned to the
8080 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
8084 dm_request_for_irq_injection(vcpu) &&
8085 kvm_cpu_accept_dm_intr(vcpu);
8086 enum exit_fastpath_completion exit_fastpath = EXIT_FASTPATH_NONE;
8088 bool req_immediate_exit = false;
8090 if (kvm_request_pending(vcpu)) {
8091 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
8092 if (unlikely(!kvm_x86_ops->get_vmcs12_pages(vcpu))) {
8097 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
8098 kvm_mmu_unload(vcpu);
8099 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
8100 __kvm_migrate_timers(vcpu);
8101 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
8102 kvm_gen_update_masterclock(vcpu->kvm);
8103 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
8104 kvm_gen_kvmclock_update(vcpu);
8105 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
8106 r = kvm_guest_time_update(vcpu);
8110 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
8111 kvm_mmu_sync_roots(vcpu);
8112 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
8113 kvm_mmu_load_cr3(vcpu);
8114 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
8115 kvm_vcpu_flush_tlb(vcpu, true);
8116 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
8117 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
8121 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8122 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8123 vcpu->mmio_needed = 0;
8127 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8128 /* Page is swapped out. Do synthetic halt */
8129 vcpu->arch.apf.halted = true;
8133 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8134 record_steal_time(vcpu);
8135 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8137 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8139 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8140 kvm_pmu_handle_event(vcpu);
8141 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8142 kvm_pmu_deliver_pmi(vcpu);
8143 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8144 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8145 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8146 vcpu->arch.ioapic_handled_vectors)) {
8147 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8148 vcpu->run->eoi.vector =
8149 vcpu->arch.pending_ioapic_eoi;
8154 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8155 vcpu_scan_ioapic(vcpu);
8156 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8157 vcpu_load_eoi_exitmap(vcpu);
8158 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8159 kvm_vcpu_reload_apic_access_page(vcpu);
8160 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8161 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8162 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8166 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8167 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8168 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8172 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8173 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8174 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8180 * KVM_REQ_HV_STIMER has to be processed after
8181 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8182 * depend on the guest clock being up-to-date
8184 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8185 kvm_hv_process_stimers(vcpu);
8188 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8189 ++vcpu->stat.req_event;
8190 kvm_apic_accept_events(vcpu);
8191 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8196 if (inject_pending_event(vcpu, req_int_win) != 0)
8197 req_immediate_exit = true;
8199 /* Enable SMI/NMI/IRQ window open exits if needed.
8201 * SMIs have three cases:
8202 * 1) They can be nested, and then there is nothing to
8203 * do here because RSM will cause a vmexit anyway.
8204 * 2) There is an ISA-specific reason why SMI cannot be
8205 * injected, and the moment when this changes can be
8207 * 3) Or the SMI can be pending because
8208 * inject_pending_event has completed the injection
8209 * of an IRQ or NMI from the previous vmexit, and
8210 * then we request an immediate exit to inject the
8213 if (vcpu->arch.smi_pending && !is_smm(vcpu))
8214 if (!kvm_x86_ops->enable_smi_window(vcpu))
8215 req_immediate_exit = true;
8216 if (vcpu->arch.nmi_pending)
8217 kvm_x86_ops->enable_nmi_window(vcpu);
8218 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8219 kvm_x86_ops->enable_irq_window(vcpu);
8220 WARN_ON(vcpu->arch.exception.pending);
8223 if (kvm_lapic_enabled(vcpu)) {
8224 update_cr8_intercept(vcpu);
8225 kvm_lapic_sync_to_vapic(vcpu);
8229 r = kvm_mmu_reload(vcpu);
8231 goto cancel_injection;
8236 kvm_x86_ops->prepare_guest_switch(vcpu);
8239 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8240 * IPI are then delayed after guest entry, which ensures that they
8241 * result in virtual interrupt delivery.
8243 local_irq_disable();
8244 vcpu->mode = IN_GUEST_MODE;
8246 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8249 * 1) We should set ->mode before checking ->requests. Please see
8250 * the comment in kvm_vcpu_exiting_guest_mode().
8252 * 2) For APICv, we should set ->mode before checking PID.ON. This
8253 * pairs with the memory barrier implicit in pi_test_and_set_on
8254 * (see vmx_deliver_posted_interrupt).
8256 * 3) This also orders the write to mode from any reads to the page
8257 * tables done while the VCPU is running. Please see the comment
8258 * in kvm_flush_remote_tlbs.
8260 smp_mb__after_srcu_read_unlock();
8263 * This handles the case where a posted interrupt was
8264 * notified with kvm_vcpu_kick.
8266 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8267 kvm_x86_ops->sync_pir_to_irr(vcpu);
8269 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8270 || need_resched() || signal_pending(current)) {
8271 vcpu->mode = OUTSIDE_GUEST_MODE;
8275 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8277 goto cancel_injection;
8280 if (req_immediate_exit) {
8281 kvm_make_request(KVM_REQ_EVENT, vcpu);
8282 kvm_x86_ops->request_immediate_exit(vcpu);
8285 trace_kvm_entry(vcpu->vcpu_id);
8286 guest_enter_irqoff();
8288 /* The preempt notifier should have taken care of the FPU already. */
8289 WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8291 if (unlikely(vcpu->arch.switch_db_regs)) {
8293 set_debugreg(vcpu->arch.eff_db[0], 0);
8294 set_debugreg(vcpu->arch.eff_db[1], 1);
8295 set_debugreg(vcpu->arch.eff_db[2], 2);
8296 set_debugreg(vcpu->arch.eff_db[3], 3);
8297 set_debugreg(vcpu->arch.dr6, 6);
8298 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8301 kvm_x86_ops->run(vcpu);
8304 * Do this here before restoring debug registers on the host. And
8305 * since we do this before handling the vmexit, a DR access vmexit
8306 * can (a) read the correct value of the debug registers, (b) set
8307 * KVM_DEBUGREG_WONT_EXIT again.
8309 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8310 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8311 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8312 kvm_update_dr0123(vcpu);
8313 kvm_update_dr6(vcpu);
8314 kvm_update_dr7(vcpu);
8315 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8319 * If the guest has used debug registers, at least dr7
8320 * will be disabled while returning to the host.
8321 * If we don't have active breakpoints in the host, we don't
8322 * care about the messed up debug address registers. But if
8323 * we have some of them active, restore the old state.
8325 if (hw_breakpoint_active())
8326 hw_breakpoint_restore();
8328 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8330 vcpu->mode = OUTSIDE_GUEST_MODE;
8333 kvm_x86_ops->handle_exit_irqoff(vcpu, &exit_fastpath);
8336 * Consume any pending interrupts, including the possible source of
8337 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8338 * An instruction is required after local_irq_enable() to fully unblock
8339 * interrupts on processors that implement an interrupt shadow, the
8340 * stat.exits increment will do nicely.
8342 kvm_before_interrupt(vcpu);
8345 local_irq_disable();
8346 kvm_after_interrupt(vcpu);
8348 guest_exit_irqoff();
8349 if (lapic_in_kernel(vcpu)) {
8350 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8351 if (delta != S64_MIN) {
8352 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8353 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8360 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8363 * Profile KVM exit RIPs:
8365 if (unlikely(prof_on == KVM_PROFILING)) {
8366 unsigned long rip = kvm_rip_read(vcpu);
8367 profile_hit(KVM_PROFILING, (void *)rip);
8370 if (unlikely(vcpu->arch.tsc_always_catchup))
8371 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8373 if (vcpu->arch.apic_attention)
8374 kvm_lapic_sync_from_vapic(vcpu);
8376 vcpu->arch.gpa_available = false;
8377 r = kvm_x86_ops->handle_exit(vcpu, exit_fastpath);
8381 kvm_x86_ops->cancel_injection(vcpu);
8382 if (unlikely(vcpu->arch.apic_attention))
8383 kvm_lapic_sync_from_vapic(vcpu);
8388 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8390 if (!kvm_arch_vcpu_runnable(vcpu) &&
8391 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8392 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8393 kvm_vcpu_block(vcpu);
8394 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8396 if (kvm_x86_ops->post_block)
8397 kvm_x86_ops->post_block(vcpu);
8399 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8403 kvm_apic_accept_events(vcpu);
8404 switch(vcpu->arch.mp_state) {
8405 case KVM_MP_STATE_HALTED:
8406 vcpu->arch.pv.pv_unhalted = false;
8407 vcpu->arch.mp_state =
8408 KVM_MP_STATE_RUNNABLE;
8410 case KVM_MP_STATE_RUNNABLE:
8411 vcpu->arch.apf.halted = false;
8413 case KVM_MP_STATE_INIT_RECEIVED:
8422 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8424 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8425 kvm_x86_ops->check_nested_events(vcpu, false);
8427 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8428 !vcpu->arch.apf.halted);
8431 static int vcpu_run(struct kvm_vcpu *vcpu)
8434 struct kvm *kvm = vcpu->kvm;
8436 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8437 vcpu->arch.l1tf_flush_l1d = true;
8440 if (kvm_vcpu_running(vcpu)) {
8441 r = vcpu_enter_guest(vcpu);
8443 r = vcpu_block(kvm, vcpu);
8449 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8450 if (kvm_cpu_has_pending_timer(vcpu))
8451 kvm_inject_pending_timer_irqs(vcpu);
8453 if (dm_request_for_irq_injection(vcpu) &&
8454 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8456 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8457 ++vcpu->stat.request_irq_exits;
8461 kvm_check_async_pf_completion(vcpu);
8463 if (signal_pending(current)) {
8465 vcpu->run->exit_reason = KVM_EXIT_INTR;
8466 ++vcpu->stat.signal_exits;
8469 if (need_resched()) {
8470 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8472 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8476 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8481 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8485 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8486 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8487 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8491 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8493 BUG_ON(!vcpu->arch.pio.count);
8495 return complete_emulated_io(vcpu);
8499 * Implements the following, as a state machine:
8503 * for each mmio piece in the fragment
8511 * for each mmio piece in the fragment
8516 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8518 struct kvm_run *run = vcpu->run;
8519 struct kvm_mmio_fragment *frag;
8522 BUG_ON(!vcpu->mmio_needed);
8524 /* Complete previous fragment */
8525 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8526 len = min(8u, frag->len);
8527 if (!vcpu->mmio_is_write)
8528 memcpy(frag->data, run->mmio.data, len);
8530 if (frag->len <= 8) {
8531 /* Switch to the next fragment. */
8533 vcpu->mmio_cur_fragment++;
8535 /* Go forward to the next mmio piece. */
8541 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8542 vcpu->mmio_needed = 0;
8544 /* FIXME: return into emulator if single-stepping. */
8545 if (vcpu->mmio_is_write)
8547 vcpu->mmio_read_completed = 1;
8548 return complete_emulated_io(vcpu);
8551 run->exit_reason = KVM_EXIT_MMIO;
8552 run->mmio.phys_addr = frag->gpa;
8553 if (vcpu->mmio_is_write)
8554 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8555 run->mmio.len = min(8u, frag->len);
8556 run->mmio.is_write = vcpu->mmio_is_write;
8557 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8561 /* Swap (qemu) user FPU context for the guest FPU context. */
8562 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8567 * Reloading userspace's FPU is handled by kvm_arch_vcpu_load(), both
8568 * for direct calls from userspace (via vcpu_load()) and if this task
8569 * is preempted (via kvm_sched_in()) between vcpu_load() and now.
8571 WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8573 copy_fpregs_to_fpstate(vcpu->arch.user_fpu);
8574 /* PKRU is separately restored in kvm_x86_ops->run. */
8575 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8576 ~XFEATURE_MASK_PKRU);
8578 fpregs_mark_activate();
8584 /* When vcpu_run ends, restore user space FPU context. */
8585 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8589 copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8590 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8592 fpregs_mark_activate();
8595 ++vcpu->stat.fpu_reload;
8599 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8604 kvm_sigset_activate(vcpu);
8605 kvm_load_guest_fpu(vcpu);
8607 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8608 if (kvm_run->immediate_exit) {
8612 kvm_vcpu_block(vcpu);
8613 kvm_apic_accept_events(vcpu);
8614 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8616 if (signal_pending(current)) {
8618 vcpu->run->exit_reason = KVM_EXIT_INTR;
8619 ++vcpu->stat.signal_exits;
8624 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8629 if (vcpu->run->kvm_dirty_regs) {
8630 r = sync_regs(vcpu);
8635 /* re-sync apic's tpr */
8636 if (!lapic_in_kernel(vcpu)) {
8637 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8643 if (unlikely(vcpu->arch.complete_userspace_io)) {
8644 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8645 vcpu->arch.complete_userspace_io = NULL;
8650 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8652 if (kvm_run->immediate_exit)
8658 kvm_put_guest_fpu(vcpu);
8659 if (vcpu->run->kvm_valid_regs)
8661 post_kvm_run_save(vcpu);
8662 kvm_sigset_deactivate(vcpu);
8668 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8670 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8672 * We are here if userspace calls get_regs() in the middle of
8673 * instruction emulation. Registers state needs to be copied
8674 * back from emulation context to vcpu. Userspace shouldn't do
8675 * that usually, but some bad designed PV devices (vmware
8676 * backdoor interface) need this to work
8678 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8679 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8681 regs->rax = kvm_rax_read(vcpu);
8682 regs->rbx = kvm_rbx_read(vcpu);
8683 regs->rcx = kvm_rcx_read(vcpu);
8684 regs->rdx = kvm_rdx_read(vcpu);
8685 regs->rsi = kvm_rsi_read(vcpu);
8686 regs->rdi = kvm_rdi_read(vcpu);
8687 regs->rsp = kvm_rsp_read(vcpu);
8688 regs->rbp = kvm_rbp_read(vcpu);
8689 #ifdef CONFIG_X86_64
8690 regs->r8 = kvm_r8_read(vcpu);
8691 regs->r9 = kvm_r9_read(vcpu);
8692 regs->r10 = kvm_r10_read(vcpu);
8693 regs->r11 = kvm_r11_read(vcpu);
8694 regs->r12 = kvm_r12_read(vcpu);
8695 regs->r13 = kvm_r13_read(vcpu);
8696 regs->r14 = kvm_r14_read(vcpu);
8697 regs->r15 = kvm_r15_read(vcpu);
8700 regs->rip = kvm_rip_read(vcpu);
8701 regs->rflags = kvm_get_rflags(vcpu);
8704 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8707 __get_regs(vcpu, regs);
8712 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8714 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8715 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8717 kvm_rax_write(vcpu, regs->rax);
8718 kvm_rbx_write(vcpu, regs->rbx);
8719 kvm_rcx_write(vcpu, regs->rcx);
8720 kvm_rdx_write(vcpu, regs->rdx);
8721 kvm_rsi_write(vcpu, regs->rsi);
8722 kvm_rdi_write(vcpu, regs->rdi);
8723 kvm_rsp_write(vcpu, regs->rsp);
8724 kvm_rbp_write(vcpu, regs->rbp);
8725 #ifdef CONFIG_X86_64
8726 kvm_r8_write(vcpu, regs->r8);
8727 kvm_r9_write(vcpu, regs->r9);
8728 kvm_r10_write(vcpu, regs->r10);
8729 kvm_r11_write(vcpu, regs->r11);
8730 kvm_r12_write(vcpu, regs->r12);
8731 kvm_r13_write(vcpu, regs->r13);
8732 kvm_r14_write(vcpu, regs->r14);
8733 kvm_r15_write(vcpu, regs->r15);
8736 kvm_rip_write(vcpu, regs->rip);
8737 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8739 vcpu->arch.exception.pending = false;
8741 kvm_make_request(KVM_REQ_EVENT, vcpu);
8744 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8747 __set_regs(vcpu, regs);
8752 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8754 struct kvm_segment cs;
8756 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8760 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8762 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8766 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8767 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8768 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8769 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8770 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8771 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8773 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8774 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8776 kvm_x86_ops->get_idt(vcpu, &dt);
8777 sregs->idt.limit = dt.size;
8778 sregs->idt.base = dt.address;
8779 kvm_x86_ops->get_gdt(vcpu, &dt);
8780 sregs->gdt.limit = dt.size;
8781 sregs->gdt.base = dt.address;
8783 sregs->cr0 = kvm_read_cr0(vcpu);
8784 sregs->cr2 = vcpu->arch.cr2;
8785 sregs->cr3 = kvm_read_cr3(vcpu);
8786 sregs->cr4 = kvm_read_cr4(vcpu);
8787 sregs->cr8 = kvm_get_cr8(vcpu);
8788 sregs->efer = vcpu->arch.efer;
8789 sregs->apic_base = kvm_get_apic_base(vcpu);
8791 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8793 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8794 set_bit(vcpu->arch.interrupt.nr,
8795 (unsigned long *)sregs->interrupt_bitmap);
8798 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8799 struct kvm_sregs *sregs)
8802 __get_sregs(vcpu, sregs);
8807 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8808 struct kvm_mp_state *mp_state)
8811 if (kvm_mpx_supported())
8812 kvm_load_guest_fpu(vcpu);
8814 kvm_apic_accept_events(vcpu);
8815 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8816 vcpu->arch.pv.pv_unhalted)
8817 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8819 mp_state->mp_state = vcpu->arch.mp_state;
8821 if (kvm_mpx_supported())
8822 kvm_put_guest_fpu(vcpu);
8827 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8828 struct kvm_mp_state *mp_state)
8834 if (!lapic_in_kernel(vcpu) &&
8835 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8839 * KVM_MP_STATE_INIT_RECEIVED means the processor is in
8840 * INIT state; latched init should be reported using
8841 * KVM_SET_VCPU_EVENTS, so reject it here.
8843 if ((kvm_vcpu_latch_init(vcpu) || vcpu->arch.smi_pending) &&
8844 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8845 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8848 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8849 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8850 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8852 vcpu->arch.mp_state = mp_state->mp_state;
8853 kvm_make_request(KVM_REQ_EVENT, vcpu);
8861 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8862 int reason, bool has_error_code, u32 error_code)
8864 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8867 init_emulate_ctxt(vcpu);
8869 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8870 has_error_code, error_code);
8872 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8873 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8874 vcpu->run->internal.ndata = 0;
8878 kvm_rip_write(vcpu, ctxt->eip);
8879 kvm_set_rflags(vcpu, ctxt->eflags);
8880 kvm_make_request(KVM_REQ_EVENT, vcpu);
8883 EXPORT_SYMBOL_GPL(kvm_task_switch);
8885 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8887 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8889 * When EFER.LME and CR0.PG are set, the processor is in
8890 * 64-bit mode (though maybe in a 32-bit code segment).
8891 * CR4.PAE and EFER.LMA must be set.
8893 if (!(sregs->cr4 & X86_CR4_PAE)
8894 || !(sregs->efer & EFER_LMA))
8898 * Not in 64-bit mode: EFER.LMA is clear and the code
8899 * segment cannot be 64-bit.
8901 if (sregs->efer & EFER_LMA || sregs->cs.l)
8905 return kvm_valid_cr4(vcpu, sregs->cr4);
8908 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8910 struct msr_data apic_base_msr;
8911 int mmu_reset_needed = 0;
8912 int cpuid_update_needed = 0;
8913 int pending_vec, max_bits, idx;
8917 if (kvm_valid_sregs(vcpu, sregs))
8920 apic_base_msr.data = sregs->apic_base;
8921 apic_base_msr.host_initiated = true;
8922 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8925 dt.size = sregs->idt.limit;
8926 dt.address = sregs->idt.base;
8927 kvm_x86_ops->set_idt(vcpu, &dt);
8928 dt.size = sregs->gdt.limit;
8929 dt.address = sregs->gdt.base;
8930 kvm_x86_ops->set_gdt(vcpu, &dt);
8932 vcpu->arch.cr2 = sregs->cr2;
8933 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8934 vcpu->arch.cr3 = sregs->cr3;
8935 kvm_register_mark_available(vcpu, VCPU_EXREG_CR3);
8937 kvm_set_cr8(vcpu, sregs->cr8);
8939 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8940 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8942 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8943 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8944 vcpu->arch.cr0 = sregs->cr0;
8946 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8947 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8948 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8949 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8950 if (cpuid_update_needed)
8951 kvm_update_cpuid(vcpu);
8953 idx = srcu_read_lock(&vcpu->kvm->srcu);
8954 if (is_pae_paging(vcpu)) {
8955 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8956 mmu_reset_needed = 1;
8958 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8960 if (mmu_reset_needed)
8961 kvm_mmu_reset_context(vcpu);
8963 max_bits = KVM_NR_INTERRUPTS;
8964 pending_vec = find_first_bit(
8965 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8966 if (pending_vec < max_bits) {
8967 kvm_queue_interrupt(vcpu, pending_vec, false);
8968 pr_debug("Set back pending irq %d\n", pending_vec);
8971 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8972 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8973 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8974 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8975 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8976 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8978 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8979 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8981 update_cr8_intercept(vcpu);
8983 /* Older userspace won't unhalt the vcpu on reset. */
8984 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8985 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8987 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8989 kvm_make_request(KVM_REQ_EVENT, vcpu);
8996 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8997 struct kvm_sregs *sregs)
9002 ret = __set_sregs(vcpu, sregs);
9007 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
9008 struct kvm_guest_debug *dbg)
9010 unsigned long rflags;
9015 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
9017 if (vcpu->arch.exception.pending)
9019 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
9020 kvm_queue_exception(vcpu, DB_VECTOR);
9022 kvm_queue_exception(vcpu, BP_VECTOR);
9026 * Read rflags as long as potentially injected trace flags are still
9029 rflags = kvm_get_rflags(vcpu);
9031 vcpu->guest_debug = dbg->control;
9032 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
9033 vcpu->guest_debug = 0;
9035 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
9036 for (i = 0; i < KVM_NR_DB_REGS; ++i)
9037 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
9038 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
9040 for (i = 0; i < KVM_NR_DB_REGS; i++)
9041 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
9043 kvm_update_dr7(vcpu);
9045 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9046 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
9047 get_segment_base(vcpu, VCPU_SREG_CS);
9050 * Trigger an rflags update that will inject or remove the trace
9053 kvm_set_rflags(vcpu, rflags);
9055 kvm_x86_ops->update_bp_intercept(vcpu);
9065 * Translate a guest virtual address to a guest physical address.
9067 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
9068 struct kvm_translation *tr)
9070 unsigned long vaddr = tr->linear_address;
9076 idx = srcu_read_lock(&vcpu->kvm->srcu);
9077 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
9078 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9079 tr->physical_address = gpa;
9080 tr->valid = gpa != UNMAPPED_GVA;
9088 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9090 struct fxregs_state *fxsave;
9094 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9095 memcpy(fpu->fpr, fxsave->st_space, 128);
9096 fpu->fcw = fxsave->cwd;
9097 fpu->fsw = fxsave->swd;
9098 fpu->ftwx = fxsave->twd;
9099 fpu->last_opcode = fxsave->fop;
9100 fpu->last_ip = fxsave->rip;
9101 fpu->last_dp = fxsave->rdp;
9102 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
9108 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
9110 struct fxregs_state *fxsave;
9114 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
9116 memcpy(fxsave->st_space, fpu->fpr, 128);
9117 fxsave->cwd = fpu->fcw;
9118 fxsave->swd = fpu->fsw;
9119 fxsave->twd = fpu->ftwx;
9120 fxsave->fop = fpu->last_opcode;
9121 fxsave->rip = fpu->last_ip;
9122 fxsave->rdp = fpu->last_dp;
9123 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
9129 static void store_regs(struct kvm_vcpu *vcpu)
9131 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
9133 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
9134 __get_regs(vcpu, &vcpu->run->s.regs.regs);
9136 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9137 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9139 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9140 kvm_vcpu_ioctl_x86_get_vcpu_events(
9141 vcpu, &vcpu->run->s.regs.events);
9144 static int sync_regs(struct kvm_vcpu *vcpu)
9146 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9149 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9150 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9151 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9153 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9154 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9156 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9158 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9159 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9160 vcpu, &vcpu->run->s.regs.events))
9162 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9168 static void fx_init(struct kvm_vcpu *vcpu)
9170 fpstate_init(&vcpu->arch.guest_fpu->state);
9171 if (boot_cpu_has(X86_FEATURE_XSAVES))
9172 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9173 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9176 * Ensure guest xcr0 is valid for loading
9178 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9180 vcpu->arch.cr0 |= X86_CR0_ET;
9183 int kvm_arch_vcpu_precreate(struct kvm *kvm, unsigned int id)
9185 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9186 pr_warn_once("kvm: SMP vm created on host with unstable TSC; "
9187 "guest TSC will not be reliable\n");
9192 int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu)
9197 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9198 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9199 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9201 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9203 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9205 r = kvm_mmu_create(vcpu);
9209 if (irqchip_in_kernel(vcpu->kvm)) {
9210 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu->kvm);
9211 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9213 goto fail_mmu_destroy;
9215 static_key_slow_inc(&kvm_no_apic_vcpu);
9219 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9221 goto fail_free_lapic;
9222 vcpu->arch.pio_data = page_address(page);
9224 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9225 GFP_KERNEL_ACCOUNT);
9226 if (!vcpu->arch.mce_banks)
9227 goto fail_free_pio_data;
9228 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9230 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9231 GFP_KERNEL_ACCOUNT))
9232 goto fail_free_mce_banks;
9234 vcpu->arch.user_fpu = kmem_cache_zalloc(x86_fpu_cache,
9235 GFP_KERNEL_ACCOUNT);
9236 if (!vcpu->arch.user_fpu) {
9237 pr_err("kvm: failed to allocate userspace's fpu\n");
9238 goto free_wbinvd_dirty_mask;
9241 vcpu->arch.guest_fpu = kmem_cache_zalloc(x86_fpu_cache,
9242 GFP_KERNEL_ACCOUNT);
9243 if (!vcpu->arch.guest_fpu) {
9244 pr_err("kvm: failed to allocate vcpu's fpu\n");
9249 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9251 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9253 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9255 kvm_async_pf_hash_reset(vcpu);
9258 vcpu->arch.pending_external_vector = -1;
9259 vcpu->arch.preempted_in_kernel = false;
9261 kvm_hv_vcpu_init(vcpu);
9263 r = kvm_x86_ops->vcpu_create(vcpu);
9265 goto free_guest_fpu;
9267 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9268 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9269 kvm_vcpu_mtrr_init(vcpu);
9271 kvm_vcpu_reset(vcpu, false);
9272 kvm_init_mmu(vcpu, false);
9277 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9279 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9280 free_wbinvd_dirty_mask:
9281 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9282 fail_free_mce_banks:
9283 kfree(vcpu->arch.mce_banks);
9285 free_page((unsigned long)vcpu->arch.pio_data);
9287 kvm_free_lapic(vcpu);
9289 kvm_mmu_destroy(vcpu);
9293 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9295 struct msr_data msr;
9296 struct kvm *kvm = vcpu->kvm;
9298 kvm_hv_vcpu_postcreate(vcpu);
9300 if (mutex_lock_killable(&vcpu->mutex))
9304 msr.index = MSR_IA32_TSC;
9305 msr.host_initiated = true;
9306 kvm_write_tsc(vcpu, &msr);
9309 /* poll control enabled by default */
9310 vcpu->arch.msr_kvm_poll_control = 1;
9312 mutex_unlock(&vcpu->mutex);
9314 if (!kvmclock_periodic_sync)
9317 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9318 KVMCLOCK_SYNC_PERIOD);
9321 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9325 kvmclock_reset(vcpu);
9327 kvm_x86_ops->vcpu_free(vcpu);
9329 free_cpumask_var(vcpu->arch.wbinvd_dirty_mask);
9330 kmem_cache_free(x86_fpu_cache, vcpu->arch.user_fpu);
9331 kmem_cache_free(x86_fpu_cache, vcpu->arch.guest_fpu);
9333 kvm_hv_vcpu_uninit(vcpu);
9334 kvm_pmu_destroy(vcpu);
9335 kfree(vcpu->arch.mce_banks);
9336 kvm_free_lapic(vcpu);
9337 idx = srcu_read_lock(&vcpu->kvm->srcu);
9338 kvm_mmu_destroy(vcpu);
9339 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9340 free_page((unsigned long)vcpu->arch.pio_data);
9341 if (!lapic_in_kernel(vcpu))
9342 static_key_slow_dec(&kvm_no_apic_vcpu);
9345 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9347 kvm_lapic_reset(vcpu, init_event);
9349 vcpu->arch.hflags = 0;
9351 vcpu->arch.smi_pending = 0;
9352 vcpu->arch.smi_count = 0;
9353 atomic_set(&vcpu->arch.nmi_queued, 0);
9354 vcpu->arch.nmi_pending = 0;
9355 vcpu->arch.nmi_injected = false;
9356 kvm_clear_interrupt_queue(vcpu);
9357 kvm_clear_exception_queue(vcpu);
9358 vcpu->arch.exception.pending = false;
9360 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9361 kvm_update_dr0123(vcpu);
9362 vcpu->arch.dr6 = DR6_INIT;
9363 kvm_update_dr6(vcpu);
9364 vcpu->arch.dr7 = DR7_FIXED_1;
9365 kvm_update_dr7(vcpu);
9369 kvm_make_request(KVM_REQ_EVENT, vcpu);
9370 vcpu->arch.apf.msr_val = 0;
9371 vcpu->arch.st.msr_val = 0;
9373 kvmclock_reset(vcpu);
9375 kvm_clear_async_pf_completion_queue(vcpu);
9376 kvm_async_pf_hash_reset(vcpu);
9377 vcpu->arch.apf.halted = false;
9379 if (kvm_mpx_supported()) {
9380 void *mpx_state_buffer;
9383 * To avoid have the INIT path from kvm_apic_has_events() that be
9384 * called with loaded FPU and does not let userspace fix the state.
9387 kvm_put_guest_fpu(vcpu);
9388 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9390 if (mpx_state_buffer)
9391 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9392 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9394 if (mpx_state_buffer)
9395 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9397 kvm_load_guest_fpu(vcpu);
9401 kvm_pmu_reset(vcpu);
9402 vcpu->arch.smbase = 0x30000;
9404 vcpu->arch.msr_misc_features_enables = 0;
9406 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9409 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9410 vcpu->arch.regs_avail = ~0;
9411 vcpu->arch.regs_dirty = ~0;
9413 vcpu->arch.ia32_xss = 0;
9415 kvm_x86_ops->vcpu_reset(vcpu, init_event);
9418 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9420 struct kvm_segment cs;
9422 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9423 cs.selector = vector << 8;
9424 cs.base = vector << 12;
9425 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9426 kvm_rip_write(vcpu, 0);
9429 int kvm_arch_hardware_enable(void)
9432 struct kvm_vcpu *vcpu;
9437 bool stable, backwards_tsc = false;
9439 kvm_shared_msr_cpu_online();
9440 ret = kvm_x86_ops->hardware_enable();
9444 local_tsc = rdtsc();
9445 stable = !kvm_check_tsc_unstable();
9446 list_for_each_entry(kvm, &vm_list, vm_list) {
9447 kvm_for_each_vcpu(i, vcpu, kvm) {
9448 if (!stable && vcpu->cpu == smp_processor_id())
9449 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9450 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9451 backwards_tsc = true;
9452 if (vcpu->arch.last_host_tsc > max_tsc)
9453 max_tsc = vcpu->arch.last_host_tsc;
9459 * Sometimes, even reliable TSCs go backwards. This happens on
9460 * platforms that reset TSC during suspend or hibernate actions, but
9461 * maintain synchronization. We must compensate. Fortunately, we can
9462 * detect that condition here, which happens early in CPU bringup,
9463 * before any KVM threads can be running. Unfortunately, we can't
9464 * bring the TSCs fully up to date with real time, as we aren't yet far
9465 * enough into CPU bringup that we know how much real time has actually
9466 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9467 * variables that haven't been updated yet.
9469 * So we simply find the maximum observed TSC above, then record the
9470 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9471 * the adjustment will be applied. Note that we accumulate
9472 * adjustments, in case multiple suspend cycles happen before some VCPU
9473 * gets a chance to run again. In the event that no KVM threads get a
9474 * chance to run, we will miss the entire elapsed period, as we'll have
9475 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9476 * loose cycle time. This isn't too big a deal, since the loss will be
9477 * uniform across all VCPUs (not to mention the scenario is extremely
9478 * unlikely). It is possible that a second hibernate recovery happens
9479 * much faster than a first, causing the observed TSC here to be
9480 * smaller; this would require additional padding adjustment, which is
9481 * why we set last_host_tsc to the local tsc observed here.
9483 * N.B. - this code below runs only on platforms with reliable TSC,
9484 * as that is the only way backwards_tsc is set above. Also note
9485 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9486 * have the same delta_cyc adjustment applied if backwards_tsc
9487 * is detected. Note further, this adjustment is only done once,
9488 * as we reset last_host_tsc on all VCPUs to stop this from being
9489 * called multiple times (one for each physical CPU bringup).
9491 * Platforms with unreliable TSCs don't have to deal with this, they
9492 * will be compensated by the logic in vcpu_load, which sets the TSC to
9493 * catchup mode. This will catchup all VCPUs to real time, but cannot
9494 * guarantee that they stay in perfect synchronization.
9496 if (backwards_tsc) {
9497 u64 delta_cyc = max_tsc - local_tsc;
9498 list_for_each_entry(kvm, &vm_list, vm_list) {
9499 kvm->arch.backwards_tsc_observed = true;
9500 kvm_for_each_vcpu(i, vcpu, kvm) {
9501 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9502 vcpu->arch.last_host_tsc = local_tsc;
9503 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9507 * We have to disable TSC offset matching.. if you were
9508 * booting a VM while issuing an S4 host suspend....
9509 * you may have some problem. Solving this issue is
9510 * left as an exercise to the reader.
9512 kvm->arch.last_tsc_nsec = 0;
9513 kvm->arch.last_tsc_write = 0;
9520 void kvm_arch_hardware_disable(void)
9522 kvm_x86_ops->hardware_disable();
9523 drop_user_return_notifiers();
9526 int kvm_arch_hardware_setup(void)
9530 r = kvm_x86_ops->hardware_setup();
9534 cr4_reserved_bits = kvm_host_cr4_reserved_bits(&boot_cpu_data);
9536 if (kvm_has_tsc_control) {
9538 * Make sure the user can only configure tsc_khz values that
9539 * fit into a signed integer.
9540 * A min value is not calculated because it will always
9541 * be 1 on all machines.
9543 u64 max = min(0x7fffffffULL,
9544 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9545 kvm_max_guest_tsc_khz = max;
9547 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9550 if (boot_cpu_has(X86_FEATURE_XSAVES))
9551 rdmsrl(MSR_IA32_XSS, host_xss);
9553 kvm_init_msr_list();
9557 void kvm_arch_hardware_unsetup(void)
9559 kvm_x86_ops->hardware_unsetup();
9562 int kvm_arch_check_processor_compat(void)
9564 struct cpuinfo_x86 *c = &cpu_data(smp_processor_id());
9566 WARN_ON(!irqs_disabled());
9568 if (kvm_host_cr4_reserved_bits(c) != cr4_reserved_bits)
9571 return kvm_x86_ops->check_processor_compatibility();
9574 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9576 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9578 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9580 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9582 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9585 struct static_key kvm_no_apic_vcpu __read_mostly;
9586 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9588 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9590 struct kvm_pmu *pmu = vcpu_to_pmu(vcpu);
9592 vcpu->arch.l1tf_flush_l1d = true;
9593 if (pmu->version && unlikely(pmu->event_count)) {
9594 pmu->need_cleanup = true;
9595 kvm_make_request(KVM_REQ_PMU, vcpu);
9597 kvm_x86_ops->sched_in(vcpu, cpu);
9600 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9605 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9606 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9607 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9608 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9609 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9610 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9612 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9613 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9614 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9615 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9616 &kvm->arch.irq_sources_bitmap);
9618 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9619 mutex_init(&kvm->arch.apic_map_lock);
9620 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9622 kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9623 pvclock_update_vm_gtod_copy(kvm);
9625 kvm->arch.guest_can_read_msr_platform_info = true;
9627 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9628 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9630 kvm_hv_init_vm(kvm);
9631 kvm_page_track_init(kvm);
9632 kvm_mmu_init_vm(kvm);
9634 return kvm_x86_ops->vm_init(kvm);
9637 int kvm_arch_post_init_vm(struct kvm *kvm)
9639 return kvm_mmu_post_init_vm(kvm);
9642 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9645 kvm_mmu_unload(vcpu);
9649 static void kvm_free_vcpus(struct kvm *kvm)
9652 struct kvm_vcpu *vcpu;
9655 * Unpin any mmu pages first.
9657 kvm_for_each_vcpu(i, vcpu, kvm) {
9658 kvm_clear_async_pf_completion_queue(vcpu);
9659 kvm_unload_vcpu_mmu(vcpu);
9661 kvm_for_each_vcpu(i, vcpu, kvm)
9662 kvm_vcpu_destroy(vcpu);
9664 mutex_lock(&kvm->lock);
9665 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9666 kvm->vcpus[i] = NULL;
9668 atomic_set(&kvm->online_vcpus, 0);
9669 mutex_unlock(&kvm->lock);
9672 void kvm_arch_sync_events(struct kvm *kvm)
9674 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9675 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9679 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9683 struct kvm_memslots *slots = kvm_memslots(kvm);
9684 struct kvm_memory_slot *slot, old;
9686 /* Called with kvm->slots_lock held. */
9687 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9690 slot = id_to_memslot(slots, id);
9696 * MAP_SHARED to prevent internal slot pages from being moved
9699 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9700 MAP_SHARED | MAP_ANONYMOUS, 0);
9701 if (IS_ERR((void *)hva))
9702 return PTR_ERR((void *)hva);
9711 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9712 struct kvm_userspace_memory_region m;
9714 m.slot = id | (i << 16);
9716 m.guest_phys_addr = gpa;
9717 m.userspace_addr = hva;
9718 m.memory_size = size;
9719 r = __kvm_set_memory_region(kvm, &m);
9725 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9729 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9731 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9735 mutex_lock(&kvm->slots_lock);
9736 r = __x86_set_memory_region(kvm, id, gpa, size);
9737 mutex_unlock(&kvm->slots_lock);
9741 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9743 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9745 kvm_mmu_pre_destroy_vm(kvm);
9748 void kvm_arch_destroy_vm(struct kvm *kvm)
9750 if (current->mm == kvm->mm) {
9752 * Free memory regions allocated on behalf of userspace,
9753 * unless the the memory map has changed due to process exit
9756 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9757 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9758 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9760 if (kvm_x86_ops->vm_destroy)
9761 kvm_x86_ops->vm_destroy(kvm);
9762 kvm_pic_destroy(kvm);
9763 kvm_ioapic_destroy(kvm);
9764 kvm_free_vcpus(kvm);
9765 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9766 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9767 kvm_mmu_uninit_vm(kvm);
9768 kvm_page_track_cleanup(kvm);
9769 kvm_hv_destroy_vm(kvm);
9772 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9773 struct kvm_memory_slot *dont)
9777 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9778 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9779 kvfree(free->arch.rmap[i]);
9780 free->arch.rmap[i] = NULL;
9785 if (!dont || free->arch.lpage_info[i - 1] !=
9786 dont->arch.lpage_info[i - 1]) {
9787 kvfree(free->arch.lpage_info[i - 1]);
9788 free->arch.lpage_info[i - 1] = NULL;
9792 kvm_page_track_free_memslot(free, dont);
9795 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9796 unsigned long npages)
9800 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9801 struct kvm_lpage_info *linfo;
9806 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9807 slot->base_gfn, level) + 1;
9809 slot->arch.rmap[i] =
9810 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9811 GFP_KERNEL_ACCOUNT);
9812 if (!slot->arch.rmap[i])
9817 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9821 slot->arch.lpage_info[i - 1] = linfo;
9823 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9824 linfo[0].disallow_lpage = 1;
9825 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9826 linfo[lpages - 1].disallow_lpage = 1;
9827 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9829 * If the gfn and userspace address are not aligned wrt each
9830 * other, or if explicitly asked to, disable large page
9831 * support for this slot
9833 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9834 !kvm_largepages_enabled()) {
9837 for (j = 0; j < lpages; ++j)
9838 linfo[j].disallow_lpage = 1;
9842 if (kvm_page_track_create_memslot(slot, npages))
9848 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9849 kvfree(slot->arch.rmap[i]);
9850 slot->arch.rmap[i] = NULL;
9854 kvfree(slot->arch.lpage_info[i - 1]);
9855 slot->arch.lpage_info[i - 1] = NULL;
9860 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9863 * memslots->generation has been incremented.
9864 * mmio generation may have reached its maximum value.
9866 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9869 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9870 struct kvm_memory_slot *memslot,
9871 const struct kvm_userspace_memory_region *mem,
9872 enum kvm_mr_change change)
9877 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9878 struct kvm_memory_slot *new)
9880 /* Still write protect RO slot */
9881 if (new->flags & KVM_MEM_READONLY) {
9882 kvm_mmu_slot_remove_write_access(kvm, new);
9887 * Call kvm_x86_ops dirty logging hooks when they are valid.
9889 * kvm_x86_ops->slot_disable_log_dirty is called when:
9891 * - KVM_MR_CREATE with dirty logging is disabled
9892 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9894 * The reason is, in case of PML, we need to set D-bit for any slots
9895 * with dirty logging disabled in order to eliminate unnecessary GPA
9896 * logging in PML buffer (and potential PML buffer full VMEXIT). This
9897 * guarantees leaving PML enabled during guest's lifetime won't have
9898 * any additional overhead from PML when guest is running with dirty
9899 * logging disabled for memory slots.
9901 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9902 * to dirty logging mode.
9904 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9906 * In case of write protect:
9908 * Write protect all pages for dirty logging.
9910 * All the sptes including the large sptes which point to this
9911 * slot are set to readonly. We can not create any new large
9912 * spte on this slot until the end of the logging.
9914 * See the comments in fast_page_fault().
9916 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9917 if (kvm_x86_ops->slot_enable_log_dirty)
9918 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9920 kvm_mmu_slot_remove_write_access(kvm, new);
9922 if (kvm_x86_ops->slot_disable_log_dirty)
9923 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9927 void kvm_arch_commit_memory_region(struct kvm *kvm,
9928 const struct kvm_userspace_memory_region *mem,
9929 const struct kvm_memory_slot *old,
9930 const struct kvm_memory_slot *new,
9931 enum kvm_mr_change change)
9933 if (!kvm->arch.n_requested_mmu_pages)
9934 kvm_mmu_change_mmu_pages(kvm,
9935 kvm_mmu_calculate_default_mmu_pages(kvm));
9938 * Dirty logging tracks sptes in 4k granularity, meaning that large
9939 * sptes have to be split. If live migration is successful, the guest
9940 * in the source machine will be destroyed and large sptes will be
9941 * created in the destination. However, if the guest continues to run
9942 * in the source machine (for example if live migration fails), small
9943 * sptes will remain around and cause bad performance.
9945 * Scan sptes if dirty logging has been stopped, dropping those
9946 * which can be collapsed into a single large-page spte. Later
9947 * page faults will create the large-page sptes.
9949 * There is no need to do this in any of the following cases:
9950 * CREATE: No dirty mappings will already exist.
9951 * MOVE/DELETE: The old mappings will already have been cleaned up by
9952 * kvm_arch_flush_shadow_memslot()
9954 if (change == KVM_MR_FLAGS_ONLY &&
9955 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9956 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9957 kvm_mmu_zap_collapsible_sptes(kvm, new);
9960 * Set up write protection and/or dirty logging for the new slot.
9962 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9963 * been zapped so no dirty logging staff is needed for old slot. For
9964 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9965 * new and it's also covered when dealing with the new slot.
9967 * FIXME: const-ify all uses of struct kvm_memory_slot.
9969 if (change != KVM_MR_DELETE)
9970 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9973 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9975 kvm_mmu_zap_all(kvm);
9978 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9979 struct kvm_memory_slot *slot)
9981 kvm_page_track_flush_slot(kvm, slot);
9984 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9986 return (is_guest_mode(vcpu) &&
9987 kvm_x86_ops->guest_apic_has_interrupt &&
9988 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9991 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9993 if (!list_empty_careful(&vcpu->async_pf.done))
9996 if (kvm_apic_has_events(vcpu))
9999 if (vcpu->arch.pv.pv_unhalted)
10002 if (vcpu->arch.exception.pending)
10005 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10006 (vcpu->arch.nmi_pending &&
10007 kvm_x86_ops->nmi_allowed(vcpu)))
10010 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
10011 (vcpu->arch.smi_pending && !is_smm(vcpu)))
10014 if (kvm_arch_interrupt_allowed(vcpu) &&
10015 (kvm_cpu_has_interrupt(vcpu) ||
10016 kvm_guest_apic_has_interrupt(vcpu)))
10019 if (kvm_hv_has_stimer_pending(vcpu))
10025 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
10027 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
10030 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
10032 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
10035 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
10036 kvm_test_request(KVM_REQ_SMI, vcpu) ||
10037 kvm_test_request(KVM_REQ_EVENT, vcpu))
10040 if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
10046 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
10048 return vcpu->arch.preempted_in_kernel;
10051 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
10053 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
10056 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
10058 return kvm_x86_ops->interrupt_allowed(vcpu);
10061 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
10063 if (is_64_bit_mode(vcpu))
10064 return kvm_rip_read(vcpu);
10065 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
10066 kvm_rip_read(vcpu));
10068 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
10070 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
10072 return kvm_get_linear_rip(vcpu) == linear_rip;
10074 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
10076 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
10078 unsigned long rflags;
10080 rflags = kvm_x86_ops->get_rflags(vcpu);
10081 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
10082 rflags &= ~X86_EFLAGS_TF;
10085 EXPORT_SYMBOL_GPL(kvm_get_rflags);
10087 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10089 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
10090 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
10091 rflags |= X86_EFLAGS_TF;
10092 kvm_x86_ops->set_rflags(vcpu, rflags);
10095 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
10097 __kvm_set_rflags(vcpu, rflags);
10098 kvm_make_request(KVM_REQ_EVENT, vcpu);
10100 EXPORT_SYMBOL_GPL(kvm_set_rflags);
10102 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
10106 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
10110 r = kvm_mmu_reload(vcpu);
10114 if (!vcpu->arch.mmu->direct_map &&
10115 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
10118 vcpu->arch.mmu->page_fault(vcpu, work->cr2_or_gpa, 0, true);
10121 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
10123 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
10126 static inline u32 kvm_async_pf_next_probe(u32 key)
10128 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
10131 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10133 u32 key = kvm_async_pf_hash_fn(gfn);
10135 while (vcpu->arch.apf.gfns[key] != ~0)
10136 key = kvm_async_pf_next_probe(key);
10138 vcpu->arch.apf.gfns[key] = gfn;
10141 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
10144 u32 key = kvm_async_pf_hash_fn(gfn);
10146 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
10147 (vcpu->arch.apf.gfns[key] != gfn &&
10148 vcpu->arch.apf.gfns[key] != ~0); i++)
10149 key = kvm_async_pf_next_probe(key);
10154 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10156 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10159 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10163 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10165 vcpu->arch.apf.gfns[i] = ~0;
10167 j = kvm_async_pf_next_probe(j);
10168 if (vcpu->arch.apf.gfns[j] == ~0)
10170 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10172 * k lies cyclically in ]i,j]
10174 * |....j i.k.| or |.k..j i...|
10176 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10177 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10182 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10185 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10189 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10192 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10196 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10198 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10201 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10202 (vcpu->arch.apf.send_user_only &&
10203 kvm_x86_ops->get_cpl(vcpu) == 0))
10209 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10211 if (unlikely(!lapic_in_kernel(vcpu) ||
10212 kvm_event_needs_reinjection(vcpu) ||
10213 vcpu->arch.exception.pending))
10216 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10220 * If interrupts are off we cannot even use an artificial
10223 return kvm_x86_ops->interrupt_allowed(vcpu);
10226 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10227 struct kvm_async_pf *work)
10229 struct x86_exception fault;
10231 trace_kvm_async_pf_not_present(work->arch.token, work->cr2_or_gpa);
10232 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10234 if (kvm_can_deliver_async_pf(vcpu) &&
10235 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10236 fault.vector = PF_VECTOR;
10237 fault.error_code_valid = true;
10238 fault.error_code = 0;
10239 fault.nested_page_fault = false;
10240 fault.address = work->arch.token;
10241 fault.async_page_fault = true;
10242 kvm_inject_page_fault(vcpu, &fault);
10245 * It is not possible to deliver a paravirtualized asynchronous
10246 * page fault, but putting the guest in an artificial halt state
10247 * can be beneficial nevertheless: if an interrupt arrives, we
10248 * can deliver it timely and perhaps the guest will schedule
10249 * another process. When the instruction that triggered a page
10250 * fault is retried, hopefully the page will be ready in the host.
10252 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10256 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10257 struct kvm_async_pf *work)
10259 struct x86_exception fault;
10262 if (work->wakeup_all)
10263 work->arch.token = ~0; /* broadcast wakeup */
10265 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10266 trace_kvm_async_pf_ready(work->arch.token, work->cr2_or_gpa);
10268 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10269 !apf_get_user(vcpu, &val)) {
10270 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10271 vcpu->arch.exception.pending &&
10272 vcpu->arch.exception.nr == PF_VECTOR &&
10273 !apf_put_user(vcpu, 0)) {
10274 vcpu->arch.exception.injected = false;
10275 vcpu->arch.exception.pending = false;
10276 vcpu->arch.exception.nr = 0;
10277 vcpu->arch.exception.has_error_code = false;
10278 vcpu->arch.exception.error_code = 0;
10279 vcpu->arch.exception.has_payload = false;
10280 vcpu->arch.exception.payload = 0;
10281 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10282 fault.vector = PF_VECTOR;
10283 fault.error_code_valid = true;
10284 fault.error_code = 0;
10285 fault.nested_page_fault = false;
10286 fault.address = work->arch.token;
10287 fault.async_page_fault = true;
10288 kvm_inject_page_fault(vcpu, &fault);
10291 vcpu->arch.apf.halted = false;
10292 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10295 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10297 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10300 return kvm_can_do_async_pf(vcpu);
10303 void kvm_arch_start_assignment(struct kvm *kvm)
10305 atomic_inc(&kvm->arch.assigned_device_count);
10307 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10309 void kvm_arch_end_assignment(struct kvm *kvm)
10311 atomic_dec(&kvm->arch.assigned_device_count);
10313 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10315 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10317 return atomic_read(&kvm->arch.assigned_device_count);
10319 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10321 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10323 atomic_inc(&kvm->arch.noncoherent_dma_count);
10325 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10327 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10329 atomic_dec(&kvm->arch.noncoherent_dma_count);
10331 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10333 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10335 return atomic_read(&kvm->arch.noncoherent_dma_count);
10337 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10339 bool kvm_arch_has_irq_bypass(void)
10344 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10345 struct irq_bypass_producer *prod)
10347 struct kvm_kernel_irqfd *irqfd =
10348 container_of(cons, struct kvm_kernel_irqfd, consumer);
10350 irqfd->producer = prod;
10352 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10353 prod->irq, irqfd->gsi, 1);
10356 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10357 struct irq_bypass_producer *prod)
10360 struct kvm_kernel_irqfd *irqfd =
10361 container_of(cons, struct kvm_kernel_irqfd, consumer);
10363 WARN_ON(irqfd->producer != prod);
10364 irqfd->producer = NULL;
10367 * When producer of consumer is unregistered, we change back to
10368 * remapped mode, so we can re-use the current implementation
10369 * when the irq is masked/disabled or the consumer side (KVM
10370 * int this case doesn't want to receive the interrupts.
10372 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10374 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10375 " fails: %d\n", irqfd->consumer.token, ret);
10378 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10379 uint32_t guest_irq, bool set)
10381 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10384 bool kvm_vector_hashing_enabled(void)
10386 return vector_hashing;
10388 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10390 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10392 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10394 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10396 u64 kvm_spec_ctrl_valid_bits(struct kvm_vcpu *vcpu)
10398 uint64_t bits = SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD;
10400 /* The STIBP bit doesn't fault even if it's not advertised */
10401 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL) &&
10402 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_IBRS))
10403 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10404 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL) &&
10405 !boot_cpu_has(X86_FEATURE_AMD_IBRS))
10406 bits &= ~(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP);
10408 if (!guest_cpuid_has(vcpu, X86_FEATURE_SPEC_CTRL_SSBD) &&
10409 !guest_cpuid_has(vcpu, X86_FEATURE_AMD_SSBD))
10410 bits &= ~SPEC_CTRL_SSBD;
10411 if (!boot_cpu_has(X86_FEATURE_SPEC_CTRL_SSBD) &&
10412 !boot_cpu_has(X86_FEATURE_AMD_SSBD))
10413 bits &= ~SPEC_CTRL_SSBD;
10417 EXPORT_SYMBOL_GPL(kvm_spec_ctrl_valid_bits);
10419 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10420 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10421 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10422 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10423 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10424 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10425 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10426 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10427 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10428 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10429 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10430 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10431 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10432 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10433 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10434 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10435 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10436 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10437 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10438 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);