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 <clocksource/hyperv_timer.h>
73 #define CREATE_TRACE_POINTS
76 #define MAX_IO_MSRS 256
77 #define KVM_MAX_MCE_BANKS 32
78 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
79 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
81 #define emul_to_vcpu(ctxt) \
82 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
85 * - enable syscall per default because its emulated by KVM
86 * - enable LME and LMA per default on 64 bit KVM
90 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
92 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
95 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
96 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
98 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
99 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
101 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
102 static void process_nmi(struct kvm_vcpu *vcpu);
103 static void enter_smm(struct kvm_vcpu *vcpu);
104 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
105 static void store_regs(struct kvm_vcpu *vcpu);
106 static int sync_regs(struct kvm_vcpu *vcpu);
108 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
109 EXPORT_SYMBOL_GPL(kvm_x86_ops);
111 static bool __read_mostly ignore_msrs = 0;
112 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
114 static bool __read_mostly report_ignored_msrs = true;
115 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
117 unsigned int min_timer_period_us = 200;
118 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
120 static bool __read_mostly kvmclock_periodic_sync = true;
121 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
123 bool __read_mostly kvm_has_tsc_control;
124 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
125 u32 __read_mostly kvm_max_guest_tsc_khz;
126 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
127 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
128 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
129 u64 __read_mostly kvm_max_tsc_scaling_ratio;
130 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
131 u64 __read_mostly kvm_default_tsc_scaling_ratio;
132 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
134 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
135 static u32 __read_mostly tsc_tolerance_ppm = 250;
136 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
139 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
140 * adaptive tuning starting from default advancment of 1000ns. '0' disables
141 * advancement entirely. Any other value is used as-is and disables adaptive
142 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
144 static int __read_mostly lapic_timer_advance_ns = -1;
145 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
147 static bool __read_mostly vector_hashing = true;
148 module_param(vector_hashing, bool, S_IRUGO);
150 bool __read_mostly enable_vmware_backdoor = false;
151 module_param(enable_vmware_backdoor, bool, S_IRUGO);
152 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
154 static bool __read_mostly force_emulation_prefix = false;
155 module_param(force_emulation_prefix, bool, S_IRUGO);
157 int __read_mostly pi_inject_timer = -1;
158 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
160 #define KVM_NR_SHARED_MSRS 16
162 struct kvm_shared_msrs_global {
164 u32 msrs[KVM_NR_SHARED_MSRS];
167 struct kvm_shared_msrs {
168 struct user_return_notifier urn;
170 struct kvm_shared_msr_values {
173 } values[KVM_NR_SHARED_MSRS];
176 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
177 static struct kvm_shared_msrs __percpu *shared_msrs;
179 struct kvm_stats_debugfs_item debugfs_entries[] = {
180 { "pf_fixed", VCPU_STAT(pf_fixed) },
181 { "pf_guest", VCPU_STAT(pf_guest) },
182 { "tlb_flush", VCPU_STAT(tlb_flush) },
183 { "invlpg", VCPU_STAT(invlpg) },
184 { "exits", VCPU_STAT(exits) },
185 { "io_exits", VCPU_STAT(io_exits) },
186 { "mmio_exits", VCPU_STAT(mmio_exits) },
187 { "signal_exits", VCPU_STAT(signal_exits) },
188 { "irq_window", VCPU_STAT(irq_window_exits) },
189 { "nmi_window", VCPU_STAT(nmi_window_exits) },
190 { "halt_exits", VCPU_STAT(halt_exits) },
191 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
192 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
193 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
194 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
195 { "hypercalls", VCPU_STAT(hypercalls) },
196 { "request_irq", VCPU_STAT(request_irq_exits) },
197 { "irq_exits", VCPU_STAT(irq_exits) },
198 { "host_state_reload", VCPU_STAT(host_state_reload) },
199 { "fpu_reload", VCPU_STAT(fpu_reload) },
200 { "insn_emulation", VCPU_STAT(insn_emulation) },
201 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
202 { "irq_injections", VCPU_STAT(irq_injections) },
203 { "nmi_injections", VCPU_STAT(nmi_injections) },
204 { "req_event", VCPU_STAT(req_event) },
205 { "l1d_flush", VCPU_STAT(l1d_flush) },
206 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
207 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
208 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
209 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
210 { "mmu_flooded", VM_STAT(mmu_flooded) },
211 { "mmu_recycled", VM_STAT(mmu_recycled) },
212 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
213 { "mmu_unsync", VM_STAT(mmu_unsync) },
214 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
215 { "largepages", VM_STAT(lpages) },
216 { "max_mmu_page_hash_collisions",
217 VM_STAT(max_mmu_page_hash_collisions) },
221 u64 __read_mostly host_xcr0;
223 struct kmem_cache *x86_fpu_cache;
224 EXPORT_SYMBOL_GPL(x86_fpu_cache);
226 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
228 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
231 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
232 vcpu->arch.apf.gfns[i] = ~0;
235 static void kvm_on_user_return(struct user_return_notifier *urn)
238 struct kvm_shared_msrs *locals
239 = container_of(urn, struct kvm_shared_msrs, urn);
240 struct kvm_shared_msr_values *values;
244 * Disabling irqs at this point since the following code could be
245 * interrupted and executed through kvm_arch_hardware_disable()
247 local_irq_save(flags);
248 if (locals->registered) {
249 locals->registered = false;
250 user_return_notifier_unregister(urn);
252 local_irq_restore(flags);
253 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
254 values = &locals->values[slot];
255 if (values->host != values->curr) {
256 wrmsrl(shared_msrs_global.msrs[slot], values->host);
257 values->curr = values->host;
262 static void shared_msr_update(unsigned slot, u32 msr)
265 unsigned int cpu = smp_processor_id();
266 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
268 /* only read, and nobody should modify it at this time,
269 * so don't need lock */
270 if (slot >= shared_msrs_global.nr) {
271 printk(KERN_ERR "kvm: invalid MSR slot!");
274 rdmsrl_safe(msr, &value);
275 smsr->values[slot].host = value;
276 smsr->values[slot].curr = value;
279 void kvm_define_shared_msr(unsigned slot, u32 msr)
281 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
282 shared_msrs_global.msrs[slot] = msr;
283 if (slot >= shared_msrs_global.nr)
284 shared_msrs_global.nr = slot + 1;
286 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
288 static void kvm_shared_msr_cpu_online(void)
292 for (i = 0; i < shared_msrs_global.nr; ++i)
293 shared_msr_update(i, shared_msrs_global.msrs[i]);
296 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
298 unsigned int cpu = smp_processor_id();
299 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
302 if (((value ^ smsr->values[slot].curr) & mask) == 0)
304 smsr->values[slot].curr = value;
305 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
309 if (!smsr->registered) {
310 smsr->urn.on_user_return = kvm_on_user_return;
311 user_return_notifier_register(&smsr->urn);
312 smsr->registered = true;
316 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
318 static void drop_user_return_notifiers(void)
320 unsigned int cpu = smp_processor_id();
321 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
323 if (smsr->registered)
324 kvm_on_user_return(&smsr->urn);
327 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
329 return vcpu->arch.apic_base;
331 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
333 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
335 return kvm_apic_mode(kvm_get_apic_base(vcpu));
337 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
339 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
341 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
342 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
343 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
344 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
346 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
348 if (!msr_info->host_initiated) {
349 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
351 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
355 kvm_lapic_set_base(vcpu, msr_info->data);
358 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
360 asmlinkage __visible void kvm_spurious_fault(void)
362 /* Fault while not rebooting. We want the trace. */
366 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
368 #define EXCPT_BENIGN 0
369 #define EXCPT_CONTRIBUTORY 1
372 static int exception_class(int vector)
382 return EXCPT_CONTRIBUTORY;
389 #define EXCPT_FAULT 0
391 #define EXCPT_ABORT 2
392 #define EXCPT_INTERRUPT 3
394 static int exception_type(int vector)
398 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
399 return EXCPT_INTERRUPT;
403 /* #DB is trap, as instruction watchpoints are handled elsewhere */
404 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
407 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
410 /* Reserved exceptions will result in fault */
414 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
416 unsigned nr = vcpu->arch.exception.nr;
417 bool has_payload = vcpu->arch.exception.has_payload;
418 unsigned long payload = vcpu->arch.exception.payload;
426 * "Certain debug exceptions may clear bit 0-3. The
427 * remaining contents of the DR6 register are never
428 * cleared by the processor".
430 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
432 * DR6.RTM is set by all #DB exceptions that don't clear it.
434 vcpu->arch.dr6 |= DR6_RTM;
435 vcpu->arch.dr6 |= payload;
437 * Bit 16 should be set in the payload whenever the #DB
438 * exception should clear DR6.RTM. This makes the payload
439 * compatible with the pending debug exceptions under VMX.
440 * Though not currently documented in the SDM, this also
441 * makes the payload compatible with the exit qualification
442 * for #DB exceptions under VMX.
444 vcpu->arch.dr6 ^= payload & DR6_RTM;
447 vcpu->arch.cr2 = payload;
451 vcpu->arch.exception.has_payload = false;
452 vcpu->arch.exception.payload = 0;
454 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
456 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
457 unsigned nr, bool has_error, u32 error_code,
458 bool has_payload, unsigned long payload, bool reinject)
463 kvm_make_request(KVM_REQ_EVENT, vcpu);
465 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
467 if (has_error && !is_protmode(vcpu))
471 * On vmentry, vcpu->arch.exception.pending is only
472 * true if an event injection was blocked by
473 * nested_run_pending. In that case, however,
474 * vcpu_enter_guest requests an immediate exit,
475 * and the guest shouldn't proceed far enough to
478 WARN_ON_ONCE(vcpu->arch.exception.pending);
479 vcpu->arch.exception.injected = true;
480 if (WARN_ON_ONCE(has_payload)) {
482 * A reinjected event has already
483 * delivered its payload.
489 vcpu->arch.exception.pending = true;
490 vcpu->arch.exception.injected = false;
492 vcpu->arch.exception.has_error_code = has_error;
493 vcpu->arch.exception.nr = nr;
494 vcpu->arch.exception.error_code = error_code;
495 vcpu->arch.exception.has_payload = has_payload;
496 vcpu->arch.exception.payload = payload;
498 * In guest mode, payload delivery should be deferred,
499 * so that the L1 hypervisor can intercept #PF before
500 * CR2 is modified (or intercept #DB before DR6 is
501 * modified under nVMX). However, for ABI
502 * compatibility with KVM_GET_VCPU_EVENTS and
503 * KVM_SET_VCPU_EVENTS, we can't delay payload
504 * delivery unless userspace has enabled this
505 * functionality via the per-VM capability,
506 * KVM_CAP_EXCEPTION_PAYLOAD.
508 if (!vcpu->kvm->arch.exception_payload_enabled ||
509 !is_guest_mode(vcpu))
510 kvm_deliver_exception_payload(vcpu);
514 /* to check exception */
515 prev_nr = vcpu->arch.exception.nr;
516 if (prev_nr == DF_VECTOR) {
517 /* triple fault -> shutdown */
518 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
521 class1 = exception_class(prev_nr);
522 class2 = exception_class(nr);
523 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
524 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
526 * Generate double fault per SDM Table 5-5. Set
527 * exception.pending = true so that the double fault
528 * can trigger a nested vmexit.
530 vcpu->arch.exception.pending = true;
531 vcpu->arch.exception.injected = false;
532 vcpu->arch.exception.has_error_code = true;
533 vcpu->arch.exception.nr = DF_VECTOR;
534 vcpu->arch.exception.error_code = 0;
535 vcpu->arch.exception.has_payload = false;
536 vcpu->arch.exception.payload = 0;
538 /* replace previous exception with a new one in a hope
539 that instruction re-execution will regenerate lost
544 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
546 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
548 EXPORT_SYMBOL_GPL(kvm_queue_exception);
550 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
552 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
554 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
556 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
557 unsigned long payload)
559 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
562 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
563 u32 error_code, unsigned long payload)
565 kvm_multiple_exception(vcpu, nr, true, error_code,
566 true, payload, false);
569 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
572 kvm_inject_gp(vcpu, 0);
574 return kvm_skip_emulated_instruction(vcpu);
578 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
580 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
582 ++vcpu->stat.pf_guest;
583 vcpu->arch.exception.nested_apf =
584 is_guest_mode(vcpu) && fault->async_page_fault;
585 if (vcpu->arch.exception.nested_apf) {
586 vcpu->arch.apf.nested_apf_token = fault->address;
587 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
589 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
593 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
595 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
597 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
598 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
600 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
602 return fault->nested_page_fault;
605 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
607 atomic_inc(&vcpu->arch.nmi_queued);
608 kvm_make_request(KVM_REQ_NMI, vcpu);
610 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
612 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
614 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
616 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
618 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
620 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
622 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
625 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
626 * a #GP and return false.
628 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
630 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
632 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
635 EXPORT_SYMBOL_GPL(kvm_require_cpl);
637 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
639 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
642 kvm_queue_exception(vcpu, UD_VECTOR);
645 EXPORT_SYMBOL_GPL(kvm_require_dr);
648 * This function will be used to read from the physical memory of the currently
649 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
650 * can read from guest physical or from the guest's guest physical memory.
652 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
653 gfn_t ngfn, void *data, int offset, int len,
656 struct x86_exception exception;
660 ngpa = gfn_to_gpa(ngfn);
661 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
662 if (real_gfn == UNMAPPED_GVA)
665 real_gfn = gpa_to_gfn(real_gfn);
667 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
669 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
671 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
672 void *data, int offset, int len, u32 access)
674 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
675 data, offset, len, access);
678 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
680 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
685 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
687 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
689 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
690 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
693 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
695 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
696 offset * sizeof(u64), sizeof(pdpte),
697 PFERR_USER_MASK|PFERR_WRITE_MASK);
702 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
703 if ((pdpte[i] & PT_PRESENT_MASK) &&
704 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
711 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
712 __set_bit(VCPU_EXREG_PDPTR,
713 (unsigned long *)&vcpu->arch.regs_avail);
714 __set_bit(VCPU_EXREG_PDPTR,
715 (unsigned long *)&vcpu->arch.regs_dirty);
720 EXPORT_SYMBOL_GPL(load_pdptrs);
722 bool pdptrs_changed(struct kvm_vcpu *vcpu)
724 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
730 if (!is_pae_paging(vcpu))
733 if (!test_bit(VCPU_EXREG_PDPTR,
734 (unsigned long *)&vcpu->arch.regs_avail))
737 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
738 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
739 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
740 PFERR_USER_MASK | PFERR_WRITE_MASK);
743 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
748 EXPORT_SYMBOL_GPL(pdptrs_changed);
750 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
752 unsigned long old_cr0 = kvm_read_cr0(vcpu);
753 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
758 if (cr0 & 0xffffffff00000000UL)
762 cr0 &= ~CR0_RESERVED_BITS;
764 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
767 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
770 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
772 if ((vcpu->arch.efer & EFER_LME)) {
777 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
782 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
787 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
790 kvm_x86_ops->set_cr0(vcpu, cr0);
792 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
793 kvm_clear_async_pf_completion_queue(vcpu);
794 kvm_async_pf_hash_reset(vcpu);
797 if ((cr0 ^ old_cr0) & update_bits)
798 kvm_mmu_reset_context(vcpu);
800 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
801 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
802 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
803 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
807 EXPORT_SYMBOL_GPL(kvm_set_cr0);
809 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
811 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
813 EXPORT_SYMBOL_GPL(kvm_lmsw);
815 void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
817 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
818 !vcpu->guest_xcr0_loaded) {
819 /* kvm_set_xcr() also depends on this */
820 if (vcpu->arch.xcr0 != host_xcr0)
821 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
822 vcpu->guest_xcr0_loaded = 1;
825 EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
827 void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
829 if (vcpu->guest_xcr0_loaded) {
830 if (vcpu->arch.xcr0 != host_xcr0)
831 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
832 vcpu->guest_xcr0_loaded = 0;
835 EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
837 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
840 u64 old_xcr0 = vcpu->arch.xcr0;
843 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
844 if (index != XCR_XFEATURE_ENABLED_MASK)
846 if (!(xcr0 & XFEATURE_MASK_FP))
848 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
852 * Do not allow the guest to set bits that we do not support
853 * saving. However, xcr0 bit 0 is always set, even if the
854 * emulated CPU does not support XSAVE (see fx_init).
856 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
857 if (xcr0 & ~valid_bits)
860 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
861 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
864 if (xcr0 & XFEATURE_MASK_AVX512) {
865 if (!(xcr0 & XFEATURE_MASK_YMM))
867 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
870 vcpu->arch.xcr0 = xcr0;
872 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
873 kvm_update_cpuid(vcpu);
877 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
879 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
880 __kvm_set_xcr(vcpu, index, xcr)) {
881 kvm_inject_gp(vcpu, 0);
886 EXPORT_SYMBOL_GPL(kvm_set_xcr);
888 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
890 unsigned long old_cr4 = kvm_read_cr4(vcpu);
891 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
892 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
894 if (cr4 & CR4_RESERVED_BITS)
897 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
900 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
903 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
906 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
909 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
912 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
915 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
918 if (is_long_mode(vcpu)) {
919 if (!(cr4 & X86_CR4_PAE))
921 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
922 && ((cr4 ^ old_cr4) & pdptr_bits)
923 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
927 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
928 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
931 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
932 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
936 if (kvm_x86_ops->set_cr4(vcpu, cr4))
939 if (((cr4 ^ old_cr4) & pdptr_bits) ||
940 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
941 kvm_mmu_reset_context(vcpu);
943 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
944 kvm_update_cpuid(vcpu);
948 EXPORT_SYMBOL_GPL(kvm_set_cr4);
950 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
952 bool skip_tlb_flush = false;
954 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
957 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
958 cr3 &= ~X86_CR3_PCID_NOFLUSH;
962 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
963 if (!skip_tlb_flush) {
964 kvm_mmu_sync_roots(vcpu);
965 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
970 if (is_long_mode(vcpu) &&
971 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
973 else if (is_pae_paging(vcpu) &&
974 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
977 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
978 vcpu->arch.cr3 = cr3;
979 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
983 EXPORT_SYMBOL_GPL(kvm_set_cr3);
985 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
987 if (cr8 & CR8_RESERVED_BITS)
989 if (lapic_in_kernel(vcpu))
990 kvm_lapic_set_tpr(vcpu, cr8);
992 vcpu->arch.cr8 = cr8;
995 EXPORT_SYMBOL_GPL(kvm_set_cr8);
997 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
999 if (lapic_in_kernel(vcpu))
1000 return kvm_lapic_get_cr8(vcpu);
1002 return vcpu->arch.cr8;
1004 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1006 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1010 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1011 for (i = 0; i < KVM_NR_DB_REGS; i++)
1012 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1013 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1017 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1019 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1020 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1023 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1027 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1028 dr7 = vcpu->arch.guest_debug_dr7;
1030 dr7 = vcpu->arch.dr7;
1031 kvm_x86_ops->set_dr7(vcpu, dr7);
1032 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1033 if (dr7 & DR7_BP_EN_MASK)
1034 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1037 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1039 u64 fixed = DR6_FIXED_1;
1041 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1046 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1050 vcpu->arch.db[dr] = val;
1051 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1052 vcpu->arch.eff_db[dr] = val;
1057 if (val & 0xffffffff00000000ULL)
1058 return -1; /* #GP */
1059 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1060 kvm_update_dr6(vcpu);
1065 if (val & 0xffffffff00000000ULL)
1066 return -1; /* #GP */
1067 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1068 kvm_update_dr7(vcpu);
1075 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1077 if (__kvm_set_dr(vcpu, dr, val)) {
1078 kvm_inject_gp(vcpu, 0);
1083 EXPORT_SYMBOL_GPL(kvm_set_dr);
1085 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1089 *val = vcpu->arch.db[dr];
1094 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1095 *val = vcpu->arch.dr6;
1097 *val = kvm_x86_ops->get_dr6(vcpu);
1102 *val = vcpu->arch.dr7;
1107 EXPORT_SYMBOL_GPL(kvm_get_dr);
1109 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1111 u32 ecx = kvm_rcx_read(vcpu);
1115 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1118 kvm_rax_write(vcpu, (u32)data);
1119 kvm_rdx_write(vcpu, data >> 32);
1122 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1125 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1126 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1128 * This list is modified at module load time to reflect the
1129 * capabilities of the host cpu. This capabilities test skips MSRs that are
1130 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1131 * may depend on host virtualization features rather than host cpu features.
1134 static u32 msrs_to_save[] = {
1135 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1137 #ifdef CONFIG_X86_64
1138 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1140 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1141 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1143 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1144 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1145 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1146 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1147 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1148 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1149 MSR_IA32_UMWAIT_CONTROL,
1151 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1152 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1153 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1154 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1155 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1156 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1157 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1158 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1159 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1160 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1161 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1162 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1163 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1164 MSR_ARCH_PERFMON_PERFCTR0 + 18, MSR_ARCH_PERFMON_PERFCTR0 + 19,
1165 MSR_ARCH_PERFMON_PERFCTR0 + 20, MSR_ARCH_PERFMON_PERFCTR0 + 21,
1166 MSR_ARCH_PERFMON_PERFCTR0 + 22, MSR_ARCH_PERFMON_PERFCTR0 + 23,
1167 MSR_ARCH_PERFMON_PERFCTR0 + 24, MSR_ARCH_PERFMON_PERFCTR0 + 25,
1168 MSR_ARCH_PERFMON_PERFCTR0 + 26, MSR_ARCH_PERFMON_PERFCTR0 + 27,
1169 MSR_ARCH_PERFMON_PERFCTR0 + 28, MSR_ARCH_PERFMON_PERFCTR0 + 29,
1170 MSR_ARCH_PERFMON_PERFCTR0 + 30, MSR_ARCH_PERFMON_PERFCTR0 + 31,
1171 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1172 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1173 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1174 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1175 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1176 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1177 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1178 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1179 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1180 MSR_ARCH_PERFMON_EVENTSEL0 + 18, MSR_ARCH_PERFMON_EVENTSEL0 + 19,
1181 MSR_ARCH_PERFMON_EVENTSEL0 + 20, MSR_ARCH_PERFMON_EVENTSEL0 + 21,
1182 MSR_ARCH_PERFMON_EVENTSEL0 + 22, MSR_ARCH_PERFMON_EVENTSEL0 + 23,
1183 MSR_ARCH_PERFMON_EVENTSEL0 + 24, MSR_ARCH_PERFMON_EVENTSEL0 + 25,
1184 MSR_ARCH_PERFMON_EVENTSEL0 + 26, MSR_ARCH_PERFMON_EVENTSEL0 + 27,
1185 MSR_ARCH_PERFMON_EVENTSEL0 + 28, MSR_ARCH_PERFMON_EVENTSEL0 + 29,
1186 MSR_ARCH_PERFMON_EVENTSEL0 + 30, MSR_ARCH_PERFMON_EVENTSEL0 + 31,
1189 static unsigned num_msrs_to_save;
1191 static u32 emulated_msrs[] = {
1192 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1193 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1194 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1195 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1196 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1197 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1198 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1200 HV_X64_MSR_VP_INDEX,
1201 HV_X64_MSR_VP_RUNTIME,
1202 HV_X64_MSR_SCONTROL,
1203 HV_X64_MSR_STIMER0_CONFIG,
1204 HV_X64_MSR_VP_ASSIST_PAGE,
1205 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1206 HV_X64_MSR_TSC_EMULATION_STATUS,
1208 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1211 MSR_IA32_TSC_ADJUST,
1212 MSR_IA32_TSCDEADLINE,
1213 MSR_IA32_ARCH_CAPABILITIES,
1214 MSR_IA32_MISC_ENABLE,
1215 MSR_IA32_MCG_STATUS,
1217 MSR_IA32_MCG_EXT_CTL,
1221 MSR_MISC_FEATURES_ENABLES,
1222 MSR_AMD64_VIRT_SPEC_CTRL,
1226 * The following list leaves out MSRs whose values are determined
1227 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1228 * We always support the "true" VMX control MSRs, even if the host
1229 * processor does not, so I am putting these registers here rather
1230 * than in msrs_to_save.
1233 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1234 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1235 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1236 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1238 MSR_IA32_VMX_CR0_FIXED0,
1239 MSR_IA32_VMX_CR4_FIXED0,
1240 MSR_IA32_VMX_VMCS_ENUM,
1241 MSR_IA32_VMX_PROCBASED_CTLS2,
1242 MSR_IA32_VMX_EPT_VPID_CAP,
1243 MSR_IA32_VMX_VMFUNC,
1246 MSR_KVM_POLL_CONTROL,
1249 static unsigned num_emulated_msrs;
1252 * List of msr numbers which are used to expose MSR-based features that
1253 * can be used by a hypervisor to validate requested CPU features.
1255 static u32 msr_based_features[] = {
1257 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1258 MSR_IA32_VMX_PINBASED_CTLS,
1259 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1260 MSR_IA32_VMX_PROCBASED_CTLS,
1261 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1262 MSR_IA32_VMX_EXIT_CTLS,
1263 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1264 MSR_IA32_VMX_ENTRY_CTLS,
1266 MSR_IA32_VMX_CR0_FIXED0,
1267 MSR_IA32_VMX_CR0_FIXED1,
1268 MSR_IA32_VMX_CR4_FIXED0,
1269 MSR_IA32_VMX_CR4_FIXED1,
1270 MSR_IA32_VMX_VMCS_ENUM,
1271 MSR_IA32_VMX_PROCBASED_CTLS2,
1272 MSR_IA32_VMX_EPT_VPID_CAP,
1273 MSR_IA32_VMX_VMFUNC,
1277 MSR_IA32_ARCH_CAPABILITIES,
1280 static unsigned int num_msr_based_features;
1282 static u64 kvm_get_arch_capabilities(void)
1286 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1287 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1290 * If we're doing cache flushes (either "always" or "cond")
1291 * we will do one whenever the guest does a vmlaunch/vmresume.
1292 * If an outer hypervisor is doing the cache flush for us
1293 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1294 * capability to the guest too, and if EPT is disabled we're not
1295 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1296 * require a nested hypervisor to do a flush of its own.
1298 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1299 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1301 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1302 data |= ARCH_CAP_RDCL_NO;
1303 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1304 data |= ARCH_CAP_SSB_NO;
1305 if (!boot_cpu_has_bug(X86_BUG_MDS))
1306 data |= ARCH_CAP_MDS_NO;
1311 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1313 switch (msr->index) {
1314 case MSR_IA32_ARCH_CAPABILITIES:
1315 msr->data = kvm_get_arch_capabilities();
1317 case MSR_IA32_UCODE_REV:
1318 rdmsrl_safe(msr->index, &msr->data);
1321 if (kvm_x86_ops->get_msr_feature(msr))
1327 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1329 struct kvm_msr_entry msr;
1333 r = kvm_get_msr_feature(&msr);
1342 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1344 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1347 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1350 if (efer & (EFER_LME | EFER_LMA) &&
1351 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1354 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1360 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1362 if (efer & efer_reserved_bits)
1365 return __kvm_valid_efer(vcpu, efer);
1367 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1369 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1371 u64 old_efer = vcpu->arch.efer;
1372 u64 efer = msr_info->data;
1374 if (efer & efer_reserved_bits)
1377 if (!msr_info->host_initiated) {
1378 if (!__kvm_valid_efer(vcpu, efer))
1381 if (is_paging(vcpu) &&
1382 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1387 efer |= vcpu->arch.efer & EFER_LMA;
1389 kvm_x86_ops->set_efer(vcpu, efer);
1391 /* Update reserved bits */
1392 if ((efer ^ old_efer) & EFER_NX)
1393 kvm_mmu_reset_context(vcpu);
1398 void kvm_enable_efer_bits(u64 mask)
1400 efer_reserved_bits &= ~mask;
1402 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1405 * Write @data into the MSR specified by @index. Select MSR specific fault
1406 * checks are bypassed if @host_initiated is %true.
1407 * Returns 0 on success, non-0 otherwise.
1408 * Assumes vcpu_load() was already called.
1410 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1411 bool host_initiated)
1413 struct msr_data msr;
1418 case MSR_KERNEL_GS_BASE:
1421 if (is_noncanonical_address(data, vcpu))
1424 case MSR_IA32_SYSENTER_EIP:
1425 case MSR_IA32_SYSENTER_ESP:
1427 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1428 * non-canonical address is written on Intel but not on
1429 * AMD (which ignores the top 32-bits, because it does
1430 * not implement 64-bit SYSENTER).
1432 * 64-bit code should hence be able to write a non-canonical
1433 * value on AMD. Making the address canonical ensures that
1434 * vmentry does not fail on Intel after writing a non-canonical
1435 * value, and that something deterministic happens if the guest
1436 * invokes 64-bit SYSENTER.
1438 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1443 msr.host_initiated = host_initiated;
1445 return kvm_x86_ops->set_msr(vcpu, &msr);
1449 * Read the MSR specified by @index into @data. Select MSR specific fault
1450 * checks are bypassed if @host_initiated is %true.
1451 * Returns 0 on success, non-0 otherwise.
1452 * Assumes vcpu_load() was already called.
1454 static int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1455 bool host_initiated)
1457 struct msr_data msr;
1461 msr.host_initiated = host_initiated;
1463 ret = kvm_x86_ops->get_msr(vcpu, &msr);
1469 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1471 return __kvm_get_msr(vcpu, index, data, false);
1473 EXPORT_SYMBOL_GPL(kvm_get_msr);
1475 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1477 return __kvm_set_msr(vcpu, index, data, false);
1479 EXPORT_SYMBOL_GPL(kvm_set_msr);
1481 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1483 u32 ecx = kvm_rcx_read(vcpu);
1486 if (kvm_get_msr(vcpu, ecx, &data)) {
1487 trace_kvm_msr_read_ex(ecx);
1488 kvm_inject_gp(vcpu, 0);
1492 trace_kvm_msr_read(ecx, data);
1494 kvm_rax_write(vcpu, data & -1u);
1495 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1496 return kvm_skip_emulated_instruction(vcpu);
1498 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1500 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1502 u32 ecx = kvm_rcx_read(vcpu);
1503 u64 data = kvm_read_edx_eax(vcpu);
1505 if (kvm_set_msr(vcpu, ecx, data)) {
1506 trace_kvm_msr_write_ex(ecx, data);
1507 kvm_inject_gp(vcpu, 0);
1511 trace_kvm_msr_write(ecx, data);
1512 return kvm_skip_emulated_instruction(vcpu);
1514 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1517 * Adapt set_msr() to msr_io()'s calling convention
1519 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1521 return __kvm_get_msr(vcpu, index, data, true);
1524 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1526 return __kvm_set_msr(vcpu, index, *data, true);
1529 #ifdef CONFIG_X86_64
1530 struct pvclock_gtod_data {
1533 struct { /* extract of a clocksource struct */
1546 static struct pvclock_gtod_data pvclock_gtod_data;
1548 static void update_pvclock_gtod(struct timekeeper *tk)
1550 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1553 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1555 write_seqcount_begin(&vdata->seq);
1557 /* copy pvclock gtod data */
1558 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1559 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1560 vdata->clock.mask = tk->tkr_mono.mask;
1561 vdata->clock.mult = tk->tkr_mono.mult;
1562 vdata->clock.shift = tk->tkr_mono.shift;
1564 vdata->boot_ns = boot_ns;
1565 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1567 vdata->wall_time_sec = tk->xtime_sec;
1569 write_seqcount_end(&vdata->seq);
1573 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1575 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1576 kvm_vcpu_kick(vcpu);
1579 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1583 struct pvclock_wall_clock wc;
1584 struct timespec64 boot;
1589 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1594 ++version; /* first time write, random junk */
1598 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1602 * The guest calculates current wall clock time by adding
1603 * system time (updated by kvm_guest_time_update below) to the
1604 * wall clock specified here. guest system time equals host
1605 * system time for us, thus we must fill in host boot time here.
1607 getboottime64(&boot);
1609 if (kvm->arch.kvmclock_offset) {
1610 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1611 boot = timespec64_sub(boot, ts);
1613 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1614 wc.nsec = boot.tv_nsec;
1615 wc.version = version;
1617 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1620 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1623 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1625 do_shl32_div32(dividend, divisor);
1629 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1630 s8 *pshift, u32 *pmultiplier)
1638 scaled64 = scaled_hz;
1639 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1644 tps32 = (uint32_t)tps64;
1645 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1646 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1654 *pmultiplier = div_frac(scaled64, tps32);
1657 #ifdef CONFIG_X86_64
1658 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1661 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1662 static unsigned long max_tsc_khz;
1664 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1666 u64 v = (u64)khz * (1000000 + ppm);
1671 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1675 /* Guest TSC same frequency as host TSC? */
1677 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1681 /* TSC scaling supported? */
1682 if (!kvm_has_tsc_control) {
1683 if (user_tsc_khz > tsc_khz) {
1684 vcpu->arch.tsc_catchup = 1;
1685 vcpu->arch.tsc_always_catchup = 1;
1688 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1693 /* TSC scaling required - calculate ratio */
1694 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1695 user_tsc_khz, tsc_khz);
1697 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1698 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1703 vcpu->arch.tsc_scaling_ratio = ratio;
1707 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1709 u32 thresh_lo, thresh_hi;
1710 int use_scaling = 0;
1712 /* tsc_khz can be zero if TSC calibration fails */
1713 if (user_tsc_khz == 0) {
1714 /* set tsc_scaling_ratio to a safe value */
1715 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1719 /* Compute a scale to convert nanoseconds in TSC cycles */
1720 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1721 &vcpu->arch.virtual_tsc_shift,
1722 &vcpu->arch.virtual_tsc_mult);
1723 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1726 * Compute the variation in TSC rate which is acceptable
1727 * within the range of tolerance and decide if the
1728 * rate being applied is within that bounds of the hardware
1729 * rate. If so, no scaling or compensation need be done.
1731 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1732 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1733 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1734 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1737 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1740 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1742 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1743 vcpu->arch.virtual_tsc_mult,
1744 vcpu->arch.virtual_tsc_shift);
1745 tsc += vcpu->arch.this_tsc_write;
1749 static inline int gtod_is_based_on_tsc(int mode)
1751 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1754 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1756 #ifdef CONFIG_X86_64
1758 struct kvm_arch *ka = &vcpu->kvm->arch;
1759 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1761 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1762 atomic_read(&vcpu->kvm->online_vcpus));
1765 * Once the masterclock is enabled, always perform request in
1766 * order to update it.
1768 * In order to enable masterclock, the host clocksource must be TSC
1769 * and the vcpus need to have matched TSCs. When that happens,
1770 * perform request to enable masterclock.
1772 if (ka->use_master_clock ||
1773 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1774 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1776 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1777 atomic_read(&vcpu->kvm->online_vcpus),
1778 ka->use_master_clock, gtod->clock.vclock_mode);
1782 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1784 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1785 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1789 * Multiply tsc by a fixed point number represented by ratio.
1791 * The most significant 64-N bits (mult) of ratio represent the
1792 * integral part of the fixed point number; the remaining N bits
1793 * (frac) represent the fractional part, ie. ratio represents a fixed
1794 * point number (mult + frac * 2^(-N)).
1796 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1798 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1800 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1803 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1806 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1808 if (ratio != kvm_default_tsc_scaling_ratio)
1809 _tsc = __scale_tsc(ratio, tsc);
1813 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1815 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1819 tsc = kvm_scale_tsc(vcpu, rdtsc());
1821 return target_tsc - tsc;
1824 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1826 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1828 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1830 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1832 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1834 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1837 static inline bool kvm_check_tsc_unstable(void)
1839 #ifdef CONFIG_X86_64
1841 * TSC is marked unstable when we're running on Hyper-V,
1842 * 'TSC page' clocksource is good.
1844 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1847 return check_tsc_unstable();
1850 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1852 struct kvm *kvm = vcpu->kvm;
1853 u64 offset, ns, elapsed;
1854 unsigned long flags;
1856 bool already_matched;
1857 u64 data = msr->data;
1858 bool synchronizing = false;
1860 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1861 offset = kvm_compute_tsc_offset(vcpu, data);
1862 ns = ktime_get_boottime_ns();
1863 elapsed = ns - kvm->arch.last_tsc_nsec;
1865 if (vcpu->arch.virtual_tsc_khz) {
1866 if (data == 0 && msr->host_initiated) {
1868 * detection of vcpu initialization -- need to sync
1869 * with other vCPUs. This particularly helps to keep
1870 * kvm_clock stable after CPU hotplug
1872 synchronizing = true;
1874 u64 tsc_exp = kvm->arch.last_tsc_write +
1875 nsec_to_cycles(vcpu, elapsed);
1876 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1878 * Special case: TSC write with a small delta (1 second)
1879 * of virtual cycle time against real time is
1880 * interpreted as an attempt to synchronize the CPU.
1882 synchronizing = data < tsc_exp + tsc_hz &&
1883 data + tsc_hz > tsc_exp;
1888 * For a reliable TSC, we can match TSC offsets, and for an unstable
1889 * TSC, we add elapsed time in this computation. We could let the
1890 * compensation code attempt to catch up if we fall behind, but
1891 * it's better to try to match offsets from the beginning.
1893 if (synchronizing &&
1894 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1895 if (!kvm_check_tsc_unstable()) {
1896 offset = kvm->arch.cur_tsc_offset;
1898 u64 delta = nsec_to_cycles(vcpu, elapsed);
1900 offset = kvm_compute_tsc_offset(vcpu, data);
1903 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1906 * We split periods of matched TSC writes into generations.
1907 * For each generation, we track the original measured
1908 * nanosecond time, offset, and write, so if TSCs are in
1909 * sync, we can match exact offset, and if not, we can match
1910 * exact software computation in compute_guest_tsc()
1912 * These values are tracked in kvm->arch.cur_xxx variables.
1914 kvm->arch.cur_tsc_generation++;
1915 kvm->arch.cur_tsc_nsec = ns;
1916 kvm->arch.cur_tsc_write = data;
1917 kvm->arch.cur_tsc_offset = offset;
1922 * We also track th most recent recorded KHZ, write and time to
1923 * allow the matching interval to be extended at each write.
1925 kvm->arch.last_tsc_nsec = ns;
1926 kvm->arch.last_tsc_write = data;
1927 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1929 vcpu->arch.last_guest_tsc = data;
1931 /* Keep track of which generation this VCPU has synchronized to */
1932 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1933 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1934 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1936 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1937 update_ia32_tsc_adjust_msr(vcpu, offset);
1939 kvm_vcpu_write_tsc_offset(vcpu, offset);
1940 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1942 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1944 kvm->arch.nr_vcpus_matched_tsc = 0;
1945 } else if (!already_matched) {
1946 kvm->arch.nr_vcpus_matched_tsc++;
1949 kvm_track_tsc_matching(vcpu);
1950 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1953 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1955 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1958 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1959 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1962 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1964 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1965 WARN_ON(adjustment < 0);
1966 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1967 adjust_tsc_offset_guest(vcpu, adjustment);
1970 #ifdef CONFIG_X86_64
1972 static u64 read_tsc(void)
1974 u64 ret = (u64)rdtsc_ordered();
1975 u64 last = pvclock_gtod_data.clock.cycle_last;
1977 if (likely(ret >= last))
1981 * GCC likes to generate cmov here, but this branch is extremely
1982 * predictable (it's just a function of time and the likely is
1983 * very likely) and there's a data dependence, so force GCC
1984 * to generate a branch instead. I don't barrier() because
1985 * we don't actually need a barrier, and if this function
1986 * ever gets inlined it will generate worse code.
1992 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1995 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1998 switch (gtod->clock.vclock_mode) {
1999 case VCLOCK_HVCLOCK:
2000 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2002 if (tsc_pg_val != U64_MAX) {
2003 /* TSC page valid */
2004 *mode = VCLOCK_HVCLOCK;
2005 v = (tsc_pg_val - gtod->clock.cycle_last) &
2008 /* TSC page invalid */
2009 *mode = VCLOCK_NONE;
2014 *tsc_timestamp = read_tsc();
2015 v = (*tsc_timestamp - gtod->clock.cycle_last) &
2019 *mode = VCLOCK_NONE;
2022 if (*mode == VCLOCK_NONE)
2023 *tsc_timestamp = v = 0;
2025 return v * gtod->clock.mult;
2028 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
2030 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2036 seq = read_seqcount_begin(>od->seq);
2037 ns = gtod->nsec_base;
2038 ns += vgettsc(tsc_timestamp, &mode);
2039 ns >>= gtod->clock.shift;
2040 ns += gtod->boot_ns;
2041 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2047 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2049 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2055 seq = read_seqcount_begin(>od->seq);
2056 ts->tv_sec = gtod->wall_time_sec;
2057 ns = gtod->nsec_base;
2058 ns += vgettsc(tsc_timestamp, &mode);
2059 ns >>= gtod->clock.shift;
2060 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2062 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2068 /* returns true if host is using TSC based clocksource */
2069 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2071 /* checked again under seqlock below */
2072 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2075 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
2079 /* returns true if host is using TSC based clocksource */
2080 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2083 /* checked again under seqlock below */
2084 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2087 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2093 * Assuming a stable TSC across physical CPUS, and a stable TSC
2094 * across virtual CPUs, the following condition is possible.
2095 * Each numbered line represents an event visible to both
2096 * CPUs at the next numbered event.
2098 * "timespecX" represents host monotonic time. "tscX" represents
2101 * VCPU0 on CPU0 | VCPU1 on CPU1
2103 * 1. read timespec0,tsc0
2104 * 2. | timespec1 = timespec0 + N
2106 * 3. transition to guest | transition to guest
2107 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2108 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2109 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2111 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2114 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2116 * - 0 < N - M => M < N
2118 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2119 * always the case (the difference between two distinct xtime instances
2120 * might be smaller then the difference between corresponding TSC reads,
2121 * when updating guest vcpus pvclock areas).
2123 * To avoid that problem, do not allow visibility of distinct
2124 * system_timestamp/tsc_timestamp values simultaneously: use a master
2125 * copy of host monotonic time values. Update that master copy
2128 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2132 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2134 #ifdef CONFIG_X86_64
2135 struct kvm_arch *ka = &kvm->arch;
2137 bool host_tsc_clocksource, vcpus_matched;
2139 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2140 atomic_read(&kvm->online_vcpus));
2143 * If the host uses TSC clock, then passthrough TSC as stable
2146 host_tsc_clocksource = kvm_get_time_and_clockread(
2147 &ka->master_kernel_ns,
2148 &ka->master_cycle_now);
2150 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2151 && !ka->backwards_tsc_observed
2152 && !ka->boot_vcpu_runs_old_kvmclock;
2154 if (ka->use_master_clock)
2155 atomic_set(&kvm_guest_has_master_clock, 1);
2157 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2158 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2163 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2165 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2168 static void kvm_gen_update_masterclock(struct kvm *kvm)
2170 #ifdef CONFIG_X86_64
2172 struct kvm_vcpu *vcpu;
2173 struct kvm_arch *ka = &kvm->arch;
2175 spin_lock(&ka->pvclock_gtod_sync_lock);
2176 kvm_make_mclock_inprogress_request(kvm);
2177 /* no guest entries from this point */
2178 pvclock_update_vm_gtod_copy(kvm);
2180 kvm_for_each_vcpu(i, vcpu, kvm)
2181 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2183 /* guest entries allowed */
2184 kvm_for_each_vcpu(i, vcpu, kvm)
2185 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2187 spin_unlock(&ka->pvclock_gtod_sync_lock);
2191 u64 get_kvmclock_ns(struct kvm *kvm)
2193 struct kvm_arch *ka = &kvm->arch;
2194 struct pvclock_vcpu_time_info hv_clock;
2197 spin_lock(&ka->pvclock_gtod_sync_lock);
2198 if (!ka->use_master_clock) {
2199 spin_unlock(&ka->pvclock_gtod_sync_lock);
2200 return ktime_get_boottime_ns() + ka->kvmclock_offset;
2203 hv_clock.tsc_timestamp = ka->master_cycle_now;
2204 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2205 spin_unlock(&ka->pvclock_gtod_sync_lock);
2207 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2210 if (__this_cpu_read(cpu_tsc_khz)) {
2211 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2212 &hv_clock.tsc_shift,
2213 &hv_clock.tsc_to_system_mul);
2214 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2216 ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2223 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2225 struct kvm_vcpu_arch *vcpu = &v->arch;
2226 struct pvclock_vcpu_time_info guest_hv_clock;
2228 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2229 &guest_hv_clock, sizeof(guest_hv_clock))))
2232 /* This VCPU is paused, but it's legal for a guest to read another
2233 * VCPU's kvmclock, so we really have to follow the specification where
2234 * it says that version is odd if data is being modified, and even after
2237 * Version field updates must be kept separate. This is because
2238 * kvm_write_guest_cached might use a "rep movs" instruction, and
2239 * writes within a string instruction are weakly ordered. So there
2240 * are three writes overall.
2242 * As a small optimization, only write the version field in the first
2243 * and third write. The vcpu->pv_time cache is still valid, because the
2244 * version field is the first in the struct.
2246 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2248 if (guest_hv_clock.version & 1)
2249 ++guest_hv_clock.version; /* first time write, random junk */
2251 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2252 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2254 sizeof(vcpu->hv_clock.version));
2258 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2259 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2261 if (vcpu->pvclock_set_guest_stopped_request) {
2262 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2263 vcpu->pvclock_set_guest_stopped_request = false;
2266 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2268 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2270 sizeof(vcpu->hv_clock));
2274 vcpu->hv_clock.version++;
2275 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2277 sizeof(vcpu->hv_clock.version));
2280 static int kvm_guest_time_update(struct kvm_vcpu *v)
2282 unsigned long flags, tgt_tsc_khz;
2283 struct kvm_vcpu_arch *vcpu = &v->arch;
2284 struct kvm_arch *ka = &v->kvm->arch;
2286 u64 tsc_timestamp, host_tsc;
2288 bool use_master_clock;
2294 * If the host uses TSC clock, then passthrough TSC as stable
2297 spin_lock(&ka->pvclock_gtod_sync_lock);
2298 use_master_clock = ka->use_master_clock;
2299 if (use_master_clock) {
2300 host_tsc = ka->master_cycle_now;
2301 kernel_ns = ka->master_kernel_ns;
2303 spin_unlock(&ka->pvclock_gtod_sync_lock);
2305 /* Keep irq disabled to prevent changes to the clock */
2306 local_irq_save(flags);
2307 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2308 if (unlikely(tgt_tsc_khz == 0)) {
2309 local_irq_restore(flags);
2310 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2313 if (!use_master_clock) {
2315 kernel_ns = ktime_get_boottime_ns();
2318 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2321 * We may have to catch up the TSC to match elapsed wall clock
2322 * time for two reasons, even if kvmclock is used.
2323 * 1) CPU could have been running below the maximum TSC rate
2324 * 2) Broken TSC compensation resets the base at each VCPU
2325 * entry to avoid unknown leaps of TSC even when running
2326 * again on the same CPU. This may cause apparent elapsed
2327 * time to disappear, and the guest to stand still or run
2330 if (vcpu->tsc_catchup) {
2331 u64 tsc = compute_guest_tsc(v, kernel_ns);
2332 if (tsc > tsc_timestamp) {
2333 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2334 tsc_timestamp = tsc;
2338 local_irq_restore(flags);
2340 /* With all the info we got, fill in the values */
2342 if (kvm_has_tsc_control)
2343 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2345 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2346 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2347 &vcpu->hv_clock.tsc_shift,
2348 &vcpu->hv_clock.tsc_to_system_mul);
2349 vcpu->hw_tsc_khz = tgt_tsc_khz;
2352 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2353 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2354 vcpu->last_guest_tsc = tsc_timestamp;
2356 /* If the host uses TSC clocksource, then it is stable */
2358 if (use_master_clock)
2359 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2361 vcpu->hv_clock.flags = pvclock_flags;
2363 if (vcpu->pv_time_enabled)
2364 kvm_setup_pvclock_page(v);
2365 if (v == kvm_get_vcpu(v->kvm, 0))
2366 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2371 * kvmclock updates which are isolated to a given vcpu, such as
2372 * vcpu->cpu migration, should not allow system_timestamp from
2373 * the rest of the vcpus to remain static. Otherwise ntp frequency
2374 * correction applies to one vcpu's system_timestamp but not
2377 * So in those cases, request a kvmclock update for all vcpus.
2378 * We need to rate-limit these requests though, as they can
2379 * considerably slow guests that have a large number of vcpus.
2380 * The time for a remote vcpu to update its kvmclock is bound
2381 * by the delay we use to rate-limit the updates.
2384 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2386 static void kvmclock_update_fn(struct work_struct *work)
2389 struct delayed_work *dwork = to_delayed_work(work);
2390 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2391 kvmclock_update_work);
2392 struct kvm *kvm = container_of(ka, struct kvm, arch);
2393 struct kvm_vcpu *vcpu;
2395 kvm_for_each_vcpu(i, vcpu, kvm) {
2396 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2397 kvm_vcpu_kick(vcpu);
2401 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2403 struct kvm *kvm = v->kvm;
2405 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2406 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2407 KVMCLOCK_UPDATE_DELAY);
2410 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2412 static void kvmclock_sync_fn(struct work_struct *work)
2414 struct delayed_work *dwork = to_delayed_work(work);
2415 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2416 kvmclock_sync_work);
2417 struct kvm *kvm = container_of(ka, struct kvm, arch);
2419 if (!kvmclock_periodic_sync)
2422 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2423 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2424 KVMCLOCK_SYNC_PERIOD);
2428 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2430 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2432 /* McStatusWrEn enabled? */
2433 if (guest_cpuid_is_amd(vcpu))
2434 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2439 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2441 u64 mcg_cap = vcpu->arch.mcg_cap;
2442 unsigned bank_num = mcg_cap & 0xff;
2443 u32 msr = msr_info->index;
2444 u64 data = msr_info->data;
2447 case MSR_IA32_MCG_STATUS:
2448 vcpu->arch.mcg_status = data;
2450 case MSR_IA32_MCG_CTL:
2451 if (!(mcg_cap & MCG_CTL_P) &&
2452 (data || !msr_info->host_initiated))
2454 if (data != 0 && data != ~(u64)0)
2456 vcpu->arch.mcg_ctl = data;
2459 if (msr >= MSR_IA32_MC0_CTL &&
2460 msr < MSR_IA32_MCx_CTL(bank_num)) {
2461 u32 offset = msr - MSR_IA32_MC0_CTL;
2462 /* only 0 or all 1s can be written to IA32_MCi_CTL
2463 * some Linux kernels though clear bit 10 in bank 4 to
2464 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2465 * this to avoid an uncatched #GP in the guest
2467 if ((offset & 0x3) == 0 &&
2468 data != 0 && (data | (1 << 10)) != ~(u64)0)
2472 if (!msr_info->host_initiated &&
2473 (offset & 0x3) == 1 && data != 0) {
2474 if (!can_set_mci_status(vcpu))
2478 vcpu->arch.mce_banks[offset] = data;
2486 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2488 struct kvm *kvm = vcpu->kvm;
2489 int lm = is_long_mode(vcpu);
2490 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2491 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2492 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2493 : kvm->arch.xen_hvm_config.blob_size_32;
2494 u32 page_num = data & ~PAGE_MASK;
2495 u64 page_addr = data & PAGE_MASK;
2500 if (page_num >= blob_size)
2503 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2508 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2517 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2519 gpa_t gpa = data & ~0x3f;
2521 /* Bits 3:5 are reserved, Should be zero */
2525 vcpu->arch.apf.msr_val = data;
2527 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2528 kvm_clear_async_pf_completion_queue(vcpu);
2529 kvm_async_pf_hash_reset(vcpu);
2533 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2537 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2538 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2539 kvm_async_pf_wakeup_all(vcpu);
2543 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2545 vcpu->arch.pv_time_enabled = false;
2548 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2550 ++vcpu->stat.tlb_flush;
2551 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2554 static void record_steal_time(struct kvm_vcpu *vcpu)
2556 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2559 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2560 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2564 * Doing a TLB flush here, on the guest's behalf, can avoid
2567 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2568 vcpu->arch.st.steal.preempted & KVM_VCPU_FLUSH_TLB);
2569 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2570 kvm_vcpu_flush_tlb(vcpu, false);
2572 if (vcpu->arch.st.steal.version & 1)
2573 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2575 vcpu->arch.st.steal.version += 1;
2577 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2578 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2582 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2583 vcpu->arch.st.last_steal;
2584 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2586 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2587 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2591 vcpu->arch.st.steal.version += 1;
2593 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2594 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2597 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2600 u32 msr = msr_info->index;
2601 u64 data = msr_info->data;
2604 case MSR_AMD64_NB_CFG:
2605 case MSR_IA32_UCODE_WRITE:
2606 case MSR_VM_HSAVE_PA:
2607 case MSR_AMD64_PATCH_LOADER:
2608 case MSR_AMD64_BU_CFG2:
2609 case MSR_AMD64_DC_CFG:
2610 case MSR_F15H_EX_CFG:
2613 case MSR_IA32_UCODE_REV:
2614 if (msr_info->host_initiated)
2615 vcpu->arch.microcode_version = data;
2617 case MSR_IA32_ARCH_CAPABILITIES:
2618 if (!msr_info->host_initiated)
2620 vcpu->arch.arch_capabilities = data;
2623 return set_efer(vcpu, msr_info);
2625 data &= ~(u64)0x40; /* ignore flush filter disable */
2626 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2627 data &= ~(u64)0x8; /* ignore TLB cache disable */
2629 /* Handle McStatusWrEn */
2630 if (data == BIT_ULL(18)) {
2631 vcpu->arch.msr_hwcr = data;
2632 } else if (data != 0) {
2633 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2638 case MSR_FAM10H_MMIO_CONF_BASE:
2640 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2645 case MSR_IA32_DEBUGCTLMSR:
2647 /* We support the non-activated case already */
2649 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2650 /* Values other than LBR and BTF are vendor-specific,
2651 thus reserved and should throw a #GP */
2654 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2657 case 0x200 ... 0x2ff:
2658 return kvm_mtrr_set_msr(vcpu, msr, data);
2659 case MSR_IA32_APICBASE:
2660 return kvm_set_apic_base(vcpu, msr_info);
2661 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2662 return kvm_x2apic_msr_write(vcpu, msr, data);
2663 case MSR_IA32_TSCDEADLINE:
2664 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2666 case MSR_IA32_TSC_ADJUST:
2667 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2668 if (!msr_info->host_initiated) {
2669 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2670 adjust_tsc_offset_guest(vcpu, adj);
2672 vcpu->arch.ia32_tsc_adjust_msr = data;
2675 case MSR_IA32_MISC_ENABLE:
2676 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2677 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2678 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2680 vcpu->arch.ia32_misc_enable_msr = data;
2681 kvm_update_cpuid(vcpu);
2683 vcpu->arch.ia32_misc_enable_msr = data;
2686 case MSR_IA32_SMBASE:
2687 if (!msr_info->host_initiated)
2689 vcpu->arch.smbase = data;
2691 case MSR_IA32_POWER_CTL:
2692 vcpu->arch.msr_ia32_power_ctl = data;
2695 kvm_write_tsc(vcpu, msr_info);
2698 if (!msr_info->host_initiated)
2700 vcpu->arch.smi_count = data;
2702 case MSR_KVM_WALL_CLOCK_NEW:
2703 case MSR_KVM_WALL_CLOCK:
2704 vcpu->kvm->arch.wall_clock = data;
2705 kvm_write_wall_clock(vcpu->kvm, data);
2707 case MSR_KVM_SYSTEM_TIME_NEW:
2708 case MSR_KVM_SYSTEM_TIME: {
2709 struct kvm_arch *ka = &vcpu->kvm->arch;
2711 kvmclock_reset(vcpu);
2713 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2714 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2716 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2717 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2719 ka->boot_vcpu_runs_old_kvmclock = tmp;
2722 vcpu->arch.time = data;
2723 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2725 /* we verify if the enable bit is set... */
2729 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2730 &vcpu->arch.pv_time, data & ~1ULL,
2731 sizeof(struct pvclock_vcpu_time_info)))
2732 vcpu->arch.pv_time_enabled = false;
2734 vcpu->arch.pv_time_enabled = true;
2738 case MSR_KVM_ASYNC_PF_EN:
2739 if (kvm_pv_enable_async_pf(vcpu, data))
2742 case MSR_KVM_STEAL_TIME:
2744 if (unlikely(!sched_info_on()))
2747 if (data & KVM_STEAL_RESERVED_MASK)
2750 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2751 data & KVM_STEAL_VALID_BITS,
2752 sizeof(struct kvm_steal_time)))
2755 vcpu->arch.st.msr_val = data;
2757 if (!(data & KVM_MSR_ENABLED))
2760 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2763 case MSR_KVM_PV_EOI_EN:
2764 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2768 case MSR_KVM_POLL_CONTROL:
2769 /* only enable bit supported */
2770 if (data & (-1ULL << 1))
2773 vcpu->arch.msr_kvm_poll_control = data;
2776 case MSR_IA32_MCG_CTL:
2777 case MSR_IA32_MCG_STATUS:
2778 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2779 return set_msr_mce(vcpu, msr_info);
2781 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2782 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2783 pr = true; /* fall through */
2784 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2785 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2786 if (kvm_pmu_is_valid_msr(vcpu, msr))
2787 return kvm_pmu_set_msr(vcpu, msr_info);
2789 if (pr || data != 0)
2790 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2791 "0x%x data 0x%llx\n", msr, data);
2793 case MSR_K7_CLK_CTL:
2795 * Ignore all writes to this no longer documented MSR.
2796 * Writes are only relevant for old K7 processors,
2797 * all pre-dating SVM, but a recommended workaround from
2798 * AMD for these chips. It is possible to specify the
2799 * affected processor models on the command line, hence
2800 * the need to ignore the workaround.
2803 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2804 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2805 case HV_X64_MSR_CRASH_CTL:
2806 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2807 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2808 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2809 case HV_X64_MSR_TSC_EMULATION_STATUS:
2810 return kvm_hv_set_msr_common(vcpu, msr, data,
2811 msr_info->host_initiated);
2812 case MSR_IA32_BBL_CR_CTL3:
2813 /* Drop writes to this legacy MSR -- see rdmsr
2814 * counterpart for further detail.
2816 if (report_ignored_msrs)
2817 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2820 case MSR_AMD64_OSVW_ID_LENGTH:
2821 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2823 vcpu->arch.osvw.length = data;
2825 case MSR_AMD64_OSVW_STATUS:
2826 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2828 vcpu->arch.osvw.status = data;
2830 case MSR_PLATFORM_INFO:
2831 if (!msr_info->host_initiated ||
2832 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2833 cpuid_fault_enabled(vcpu)))
2835 vcpu->arch.msr_platform_info = data;
2837 case MSR_MISC_FEATURES_ENABLES:
2838 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2839 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2840 !supports_cpuid_fault(vcpu)))
2842 vcpu->arch.msr_misc_features_enables = data;
2845 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2846 return xen_hvm_config(vcpu, data);
2847 if (kvm_pmu_is_valid_msr(vcpu, msr))
2848 return kvm_pmu_set_msr(vcpu, msr_info);
2850 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2854 if (report_ignored_msrs)
2856 "ignored wrmsr: 0x%x data 0x%llx\n",
2863 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2865 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2868 u64 mcg_cap = vcpu->arch.mcg_cap;
2869 unsigned bank_num = mcg_cap & 0xff;
2872 case MSR_IA32_P5_MC_ADDR:
2873 case MSR_IA32_P5_MC_TYPE:
2876 case MSR_IA32_MCG_CAP:
2877 data = vcpu->arch.mcg_cap;
2879 case MSR_IA32_MCG_CTL:
2880 if (!(mcg_cap & MCG_CTL_P) && !host)
2882 data = vcpu->arch.mcg_ctl;
2884 case MSR_IA32_MCG_STATUS:
2885 data = vcpu->arch.mcg_status;
2888 if (msr >= MSR_IA32_MC0_CTL &&
2889 msr < MSR_IA32_MCx_CTL(bank_num)) {
2890 u32 offset = msr - MSR_IA32_MC0_CTL;
2891 data = vcpu->arch.mce_banks[offset];
2900 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2902 switch (msr_info->index) {
2903 case MSR_IA32_PLATFORM_ID:
2904 case MSR_IA32_EBL_CR_POWERON:
2905 case MSR_IA32_DEBUGCTLMSR:
2906 case MSR_IA32_LASTBRANCHFROMIP:
2907 case MSR_IA32_LASTBRANCHTOIP:
2908 case MSR_IA32_LASTINTFROMIP:
2909 case MSR_IA32_LASTINTTOIP:
2911 case MSR_K8_TSEG_ADDR:
2912 case MSR_K8_TSEG_MASK:
2913 case MSR_VM_HSAVE_PA:
2914 case MSR_K8_INT_PENDING_MSG:
2915 case MSR_AMD64_NB_CFG:
2916 case MSR_FAM10H_MMIO_CONF_BASE:
2917 case MSR_AMD64_BU_CFG2:
2918 case MSR_IA32_PERF_CTL:
2919 case MSR_AMD64_DC_CFG:
2920 case MSR_F15H_EX_CFG:
2923 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2924 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2925 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2926 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2927 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2928 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2929 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2932 case MSR_IA32_UCODE_REV:
2933 msr_info->data = vcpu->arch.microcode_version;
2935 case MSR_IA32_ARCH_CAPABILITIES:
2936 if (!msr_info->host_initiated &&
2937 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2939 msr_info->data = vcpu->arch.arch_capabilities;
2941 case MSR_IA32_POWER_CTL:
2942 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2945 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2948 case 0x200 ... 0x2ff:
2949 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2950 case 0xcd: /* fsb frequency */
2954 * MSR_EBC_FREQUENCY_ID
2955 * Conservative value valid for even the basic CPU models.
2956 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2957 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2958 * and 266MHz for model 3, or 4. Set Core Clock
2959 * Frequency to System Bus Frequency Ratio to 1 (bits
2960 * 31:24) even though these are only valid for CPU
2961 * models > 2, however guests may end up dividing or
2962 * multiplying by zero otherwise.
2964 case MSR_EBC_FREQUENCY_ID:
2965 msr_info->data = 1 << 24;
2967 case MSR_IA32_APICBASE:
2968 msr_info->data = kvm_get_apic_base(vcpu);
2970 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2971 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2973 case MSR_IA32_TSCDEADLINE:
2974 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2976 case MSR_IA32_TSC_ADJUST:
2977 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2979 case MSR_IA32_MISC_ENABLE:
2980 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2982 case MSR_IA32_SMBASE:
2983 if (!msr_info->host_initiated)
2985 msr_info->data = vcpu->arch.smbase;
2988 msr_info->data = vcpu->arch.smi_count;
2990 case MSR_IA32_PERF_STATUS:
2991 /* TSC increment by tick */
2992 msr_info->data = 1000ULL;
2993 /* CPU multiplier */
2994 msr_info->data |= (((uint64_t)4ULL) << 40);
2997 msr_info->data = vcpu->arch.efer;
2999 case MSR_KVM_WALL_CLOCK:
3000 case MSR_KVM_WALL_CLOCK_NEW:
3001 msr_info->data = vcpu->kvm->arch.wall_clock;
3003 case MSR_KVM_SYSTEM_TIME:
3004 case MSR_KVM_SYSTEM_TIME_NEW:
3005 msr_info->data = vcpu->arch.time;
3007 case MSR_KVM_ASYNC_PF_EN:
3008 msr_info->data = vcpu->arch.apf.msr_val;
3010 case MSR_KVM_STEAL_TIME:
3011 msr_info->data = vcpu->arch.st.msr_val;
3013 case MSR_KVM_PV_EOI_EN:
3014 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3016 case MSR_KVM_POLL_CONTROL:
3017 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3019 case MSR_IA32_P5_MC_ADDR:
3020 case MSR_IA32_P5_MC_TYPE:
3021 case MSR_IA32_MCG_CAP:
3022 case MSR_IA32_MCG_CTL:
3023 case MSR_IA32_MCG_STATUS:
3024 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3025 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3026 msr_info->host_initiated);
3027 case MSR_K7_CLK_CTL:
3029 * Provide expected ramp-up count for K7. All other
3030 * are set to zero, indicating minimum divisors for
3033 * This prevents guest kernels on AMD host with CPU
3034 * type 6, model 8 and higher from exploding due to
3035 * the rdmsr failing.
3037 msr_info->data = 0x20000000;
3039 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3040 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3041 case HV_X64_MSR_CRASH_CTL:
3042 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3043 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3044 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3045 case HV_X64_MSR_TSC_EMULATION_STATUS:
3046 return kvm_hv_get_msr_common(vcpu,
3047 msr_info->index, &msr_info->data,
3048 msr_info->host_initiated);
3050 case MSR_IA32_BBL_CR_CTL3:
3051 /* This legacy MSR exists but isn't fully documented in current
3052 * silicon. It is however accessed by winxp in very narrow
3053 * scenarios where it sets bit #19, itself documented as
3054 * a "reserved" bit. Best effort attempt to source coherent
3055 * read data here should the balance of the register be
3056 * interpreted by the guest:
3058 * L2 cache control register 3: 64GB range, 256KB size,
3059 * enabled, latency 0x1, configured
3061 msr_info->data = 0xbe702111;
3063 case MSR_AMD64_OSVW_ID_LENGTH:
3064 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3066 msr_info->data = vcpu->arch.osvw.length;
3068 case MSR_AMD64_OSVW_STATUS:
3069 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3071 msr_info->data = vcpu->arch.osvw.status;
3073 case MSR_PLATFORM_INFO:
3074 if (!msr_info->host_initiated &&
3075 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3077 msr_info->data = vcpu->arch.msr_platform_info;
3079 case MSR_MISC_FEATURES_ENABLES:
3080 msr_info->data = vcpu->arch.msr_misc_features_enables;
3083 msr_info->data = vcpu->arch.msr_hwcr;
3086 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3087 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3089 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3093 if (report_ignored_msrs)
3094 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3102 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3105 * Read or write a bunch of msrs. All parameters are kernel addresses.
3107 * @return number of msrs set successfully.
3109 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3110 struct kvm_msr_entry *entries,
3111 int (*do_msr)(struct kvm_vcpu *vcpu,
3112 unsigned index, u64 *data))
3116 for (i = 0; i < msrs->nmsrs; ++i)
3117 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3124 * Read or write a bunch of msrs. Parameters are user addresses.
3126 * @return number of msrs set successfully.
3128 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3129 int (*do_msr)(struct kvm_vcpu *vcpu,
3130 unsigned index, u64 *data),
3133 struct kvm_msrs msrs;
3134 struct kvm_msr_entry *entries;
3139 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3143 if (msrs.nmsrs >= MAX_IO_MSRS)
3146 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3147 entries = memdup_user(user_msrs->entries, size);
3148 if (IS_ERR(entries)) {
3149 r = PTR_ERR(entries);
3153 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3158 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3169 static inline bool kvm_can_mwait_in_guest(void)
3171 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3172 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3173 boot_cpu_has(X86_FEATURE_ARAT);
3176 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3181 case KVM_CAP_IRQCHIP:
3183 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3184 case KVM_CAP_SET_TSS_ADDR:
3185 case KVM_CAP_EXT_CPUID:
3186 case KVM_CAP_EXT_EMUL_CPUID:
3187 case KVM_CAP_CLOCKSOURCE:
3189 case KVM_CAP_NOP_IO_DELAY:
3190 case KVM_CAP_MP_STATE:
3191 case KVM_CAP_SYNC_MMU:
3192 case KVM_CAP_USER_NMI:
3193 case KVM_CAP_REINJECT_CONTROL:
3194 case KVM_CAP_IRQ_INJECT_STATUS:
3195 case KVM_CAP_IOEVENTFD:
3196 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3198 case KVM_CAP_PIT_STATE2:
3199 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3200 case KVM_CAP_XEN_HVM:
3201 case KVM_CAP_VCPU_EVENTS:
3202 case KVM_CAP_HYPERV:
3203 case KVM_CAP_HYPERV_VAPIC:
3204 case KVM_CAP_HYPERV_SPIN:
3205 case KVM_CAP_HYPERV_SYNIC:
3206 case KVM_CAP_HYPERV_SYNIC2:
3207 case KVM_CAP_HYPERV_VP_INDEX:
3208 case KVM_CAP_HYPERV_EVENTFD:
3209 case KVM_CAP_HYPERV_TLBFLUSH:
3210 case KVM_CAP_HYPERV_SEND_IPI:
3211 case KVM_CAP_HYPERV_CPUID:
3212 case KVM_CAP_PCI_SEGMENT:
3213 case KVM_CAP_DEBUGREGS:
3214 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3216 case KVM_CAP_ASYNC_PF:
3217 case KVM_CAP_GET_TSC_KHZ:
3218 case KVM_CAP_KVMCLOCK_CTRL:
3219 case KVM_CAP_READONLY_MEM:
3220 case KVM_CAP_HYPERV_TIME:
3221 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3222 case KVM_CAP_TSC_DEADLINE_TIMER:
3223 case KVM_CAP_DISABLE_QUIRKS:
3224 case KVM_CAP_SET_BOOT_CPU_ID:
3225 case KVM_CAP_SPLIT_IRQCHIP:
3226 case KVM_CAP_IMMEDIATE_EXIT:
3227 case KVM_CAP_PMU_EVENT_FILTER:
3228 case KVM_CAP_GET_MSR_FEATURES:
3229 case KVM_CAP_MSR_PLATFORM_INFO:
3230 case KVM_CAP_EXCEPTION_PAYLOAD:
3233 case KVM_CAP_SYNC_REGS:
3234 r = KVM_SYNC_X86_VALID_FIELDS;
3236 case KVM_CAP_ADJUST_CLOCK:
3237 r = KVM_CLOCK_TSC_STABLE;
3239 case KVM_CAP_X86_DISABLE_EXITS:
3240 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3241 KVM_X86_DISABLE_EXITS_CSTATE;
3242 if(kvm_can_mwait_in_guest())
3243 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3245 case KVM_CAP_X86_SMM:
3246 /* SMBASE is usually relocated above 1M on modern chipsets,
3247 * and SMM handlers might indeed rely on 4G segment limits,
3248 * so do not report SMM to be available if real mode is
3249 * emulated via vm86 mode. Still, do not go to great lengths
3250 * to avoid userspace's usage of the feature, because it is a
3251 * fringe case that is not enabled except via specific settings
3252 * of the module parameters.
3254 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3257 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3259 case KVM_CAP_NR_VCPUS:
3260 r = KVM_SOFT_MAX_VCPUS;
3262 case KVM_CAP_MAX_VCPUS:
3265 case KVM_CAP_MAX_VCPU_ID:
3266 r = KVM_MAX_VCPU_ID;
3268 case KVM_CAP_PV_MMU: /* obsolete */
3272 r = KVM_MAX_MCE_BANKS;
3275 r = boot_cpu_has(X86_FEATURE_XSAVE);
3277 case KVM_CAP_TSC_CONTROL:
3278 r = kvm_has_tsc_control;
3280 case KVM_CAP_X2APIC_API:
3281 r = KVM_X2APIC_API_VALID_FLAGS;
3283 case KVM_CAP_NESTED_STATE:
3284 r = kvm_x86_ops->get_nested_state ?
3285 kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3287 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3288 r = kvm_x86_ops->enable_direct_tlbflush != NULL;
3290 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3291 r = kvm_x86_ops->nested_enable_evmcs != NULL;
3300 long kvm_arch_dev_ioctl(struct file *filp,
3301 unsigned int ioctl, unsigned long arg)
3303 void __user *argp = (void __user *)arg;
3307 case KVM_GET_MSR_INDEX_LIST: {
3308 struct kvm_msr_list __user *user_msr_list = argp;
3309 struct kvm_msr_list msr_list;
3313 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3316 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3317 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3320 if (n < msr_list.nmsrs)
3323 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3324 num_msrs_to_save * sizeof(u32)))
3326 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3328 num_emulated_msrs * sizeof(u32)))
3333 case KVM_GET_SUPPORTED_CPUID:
3334 case KVM_GET_EMULATED_CPUID: {
3335 struct kvm_cpuid2 __user *cpuid_arg = argp;
3336 struct kvm_cpuid2 cpuid;
3339 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3342 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3348 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3353 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3355 if (copy_to_user(argp, &kvm_mce_cap_supported,
3356 sizeof(kvm_mce_cap_supported)))
3360 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3361 struct kvm_msr_list __user *user_msr_list = argp;
3362 struct kvm_msr_list msr_list;
3366 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3369 msr_list.nmsrs = num_msr_based_features;
3370 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3373 if (n < msr_list.nmsrs)
3376 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3377 num_msr_based_features * sizeof(u32)))
3383 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3393 static void wbinvd_ipi(void *garbage)
3398 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3400 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3403 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3405 /* Address WBINVD may be executed by guest */
3406 if (need_emulate_wbinvd(vcpu)) {
3407 if (kvm_x86_ops->has_wbinvd_exit())
3408 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3409 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3410 smp_call_function_single(vcpu->cpu,
3411 wbinvd_ipi, NULL, 1);
3414 kvm_x86_ops->vcpu_load(vcpu, cpu);
3416 fpregs_assert_state_consistent();
3417 if (test_thread_flag(TIF_NEED_FPU_LOAD))
3418 switch_fpu_return();
3420 /* Apply any externally detected TSC adjustments (due to suspend) */
3421 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3422 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3423 vcpu->arch.tsc_offset_adjustment = 0;
3424 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3427 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3428 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3429 rdtsc() - vcpu->arch.last_host_tsc;
3431 mark_tsc_unstable("KVM discovered backwards TSC");
3433 if (kvm_check_tsc_unstable()) {
3434 u64 offset = kvm_compute_tsc_offset(vcpu,
3435 vcpu->arch.last_guest_tsc);
3436 kvm_vcpu_write_tsc_offset(vcpu, offset);
3437 vcpu->arch.tsc_catchup = 1;
3440 if (kvm_lapic_hv_timer_in_use(vcpu))
3441 kvm_lapic_restart_hv_timer(vcpu);
3444 * On a host with synchronized TSC, there is no need to update
3445 * kvmclock on vcpu->cpu migration
3447 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3448 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3449 if (vcpu->cpu != cpu)
3450 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3454 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3457 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3459 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3462 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3464 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3465 &vcpu->arch.st.steal.preempted,
3466 offsetof(struct kvm_steal_time, preempted),
3467 sizeof(vcpu->arch.st.steal.preempted));
3470 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3474 if (vcpu->preempted)
3475 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3478 * Disable page faults because we're in atomic context here.
3479 * kvm_write_guest_offset_cached() would call might_fault()
3480 * that relies on pagefault_disable() to tell if there's a
3481 * bug. NOTE: the write to guest memory may not go through if
3482 * during postcopy live migration or if there's heavy guest
3485 pagefault_disable();
3487 * kvm_memslots() will be called by
3488 * kvm_write_guest_offset_cached() so take the srcu lock.
3490 idx = srcu_read_lock(&vcpu->kvm->srcu);
3491 kvm_steal_time_set_preempted(vcpu);
3492 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3494 kvm_x86_ops->vcpu_put(vcpu);
3495 vcpu->arch.last_host_tsc = rdtsc();
3497 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3498 * on every vmexit, but if not, we might have a stale dr6 from the
3499 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3504 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3505 struct kvm_lapic_state *s)
3507 if (vcpu->arch.apicv_active)
3508 kvm_x86_ops->sync_pir_to_irr(vcpu);
3510 return kvm_apic_get_state(vcpu, s);
3513 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3514 struct kvm_lapic_state *s)
3518 r = kvm_apic_set_state(vcpu, s);
3521 update_cr8_intercept(vcpu);
3526 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3528 return (!lapic_in_kernel(vcpu) ||
3529 kvm_apic_accept_pic_intr(vcpu));
3533 * if userspace requested an interrupt window, check that the
3534 * interrupt window is open.
3536 * No need to exit to userspace if we already have an interrupt queued.
3538 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3540 return kvm_arch_interrupt_allowed(vcpu) &&
3541 !kvm_cpu_has_interrupt(vcpu) &&
3542 !kvm_event_needs_reinjection(vcpu) &&
3543 kvm_cpu_accept_dm_intr(vcpu);
3546 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3547 struct kvm_interrupt *irq)
3549 if (irq->irq >= KVM_NR_INTERRUPTS)
3552 if (!irqchip_in_kernel(vcpu->kvm)) {
3553 kvm_queue_interrupt(vcpu, irq->irq, false);
3554 kvm_make_request(KVM_REQ_EVENT, vcpu);
3559 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3560 * fail for in-kernel 8259.
3562 if (pic_in_kernel(vcpu->kvm))
3565 if (vcpu->arch.pending_external_vector != -1)
3568 vcpu->arch.pending_external_vector = irq->irq;
3569 kvm_make_request(KVM_REQ_EVENT, vcpu);
3573 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3575 kvm_inject_nmi(vcpu);
3580 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3582 kvm_make_request(KVM_REQ_SMI, vcpu);
3587 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3588 struct kvm_tpr_access_ctl *tac)
3592 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3596 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3600 unsigned bank_num = mcg_cap & 0xff, bank;
3603 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3605 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3608 vcpu->arch.mcg_cap = mcg_cap;
3609 /* Init IA32_MCG_CTL to all 1s */
3610 if (mcg_cap & MCG_CTL_P)
3611 vcpu->arch.mcg_ctl = ~(u64)0;
3612 /* Init IA32_MCi_CTL to all 1s */
3613 for (bank = 0; bank < bank_num; bank++)
3614 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3616 kvm_x86_ops->setup_mce(vcpu);
3621 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3622 struct kvm_x86_mce *mce)
3624 u64 mcg_cap = vcpu->arch.mcg_cap;
3625 unsigned bank_num = mcg_cap & 0xff;
3626 u64 *banks = vcpu->arch.mce_banks;
3628 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3631 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3632 * reporting is disabled
3634 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3635 vcpu->arch.mcg_ctl != ~(u64)0)
3637 banks += 4 * mce->bank;
3639 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3640 * reporting is disabled for the bank
3642 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3644 if (mce->status & MCI_STATUS_UC) {
3645 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3646 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3647 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3650 if (banks[1] & MCI_STATUS_VAL)
3651 mce->status |= MCI_STATUS_OVER;
3652 banks[2] = mce->addr;
3653 banks[3] = mce->misc;
3654 vcpu->arch.mcg_status = mce->mcg_status;
3655 banks[1] = mce->status;
3656 kvm_queue_exception(vcpu, MC_VECTOR);
3657 } else if (!(banks[1] & MCI_STATUS_VAL)
3658 || !(banks[1] & MCI_STATUS_UC)) {
3659 if (banks[1] & MCI_STATUS_VAL)
3660 mce->status |= MCI_STATUS_OVER;
3661 banks[2] = mce->addr;
3662 banks[3] = mce->misc;
3663 banks[1] = mce->status;
3665 banks[1] |= MCI_STATUS_OVER;
3669 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3670 struct kvm_vcpu_events *events)
3675 * The API doesn't provide the instruction length for software
3676 * exceptions, so don't report them. As long as the guest RIP
3677 * isn't advanced, we should expect to encounter the exception
3680 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3681 events->exception.injected = 0;
3682 events->exception.pending = 0;
3684 events->exception.injected = vcpu->arch.exception.injected;
3685 events->exception.pending = vcpu->arch.exception.pending;
3687 * For ABI compatibility, deliberately conflate
3688 * pending and injected exceptions when
3689 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3691 if (!vcpu->kvm->arch.exception_payload_enabled)
3692 events->exception.injected |=
3693 vcpu->arch.exception.pending;
3695 events->exception.nr = vcpu->arch.exception.nr;
3696 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3697 events->exception.error_code = vcpu->arch.exception.error_code;
3698 events->exception_has_payload = vcpu->arch.exception.has_payload;
3699 events->exception_payload = vcpu->arch.exception.payload;
3701 events->interrupt.injected =
3702 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3703 events->interrupt.nr = vcpu->arch.interrupt.nr;
3704 events->interrupt.soft = 0;
3705 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3707 events->nmi.injected = vcpu->arch.nmi_injected;
3708 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3709 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3710 events->nmi.pad = 0;
3712 events->sipi_vector = 0; /* never valid when reporting to user space */
3714 events->smi.smm = is_smm(vcpu);
3715 events->smi.pending = vcpu->arch.smi_pending;
3716 events->smi.smm_inside_nmi =
3717 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3718 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3720 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3721 | KVM_VCPUEVENT_VALID_SHADOW
3722 | KVM_VCPUEVENT_VALID_SMM);
3723 if (vcpu->kvm->arch.exception_payload_enabled)
3724 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3726 memset(&events->reserved, 0, sizeof(events->reserved));
3729 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3731 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3732 struct kvm_vcpu_events *events)
3734 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3735 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3736 | KVM_VCPUEVENT_VALID_SHADOW
3737 | KVM_VCPUEVENT_VALID_SMM
3738 | KVM_VCPUEVENT_VALID_PAYLOAD))
3741 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3742 if (!vcpu->kvm->arch.exception_payload_enabled)
3744 if (events->exception.pending)
3745 events->exception.injected = 0;
3747 events->exception_has_payload = 0;
3749 events->exception.pending = 0;
3750 events->exception_has_payload = 0;
3753 if ((events->exception.injected || events->exception.pending) &&
3754 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3757 /* INITs are latched while in SMM */
3758 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3759 (events->smi.smm || events->smi.pending) &&
3760 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3764 vcpu->arch.exception.injected = events->exception.injected;
3765 vcpu->arch.exception.pending = events->exception.pending;
3766 vcpu->arch.exception.nr = events->exception.nr;
3767 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3768 vcpu->arch.exception.error_code = events->exception.error_code;
3769 vcpu->arch.exception.has_payload = events->exception_has_payload;
3770 vcpu->arch.exception.payload = events->exception_payload;
3772 vcpu->arch.interrupt.injected = events->interrupt.injected;
3773 vcpu->arch.interrupt.nr = events->interrupt.nr;
3774 vcpu->arch.interrupt.soft = events->interrupt.soft;
3775 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3776 kvm_x86_ops->set_interrupt_shadow(vcpu,
3777 events->interrupt.shadow);
3779 vcpu->arch.nmi_injected = events->nmi.injected;
3780 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3781 vcpu->arch.nmi_pending = events->nmi.pending;
3782 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3784 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3785 lapic_in_kernel(vcpu))
3786 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3788 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3789 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3790 if (events->smi.smm)
3791 vcpu->arch.hflags |= HF_SMM_MASK;
3793 vcpu->arch.hflags &= ~HF_SMM_MASK;
3794 kvm_smm_changed(vcpu);
3797 vcpu->arch.smi_pending = events->smi.pending;
3799 if (events->smi.smm) {
3800 if (events->smi.smm_inside_nmi)
3801 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3803 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3804 if (lapic_in_kernel(vcpu)) {
3805 if (events->smi.latched_init)
3806 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3808 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3813 kvm_make_request(KVM_REQ_EVENT, vcpu);
3818 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3819 struct kvm_debugregs *dbgregs)
3823 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3824 kvm_get_dr(vcpu, 6, &val);
3826 dbgregs->dr7 = vcpu->arch.dr7;
3828 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3831 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3832 struct kvm_debugregs *dbgregs)
3837 if (dbgregs->dr6 & ~0xffffffffull)
3839 if (dbgregs->dr7 & ~0xffffffffull)
3842 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3843 kvm_update_dr0123(vcpu);
3844 vcpu->arch.dr6 = dbgregs->dr6;
3845 kvm_update_dr6(vcpu);
3846 vcpu->arch.dr7 = dbgregs->dr7;
3847 kvm_update_dr7(vcpu);
3852 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3854 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3856 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3857 u64 xstate_bv = xsave->header.xfeatures;
3861 * Copy legacy XSAVE area, to avoid complications with CPUID
3862 * leaves 0 and 1 in the loop below.
3864 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3867 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3868 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3871 * Copy each region from the possibly compacted offset to the
3872 * non-compacted offset.
3874 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3876 u64 xfeature_mask = valid & -valid;
3877 int xfeature_nr = fls64(xfeature_mask) - 1;
3878 void *src = get_xsave_addr(xsave, xfeature_nr);
3881 u32 size, offset, ecx, edx;
3882 cpuid_count(XSTATE_CPUID, xfeature_nr,
3883 &size, &offset, &ecx, &edx);
3884 if (xfeature_nr == XFEATURE_PKRU)
3885 memcpy(dest + offset, &vcpu->arch.pkru,
3886 sizeof(vcpu->arch.pkru));
3888 memcpy(dest + offset, src, size);
3892 valid -= xfeature_mask;
3896 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3898 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3899 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3903 * Copy legacy XSAVE area, to avoid complications with CPUID
3904 * leaves 0 and 1 in the loop below.
3906 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3908 /* Set XSTATE_BV and possibly XCOMP_BV. */
3909 xsave->header.xfeatures = xstate_bv;
3910 if (boot_cpu_has(X86_FEATURE_XSAVES))
3911 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3914 * Copy each region from the non-compacted offset to the
3915 * possibly compacted offset.
3917 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3919 u64 xfeature_mask = valid & -valid;
3920 int xfeature_nr = fls64(xfeature_mask) - 1;
3921 void *dest = get_xsave_addr(xsave, xfeature_nr);
3924 u32 size, offset, ecx, edx;
3925 cpuid_count(XSTATE_CPUID, xfeature_nr,
3926 &size, &offset, &ecx, &edx);
3927 if (xfeature_nr == XFEATURE_PKRU)
3928 memcpy(&vcpu->arch.pkru, src + offset,
3929 sizeof(vcpu->arch.pkru));
3931 memcpy(dest, src + offset, size);
3934 valid -= xfeature_mask;
3938 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3939 struct kvm_xsave *guest_xsave)
3941 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3942 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3943 fill_xsave((u8 *) guest_xsave->region, vcpu);
3945 memcpy(guest_xsave->region,
3946 &vcpu->arch.guest_fpu->state.fxsave,
3947 sizeof(struct fxregs_state));
3948 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3949 XFEATURE_MASK_FPSSE;
3953 #define XSAVE_MXCSR_OFFSET 24
3955 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3956 struct kvm_xsave *guest_xsave)
3959 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3960 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3962 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3964 * Here we allow setting states that are not present in
3965 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3966 * with old userspace.
3968 if (xstate_bv & ~kvm_supported_xcr0() ||
3969 mxcsr & ~mxcsr_feature_mask)
3971 load_xsave(vcpu, (u8 *)guest_xsave->region);
3973 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3974 mxcsr & ~mxcsr_feature_mask)
3976 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3977 guest_xsave->region, sizeof(struct fxregs_state));
3982 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3983 struct kvm_xcrs *guest_xcrs)
3985 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3986 guest_xcrs->nr_xcrs = 0;
3990 guest_xcrs->nr_xcrs = 1;
3991 guest_xcrs->flags = 0;
3992 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3993 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3996 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3997 struct kvm_xcrs *guest_xcrs)
4001 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4004 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4007 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4008 /* Only support XCR0 currently */
4009 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4010 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4011 guest_xcrs->xcrs[i].value);
4020 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4021 * stopped by the hypervisor. This function will be called from the host only.
4022 * EINVAL is returned when the host attempts to set the flag for a guest that
4023 * does not support pv clocks.
4025 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4027 if (!vcpu->arch.pv_time_enabled)
4029 vcpu->arch.pvclock_set_guest_stopped_request = true;
4030 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4034 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4035 struct kvm_enable_cap *cap)
4038 uint16_t vmcs_version;
4039 void __user *user_ptr;
4045 case KVM_CAP_HYPERV_SYNIC2:
4050 case KVM_CAP_HYPERV_SYNIC:
4051 if (!irqchip_in_kernel(vcpu->kvm))
4053 return kvm_hv_activate_synic(vcpu, cap->cap ==
4054 KVM_CAP_HYPERV_SYNIC2);
4055 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4056 if (!kvm_x86_ops->nested_enable_evmcs)
4058 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4060 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4061 if (copy_to_user(user_ptr, &vmcs_version,
4062 sizeof(vmcs_version)))
4066 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4067 if (!kvm_x86_ops->enable_direct_tlbflush)
4070 return kvm_x86_ops->enable_direct_tlbflush(vcpu);
4077 long kvm_arch_vcpu_ioctl(struct file *filp,
4078 unsigned int ioctl, unsigned long arg)
4080 struct kvm_vcpu *vcpu = filp->private_data;
4081 void __user *argp = (void __user *)arg;
4084 struct kvm_lapic_state *lapic;
4085 struct kvm_xsave *xsave;
4086 struct kvm_xcrs *xcrs;
4094 case KVM_GET_LAPIC: {
4096 if (!lapic_in_kernel(vcpu))
4098 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4099 GFP_KERNEL_ACCOUNT);
4104 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4108 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4113 case KVM_SET_LAPIC: {
4115 if (!lapic_in_kernel(vcpu))
4117 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4118 if (IS_ERR(u.lapic)) {
4119 r = PTR_ERR(u.lapic);
4123 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4126 case KVM_INTERRUPT: {
4127 struct kvm_interrupt irq;
4130 if (copy_from_user(&irq, argp, sizeof(irq)))
4132 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4136 r = kvm_vcpu_ioctl_nmi(vcpu);
4140 r = kvm_vcpu_ioctl_smi(vcpu);
4143 case KVM_SET_CPUID: {
4144 struct kvm_cpuid __user *cpuid_arg = argp;
4145 struct kvm_cpuid cpuid;
4148 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4150 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4153 case KVM_SET_CPUID2: {
4154 struct kvm_cpuid2 __user *cpuid_arg = argp;
4155 struct kvm_cpuid2 cpuid;
4158 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4160 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4161 cpuid_arg->entries);
4164 case KVM_GET_CPUID2: {
4165 struct kvm_cpuid2 __user *cpuid_arg = argp;
4166 struct kvm_cpuid2 cpuid;
4169 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4171 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4172 cpuid_arg->entries);
4176 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4181 case KVM_GET_MSRS: {
4182 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4183 r = msr_io(vcpu, argp, do_get_msr, 1);
4184 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4187 case KVM_SET_MSRS: {
4188 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4189 r = msr_io(vcpu, argp, do_set_msr, 0);
4190 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4193 case KVM_TPR_ACCESS_REPORTING: {
4194 struct kvm_tpr_access_ctl tac;
4197 if (copy_from_user(&tac, argp, sizeof(tac)))
4199 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4203 if (copy_to_user(argp, &tac, sizeof(tac)))
4208 case KVM_SET_VAPIC_ADDR: {
4209 struct kvm_vapic_addr va;
4213 if (!lapic_in_kernel(vcpu))
4216 if (copy_from_user(&va, argp, sizeof(va)))
4218 idx = srcu_read_lock(&vcpu->kvm->srcu);
4219 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4220 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4223 case KVM_X86_SETUP_MCE: {
4227 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4229 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4232 case KVM_X86_SET_MCE: {
4233 struct kvm_x86_mce mce;
4236 if (copy_from_user(&mce, argp, sizeof(mce)))
4238 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4241 case KVM_GET_VCPU_EVENTS: {
4242 struct kvm_vcpu_events events;
4244 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4247 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4252 case KVM_SET_VCPU_EVENTS: {
4253 struct kvm_vcpu_events events;
4256 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4259 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4262 case KVM_GET_DEBUGREGS: {
4263 struct kvm_debugregs dbgregs;
4265 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4268 if (copy_to_user(argp, &dbgregs,
4269 sizeof(struct kvm_debugregs)))
4274 case KVM_SET_DEBUGREGS: {
4275 struct kvm_debugregs dbgregs;
4278 if (copy_from_user(&dbgregs, argp,
4279 sizeof(struct kvm_debugregs)))
4282 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4285 case KVM_GET_XSAVE: {
4286 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4291 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4294 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4299 case KVM_SET_XSAVE: {
4300 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4301 if (IS_ERR(u.xsave)) {
4302 r = PTR_ERR(u.xsave);
4306 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4309 case KVM_GET_XCRS: {
4310 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4315 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4318 if (copy_to_user(argp, u.xcrs,
4319 sizeof(struct kvm_xcrs)))
4324 case KVM_SET_XCRS: {
4325 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4326 if (IS_ERR(u.xcrs)) {
4327 r = PTR_ERR(u.xcrs);
4331 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4334 case KVM_SET_TSC_KHZ: {
4338 user_tsc_khz = (u32)arg;
4340 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4343 if (user_tsc_khz == 0)
4344 user_tsc_khz = tsc_khz;
4346 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4351 case KVM_GET_TSC_KHZ: {
4352 r = vcpu->arch.virtual_tsc_khz;
4355 case KVM_KVMCLOCK_CTRL: {
4356 r = kvm_set_guest_paused(vcpu);
4359 case KVM_ENABLE_CAP: {
4360 struct kvm_enable_cap cap;
4363 if (copy_from_user(&cap, argp, sizeof(cap)))
4365 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4368 case KVM_GET_NESTED_STATE: {
4369 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4373 if (!kvm_x86_ops->get_nested_state)
4376 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4378 if (get_user(user_data_size, &user_kvm_nested_state->size))
4381 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4386 if (r > user_data_size) {
4387 if (put_user(r, &user_kvm_nested_state->size))
4397 case KVM_SET_NESTED_STATE: {
4398 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4399 struct kvm_nested_state kvm_state;
4402 if (!kvm_x86_ops->set_nested_state)
4406 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4410 if (kvm_state.size < sizeof(kvm_state))
4413 if (kvm_state.flags &
4414 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4415 | KVM_STATE_NESTED_EVMCS))
4418 /* nested_run_pending implies guest_mode. */
4419 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4420 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4423 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4426 case KVM_GET_SUPPORTED_HV_CPUID: {
4427 struct kvm_cpuid2 __user *cpuid_arg = argp;
4428 struct kvm_cpuid2 cpuid;
4431 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4434 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4435 cpuid_arg->entries);
4440 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4455 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4457 return VM_FAULT_SIGBUS;
4460 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4464 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4466 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4470 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4473 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4476 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4477 unsigned long kvm_nr_mmu_pages)
4479 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4482 mutex_lock(&kvm->slots_lock);
4484 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4485 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4487 mutex_unlock(&kvm->slots_lock);
4491 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4493 return kvm->arch.n_max_mmu_pages;
4496 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4498 struct kvm_pic *pic = kvm->arch.vpic;
4502 switch (chip->chip_id) {
4503 case KVM_IRQCHIP_PIC_MASTER:
4504 memcpy(&chip->chip.pic, &pic->pics[0],
4505 sizeof(struct kvm_pic_state));
4507 case KVM_IRQCHIP_PIC_SLAVE:
4508 memcpy(&chip->chip.pic, &pic->pics[1],
4509 sizeof(struct kvm_pic_state));
4511 case KVM_IRQCHIP_IOAPIC:
4512 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4521 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4523 struct kvm_pic *pic = kvm->arch.vpic;
4527 switch (chip->chip_id) {
4528 case KVM_IRQCHIP_PIC_MASTER:
4529 spin_lock(&pic->lock);
4530 memcpy(&pic->pics[0], &chip->chip.pic,
4531 sizeof(struct kvm_pic_state));
4532 spin_unlock(&pic->lock);
4534 case KVM_IRQCHIP_PIC_SLAVE:
4535 spin_lock(&pic->lock);
4536 memcpy(&pic->pics[1], &chip->chip.pic,
4537 sizeof(struct kvm_pic_state));
4538 spin_unlock(&pic->lock);
4540 case KVM_IRQCHIP_IOAPIC:
4541 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4547 kvm_pic_update_irq(pic);
4551 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4553 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4555 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4557 mutex_lock(&kps->lock);
4558 memcpy(ps, &kps->channels, sizeof(*ps));
4559 mutex_unlock(&kps->lock);
4563 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4566 struct kvm_pit *pit = kvm->arch.vpit;
4568 mutex_lock(&pit->pit_state.lock);
4569 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4570 for (i = 0; i < 3; i++)
4571 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4572 mutex_unlock(&pit->pit_state.lock);
4576 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4578 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4579 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4580 sizeof(ps->channels));
4581 ps->flags = kvm->arch.vpit->pit_state.flags;
4582 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4583 memset(&ps->reserved, 0, sizeof(ps->reserved));
4587 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4591 u32 prev_legacy, cur_legacy;
4592 struct kvm_pit *pit = kvm->arch.vpit;
4594 mutex_lock(&pit->pit_state.lock);
4595 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4596 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4597 if (!prev_legacy && cur_legacy)
4599 memcpy(&pit->pit_state.channels, &ps->channels,
4600 sizeof(pit->pit_state.channels));
4601 pit->pit_state.flags = ps->flags;
4602 for (i = 0; i < 3; i++)
4603 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4605 mutex_unlock(&pit->pit_state.lock);
4609 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4610 struct kvm_reinject_control *control)
4612 struct kvm_pit *pit = kvm->arch.vpit;
4617 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4618 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4619 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4621 mutex_lock(&pit->pit_state.lock);
4622 kvm_pit_set_reinject(pit, control->pit_reinject);
4623 mutex_unlock(&pit->pit_state.lock);
4629 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4630 * @kvm: kvm instance
4631 * @log: slot id and address to which we copy the log
4633 * Steps 1-4 below provide general overview of dirty page logging. See
4634 * kvm_get_dirty_log_protect() function description for additional details.
4636 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4637 * always flush the TLB (step 4) even if previous step failed and the dirty
4638 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4639 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4640 * writes will be marked dirty for next log read.
4642 * 1. Take a snapshot of the bit and clear it if needed.
4643 * 2. Write protect the corresponding page.
4644 * 3. Copy the snapshot to the userspace.
4645 * 4. Flush TLB's if needed.
4647 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4652 mutex_lock(&kvm->slots_lock);
4655 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4657 if (kvm_x86_ops->flush_log_dirty)
4658 kvm_x86_ops->flush_log_dirty(kvm);
4660 r = kvm_get_dirty_log_protect(kvm, log, &flush);
4663 * All the TLBs can be flushed out of mmu lock, see the comments in
4664 * kvm_mmu_slot_remove_write_access().
4666 lockdep_assert_held(&kvm->slots_lock);
4668 kvm_flush_remote_tlbs(kvm);
4670 mutex_unlock(&kvm->slots_lock);
4674 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4679 mutex_lock(&kvm->slots_lock);
4682 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4684 if (kvm_x86_ops->flush_log_dirty)
4685 kvm_x86_ops->flush_log_dirty(kvm);
4687 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4690 * All the TLBs can be flushed out of mmu lock, see the comments in
4691 * kvm_mmu_slot_remove_write_access().
4693 lockdep_assert_held(&kvm->slots_lock);
4695 kvm_flush_remote_tlbs(kvm);
4697 mutex_unlock(&kvm->slots_lock);
4701 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4704 if (!irqchip_in_kernel(kvm))
4707 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4708 irq_event->irq, irq_event->level,
4713 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4714 struct kvm_enable_cap *cap)
4722 case KVM_CAP_DISABLE_QUIRKS:
4723 kvm->arch.disabled_quirks = cap->args[0];
4726 case KVM_CAP_SPLIT_IRQCHIP: {
4727 mutex_lock(&kvm->lock);
4729 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4730 goto split_irqchip_unlock;
4732 if (irqchip_in_kernel(kvm))
4733 goto split_irqchip_unlock;
4734 if (kvm->created_vcpus)
4735 goto split_irqchip_unlock;
4736 r = kvm_setup_empty_irq_routing(kvm);
4738 goto split_irqchip_unlock;
4739 /* Pairs with irqchip_in_kernel. */
4741 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4742 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4744 split_irqchip_unlock:
4745 mutex_unlock(&kvm->lock);
4748 case KVM_CAP_X2APIC_API:
4750 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4753 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4754 kvm->arch.x2apic_format = true;
4755 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4756 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4760 case KVM_CAP_X86_DISABLE_EXITS:
4762 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4765 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4766 kvm_can_mwait_in_guest())
4767 kvm->arch.mwait_in_guest = true;
4768 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4769 kvm->arch.hlt_in_guest = true;
4770 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4771 kvm->arch.pause_in_guest = true;
4772 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4773 kvm->arch.cstate_in_guest = true;
4776 case KVM_CAP_MSR_PLATFORM_INFO:
4777 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4780 case KVM_CAP_EXCEPTION_PAYLOAD:
4781 kvm->arch.exception_payload_enabled = cap->args[0];
4791 long kvm_arch_vm_ioctl(struct file *filp,
4792 unsigned int ioctl, unsigned long arg)
4794 struct kvm *kvm = filp->private_data;
4795 void __user *argp = (void __user *)arg;
4798 * This union makes it completely explicit to gcc-3.x
4799 * that these two variables' stack usage should be
4800 * combined, not added together.
4803 struct kvm_pit_state ps;
4804 struct kvm_pit_state2 ps2;
4805 struct kvm_pit_config pit_config;
4809 case KVM_SET_TSS_ADDR:
4810 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4812 case KVM_SET_IDENTITY_MAP_ADDR: {
4815 mutex_lock(&kvm->lock);
4817 if (kvm->created_vcpus)
4818 goto set_identity_unlock;
4820 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4821 goto set_identity_unlock;
4822 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4823 set_identity_unlock:
4824 mutex_unlock(&kvm->lock);
4827 case KVM_SET_NR_MMU_PAGES:
4828 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4830 case KVM_GET_NR_MMU_PAGES:
4831 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4833 case KVM_CREATE_IRQCHIP: {
4834 mutex_lock(&kvm->lock);
4837 if (irqchip_in_kernel(kvm))
4838 goto create_irqchip_unlock;
4841 if (kvm->created_vcpus)
4842 goto create_irqchip_unlock;
4844 r = kvm_pic_init(kvm);
4846 goto create_irqchip_unlock;
4848 r = kvm_ioapic_init(kvm);
4850 kvm_pic_destroy(kvm);
4851 goto create_irqchip_unlock;
4854 r = kvm_setup_default_irq_routing(kvm);
4856 kvm_ioapic_destroy(kvm);
4857 kvm_pic_destroy(kvm);
4858 goto create_irqchip_unlock;
4860 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4862 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4863 create_irqchip_unlock:
4864 mutex_unlock(&kvm->lock);
4867 case KVM_CREATE_PIT:
4868 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4870 case KVM_CREATE_PIT2:
4872 if (copy_from_user(&u.pit_config, argp,
4873 sizeof(struct kvm_pit_config)))
4876 mutex_lock(&kvm->lock);
4879 goto create_pit_unlock;
4881 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4885 mutex_unlock(&kvm->lock);
4887 case KVM_GET_IRQCHIP: {
4888 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4889 struct kvm_irqchip *chip;
4891 chip = memdup_user(argp, sizeof(*chip));
4898 if (!irqchip_kernel(kvm))
4899 goto get_irqchip_out;
4900 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4902 goto get_irqchip_out;
4904 if (copy_to_user(argp, chip, sizeof(*chip)))
4905 goto get_irqchip_out;
4911 case KVM_SET_IRQCHIP: {
4912 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4913 struct kvm_irqchip *chip;
4915 chip = memdup_user(argp, sizeof(*chip));
4922 if (!irqchip_kernel(kvm))
4923 goto set_irqchip_out;
4924 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4926 goto set_irqchip_out;
4934 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4937 if (!kvm->arch.vpit)
4939 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4943 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4950 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4953 if (!kvm->arch.vpit)
4955 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4958 case KVM_GET_PIT2: {
4960 if (!kvm->arch.vpit)
4962 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4966 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4971 case KVM_SET_PIT2: {
4973 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4976 if (!kvm->arch.vpit)
4978 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4981 case KVM_REINJECT_CONTROL: {
4982 struct kvm_reinject_control control;
4984 if (copy_from_user(&control, argp, sizeof(control)))
4986 r = kvm_vm_ioctl_reinject(kvm, &control);
4989 case KVM_SET_BOOT_CPU_ID:
4991 mutex_lock(&kvm->lock);
4992 if (kvm->created_vcpus)
4995 kvm->arch.bsp_vcpu_id = arg;
4996 mutex_unlock(&kvm->lock);
4998 case KVM_XEN_HVM_CONFIG: {
4999 struct kvm_xen_hvm_config xhc;
5001 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5006 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5010 case KVM_SET_CLOCK: {
5011 struct kvm_clock_data user_ns;
5015 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5024 * TODO: userspace has to take care of races with VCPU_RUN, so
5025 * kvm_gen_update_masterclock() can be cut down to locked
5026 * pvclock_update_vm_gtod_copy().
5028 kvm_gen_update_masterclock(kvm);
5029 now_ns = get_kvmclock_ns(kvm);
5030 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5031 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5034 case KVM_GET_CLOCK: {
5035 struct kvm_clock_data user_ns;
5038 now_ns = get_kvmclock_ns(kvm);
5039 user_ns.clock = now_ns;
5040 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5041 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5044 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5049 case KVM_MEMORY_ENCRYPT_OP: {
5051 if (kvm_x86_ops->mem_enc_op)
5052 r = kvm_x86_ops->mem_enc_op(kvm, argp);
5055 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5056 struct kvm_enc_region region;
5059 if (copy_from_user(®ion, argp, sizeof(region)))
5063 if (kvm_x86_ops->mem_enc_reg_region)
5064 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
5067 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5068 struct kvm_enc_region region;
5071 if (copy_from_user(®ion, argp, sizeof(region)))
5075 if (kvm_x86_ops->mem_enc_unreg_region)
5076 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
5079 case KVM_HYPERV_EVENTFD: {
5080 struct kvm_hyperv_eventfd hvevfd;
5083 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5085 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5088 case KVM_SET_PMU_EVENT_FILTER:
5089 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5098 static void kvm_init_msr_list(void)
5103 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5104 "Please update the fixed PMCs in msrs_to_save[]");
5105 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_GENERIC != 32,
5106 "Please update the generic perfctr/eventsel MSRs in msrs_to_save[]");
5108 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
5109 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
5113 * Even MSRs that are valid in the host may not be exposed
5114 * to the guests in some cases.
5116 switch (msrs_to_save[i]) {
5117 case MSR_IA32_BNDCFGS:
5118 if (!kvm_mpx_supported())
5122 if (!kvm_x86_ops->rdtscp_supported())
5125 case MSR_IA32_RTIT_CTL:
5126 case MSR_IA32_RTIT_STATUS:
5127 if (!kvm_x86_ops->pt_supported())
5130 case MSR_IA32_RTIT_CR3_MATCH:
5131 if (!kvm_x86_ops->pt_supported() ||
5132 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5135 case MSR_IA32_RTIT_OUTPUT_BASE:
5136 case MSR_IA32_RTIT_OUTPUT_MASK:
5137 if (!kvm_x86_ops->pt_supported() ||
5138 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5139 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5142 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5143 if (!kvm_x86_ops->pt_supported() ||
5144 msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
5145 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5154 msrs_to_save[j] = msrs_to_save[i];
5157 num_msrs_to_save = j;
5159 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
5160 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
5164 emulated_msrs[j] = emulated_msrs[i];
5167 num_emulated_msrs = j;
5169 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
5170 struct kvm_msr_entry msr;
5172 msr.index = msr_based_features[i];
5173 if (kvm_get_msr_feature(&msr))
5177 msr_based_features[j] = msr_based_features[i];
5180 num_msr_based_features = j;
5183 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5191 if (!(lapic_in_kernel(vcpu) &&
5192 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5193 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5204 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5211 if (!(lapic_in_kernel(vcpu) &&
5212 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5214 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5216 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5226 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5227 struct kvm_segment *var, int seg)
5229 kvm_x86_ops->set_segment(vcpu, var, seg);
5232 void kvm_get_segment(struct kvm_vcpu *vcpu,
5233 struct kvm_segment *var, int seg)
5235 kvm_x86_ops->get_segment(vcpu, var, seg);
5238 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5239 struct x86_exception *exception)
5243 BUG_ON(!mmu_is_nested(vcpu));
5245 /* NPT walks are always user-walks */
5246 access |= PFERR_USER_MASK;
5247 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5252 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5253 struct x86_exception *exception)
5255 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5256 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5259 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5260 struct x86_exception *exception)
5262 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5263 access |= PFERR_FETCH_MASK;
5264 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5267 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5268 struct x86_exception *exception)
5270 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5271 access |= PFERR_WRITE_MASK;
5272 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5275 /* uses this to access any guest's mapped memory without checking CPL */
5276 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5277 struct x86_exception *exception)
5279 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5282 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5283 struct kvm_vcpu *vcpu, u32 access,
5284 struct x86_exception *exception)
5287 int r = X86EMUL_CONTINUE;
5290 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5292 unsigned offset = addr & (PAGE_SIZE-1);
5293 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5296 if (gpa == UNMAPPED_GVA)
5297 return X86EMUL_PROPAGATE_FAULT;
5298 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5301 r = X86EMUL_IO_NEEDED;
5313 /* used for instruction fetching */
5314 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5315 gva_t addr, void *val, unsigned int bytes,
5316 struct x86_exception *exception)
5318 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5319 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5323 /* Inline kvm_read_guest_virt_helper for speed. */
5324 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5326 if (unlikely(gpa == UNMAPPED_GVA))
5327 return X86EMUL_PROPAGATE_FAULT;
5329 offset = addr & (PAGE_SIZE-1);
5330 if (WARN_ON(offset + bytes > PAGE_SIZE))
5331 bytes = (unsigned)PAGE_SIZE - offset;
5332 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5334 if (unlikely(ret < 0))
5335 return X86EMUL_IO_NEEDED;
5337 return X86EMUL_CONTINUE;
5340 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5341 gva_t addr, void *val, unsigned int bytes,
5342 struct x86_exception *exception)
5344 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5347 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5348 * is returned, but our callers are not ready for that and they blindly
5349 * call kvm_inject_page_fault. Ensure that they at least do not leak
5350 * uninitialized kernel stack memory into cr2 and error code.
5352 memset(exception, 0, sizeof(*exception));
5353 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5356 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5358 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5359 gva_t addr, void *val, unsigned int bytes,
5360 struct x86_exception *exception, bool system)
5362 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5365 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5366 access |= PFERR_USER_MASK;
5368 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5371 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5372 unsigned long addr, void *val, unsigned int bytes)
5374 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5375 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5377 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5380 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5381 struct kvm_vcpu *vcpu, u32 access,
5382 struct x86_exception *exception)
5385 int r = X86EMUL_CONTINUE;
5388 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5391 unsigned offset = addr & (PAGE_SIZE-1);
5392 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5395 if (gpa == UNMAPPED_GVA)
5396 return X86EMUL_PROPAGATE_FAULT;
5397 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5399 r = X86EMUL_IO_NEEDED;
5411 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5412 unsigned int bytes, struct x86_exception *exception,
5415 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5416 u32 access = PFERR_WRITE_MASK;
5418 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5419 access |= PFERR_USER_MASK;
5421 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5425 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5426 unsigned int bytes, struct x86_exception *exception)
5428 /* kvm_write_guest_virt_system can pull in tons of pages. */
5429 vcpu->arch.l1tf_flush_l1d = true;
5432 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5433 * is returned, but our callers are not ready for that and they blindly
5434 * call kvm_inject_page_fault. Ensure that they at least do not leak
5435 * uninitialized kernel stack memory into cr2 and error code.
5437 memset(exception, 0, sizeof(*exception));
5438 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5439 PFERR_WRITE_MASK, exception);
5441 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5443 int handle_ud(struct kvm_vcpu *vcpu)
5445 int emul_type = EMULTYPE_TRAP_UD;
5446 char sig[5]; /* ud2; .ascii "kvm" */
5447 struct x86_exception e;
5449 if (force_emulation_prefix &&
5450 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5451 sig, sizeof(sig), &e) == 0 &&
5452 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5453 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5454 emul_type = EMULTYPE_TRAP_UD_FORCED;
5457 return kvm_emulate_instruction(vcpu, emul_type);
5459 EXPORT_SYMBOL_GPL(handle_ud);
5461 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5462 gpa_t gpa, bool write)
5464 /* For APIC access vmexit */
5465 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5468 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5469 trace_vcpu_match_mmio(gva, gpa, write, true);
5476 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5477 gpa_t *gpa, struct x86_exception *exception,
5480 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5481 | (write ? PFERR_WRITE_MASK : 0);
5484 * currently PKRU is only applied to ept enabled guest so
5485 * there is no pkey in EPT page table for L1 guest or EPT
5486 * shadow page table for L2 guest.
5488 if (vcpu_match_mmio_gva(vcpu, gva)
5489 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5490 vcpu->arch.mmio_access, 0, access)) {
5491 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5492 (gva & (PAGE_SIZE - 1));
5493 trace_vcpu_match_mmio(gva, *gpa, write, false);
5497 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5499 if (*gpa == UNMAPPED_GVA)
5502 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5505 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5506 const void *val, int bytes)
5510 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5513 kvm_page_track_write(vcpu, gpa, val, bytes);
5517 struct read_write_emulator_ops {
5518 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5520 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5521 void *val, int bytes);
5522 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5523 int bytes, void *val);
5524 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5525 void *val, int bytes);
5529 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5531 if (vcpu->mmio_read_completed) {
5532 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5533 vcpu->mmio_fragments[0].gpa, val);
5534 vcpu->mmio_read_completed = 0;
5541 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5542 void *val, int bytes)
5544 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5547 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5548 void *val, int bytes)
5550 return emulator_write_phys(vcpu, gpa, val, bytes);
5553 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5555 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5556 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5559 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5560 void *val, int bytes)
5562 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5563 return X86EMUL_IO_NEEDED;
5566 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5567 void *val, int bytes)
5569 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5571 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5572 return X86EMUL_CONTINUE;
5575 static const struct read_write_emulator_ops read_emultor = {
5576 .read_write_prepare = read_prepare,
5577 .read_write_emulate = read_emulate,
5578 .read_write_mmio = vcpu_mmio_read,
5579 .read_write_exit_mmio = read_exit_mmio,
5582 static const struct read_write_emulator_ops write_emultor = {
5583 .read_write_emulate = write_emulate,
5584 .read_write_mmio = write_mmio,
5585 .read_write_exit_mmio = write_exit_mmio,
5589 static int emulator_read_write_onepage(unsigned long addr, void *val,
5591 struct x86_exception *exception,
5592 struct kvm_vcpu *vcpu,
5593 const struct read_write_emulator_ops *ops)
5597 bool write = ops->write;
5598 struct kvm_mmio_fragment *frag;
5599 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5602 * If the exit was due to a NPF we may already have a GPA.
5603 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5604 * Note, this cannot be used on string operations since string
5605 * operation using rep will only have the initial GPA from the NPF
5608 if (vcpu->arch.gpa_available &&
5609 emulator_can_use_gpa(ctxt) &&
5610 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5611 gpa = vcpu->arch.gpa_val;
5612 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5614 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5616 return X86EMUL_PROPAGATE_FAULT;
5619 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5620 return X86EMUL_CONTINUE;
5623 * Is this MMIO handled locally?
5625 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5626 if (handled == bytes)
5627 return X86EMUL_CONTINUE;
5633 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5634 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5638 return X86EMUL_CONTINUE;
5641 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5643 void *val, unsigned int bytes,
5644 struct x86_exception *exception,
5645 const struct read_write_emulator_ops *ops)
5647 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5651 if (ops->read_write_prepare &&
5652 ops->read_write_prepare(vcpu, val, bytes))
5653 return X86EMUL_CONTINUE;
5655 vcpu->mmio_nr_fragments = 0;
5657 /* Crossing a page boundary? */
5658 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5661 now = -addr & ~PAGE_MASK;
5662 rc = emulator_read_write_onepage(addr, val, now, exception,
5665 if (rc != X86EMUL_CONTINUE)
5668 if (ctxt->mode != X86EMUL_MODE_PROT64)
5674 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5676 if (rc != X86EMUL_CONTINUE)
5679 if (!vcpu->mmio_nr_fragments)
5682 gpa = vcpu->mmio_fragments[0].gpa;
5684 vcpu->mmio_needed = 1;
5685 vcpu->mmio_cur_fragment = 0;
5687 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5688 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5689 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5690 vcpu->run->mmio.phys_addr = gpa;
5692 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5695 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5699 struct x86_exception *exception)
5701 return emulator_read_write(ctxt, addr, val, bytes,
5702 exception, &read_emultor);
5705 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5709 struct x86_exception *exception)
5711 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5712 exception, &write_emultor);
5715 #define CMPXCHG_TYPE(t, ptr, old, new) \
5716 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5718 #ifdef CONFIG_X86_64
5719 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5721 # define CMPXCHG64(ptr, old, new) \
5722 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5725 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5730 struct x86_exception *exception)
5732 struct kvm_host_map map;
5733 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5738 /* guests cmpxchg8b have to be emulated atomically */
5739 if (bytes > 8 || (bytes & (bytes - 1)))
5742 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5744 if (gpa == UNMAPPED_GVA ||
5745 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5748 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5751 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5754 kaddr = map.hva + offset_in_page(gpa);
5758 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5761 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5764 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5767 exchanged = CMPXCHG64(kaddr, old, new);
5773 kvm_vcpu_unmap(vcpu, &map, true);
5776 return X86EMUL_CMPXCHG_FAILED;
5778 kvm_page_track_write(vcpu, gpa, new, bytes);
5780 return X86EMUL_CONTINUE;
5783 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5785 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5788 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5792 for (i = 0; i < vcpu->arch.pio.count; i++) {
5793 if (vcpu->arch.pio.in)
5794 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5795 vcpu->arch.pio.size, pd);
5797 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5798 vcpu->arch.pio.port, vcpu->arch.pio.size,
5802 pd += vcpu->arch.pio.size;
5807 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5808 unsigned short port, void *val,
5809 unsigned int count, bool in)
5811 vcpu->arch.pio.port = port;
5812 vcpu->arch.pio.in = in;
5813 vcpu->arch.pio.count = count;
5814 vcpu->arch.pio.size = size;
5816 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5817 vcpu->arch.pio.count = 0;
5821 vcpu->run->exit_reason = KVM_EXIT_IO;
5822 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5823 vcpu->run->io.size = size;
5824 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5825 vcpu->run->io.count = count;
5826 vcpu->run->io.port = port;
5831 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5832 int size, unsigned short port, void *val,
5835 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5838 if (vcpu->arch.pio.count)
5841 memset(vcpu->arch.pio_data, 0, size * count);
5843 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5846 memcpy(val, vcpu->arch.pio_data, size * count);
5847 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5848 vcpu->arch.pio.count = 0;
5855 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5856 int size, unsigned short port,
5857 const void *val, unsigned int count)
5859 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5861 memcpy(vcpu->arch.pio_data, val, size * count);
5862 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5863 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5866 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5868 return kvm_x86_ops->get_segment_base(vcpu, seg);
5871 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5873 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5876 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5878 if (!need_emulate_wbinvd(vcpu))
5879 return X86EMUL_CONTINUE;
5881 if (kvm_x86_ops->has_wbinvd_exit()) {
5882 int cpu = get_cpu();
5884 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5885 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5886 wbinvd_ipi, NULL, 1);
5888 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5891 return X86EMUL_CONTINUE;
5894 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5896 kvm_emulate_wbinvd_noskip(vcpu);
5897 return kvm_skip_emulated_instruction(vcpu);
5899 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5903 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5905 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5908 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5909 unsigned long *dest)
5911 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5914 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5915 unsigned long value)
5918 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5921 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5923 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5926 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5928 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5929 unsigned long value;
5933 value = kvm_read_cr0(vcpu);
5936 value = vcpu->arch.cr2;
5939 value = kvm_read_cr3(vcpu);
5942 value = kvm_read_cr4(vcpu);
5945 value = kvm_get_cr8(vcpu);
5948 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5955 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5957 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5962 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5965 vcpu->arch.cr2 = val;
5968 res = kvm_set_cr3(vcpu, val);
5971 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5974 res = kvm_set_cr8(vcpu, val);
5977 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5984 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5986 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5989 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5991 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5994 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5996 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5999 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6001 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
6004 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6006 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
6009 static unsigned long emulator_get_cached_segment_base(
6010 struct x86_emulate_ctxt *ctxt, int seg)
6012 return get_segment_base(emul_to_vcpu(ctxt), seg);
6015 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6016 struct desc_struct *desc, u32 *base3,
6019 struct kvm_segment var;
6021 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6022 *selector = var.selector;
6025 memset(desc, 0, sizeof(*desc));
6033 set_desc_limit(desc, var.limit);
6034 set_desc_base(desc, (unsigned long)var.base);
6035 #ifdef CONFIG_X86_64
6037 *base3 = var.base >> 32;
6039 desc->type = var.type;
6041 desc->dpl = var.dpl;
6042 desc->p = var.present;
6043 desc->avl = var.avl;
6051 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6052 struct desc_struct *desc, u32 base3,
6055 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6056 struct kvm_segment var;
6058 var.selector = selector;
6059 var.base = get_desc_base(desc);
6060 #ifdef CONFIG_X86_64
6061 var.base |= ((u64)base3) << 32;
6063 var.limit = get_desc_limit(desc);
6065 var.limit = (var.limit << 12) | 0xfff;
6066 var.type = desc->type;
6067 var.dpl = desc->dpl;
6072 var.avl = desc->avl;
6073 var.present = desc->p;
6074 var.unusable = !var.present;
6077 kvm_set_segment(vcpu, &var, seg);
6081 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6082 u32 msr_index, u64 *pdata)
6084 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6087 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6088 u32 msr_index, u64 data)
6090 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6093 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6095 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6097 return vcpu->arch.smbase;
6100 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6102 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6104 vcpu->arch.smbase = smbase;
6107 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6110 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
6113 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6114 u32 pmc, u64 *pdata)
6116 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6119 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6121 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6124 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6125 struct x86_instruction_info *info,
6126 enum x86_intercept_stage stage)
6128 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6131 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6132 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6134 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6137 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6139 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6142 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6144 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6147 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6149 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6152 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6154 return emul_to_vcpu(ctxt)->arch.hflags;
6157 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6159 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6162 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6163 const char *smstate)
6165 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6168 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6170 kvm_smm_changed(emul_to_vcpu(ctxt));
6173 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6175 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6178 static const struct x86_emulate_ops emulate_ops = {
6179 .read_gpr = emulator_read_gpr,
6180 .write_gpr = emulator_write_gpr,
6181 .read_std = emulator_read_std,
6182 .write_std = emulator_write_std,
6183 .read_phys = kvm_read_guest_phys_system,
6184 .fetch = kvm_fetch_guest_virt,
6185 .read_emulated = emulator_read_emulated,
6186 .write_emulated = emulator_write_emulated,
6187 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6188 .invlpg = emulator_invlpg,
6189 .pio_in_emulated = emulator_pio_in_emulated,
6190 .pio_out_emulated = emulator_pio_out_emulated,
6191 .get_segment = emulator_get_segment,
6192 .set_segment = emulator_set_segment,
6193 .get_cached_segment_base = emulator_get_cached_segment_base,
6194 .get_gdt = emulator_get_gdt,
6195 .get_idt = emulator_get_idt,
6196 .set_gdt = emulator_set_gdt,
6197 .set_idt = emulator_set_idt,
6198 .get_cr = emulator_get_cr,
6199 .set_cr = emulator_set_cr,
6200 .cpl = emulator_get_cpl,
6201 .get_dr = emulator_get_dr,
6202 .set_dr = emulator_set_dr,
6203 .get_smbase = emulator_get_smbase,
6204 .set_smbase = emulator_set_smbase,
6205 .set_msr = emulator_set_msr,
6206 .get_msr = emulator_get_msr,
6207 .check_pmc = emulator_check_pmc,
6208 .read_pmc = emulator_read_pmc,
6209 .halt = emulator_halt,
6210 .wbinvd = emulator_wbinvd,
6211 .fix_hypercall = emulator_fix_hypercall,
6212 .intercept = emulator_intercept,
6213 .get_cpuid = emulator_get_cpuid,
6214 .set_nmi_mask = emulator_set_nmi_mask,
6215 .get_hflags = emulator_get_hflags,
6216 .set_hflags = emulator_set_hflags,
6217 .pre_leave_smm = emulator_pre_leave_smm,
6218 .post_leave_smm = emulator_post_leave_smm,
6219 .set_xcr = emulator_set_xcr,
6222 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6224 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6226 * an sti; sti; sequence only disable interrupts for the first
6227 * instruction. So, if the last instruction, be it emulated or
6228 * not, left the system with the INT_STI flag enabled, it
6229 * means that the last instruction is an sti. We should not
6230 * leave the flag on in this case. The same goes for mov ss
6232 if (int_shadow & mask)
6234 if (unlikely(int_shadow || mask)) {
6235 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6237 kvm_make_request(KVM_REQ_EVENT, vcpu);
6241 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6243 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6244 if (ctxt->exception.vector == PF_VECTOR)
6245 return kvm_propagate_fault(vcpu, &ctxt->exception);
6247 if (ctxt->exception.error_code_valid)
6248 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6249 ctxt->exception.error_code);
6251 kvm_queue_exception(vcpu, ctxt->exception.vector);
6255 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6257 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6260 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6262 ctxt->eflags = kvm_get_rflags(vcpu);
6263 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6265 ctxt->eip = kvm_rip_read(vcpu);
6266 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6267 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6268 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6269 cs_db ? X86EMUL_MODE_PROT32 :
6270 X86EMUL_MODE_PROT16;
6271 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6272 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6273 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6275 init_decode_cache(ctxt);
6276 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6279 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6281 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6284 init_emulate_ctxt(vcpu);
6288 ctxt->_eip = ctxt->eip + inc_eip;
6289 ret = emulate_int_real(ctxt, irq);
6291 if (ret != X86EMUL_CONTINUE) {
6292 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6294 ctxt->eip = ctxt->_eip;
6295 kvm_rip_write(vcpu, ctxt->eip);
6296 kvm_set_rflags(vcpu, ctxt->eflags);
6299 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6301 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6303 ++vcpu->stat.insn_emulation_fail;
6304 trace_kvm_emulate_insn_failed(vcpu);
6306 if (emulation_type & EMULTYPE_VMWARE_GP) {
6307 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6311 if (emulation_type & EMULTYPE_SKIP) {
6312 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6313 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6314 vcpu->run->internal.ndata = 0;
6318 kvm_queue_exception(vcpu, UD_VECTOR);
6320 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6321 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6322 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6323 vcpu->run->internal.ndata = 0;
6330 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6331 bool write_fault_to_shadow_pgtable,
6337 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6340 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6343 if (!vcpu->arch.mmu->direct_map) {
6345 * Write permission should be allowed since only
6346 * write access need to be emulated.
6348 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6351 * If the mapping is invalid in guest, let cpu retry
6352 * it to generate fault.
6354 if (gpa == UNMAPPED_GVA)
6359 * Do not retry the unhandleable instruction if it faults on the
6360 * readonly host memory, otherwise it will goto a infinite loop:
6361 * retry instruction -> write #PF -> emulation fail -> retry
6362 * instruction -> ...
6364 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6367 * If the instruction failed on the error pfn, it can not be fixed,
6368 * report the error to userspace.
6370 if (is_error_noslot_pfn(pfn))
6373 kvm_release_pfn_clean(pfn);
6375 /* The instructions are well-emulated on direct mmu. */
6376 if (vcpu->arch.mmu->direct_map) {
6377 unsigned int indirect_shadow_pages;
6379 spin_lock(&vcpu->kvm->mmu_lock);
6380 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6381 spin_unlock(&vcpu->kvm->mmu_lock);
6383 if (indirect_shadow_pages)
6384 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6390 * if emulation was due to access to shadowed page table
6391 * and it failed try to unshadow page and re-enter the
6392 * guest to let CPU execute the instruction.
6394 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6397 * If the access faults on its page table, it can not
6398 * be fixed by unprotecting shadow page and it should
6399 * be reported to userspace.
6401 return !write_fault_to_shadow_pgtable;
6404 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6405 unsigned long cr2, int emulation_type)
6407 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6408 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6410 last_retry_eip = vcpu->arch.last_retry_eip;
6411 last_retry_addr = vcpu->arch.last_retry_addr;
6414 * If the emulation is caused by #PF and it is non-page_table
6415 * writing instruction, it means the VM-EXIT is caused by shadow
6416 * page protected, we can zap the shadow page and retry this
6417 * instruction directly.
6419 * Note: if the guest uses a non-page-table modifying instruction
6420 * on the PDE that points to the instruction, then we will unmap
6421 * the instruction and go to an infinite loop. So, we cache the
6422 * last retried eip and the last fault address, if we meet the eip
6423 * and the address again, we can break out of the potential infinite
6426 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6428 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6431 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6434 if (x86_page_table_writing_insn(ctxt))
6437 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6440 vcpu->arch.last_retry_eip = ctxt->eip;
6441 vcpu->arch.last_retry_addr = cr2;
6443 if (!vcpu->arch.mmu->direct_map)
6444 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6446 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6451 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6452 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6454 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6456 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6457 /* This is a good place to trace that we are exiting SMM. */
6458 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6460 /* Process a latched INIT or SMI, if any. */
6461 kvm_make_request(KVM_REQ_EVENT, vcpu);
6464 kvm_mmu_reset_context(vcpu);
6467 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6476 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6477 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6482 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6484 struct kvm_run *kvm_run = vcpu->run;
6486 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6487 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6488 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6489 kvm_run->debug.arch.exception = DB_VECTOR;
6490 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6493 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6497 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6499 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6502 r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6507 * rflags is the old, "raw" value of the flags. The new value has
6508 * not been saved yet.
6510 * This is correct even for TF set by the guest, because "the
6511 * processor will not generate this exception after the instruction
6512 * that sets the TF flag".
6514 if (unlikely(rflags & X86_EFLAGS_TF))
6515 r = kvm_vcpu_do_singlestep(vcpu);
6518 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6520 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6522 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6523 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6524 struct kvm_run *kvm_run = vcpu->run;
6525 unsigned long eip = kvm_get_linear_rip(vcpu);
6526 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6527 vcpu->arch.guest_debug_dr7,
6531 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6532 kvm_run->debug.arch.pc = eip;
6533 kvm_run->debug.arch.exception = DB_VECTOR;
6534 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6540 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6541 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6542 unsigned long eip = kvm_get_linear_rip(vcpu);
6543 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6548 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6549 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6550 kvm_queue_exception(vcpu, DB_VECTOR);
6559 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6561 switch (ctxt->opcode_len) {
6568 case 0xe6: /* OUT */
6572 case 0x6c: /* INS */
6574 case 0x6e: /* OUTS */
6581 case 0x33: /* RDPMC */
6590 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6597 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6598 bool writeback = true;
6599 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6601 vcpu->arch.l1tf_flush_l1d = true;
6604 * Clear write_fault_to_shadow_pgtable here to ensure it is
6607 vcpu->arch.write_fault_to_shadow_pgtable = false;
6608 kvm_clear_exception_queue(vcpu);
6610 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6611 init_emulate_ctxt(vcpu);
6614 * We will reenter on the same instruction since
6615 * we do not set complete_userspace_io. This does not
6616 * handle watchpoints yet, those would be handled in
6619 if (!(emulation_type & EMULTYPE_SKIP) &&
6620 kvm_vcpu_check_breakpoint(vcpu, &r))
6623 ctxt->interruptibility = 0;
6624 ctxt->have_exception = false;
6625 ctxt->exception.vector = -1;
6626 ctxt->perm_ok = false;
6628 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6630 r = x86_decode_insn(ctxt, insn, insn_len);
6632 trace_kvm_emulate_insn_start(vcpu);
6633 ++vcpu->stat.insn_emulation;
6634 if (r != EMULATION_OK) {
6635 if ((emulation_type & EMULTYPE_TRAP_UD) ||
6636 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6637 kvm_queue_exception(vcpu, UD_VECTOR);
6640 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6643 if (ctxt->have_exception) {
6645 * #UD should result in just EMULATION_FAILED, and trap-like
6646 * exception should not be encountered during decode.
6648 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6649 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6650 inject_emulated_exception(vcpu);
6653 return handle_emulation_failure(vcpu, emulation_type);
6657 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6658 !is_vmware_backdoor_opcode(ctxt)) {
6659 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6664 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6665 * for kvm_skip_emulated_instruction(). The caller is responsible for
6666 * updating interruptibility state and injecting single-step #DBs.
6668 if (emulation_type & EMULTYPE_SKIP) {
6669 kvm_rip_write(vcpu, ctxt->_eip);
6670 if (ctxt->eflags & X86_EFLAGS_RF)
6671 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6675 if (retry_instruction(ctxt, cr2, emulation_type))
6678 /* this is needed for vmware backdoor interface to work since it
6679 changes registers values during IO operation */
6680 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6681 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6682 emulator_invalidate_register_cache(ctxt);
6686 /* Save the faulting GPA (cr2) in the address field */
6687 ctxt->exception.address = cr2;
6689 r = x86_emulate_insn(ctxt);
6691 if (r == EMULATION_INTERCEPTED)
6694 if (r == EMULATION_FAILED) {
6695 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6699 return handle_emulation_failure(vcpu, emulation_type);
6702 if (ctxt->have_exception) {
6704 if (inject_emulated_exception(vcpu))
6706 } else if (vcpu->arch.pio.count) {
6707 if (!vcpu->arch.pio.in) {
6708 /* FIXME: return into emulator if single-stepping. */
6709 vcpu->arch.pio.count = 0;
6712 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6715 } else if (vcpu->mmio_needed) {
6716 ++vcpu->stat.mmio_exits;
6718 if (!vcpu->mmio_is_write)
6721 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6722 } else if (r == EMULATION_RESTART)
6728 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6729 toggle_interruptibility(vcpu, ctxt->interruptibility);
6730 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6731 if (!ctxt->have_exception ||
6732 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6733 kvm_rip_write(vcpu, ctxt->eip);
6735 r = kvm_vcpu_do_singlestep(vcpu);
6736 __kvm_set_rflags(vcpu, ctxt->eflags);
6740 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6741 * do nothing, and it will be requested again as soon as
6742 * the shadow expires. But we still need to check here,
6743 * because POPF has no interrupt shadow.
6745 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6746 kvm_make_request(KVM_REQ_EVENT, vcpu);
6748 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6753 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6755 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6757 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6759 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6760 void *insn, int insn_len)
6762 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6764 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6766 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6768 vcpu->arch.pio.count = 0;
6772 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6774 vcpu->arch.pio.count = 0;
6776 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6779 return kvm_skip_emulated_instruction(vcpu);
6782 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6783 unsigned short port)
6785 unsigned long val = kvm_rax_read(vcpu);
6786 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6787 size, port, &val, 1);
6792 * Workaround userspace that relies on old KVM behavior of %rip being
6793 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6796 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6797 vcpu->arch.complete_userspace_io =
6798 complete_fast_pio_out_port_0x7e;
6799 kvm_skip_emulated_instruction(vcpu);
6801 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6802 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6807 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6811 /* We should only ever be called with arch.pio.count equal to 1 */
6812 BUG_ON(vcpu->arch.pio.count != 1);
6814 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6815 vcpu->arch.pio.count = 0;
6819 /* For size less than 4 we merge, else we zero extend */
6820 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6823 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6824 * the copy and tracing
6826 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6827 vcpu->arch.pio.port, &val, 1);
6828 kvm_rax_write(vcpu, val);
6830 return kvm_skip_emulated_instruction(vcpu);
6833 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6834 unsigned short port)
6839 /* For size less than 4 we merge, else we zero extend */
6840 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6842 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6845 kvm_rax_write(vcpu, val);
6849 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6850 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6855 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6860 ret = kvm_fast_pio_in(vcpu, size, port);
6862 ret = kvm_fast_pio_out(vcpu, size, port);
6863 return ret && kvm_skip_emulated_instruction(vcpu);
6865 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6867 static int kvmclock_cpu_down_prep(unsigned int cpu)
6869 __this_cpu_write(cpu_tsc_khz, 0);
6873 static void tsc_khz_changed(void *data)
6875 struct cpufreq_freqs *freq = data;
6876 unsigned long khz = 0;
6880 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6881 khz = cpufreq_quick_get(raw_smp_processor_id());
6884 __this_cpu_write(cpu_tsc_khz, khz);
6887 #ifdef CONFIG_X86_64
6888 static void kvm_hyperv_tsc_notifier(void)
6891 struct kvm_vcpu *vcpu;
6894 mutex_lock(&kvm_lock);
6895 list_for_each_entry(kvm, &vm_list, vm_list)
6896 kvm_make_mclock_inprogress_request(kvm);
6898 hyperv_stop_tsc_emulation();
6900 /* TSC frequency always matches when on Hyper-V */
6901 for_each_present_cpu(cpu)
6902 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6903 kvm_max_guest_tsc_khz = tsc_khz;
6905 list_for_each_entry(kvm, &vm_list, vm_list) {
6906 struct kvm_arch *ka = &kvm->arch;
6908 spin_lock(&ka->pvclock_gtod_sync_lock);
6910 pvclock_update_vm_gtod_copy(kvm);
6912 kvm_for_each_vcpu(cpu, vcpu, kvm)
6913 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6915 kvm_for_each_vcpu(cpu, vcpu, kvm)
6916 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6918 spin_unlock(&ka->pvclock_gtod_sync_lock);
6920 mutex_unlock(&kvm_lock);
6924 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6927 struct kvm_vcpu *vcpu;
6928 int i, send_ipi = 0;
6931 * We allow guests to temporarily run on slowing clocks,
6932 * provided we notify them after, or to run on accelerating
6933 * clocks, provided we notify them before. Thus time never
6936 * However, we have a problem. We can't atomically update
6937 * the frequency of a given CPU from this function; it is
6938 * merely a notifier, which can be called from any CPU.
6939 * Changing the TSC frequency at arbitrary points in time
6940 * requires a recomputation of local variables related to
6941 * the TSC for each VCPU. We must flag these local variables
6942 * to be updated and be sure the update takes place with the
6943 * new frequency before any guests proceed.
6945 * Unfortunately, the combination of hotplug CPU and frequency
6946 * change creates an intractable locking scenario; the order
6947 * of when these callouts happen is undefined with respect to
6948 * CPU hotplug, and they can race with each other. As such,
6949 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6950 * undefined; you can actually have a CPU frequency change take
6951 * place in between the computation of X and the setting of the
6952 * variable. To protect against this problem, all updates of
6953 * the per_cpu tsc_khz variable are done in an interrupt
6954 * protected IPI, and all callers wishing to update the value
6955 * must wait for a synchronous IPI to complete (which is trivial
6956 * if the caller is on the CPU already). This establishes the
6957 * necessary total order on variable updates.
6959 * Note that because a guest time update may take place
6960 * anytime after the setting of the VCPU's request bit, the
6961 * correct TSC value must be set before the request. However,
6962 * to ensure the update actually makes it to any guest which
6963 * starts running in hardware virtualization between the set
6964 * and the acquisition of the spinlock, we must also ping the
6965 * CPU after setting the request bit.
6969 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6971 mutex_lock(&kvm_lock);
6972 list_for_each_entry(kvm, &vm_list, vm_list) {
6973 kvm_for_each_vcpu(i, vcpu, kvm) {
6974 if (vcpu->cpu != cpu)
6976 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6977 if (vcpu->cpu != raw_smp_processor_id())
6981 mutex_unlock(&kvm_lock);
6983 if (freq->old < freq->new && send_ipi) {
6985 * We upscale the frequency. Must make the guest
6986 * doesn't see old kvmclock values while running with
6987 * the new frequency, otherwise we risk the guest sees
6988 * time go backwards.
6990 * In case we update the frequency for another cpu
6991 * (which might be in guest context) send an interrupt
6992 * to kick the cpu out of guest context. Next time
6993 * guest context is entered kvmclock will be updated,
6994 * so the guest will not see stale values.
6996 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7000 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7003 struct cpufreq_freqs *freq = data;
7006 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7008 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7011 for_each_cpu(cpu, freq->policy->cpus)
7012 __kvmclock_cpufreq_notifier(freq, cpu);
7017 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7018 .notifier_call = kvmclock_cpufreq_notifier
7021 static int kvmclock_cpu_online(unsigned int cpu)
7023 tsc_khz_changed(NULL);
7027 static void kvm_timer_init(void)
7029 max_tsc_khz = tsc_khz;
7031 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7032 #ifdef CONFIG_CPU_FREQ
7033 struct cpufreq_policy policy;
7036 memset(&policy, 0, sizeof(policy));
7038 cpufreq_get_policy(&policy, cpu);
7039 if (policy.cpuinfo.max_freq)
7040 max_tsc_khz = policy.cpuinfo.max_freq;
7043 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7044 CPUFREQ_TRANSITION_NOTIFIER);
7047 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7048 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7051 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7052 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7054 int kvm_is_in_guest(void)
7056 return __this_cpu_read(current_vcpu) != NULL;
7059 static int kvm_is_user_mode(void)
7063 if (__this_cpu_read(current_vcpu))
7064 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7066 return user_mode != 0;
7069 static unsigned long kvm_get_guest_ip(void)
7071 unsigned long ip = 0;
7073 if (__this_cpu_read(current_vcpu))
7074 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7079 static void kvm_handle_intel_pt_intr(void)
7081 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7083 kvm_make_request(KVM_REQ_PMI, vcpu);
7084 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7085 (unsigned long *)&vcpu->arch.pmu.global_status);
7088 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7089 .is_in_guest = kvm_is_in_guest,
7090 .is_user_mode = kvm_is_user_mode,
7091 .get_guest_ip = kvm_get_guest_ip,
7092 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7095 #ifdef CONFIG_X86_64
7096 static void pvclock_gtod_update_fn(struct work_struct *work)
7100 struct kvm_vcpu *vcpu;
7103 mutex_lock(&kvm_lock);
7104 list_for_each_entry(kvm, &vm_list, vm_list)
7105 kvm_for_each_vcpu(i, vcpu, kvm)
7106 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7107 atomic_set(&kvm_guest_has_master_clock, 0);
7108 mutex_unlock(&kvm_lock);
7111 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7114 * Notification about pvclock gtod data update.
7116 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7119 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7120 struct timekeeper *tk = priv;
7122 update_pvclock_gtod(tk);
7124 /* disable master clock if host does not trust, or does not
7125 * use, TSC based clocksource.
7127 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7128 atomic_read(&kvm_guest_has_master_clock) != 0)
7129 queue_work(system_long_wq, &pvclock_gtod_work);
7134 static struct notifier_block pvclock_gtod_notifier = {
7135 .notifier_call = pvclock_gtod_notify,
7139 int kvm_arch_init(void *opaque)
7142 struct kvm_x86_ops *ops = opaque;
7145 printk(KERN_ERR "kvm: already loaded the other module\n");
7150 if (!ops->cpu_has_kvm_support()) {
7151 printk(KERN_ERR "kvm: no hardware support\n");
7155 if (ops->disabled_by_bios()) {
7156 printk(KERN_ERR "kvm: disabled by bios\n");
7162 * KVM explicitly assumes that the guest has an FPU and
7163 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7164 * vCPU's FPU state as a fxregs_state struct.
7166 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7167 printk(KERN_ERR "kvm: inadequate fpu\n");
7173 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7174 __alignof__(struct fpu), SLAB_ACCOUNT,
7176 if (!x86_fpu_cache) {
7177 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7181 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7183 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7184 goto out_free_x86_fpu_cache;
7187 r = kvm_mmu_module_init();
7189 goto out_free_percpu;
7193 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7194 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7195 PT_PRESENT_MASK, 0, sme_me_mask);
7198 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7200 if (boot_cpu_has(X86_FEATURE_XSAVE))
7201 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7204 if (pi_inject_timer == -1)
7205 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7206 #ifdef CONFIG_X86_64
7207 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7209 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7210 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7216 free_percpu(shared_msrs);
7217 out_free_x86_fpu_cache:
7218 kmem_cache_destroy(x86_fpu_cache);
7223 void kvm_arch_exit(void)
7225 #ifdef CONFIG_X86_64
7226 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7227 clear_hv_tscchange_cb();
7230 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7232 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7233 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7234 CPUFREQ_TRANSITION_NOTIFIER);
7235 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7236 #ifdef CONFIG_X86_64
7237 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7240 kvm_mmu_module_exit();
7241 free_percpu(shared_msrs);
7242 kmem_cache_destroy(x86_fpu_cache);
7245 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7247 ++vcpu->stat.halt_exits;
7248 if (lapic_in_kernel(vcpu)) {
7249 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7252 vcpu->run->exit_reason = KVM_EXIT_HLT;
7256 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7258 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7260 int ret = kvm_skip_emulated_instruction(vcpu);
7262 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7263 * KVM_EXIT_DEBUG here.
7265 return kvm_vcpu_halt(vcpu) && ret;
7267 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7269 #ifdef CONFIG_X86_64
7270 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7271 unsigned long clock_type)
7273 struct kvm_clock_pairing clock_pairing;
7274 struct timespec64 ts;
7278 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7279 return -KVM_EOPNOTSUPP;
7281 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7282 return -KVM_EOPNOTSUPP;
7284 clock_pairing.sec = ts.tv_sec;
7285 clock_pairing.nsec = ts.tv_nsec;
7286 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7287 clock_pairing.flags = 0;
7288 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7291 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7292 sizeof(struct kvm_clock_pairing)))
7300 * kvm_pv_kick_cpu_op: Kick a vcpu.
7302 * @apicid - apicid of vcpu to be kicked.
7304 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7306 struct kvm_lapic_irq lapic_irq;
7308 lapic_irq.shorthand = 0;
7309 lapic_irq.dest_mode = 0;
7310 lapic_irq.level = 0;
7311 lapic_irq.dest_id = apicid;
7312 lapic_irq.msi_redir_hint = false;
7314 lapic_irq.delivery_mode = APIC_DM_REMRD;
7315 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7318 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7320 if (!lapic_in_kernel(vcpu)) {
7321 WARN_ON_ONCE(vcpu->arch.apicv_active);
7324 if (!vcpu->arch.apicv_active)
7327 vcpu->arch.apicv_active = false;
7328 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7331 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7333 struct kvm_vcpu *target = NULL;
7334 struct kvm_apic_map *map;
7337 map = rcu_dereference(kvm->arch.apic_map);
7339 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7340 target = map->phys_map[dest_id]->vcpu;
7344 if (target && READ_ONCE(target->ready))
7345 kvm_vcpu_yield_to(target);
7348 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7350 unsigned long nr, a0, a1, a2, a3, ret;
7353 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7354 return kvm_hv_hypercall(vcpu);
7356 nr = kvm_rax_read(vcpu);
7357 a0 = kvm_rbx_read(vcpu);
7358 a1 = kvm_rcx_read(vcpu);
7359 a2 = kvm_rdx_read(vcpu);
7360 a3 = kvm_rsi_read(vcpu);
7362 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7364 op_64_bit = is_64_bit_mode(vcpu);
7373 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7379 case KVM_HC_VAPIC_POLL_IRQ:
7382 case KVM_HC_KICK_CPU:
7383 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7384 kvm_sched_yield(vcpu->kvm, a1);
7387 #ifdef CONFIG_X86_64
7388 case KVM_HC_CLOCK_PAIRING:
7389 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7392 case KVM_HC_SEND_IPI:
7393 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7395 case KVM_HC_SCHED_YIELD:
7396 kvm_sched_yield(vcpu->kvm, a0);
7406 kvm_rax_write(vcpu, ret);
7408 ++vcpu->stat.hypercalls;
7409 return kvm_skip_emulated_instruction(vcpu);
7411 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7413 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7415 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7416 char instruction[3];
7417 unsigned long rip = kvm_rip_read(vcpu);
7419 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7421 return emulator_write_emulated(ctxt, rip, instruction, 3,
7425 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7427 return vcpu->run->request_interrupt_window &&
7428 likely(!pic_in_kernel(vcpu->kvm));
7431 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7433 struct kvm_run *kvm_run = vcpu->run;
7435 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7436 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7437 kvm_run->cr8 = kvm_get_cr8(vcpu);
7438 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7439 kvm_run->ready_for_interrupt_injection =
7440 pic_in_kernel(vcpu->kvm) ||
7441 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7444 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7448 if (!kvm_x86_ops->update_cr8_intercept)
7451 if (!lapic_in_kernel(vcpu))
7454 if (vcpu->arch.apicv_active)
7457 if (!vcpu->arch.apic->vapic_addr)
7458 max_irr = kvm_lapic_find_highest_irr(vcpu);
7465 tpr = kvm_lapic_get_cr8(vcpu);
7467 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7470 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7474 /* try to reinject previous events if any */
7476 if (vcpu->arch.exception.injected)
7477 kvm_x86_ops->queue_exception(vcpu);
7479 * Do not inject an NMI or interrupt if there is a pending
7480 * exception. Exceptions and interrupts are recognized at
7481 * instruction boundaries, i.e. the start of an instruction.
7482 * Trap-like exceptions, e.g. #DB, have higher priority than
7483 * NMIs and interrupts, i.e. traps are recognized before an
7484 * NMI/interrupt that's pending on the same instruction.
7485 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7486 * priority, but are only generated (pended) during instruction
7487 * execution, i.e. a pending fault-like exception means the
7488 * fault occurred on the *previous* instruction and must be
7489 * serviced prior to recognizing any new events in order to
7490 * fully complete the previous instruction.
7492 else if (!vcpu->arch.exception.pending) {
7493 if (vcpu->arch.nmi_injected)
7494 kvm_x86_ops->set_nmi(vcpu);
7495 else if (vcpu->arch.interrupt.injected)
7496 kvm_x86_ops->set_irq(vcpu);
7500 * Call check_nested_events() even if we reinjected a previous event
7501 * in order for caller to determine if it should require immediate-exit
7502 * from L2 to L1 due to pending L1 events which require exit
7505 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7506 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7511 /* try to inject new event if pending */
7512 if (vcpu->arch.exception.pending) {
7513 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7514 vcpu->arch.exception.has_error_code,
7515 vcpu->arch.exception.error_code);
7517 WARN_ON_ONCE(vcpu->arch.exception.injected);
7518 vcpu->arch.exception.pending = false;
7519 vcpu->arch.exception.injected = true;
7521 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7522 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7525 if (vcpu->arch.exception.nr == DB_VECTOR) {
7527 * This code assumes that nSVM doesn't use
7528 * check_nested_events(). If it does, the
7529 * DR6/DR7 changes should happen before L1
7530 * gets a #VMEXIT for an intercepted #DB in
7531 * L2. (Under VMX, on the other hand, the
7532 * DR6/DR7 changes should not happen in the
7533 * event of a VM-exit to L1 for an intercepted
7536 kvm_deliver_exception_payload(vcpu);
7537 if (vcpu->arch.dr7 & DR7_GD) {
7538 vcpu->arch.dr7 &= ~DR7_GD;
7539 kvm_update_dr7(vcpu);
7543 kvm_x86_ops->queue_exception(vcpu);
7546 /* Don't consider new event if we re-injected an event */
7547 if (kvm_event_needs_reinjection(vcpu))
7550 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7551 kvm_x86_ops->smi_allowed(vcpu)) {
7552 vcpu->arch.smi_pending = false;
7553 ++vcpu->arch.smi_count;
7555 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7556 --vcpu->arch.nmi_pending;
7557 vcpu->arch.nmi_injected = true;
7558 kvm_x86_ops->set_nmi(vcpu);
7559 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7561 * Because interrupts can be injected asynchronously, we are
7562 * calling check_nested_events again here to avoid a race condition.
7563 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7564 * proposal and current concerns. Perhaps we should be setting
7565 * KVM_REQ_EVENT only on certain events and not unconditionally?
7567 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7568 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7572 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7573 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7575 kvm_x86_ops->set_irq(vcpu);
7582 static void process_nmi(struct kvm_vcpu *vcpu)
7587 * x86 is limited to one NMI running, and one NMI pending after it.
7588 * If an NMI is already in progress, limit further NMIs to just one.
7589 * Otherwise, allow two (and we'll inject the first one immediately).
7591 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7594 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7595 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7596 kvm_make_request(KVM_REQ_EVENT, vcpu);
7599 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7602 flags |= seg->g << 23;
7603 flags |= seg->db << 22;
7604 flags |= seg->l << 21;
7605 flags |= seg->avl << 20;
7606 flags |= seg->present << 15;
7607 flags |= seg->dpl << 13;
7608 flags |= seg->s << 12;
7609 flags |= seg->type << 8;
7613 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7615 struct kvm_segment seg;
7618 kvm_get_segment(vcpu, &seg, n);
7619 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7622 offset = 0x7f84 + n * 12;
7624 offset = 0x7f2c + (n - 3) * 12;
7626 put_smstate(u32, buf, offset + 8, seg.base);
7627 put_smstate(u32, buf, offset + 4, seg.limit);
7628 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7631 #ifdef CONFIG_X86_64
7632 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7634 struct kvm_segment seg;
7638 kvm_get_segment(vcpu, &seg, n);
7639 offset = 0x7e00 + n * 16;
7641 flags = enter_smm_get_segment_flags(&seg) >> 8;
7642 put_smstate(u16, buf, offset, seg.selector);
7643 put_smstate(u16, buf, offset + 2, flags);
7644 put_smstate(u32, buf, offset + 4, seg.limit);
7645 put_smstate(u64, buf, offset + 8, seg.base);
7649 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7652 struct kvm_segment seg;
7656 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7657 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7658 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7659 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7661 for (i = 0; i < 8; i++)
7662 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7664 kvm_get_dr(vcpu, 6, &val);
7665 put_smstate(u32, buf, 0x7fcc, (u32)val);
7666 kvm_get_dr(vcpu, 7, &val);
7667 put_smstate(u32, buf, 0x7fc8, (u32)val);
7669 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7670 put_smstate(u32, buf, 0x7fc4, seg.selector);
7671 put_smstate(u32, buf, 0x7f64, seg.base);
7672 put_smstate(u32, buf, 0x7f60, seg.limit);
7673 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7675 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7676 put_smstate(u32, buf, 0x7fc0, seg.selector);
7677 put_smstate(u32, buf, 0x7f80, seg.base);
7678 put_smstate(u32, buf, 0x7f7c, seg.limit);
7679 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7681 kvm_x86_ops->get_gdt(vcpu, &dt);
7682 put_smstate(u32, buf, 0x7f74, dt.address);
7683 put_smstate(u32, buf, 0x7f70, dt.size);
7685 kvm_x86_ops->get_idt(vcpu, &dt);
7686 put_smstate(u32, buf, 0x7f58, dt.address);
7687 put_smstate(u32, buf, 0x7f54, dt.size);
7689 for (i = 0; i < 6; i++)
7690 enter_smm_save_seg_32(vcpu, buf, i);
7692 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7695 put_smstate(u32, buf, 0x7efc, 0x00020000);
7696 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7699 #ifdef CONFIG_X86_64
7700 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7703 struct kvm_segment seg;
7707 for (i = 0; i < 16; i++)
7708 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7710 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7711 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7713 kvm_get_dr(vcpu, 6, &val);
7714 put_smstate(u64, buf, 0x7f68, val);
7715 kvm_get_dr(vcpu, 7, &val);
7716 put_smstate(u64, buf, 0x7f60, val);
7718 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7719 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7720 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7722 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7725 put_smstate(u32, buf, 0x7efc, 0x00020064);
7727 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7729 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7730 put_smstate(u16, buf, 0x7e90, seg.selector);
7731 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7732 put_smstate(u32, buf, 0x7e94, seg.limit);
7733 put_smstate(u64, buf, 0x7e98, seg.base);
7735 kvm_x86_ops->get_idt(vcpu, &dt);
7736 put_smstate(u32, buf, 0x7e84, dt.size);
7737 put_smstate(u64, buf, 0x7e88, dt.address);
7739 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7740 put_smstate(u16, buf, 0x7e70, seg.selector);
7741 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7742 put_smstate(u32, buf, 0x7e74, seg.limit);
7743 put_smstate(u64, buf, 0x7e78, seg.base);
7745 kvm_x86_ops->get_gdt(vcpu, &dt);
7746 put_smstate(u32, buf, 0x7e64, dt.size);
7747 put_smstate(u64, buf, 0x7e68, dt.address);
7749 for (i = 0; i < 6; i++)
7750 enter_smm_save_seg_64(vcpu, buf, i);
7754 static void enter_smm(struct kvm_vcpu *vcpu)
7756 struct kvm_segment cs, ds;
7761 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7762 memset(buf, 0, 512);
7763 #ifdef CONFIG_X86_64
7764 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7765 enter_smm_save_state_64(vcpu, buf);
7768 enter_smm_save_state_32(vcpu, buf);
7771 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7772 * vCPU state (e.g. leave guest mode) after we've saved the state into
7773 * the SMM state-save area.
7775 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7777 vcpu->arch.hflags |= HF_SMM_MASK;
7778 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7780 if (kvm_x86_ops->get_nmi_mask(vcpu))
7781 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7783 kvm_x86_ops->set_nmi_mask(vcpu, true);
7785 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7786 kvm_rip_write(vcpu, 0x8000);
7788 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7789 kvm_x86_ops->set_cr0(vcpu, cr0);
7790 vcpu->arch.cr0 = cr0;
7792 kvm_x86_ops->set_cr4(vcpu, 0);
7794 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7795 dt.address = dt.size = 0;
7796 kvm_x86_ops->set_idt(vcpu, &dt);
7798 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7800 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7801 cs.base = vcpu->arch.smbase;
7806 cs.limit = ds.limit = 0xffffffff;
7807 cs.type = ds.type = 0x3;
7808 cs.dpl = ds.dpl = 0;
7813 cs.avl = ds.avl = 0;
7814 cs.present = ds.present = 1;
7815 cs.unusable = ds.unusable = 0;
7816 cs.padding = ds.padding = 0;
7818 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7819 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7820 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7821 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7822 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7823 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7825 #ifdef CONFIG_X86_64
7826 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7827 kvm_x86_ops->set_efer(vcpu, 0);
7830 kvm_update_cpuid(vcpu);
7831 kvm_mmu_reset_context(vcpu);
7834 static void process_smi(struct kvm_vcpu *vcpu)
7836 vcpu->arch.smi_pending = true;
7837 kvm_make_request(KVM_REQ_EVENT, vcpu);
7840 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7842 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7845 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7847 if (!kvm_apic_present(vcpu))
7850 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7852 if (irqchip_split(vcpu->kvm))
7853 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7855 if (vcpu->arch.apicv_active)
7856 kvm_x86_ops->sync_pir_to_irr(vcpu);
7857 if (ioapic_in_kernel(vcpu->kvm))
7858 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7861 if (is_guest_mode(vcpu))
7862 vcpu->arch.load_eoi_exitmap_pending = true;
7864 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7867 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7869 u64 eoi_exit_bitmap[4];
7871 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7874 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7875 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7876 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7879 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7880 unsigned long start, unsigned long end,
7883 unsigned long apic_address;
7886 * The physical address of apic access page is stored in the VMCS.
7887 * Update it when it becomes invalid.
7889 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7890 if (start <= apic_address && apic_address < end)
7891 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7896 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7898 struct page *page = NULL;
7900 if (!lapic_in_kernel(vcpu))
7903 if (!kvm_x86_ops->set_apic_access_page_addr)
7906 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7907 if (is_error_page(page))
7909 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7912 * Do not pin apic access page in memory, the MMU notifier
7913 * will call us again if it is migrated or swapped out.
7917 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7919 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7921 smp_send_reschedule(vcpu->cpu);
7923 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7926 * Returns 1 to let vcpu_run() continue the guest execution loop without
7927 * exiting to the userspace. Otherwise, the value will be returned to the
7930 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7934 dm_request_for_irq_injection(vcpu) &&
7935 kvm_cpu_accept_dm_intr(vcpu);
7937 bool req_immediate_exit = false;
7939 if (kvm_request_pending(vcpu)) {
7940 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7941 kvm_x86_ops->get_vmcs12_pages(vcpu);
7942 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7943 kvm_mmu_unload(vcpu);
7944 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7945 __kvm_migrate_timers(vcpu);
7946 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7947 kvm_gen_update_masterclock(vcpu->kvm);
7948 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7949 kvm_gen_kvmclock_update(vcpu);
7950 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7951 r = kvm_guest_time_update(vcpu);
7955 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7956 kvm_mmu_sync_roots(vcpu);
7957 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7958 kvm_mmu_load_cr3(vcpu);
7959 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7960 kvm_vcpu_flush_tlb(vcpu, true);
7961 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7962 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7966 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7967 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7968 vcpu->mmio_needed = 0;
7972 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7973 /* Page is swapped out. Do synthetic halt */
7974 vcpu->arch.apf.halted = true;
7978 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7979 record_steal_time(vcpu);
7980 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7982 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7984 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7985 kvm_pmu_handle_event(vcpu);
7986 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7987 kvm_pmu_deliver_pmi(vcpu);
7988 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7989 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7990 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7991 vcpu->arch.ioapic_handled_vectors)) {
7992 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7993 vcpu->run->eoi.vector =
7994 vcpu->arch.pending_ioapic_eoi;
7999 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8000 vcpu_scan_ioapic(vcpu);
8001 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8002 vcpu_load_eoi_exitmap(vcpu);
8003 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8004 kvm_vcpu_reload_apic_access_page(vcpu);
8005 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8006 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8007 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8011 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8012 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8013 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8017 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8018 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8019 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8025 * KVM_REQ_HV_STIMER has to be processed after
8026 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8027 * depend on the guest clock being up-to-date
8029 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8030 kvm_hv_process_stimers(vcpu);
8033 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8034 ++vcpu->stat.req_event;
8035 kvm_apic_accept_events(vcpu);
8036 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8041 if (inject_pending_event(vcpu, req_int_win) != 0)
8042 req_immediate_exit = true;
8044 /* Enable SMI/NMI/IRQ window open exits if needed.
8046 * SMIs have three cases:
8047 * 1) They can be nested, and then there is nothing to
8048 * do here because RSM will cause a vmexit anyway.
8049 * 2) There is an ISA-specific reason why SMI cannot be
8050 * injected, and the moment when this changes can be
8052 * 3) Or the SMI can be pending because
8053 * inject_pending_event has completed the injection
8054 * of an IRQ or NMI from the previous vmexit, and
8055 * then we request an immediate exit to inject the
8058 if (vcpu->arch.smi_pending && !is_smm(vcpu))
8059 if (!kvm_x86_ops->enable_smi_window(vcpu))
8060 req_immediate_exit = true;
8061 if (vcpu->arch.nmi_pending)
8062 kvm_x86_ops->enable_nmi_window(vcpu);
8063 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8064 kvm_x86_ops->enable_irq_window(vcpu);
8065 WARN_ON(vcpu->arch.exception.pending);
8068 if (kvm_lapic_enabled(vcpu)) {
8069 update_cr8_intercept(vcpu);
8070 kvm_lapic_sync_to_vapic(vcpu);
8074 r = kvm_mmu_reload(vcpu);
8076 goto cancel_injection;
8081 kvm_x86_ops->prepare_guest_switch(vcpu);
8084 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8085 * IPI are then delayed after guest entry, which ensures that they
8086 * result in virtual interrupt delivery.
8088 local_irq_disable();
8089 vcpu->mode = IN_GUEST_MODE;
8091 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8094 * 1) We should set ->mode before checking ->requests. Please see
8095 * the comment in kvm_vcpu_exiting_guest_mode().
8097 * 2) For APICv, we should set ->mode before checking PID.ON. This
8098 * pairs with the memory barrier implicit in pi_test_and_set_on
8099 * (see vmx_deliver_posted_interrupt).
8101 * 3) This also orders the write to mode from any reads to the page
8102 * tables done while the VCPU is running. Please see the comment
8103 * in kvm_flush_remote_tlbs.
8105 smp_mb__after_srcu_read_unlock();
8108 * This handles the case where a posted interrupt was
8109 * notified with kvm_vcpu_kick.
8111 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8112 kvm_x86_ops->sync_pir_to_irr(vcpu);
8114 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8115 || need_resched() || signal_pending(current)) {
8116 vcpu->mode = OUTSIDE_GUEST_MODE;
8120 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8122 goto cancel_injection;
8125 if (req_immediate_exit) {
8126 kvm_make_request(KVM_REQ_EVENT, vcpu);
8127 kvm_x86_ops->request_immediate_exit(vcpu);
8130 trace_kvm_entry(vcpu->vcpu_id);
8131 guest_enter_irqoff();
8133 /* The preempt notifier should have taken care of the FPU already. */
8134 WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8136 if (unlikely(vcpu->arch.switch_db_regs)) {
8138 set_debugreg(vcpu->arch.eff_db[0], 0);
8139 set_debugreg(vcpu->arch.eff_db[1], 1);
8140 set_debugreg(vcpu->arch.eff_db[2], 2);
8141 set_debugreg(vcpu->arch.eff_db[3], 3);
8142 set_debugreg(vcpu->arch.dr6, 6);
8143 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8146 kvm_x86_ops->run(vcpu);
8149 * Do this here before restoring debug registers on the host. And
8150 * since we do this before handling the vmexit, a DR access vmexit
8151 * can (a) read the correct value of the debug registers, (b) set
8152 * KVM_DEBUGREG_WONT_EXIT again.
8154 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8155 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8156 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8157 kvm_update_dr0123(vcpu);
8158 kvm_update_dr6(vcpu);
8159 kvm_update_dr7(vcpu);
8160 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8164 * If the guest has used debug registers, at least dr7
8165 * will be disabled while returning to the host.
8166 * If we don't have active breakpoints in the host, we don't
8167 * care about the messed up debug address registers. But if
8168 * we have some of them active, restore the old state.
8170 if (hw_breakpoint_active())
8171 hw_breakpoint_restore();
8173 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8175 vcpu->mode = OUTSIDE_GUEST_MODE;
8178 kvm_x86_ops->handle_exit_irqoff(vcpu);
8181 * Consume any pending interrupts, including the possible source of
8182 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8183 * An instruction is required after local_irq_enable() to fully unblock
8184 * interrupts on processors that implement an interrupt shadow, the
8185 * stat.exits increment will do nicely.
8187 kvm_before_interrupt(vcpu);
8190 local_irq_disable();
8191 kvm_after_interrupt(vcpu);
8193 guest_exit_irqoff();
8194 if (lapic_in_kernel(vcpu)) {
8195 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8196 if (delta != S64_MIN) {
8197 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8198 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8205 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8208 * Profile KVM exit RIPs:
8210 if (unlikely(prof_on == KVM_PROFILING)) {
8211 unsigned long rip = kvm_rip_read(vcpu);
8212 profile_hit(KVM_PROFILING, (void *)rip);
8215 if (unlikely(vcpu->arch.tsc_always_catchup))
8216 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8218 if (vcpu->arch.apic_attention)
8219 kvm_lapic_sync_from_vapic(vcpu);
8221 vcpu->arch.gpa_available = false;
8222 r = kvm_x86_ops->handle_exit(vcpu);
8226 kvm_x86_ops->cancel_injection(vcpu);
8227 if (unlikely(vcpu->arch.apic_attention))
8228 kvm_lapic_sync_from_vapic(vcpu);
8233 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8235 if (!kvm_arch_vcpu_runnable(vcpu) &&
8236 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8237 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8238 kvm_vcpu_block(vcpu);
8239 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8241 if (kvm_x86_ops->post_block)
8242 kvm_x86_ops->post_block(vcpu);
8244 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8248 kvm_apic_accept_events(vcpu);
8249 switch(vcpu->arch.mp_state) {
8250 case KVM_MP_STATE_HALTED:
8251 vcpu->arch.pv.pv_unhalted = false;
8252 vcpu->arch.mp_state =
8253 KVM_MP_STATE_RUNNABLE;
8255 case KVM_MP_STATE_RUNNABLE:
8256 vcpu->arch.apf.halted = false;
8258 case KVM_MP_STATE_INIT_RECEIVED:
8267 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8269 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8270 kvm_x86_ops->check_nested_events(vcpu, false);
8272 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8273 !vcpu->arch.apf.halted);
8276 static int vcpu_run(struct kvm_vcpu *vcpu)
8279 struct kvm *kvm = vcpu->kvm;
8281 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8282 vcpu->arch.l1tf_flush_l1d = true;
8285 if (kvm_vcpu_running(vcpu)) {
8286 r = vcpu_enter_guest(vcpu);
8288 r = vcpu_block(kvm, vcpu);
8294 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8295 if (kvm_cpu_has_pending_timer(vcpu))
8296 kvm_inject_pending_timer_irqs(vcpu);
8298 if (dm_request_for_irq_injection(vcpu) &&
8299 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8301 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8302 ++vcpu->stat.request_irq_exits;
8306 kvm_check_async_pf_completion(vcpu);
8308 if (signal_pending(current)) {
8310 vcpu->run->exit_reason = KVM_EXIT_INTR;
8311 ++vcpu->stat.signal_exits;
8314 if (need_resched()) {
8315 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8317 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8321 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8326 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8330 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8331 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8332 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8336 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8338 BUG_ON(!vcpu->arch.pio.count);
8340 return complete_emulated_io(vcpu);
8344 * Implements the following, as a state machine:
8348 * for each mmio piece in the fragment
8356 * for each mmio piece in the fragment
8361 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8363 struct kvm_run *run = vcpu->run;
8364 struct kvm_mmio_fragment *frag;
8367 BUG_ON(!vcpu->mmio_needed);
8369 /* Complete previous fragment */
8370 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8371 len = min(8u, frag->len);
8372 if (!vcpu->mmio_is_write)
8373 memcpy(frag->data, run->mmio.data, len);
8375 if (frag->len <= 8) {
8376 /* Switch to the next fragment. */
8378 vcpu->mmio_cur_fragment++;
8380 /* Go forward to the next mmio piece. */
8386 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8387 vcpu->mmio_needed = 0;
8389 /* FIXME: return into emulator if single-stepping. */
8390 if (vcpu->mmio_is_write)
8392 vcpu->mmio_read_completed = 1;
8393 return complete_emulated_io(vcpu);
8396 run->exit_reason = KVM_EXIT_MMIO;
8397 run->mmio.phys_addr = frag->gpa;
8398 if (vcpu->mmio_is_write)
8399 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8400 run->mmio.len = min(8u, frag->len);
8401 run->mmio.is_write = vcpu->mmio_is_write;
8402 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8406 /* Swap (qemu) user FPU context for the guest FPU context. */
8407 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8411 copy_fpregs_to_fpstate(vcpu->arch.user_fpu);
8412 /* PKRU is separately restored in kvm_x86_ops->run. */
8413 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8414 ~XFEATURE_MASK_PKRU);
8416 fpregs_mark_activate();
8422 /* When vcpu_run ends, restore user space FPU context. */
8423 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8427 copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8428 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8430 fpregs_mark_activate();
8433 ++vcpu->stat.fpu_reload;
8437 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8442 kvm_sigset_activate(vcpu);
8443 kvm_load_guest_fpu(vcpu);
8445 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8446 if (kvm_run->immediate_exit) {
8450 kvm_vcpu_block(vcpu);
8451 kvm_apic_accept_events(vcpu);
8452 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8454 if (signal_pending(current)) {
8456 vcpu->run->exit_reason = KVM_EXIT_INTR;
8457 ++vcpu->stat.signal_exits;
8462 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8467 if (vcpu->run->kvm_dirty_regs) {
8468 r = sync_regs(vcpu);
8473 /* re-sync apic's tpr */
8474 if (!lapic_in_kernel(vcpu)) {
8475 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8481 if (unlikely(vcpu->arch.complete_userspace_io)) {
8482 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8483 vcpu->arch.complete_userspace_io = NULL;
8488 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8490 if (kvm_run->immediate_exit)
8496 kvm_put_guest_fpu(vcpu);
8497 if (vcpu->run->kvm_valid_regs)
8499 post_kvm_run_save(vcpu);
8500 kvm_sigset_deactivate(vcpu);
8506 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8508 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8510 * We are here if userspace calls get_regs() in the middle of
8511 * instruction emulation. Registers state needs to be copied
8512 * back from emulation context to vcpu. Userspace shouldn't do
8513 * that usually, but some bad designed PV devices (vmware
8514 * backdoor interface) need this to work
8516 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8517 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8519 regs->rax = kvm_rax_read(vcpu);
8520 regs->rbx = kvm_rbx_read(vcpu);
8521 regs->rcx = kvm_rcx_read(vcpu);
8522 regs->rdx = kvm_rdx_read(vcpu);
8523 regs->rsi = kvm_rsi_read(vcpu);
8524 regs->rdi = kvm_rdi_read(vcpu);
8525 regs->rsp = kvm_rsp_read(vcpu);
8526 regs->rbp = kvm_rbp_read(vcpu);
8527 #ifdef CONFIG_X86_64
8528 regs->r8 = kvm_r8_read(vcpu);
8529 regs->r9 = kvm_r9_read(vcpu);
8530 regs->r10 = kvm_r10_read(vcpu);
8531 regs->r11 = kvm_r11_read(vcpu);
8532 regs->r12 = kvm_r12_read(vcpu);
8533 regs->r13 = kvm_r13_read(vcpu);
8534 regs->r14 = kvm_r14_read(vcpu);
8535 regs->r15 = kvm_r15_read(vcpu);
8538 regs->rip = kvm_rip_read(vcpu);
8539 regs->rflags = kvm_get_rflags(vcpu);
8542 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8545 __get_regs(vcpu, regs);
8550 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8552 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8553 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8555 kvm_rax_write(vcpu, regs->rax);
8556 kvm_rbx_write(vcpu, regs->rbx);
8557 kvm_rcx_write(vcpu, regs->rcx);
8558 kvm_rdx_write(vcpu, regs->rdx);
8559 kvm_rsi_write(vcpu, regs->rsi);
8560 kvm_rdi_write(vcpu, regs->rdi);
8561 kvm_rsp_write(vcpu, regs->rsp);
8562 kvm_rbp_write(vcpu, regs->rbp);
8563 #ifdef CONFIG_X86_64
8564 kvm_r8_write(vcpu, regs->r8);
8565 kvm_r9_write(vcpu, regs->r9);
8566 kvm_r10_write(vcpu, regs->r10);
8567 kvm_r11_write(vcpu, regs->r11);
8568 kvm_r12_write(vcpu, regs->r12);
8569 kvm_r13_write(vcpu, regs->r13);
8570 kvm_r14_write(vcpu, regs->r14);
8571 kvm_r15_write(vcpu, regs->r15);
8574 kvm_rip_write(vcpu, regs->rip);
8575 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8577 vcpu->arch.exception.pending = false;
8579 kvm_make_request(KVM_REQ_EVENT, vcpu);
8582 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8585 __set_regs(vcpu, regs);
8590 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8592 struct kvm_segment cs;
8594 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8598 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8600 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8604 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8605 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8606 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8607 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8608 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8609 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8611 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8612 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8614 kvm_x86_ops->get_idt(vcpu, &dt);
8615 sregs->idt.limit = dt.size;
8616 sregs->idt.base = dt.address;
8617 kvm_x86_ops->get_gdt(vcpu, &dt);
8618 sregs->gdt.limit = dt.size;
8619 sregs->gdt.base = dt.address;
8621 sregs->cr0 = kvm_read_cr0(vcpu);
8622 sregs->cr2 = vcpu->arch.cr2;
8623 sregs->cr3 = kvm_read_cr3(vcpu);
8624 sregs->cr4 = kvm_read_cr4(vcpu);
8625 sregs->cr8 = kvm_get_cr8(vcpu);
8626 sregs->efer = vcpu->arch.efer;
8627 sregs->apic_base = kvm_get_apic_base(vcpu);
8629 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8631 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8632 set_bit(vcpu->arch.interrupt.nr,
8633 (unsigned long *)sregs->interrupt_bitmap);
8636 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8637 struct kvm_sregs *sregs)
8640 __get_sregs(vcpu, sregs);
8645 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8646 struct kvm_mp_state *mp_state)
8650 kvm_apic_accept_events(vcpu);
8651 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8652 vcpu->arch.pv.pv_unhalted)
8653 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8655 mp_state->mp_state = vcpu->arch.mp_state;
8661 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8662 struct kvm_mp_state *mp_state)
8668 if (!lapic_in_kernel(vcpu) &&
8669 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8672 /* INITs are latched while in SMM */
8673 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8674 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8675 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8678 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8679 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8680 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8682 vcpu->arch.mp_state = mp_state->mp_state;
8683 kvm_make_request(KVM_REQ_EVENT, vcpu);
8691 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8692 int reason, bool has_error_code, u32 error_code)
8694 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8697 init_emulate_ctxt(vcpu);
8699 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8700 has_error_code, error_code);
8702 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8703 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8704 vcpu->run->internal.ndata = 0;
8708 kvm_rip_write(vcpu, ctxt->eip);
8709 kvm_set_rflags(vcpu, ctxt->eflags);
8710 kvm_make_request(KVM_REQ_EVENT, vcpu);
8713 EXPORT_SYMBOL_GPL(kvm_task_switch);
8715 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8717 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8718 (sregs->cr4 & X86_CR4_OSXSAVE))
8721 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8723 * When EFER.LME and CR0.PG are set, the processor is in
8724 * 64-bit mode (though maybe in a 32-bit code segment).
8725 * CR4.PAE and EFER.LMA must be set.
8727 if (!(sregs->cr4 & X86_CR4_PAE)
8728 || !(sregs->efer & EFER_LMA))
8732 * Not in 64-bit mode: EFER.LMA is clear and the code
8733 * segment cannot be 64-bit.
8735 if (sregs->efer & EFER_LMA || sregs->cs.l)
8742 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8744 struct msr_data apic_base_msr;
8745 int mmu_reset_needed = 0;
8746 int cpuid_update_needed = 0;
8747 int pending_vec, max_bits, idx;
8751 if (kvm_valid_sregs(vcpu, sregs))
8754 apic_base_msr.data = sregs->apic_base;
8755 apic_base_msr.host_initiated = true;
8756 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8759 dt.size = sregs->idt.limit;
8760 dt.address = sregs->idt.base;
8761 kvm_x86_ops->set_idt(vcpu, &dt);
8762 dt.size = sregs->gdt.limit;
8763 dt.address = sregs->gdt.base;
8764 kvm_x86_ops->set_gdt(vcpu, &dt);
8766 vcpu->arch.cr2 = sregs->cr2;
8767 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8768 vcpu->arch.cr3 = sregs->cr3;
8769 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8771 kvm_set_cr8(vcpu, sregs->cr8);
8773 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8774 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8776 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8777 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8778 vcpu->arch.cr0 = sregs->cr0;
8780 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8781 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8782 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8783 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8784 if (cpuid_update_needed)
8785 kvm_update_cpuid(vcpu);
8787 idx = srcu_read_lock(&vcpu->kvm->srcu);
8788 if (is_pae_paging(vcpu)) {
8789 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8790 mmu_reset_needed = 1;
8792 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8794 if (mmu_reset_needed)
8795 kvm_mmu_reset_context(vcpu);
8797 max_bits = KVM_NR_INTERRUPTS;
8798 pending_vec = find_first_bit(
8799 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8800 if (pending_vec < max_bits) {
8801 kvm_queue_interrupt(vcpu, pending_vec, false);
8802 pr_debug("Set back pending irq %d\n", pending_vec);
8805 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8806 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8807 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8808 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8809 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8810 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8812 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8813 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8815 update_cr8_intercept(vcpu);
8817 /* Older userspace won't unhalt the vcpu on reset. */
8818 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8819 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8821 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8823 kvm_make_request(KVM_REQ_EVENT, vcpu);
8830 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8831 struct kvm_sregs *sregs)
8836 ret = __set_sregs(vcpu, sregs);
8841 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8842 struct kvm_guest_debug *dbg)
8844 unsigned long rflags;
8849 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8851 if (vcpu->arch.exception.pending)
8853 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8854 kvm_queue_exception(vcpu, DB_VECTOR);
8856 kvm_queue_exception(vcpu, BP_VECTOR);
8860 * Read rflags as long as potentially injected trace flags are still
8863 rflags = kvm_get_rflags(vcpu);
8865 vcpu->guest_debug = dbg->control;
8866 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8867 vcpu->guest_debug = 0;
8869 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8870 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8871 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8872 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8874 for (i = 0; i < KVM_NR_DB_REGS; i++)
8875 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8877 kvm_update_dr7(vcpu);
8879 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8880 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8881 get_segment_base(vcpu, VCPU_SREG_CS);
8884 * Trigger an rflags update that will inject or remove the trace
8887 kvm_set_rflags(vcpu, rflags);
8889 kvm_x86_ops->update_bp_intercept(vcpu);
8899 * Translate a guest virtual address to a guest physical address.
8901 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8902 struct kvm_translation *tr)
8904 unsigned long vaddr = tr->linear_address;
8910 idx = srcu_read_lock(&vcpu->kvm->srcu);
8911 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8912 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8913 tr->physical_address = gpa;
8914 tr->valid = gpa != UNMAPPED_GVA;
8922 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8924 struct fxregs_state *fxsave;
8928 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8929 memcpy(fpu->fpr, fxsave->st_space, 128);
8930 fpu->fcw = fxsave->cwd;
8931 fpu->fsw = fxsave->swd;
8932 fpu->ftwx = fxsave->twd;
8933 fpu->last_opcode = fxsave->fop;
8934 fpu->last_ip = fxsave->rip;
8935 fpu->last_dp = fxsave->rdp;
8936 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8942 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8944 struct fxregs_state *fxsave;
8948 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8950 memcpy(fxsave->st_space, fpu->fpr, 128);
8951 fxsave->cwd = fpu->fcw;
8952 fxsave->swd = fpu->fsw;
8953 fxsave->twd = fpu->ftwx;
8954 fxsave->fop = fpu->last_opcode;
8955 fxsave->rip = fpu->last_ip;
8956 fxsave->rdp = fpu->last_dp;
8957 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8963 static void store_regs(struct kvm_vcpu *vcpu)
8965 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8967 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8968 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8970 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8971 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8973 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8974 kvm_vcpu_ioctl_x86_get_vcpu_events(
8975 vcpu, &vcpu->run->s.regs.events);
8978 static int sync_regs(struct kvm_vcpu *vcpu)
8980 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8983 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8984 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8985 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8987 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8988 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8990 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8992 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8993 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8994 vcpu, &vcpu->run->s.regs.events))
8996 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9002 static void fx_init(struct kvm_vcpu *vcpu)
9004 fpstate_init(&vcpu->arch.guest_fpu->state);
9005 if (boot_cpu_has(X86_FEATURE_XSAVES))
9006 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9007 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9010 * Ensure guest xcr0 is valid for loading
9012 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9014 vcpu->arch.cr0 |= X86_CR0_ET;
9017 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
9019 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
9021 kvmclock_reset(vcpu);
9023 kvm_x86_ops->vcpu_free(vcpu);
9024 free_cpumask_var(wbinvd_dirty_mask);
9027 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
9030 struct kvm_vcpu *vcpu;
9032 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9033 printk_once(KERN_WARNING
9034 "kvm: SMP vm created on host with unstable TSC; "
9035 "guest TSC will not be reliable\n");
9037 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
9042 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
9044 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9045 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9046 kvm_vcpu_mtrr_init(vcpu);
9048 kvm_vcpu_reset(vcpu, false);
9049 kvm_init_mmu(vcpu, false);
9054 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9056 struct msr_data msr;
9057 struct kvm *kvm = vcpu->kvm;
9059 kvm_hv_vcpu_postcreate(vcpu);
9061 if (mutex_lock_killable(&vcpu->mutex))
9065 msr.index = MSR_IA32_TSC;
9066 msr.host_initiated = true;
9067 kvm_write_tsc(vcpu, &msr);
9070 /* poll control enabled by default */
9071 vcpu->arch.msr_kvm_poll_control = 1;
9073 mutex_unlock(&vcpu->mutex);
9075 if (!kvmclock_periodic_sync)
9078 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9079 KVMCLOCK_SYNC_PERIOD);
9082 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9084 vcpu->arch.apf.msr_val = 0;
9087 kvm_mmu_unload(vcpu);
9090 kvm_x86_ops->vcpu_free(vcpu);
9093 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9095 kvm_lapic_reset(vcpu, init_event);
9097 vcpu->arch.hflags = 0;
9099 vcpu->arch.smi_pending = 0;
9100 vcpu->arch.smi_count = 0;
9101 atomic_set(&vcpu->arch.nmi_queued, 0);
9102 vcpu->arch.nmi_pending = 0;
9103 vcpu->arch.nmi_injected = false;
9104 kvm_clear_interrupt_queue(vcpu);
9105 kvm_clear_exception_queue(vcpu);
9106 vcpu->arch.exception.pending = false;
9108 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9109 kvm_update_dr0123(vcpu);
9110 vcpu->arch.dr6 = DR6_INIT;
9111 kvm_update_dr6(vcpu);
9112 vcpu->arch.dr7 = DR7_FIXED_1;
9113 kvm_update_dr7(vcpu);
9117 kvm_make_request(KVM_REQ_EVENT, vcpu);
9118 vcpu->arch.apf.msr_val = 0;
9119 vcpu->arch.st.msr_val = 0;
9121 kvmclock_reset(vcpu);
9123 kvm_clear_async_pf_completion_queue(vcpu);
9124 kvm_async_pf_hash_reset(vcpu);
9125 vcpu->arch.apf.halted = false;
9127 if (kvm_mpx_supported()) {
9128 void *mpx_state_buffer;
9131 * To avoid have the INIT path from kvm_apic_has_events() that be
9132 * called with loaded FPU and does not let userspace fix the state.
9135 kvm_put_guest_fpu(vcpu);
9136 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9138 if (mpx_state_buffer)
9139 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9140 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9142 if (mpx_state_buffer)
9143 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9145 kvm_load_guest_fpu(vcpu);
9149 kvm_pmu_reset(vcpu);
9150 vcpu->arch.smbase = 0x30000;
9152 vcpu->arch.msr_misc_features_enables = 0;
9154 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9157 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9158 vcpu->arch.regs_avail = ~0;
9159 vcpu->arch.regs_dirty = ~0;
9161 vcpu->arch.ia32_xss = 0;
9163 kvm_x86_ops->vcpu_reset(vcpu, init_event);
9166 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9168 struct kvm_segment cs;
9170 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9171 cs.selector = vector << 8;
9172 cs.base = vector << 12;
9173 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9174 kvm_rip_write(vcpu, 0);
9177 int kvm_arch_hardware_enable(void)
9180 struct kvm_vcpu *vcpu;
9185 bool stable, backwards_tsc = false;
9187 kvm_shared_msr_cpu_online();
9188 ret = kvm_x86_ops->hardware_enable();
9192 local_tsc = rdtsc();
9193 stable = !kvm_check_tsc_unstable();
9194 list_for_each_entry(kvm, &vm_list, vm_list) {
9195 kvm_for_each_vcpu(i, vcpu, kvm) {
9196 if (!stable && vcpu->cpu == smp_processor_id())
9197 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9198 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9199 backwards_tsc = true;
9200 if (vcpu->arch.last_host_tsc > max_tsc)
9201 max_tsc = vcpu->arch.last_host_tsc;
9207 * Sometimes, even reliable TSCs go backwards. This happens on
9208 * platforms that reset TSC during suspend or hibernate actions, but
9209 * maintain synchronization. We must compensate. Fortunately, we can
9210 * detect that condition here, which happens early in CPU bringup,
9211 * before any KVM threads can be running. Unfortunately, we can't
9212 * bring the TSCs fully up to date with real time, as we aren't yet far
9213 * enough into CPU bringup that we know how much real time has actually
9214 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9215 * variables that haven't been updated yet.
9217 * So we simply find the maximum observed TSC above, then record the
9218 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9219 * the adjustment will be applied. Note that we accumulate
9220 * adjustments, in case multiple suspend cycles happen before some VCPU
9221 * gets a chance to run again. In the event that no KVM threads get a
9222 * chance to run, we will miss the entire elapsed period, as we'll have
9223 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9224 * loose cycle time. This isn't too big a deal, since the loss will be
9225 * uniform across all VCPUs (not to mention the scenario is extremely
9226 * unlikely). It is possible that a second hibernate recovery happens
9227 * much faster than a first, causing the observed TSC here to be
9228 * smaller; this would require additional padding adjustment, which is
9229 * why we set last_host_tsc to the local tsc observed here.
9231 * N.B. - this code below runs only on platforms with reliable TSC,
9232 * as that is the only way backwards_tsc is set above. Also note
9233 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9234 * have the same delta_cyc adjustment applied if backwards_tsc
9235 * is detected. Note further, this adjustment is only done once,
9236 * as we reset last_host_tsc on all VCPUs to stop this from being
9237 * called multiple times (one for each physical CPU bringup).
9239 * Platforms with unreliable TSCs don't have to deal with this, they
9240 * will be compensated by the logic in vcpu_load, which sets the TSC to
9241 * catchup mode. This will catchup all VCPUs to real time, but cannot
9242 * guarantee that they stay in perfect synchronization.
9244 if (backwards_tsc) {
9245 u64 delta_cyc = max_tsc - local_tsc;
9246 list_for_each_entry(kvm, &vm_list, vm_list) {
9247 kvm->arch.backwards_tsc_observed = true;
9248 kvm_for_each_vcpu(i, vcpu, kvm) {
9249 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9250 vcpu->arch.last_host_tsc = local_tsc;
9251 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9255 * We have to disable TSC offset matching.. if you were
9256 * booting a VM while issuing an S4 host suspend....
9257 * you may have some problem. Solving this issue is
9258 * left as an exercise to the reader.
9260 kvm->arch.last_tsc_nsec = 0;
9261 kvm->arch.last_tsc_write = 0;
9268 void kvm_arch_hardware_disable(void)
9270 kvm_x86_ops->hardware_disable();
9271 drop_user_return_notifiers();
9274 int kvm_arch_hardware_setup(void)
9278 r = kvm_x86_ops->hardware_setup();
9282 if (kvm_has_tsc_control) {
9284 * Make sure the user can only configure tsc_khz values that
9285 * fit into a signed integer.
9286 * A min value is not calculated because it will always
9287 * be 1 on all machines.
9289 u64 max = min(0x7fffffffULL,
9290 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9291 kvm_max_guest_tsc_khz = max;
9293 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9296 kvm_init_msr_list();
9300 void kvm_arch_hardware_unsetup(void)
9302 kvm_x86_ops->hardware_unsetup();
9305 int kvm_arch_check_processor_compat(void)
9307 return kvm_x86_ops->check_processor_compatibility();
9310 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9312 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9314 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9316 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9318 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9321 struct static_key kvm_no_apic_vcpu __read_mostly;
9322 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9324 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9329 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9330 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9331 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9333 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9335 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9340 vcpu->arch.pio_data = page_address(page);
9342 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9344 r = kvm_mmu_create(vcpu);
9346 goto fail_free_pio_data;
9348 if (irqchip_in_kernel(vcpu->kvm)) {
9349 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9350 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9352 goto fail_mmu_destroy;
9354 static_key_slow_inc(&kvm_no_apic_vcpu);
9356 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9357 GFP_KERNEL_ACCOUNT);
9358 if (!vcpu->arch.mce_banks) {
9360 goto fail_free_lapic;
9362 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9364 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9365 GFP_KERNEL_ACCOUNT)) {
9367 goto fail_free_mce_banks;
9372 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9374 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9376 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9378 kvm_async_pf_hash_reset(vcpu);
9381 vcpu->arch.pending_external_vector = -1;
9382 vcpu->arch.preempted_in_kernel = false;
9384 kvm_hv_vcpu_init(vcpu);
9388 fail_free_mce_banks:
9389 kfree(vcpu->arch.mce_banks);
9391 kvm_free_lapic(vcpu);
9393 kvm_mmu_destroy(vcpu);
9395 free_page((unsigned long)vcpu->arch.pio_data);
9400 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9404 kvm_hv_vcpu_uninit(vcpu);
9405 kvm_pmu_destroy(vcpu);
9406 kfree(vcpu->arch.mce_banks);
9407 kvm_free_lapic(vcpu);
9408 idx = srcu_read_lock(&vcpu->kvm->srcu);
9409 kvm_mmu_destroy(vcpu);
9410 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9411 free_page((unsigned long)vcpu->arch.pio_data);
9412 if (!lapic_in_kernel(vcpu))
9413 static_key_slow_dec(&kvm_no_apic_vcpu);
9416 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9418 vcpu->arch.l1tf_flush_l1d = true;
9419 kvm_x86_ops->sched_in(vcpu, cpu);
9422 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9427 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9428 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9429 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9430 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9431 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9433 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9434 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9435 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9436 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9437 &kvm->arch.irq_sources_bitmap);
9439 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9440 mutex_init(&kvm->arch.apic_map_lock);
9441 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9443 kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9444 pvclock_update_vm_gtod_copy(kvm);
9446 kvm->arch.guest_can_read_msr_platform_info = true;
9448 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9449 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9451 kvm_hv_init_vm(kvm);
9452 kvm_page_track_init(kvm);
9453 kvm_mmu_init_vm(kvm);
9455 return kvm_x86_ops->vm_init(kvm);
9458 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9461 kvm_mmu_unload(vcpu);
9465 static void kvm_free_vcpus(struct kvm *kvm)
9468 struct kvm_vcpu *vcpu;
9471 * Unpin any mmu pages first.
9473 kvm_for_each_vcpu(i, vcpu, kvm) {
9474 kvm_clear_async_pf_completion_queue(vcpu);
9475 kvm_unload_vcpu_mmu(vcpu);
9477 kvm_for_each_vcpu(i, vcpu, kvm)
9478 kvm_arch_vcpu_free(vcpu);
9480 mutex_lock(&kvm->lock);
9481 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9482 kvm->vcpus[i] = NULL;
9484 atomic_set(&kvm->online_vcpus, 0);
9485 mutex_unlock(&kvm->lock);
9488 void kvm_arch_sync_events(struct kvm *kvm)
9490 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9491 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9495 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9499 struct kvm_memslots *slots = kvm_memslots(kvm);
9500 struct kvm_memory_slot *slot, old;
9502 /* Called with kvm->slots_lock held. */
9503 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9506 slot = id_to_memslot(slots, id);
9512 * MAP_SHARED to prevent internal slot pages from being moved
9515 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9516 MAP_SHARED | MAP_ANONYMOUS, 0);
9517 if (IS_ERR((void *)hva))
9518 return PTR_ERR((void *)hva);
9527 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9528 struct kvm_userspace_memory_region m;
9530 m.slot = id | (i << 16);
9532 m.guest_phys_addr = gpa;
9533 m.userspace_addr = hva;
9534 m.memory_size = size;
9535 r = __kvm_set_memory_region(kvm, &m);
9541 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9545 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9547 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9551 mutex_lock(&kvm->slots_lock);
9552 r = __x86_set_memory_region(kvm, id, gpa, size);
9553 mutex_unlock(&kvm->slots_lock);
9557 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9559 void kvm_arch_destroy_vm(struct kvm *kvm)
9561 if (current->mm == kvm->mm) {
9563 * Free memory regions allocated on behalf of userspace,
9564 * unless the the memory map has changed due to process exit
9567 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9568 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9569 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9571 if (kvm_x86_ops->vm_destroy)
9572 kvm_x86_ops->vm_destroy(kvm);
9573 kvm_pic_destroy(kvm);
9574 kvm_ioapic_destroy(kvm);
9575 kvm_free_vcpus(kvm);
9576 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9577 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9578 kvm_mmu_uninit_vm(kvm);
9579 kvm_page_track_cleanup(kvm);
9580 kvm_hv_destroy_vm(kvm);
9583 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9584 struct kvm_memory_slot *dont)
9588 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9589 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9590 kvfree(free->arch.rmap[i]);
9591 free->arch.rmap[i] = NULL;
9596 if (!dont || free->arch.lpage_info[i - 1] !=
9597 dont->arch.lpage_info[i - 1]) {
9598 kvfree(free->arch.lpage_info[i - 1]);
9599 free->arch.lpage_info[i - 1] = NULL;
9603 kvm_page_track_free_memslot(free, dont);
9606 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9607 unsigned long npages)
9611 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9612 struct kvm_lpage_info *linfo;
9617 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9618 slot->base_gfn, level) + 1;
9620 slot->arch.rmap[i] =
9621 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9622 GFP_KERNEL_ACCOUNT);
9623 if (!slot->arch.rmap[i])
9628 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9632 slot->arch.lpage_info[i - 1] = linfo;
9634 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9635 linfo[0].disallow_lpage = 1;
9636 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9637 linfo[lpages - 1].disallow_lpage = 1;
9638 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9640 * If the gfn and userspace address are not aligned wrt each
9641 * other, or if explicitly asked to, disable large page
9642 * support for this slot
9644 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9645 !kvm_largepages_enabled()) {
9648 for (j = 0; j < lpages; ++j)
9649 linfo[j].disallow_lpage = 1;
9653 if (kvm_page_track_create_memslot(slot, npages))
9659 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9660 kvfree(slot->arch.rmap[i]);
9661 slot->arch.rmap[i] = NULL;
9665 kvfree(slot->arch.lpage_info[i - 1]);
9666 slot->arch.lpage_info[i - 1] = NULL;
9671 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9674 * memslots->generation has been incremented.
9675 * mmio generation may have reached its maximum value.
9677 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9680 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9681 struct kvm_memory_slot *memslot,
9682 const struct kvm_userspace_memory_region *mem,
9683 enum kvm_mr_change change)
9688 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9689 struct kvm_memory_slot *new)
9691 /* Still write protect RO slot */
9692 if (new->flags & KVM_MEM_READONLY) {
9693 kvm_mmu_slot_remove_write_access(kvm, new);
9698 * Call kvm_x86_ops dirty logging hooks when they are valid.
9700 * kvm_x86_ops->slot_disable_log_dirty is called when:
9702 * - KVM_MR_CREATE with dirty logging is disabled
9703 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9705 * The reason is, in case of PML, we need to set D-bit for any slots
9706 * with dirty logging disabled in order to eliminate unnecessary GPA
9707 * logging in PML buffer (and potential PML buffer full VMEXT). This
9708 * guarantees leaving PML enabled during guest's lifetime won't have
9709 * any additional overhead from PML when guest is running with dirty
9710 * logging disabled for memory slots.
9712 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9713 * to dirty logging mode.
9715 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9717 * In case of write protect:
9719 * Write protect all pages for dirty logging.
9721 * All the sptes including the large sptes which point to this
9722 * slot are set to readonly. We can not create any new large
9723 * spte on this slot until the end of the logging.
9725 * See the comments in fast_page_fault().
9727 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9728 if (kvm_x86_ops->slot_enable_log_dirty)
9729 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9731 kvm_mmu_slot_remove_write_access(kvm, new);
9733 if (kvm_x86_ops->slot_disable_log_dirty)
9734 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9738 void kvm_arch_commit_memory_region(struct kvm *kvm,
9739 const struct kvm_userspace_memory_region *mem,
9740 const struct kvm_memory_slot *old,
9741 const struct kvm_memory_slot *new,
9742 enum kvm_mr_change change)
9744 if (!kvm->arch.n_requested_mmu_pages)
9745 kvm_mmu_change_mmu_pages(kvm,
9746 kvm_mmu_calculate_default_mmu_pages(kvm));
9749 * Dirty logging tracks sptes in 4k granularity, meaning that large
9750 * sptes have to be split. If live migration is successful, the guest
9751 * in the source machine will be destroyed and large sptes will be
9752 * created in the destination. However, if the guest continues to run
9753 * in the source machine (for example if live migration fails), small
9754 * sptes will remain around and cause bad performance.
9756 * Scan sptes if dirty logging has been stopped, dropping those
9757 * which can be collapsed into a single large-page spte. Later
9758 * page faults will create the large-page sptes.
9760 * There is no need to do this in any of the following cases:
9761 * CREATE: No dirty mappings will already exist.
9762 * MOVE/DELETE: The old mappings will already have been cleaned up by
9763 * kvm_arch_flush_shadow_memslot()
9765 if (change == KVM_MR_FLAGS_ONLY &&
9766 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9767 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9768 kvm_mmu_zap_collapsible_sptes(kvm, new);
9771 * Set up write protection and/or dirty logging for the new slot.
9773 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9774 * been zapped so no dirty logging staff is needed for old slot. For
9775 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9776 * new and it's also covered when dealing with the new slot.
9778 * FIXME: const-ify all uses of struct kvm_memory_slot.
9780 if (change != KVM_MR_DELETE)
9781 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9784 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9786 kvm_mmu_zap_all(kvm);
9789 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9790 struct kvm_memory_slot *slot)
9792 kvm_page_track_flush_slot(kvm, slot);
9795 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9797 return (is_guest_mode(vcpu) &&
9798 kvm_x86_ops->guest_apic_has_interrupt &&
9799 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9802 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9804 if (!list_empty_careful(&vcpu->async_pf.done))
9807 if (kvm_apic_has_events(vcpu))
9810 if (vcpu->arch.pv.pv_unhalted)
9813 if (vcpu->arch.exception.pending)
9816 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9817 (vcpu->arch.nmi_pending &&
9818 kvm_x86_ops->nmi_allowed(vcpu)))
9821 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9822 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9825 if (kvm_arch_interrupt_allowed(vcpu) &&
9826 (kvm_cpu_has_interrupt(vcpu) ||
9827 kvm_guest_apic_has_interrupt(vcpu)))
9830 if (kvm_hv_has_stimer_pending(vcpu))
9836 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9838 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9841 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
9843 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
9846 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9847 kvm_test_request(KVM_REQ_SMI, vcpu) ||
9848 kvm_test_request(KVM_REQ_EVENT, vcpu))
9851 if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
9857 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9859 return vcpu->arch.preempted_in_kernel;
9862 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9864 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9867 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9869 return kvm_x86_ops->interrupt_allowed(vcpu);
9872 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9874 if (is_64_bit_mode(vcpu))
9875 return kvm_rip_read(vcpu);
9876 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9877 kvm_rip_read(vcpu));
9879 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9881 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9883 return kvm_get_linear_rip(vcpu) == linear_rip;
9885 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9887 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9889 unsigned long rflags;
9891 rflags = kvm_x86_ops->get_rflags(vcpu);
9892 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9893 rflags &= ~X86_EFLAGS_TF;
9896 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9898 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9900 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9901 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9902 rflags |= X86_EFLAGS_TF;
9903 kvm_x86_ops->set_rflags(vcpu, rflags);
9906 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9908 __kvm_set_rflags(vcpu, rflags);
9909 kvm_make_request(KVM_REQ_EVENT, vcpu);
9911 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9913 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9917 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9921 r = kvm_mmu_reload(vcpu);
9925 if (!vcpu->arch.mmu->direct_map &&
9926 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9929 vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9932 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9934 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9937 static inline u32 kvm_async_pf_next_probe(u32 key)
9939 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9942 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9944 u32 key = kvm_async_pf_hash_fn(gfn);
9946 while (vcpu->arch.apf.gfns[key] != ~0)
9947 key = kvm_async_pf_next_probe(key);
9949 vcpu->arch.apf.gfns[key] = gfn;
9952 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9955 u32 key = kvm_async_pf_hash_fn(gfn);
9957 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9958 (vcpu->arch.apf.gfns[key] != gfn &&
9959 vcpu->arch.apf.gfns[key] != ~0); i++)
9960 key = kvm_async_pf_next_probe(key);
9965 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9967 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9970 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9974 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9976 vcpu->arch.apf.gfns[i] = ~0;
9978 j = kvm_async_pf_next_probe(j);
9979 if (vcpu->arch.apf.gfns[j] == ~0)
9981 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9983 * k lies cyclically in ]i,j]
9985 * |....j i.k.| or |.k..j i...|
9987 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9988 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9993 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9996 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10000 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10003 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10007 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10009 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10012 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10013 (vcpu->arch.apf.send_user_only &&
10014 kvm_x86_ops->get_cpl(vcpu) == 0))
10020 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10022 if (unlikely(!lapic_in_kernel(vcpu) ||
10023 kvm_event_needs_reinjection(vcpu) ||
10024 vcpu->arch.exception.pending))
10027 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10031 * If interrupts are off we cannot even use an artificial
10034 return kvm_x86_ops->interrupt_allowed(vcpu);
10037 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10038 struct kvm_async_pf *work)
10040 struct x86_exception fault;
10042 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
10043 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10045 if (kvm_can_deliver_async_pf(vcpu) &&
10046 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10047 fault.vector = PF_VECTOR;
10048 fault.error_code_valid = true;
10049 fault.error_code = 0;
10050 fault.nested_page_fault = false;
10051 fault.address = work->arch.token;
10052 fault.async_page_fault = true;
10053 kvm_inject_page_fault(vcpu, &fault);
10056 * It is not possible to deliver a paravirtualized asynchronous
10057 * page fault, but putting the guest in an artificial halt state
10058 * can be beneficial nevertheless: if an interrupt arrives, we
10059 * can deliver it timely and perhaps the guest will schedule
10060 * another process. When the instruction that triggered a page
10061 * fault is retried, hopefully the page will be ready in the host.
10063 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10067 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10068 struct kvm_async_pf *work)
10070 struct x86_exception fault;
10073 if (work->wakeup_all)
10074 work->arch.token = ~0; /* broadcast wakeup */
10076 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10077 trace_kvm_async_pf_ready(work->arch.token, work->gva);
10079 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10080 !apf_get_user(vcpu, &val)) {
10081 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10082 vcpu->arch.exception.pending &&
10083 vcpu->arch.exception.nr == PF_VECTOR &&
10084 !apf_put_user(vcpu, 0)) {
10085 vcpu->arch.exception.injected = false;
10086 vcpu->arch.exception.pending = false;
10087 vcpu->arch.exception.nr = 0;
10088 vcpu->arch.exception.has_error_code = false;
10089 vcpu->arch.exception.error_code = 0;
10090 vcpu->arch.exception.has_payload = false;
10091 vcpu->arch.exception.payload = 0;
10092 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10093 fault.vector = PF_VECTOR;
10094 fault.error_code_valid = true;
10095 fault.error_code = 0;
10096 fault.nested_page_fault = false;
10097 fault.address = work->arch.token;
10098 fault.async_page_fault = true;
10099 kvm_inject_page_fault(vcpu, &fault);
10102 vcpu->arch.apf.halted = false;
10103 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10106 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10108 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10111 return kvm_can_do_async_pf(vcpu);
10114 void kvm_arch_start_assignment(struct kvm *kvm)
10116 atomic_inc(&kvm->arch.assigned_device_count);
10118 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10120 void kvm_arch_end_assignment(struct kvm *kvm)
10122 atomic_dec(&kvm->arch.assigned_device_count);
10124 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10126 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10128 return atomic_read(&kvm->arch.assigned_device_count);
10130 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10132 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10134 atomic_inc(&kvm->arch.noncoherent_dma_count);
10136 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10138 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10140 atomic_dec(&kvm->arch.noncoherent_dma_count);
10142 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10144 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10146 return atomic_read(&kvm->arch.noncoherent_dma_count);
10148 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10150 bool kvm_arch_has_irq_bypass(void)
10155 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10156 struct irq_bypass_producer *prod)
10158 struct kvm_kernel_irqfd *irqfd =
10159 container_of(cons, struct kvm_kernel_irqfd, consumer);
10161 irqfd->producer = prod;
10163 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10164 prod->irq, irqfd->gsi, 1);
10167 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10168 struct irq_bypass_producer *prod)
10171 struct kvm_kernel_irqfd *irqfd =
10172 container_of(cons, struct kvm_kernel_irqfd, consumer);
10174 WARN_ON(irqfd->producer != prod);
10175 irqfd->producer = NULL;
10178 * When producer of consumer is unregistered, we change back to
10179 * remapped mode, so we can re-use the current implementation
10180 * when the irq is masked/disabled or the consumer side (KVM
10181 * int this case doesn't want to receive the interrupts.
10183 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10185 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10186 " fails: %d\n", irqfd->consumer.token, ret);
10189 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10190 uint32_t guest_irq, bool set)
10192 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10195 bool kvm_vector_hashing_enabled(void)
10197 return vector_hashing;
10199 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10201 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10203 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10205 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10208 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10209 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10210 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10211 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10212 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10213 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10214 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10215 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10216 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10217 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10218 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10219 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10220 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10221 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10222 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10223 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10224 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10225 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10226 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10227 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);