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, ## __VA_ARGS__
96 #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__
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, .mode = 0444) },
216 { "nx_largepages_splitted", VM_STAT(nx_lpage_splits, .mode = 0444) },
217 { "max_mmu_page_hash_collisions",
218 VM_STAT(max_mmu_page_hash_collisions) },
222 u64 __read_mostly host_xcr0;
224 struct kmem_cache *x86_fpu_cache;
225 EXPORT_SYMBOL_GPL(x86_fpu_cache);
227 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
229 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
232 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
233 vcpu->arch.apf.gfns[i] = ~0;
236 static void kvm_on_user_return(struct user_return_notifier *urn)
239 struct kvm_shared_msrs *locals
240 = container_of(urn, struct kvm_shared_msrs, urn);
241 struct kvm_shared_msr_values *values;
245 * Disabling irqs at this point since the following code could be
246 * interrupted and executed through kvm_arch_hardware_disable()
248 local_irq_save(flags);
249 if (locals->registered) {
250 locals->registered = false;
251 user_return_notifier_unregister(urn);
253 local_irq_restore(flags);
254 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
255 values = &locals->values[slot];
256 if (values->host != values->curr) {
257 wrmsrl(shared_msrs_global.msrs[slot], values->host);
258 values->curr = values->host;
263 static void shared_msr_update(unsigned slot, u32 msr)
266 unsigned int cpu = smp_processor_id();
267 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
269 /* only read, and nobody should modify it at this time,
270 * so don't need lock */
271 if (slot >= shared_msrs_global.nr) {
272 printk(KERN_ERR "kvm: invalid MSR slot!");
275 rdmsrl_safe(msr, &value);
276 smsr->values[slot].host = value;
277 smsr->values[slot].curr = value;
280 void kvm_define_shared_msr(unsigned slot, u32 msr)
282 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
283 shared_msrs_global.msrs[slot] = msr;
284 if (slot >= shared_msrs_global.nr)
285 shared_msrs_global.nr = slot + 1;
287 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
289 static void kvm_shared_msr_cpu_online(void)
293 for (i = 0; i < shared_msrs_global.nr; ++i)
294 shared_msr_update(i, shared_msrs_global.msrs[i]);
297 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
299 unsigned int cpu = smp_processor_id();
300 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
303 if (((value ^ smsr->values[slot].curr) & mask) == 0)
305 smsr->values[slot].curr = value;
306 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
310 if (!smsr->registered) {
311 smsr->urn.on_user_return = kvm_on_user_return;
312 user_return_notifier_register(&smsr->urn);
313 smsr->registered = true;
317 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
319 static void drop_user_return_notifiers(void)
321 unsigned int cpu = smp_processor_id();
322 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
324 if (smsr->registered)
325 kvm_on_user_return(&smsr->urn);
328 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
330 return vcpu->arch.apic_base;
332 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
334 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
336 return kvm_apic_mode(kvm_get_apic_base(vcpu));
338 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
340 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
342 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
343 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
344 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
345 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
347 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
349 if (!msr_info->host_initiated) {
350 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
352 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
356 kvm_lapic_set_base(vcpu, msr_info->data);
359 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
361 asmlinkage __visible void kvm_spurious_fault(void)
363 /* Fault while not rebooting. We want the trace. */
364 BUG_ON(!kvm_rebooting);
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 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
890 if (cr4 & CR4_RESERVED_BITS)
893 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
896 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
899 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
902 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
905 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
908 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
911 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
917 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
919 unsigned long old_cr4 = kvm_read_cr4(vcpu);
920 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
921 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
923 if (kvm_valid_cr4(vcpu, cr4))
926 if (is_long_mode(vcpu)) {
927 if (!(cr4 & X86_CR4_PAE))
929 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
930 && ((cr4 ^ old_cr4) & pdptr_bits)
931 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
935 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
936 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
939 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
940 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
944 if (kvm_x86_ops->set_cr4(vcpu, cr4))
947 if (((cr4 ^ old_cr4) & pdptr_bits) ||
948 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
949 kvm_mmu_reset_context(vcpu);
951 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
952 kvm_update_cpuid(vcpu);
956 EXPORT_SYMBOL_GPL(kvm_set_cr4);
958 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
960 bool skip_tlb_flush = false;
962 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
965 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
966 cr3 &= ~X86_CR3_PCID_NOFLUSH;
970 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
971 if (!skip_tlb_flush) {
972 kvm_mmu_sync_roots(vcpu);
973 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
978 if (is_long_mode(vcpu) &&
979 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
981 else if (is_pae_paging(vcpu) &&
982 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
985 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
986 vcpu->arch.cr3 = cr3;
987 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
991 EXPORT_SYMBOL_GPL(kvm_set_cr3);
993 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
995 if (cr8 & CR8_RESERVED_BITS)
997 if (lapic_in_kernel(vcpu))
998 kvm_lapic_set_tpr(vcpu, cr8);
1000 vcpu->arch.cr8 = cr8;
1003 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1005 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1007 if (lapic_in_kernel(vcpu))
1008 return kvm_lapic_get_cr8(vcpu);
1010 return vcpu->arch.cr8;
1012 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1014 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1018 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1019 for (i = 0; i < KVM_NR_DB_REGS; i++)
1020 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1021 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1025 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1027 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1028 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1031 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1035 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1036 dr7 = vcpu->arch.guest_debug_dr7;
1038 dr7 = vcpu->arch.dr7;
1039 kvm_x86_ops->set_dr7(vcpu, dr7);
1040 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1041 if (dr7 & DR7_BP_EN_MASK)
1042 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1045 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1047 u64 fixed = DR6_FIXED_1;
1049 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1054 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1058 vcpu->arch.db[dr] = val;
1059 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1060 vcpu->arch.eff_db[dr] = val;
1065 if (val & 0xffffffff00000000ULL)
1066 return -1; /* #GP */
1067 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1068 kvm_update_dr6(vcpu);
1073 if (val & 0xffffffff00000000ULL)
1074 return -1; /* #GP */
1075 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1076 kvm_update_dr7(vcpu);
1083 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1085 if (__kvm_set_dr(vcpu, dr, val)) {
1086 kvm_inject_gp(vcpu, 0);
1091 EXPORT_SYMBOL_GPL(kvm_set_dr);
1093 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1097 *val = vcpu->arch.db[dr];
1102 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1103 *val = vcpu->arch.dr6;
1105 *val = kvm_x86_ops->get_dr6(vcpu);
1110 *val = vcpu->arch.dr7;
1115 EXPORT_SYMBOL_GPL(kvm_get_dr);
1117 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1119 u32 ecx = kvm_rcx_read(vcpu);
1123 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1126 kvm_rax_write(vcpu, (u32)data);
1127 kvm_rdx_write(vcpu, data >> 32);
1130 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1133 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1134 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1136 * The three MSR lists(msrs_to_save, emulated_msrs, msr_based_features)
1137 * extract the supported MSRs from the related const lists.
1138 * msrs_to_save is selected from the msrs_to_save_all to reflect the
1139 * capabilities of the host cpu. This capabilities test skips MSRs that are
1140 * kvm-specific. Those are put in emulated_msrs_all; filtering of emulated_msrs
1141 * may depend on host virtualization features rather than host cpu features.
1144 static const u32 msrs_to_save_all[] = {
1145 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1147 #ifdef CONFIG_X86_64
1148 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1150 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1151 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1153 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1154 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1155 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1156 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1157 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1158 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1159 MSR_IA32_UMWAIT_CONTROL,
1161 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1162 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1163 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1164 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1165 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1166 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1167 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1168 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1169 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1170 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1171 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1172 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1173 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1174 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1175 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1176 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1177 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1178 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1179 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1180 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1181 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1182 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1185 static u32 msrs_to_save[ARRAY_SIZE(msrs_to_save_all)];
1186 static unsigned num_msrs_to_save;
1188 static const u32 emulated_msrs_all[] = {
1189 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1190 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1191 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1192 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1193 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1194 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1195 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1197 HV_X64_MSR_VP_INDEX,
1198 HV_X64_MSR_VP_RUNTIME,
1199 HV_X64_MSR_SCONTROL,
1200 HV_X64_MSR_STIMER0_CONFIG,
1201 HV_X64_MSR_VP_ASSIST_PAGE,
1202 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1203 HV_X64_MSR_TSC_EMULATION_STATUS,
1205 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1208 MSR_IA32_TSC_ADJUST,
1209 MSR_IA32_TSCDEADLINE,
1210 MSR_IA32_ARCH_CAPABILITIES,
1211 MSR_IA32_MISC_ENABLE,
1212 MSR_IA32_MCG_STATUS,
1214 MSR_IA32_MCG_EXT_CTL,
1218 MSR_MISC_FEATURES_ENABLES,
1219 MSR_AMD64_VIRT_SPEC_CTRL,
1223 * The following list leaves out MSRs whose values are determined
1224 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1225 * We always support the "true" VMX control MSRs, even if the host
1226 * processor does not, so I am putting these registers here rather
1227 * than in msrs_to_save_all.
1230 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1231 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1232 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1233 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1235 MSR_IA32_VMX_CR0_FIXED0,
1236 MSR_IA32_VMX_CR4_FIXED0,
1237 MSR_IA32_VMX_VMCS_ENUM,
1238 MSR_IA32_VMX_PROCBASED_CTLS2,
1239 MSR_IA32_VMX_EPT_VPID_CAP,
1240 MSR_IA32_VMX_VMFUNC,
1243 MSR_KVM_POLL_CONTROL,
1246 static u32 emulated_msrs[ARRAY_SIZE(emulated_msrs_all)];
1247 static unsigned num_emulated_msrs;
1250 * List of msr numbers which are used to expose MSR-based features that
1251 * can be used by a hypervisor to validate requested CPU features.
1253 static const u32 msr_based_features_all[] = {
1255 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1256 MSR_IA32_VMX_PINBASED_CTLS,
1257 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1258 MSR_IA32_VMX_PROCBASED_CTLS,
1259 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1260 MSR_IA32_VMX_EXIT_CTLS,
1261 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1262 MSR_IA32_VMX_ENTRY_CTLS,
1264 MSR_IA32_VMX_CR0_FIXED0,
1265 MSR_IA32_VMX_CR0_FIXED1,
1266 MSR_IA32_VMX_CR4_FIXED0,
1267 MSR_IA32_VMX_CR4_FIXED1,
1268 MSR_IA32_VMX_VMCS_ENUM,
1269 MSR_IA32_VMX_PROCBASED_CTLS2,
1270 MSR_IA32_VMX_EPT_VPID_CAP,
1271 MSR_IA32_VMX_VMFUNC,
1275 MSR_IA32_ARCH_CAPABILITIES,
1278 static u32 msr_based_features[ARRAY_SIZE(msr_based_features_all)];
1279 static unsigned int num_msr_based_features;
1281 static u64 kvm_get_arch_capabilities(void)
1285 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1286 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1289 * If nx_huge_pages is enabled, KVM's shadow paging will ensure that
1290 * the nested hypervisor runs with NX huge pages. If it is not,
1291 * L1 is anyway vulnerable to ITLB_MULTIHIT explots from other
1292 * L1 guests, so it need not worry about its own (L2) guests.
1294 data |= ARCH_CAP_PSCHANGE_MC_NO;
1297 * If we're doing cache flushes (either "always" or "cond")
1298 * we will do one whenever the guest does a vmlaunch/vmresume.
1299 * If an outer hypervisor is doing the cache flush for us
1300 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1301 * capability to the guest too, and if EPT is disabled we're not
1302 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1303 * require a nested hypervisor to do a flush of its own.
1305 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1306 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1308 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1309 data |= ARCH_CAP_RDCL_NO;
1310 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1311 data |= ARCH_CAP_SSB_NO;
1312 if (!boot_cpu_has_bug(X86_BUG_MDS))
1313 data |= ARCH_CAP_MDS_NO;
1316 * On TAA affected systems, export MDS_NO=0 when:
1317 * - TSX is enabled on the host, i.e. X86_FEATURE_RTM=1.
1318 * - Updated microcode is present. This is detected by
1319 * the presence of ARCH_CAP_TSX_CTRL_MSR and ensures
1320 * that VERW clears CPU buffers.
1322 * When MDS_NO=0 is exported, guests deploy clear CPU buffer
1323 * mitigation and don't complain:
1325 * "Vulnerable: Clear CPU buffers attempted, no microcode"
1327 * If TSX is disabled on the system, guests are also mitigated against
1328 * TAA and clear CPU buffer mitigation is not required for guests.
1330 if (boot_cpu_has_bug(X86_BUG_TAA) && boot_cpu_has(X86_FEATURE_RTM) &&
1331 (data & ARCH_CAP_TSX_CTRL_MSR))
1332 data &= ~ARCH_CAP_MDS_NO;
1337 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1339 switch (msr->index) {
1340 case MSR_IA32_ARCH_CAPABILITIES:
1341 msr->data = kvm_get_arch_capabilities();
1343 case MSR_IA32_UCODE_REV:
1344 rdmsrl_safe(msr->index, &msr->data);
1347 if (kvm_x86_ops->get_msr_feature(msr))
1353 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1355 struct kvm_msr_entry msr;
1359 r = kvm_get_msr_feature(&msr);
1368 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1370 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1373 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1376 if (efer & (EFER_LME | EFER_LMA) &&
1377 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1380 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1386 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1388 if (efer & efer_reserved_bits)
1391 return __kvm_valid_efer(vcpu, efer);
1393 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1395 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1397 u64 old_efer = vcpu->arch.efer;
1398 u64 efer = msr_info->data;
1400 if (efer & efer_reserved_bits)
1403 if (!msr_info->host_initiated) {
1404 if (!__kvm_valid_efer(vcpu, efer))
1407 if (is_paging(vcpu) &&
1408 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1413 efer |= vcpu->arch.efer & EFER_LMA;
1415 kvm_x86_ops->set_efer(vcpu, efer);
1417 /* Update reserved bits */
1418 if ((efer ^ old_efer) & EFER_NX)
1419 kvm_mmu_reset_context(vcpu);
1424 void kvm_enable_efer_bits(u64 mask)
1426 efer_reserved_bits &= ~mask;
1428 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1431 * Write @data into the MSR specified by @index. Select MSR specific fault
1432 * checks are bypassed if @host_initiated is %true.
1433 * Returns 0 on success, non-0 otherwise.
1434 * Assumes vcpu_load() was already called.
1436 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1437 bool host_initiated)
1439 struct msr_data msr;
1444 case MSR_KERNEL_GS_BASE:
1447 if (is_noncanonical_address(data, vcpu))
1450 case MSR_IA32_SYSENTER_EIP:
1451 case MSR_IA32_SYSENTER_ESP:
1453 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1454 * non-canonical address is written on Intel but not on
1455 * AMD (which ignores the top 32-bits, because it does
1456 * not implement 64-bit SYSENTER).
1458 * 64-bit code should hence be able to write a non-canonical
1459 * value on AMD. Making the address canonical ensures that
1460 * vmentry does not fail on Intel after writing a non-canonical
1461 * value, and that something deterministic happens if the guest
1462 * invokes 64-bit SYSENTER.
1464 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1469 msr.host_initiated = host_initiated;
1471 return kvm_x86_ops->set_msr(vcpu, &msr);
1475 * Read the MSR specified by @index into @data. Select MSR specific fault
1476 * checks are bypassed if @host_initiated is %true.
1477 * Returns 0 on success, non-0 otherwise.
1478 * Assumes vcpu_load() was already called.
1480 static int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1481 bool host_initiated)
1483 struct msr_data msr;
1487 msr.host_initiated = host_initiated;
1489 ret = kvm_x86_ops->get_msr(vcpu, &msr);
1495 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1497 return __kvm_get_msr(vcpu, index, data, false);
1499 EXPORT_SYMBOL_GPL(kvm_get_msr);
1501 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1503 return __kvm_set_msr(vcpu, index, data, false);
1505 EXPORT_SYMBOL_GPL(kvm_set_msr);
1507 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1509 u32 ecx = kvm_rcx_read(vcpu);
1512 if (kvm_get_msr(vcpu, ecx, &data)) {
1513 trace_kvm_msr_read_ex(ecx);
1514 kvm_inject_gp(vcpu, 0);
1518 trace_kvm_msr_read(ecx, data);
1520 kvm_rax_write(vcpu, data & -1u);
1521 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1522 return kvm_skip_emulated_instruction(vcpu);
1524 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1526 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1528 u32 ecx = kvm_rcx_read(vcpu);
1529 u64 data = kvm_read_edx_eax(vcpu);
1531 if (kvm_set_msr(vcpu, ecx, data)) {
1532 trace_kvm_msr_write_ex(ecx, data);
1533 kvm_inject_gp(vcpu, 0);
1537 trace_kvm_msr_write(ecx, data);
1538 return kvm_skip_emulated_instruction(vcpu);
1540 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1543 * Adapt set_msr() to msr_io()'s calling convention
1545 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1547 return __kvm_get_msr(vcpu, index, data, true);
1550 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1552 return __kvm_set_msr(vcpu, index, *data, true);
1555 #ifdef CONFIG_X86_64
1556 struct pvclock_gtod_data {
1559 struct { /* extract of a clocksource struct */
1572 static struct pvclock_gtod_data pvclock_gtod_data;
1574 static void update_pvclock_gtod(struct timekeeper *tk)
1576 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1579 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1581 write_seqcount_begin(&vdata->seq);
1583 /* copy pvclock gtod data */
1584 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1585 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1586 vdata->clock.mask = tk->tkr_mono.mask;
1587 vdata->clock.mult = tk->tkr_mono.mult;
1588 vdata->clock.shift = tk->tkr_mono.shift;
1590 vdata->boot_ns = boot_ns;
1591 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1593 vdata->wall_time_sec = tk->xtime_sec;
1595 write_seqcount_end(&vdata->seq);
1599 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1601 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1602 kvm_vcpu_kick(vcpu);
1605 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1609 struct pvclock_wall_clock wc;
1610 struct timespec64 boot;
1615 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1620 ++version; /* first time write, random junk */
1624 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1628 * The guest calculates current wall clock time by adding
1629 * system time (updated by kvm_guest_time_update below) to the
1630 * wall clock specified here. guest system time equals host
1631 * system time for us, thus we must fill in host boot time here.
1633 getboottime64(&boot);
1635 if (kvm->arch.kvmclock_offset) {
1636 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1637 boot = timespec64_sub(boot, ts);
1639 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1640 wc.nsec = boot.tv_nsec;
1641 wc.version = version;
1643 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1646 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1649 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1651 do_shl32_div32(dividend, divisor);
1655 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1656 s8 *pshift, u32 *pmultiplier)
1664 scaled64 = scaled_hz;
1665 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1670 tps32 = (uint32_t)tps64;
1671 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1672 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1680 *pmultiplier = div_frac(scaled64, tps32);
1683 #ifdef CONFIG_X86_64
1684 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1687 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1688 static unsigned long max_tsc_khz;
1690 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1692 u64 v = (u64)khz * (1000000 + ppm);
1697 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1701 /* Guest TSC same frequency as host TSC? */
1703 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1707 /* TSC scaling supported? */
1708 if (!kvm_has_tsc_control) {
1709 if (user_tsc_khz > tsc_khz) {
1710 vcpu->arch.tsc_catchup = 1;
1711 vcpu->arch.tsc_always_catchup = 1;
1714 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1719 /* TSC scaling required - calculate ratio */
1720 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1721 user_tsc_khz, tsc_khz);
1723 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1724 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1729 vcpu->arch.tsc_scaling_ratio = ratio;
1733 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1735 u32 thresh_lo, thresh_hi;
1736 int use_scaling = 0;
1738 /* tsc_khz can be zero if TSC calibration fails */
1739 if (user_tsc_khz == 0) {
1740 /* set tsc_scaling_ratio to a safe value */
1741 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1745 /* Compute a scale to convert nanoseconds in TSC cycles */
1746 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1747 &vcpu->arch.virtual_tsc_shift,
1748 &vcpu->arch.virtual_tsc_mult);
1749 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1752 * Compute the variation in TSC rate which is acceptable
1753 * within the range of tolerance and decide if the
1754 * rate being applied is within that bounds of the hardware
1755 * rate. If so, no scaling or compensation need be done.
1757 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1758 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1759 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1760 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1763 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1766 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1768 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1769 vcpu->arch.virtual_tsc_mult,
1770 vcpu->arch.virtual_tsc_shift);
1771 tsc += vcpu->arch.this_tsc_write;
1775 static inline int gtod_is_based_on_tsc(int mode)
1777 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1780 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1782 #ifdef CONFIG_X86_64
1784 struct kvm_arch *ka = &vcpu->kvm->arch;
1785 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1787 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1788 atomic_read(&vcpu->kvm->online_vcpus));
1791 * Once the masterclock is enabled, always perform request in
1792 * order to update it.
1794 * In order to enable masterclock, the host clocksource must be TSC
1795 * and the vcpus need to have matched TSCs. When that happens,
1796 * perform request to enable masterclock.
1798 if (ka->use_master_clock ||
1799 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1800 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1802 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1803 atomic_read(&vcpu->kvm->online_vcpus),
1804 ka->use_master_clock, gtod->clock.vclock_mode);
1808 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1810 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1811 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1815 * Multiply tsc by a fixed point number represented by ratio.
1817 * The most significant 64-N bits (mult) of ratio represent the
1818 * integral part of the fixed point number; the remaining N bits
1819 * (frac) represent the fractional part, ie. ratio represents a fixed
1820 * point number (mult + frac * 2^(-N)).
1822 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1824 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1826 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1829 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1832 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1834 if (ratio != kvm_default_tsc_scaling_ratio)
1835 _tsc = __scale_tsc(ratio, tsc);
1839 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1841 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1845 tsc = kvm_scale_tsc(vcpu, rdtsc());
1847 return target_tsc - tsc;
1850 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1852 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1854 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1856 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1858 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1860 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1863 static inline bool kvm_check_tsc_unstable(void)
1865 #ifdef CONFIG_X86_64
1867 * TSC is marked unstable when we're running on Hyper-V,
1868 * 'TSC page' clocksource is good.
1870 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1873 return check_tsc_unstable();
1876 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1878 struct kvm *kvm = vcpu->kvm;
1879 u64 offset, ns, elapsed;
1880 unsigned long flags;
1882 bool already_matched;
1883 u64 data = msr->data;
1884 bool synchronizing = false;
1886 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1887 offset = kvm_compute_tsc_offset(vcpu, data);
1888 ns = ktime_get_boottime_ns();
1889 elapsed = ns - kvm->arch.last_tsc_nsec;
1891 if (vcpu->arch.virtual_tsc_khz) {
1892 if (data == 0 && msr->host_initiated) {
1894 * detection of vcpu initialization -- need to sync
1895 * with other vCPUs. This particularly helps to keep
1896 * kvm_clock stable after CPU hotplug
1898 synchronizing = true;
1900 u64 tsc_exp = kvm->arch.last_tsc_write +
1901 nsec_to_cycles(vcpu, elapsed);
1902 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1904 * Special case: TSC write with a small delta (1 second)
1905 * of virtual cycle time against real time is
1906 * interpreted as an attempt to synchronize the CPU.
1908 synchronizing = data < tsc_exp + tsc_hz &&
1909 data + tsc_hz > tsc_exp;
1914 * For a reliable TSC, we can match TSC offsets, and for an unstable
1915 * TSC, we add elapsed time in this computation. We could let the
1916 * compensation code attempt to catch up if we fall behind, but
1917 * it's better to try to match offsets from the beginning.
1919 if (synchronizing &&
1920 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1921 if (!kvm_check_tsc_unstable()) {
1922 offset = kvm->arch.cur_tsc_offset;
1924 u64 delta = nsec_to_cycles(vcpu, elapsed);
1926 offset = kvm_compute_tsc_offset(vcpu, data);
1929 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1932 * We split periods of matched TSC writes into generations.
1933 * For each generation, we track the original measured
1934 * nanosecond time, offset, and write, so if TSCs are in
1935 * sync, we can match exact offset, and if not, we can match
1936 * exact software computation in compute_guest_tsc()
1938 * These values are tracked in kvm->arch.cur_xxx variables.
1940 kvm->arch.cur_tsc_generation++;
1941 kvm->arch.cur_tsc_nsec = ns;
1942 kvm->arch.cur_tsc_write = data;
1943 kvm->arch.cur_tsc_offset = offset;
1948 * We also track th most recent recorded KHZ, write and time to
1949 * allow the matching interval to be extended at each write.
1951 kvm->arch.last_tsc_nsec = ns;
1952 kvm->arch.last_tsc_write = data;
1953 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1955 vcpu->arch.last_guest_tsc = data;
1957 /* Keep track of which generation this VCPU has synchronized to */
1958 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1959 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1960 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1962 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1963 update_ia32_tsc_adjust_msr(vcpu, offset);
1965 kvm_vcpu_write_tsc_offset(vcpu, offset);
1966 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1968 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1970 kvm->arch.nr_vcpus_matched_tsc = 0;
1971 } else if (!already_matched) {
1972 kvm->arch.nr_vcpus_matched_tsc++;
1975 kvm_track_tsc_matching(vcpu);
1976 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1979 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1981 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1984 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1985 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1988 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1990 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1991 WARN_ON(adjustment < 0);
1992 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1993 adjust_tsc_offset_guest(vcpu, adjustment);
1996 #ifdef CONFIG_X86_64
1998 static u64 read_tsc(void)
2000 u64 ret = (u64)rdtsc_ordered();
2001 u64 last = pvclock_gtod_data.clock.cycle_last;
2003 if (likely(ret >= last))
2007 * GCC likes to generate cmov here, but this branch is extremely
2008 * predictable (it's just a function of time and the likely is
2009 * very likely) and there's a data dependence, so force GCC
2010 * to generate a branch instead. I don't barrier() because
2011 * we don't actually need a barrier, and if this function
2012 * ever gets inlined it will generate worse code.
2018 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
2021 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2024 switch (gtod->clock.vclock_mode) {
2025 case VCLOCK_HVCLOCK:
2026 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2028 if (tsc_pg_val != U64_MAX) {
2029 /* TSC page valid */
2030 *mode = VCLOCK_HVCLOCK;
2031 v = (tsc_pg_val - gtod->clock.cycle_last) &
2034 /* TSC page invalid */
2035 *mode = VCLOCK_NONE;
2040 *tsc_timestamp = read_tsc();
2041 v = (*tsc_timestamp - gtod->clock.cycle_last) &
2045 *mode = VCLOCK_NONE;
2048 if (*mode == VCLOCK_NONE)
2049 *tsc_timestamp = v = 0;
2051 return v * gtod->clock.mult;
2054 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
2056 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2062 seq = read_seqcount_begin(>od->seq);
2063 ns = gtod->nsec_base;
2064 ns += vgettsc(tsc_timestamp, &mode);
2065 ns >>= gtod->clock.shift;
2066 ns += gtod->boot_ns;
2067 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2073 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2075 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2081 seq = read_seqcount_begin(>od->seq);
2082 ts->tv_sec = gtod->wall_time_sec;
2083 ns = gtod->nsec_base;
2084 ns += vgettsc(tsc_timestamp, &mode);
2085 ns >>= gtod->clock.shift;
2086 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2088 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2094 /* returns true if host is using TSC based clocksource */
2095 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2097 /* checked again under seqlock below */
2098 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2101 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
2105 /* returns true if host is using TSC based clocksource */
2106 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2109 /* checked again under seqlock below */
2110 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2113 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2119 * Assuming a stable TSC across physical CPUS, and a stable TSC
2120 * across virtual CPUs, the following condition is possible.
2121 * Each numbered line represents an event visible to both
2122 * CPUs at the next numbered event.
2124 * "timespecX" represents host monotonic time. "tscX" represents
2127 * VCPU0 on CPU0 | VCPU1 on CPU1
2129 * 1. read timespec0,tsc0
2130 * 2. | timespec1 = timespec0 + N
2132 * 3. transition to guest | transition to guest
2133 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2134 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2135 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2137 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2140 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2142 * - 0 < N - M => M < N
2144 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2145 * always the case (the difference between two distinct xtime instances
2146 * might be smaller then the difference between corresponding TSC reads,
2147 * when updating guest vcpus pvclock areas).
2149 * To avoid that problem, do not allow visibility of distinct
2150 * system_timestamp/tsc_timestamp values simultaneously: use a master
2151 * copy of host monotonic time values. Update that master copy
2154 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2158 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2160 #ifdef CONFIG_X86_64
2161 struct kvm_arch *ka = &kvm->arch;
2163 bool host_tsc_clocksource, vcpus_matched;
2165 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2166 atomic_read(&kvm->online_vcpus));
2169 * If the host uses TSC clock, then passthrough TSC as stable
2172 host_tsc_clocksource = kvm_get_time_and_clockread(
2173 &ka->master_kernel_ns,
2174 &ka->master_cycle_now);
2176 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2177 && !ka->backwards_tsc_observed
2178 && !ka->boot_vcpu_runs_old_kvmclock;
2180 if (ka->use_master_clock)
2181 atomic_set(&kvm_guest_has_master_clock, 1);
2183 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2184 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2189 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2191 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2194 static void kvm_gen_update_masterclock(struct kvm *kvm)
2196 #ifdef CONFIG_X86_64
2198 struct kvm_vcpu *vcpu;
2199 struct kvm_arch *ka = &kvm->arch;
2201 spin_lock(&ka->pvclock_gtod_sync_lock);
2202 kvm_make_mclock_inprogress_request(kvm);
2203 /* no guest entries from this point */
2204 pvclock_update_vm_gtod_copy(kvm);
2206 kvm_for_each_vcpu(i, vcpu, kvm)
2207 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2209 /* guest entries allowed */
2210 kvm_for_each_vcpu(i, vcpu, kvm)
2211 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2213 spin_unlock(&ka->pvclock_gtod_sync_lock);
2217 u64 get_kvmclock_ns(struct kvm *kvm)
2219 struct kvm_arch *ka = &kvm->arch;
2220 struct pvclock_vcpu_time_info hv_clock;
2223 spin_lock(&ka->pvclock_gtod_sync_lock);
2224 if (!ka->use_master_clock) {
2225 spin_unlock(&ka->pvclock_gtod_sync_lock);
2226 return ktime_get_boottime_ns() + ka->kvmclock_offset;
2229 hv_clock.tsc_timestamp = ka->master_cycle_now;
2230 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2231 spin_unlock(&ka->pvclock_gtod_sync_lock);
2233 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2236 if (__this_cpu_read(cpu_tsc_khz)) {
2237 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2238 &hv_clock.tsc_shift,
2239 &hv_clock.tsc_to_system_mul);
2240 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2242 ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2249 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2251 struct kvm_vcpu_arch *vcpu = &v->arch;
2252 struct pvclock_vcpu_time_info guest_hv_clock;
2254 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2255 &guest_hv_clock, sizeof(guest_hv_clock))))
2258 /* This VCPU is paused, but it's legal for a guest to read another
2259 * VCPU's kvmclock, so we really have to follow the specification where
2260 * it says that version is odd if data is being modified, and even after
2263 * Version field updates must be kept separate. This is because
2264 * kvm_write_guest_cached might use a "rep movs" instruction, and
2265 * writes within a string instruction are weakly ordered. So there
2266 * are three writes overall.
2268 * As a small optimization, only write the version field in the first
2269 * and third write. The vcpu->pv_time cache is still valid, because the
2270 * version field is the first in the struct.
2272 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2274 if (guest_hv_clock.version & 1)
2275 ++guest_hv_clock.version; /* first time write, random junk */
2277 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2278 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2280 sizeof(vcpu->hv_clock.version));
2284 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2285 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2287 if (vcpu->pvclock_set_guest_stopped_request) {
2288 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2289 vcpu->pvclock_set_guest_stopped_request = false;
2292 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2294 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2296 sizeof(vcpu->hv_clock));
2300 vcpu->hv_clock.version++;
2301 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2303 sizeof(vcpu->hv_clock.version));
2306 static int kvm_guest_time_update(struct kvm_vcpu *v)
2308 unsigned long flags, tgt_tsc_khz;
2309 struct kvm_vcpu_arch *vcpu = &v->arch;
2310 struct kvm_arch *ka = &v->kvm->arch;
2312 u64 tsc_timestamp, host_tsc;
2314 bool use_master_clock;
2320 * If the host uses TSC clock, then passthrough TSC as stable
2323 spin_lock(&ka->pvclock_gtod_sync_lock);
2324 use_master_clock = ka->use_master_clock;
2325 if (use_master_clock) {
2326 host_tsc = ka->master_cycle_now;
2327 kernel_ns = ka->master_kernel_ns;
2329 spin_unlock(&ka->pvclock_gtod_sync_lock);
2331 /* Keep irq disabled to prevent changes to the clock */
2332 local_irq_save(flags);
2333 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2334 if (unlikely(tgt_tsc_khz == 0)) {
2335 local_irq_restore(flags);
2336 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2339 if (!use_master_clock) {
2341 kernel_ns = ktime_get_boottime_ns();
2344 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2347 * We may have to catch up the TSC to match elapsed wall clock
2348 * time for two reasons, even if kvmclock is used.
2349 * 1) CPU could have been running below the maximum TSC rate
2350 * 2) Broken TSC compensation resets the base at each VCPU
2351 * entry to avoid unknown leaps of TSC even when running
2352 * again on the same CPU. This may cause apparent elapsed
2353 * time to disappear, and the guest to stand still or run
2356 if (vcpu->tsc_catchup) {
2357 u64 tsc = compute_guest_tsc(v, kernel_ns);
2358 if (tsc > tsc_timestamp) {
2359 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2360 tsc_timestamp = tsc;
2364 local_irq_restore(flags);
2366 /* With all the info we got, fill in the values */
2368 if (kvm_has_tsc_control)
2369 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2371 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2372 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2373 &vcpu->hv_clock.tsc_shift,
2374 &vcpu->hv_clock.tsc_to_system_mul);
2375 vcpu->hw_tsc_khz = tgt_tsc_khz;
2378 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2379 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2380 vcpu->last_guest_tsc = tsc_timestamp;
2382 /* If the host uses TSC clocksource, then it is stable */
2384 if (use_master_clock)
2385 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2387 vcpu->hv_clock.flags = pvclock_flags;
2389 if (vcpu->pv_time_enabled)
2390 kvm_setup_pvclock_page(v);
2391 if (v == kvm_get_vcpu(v->kvm, 0))
2392 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2397 * kvmclock updates which are isolated to a given vcpu, such as
2398 * vcpu->cpu migration, should not allow system_timestamp from
2399 * the rest of the vcpus to remain static. Otherwise ntp frequency
2400 * correction applies to one vcpu's system_timestamp but not
2403 * So in those cases, request a kvmclock update for all vcpus.
2404 * We need to rate-limit these requests though, as they can
2405 * considerably slow guests that have a large number of vcpus.
2406 * The time for a remote vcpu to update its kvmclock is bound
2407 * by the delay we use to rate-limit the updates.
2410 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2412 static void kvmclock_update_fn(struct work_struct *work)
2415 struct delayed_work *dwork = to_delayed_work(work);
2416 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2417 kvmclock_update_work);
2418 struct kvm *kvm = container_of(ka, struct kvm, arch);
2419 struct kvm_vcpu *vcpu;
2421 kvm_for_each_vcpu(i, vcpu, kvm) {
2422 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2423 kvm_vcpu_kick(vcpu);
2427 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2429 struct kvm *kvm = v->kvm;
2431 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2432 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2433 KVMCLOCK_UPDATE_DELAY);
2436 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2438 static void kvmclock_sync_fn(struct work_struct *work)
2440 struct delayed_work *dwork = to_delayed_work(work);
2441 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2442 kvmclock_sync_work);
2443 struct kvm *kvm = container_of(ka, struct kvm, arch);
2445 if (!kvmclock_periodic_sync)
2448 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2449 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2450 KVMCLOCK_SYNC_PERIOD);
2454 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2456 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2458 /* McStatusWrEn enabled? */
2459 if (guest_cpuid_is_amd(vcpu))
2460 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2465 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2467 u64 mcg_cap = vcpu->arch.mcg_cap;
2468 unsigned bank_num = mcg_cap & 0xff;
2469 u32 msr = msr_info->index;
2470 u64 data = msr_info->data;
2473 case MSR_IA32_MCG_STATUS:
2474 vcpu->arch.mcg_status = data;
2476 case MSR_IA32_MCG_CTL:
2477 if (!(mcg_cap & MCG_CTL_P) &&
2478 (data || !msr_info->host_initiated))
2480 if (data != 0 && data != ~(u64)0)
2482 vcpu->arch.mcg_ctl = data;
2485 if (msr >= MSR_IA32_MC0_CTL &&
2486 msr < MSR_IA32_MCx_CTL(bank_num)) {
2487 u32 offset = msr - MSR_IA32_MC0_CTL;
2488 /* only 0 or all 1s can be written to IA32_MCi_CTL
2489 * some Linux kernels though clear bit 10 in bank 4 to
2490 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2491 * this to avoid an uncatched #GP in the guest
2493 if ((offset & 0x3) == 0 &&
2494 data != 0 && (data | (1 << 10)) != ~(u64)0)
2498 if (!msr_info->host_initiated &&
2499 (offset & 0x3) == 1 && data != 0) {
2500 if (!can_set_mci_status(vcpu))
2504 vcpu->arch.mce_banks[offset] = data;
2512 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2514 struct kvm *kvm = vcpu->kvm;
2515 int lm = is_long_mode(vcpu);
2516 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2517 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2518 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2519 : kvm->arch.xen_hvm_config.blob_size_32;
2520 u32 page_num = data & ~PAGE_MASK;
2521 u64 page_addr = data & PAGE_MASK;
2526 if (page_num >= blob_size)
2529 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2534 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2543 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2545 gpa_t gpa = data & ~0x3f;
2547 /* Bits 3:5 are reserved, Should be zero */
2551 vcpu->arch.apf.msr_val = data;
2553 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2554 kvm_clear_async_pf_completion_queue(vcpu);
2555 kvm_async_pf_hash_reset(vcpu);
2559 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2563 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2564 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2565 kvm_async_pf_wakeup_all(vcpu);
2569 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2571 vcpu->arch.pv_time_enabled = false;
2572 vcpu->arch.time = 0;
2575 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2577 ++vcpu->stat.tlb_flush;
2578 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2581 static void record_steal_time(struct kvm_vcpu *vcpu)
2583 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2586 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2587 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2591 * Doing a TLB flush here, on the guest's behalf, can avoid
2594 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2595 vcpu->arch.st.steal.preempted & KVM_VCPU_FLUSH_TLB);
2596 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2597 kvm_vcpu_flush_tlb(vcpu, false);
2599 if (vcpu->arch.st.steal.version & 1)
2600 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2602 vcpu->arch.st.steal.version += 1;
2604 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2605 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2609 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2610 vcpu->arch.st.last_steal;
2611 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2613 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2614 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2618 vcpu->arch.st.steal.version += 1;
2620 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2621 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2624 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2627 u32 msr = msr_info->index;
2628 u64 data = msr_info->data;
2631 case MSR_AMD64_NB_CFG:
2632 case MSR_IA32_UCODE_WRITE:
2633 case MSR_VM_HSAVE_PA:
2634 case MSR_AMD64_PATCH_LOADER:
2635 case MSR_AMD64_BU_CFG2:
2636 case MSR_AMD64_DC_CFG:
2637 case MSR_F15H_EX_CFG:
2640 case MSR_IA32_UCODE_REV:
2641 if (msr_info->host_initiated)
2642 vcpu->arch.microcode_version = data;
2644 case MSR_IA32_ARCH_CAPABILITIES:
2645 if (!msr_info->host_initiated)
2647 vcpu->arch.arch_capabilities = data;
2650 return set_efer(vcpu, msr_info);
2652 data &= ~(u64)0x40; /* ignore flush filter disable */
2653 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2654 data &= ~(u64)0x8; /* ignore TLB cache disable */
2656 /* Handle McStatusWrEn */
2657 if (data == BIT_ULL(18)) {
2658 vcpu->arch.msr_hwcr = data;
2659 } else if (data != 0) {
2660 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2665 case MSR_FAM10H_MMIO_CONF_BASE:
2667 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2672 case MSR_IA32_DEBUGCTLMSR:
2674 /* We support the non-activated case already */
2676 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2677 /* Values other than LBR and BTF are vendor-specific,
2678 thus reserved and should throw a #GP */
2681 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2684 case 0x200 ... 0x2ff:
2685 return kvm_mtrr_set_msr(vcpu, msr, data);
2686 case MSR_IA32_APICBASE:
2687 return kvm_set_apic_base(vcpu, msr_info);
2688 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2689 return kvm_x2apic_msr_write(vcpu, msr, data);
2690 case MSR_IA32_TSCDEADLINE:
2691 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2693 case MSR_IA32_TSC_ADJUST:
2694 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2695 if (!msr_info->host_initiated) {
2696 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2697 adjust_tsc_offset_guest(vcpu, adj);
2699 vcpu->arch.ia32_tsc_adjust_msr = data;
2702 case MSR_IA32_MISC_ENABLE:
2703 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2704 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2705 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2707 vcpu->arch.ia32_misc_enable_msr = data;
2708 kvm_update_cpuid(vcpu);
2710 vcpu->arch.ia32_misc_enable_msr = data;
2713 case MSR_IA32_SMBASE:
2714 if (!msr_info->host_initiated)
2716 vcpu->arch.smbase = data;
2718 case MSR_IA32_POWER_CTL:
2719 vcpu->arch.msr_ia32_power_ctl = data;
2722 kvm_write_tsc(vcpu, msr_info);
2725 if (!msr_info->host_initiated)
2727 vcpu->arch.smi_count = data;
2729 case MSR_KVM_WALL_CLOCK_NEW:
2730 case MSR_KVM_WALL_CLOCK:
2731 vcpu->kvm->arch.wall_clock = data;
2732 kvm_write_wall_clock(vcpu->kvm, data);
2734 case MSR_KVM_SYSTEM_TIME_NEW:
2735 case MSR_KVM_SYSTEM_TIME: {
2736 struct kvm_arch *ka = &vcpu->kvm->arch;
2738 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2739 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2741 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2742 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2744 ka->boot_vcpu_runs_old_kvmclock = tmp;
2747 vcpu->arch.time = data;
2748 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2750 /* we verify if the enable bit is set... */
2751 vcpu->arch.pv_time_enabled = false;
2755 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2756 &vcpu->arch.pv_time, data & ~1ULL,
2757 sizeof(struct pvclock_vcpu_time_info)))
2758 vcpu->arch.pv_time_enabled = true;
2762 case MSR_KVM_ASYNC_PF_EN:
2763 if (kvm_pv_enable_async_pf(vcpu, data))
2766 case MSR_KVM_STEAL_TIME:
2768 if (unlikely(!sched_info_on()))
2771 if (data & KVM_STEAL_RESERVED_MASK)
2774 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2775 data & KVM_STEAL_VALID_BITS,
2776 sizeof(struct kvm_steal_time)))
2779 vcpu->arch.st.msr_val = data;
2781 if (!(data & KVM_MSR_ENABLED))
2784 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2787 case MSR_KVM_PV_EOI_EN:
2788 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2792 case MSR_KVM_POLL_CONTROL:
2793 /* only enable bit supported */
2794 if (data & (-1ULL << 1))
2797 vcpu->arch.msr_kvm_poll_control = data;
2800 case MSR_IA32_MCG_CTL:
2801 case MSR_IA32_MCG_STATUS:
2802 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2803 return set_msr_mce(vcpu, msr_info);
2805 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2806 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2807 pr = true; /* fall through */
2808 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2809 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2810 if (kvm_pmu_is_valid_msr(vcpu, msr))
2811 return kvm_pmu_set_msr(vcpu, msr_info);
2813 if (pr || data != 0)
2814 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2815 "0x%x data 0x%llx\n", msr, data);
2817 case MSR_K7_CLK_CTL:
2819 * Ignore all writes to this no longer documented MSR.
2820 * Writes are only relevant for old K7 processors,
2821 * all pre-dating SVM, but a recommended workaround from
2822 * AMD for these chips. It is possible to specify the
2823 * affected processor models on the command line, hence
2824 * the need to ignore the workaround.
2827 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2828 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2829 case HV_X64_MSR_CRASH_CTL:
2830 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2831 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2832 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2833 case HV_X64_MSR_TSC_EMULATION_STATUS:
2834 return kvm_hv_set_msr_common(vcpu, msr, data,
2835 msr_info->host_initiated);
2836 case MSR_IA32_BBL_CR_CTL3:
2837 /* Drop writes to this legacy MSR -- see rdmsr
2838 * counterpart for further detail.
2840 if (report_ignored_msrs)
2841 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2844 case MSR_AMD64_OSVW_ID_LENGTH:
2845 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2847 vcpu->arch.osvw.length = data;
2849 case MSR_AMD64_OSVW_STATUS:
2850 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2852 vcpu->arch.osvw.status = data;
2854 case MSR_PLATFORM_INFO:
2855 if (!msr_info->host_initiated ||
2856 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2857 cpuid_fault_enabled(vcpu)))
2859 vcpu->arch.msr_platform_info = data;
2861 case MSR_MISC_FEATURES_ENABLES:
2862 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2863 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2864 !supports_cpuid_fault(vcpu)))
2866 vcpu->arch.msr_misc_features_enables = data;
2869 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2870 return xen_hvm_config(vcpu, data);
2871 if (kvm_pmu_is_valid_msr(vcpu, msr))
2872 return kvm_pmu_set_msr(vcpu, msr_info);
2874 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2878 if (report_ignored_msrs)
2880 "ignored wrmsr: 0x%x data 0x%llx\n",
2887 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2889 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2892 u64 mcg_cap = vcpu->arch.mcg_cap;
2893 unsigned bank_num = mcg_cap & 0xff;
2896 case MSR_IA32_P5_MC_ADDR:
2897 case MSR_IA32_P5_MC_TYPE:
2900 case MSR_IA32_MCG_CAP:
2901 data = vcpu->arch.mcg_cap;
2903 case MSR_IA32_MCG_CTL:
2904 if (!(mcg_cap & MCG_CTL_P) && !host)
2906 data = vcpu->arch.mcg_ctl;
2908 case MSR_IA32_MCG_STATUS:
2909 data = vcpu->arch.mcg_status;
2912 if (msr >= MSR_IA32_MC0_CTL &&
2913 msr < MSR_IA32_MCx_CTL(bank_num)) {
2914 u32 offset = msr - MSR_IA32_MC0_CTL;
2915 data = vcpu->arch.mce_banks[offset];
2924 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2926 switch (msr_info->index) {
2927 case MSR_IA32_PLATFORM_ID:
2928 case MSR_IA32_EBL_CR_POWERON:
2929 case MSR_IA32_DEBUGCTLMSR:
2930 case MSR_IA32_LASTBRANCHFROMIP:
2931 case MSR_IA32_LASTBRANCHTOIP:
2932 case MSR_IA32_LASTINTFROMIP:
2933 case MSR_IA32_LASTINTTOIP:
2935 case MSR_K8_TSEG_ADDR:
2936 case MSR_K8_TSEG_MASK:
2937 case MSR_VM_HSAVE_PA:
2938 case MSR_K8_INT_PENDING_MSG:
2939 case MSR_AMD64_NB_CFG:
2940 case MSR_FAM10H_MMIO_CONF_BASE:
2941 case MSR_AMD64_BU_CFG2:
2942 case MSR_IA32_PERF_CTL:
2943 case MSR_AMD64_DC_CFG:
2944 case MSR_F15H_EX_CFG:
2947 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2948 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2949 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2950 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2951 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2952 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2953 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2956 case MSR_IA32_UCODE_REV:
2957 msr_info->data = vcpu->arch.microcode_version;
2959 case MSR_IA32_ARCH_CAPABILITIES:
2960 if (!msr_info->host_initiated &&
2961 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2963 msr_info->data = vcpu->arch.arch_capabilities;
2965 case MSR_IA32_POWER_CTL:
2966 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2969 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2972 case 0x200 ... 0x2ff:
2973 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2974 case 0xcd: /* fsb frequency */
2978 * MSR_EBC_FREQUENCY_ID
2979 * Conservative value valid for even the basic CPU models.
2980 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2981 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2982 * and 266MHz for model 3, or 4. Set Core Clock
2983 * Frequency to System Bus Frequency Ratio to 1 (bits
2984 * 31:24) even though these are only valid for CPU
2985 * models > 2, however guests may end up dividing or
2986 * multiplying by zero otherwise.
2988 case MSR_EBC_FREQUENCY_ID:
2989 msr_info->data = 1 << 24;
2991 case MSR_IA32_APICBASE:
2992 msr_info->data = kvm_get_apic_base(vcpu);
2994 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2995 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2997 case MSR_IA32_TSCDEADLINE:
2998 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
3000 case MSR_IA32_TSC_ADJUST:
3001 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
3003 case MSR_IA32_MISC_ENABLE:
3004 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
3006 case MSR_IA32_SMBASE:
3007 if (!msr_info->host_initiated)
3009 msr_info->data = vcpu->arch.smbase;
3012 msr_info->data = vcpu->arch.smi_count;
3014 case MSR_IA32_PERF_STATUS:
3015 /* TSC increment by tick */
3016 msr_info->data = 1000ULL;
3017 /* CPU multiplier */
3018 msr_info->data |= (((uint64_t)4ULL) << 40);
3021 msr_info->data = vcpu->arch.efer;
3023 case MSR_KVM_WALL_CLOCK:
3024 case MSR_KVM_WALL_CLOCK_NEW:
3025 msr_info->data = vcpu->kvm->arch.wall_clock;
3027 case MSR_KVM_SYSTEM_TIME:
3028 case MSR_KVM_SYSTEM_TIME_NEW:
3029 msr_info->data = vcpu->arch.time;
3031 case MSR_KVM_ASYNC_PF_EN:
3032 msr_info->data = vcpu->arch.apf.msr_val;
3034 case MSR_KVM_STEAL_TIME:
3035 msr_info->data = vcpu->arch.st.msr_val;
3037 case MSR_KVM_PV_EOI_EN:
3038 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3040 case MSR_KVM_POLL_CONTROL:
3041 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3043 case MSR_IA32_P5_MC_ADDR:
3044 case MSR_IA32_P5_MC_TYPE:
3045 case MSR_IA32_MCG_CAP:
3046 case MSR_IA32_MCG_CTL:
3047 case MSR_IA32_MCG_STATUS:
3048 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3049 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3050 msr_info->host_initiated);
3051 case MSR_K7_CLK_CTL:
3053 * Provide expected ramp-up count for K7. All other
3054 * are set to zero, indicating minimum divisors for
3057 * This prevents guest kernels on AMD host with CPU
3058 * type 6, model 8 and higher from exploding due to
3059 * the rdmsr failing.
3061 msr_info->data = 0x20000000;
3063 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3064 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3065 case HV_X64_MSR_CRASH_CTL:
3066 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3067 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3068 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3069 case HV_X64_MSR_TSC_EMULATION_STATUS:
3070 return kvm_hv_get_msr_common(vcpu,
3071 msr_info->index, &msr_info->data,
3072 msr_info->host_initiated);
3074 case MSR_IA32_BBL_CR_CTL3:
3075 /* This legacy MSR exists but isn't fully documented in current
3076 * silicon. It is however accessed by winxp in very narrow
3077 * scenarios where it sets bit #19, itself documented as
3078 * a "reserved" bit. Best effort attempt to source coherent
3079 * read data here should the balance of the register be
3080 * interpreted by the guest:
3082 * L2 cache control register 3: 64GB range, 256KB size,
3083 * enabled, latency 0x1, configured
3085 msr_info->data = 0xbe702111;
3087 case MSR_AMD64_OSVW_ID_LENGTH:
3088 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3090 msr_info->data = vcpu->arch.osvw.length;
3092 case MSR_AMD64_OSVW_STATUS:
3093 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3095 msr_info->data = vcpu->arch.osvw.status;
3097 case MSR_PLATFORM_INFO:
3098 if (!msr_info->host_initiated &&
3099 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3101 msr_info->data = vcpu->arch.msr_platform_info;
3103 case MSR_MISC_FEATURES_ENABLES:
3104 msr_info->data = vcpu->arch.msr_misc_features_enables;
3107 msr_info->data = vcpu->arch.msr_hwcr;
3110 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3111 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3113 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3117 if (report_ignored_msrs)
3118 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3126 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3129 * Read or write a bunch of msrs. All parameters are kernel addresses.
3131 * @return number of msrs set successfully.
3133 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3134 struct kvm_msr_entry *entries,
3135 int (*do_msr)(struct kvm_vcpu *vcpu,
3136 unsigned index, u64 *data))
3140 for (i = 0; i < msrs->nmsrs; ++i)
3141 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3148 * Read or write a bunch of msrs. Parameters are user addresses.
3150 * @return number of msrs set successfully.
3152 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3153 int (*do_msr)(struct kvm_vcpu *vcpu,
3154 unsigned index, u64 *data),
3157 struct kvm_msrs msrs;
3158 struct kvm_msr_entry *entries;
3163 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3167 if (msrs.nmsrs >= MAX_IO_MSRS)
3170 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3171 entries = memdup_user(user_msrs->entries, size);
3172 if (IS_ERR(entries)) {
3173 r = PTR_ERR(entries);
3177 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3182 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3193 static inline bool kvm_can_mwait_in_guest(void)
3195 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3196 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3197 boot_cpu_has(X86_FEATURE_ARAT);
3200 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3205 case KVM_CAP_IRQCHIP:
3207 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3208 case KVM_CAP_SET_TSS_ADDR:
3209 case KVM_CAP_EXT_CPUID:
3210 case KVM_CAP_EXT_EMUL_CPUID:
3211 case KVM_CAP_CLOCKSOURCE:
3213 case KVM_CAP_NOP_IO_DELAY:
3214 case KVM_CAP_MP_STATE:
3215 case KVM_CAP_SYNC_MMU:
3216 case KVM_CAP_USER_NMI:
3217 case KVM_CAP_REINJECT_CONTROL:
3218 case KVM_CAP_IRQ_INJECT_STATUS:
3219 case KVM_CAP_IOEVENTFD:
3220 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3222 case KVM_CAP_PIT_STATE2:
3223 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3224 case KVM_CAP_XEN_HVM:
3225 case KVM_CAP_VCPU_EVENTS:
3226 case KVM_CAP_HYPERV:
3227 case KVM_CAP_HYPERV_VAPIC:
3228 case KVM_CAP_HYPERV_SPIN:
3229 case KVM_CAP_HYPERV_SYNIC:
3230 case KVM_CAP_HYPERV_SYNIC2:
3231 case KVM_CAP_HYPERV_VP_INDEX:
3232 case KVM_CAP_HYPERV_EVENTFD:
3233 case KVM_CAP_HYPERV_TLBFLUSH:
3234 case KVM_CAP_HYPERV_SEND_IPI:
3235 case KVM_CAP_HYPERV_CPUID:
3236 case KVM_CAP_PCI_SEGMENT:
3237 case KVM_CAP_DEBUGREGS:
3238 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3240 case KVM_CAP_ASYNC_PF:
3241 case KVM_CAP_GET_TSC_KHZ:
3242 case KVM_CAP_KVMCLOCK_CTRL:
3243 case KVM_CAP_READONLY_MEM:
3244 case KVM_CAP_HYPERV_TIME:
3245 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3246 case KVM_CAP_TSC_DEADLINE_TIMER:
3247 case KVM_CAP_DISABLE_QUIRKS:
3248 case KVM_CAP_SET_BOOT_CPU_ID:
3249 case KVM_CAP_SPLIT_IRQCHIP:
3250 case KVM_CAP_IMMEDIATE_EXIT:
3251 case KVM_CAP_PMU_EVENT_FILTER:
3252 case KVM_CAP_GET_MSR_FEATURES:
3253 case KVM_CAP_MSR_PLATFORM_INFO:
3254 case KVM_CAP_EXCEPTION_PAYLOAD:
3257 case KVM_CAP_SYNC_REGS:
3258 r = KVM_SYNC_X86_VALID_FIELDS;
3260 case KVM_CAP_ADJUST_CLOCK:
3261 r = KVM_CLOCK_TSC_STABLE;
3263 case KVM_CAP_X86_DISABLE_EXITS:
3264 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3265 KVM_X86_DISABLE_EXITS_CSTATE;
3266 if(kvm_can_mwait_in_guest())
3267 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3269 case KVM_CAP_X86_SMM:
3270 /* SMBASE is usually relocated above 1M on modern chipsets,
3271 * and SMM handlers might indeed rely on 4G segment limits,
3272 * so do not report SMM to be available if real mode is
3273 * emulated via vm86 mode. Still, do not go to great lengths
3274 * to avoid userspace's usage of the feature, because it is a
3275 * fringe case that is not enabled except via specific settings
3276 * of the module parameters.
3278 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3281 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3283 case KVM_CAP_NR_VCPUS:
3284 r = KVM_SOFT_MAX_VCPUS;
3286 case KVM_CAP_MAX_VCPUS:
3289 case KVM_CAP_MAX_VCPU_ID:
3290 r = KVM_MAX_VCPU_ID;
3292 case KVM_CAP_PV_MMU: /* obsolete */
3296 r = KVM_MAX_MCE_BANKS;
3299 r = boot_cpu_has(X86_FEATURE_XSAVE);
3301 case KVM_CAP_TSC_CONTROL:
3302 r = kvm_has_tsc_control;
3304 case KVM_CAP_X2APIC_API:
3305 r = KVM_X2APIC_API_VALID_FLAGS;
3307 case KVM_CAP_NESTED_STATE:
3308 r = kvm_x86_ops->get_nested_state ?
3309 kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3311 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3312 r = kvm_x86_ops->enable_direct_tlbflush != NULL;
3314 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3315 r = kvm_x86_ops->nested_enable_evmcs != NULL;
3324 long kvm_arch_dev_ioctl(struct file *filp,
3325 unsigned int ioctl, unsigned long arg)
3327 void __user *argp = (void __user *)arg;
3331 case KVM_GET_MSR_INDEX_LIST: {
3332 struct kvm_msr_list __user *user_msr_list = argp;
3333 struct kvm_msr_list msr_list;
3337 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3340 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3341 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3344 if (n < msr_list.nmsrs)
3347 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3348 num_msrs_to_save * sizeof(u32)))
3350 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3352 num_emulated_msrs * sizeof(u32)))
3357 case KVM_GET_SUPPORTED_CPUID:
3358 case KVM_GET_EMULATED_CPUID: {
3359 struct kvm_cpuid2 __user *cpuid_arg = argp;
3360 struct kvm_cpuid2 cpuid;
3363 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3366 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3372 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3377 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3379 if (copy_to_user(argp, &kvm_mce_cap_supported,
3380 sizeof(kvm_mce_cap_supported)))
3384 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3385 struct kvm_msr_list __user *user_msr_list = argp;
3386 struct kvm_msr_list msr_list;
3390 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3393 msr_list.nmsrs = num_msr_based_features;
3394 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3397 if (n < msr_list.nmsrs)
3400 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3401 num_msr_based_features * sizeof(u32)))
3407 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3417 static void wbinvd_ipi(void *garbage)
3422 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3424 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3427 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3429 /* Address WBINVD may be executed by guest */
3430 if (need_emulate_wbinvd(vcpu)) {
3431 if (kvm_x86_ops->has_wbinvd_exit())
3432 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3433 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3434 smp_call_function_single(vcpu->cpu,
3435 wbinvd_ipi, NULL, 1);
3438 kvm_x86_ops->vcpu_load(vcpu, cpu);
3440 fpregs_assert_state_consistent();
3441 if (test_thread_flag(TIF_NEED_FPU_LOAD))
3442 switch_fpu_return();
3444 /* Apply any externally detected TSC adjustments (due to suspend) */
3445 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3446 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3447 vcpu->arch.tsc_offset_adjustment = 0;
3448 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3451 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3452 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3453 rdtsc() - vcpu->arch.last_host_tsc;
3455 mark_tsc_unstable("KVM discovered backwards TSC");
3457 if (kvm_check_tsc_unstable()) {
3458 u64 offset = kvm_compute_tsc_offset(vcpu,
3459 vcpu->arch.last_guest_tsc);
3460 kvm_vcpu_write_tsc_offset(vcpu, offset);
3461 vcpu->arch.tsc_catchup = 1;
3464 if (kvm_lapic_hv_timer_in_use(vcpu))
3465 kvm_lapic_restart_hv_timer(vcpu);
3468 * On a host with synchronized TSC, there is no need to update
3469 * kvmclock on vcpu->cpu migration
3471 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3472 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3473 if (vcpu->cpu != cpu)
3474 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3478 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3481 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3483 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3486 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3488 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3489 &vcpu->arch.st.steal.preempted,
3490 offsetof(struct kvm_steal_time, preempted),
3491 sizeof(vcpu->arch.st.steal.preempted));
3494 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3498 if (vcpu->preempted)
3499 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3502 * Disable page faults because we're in atomic context here.
3503 * kvm_write_guest_offset_cached() would call might_fault()
3504 * that relies on pagefault_disable() to tell if there's a
3505 * bug. NOTE: the write to guest memory may not go through if
3506 * during postcopy live migration or if there's heavy guest
3509 pagefault_disable();
3511 * kvm_memslots() will be called by
3512 * kvm_write_guest_offset_cached() so take the srcu lock.
3514 idx = srcu_read_lock(&vcpu->kvm->srcu);
3515 kvm_steal_time_set_preempted(vcpu);
3516 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3518 kvm_x86_ops->vcpu_put(vcpu);
3519 vcpu->arch.last_host_tsc = rdtsc();
3521 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3522 * on every vmexit, but if not, we might have a stale dr6 from the
3523 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3528 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3529 struct kvm_lapic_state *s)
3531 if (vcpu->arch.apicv_active)
3532 kvm_x86_ops->sync_pir_to_irr(vcpu);
3534 return kvm_apic_get_state(vcpu, s);
3537 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3538 struct kvm_lapic_state *s)
3542 r = kvm_apic_set_state(vcpu, s);
3545 update_cr8_intercept(vcpu);
3550 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3552 return (!lapic_in_kernel(vcpu) ||
3553 kvm_apic_accept_pic_intr(vcpu));
3557 * if userspace requested an interrupt window, check that the
3558 * interrupt window is open.
3560 * No need to exit to userspace if we already have an interrupt queued.
3562 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3564 return kvm_arch_interrupt_allowed(vcpu) &&
3565 !kvm_cpu_has_interrupt(vcpu) &&
3566 !kvm_event_needs_reinjection(vcpu) &&
3567 kvm_cpu_accept_dm_intr(vcpu);
3570 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3571 struct kvm_interrupt *irq)
3573 if (irq->irq >= KVM_NR_INTERRUPTS)
3576 if (!irqchip_in_kernel(vcpu->kvm)) {
3577 kvm_queue_interrupt(vcpu, irq->irq, false);
3578 kvm_make_request(KVM_REQ_EVENT, vcpu);
3583 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3584 * fail for in-kernel 8259.
3586 if (pic_in_kernel(vcpu->kvm))
3589 if (vcpu->arch.pending_external_vector != -1)
3592 vcpu->arch.pending_external_vector = irq->irq;
3593 kvm_make_request(KVM_REQ_EVENT, vcpu);
3597 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3599 kvm_inject_nmi(vcpu);
3604 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3606 kvm_make_request(KVM_REQ_SMI, vcpu);
3611 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3612 struct kvm_tpr_access_ctl *tac)
3616 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3620 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3624 unsigned bank_num = mcg_cap & 0xff, bank;
3627 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3629 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3632 vcpu->arch.mcg_cap = mcg_cap;
3633 /* Init IA32_MCG_CTL to all 1s */
3634 if (mcg_cap & MCG_CTL_P)
3635 vcpu->arch.mcg_ctl = ~(u64)0;
3636 /* Init IA32_MCi_CTL to all 1s */
3637 for (bank = 0; bank < bank_num; bank++)
3638 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3640 kvm_x86_ops->setup_mce(vcpu);
3645 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3646 struct kvm_x86_mce *mce)
3648 u64 mcg_cap = vcpu->arch.mcg_cap;
3649 unsigned bank_num = mcg_cap & 0xff;
3650 u64 *banks = vcpu->arch.mce_banks;
3652 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3655 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3656 * reporting is disabled
3658 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3659 vcpu->arch.mcg_ctl != ~(u64)0)
3661 banks += 4 * mce->bank;
3663 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3664 * reporting is disabled for the bank
3666 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3668 if (mce->status & MCI_STATUS_UC) {
3669 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3670 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3671 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3674 if (banks[1] & MCI_STATUS_VAL)
3675 mce->status |= MCI_STATUS_OVER;
3676 banks[2] = mce->addr;
3677 banks[3] = mce->misc;
3678 vcpu->arch.mcg_status = mce->mcg_status;
3679 banks[1] = mce->status;
3680 kvm_queue_exception(vcpu, MC_VECTOR);
3681 } else if (!(banks[1] & MCI_STATUS_VAL)
3682 || !(banks[1] & MCI_STATUS_UC)) {
3683 if (banks[1] & MCI_STATUS_VAL)
3684 mce->status |= MCI_STATUS_OVER;
3685 banks[2] = mce->addr;
3686 banks[3] = mce->misc;
3687 banks[1] = mce->status;
3689 banks[1] |= MCI_STATUS_OVER;
3693 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3694 struct kvm_vcpu_events *events)
3699 * The API doesn't provide the instruction length for software
3700 * exceptions, so don't report them. As long as the guest RIP
3701 * isn't advanced, we should expect to encounter the exception
3704 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3705 events->exception.injected = 0;
3706 events->exception.pending = 0;
3708 events->exception.injected = vcpu->arch.exception.injected;
3709 events->exception.pending = vcpu->arch.exception.pending;
3711 * For ABI compatibility, deliberately conflate
3712 * pending and injected exceptions when
3713 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3715 if (!vcpu->kvm->arch.exception_payload_enabled)
3716 events->exception.injected |=
3717 vcpu->arch.exception.pending;
3719 events->exception.nr = vcpu->arch.exception.nr;
3720 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3721 events->exception.error_code = vcpu->arch.exception.error_code;
3722 events->exception_has_payload = vcpu->arch.exception.has_payload;
3723 events->exception_payload = vcpu->arch.exception.payload;
3725 events->interrupt.injected =
3726 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3727 events->interrupt.nr = vcpu->arch.interrupt.nr;
3728 events->interrupt.soft = 0;
3729 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3731 events->nmi.injected = vcpu->arch.nmi_injected;
3732 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3733 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3734 events->nmi.pad = 0;
3736 events->sipi_vector = 0; /* never valid when reporting to user space */
3738 events->smi.smm = is_smm(vcpu);
3739 events->smi.pending = vcpu->arch.smi_pending;
3740 events->smi.smm_inside_nmi =
3741 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3742 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3744 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3745 | KVM_VCPUEVENT_VALID_SHADOW
3746 | KVM_VCPUEVENT_VALID_SMM);
3747 if (vcpu->kvm->arch.exception_payload_enabled)
3748 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3750 memset(&events->reserved, 0, sizeof(events->reserved));
3753 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3755 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3756 struct kvm_vcpu_events *events)
3758 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3759 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3760 | KVM_VCPUEVENT_VALID_SHADOW
3761 | KVM_VCPUEVENT_VALID_SMM
3762 | KVM_VCPUEVENT_VALID_PAYLOAD))
3765 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3766 if (!vcpu->kvm->arch.exception_payload_enabled)
3768 if (events->exception.pending)
3769 events->exception.injected = 0;
3771 events->exception_has_payload = 0;
3773 events->exception.pending = 0;
3774 events->exception_has_payload = 0;
3777 if ((events->exception.injected || events->exception.pending) &&
3778 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3781 /* INITs are latched while in SMM */
3782 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3783 (events->smi.smm || events->smi.pending) &&
3784 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3788 vcpu->arch.exception.injected = events->exception.injected;
3789 vcpu->arch.exception.pending = events->exception.pending;
3790 vcpu->arch.exception.nr = events->exception.nr;
3791 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3792 vcpu->arch.exception.error_code = events->exception.error_code;
3793 vcpu->arch.exception.has_payload = events->exception_has_payload;
3794 vcpu->arch.exception.payload = events->exception_payload;
3796 vcpu->arch.interrupt.injected = events->interrupt.injected;
3797 vcpu->arch.interrupt.nr = events->interrupt.nr;
3798 vcpu->arch.interrupt.soft = events->interrupt.soft;
3799 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3800 kvm_x86_ops->set_interrupt_shadow(vcpu,
3801 events->interrupt.shadow);
3803 vcpu->arch.nmi_injected = events->nmi.injected;
3804 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3805 vcpu->arch.nmi_pending = events->nmi.pending;
3806 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3808 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3809 lapic_in_kernel(vcpu))
3810 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3812 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3813 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3814 if (events->smi.smm)
3815 vcpu->arch.hflags |= HF_SMM_MASK;
3817 vcpu->arch.hflags &= ~HF_SMM_MASK;
3818 kvm_smm_changed(vcpu);
3821 vcpu->arch.smi_pending = events->smi.pending;
3823 if (events->smi.smm) {
3824 if (events->smi.smm_inside_nmi)
3825 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3827 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3828 if (lapic_in_kernel(vcpu)) {
3829 if (events->smi.latched_init)
3830 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3832 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3837 kvm_make_request(KVM_REQ_EVENT, vcpu);
3842 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3843 struct kvm_debugregs *dbgregs)
3847 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3848 kvm_get_dr(vcpu, 6, &val);
3850 dbgregs->dr7 = vcpu->arch.dr7;
3852 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3855 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3856 struct kvm_debugregs *dbgregs)
3861 if (dbgregs->dr6 & ~0xffffffffull)
3863 if (dbgregs->dr7 & ~0xffffffffull)
3866 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3867 kvm_update_dr0123(vcpu);
3868 vcpu->arch.dr6 = dbgregs->dr6;
3869 kvm_update_dr6(vcpu);
3870 vcpu->arch.dr7 = dbgregs->dr7;
3871 kvm_update_dr7(vcpu);
3876 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3878 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3880 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3881 u64 xstate_bv = xsave->header.xfeatures;
3885 * Copy legacy XSAVE area, to avoid complications with CPUID
3886 * leaves 0 and 1 in the loop below.
3888 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3891 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3892 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3895 * Copy each region from the possibly compacted offset to the
3896 * non-compacted offset.
3898 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3900 u64 xfeature_mask = valid & -valid;
3901 int xfeature_nr = fls64(xfeature_mask) - 1;
3902 void *src = get_xsave_addr(xsave, xfeature_nr);
3905 u32 size, offset, ecx, edx;
3906 cpuid_count(XSTATE_CPUID, xfeature_nr,
3907 &size, &offset, &ecx, &edx);
3908 if (xfeature_nr == XFEATURE_PKRU)
3909 memcpy(dest + offset, &vcpu->arch.pkru,
3910 sizeof(vcpu->arch.pkru));
3912 memcpy(dest + offset, src, size);
3916 valid -= xfeature_mask;
3920 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3922 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3923 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3927 * Copy legacy XSAVE area, to avoid complications with CPUID
3928 * leaves 0 and 1 in the loop below.
3930 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3932 /* Set XSTATE_BV and possibly XCOMP_BV. */
3933 xsave->header.xfeatures = xstate_bv;
3934 if (boot_cpu_has(X86_FEATURE_XSAVES))
3935 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3938 * Copy each region from the non-compacted offset to the
3939 * possibly compacted offset.
3941 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3943 u64 xfeature_mask = valid & -valid;
3944 int xfeature_nr = fls64(xfeature_mask) - 1;
3945 void *dest = get_xsave_addr(xsave, xfeature_nr);
3948 u32 size, offset, ecx, edx;
3949 cpuid_count(XSTATE_CPUID, xfeature_nr,
3950 &size, &offset, &ecx, &edx);
3951 if (xfeature_nr == XFEATURE_PKRU)
3952 memcpy(&vcpu->arch.pkru, src + offset,
3953 sizeof(vcpu->arch.pkru));
3955 memcpy(dest, src + offset, size);
3958 valid -= xfeature_mask;
3962 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3963 struct kvm_xsave *guest_xsave)
3965 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3966 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3967 fill_xsave((u8 *) guest_xsave->region, vcpu);
3969 memcpy(guest_xsave->region,
3970 &vcpu->arch.guest_fpu->state.fxsave,
3971 sizeof(struct fxregs_state));
3972 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3973 XFEATURE_MASK_FPSSE;
3977 #define XSAVE_MXCSR_OFFSET 24
3979 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3980 struct kvm_xsave *guest_xsave)
3983 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3984 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3986 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3988 * Here we allow setting states that are not present in
3989 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3990 * with old userspace.
3992 if (xstate_bv & ~kvm_supported_xcr0() ||
3993 mxcsr & ~mxcsr_feature_mask)
3995 load_xsave(vcpu, (u8 *)guest_xsave->region);
3997 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3998 mxcsr & ~mxcsr_feature_mask)
4000 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
4001 guest_xsave->region, sizeof(struct fxregs_state));
4006 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
4007 struct kvm_xcrs *guest_xcrs)
4009 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
4010 guest_xcrs->nr_xcrs = 0;
4014 guest_xcrs->nr_xcrs = 1;
4015 guest_xcrs->flags = 0;
4016 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4017 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4020 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4021 struct kvm_xcrs *guest_xcrs)
4025 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4028 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4031 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4032 /* Only support XCR0 currently */
4033 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4034 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4035 guest_xcrs->xcrs[i].value);
4044 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4045 * stopped by the hypervisor. This function will be called from the host only.
4046 * EINVAL is returned when the host attempts to set the flag for a guest that
4047 * does not support pv clocks.
4049 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4051 if (!vcpu->arch.pv_time_enabled)
4053 vcpu->arch.pvclock_set_guest_stopped_request = true;
4054 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4058 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4059 struct kvm_enable_cap *cap)
4062 uint16_t vmcs_version;
4063 void __user *user_ptr;
4069 case KVM_CAP_HYPERV_SYNIC2:
4074 case KVM_CAP_HYPERV_SYNIC:
4075 if (!irqchip_in_kernel(vcpu->kvm))
4077 return kvm_hv_activate_synic(vcpu, cap->cap ==
4078 KVM_CAP_HYPERV_SYNIC2);
4079 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4080 if (!kvm_x86_ops->nested_enable_evmcs)
4082 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4084 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4085 if (copy_to_user(user_ptr, &vmcs_version,
4086 sizeof(vmcs_version)))
4090 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4091 if (!kvm_x86_ops->enable_direct_tlbflush)
4094 return kvm_x86_ops->enable_direct_tlbflush(vcpu);
4101 long kvm_arch_vcpu_ioctl(struct file *filp,
4102 unsigned int ioctl, unsigned long arg)
4104 struct kvm_vcpu *vcpu = filp->private_data;
4105 void __user *argp = (void __user *)arg;
4108 struct kvm_lapic_state *lapic;
4109 struct kvm_xsave *xsave;
4110 struct kvm_xcrs *xcrs;
4118 case KVM_GET_LAPIC: {
4120 if (!lapic_in_kernel(vcpu))
4122 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4123 GFP_KERNEL_ACCOUNT);
4128 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4132 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4137 case KVM_SET_LAPIC: {
4139 if (!lapic_in_kernel(vcpu))
4141 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4142 if (IS_ERR(u.lapic)) {
4143 r = PTR_ERR(u.lapic);
4147 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4150 case KVM_INTERRUPT: {
4151 struct kvm_interrupt irq;
4154 if (copy_from_user(&irq, argp, sizeof(irq)))
4156 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4160 r = kvm_vcpu_ioctl_nmi(vcpu);
4164 r = kvm_vcpu_ioctl_smi(vcpu);
4167 case KVM_SET_CPUID: {
4168 struct kvm_cpuid __user *cpuid_arg = argp;
4169 struct kvm_cpuid cpuid;
4172 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4174 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4177 case KVM_SET_CPUID2: {
4178 struct kvm_cpuid2 __user *cpuid_arg = argp;
4179 struct kvm_cpuid2 cpuid;
4182 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4184 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4185 cpuid_arg->entries);
4188 case KVM_GET_CPUID2: {
4189 struct kvm_cpuid2 __user *cpuid_arg = argp;
4190 struct kvm_cpuid2 cpuid;
4193 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4195 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4196 cpuid_arg->entries);
4200 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4205 case KVM_GET_MSRS: {
4206 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4207 r = msr_io(vcpu, argp, do_get_msr, 1);
4208 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4211 case KVM_SET_MSRS: {
4212 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4213 r = msr_io(vcpu, argp, do_set_msr, 0);
4214 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4217 case KVM_TPR_ACCESS_REPORTING: {
4218 struct kvm_tpr_access_ctl tac;
4221 if (copy_from_user(&tac, argp, sizeof(tac)))
4223 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4227 if (copy_to_user(argp, &tac, sizeof(tac)))
4232 case KVM_SET_VAPIC_ADDR: {
4233 struct kvm_vapic_addr va;
4237 if (!lapic_in_kernel(vcpu))
4240 if (copy_from_user(&va, argp, sizeof(va)))
4242 idx = srcu_read_lock(&vcpu->kvm->srcu);
4243 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4244 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4247 case KVM_X86_SETUP_MCE: {
4251 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4253 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4256 case KVM_X86_SET_MCE: {
4257 struct kvm_x86_mce mce;
4260 if (copy_from_user(&mce, argp, sizeof(mce)))
4262 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4265 case KVM_GET_VCPU_EVENTS: {
4266 struct kvm_vcpu_events events;
4268 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4271 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4276 case KVM_SET_VCPU_EVENTS: {
4277 struct kvm_vcpu_events events;
4280 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4283 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4286 case KVM_GET_DEBUGREGS: {
4287 struct kvm_debugregs dbgregs;
4289 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4292 if (copy_to_user(argp, &dbgregs,
4293 sizeof(struct kvm_debugregs)))
4298 case KVM_SET_DEBUGREGS: {
4299 struct kvm_debugregs dbgregs;
4302 if (copy_from_user(&dbgregs, argp,
4303 sizeof(struct kvm_debugregs)))
4306 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4309 case KVM_GET_XSAVE: {
4310 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4315 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4318 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4323 case KVM_SET_XSAVE: {
4324 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4325 if (IS_ERR(u.xsave)) {
4326 r = PTR_ERR(u.xsave);
4330 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4333 case KVM_GET_XCRS: {
4334 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4339 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4342 if (copy_to_user(argp, u.xcrs,
4343 sizeof(struct kvm_xcrs)))
4348 case KVM_SET_XCRS: {
4349 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4350 if (IS_ERR(u.xcrs)) {
4351 r = PTR_ERR(u.xcrs);
4355 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4358 case KVM_SET_TSC_KHZ: {
4362 user_tsc_khz = (u32)arg;
4364 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4367 if (user_tsc_khz == 0)
4368 user_tsc_khz = tsc_khz;
4370 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4375 case KVM_GET_TSC_KHZ: {
4376 r = vcpu->arch.virtual_tsc_khz;
4379 case KVM_KVMCLOCK_CTRL: {
4380 r = kvm_set_guest_paused(vcpu);
4383 case KVM_ENABLE_CAP: {
4384 struct kvm_enable_cap cap;
4387 if (copy_from_user(&cap, argp, sizeof(cap)))
4389 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4392 case KVM_GET_NESTED_STATE: {
4393 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4397 if (!kvm_x86_ops->get_nested_state)
4400 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4402 if (get_user(user_data_size, &user_kvm_nested_state->size))
4405 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4410 if (r > user_data_size) {
4411 if (put_user(r, &user_kvm_nested_state->size))
4421 case KVM_SET_NESTED_STATE: {
4422 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4423 struct kvm_nested_state kvm_state;
4426 if (!kvm_x86_ops->set_nested_state)
4430 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4434 if (kvm_state.size < sizeof(kvm_state))
4437 if (kvm_state.flags &
4438 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4439 | KVM_STATE_NESTED_EVMCS))
4442 /* nested_run_pending implies guest_mode. */
4443 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4444 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4447 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4450 case KVM_GET_SUPPORTED_HV_CPUID: {
4451 struct kvm_cpuid2 __user *cpuid_arg = argp;
4452 struct kvm_cpuid2 cpuid;
4455 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4458 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4459 cpuid_arg->entries);
4464 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4479 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4481 return VM_FAULT_SIGBUS;
4484 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4488 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4490 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4494 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4497 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4500 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4501 unsigned long kvm_nr_mmu_pages)
4503 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4506 mutex_lock(&kvm->slots_lock);
4508 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4509 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4511 mutex_unlock(&kvm->slots_lock);
4515 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4517 return kvm->arch.n_max_mmu_pages;
4520 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4522 struct kvm_pic *pic = kvm->arch.vpic;
4526 switch (chip->chip_id) {
4527 case KVM_IRQCHIP_PIC_MASTER:
4528 memcpy(&chip->chip.pic, &pic->pics[0],
4529 sizeof(struct kvm_pic_state));
4531 case KVM_IRQCHIP_PIC_SLAVE:
4532 memcpy(&chip->chip.pic, &pic->pics[1],
4533 sizeof(struct kvm_pic_state));
4535 case KVM_IRQCHIP_IOAPIC:
4536 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4545 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4547 struct kvm_pic *pic = kvm->arch.vpic;
4551 switch (chip->chip_id) {
4552 case KVM_IRQCHIP_PIC_MASTER:
4553 spin_lock(&pic->lock);
4554 memcpy(&pic->pics[0], &chip->chip.pic,
4555 sizeof(struct kvm_pic_state));
4556 spin_unlock(&pic->lock);
4558 case KVM_IRQCHIP_PIC_SLAVE:
4559 spin_lock(&pic->lock);
4560 memcpy(&pic->pics[1], &chip->chip.pic,
4561 sizeof(struct kvm_pic_state));
4562 spin_unlock(&pic->lock);
4564 case KVM_IRQCHIP_IOAPIC:
4565 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4571 kvm_pic_update_irq(pic);
4575 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4577 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4579 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4581 mutex_lock(&kps->lock);
4582 memcpy(ps, &kps->channels, sizeof(*ps));
4583 mutex_unlock(&kps->lock);
4587 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4590 struct kvm_pit *pit = kvm->arch.vpit;
4592 mutex_lock(&pit->pit_state.lock);
4593 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4594 for (i = 0; i < 3; i++)
4595 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4596 mutex_unlock(&pit->pit_state.lock);
4600 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4602 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4603 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4604 sizeof(ps->channels));
4605 ps->flags = kvm->arch.vpit->pit_state.flags;
4606 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4607 memset(&ps->reserved, 0, sizeof(ps->reserved));
4611 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4615 u32 prev_legacy, cur_legacy;
4616 struct kvm_pit *pit = kvm->arch.vpit;
4618 mutex_lock(&pit->pit_state.lock);
4619 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4620 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4621 if (!prev_legacy && cur_legacy)
4623 memcpy(&pit->pit_state.channels, &ps->channels,
4624 sizeof(pit->pit_state.channels));
4625 pit->pit_state.flags = ps->flags;
4626 for (i = 0; i < 3; i++)
4627 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4629 mutex_unlock(&pit->pit_state.lock);
4633 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4634 struct kvm_reinject_control *control)
4636 struct kvm_pit *pit = kvm->arch.vpit;
4641 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4642 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4643 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4645 mutex_lock(&pit->pit_state.lock);
4646 kvm_pit_set_reinject(pit, control->pit_reinject);
4647 mutex_unlock(&pit->pit_state.lock);
4653 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4654 * @kvm: kvm instance
4655 * @log: slot id and address to which we copy the log
4657 * Steps 1-4 below provide general overview of dirty page logging. See
4658 * kvm_get_dirty_log_protect() function description for additional details.
4660 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4661 * always flush the TLB (step 4) even if previous step failed and the dirty
4662 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4663 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4664 * writes will be marked dirty for next log read.
4666 * 1. Take a snapshot of the bit and clear it if needed.
4667 * 2. Write protect the corresponding page.
4668 * 3. Copy the snapshot to the userspace.
4669 * 4. Flush TLB's if needed.
4671 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4676 mutex_lock(&kvm->slots_lock);
4679 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4681 if (kvm_x86_ops->flush_log_dirty)
4682 kvm_x86_ops->flush_log_dirty(kvm);
4684 r = kvm_get_dirty_log_protect(kvm, log, &flush);
4687 * All the TLBs can be flushed out of mmu lock, see the comments in
4688 * kvm_mmu_slot_remove_write_access().
4690 lockdep_assert_held(&kvm->slots_lock);
4692 kvm_flush_remote_tlbs(kvm);
4694 mutex_unlock(&kvm->slots_lock);
4698 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4703 mutex_lock(&kvm->slots_lock);
4706 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4708 if (kvm_x86_ops->flush_log_dirty)
4709 kvm_x86_ops->flush_log_dirty(kvm);
4711 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4714 * All the TLBs can be flushed out of mmu lock, see the comments in
4715 * kvm_mmu_slot_remove_write_access().
4717 lockdep_assert_held(&kvm->slots_lock);
4719 kvm_flush_remote_tlbs(kvm);
4721 mutex_unlock(&kvm->slots_lock);
4725 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4728 if (!irqchip_in_kernel(kvm))
4731 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4732 irq_event->irq, irq_event->level,
4737 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4738 struct kvm_enable_cap *cap)
4746 case KVM_CAP_DISABLE_QUIRKS:
4747 kvm->arch.disabled_quirks = cap->args[0];
4750 case KVM_CAP_SPLIT_IRQCHIP: {
4751 mutex_lock(&kvm->lock);
4753 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4754 goto split_irqchip_unlock;
4756 if (irqchip_in_kernel(kvm))
4757 goto split_irqchip_unlock;
4758 if (kvm->created_vcpus)
4759 goto split_irqchip_unlock;
4760 r = kvm_setup_empty_irq_routing(kvm);
4762 goto split_irqchip_unlock;
4763 /* Pairs with irqchip_in_kernel. */
4765 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4766 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4768 split_irqchip_unlock:
4769 mutex_unlock(&kvm->lock);
4772 case KVM_CAP_X2APIC_API:
4774 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4777 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4778 kvm->arch.x2apic_format = true;
4779 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4780 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4784 case KVM_CAP_X86_DISABLE_EXITS:
4786 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4789 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4790 kvm_can_mwait_in_guest())
4791 kvm->arch.mwait_in_guest = true;
4792 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4793 kvm->arch.hlt_in_guest = true;
4794 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4795 kvm->arch.pause_in_guest = true;
4796 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4797 kvm->arch.cstate_in_guest = true;
4800 case KVM_CAP_MSR_PLATFORM_INFO:
4801 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4804 case KVM_CAP_EXCEPTION_PAYLOAD:
4805 kvm->arch.exception_payload_enabled = cap->args[0];
4815 long kvm_arch_vm_ioctl(struct file *filp,
4816 unsigned int ioctl, unsigned long arg)
4818 struct kvm *kvm = filp->private_data;
4819 void __user *argp = (void __user *)arg;
4822 * This union makes it completely explicit to gcc-3.x
4823 * that these two variables' stack usage should be
4824 * combined, not added together.
4827 struct kvm_pit_state ps;
4828 struct kvm_pit_state2 ps2;
4829 struct kvm_pit_config pit_config;
4833 case KVM_SET_TSS_ADDR:
4834 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4836 case KVM_SET_IDENTITY_MAP_ADDR: {
4839 mutex_lock(&kvm->lock);
4841 if (kvm->created_vcpus)
4842 goto set_identity_unlock;
4844 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4845 goto set_identity_unlock;
4846 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4847 set_identity_unlock:
4848 mutex_unlock(&kvm->lock);
4851 case KVM_SET_NR_MMU_PAGES:
4852 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4854 case KVM_GET_NR_MMU_PAGES:
4855 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4857 case KVM_CREATE_IRQCHIP: {
4858 mutex_lock(&kvm->lock);
4861 if (irqchip_in_kernel(kvm))
4862 goto create_irqchip_unlock;
4865 if (kvm->created_vcpus)
4866 goto create_irqchip_unlock;
4868 r = kvm_pic_init(kvm);
4870 goto create_irqchip_unlock;
4872 r = kvm_ioapic_init(kvm);
4874 kvm_pic_destroy(kvm);
4875 goto create_irqchip_unlock;
4878 r = kvm_setup_default_irq_routing(kvm);
4880 kvm_ioapic_destroy(kvm);
4881 kvm_pic_destroy(kvm);
4882 goto create_irqchip_unlock;
4884 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4886 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4887 create_irqchip_unlock:
4888 mutex_unlock(&kvm->lock);
4891 case KVM_CREATE_PIT:
4892 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4894 case KVM_CREATE_PIT2:
4896 if (copy_from_user(&u.pit_config, argp,
4897 sizeof(struct kvm_pit_config)))
4900 mutex_lock(&kvm->lock);
4903 goto create_pit_unlock;
4905 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4909 mutex_unlock(&kvm->lock);
4911 case KVM_GET_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 get_irqchip_out;
4924 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4926 goto get_irqchip_out;
4928 if (copy_to_user(argp, chip, sizeof(*chip)))
4929 goto get_irqchip_out;
4935 case KVM_SET_IRQCHIP: {
4936 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4937 struct kvm_irqchip *chip;
4939 chip = memdup_user(argp, sizeof(*chip));
4946 if (!irqchip_kernel(kvm))
4947 goto set_irqchip_out;
4948 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4950 goto set_irqchip_out;
4958 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4961 if (!kvm->arch.vpit)
4963 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4967 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4974 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4977 if (!kvm->arch.vpit)
4979 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4982 case KVM_GET_PIT2: {
4984 if (!kvm->arch.vpit)
4986 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4990 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4995 case KVM_SET_PIT2: {
4997 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
5000 if (!kvm->arch.vpit)
5002 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
5005 case KVM_REINJECT_CONTROL: {
5006 struct kvm_reinject_control control;
5008 if (copy_from_user(&control, argp, sizeof(control)))
5010 r = kvm_vm_ioctl_reinject(kvm, &control);
5013 case KVM_SET_BOOT_CPU_ID:
5015 mutex_lock(&kvm->lock);
5016 if (kvm->created_vcpus)
5019 kvm->arch.bsp_vcpu_id = arg;
5020 mutex_unlock(&kvm->lock);
5022 case KVM_XEN_HVM_CONFIG: {
5023 struct kvm_xen_hvm_config xhc;
5025 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5030 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5034 case KVM_SET_CLOCK: {
5035 struct kvm_clock_data user_ns;
5039 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5048 * TODO: userspace has to take care of races with VCPU_RUN, so
5049 * kvm_gen_update_masterclock() can be cut down to locked
5050 * pvclock_update_vm_gtod_copy().
5052 kvm_gen_update_masterclock(kvm);
5053 now_ns = get_kvmclock_ns(kvm);
5054 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5055 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5058 case KVM_GET_CLOCK: {
5059 struct kvm_clock_data user_ns;
5062 now_ns = get_kvmclock_ns(kvm);
5063 user_ns.clock = now_ns;
5064 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5065 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5068 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5073 case KVM_MEMORY_ENCRYPT_OP: {
5075 if (kvm_x86_ops->mem_enc_op)
5076 r = kvm_x86_ops->mem_enc_op(kvm, argp);
5079 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5080 struct kvm_enc_region region;
5083 if (copy_from_user(®ion, argp, sizeof(region)))
5087 if (kvm_x86_ops->mem_enc_reg_region)
5088 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
5091 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5092 struct kvm_enc_region region;
5095 if (copy_from_user(®ion, argp, sizeof(region)))
5099 if (kvm_x86_ops->mem_enc_unreg_region)
5100 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
5103 case KVM_HYPERV_EVENTFD: {
5104 struct kvm_hyperv_eventfd hvevfd;
5107 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5109 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5112 case KVM_SET_PMU_EVENT_FILTER:
5113 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5122 static void kvm_init_msr_list(void)
5124 struct x86_pmu_capability x86_pmu;
5128 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5129 "Please update the fixed PMCs in msrs_to_saved_all[]");
5131 perf_get_x86_pmu_capability(&x86_pmu);
5133 num_msrs_to_save = 0;
5134 num_emulated_msrs = 0;
5135 num_msr_based_features = 0;
5137 for (i = 0; i < ARRAY_SIZE(msrs_to_save_all); i++) {
5138 if (rdmsr_safe(msrs_to_save_all[i], &dummy[0], &dummy[1]) < 0)
5142 * Even MSRs that are valid in the host may not be exposed
5143 * to the guests in some cases.
5145 switch (msrs_to_save_all[i]) {
5146 case MSR_IA32_BNDCFGS:
5147 if (!kvm_mpx_supported())
5151 if (!kvm_x86_ops->rdtscp_supported())
5154 case MSR_IA32_RTIT_CTL:
5155 case MSR_IA32_RTIT_STATUS:
5156 if (!kvm_x86_ops->pt_supported())
5159 case MSR_IA32_RTIT_CR3_MATCH:
5160 if (!kvm_x86_ops->pt_supported() ||
5161 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5164 case MSR_IA32_RTIT_OUTPUT_BASE:
5165 case MSR_IA32_RTIT_OUTPUT_MASK:
5166 if (!kvm_x86_ops->pt_supported() ||
5167 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5168 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5171 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5172 if (!kvm_x86_ops->pt_supported() ||
5173 msrs_to_save_all[i] - MSR_IA32_RTIT_ADDR0_A >=
5174 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5177 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5178 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5179 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5182 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5183 if (msrs_to_save_all[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5184 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5191 msrs_to_save[num_msrs_to_save++] = msrs_to_save_all[i];
5194 for (i = 0; i < ARRAY_SIZE(emulated_msrs_all); i++) {
5195 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs_all[i]))
5198 emulated_msrs[num_emulated_msrs++] = emulated_msrs_all[i];
5201 for (i = 0; i < ARRAY_SIZE(msr_based_features_all); i++) {
5202 struct kvm_msr_entry msr;
5204 msr.index = msr_based_features_all[i];
5205 if (kvm_get_msr_feature(&msr))
5208 msr_based_features[num_msr_based_features++] = msr_based_features_all[i];
5212 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5220 if (!(lapic_in_kernel(vcpu) &&
5221 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5222 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5233 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5240 if (!(lapic_in_kernel(vcpu) &&
5241 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5243 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5245 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5255 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5256 struct kvm_segment *var, int seg)
5258 kvm_x86_ops->set_segment(vcpu, var, seg);
5261 void kvm_get_segment(struct kvm_vcpu *vcpu,
5262 struct kvm_segment *var, int seg)
5264 kvm_x86_ops->get_segment(vcpu, var, seg);
5267 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5268 struct x86_exception *exception)
5272 BUG_ON(!mmu_is_nested(vcpu));
5274 /* NPT walks are always user-walks */
5275 access |= PFERR_USER_MASK;
5276 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5281 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5282 struct x86_exception *exception)
5284 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5285 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5288 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5289 struct x86_exception *exception)
5291 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5292 access |= PFERR_FETCH_MASK;
5293 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5296 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5297 struct x86_exception *exception)
5299 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5300 access |= PFERR_WRITE_MASK;
5301 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5304 /* uses this to access any guest's mapped memory without checking CPL */
5305 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5306 struct x86_exception *exception)
5308 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5311 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5312 struct kvm_vcpu *vcpu, u32 access,
5313 struct x86_exception *exception)
5316 int r = X86EMUL_CONTINUE;
5319 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5321 unsigned offset = addr & (PAGE_SIZE-1);
5322 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5325 if (gpa == UNMAPPED_GVA)
5326 return X86EMUL_PROPAGATE_FAULT;
5327 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5330 r = X86EMUL_IO_NEEDED;
5342 /* used for instruction fetching */
5343 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5344 gva_t addr, void *val, unsigned int bytes,
5345 struct x86_exception *exception)
5347 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5348 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5352 /* Inline kvm_read_guest_virt_helper for speed. */
5353 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5355 if (unlikely(gpa == UNMAPPED_GVA))
5356 return X86EMUL_PROPAGATE_FAULT;
5358 offset = addr & (PAGE_SIZE-1);
5359 if (WARN_ON(offset + bytes > PAGE_SIZE))
5360 bytes = (unsigned)PAGE_SIZE - offset;
5361 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5363 if (unlikely(ret < 0))
5364 return X86EMUL_IO_NEEDED;
5366 return X86EMUL_CONTINUE;
5369 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5370 gva_t addr, void *val, unsigned int bytes,
5371 struct x86_exception *exception)
5373 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5376 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5377 * is returned, but our callers are not ready for that and they blindly
5378 * call kvm_inject_page_fault. Ensure that they at least do not leak
5379 * uninitialized kernel stack memory into cr2 and error code.
5381 memset(exception, 0, sizeof(*exception));
5382 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5385 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5387 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5388 gva_t addr, void *val, unsigned int bytes,
5389 struct x86_exception *exception, bool system)
5391 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5394 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5395 access |= PFERR_USER_MASK;
5397 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5400 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5401 unsigned long addr, void *val, unsigned int bytes)
5403 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5404 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5406 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5409 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5410 struct kvm_vcpu *vcpu, u32 access,
5411 struct x86_exception *exception)
5414 int r = X86EMUL_CONTINUE;
5417 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5420 unsigned offset = addr & (PAGE_SIZE-1);
5421 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5424 if (gpa == UNMAPPED_GVA)
5425 return X86EMUL_PROPAGATE_FAULT;
5426 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5428 r = X86EMUL_IO_NEEDED;
5440 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5441 unsigned int bytes, struct x86_exception *exception,
5444 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5445 u32 access = PFERR_WRITE_MASK;
5447 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5448 access |= PFERR_USER_MASK;
5450 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5454 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5455 unsigned int bytes, struct x86_exception *exception)
5457 /* kvm_write_guest_virt_system can pull in tons of pages. */
5458 vcpu->arch.l1tf_flush_l1d = true;
5461 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5462 * is returned, but our callers are not ready for that and they blindly
5463 * call kvm_inject_page_fault. Ensure that they at least do not leak
5464 * uninitialized kernel stack memory into cr2 and error code.
5466 memset(exception, 0, sizeof(*exception));
5467 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5468 PFERR_WRITE_MASK, exception);
5470 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5472 int handle_ud(struct kvm_vcpu *vcpu)
5474 int emul_type = EMULTYPE_TRAP_UD;
5475 char sig[5]; /* ud2; .ascii "kvm" */
5476 struct x86_exception e;
5478 if (force_emulation_prefix &&
5479 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5480 sig, sizeof(sig), &e) == 0 &&
5481 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5482 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5483 emul_type = EMULTYPE_TRAP_UD_FORCED;
5486 return kvm_emulate_instruction(vcpu, emul_type);
5488 EXPORT_SYMBOL_GPL(handle_ud);
5490 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5491 gpa_t gpa, bool write)
5493 /* For APIC access vmexit */
5494 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5497 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5498 trace_vcpu_match_mmio(gva, gpa, write, true);
5505 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5506 gpa_t *gpa, struct x86_exception *exception,
5509 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5510 | (write ? PFERR_WRITE_MASK : 0);
5513 * currently PKRU is only applied to ept enabled guest so
5514 * there is no pkey in EPT page table for L1 guest or EPT
5515 * shadow page table for L2 guest.
5517 if (vcpu_match_mmio_gva(vcpu, gva)
5518 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5519 vcpu->arch.mmio_access, 0, access)) {
5520 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5521 (gva & (PAGE_SIZE - 1));
5522 trace_vcpu_match_mmio(gva, *gpa, write, false);
5526 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5528 if (*gpa == UNMAPPED_GVA)
5531 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5534 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5535 const void *val, int bytes)
5539 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5542 kvm_page_track_write(vcpu, gpa, val, bytes);
5546 struct read_write_emulator_ops {
5547 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5549 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5550 void *val, int bytes);
5551 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5552 int bytes, void *val);
5553 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5554 void *val, int bytes);
5558 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5560 if (vcpu->mmio_read_completed) {
5561 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5562 vcpu->mmio_fragments[0].gpa, val);
5563 vcpu->mmio_read_completed = 0;
5570 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5571 void *val, int bytes)
5573 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5576 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5577 void *val, int bytes)
5579 return emulator_write_phys(vcpu, gpa, val, bytes);
5582 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5584 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5585 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5588 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5589 void *val, int bytes)
5591 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5592 return X86EMUL_IO_NEEDED;
5595 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5596 void *val, int bytes)
5598 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5600 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5601 return X86EMUL_CONTINUE;
5604 static const struct read_write_emulator_ops read_emultor = {
5605 .read_write_prepare = read_prepare,
5606 .read_write_emulate = read_emulate,
5607 .read_write_mmio = vcpu_mmio_read,
5608 .read_write_exit_mmio = read_exit_mmio,
5611 static const struct read_write_emulator_ops write_emultor = {
5612 .read_write_emulate = write_emulate,
5613 .read_write_mmio = write_mmio,
5614 .read_write_exit_mmio = write_exit_mmio,
5618 static int emulator_read_write_onepage(unsigned long addr, void *val,
5620 struct x86_exception *exception,
5621 struct kvm_vcpu *vcpu,
5622 const struct read_write_emulator_ops *ops)
5626 bool write = ops->write;
5627 struct kvm_mmio_fragment *frag;
5628 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5631 * If the exit was due to a NPF we may already have a GPA.
5632 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5633 * Note, this cannot be used on string operations since string
5634 * operation using rep will only have the initial GPA from the NPF
5637 if (vcpu->arch.gpa_available &&
5638 emulator_can_use_gpa(ctxt) &&
5639 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5640 gpa = vcpu->arch.gpa_val;
5641 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5643 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5645 return X86EMUL_PROPAGATE_FAULT;
5648 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5649 return X86EMUL_CONTINUE;
5652 * Is this MMIO handled locally?
5654 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5655 if (handled == bytes)
5656 return X86EMUL_CONTINUE;
5662 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5663 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5667 return X86EMUL_CONTINUE;
5670 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5672 void *val, unsigned int bytes,
5673 struct x86_exception *exception,
5674 const struct read_write_emulator_ops *ops)
5676 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5680 if (ops->read_write_prepare &&
5681 ops->read_write_prepare(vcpu, val, bytes))
5682 return X86EMUL_CONTINUE;
5684 vcpu->mmio_nr_fragments = 0;
5686 /* Crossing a page boundary? */
5687 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5690 now = -addr & ~PAGE_MASK;
5691 rc = emulator_read_write_onepage(addr, val, now, exception,
5694 if (rc != X86EMUL_CONTINUE)
5697 if (ctxt->mode != X86EMUL_MODE_PROT64)
5703 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5705 if (rc != X86EMUL_CONTINUE)
5708 if (!vcpu->mmio_nr_fragments)
5711 gpa = vcpu->mmio_fragments[0].gpa;
5713 vcpu->mmio_needed = 1;
5714 vcpu->mmio_cur_fragment = 0;
5716 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5717 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5718 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5719 vcpu->run->mmio.phys_addr = gpa;
5721 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5724 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5728 struct x86_exception *exception)
5730 return emulator_read_write(ctxt, addr, val, bytes,
5731 exception, &read_emultor);
5734 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5738 struct x86_exception *exception)
5740 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5741 exception, &write_emultor);
5744 #define CMPXCHG_TYPE(t, ptr, old, new) \
5745 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5747 #ifdef CONFIG_X86_64
5748 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5750 # define CMPXCHG64(ptr, old, new) \
5751 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5754 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5759 struct x86_exception *exception)
5761 struct kvm_host_map map;
5762 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5767 /* guests cmpxchg8b have to be emulated atomically */
5768 if (bytes > 8 || (bytes & (bytes - 1)))
5771 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5773 if (gpa == UNMAPPED_GVA ||
5774 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5777 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5780 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5783 kaddr = map.hva + offset_in_page(gpa);
5787 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5790 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5793 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5796 exchanged = CMPXCHG64(kaddr, old, new);
5802 kvm_vcpu_unmap(vcpu, &map, true);
5805 return X86EMUL_CMPXCHG_FAILED;
5807 kvm_page_track_write(vcpu, gpa, new, bytes);
5809 return X86EMUL_CONTINUE;
5812 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5814 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5817 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5821 for (i = 0; i < vcpu->arch.pio.count; i++) {
5822 if (vcpu->arch.pio.in)
5823 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5824 vcpu->arch.pio.size, pd);
5826 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5827 vcpu->arch.pio.port, vcpu->arch.pio.size,
5831 pd += vcpu->arch.pio.size;
5836 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5837 unsigned short port, void *val,
5838 unsigned int count, bool in)
5840 vcpu->arch.pio.port = port;
5841 vcpu->arch.pio.in = in;
5842 vcpu->arch.pio.count = count;
5843 vcpu->arch.pio.size = size;
5845 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5846 vcpu->arch.pio.count = 0;
5850 vcpu->run->exit_reason = KVM_EXIT_IO;
5851 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5852 vcpu->run->io.size = size;
5853 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5854 vcpu->run->io.count = count;
5855 vcpu->run->io.port = port;
5860 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5861 int size, unsigned short port, void *val,
5864 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5867 if (vcpu->arch.pio.count)
5870 memset(vcpu->arch.pio_data, 0, size * count);
5872 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5875 memcpy(val, vcpu->arch.pio_data, size * count);
5876 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5877 vcpu->arch.pio.count = 0;
5884 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5885 int size, unsigned short port,
5886 const void *val, unsigned int count)
5888 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5890 memcpy(vcpu->arch.pio_data, val, size * count);
5891 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5892 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5895 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5897 return kvm_x86_ops->get_segment_base(vcpu, seg);
5900 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5902 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5905 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5907 if (!need_emulate_wbinvd(vcpu))
5908 return X86EMUL_CONTINUE;
5910 if (kvm_x86_ops->has_wbinvd_exit()) {
5911 int cpu = get_cpu();
5913 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5914 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5915 wbinvd_ipi, NULL, 1);
5917 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5920 return X86EMUL_CONTINUE;
5923 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5925 kvm_emulate_wbinvd_noskip(vcpu);
5926 return kvm_skip_emulated_instruction(vcpu);
5928 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5932 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5934 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5937 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5938 unsigned long *dest)
5940 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5943 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5944 unsigned long value)
5947 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5950 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5952 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5955 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5957 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5958 unsigned long value;
5962 value = kvm_read_cr0(vcpu);
5965 value = vcpu->arch.cr2;
5968 value = kvm_read_cr3(vcpu);
5971 value = kvm_read_cr4(vcpu);
5974 value = kvm_get_cr8(vcpu);
5977 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5984 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5986 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5991 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5994 vcpu->arch.cr2 = val;
5997 res = kvm_set_cr3(vcpu, val);
6000 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
6003 res = kvm_set_cr8(vcpu, val);
6006 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6013 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6015 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
6018 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6020 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
6023 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6025 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
6028 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6030 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
6033 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6035 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
6038 static unsigned long emulator_get_cached_segment_base(
6039 struct x86_emulate_ctxt *ctxt, int seg)
6041 return get_segment_base(emul_to_vcpu(ctxt), seg);
6044 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6045 struct desc_struct *desc, u32 *base3,
6048 struct kvm_segment var;
6050 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6051 *selector = var.selector;
6054 memset(desc, 0, sizeof(*desc));
6062 set_desc_limit(desc, var.limit);
6063 set_desc_base(desc, (unsigned long)var.base);
6064 #ifdef CONFIG_X86_64
6066 *base3 = var.base >> 32;
6068 desc->type = var.type;
6070 desc->dpl = var.dpl;
6071 desc->p = var.present;
6072 desc->avl = var.avl;
6080 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6081 struct desc_struct *desc, u32 base3,
6084 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6085 struct kvm_segment var;
6087 var.selector = selector;
6088 var.base = get_desc_base(desc);
6089 #ifdef CONFIG_X86_64
6090 var.base |= ((u64)base3) << 32;
6092 var.limit = get_desc_limit(desc);
6094 var.limit = (var.limit << 12) | 0xfff;
6095 var.type = desc->type;
6096 var.dpl = desc->dpl;
6101 var.avl = desc->avl;
6102 var.present = desc->p;
6103 var.unusable = !var.present;
6106 kvm_set_segment(vcpu, &var, seg);
6110 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6111 u32 msr_index, u64 *pdata)
6113 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6116 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6117 u32 msr_index, u64 data)
6119 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6122 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6124 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6126 return vcpu->arch.smbase;
6129 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6131 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6133 vcpu->arch.smbase = smbase;
6136 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6139 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
6142 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6143 u32 pmc, u64 *pdata)
6145 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6148 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6150 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6153 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6154 struct x86_instruction_info *info,
6155 enum x86_intercept_stage stage)
6157 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6160 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6161 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6163 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6166 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6168 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6171 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6173 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6176 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6178 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6181 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6183 return emul_to_vcpu(ctxt)->arch.hflags;
6186 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6188 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6191 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6192 const char *smstate)
6194 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6197 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6199 kvm_smm_changed(emul_to_vcpu(ctxt));
6202 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6204 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6207 static const struct x86_emulate_ops emulate_ops = {
6208 .read_gpr = emulator_read_gpr,
6209 .write_gpr = emulator_write_gpr,
6210 .read_std = emulator_read_std,
6211 .write_std = emulator_write_std,
6212 .read_phys = kvm_read_guest_phys_system,
6213 .fetch = kvm_fetch_guest_virt,
6214 .read_emulated = emulator_read_emulated,
6215 .write_emulated = emulator_write_emulated,
6216 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6217 .invlpg = emulator_invlpg,
6218 .pio_in_emulated = emulator_pio_in_emulated,
6219 .pio_out_emulated = emulator_pio_out_emulated,
6220 .get_segment = emulator_get_segment,
6221 .set_segment = emulator_set_segment,
6222 .get_cached_segment_base = emulator_get_cached_segment_base,
6223 .get_gdt = emulator_get_gdt,
6224 .get_idt = emulator_get_idt,
6225 .set_gdt = emulator_set_gdt,
6226 .set_idt = emulator_set_idt,
6227 .get_cr = emulator_get_cr,
6228 .set_cr = emulator_set_cr,
6229 .cpl = emulator_get_cpl,
6230 .get_dr = emulator_get_dr,
6231 .set_dr = emulator_set_dr,
6232 .get_smbase = emulator_get_smbase,
6233 .set_smbase = emulator_set_smbase,
6234 .set_msr = emulator_set_msr,
6235 .get_msr = emulator_get_msr,
6236 .check_pmc = emulator_check_pmc,
6237 .read_pmc = emulator_read_pmc,
6238 .halt = emulator_halt,
6239 .wbinvd = emulator_wbinvd,
6240 .fix_hypercall = emulator_fix_hypercall,
6241 .intercept = emulator_intercept,
6242 .get_cpuid = emulator_get_cpuid,
6243 .set_nmi_mask = emulator_set_nmi_mask,
6244 .get_hflags = emulator_get_hflags,
6245 .set_hflags = emulator_set_hflags,
6246 .pre_leave_smm = emulator_pre_leave_smm,
6247 .post_leave_smm = emulator_post_leave_smm,
6248 .set_xcr = emulator_set_xcr,
6251 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6253 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6255 * an sti; sti; sequence only disable interrupts for the first
6256 * instruction. So, if the last instruction, be it emulated or
6257 * not, left the system with the INT_STI flag enabled, it
6258 * means that the last instruction is an sti. We should not
6259 * leave the flag on in this case. The same goes for mov ss
6261 if (int_shadow & mask)
6263 if (unlikely(int_shadow || mask)) {
6264 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6266 kvm_make_request(KVM_REQ_EVENT, vcpu);
6270 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6272 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6273 if (ctxt->exception.vector == PF_VECTOR)
6274 return kvm_propagate_fault(vcpu, &ctxt->exception);
6276 if (ctxt->exception.error_code_valid)
6277 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6278 ctxt->exception.error_code);
6280 kvm_queue_exception(vcpu, ctxt->exception.vector);
6284 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6286 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6289 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6291 ctxt->eflags = kvm_get_rflags(vcpu);
6292 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6294 ctxt->eip = kvm_rip_read(vcpu);
6295 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6296 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6297 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6298 cs_db ? X86EMUL_MODE_PROT32 :
6299 X86EMUL_MODE_PROT16;
6300 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6301 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6302 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6304 init_decode_cache(ctxt);
6305 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6308 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6310 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6313 init_emulate_ctxt(vcpu);
6317 ctxt->_eip = ctxt->eip + inc_eip;
6318 ret = emulate_int_real(ctxt, irq);
6320 if (ret != X86EMUL_CONTINUE) {
6321 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6323 ctxt->eip = ctxt->_eip;
6324 kvm_rip_write(vcpu, ctxt->eip);
6325 kvm_set_rflags(vcpu, ctxt->eflags);
6328 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6330 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6332 ++vcpu->stat.insn_emulation_fail;
6333 trace_kvm_emulate_insn_failed(vcpu);
6335 if (emulation_type & EMULTYPE_VMWARE_GP) {
6336 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6340 if (emulation_type & EMULTYPE_SKIP) {
6341 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6342 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6343 vcpu->run->internal.ndata = 0;
6347 kvm_queue_exception(vcpu, UD_VECTOR);
6349 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6350 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6351 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6352 vcpu->run->internal.ndata = 0;
6359 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6360 bool write_fault_to_shadow_pgtable,
6366 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6369 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6372 if (!vcpu->arch.mmu->direct_map) {
6374 * Write permission should be allowed since only
6375 * write access need to be emulated.
6377 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6380 * If the mapping is invalid in guest, let cpu retry
6381 * it to generate fault.
6383 if (gpa == UNMAPPED_GVA)
6388 * Do not retry the unhandleable instruction if it faults on the
6389 * readonly host memory, otherwise it will goto a infinite loop:
6390 * retry instruction -> write #PF -> emulation fail -> retry
6391 * instruction -> ...
6393 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6396 * If the instruction failed on the error pfn, it can not be fixed,
6397 * report the error to userspace.
6399 if (is_error_noslot_pfn(pfn))
6402 kvm_release_pfn_clean(pfn);
6404 /* The instructions are well-emulated on direct mmu. */
6405 if (vcpu->arch.mmu->direct_map) {
6406 unsigned int indirect_shadow_pages;
6408 spin_lock(&vcpu->kvm->mmu_lock);
6409 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6410 spin_unlock(&vcpu->kvm->mmu_lock);
6412 if (indirect_shadow_pages)
6413 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6419 * if emulation was due to access to shadowed page table
6420 * and it failed try to unshadow page and re-enter the
6421 * guest to let CPU execute the instruction.
6423 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6426 * If the access faults on its page table, it can not
6427 * be fixed by unprotecting shadow page and it should
6428 * be reported to userspace.
6430 return !write_fault_to_shadow_pgtable;
6433 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6434 unsigned long cr2, int emulation_type)
6436 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6437 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6439 last_retry_eip = vcpu->arch.last_retry_eip;
6440 last_retry_addr = vcpu->arch.last_retry_addr;
6443 * If the emulation is caused by #PF and it is non-page_table
6444 * writing instruction, it means the VM-EXIT is caused by shadow
6445 * page protected, we can zap the shadow page and retry this
6446 * instruction directly.
6448 * Note: if the guest uses a non-page-table modifying instruction
6449 * on the PDE that points to the instruction, then we will unmap
6450 * the instruction and go to an infinite loop. So, we cache the
6451 * last retried eip and the last fault address, if we meet the eip
6452 * and the address again, we can break out of the potential infinite
6455 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6457 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6460 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6463 if (x86_page_table_writing_insn(ctxt))
6466 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6469 vcpu->arch.last_retry_eip = ctxt->eip;
6470 vcpu->arch.last_retry_addr = cr2;
6472 if (!vcpu->arch.mmu->direct_map)
6473 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6475 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6480 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6481 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6483 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6485 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6486 /* This is a good place to trace that we are exiting SMM. */
6487 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6489 /* Process a latched INIT or SMI, if any. */
6490 kvm_make_request(KVM_REQ_EVENT, vcpu);
6493 kvm_mmu_reset_context(vcpu);
6496 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6505 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6506 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6511 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6513 struct kvm_run *kvm_run = vcpu->run;
6515 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6516 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6517 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6518 kvm_run->debug.arch.exception = DB_VECTOR;
6519 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6522 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6526 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6528 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6531 r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6536 * rflags is the old, "raw" value of the flags. The new value has
6537 * not been saved yet.
6539 * This is correct even for TF set by the guest, because "the
6540 * processor will not generate this exception after the instruction
6541 * that sets the TF flag".
6543 if (unlikely(rflags & X86_EFLAGS_TF))
6544 r = kvm_vcpu_do_singlestep(vcpu);
6547 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6549 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6551 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6552 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6553 struct kvm_run *kvm_run = vcpu->run;
6554 unsigned long eip = kvm_get_linear_rip(vcpu);
6555 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6556 vcpu->arch.guest_debug_dr7,
6560 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6561 kvm_run->debug.arch.pc = eip;
6562 kvm_run->debug.arch.exception = DB_VECTOR;
6563 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6569 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6570 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6571 unsigned long eip = kvm_get_linear_rip(vcpu);
6572 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6577 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6578 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6579 kvm_queue_exception(vcpu, DB_VECTOR);
6588 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6590 switch (ctxt->opcode_len) {
6597 case 0xe6: /* OUT */
6601 case 0x6c: /* INS */
6603 case 0x6e: /* OUTS */
6610 case 0x33: /* RDPMC */
6619 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6626 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6627 bool writeback = true;
6628 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6630 vcpu->arch.l1tf_flush_l1d = true;
6633 * Clear write_fault_to_shadow_pgtable here to ensure it is
6636 vcpu->arch.write_fault_to_shadow_pgtable = false;
6637 kvm_clear_exception_queue(vcpu);
6639 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6640 init_emulate_ctxt(vcpu);
6643 * We will reenter on the same instruction since
6644 * we do not set complete_userspace_io. This does not
6645 * handle watchpoints yet, those would be handled in
6648 if (!(emulation_type & EMULTYPE_SKIP) &&
6649 kvm_vcpu_check_breakpoint(vcpu, &r))
6652 ctxt->interruptibility = 0;
6653 ctxt->have_exception = false;
6654 ctxt->exception.vector = -1;
6655 ctxt->perm_ok = false;
6657 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6659 r = x86_decode_insn(ctxt, insn, insn_len);
6661 trace_kvm_emulate_insn_start(vcpu);
6662 ++vcpu->stat.insn_emulation;
6663 if (r != EMULATION_OK) {
6664 if ((emulation_type & EMULTYPE_TRAP_UD) ||
6665 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6666 kvm_queue_exception(vcpu, UD_VECTOR);
6669 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6672 if (ctxt->have_exception) {
6674 * #UD should result in just EMULATION_FAILED, and trap-like
6675 * exception should not be encountered during decode.
6677 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6678 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6679 inject_emulated_exception(vcpu);
6682 return handle_emulation_failure(vcpu, emulation_type);
6686 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6687 !is_vmware_backdoor_opcode(ctxt)) {
6688 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6693 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6694 * for kvm_skip_emulated_instruction(). The caller is responsible for
6695 * updating interruptibility state and injecting single-step #DBs.
6697 if (emulation_type & EMULTYPE_SKIP) {
6698 kvm_rip_write(vcpu, ctxt->_eip);
6699 if (ctxt->eflags & X86_EFLAGS_RF)
6700 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6704 if (retry_instruction(ctxt, cr2, emulation_type))
6707 /* this is needed for vmware backdoor interface to work since it
6708 changes registers values during IO operation */
6709 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6710 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6711 emulator_invalidate_register_cache(ctxt);
6715 /* Save the faulting GPA (cr2) in the address field */
6716 ctxt->exception.address = cr2;
6718 r = x86_emulate_insn(ctxt);
6720 if (r == EMULATION_INTERCEPTED)
6723 if (r == EMULATION_FAILED) {
6724 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6728 return handle_emulation_failure(vcpu, emulation_type);
6731 if (ctxt->have_exception) {
6733 if (inject_emulated_exception(vcpu))
6735 } else if (vcpu->arch.pio.count) {
6736 if (!vcpu->arch.pio.in) {
6737 /* FIXME: return into emulator if single-stepping. */
6738 vcpu->arch.pio.count = 0;
6741 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6744 } else if (vcpu->mmio_needed) {
6745 ++vcpu->stat.mmio_exits;
6747 if (!vcpu->mmio_is_write)
6750 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6751 } else if (r == EMULATION_RESTART)
6757 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6758 toggle_interruptibility(vcpu, ctxt->interruptibility);
6759 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6760 if (!ctxt->have_exception ||
6761 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6762 kvm_rip_write(vcpu, ctxt->eip);
6764 r = kvm_vcpu_do_singlestep(vcpu);
6765 __kvm_set_rflags(vcpu, ctxt->eflags);
6769 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6770 * do nothing, and it will be requested again as soon as
6771 * the shadow expires. But we still need to check here,
6772 * because POPF has no interrupt shadow.
6774 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6775 kvm_make_request(KVM_REQ_EVENT, vcpu);
6777 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6782 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6784 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6786 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6788 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6789 void *insn, int insn_len)
6791 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6793 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6795 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6797 vcpu->arch.pio.count = 0;
6801 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6803 vcpu->arch.pio.count = 0;
6805 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6808 return kvm_skip_emulated_instruction(vcpu);
6811 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6812 unsigned short port)
6814 unsigned long val = kvm_rax_read(vcpu);
6815 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6816 size, port, &val, 1);
6821 * Workaround userspace that relies on old KVM behavior of %rip being
6822 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6825 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6826 vcpu->arch.complete_userspace_io =
6827 complete_fast_pio_out_port_0x7e;
6828 kvm_skip_emulated_instruction(vcpu);
6830 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6831 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6836 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6840 /* We should only ever be called with arch.pio.count equal to 1 */
6841 BUG_ON(vcpu->arch.pio.count != 1);
6843 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6844 vcpu->arch.pio.count = 0;
6848 /* For size less than 4 we merge, else we zero extend */
6849 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6852 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6853 * the copy and tracing
6855 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6856 vcpu->arch.pio.port, &val, 1);
6857 kvm_rax_write(vcpu, val);
6859 return kvm_skip_emulated_instruction(vcpu);
6862 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6863 unsigned short port)
6868 /* For size less than 4 we merge, else we zero extend */
6869 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6871 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6874 kvm_rax_write(vcpu, val);
6878 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6879 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6884 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6889 ret = kvm_fast_pio_in(vcpu, size, port);
6891 ret = kvm_fast_pio_out(vcpu, size, port);
6892 return ret && kvm_skip_emulated_instruction(vcpu);
6894 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6896 static int kvmclock_cpu_down_prep(unsigned int cpu)
6898 __this_cpu_write(cpu_tsc_khz, 0);
6902 static void tsc_khz_changed(void *data)
6904 struct cpufreq_freqs *freq = data;
6905 unsigned long khz = 0;
6909 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6910 khz = cpufreq_quick_get(raw_smp_processor_id());
6913 __this_cpu_write(cpu_tsc_khz, khz);
6916 #ifdef CONFIG_X86_64
6917 static void kvm_hyperv_tsc_notifier(void)
6920 struct kvm_vcpu *vcpu;
6923 mutex_lock(&kvm_lock);
6924 list_for_each_entry(kvm, &vm_list, vm_list)
6925 kvm_make_mclock_inprogress_request(kvm);
6927 hyperv_stop_tsc_emulation();
6929 /* TSC frequency always matches when on Hyper-V */
6930 for_each_present_cpu(cpu)
6931 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6932 kvm_max_guest_tsc_khz = tsc_khz;
6934 list_for_each_entry(kvm, &vm_list, vm_list) {
6935 struct kvm_arch *ka = &kvm->arch;
6937 spin_lock(&ka->pvclock_gtod_sync_lock);
6939 pvclock_update_vm_gtod_copy(kvm);
6941 kvm_for_each_vcpu(cpu, vcpu, kvm)
6942 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6944 kvm_for_each_vcpu(cpu, vcpu, kvm)
6945 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6947 spin_unlock(&ka->pvclock_gtod_sync_lock);
6949 mutex_unlock(&kvm_lock);
6953 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6956 struct kvm_vcpu *vcpu;
6957 int i, send_ipi = 0;
6960 * We allow guests to temporarily run on slowing clocks,
6961 * provided we notify them after, or to run on accelerating
6962 * clocks, provided we notify them before. Thus time never
6965 * However, we have a problem. We can't atomically update
6966 * the frequency of a given CPU from this function; it is
6967 * merely a notifier, which can be called from any CPU.
6968 * Changing the TSC frequency at arbitrary points in time
6969 * requires a recomputation of local variables related to
6970 * the TSC for each VCPU. We must flag these local variables
6971 * to be updated and be sure the update takes place with the
6972 * new frequency before any guests proceed.
6974 * Unfortunately, the combination of hotplug CPU and frequency
6975 * change creates an intractable locking scenario; the order
6976 * of when these callouts happen is undefined with respect to
6977 * CPU hotplug, and they can race with each other. As such,
6978 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6979 * undefined; you can actually have a CPU frequency change take
6980 * place in between the computation of X and the setting of the
6981 * variable. To protect against this problem, all updates of
6982 * the per_cpu tsc_khz variable are done in an interrupt
6983 * protected IPI, and all callers wishing to update the value
6984 * must wait for a synchronous IPI to complete (which is trivial
6985 * if the caller is on the CPU already). This establishes the
6986 * necessary total order on variable updates.
6988 * Note that because a guest time update may take place
6989 * anytime after the setting of the VCPU's request bit, the
6990 * correct TSC value must be set before the request. However,
6991 * to ensure the update actually makes it to any guest which
6992 * starts running in hardware virtualization between the set
6993 * and the acquisition of the spinlock, we must also ping the
6994 * CPU after setting the request bit.
6998 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7000 mutex_lock(&kvm_lock);
7001 list_for_each_entry(kvm, &vm_list, vm_list) {
7002 kvm_for_each_vcpu(i, vcpu, kvm) {
7003 if (vcpu->cpu != cpu)
7005 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7006 if (vcpu->cpu != raw_smp_processor_id())
7010 mutex_unlock(&kvm_lock);
7012 if (freq->old < freq->new && send_ipi) {
7014 * We upscale the frequency. Must make the guest
7015 * doesn't see old kvmclock values while running with
7016 * the new frequency, otherwise we risk the guest sees
7017 * time go backwards.
7019 * In case we update the frequency for another cpu
7020 * (which might be in guest context) send an interrupt
7021 * to kick the cpu out of guest context. Next time
7022 * guest context is entered kvmclock will be updated,
7023 * so the guest will not see stale values.
7025 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7029 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7032 struct cpufreq_freqs *freq = data;
7035 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7037 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7040 for_each_cpu(cpu, freq->policy->cpus)
7041 __kvmclock_cpufreq_notifier(freq, cpu);
7046 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7047 .notifier_call = kvmclock_cpufreq_notifier
7050 static int kvmclock_cpu_online(unsigned int cpu)
7052 tsc_khz_changed(NULL);
7056 static void kvm_timer_init(void)
7058 max_tsc_khz = tsc_khz;
7060 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7061 #ifdef CONFIG_CPU_FREQ
7062 struct cpufreq_policy policy;
7065 memset(&policy, 0, sizeof(policy));
7067 cpufreq_get_policy(&policy, cpu);
7068 if (policy.cpuinfo.max_freq)
7069 max_tsc_khz = policy.cpuinfo.max_freq;
7072 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7073 CPUFREQ_TRANSITION_NOTIFIER);
7076 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7077 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7080 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7081 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7083 int kvm_is_in_guest(void)
7085 return __this_cpu_read(current_vcpu) != NULL;
7088 static int kvm_is_user_mode(void)
7092 if (__this_cpu_read(current_vcpu))
7093 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7095 return user_mode != 0;
7098 static unsigned long kvm_get_guest_ip(void)
7100 unsigned long ip = 0;
7102 if (__this_cpu_read(current_vcpu))
7103 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7108 static void kvm_handle_intel_pt_intr(void)
7110 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7112 kvm_make_request(KVM_REQ_PMI, vcpu);
7113 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7114 (unsigned long *)&vcpu->arch.pmu.global_status);
7117 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7118 .is_in_guest = kvm_is_in_guest,
7119 .is_user_mode = kvm_is_user_mode,
7120 .get_guest_ip = kvm_get_guest_ip,
7121 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7124 #ifdef CONFIG_X86_64
7125 static void pvclock_gtod_update_fn(struct work_struct *work)
7129 struct kvm_vcpu *vcpu;
7132 mutex_lock(&kvm_lock);
7133 list_for_each_entry(kvm, &vm_list, vm_list)
7134 kvm_for_each_vcpu(i, vcpu, kvm)
7135 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7136 atomic_set(&kvm_guest_has_master_clock, 0);
7137 mutex_unlock(&kvm_lock);
7140 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7143 * Notification about pvclock gtod data update.
7145 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7148 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7149 struct timekeeper *tk = priv;
7151 update_pvclock_gtod(tk);
7153 /* disable master clock if host does not trust, or does not
7154 * use, TSC based clocksource.
7156 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7157 atomic_read(&kvm_guest_has_master_clock) != 0)
7158 queue_work(system_long_wq, &pvclock_gtod_work);
7163 static struct notifier_block pvclock_gtod_notifier = {
7164 .notifier_call = pvclock_gtod_notify,
7168 int kvm_arch_init(void *opaque)
7171 struct kvm_x86_ops *ops = opaque;
7174 printk(KERN_ERR "kvm: already loaded the other module\n");
7179 if (!ops->cpu_has_kvm_support()) {
7180 printk(KERN_ERR "kvm: no hardware support\n");
7184 if (ops->disabled_by_bios()) {
7185 printk(KERN_ERR "kvm: disabled by bios\n");
7191 * KVM explicitly assumes that the guest has an FPU and
7192 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7193 * vCPU's FPU state as a fxregs_state struct.
7195 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7196 printk(KERN_ERR "kvm: inadequate fpu\n");
7202 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7203 __alignof__(struct fpu), SLAB_ACCOUNT,
7205 if (!x86_fpu_cache) {
7206 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7210 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7212 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7213 goto out_free_x86_fpu_cache;
7216 r = kvm_mmu_module_init();
7218 goto out_free_percpu;
7222 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7223 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7224 PT_PRESENT_MASK, 0, sme_me_mask);
7227 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7229 if (boot_cpu_has(X86_FEATURE_XSAVE))
7230 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7233 if (pi_inject_timer == -1)
7234 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7235 #ifdef CONFIG_X86_64
7236 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7238 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7239 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7245 free_percpu(shared_msrs);
7246 out_free_x86_fpu_cache:
7247 kmem_cache_destroy(x86_fpu_cache);
7252 void kvm_arch_exit(void)
7254 #ifdef CONFIG_X86_64
7255 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7256 clear_hv_tscchange_cb();
7259 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7261 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7262 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7263 CPUFREQ_TRANSITION_NOTIFIER);
7264 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7265 #ifdef CONFIG_X86_64
7266 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7269 kvm_mmu_module_exit();
7270 free_percpu(shared_msrs);
7271 kmem_cache_destroy(x86_fpu_cache);
7274 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7276 ++vcpu->stat.halt_exits;
7277 if (lapic_in_kernel(vcpu)) {
7278 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7281 vcpu->run->exit_reason = KVM_EXIT_HLT;
7285 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7287 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7289 int ret = kvm_skip_emulated_instruction(vcpu);
7291 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7292 * KVM_EXIT_DEBUG here.
7294 return kvm_vcpu_halt(vcpu) && ret;
7296 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7298 #ifdef CONFIG_X86_64
7299 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7300 unsigned long clock_type)
7302 struct kvm_clock_pairing clock_pairing;
7303 struct timespec64 ts;
7307 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7308 return -KVM_EOPNOTSUPP;
7310 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7311 return -KVM_EOPNOTSUPP;
7313 clock_pairing.sec = ts.tv_sec;
7314 clock_pairing.nsec = ts.tv_nsec;
7315 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7316 clock_pairing.flags = 0;
7317 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7320 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7321 sizeof(struct kvm_clock_pairing)))
7329 * kvm_pv_kick_cpu_op: Kick a vcpu.
7331 * @apicid - apicid of vcpu to be kicked.
7333 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7335 struct kvm_lapic_irq lapic_irq;
7337 lapic_irq.shorthand = 0;
7338 lapic_irq.dest_mode = 0;
7339 lapic_irq.level = 0;
7340 lapic_irq.dest_id = apicid;
7341 lapic_irq.msi_redir_hint = false;
7343 lapic_irq.delivery_mode = APIC_DM_REMRD;
7344 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7347 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7349 if (!lapic_in_kernel(vcpu)) {
7350 WARN_ON_ONCE(vcpu->arch.apicv_active);
7353 if (!vcpu->arch.apicv_active)
7356 vcpu->arch.apicv_active = false;
7357 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7360 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7362 struct kvm_vcpu *target = NULL;
7363 struct kvm_apic_map *map;
7366 map = rcu_dereference(kvm->arch.apic_map);
7368 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7369 target = map->phys_map[dest_id]->vcpu;
7373 if (target && READ_ONCE(target->ready))
7374 kvm_vcpu_yield_to(target);
7377 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7379 unsigned long nr, a0, a1, a2, a3, ret;
7382 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7383 return kvm_hv_hypercall(vcpu);
7385 nr = kvm_rax_read(vcpu);
7386 a0 = kvm_rbx_read(vcpu);
7387 a1 = kvm_rcx_read(vcpu);
7388 a2 = kvm_rdx_read(vcpu);
7389 a3 = kvm_rsi_read(vcpu);
7391 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7393 op_64_bit = is_64_bit_mode(vcpu);
7402 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7408 case KVM_HC_VAPIC_POLL_IRQ:
7411 case KVM_HC_KICK_CPU:
7412 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7413 kvm_sched_yield(vcpu->kvm, a1);
7416 #ifdef CONFIG_X86_64
7417 case KVM_HC_CLOCK_PAIRING:
7418 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7421 case KVM_HC_SEND_IPI:
7422 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7424 case KVM_HC_SCHED_YIELD:
7425 kvm_sched_yield(vcpu->kvm, a0);
7435 kvm_rax_write(vcpu, ret);
7437 ++vcpu->stat.hypercalls;
7438 return kvm_skip_emulated_instruction(vcpu);
7440 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7442 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7444 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7445 char instruction[3];
7446 unsigned long rip = kvm_rip_read(vcpu);
7448 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7450 return emulator_write_emulated(ctxt, rip, instruction, 3,
7454 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7456 return vcpu->run->request_interrupt_window &&
7457 likely(!pic_in_kernel(vcpu->kvm));
7460 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7462 struct kvm_run *kvm_run = vcpu->run;
7464 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7465 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7466 kvm_run->cr8 = kvm_get_cr8(vcpu);
7467 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7468 kvm_run->ready_for_interrupt_injection =
7469 pic_in_kernel(vcpu->kvm) ||
7470 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7473 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7477 if (!kvm_x86_ops->update_cr8_intercept)
7480 if (!lapic_in_kernel(vcpu))
7483 if (vcpu->arch.apicv_active)
7486 if (!vcpu->arch.apic->vapic_addr)
7487 max_irr = kvm_lapic_find_highest_irr(vcpu);
7494 tpr = kvm_lapic_get_cr8(vcpu);
7496 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7499 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7503 /* try to reinject previous events if any */
7505 if (vcpu->arch.exception.injected)
7506 kvm_x86_ops->queue_exception(vcpu);
7508 * Do not inject an NMI or interrupt if there is a pending
7509 * exception. Exceptions and interrupts are recognized at
7510 * instruction boundaries, i.e. the start of an instruction.
7511 * Trap-like exceptions, e.g. #DB, have higher priority than
7512 * NMIs and interrupts, i.e. traps are recognized before an
7513 * NMI/interrupt that's pending on the same instruction.
7514 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7515 * priority, but are only generated (pended) during instruction
7516 * execution, i.e. a pending fault-like exception means the
7517 * fault occurred on the *previous* instruction and must be
7518 * serviced prior to recognizing any new events in order to
7519 * fully complete the previous instruction.
7521 else if (!vcpu->arch.exception.pending) {
7522 if (vcpu->arch.nmi_injected)
7523 kvm_x86_ops->set_nmi(vcpu);
7524 else if (vcpu->arch.interrupt.injected)
7525 kvm_x86_ops->set_irq(vcpu);
7529 * Call check_nested_events() even if we reinjected a previous event
7530 * in order for caller to determine if it should require immediate-exit
7531 * from L2 to L1 due to pending L1 events which require exit
7534 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7535 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7540 /* try to inject new event if pending */
7541 if (vcpu->arch.exception.pending) {
7542 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7543 vcpu->arch.exception.has_error_code,
7544 vcpu->arch.exception.error_code);
7546 WARN_ON_ONCE(vcpu->arch.exception.injected);
7547 vcpu->arch.exception.pending = false;
7548 vcpu->arch.exception.injected = true;
7550 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7551 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7554 if (vcpu->arch.exception.nr == DB_VECTOR) {
7556 * This code assumes that nSVM doesn't use
7557 * check_nested_events(). If it does, the
7558 * DR6/DR7 changes should happen before L1
7559 * gets a #VMEXIT for an intercepted #DB in
7560 * L2. (Under VMX, on the other hand, the
7561 * DR6/DR7 changes should not happen in the
7562 * event of a VM-exit to L1 for an intercepted
7565 kvm_deliver_exception_payload(vcpu);
7566 if (vcpu->arch.dr7 & DR7_GD) {
7567 vcpu->arch.dr7 &= ~DR7_GD;
7568 kvm_update_dr7(vcpu);
7572 kvm_x86_ops->queue_exception(vcpu);
7575 /* Don't consider new event if we re-injected an event */
7576 if (kvm_event_needs_reinjection(vcpu))
7579 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7580 kvm_x86_ops->smi_allowed(vcpu)) {
7581 vcpu->arch.smi_pending = false;
7582 ++vcpu->arch.smi_count;
7584 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7585 --vcpu->arch.nmi_pending;
7586 vcpu->arch.nmi_injected = true;
7587 kvm_x86_ops->set_nmi(vcpu);
7588 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7590 * Because interrupts can be injected asynchronously, we are
7591 * calling check_nested_events again here to avoid a race condition.
7592 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7593 * proposal and current concerns. Perhaps we should be setting
7594 * KVM_REQ_EVENT only on certain events and not unconditionally?
7596 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7597 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7601 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7602 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7604 kvm_x86_ops->set_irq(vcpu);
7611 static void process_nmi(struct kvm_vcpu *vcpu)
7616 * x86 is limited to one NMI running, and one NMI pending after it.
7617 * If an NMI is already in progress, limit further NMIs to just one.
7618 * Otherwise, allow two (and we'll inject the first one immediately).
7620 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7623 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7624 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7625 kvm_make_request(KVM_REQ_EVENT, vcpu);
7628 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7631 flags |= seg->g << 23;
7632 flags |= seg->db << 22;
7633 flags |= seg->l << 21;
7634 flags |= seg->avl << 20;
7635 flags |= seg->present << 15;
7636 flags |= seg->dpl << 13;
7637 flags |= seg->s << 12;
7638 flags |= seg->type << 8;
7642 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7644 struct kvm_segment seg;
7647 kvm_get_segment(vcpu, &seg, n);
7648 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7651 offset = 0x7f84 + n * 12;
7653 offset = 0x7f2c + (n - 3) * 12;
7655 put_smstate(u32, buf, offset + 8, seg.base);
7656 put_smstate(u32, buf, offset + 4, seg.limit);
7657 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7660 #ifdef CONFIG_X86_64
7661 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7663 struct kvm_segment seg;
7667 kvm_get_segment(vcpu, &seg, n);
7668 offset = 0x7e00 + n * 16;
7670 flags = enter_smm_get_segment_flags(&seg) >> 8;
7671 put_smstate(u16, buf, offset, seg.selector);
7672 put_smstate(u16, buf, offset + 2, flags);
7673 put_smstate(u32, buf, offset + 4, seg.limit);
7674 put_smstate(u64, buf, offset + 8, seg.base);
7678 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7681 struct kvm_segment seg;
7685 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7686 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7687 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7688 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7690 for (i = 0; i < 8; i++)
7691 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7693 kvm_get_dr(vcpu, 6, &val);
7694 put_smstate(u32, buf, 0x7fcc, (u32)val);
7695 kvm_get_dr(vcpu, 7, &val);
7696 put_smstate(u32, buf, 0x7fc8, (u32)val);
7698 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7699 put_smstate(u32, buf, 0x7fc4, seg.selector);
7700 put_smstate(u32, buf, 0x7f64, seg.base);
7701 put_smstate(u32, buf, 0x7f60, seg.limit);
7702 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7704 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7705 put_smstate(u32, buf, 0x7fc0, seg.selector);
7706 put_smstate(u32, buf, 0x7f80, seg.base);
7707 put_smstate(u32, buf, 0x7f7c, seg.limit);
7708 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7710 kvm_x86_ops->get_gdt(vcpu, &dt);
7711 put_smstate(u32, buf, 0x7f74, dt.address);
7712 put_smstate(u32, buf, 0x7f70, dt.size);
7714 kvm_x86_ops->get_idt(vcpu, &dt);
7715 put_smstate(u32, buf, 0x7f58, dt.address);
7716 put_smstate(u32, buf, 0x7f54, dt.size);
7718 for (i = 0; i < 6; i++)
7719 enter_smm_save_seg_32(vcpu, buf, i);
7721 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7724 put_smstate(u32, buf, 0x7efc, 0x00020000);
7725 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7728 #ifdef CONFIG_X86_64
7729 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7732 struct kvm_segment seg;
7736 for (i = 0; i < 16; i++)
7737 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7739 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7740 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7742 kvm_get_dr(vcpu, 6, &val);
7743 put_smstate(u64, buf, 0x7f68, val);
7744 kvm_get_dr(vcpu, 7, &val);
7745 put_smstate(u64, buf, 0x7f60, val);
7747 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7748 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7749 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7751 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7754 put_smstate(u32, buf, 0x7efc, 0x00020064);
7756 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7758 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7759 put_smstate(u16, buf, 0x7e90, seg.selector);
7760 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7761 put_smstate(u32, buf, 0x7e94, seg.limit);
7762 put_smstate(u64, buf, 0x7e98, seg.base);
7764 kvm_x86_ops->get_idt(vcpu, &dt);
7765 put_smstate(u32, buf, 0x7e84, dt.size);
7766 put_smstate(u64, buf, 0x7e88, dt.address);
7768 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7769 put_smstate(u16, buf, 0x7e70, seg.selector);
7770 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7771 put_smstate(u32, buf, 0x7e74, seg.limit);
7772 put_smstate(u64, buf, 0x7e78, seg.base);
7774 kvm_x86_ops->get_gdt(vcpu, &dt);
7775 put_smstate(u32, buf, 0x7e64, dt.size);
7776 put_smstate(u64, buf, 0x7e68, dt.address);
7778 for (i = 0; i < 6; i++)
7779 enter_smm_save_seg_64(vcpu, buf, i);
7783 static void enter_smm(struct kvm_vcpu *vcpu)
7785 struct kvm_segment cs, ds;
7790 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7791 memset(buf, 0, 512);
7792 #ifdef CONFIG_X86_64
7793 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7794 enter_smm_save_state_64(vcpu, buf);
7797 enter_smm_save_state_32(vcpu, buf);
7800 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7801 * vCPU state (e.g. leave guest mode) after we've saved the state into
7802 * the SMM state-save area.
7804 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7806 vcpu->arch.hflags |= HF_SMM_MASK;
7807 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7809 if (kvm_x86_ops->get_nmi_mask(vcpu))
7810 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7812 kvm_x86_ops->set_nmi_mask(vcpu, true);
7814 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7815 kvm_rip_write(vcpu, 0x8000);
7817 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7818 kvm_x86_ops->set_cr0(vcpu, cr0);
7819 vcpu->arch.cr0 = cr0;
7821 kvm_x86_ops->set_cr4(vcpu, 0);
7823 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7824 dt.address = dt.size = 0;
7825 kvm_x86_ops->set_idt(vcpu, &dt);
7827 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7829 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7830 cs.base = vcpu->arch.smbase;
7835 cs.limit = ds.limit = 0xffffffff;
7836 cs.type = ds.type = 0x3;
7837 cs.dpl = ds.dpl = 0;
7842 cs.avl = ds.avl = 0;
7843 cs.present = ds.present = 1;
7844 cs.unusable = ds.unusable = 0;
7845 cs.padding = ds.padding = 0;
7847 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7848 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7849 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7850 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7851 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7852 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7854 #ifdef CONFIG_X86_64
7855 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7856 kvm_x86_ops->set_efer(vcpu, 0);
7859 kvm_update_cpuid(vcpu);
7860 kvm_mmu_reset_context(vcpu);
7863 static void process_smi(struct kvm_vcpu *vcpu)
7865 vcpu->arch.smi_pending = true;
7866 kvm_make_request(KVM_REQ_EVENT, vcpu);
7869 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7871 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7874 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7876 if (!kvm_apic_present(vcpu))
7879 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7881 if (irqchip_split(vcpu->kvm))
7882 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7884 if (vcpu->arch.apicv_active)
7885 kvm_x86_ops->sync_pir_to_irr(vcpu);
7886 if (ioapic_in_kernel(vcpu->kvm))
7887 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7890 if (is_guest_mode(vcpu))
7891 vcpu->arch.load_eoi_exitmap_pending = true;
7893 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7896 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7898 u64 eoi_exit_bitmap[4];
7900 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7903 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7904 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7905 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7908 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7909 unsigned long start, unsigned long end,
7912 unsigned long apic_address;
7915 * The physical address of apic access page is stored in the VMCS.
7916 * Update it when it becomes invalid.
7918 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7919 if (start <= apic_address && apic_address < end)
7920 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7925 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7927 struct page *page = NULL;
7929 if (!lapic_in_kernel(vcpu))
7932 if (!kvm_x86_ops->set_apic_access_page_addr)
7935 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7936 if (is_error_page(page))
7938 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7941 * Do not pin apic access page in memory, the MMU notifier
7942 * will call us again if it is migrated or swapped out.
7946 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7948 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7950 smp_send_reschedule(vcpu->cpu);
7952 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7955 * Returns 1 to let vcpu_run() continue the guest execution loop without
7956 * exiting to the userspace. Otherwise, the value will be returned to the
7959 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7963 dm_request_for_irq_injection(vcpu) &&
7964 kvm_cpu_accept_dm_intr(vcpu);
7966 bool req_immediate_exit = false;
7968 if (kvm_request_pending(vcpu)) {
7969 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
7970 if (unlikely(!kvm_x86_ops->get_vmcs12_pages(vcpu))) {
7975 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7976 kvm_mmu_unload(vcpu);
7977 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7978 __kvm_migrate_timers(vcpu);
7979 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7980 kvm_gen_update_masterclock(vcpu->kvm);
7981 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7982 kvm_gen_kvmclock_update(vcpu);
7983 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7984 r = kvm_guest_time_update(vcpu);
7988 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7989 kvm_mmu_sync_roots(vcpu);
7990 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7991 kvm_mmu_load_cr3(vcpu);
7992 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7993 kvm_vcpu_flush_tlb(vcpu, true);
7994 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7995 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7999 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
8000 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
8001 vcpu->mmio_needed = 0;
8005 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
8006 /* Page is swapped out. Do synthetic halt */
8007 vcpu->arch.apf.halted = true;
8011 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8012 record_steal_time(vcpu);
8013 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8015 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8017 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8018 kvm_pmu_handle_event(vcpu);
8019 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8020 kvm_pmu_deliver_pmi(vcpu);
8021 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8022 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8023 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8024 vcpu->arch.ioapic_handled_vectors)) {
8025 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8026 vcpu->run->eoi.vector =
8027 vcpu->arch.pending_ioapic_eoi;
8032 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8033 vcpu_scan_ioapic(vcpu);
8034 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8035 vcpu_load_eoi_exitmap(vcpu);
8036 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8037 kvm_vcpu_reload_apic_access_page(vcpu);
8038 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8039 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8040 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8044 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8045 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8046 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8050 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8051 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8052 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8058 * KVM_REQ_HV_STIMER has to be processed after
8059 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8060 * depend on the guest clock being up-to-date
8062 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8063 kvm_hv_process_stimers(vcpu);
8066 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8067 ++vcpu->stat.req_event;
8068 kvm_apic_accept_events(vcpu);
8069 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8074 if (inject_pending_event(vcpu, req_int_win) != 0)
8075 req_immediate_exit = true;
8077 /* Enable SMI/NMI/IRQ window open exits if needed.
8079 * SMIs have three cases:
8080 * 1) They can be nested, and then there is nothing to
8081 * do here because RSM will cause a vmexit anyway.
8082 * 2) There is an ISA-specific reason why SMI cannot be
8083 * injected, and the moment when this changes can be
8085 * 3) Or the SMI can be pending because
8086 * inject_pending_event has completed the injection
8087 * of an IRQ or NMI from the previous vmexit, and
8088 * then we request an immediate exit to inject the
8091 if (vcpu->arch.smi_pending && !is_smm(vcpu))
8092 if (!kvm_x86_ops->enable_smi_window(vcpu))
8093 req_immediate_exit = true;
8094 if (vcpu->arch.nmi_pending)
8095 kvm_x86_ops->enable_nmi_window(vcpu);
8096 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8097 kvm_x86_ops->enable_irq_window(vcpu);
8098 WARN_ON(vcpu->arch.exception.pending);
8101 if (kvm_lapic_enabled(vcpu)) {
8102 update_cr8_intercept(vcpu);
8103 kvm_lapic_sync_to_vapic(vcpu);
8107 r = kvm_mmu_reload(vcpu);
8109 goto cancel_injection;
8114 kvm_x86_ops->prepare_guest_switch(vcpu);
8117 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8118 * IPI are then delayed after guest entry, which ensures that they
8119 * result in virtual interrupt delivery.
8121 local_irq_disable();
8122 vcpu->mode = IN_GUEST_MODE;
8124 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8127 * 1) We should set ->mode before checking ->requests. Please see
8128 * the comment in kvm_vcpu_exiting_guest_mode().
8130 * 2) For APICv, we should set ->mode before checking PID.ON. This
8131 * pairs with the memory barrier implicit in pi_test_and_set_on
8132 * (see vmx_deliver_posted_interrupt).
8134 * 3) This also orders the write to mode from any reads to the page
8135 * tables done while the VCPU is running. Please see the comment
8136 * in kvm_flush_remote_tlbs.
8138 smp_mb__after_srcu_read_unlock();
8141 * This handles the case where a posted interrupt was
8142 * notified with kvm_vcpu_kick.
8144 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8145 kvm_x86_ops->sync_pir_to_irr(vcpu);
8147 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8148 || need_resched() || signal_pending(current)) {
8149 vcpu->mode = OUTSIDE_GUEST_MODE;
8153 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8155 goto cancel_injection;
8158 if (req_immediate_exit) {
8159 kvm_make_request(KVM_REQ_EVENT, vcpu);
8160 kvm_x86_ops->request_immediate_exit(vcpu);
8163 trace_kvm_entry(vcpu->vcpu_id);
8164 guest_enter_irqoff();
8166 /* The preempt notifier should have taken care of the FPU already. */
8167 WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8169 if (unlikely(vcpu->arch.switch_db_regs)) {
8171 set_debugreg(vcpu->arch.eff_db[0], 0);
8172 set_debugreg(vcpu->arch.eff_db[1], 1);
8173 set_debugreg(vcpu->arch.eff_db[2], 2);
8174 set_debugreg(vcpu->arch.eff_db[3], 3);
8175 set_debugreg(vcpu->arch.dr6, 6);
8176 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8179 kvm_x86_ops->run(vcpu);
8182 * Do this here before restoring debug registers on the host. And
8183 * since we do this before handling the vmexit, a DR access vmexit
8184 * can (a) read the correct value of the debug registers, (b) set
8185 * KVM_DEBUGREG_WONT_EXIT again.
8187 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8188 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8189 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8190 kvm_update_dr0123(vcpu);
8191 kvm_update_dr6(vcpu);
8192 kvm_update_dr7(vcpu);
8193 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8197 * If the guest has used debug registers, at least dr7
8198 * will be disabled while returning to the host.
8199 * If we don't have active breakpoints in the host, we don't
8200 * care about the messed up debug address registers. But if
8201 * we have some of them active, restore the old state.
8203 if (hw_breakpoint_active())
8204 hw_breakpoint_restore();
8206 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8208 vcpu->mode = OUTSIDE_GUEST_MODE;
8211 kvm_x86_ops->handle_exit_irqoff(vcpu);
8214 * Consume any pending interrupts, including the possible source of
8215 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8216 * An instruction is required after local_irq_enable() to fully unblock
8217 * interrupts on processors that implement an interrupt shadow, the
8218 * stat.exits increment will do nicely.
8220 kvm_before_interrupt(vcpu);
8223 local_irq_disable();
8224 kvm_after_interrupt(vcpu);
8226 guest_exit_irqoff();
8227 if (lapic_in_kernel(vcpu)) {
8228 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8229 if (delta != S64_MIN) {
8230 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8231 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8238 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8241 * Profile KVM exit RIPs:
8243 if (unlikely(prof_on == KVM_PROFILING)) {
8244 unsigned long rip = kvm_rip_read(vcpu);
8245 profile_hit(KVM_PROFILING, (void *)rip);
8248 if (unlikely(vcpu->arch.tsc_always_catchup))
8249 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8251 if (vcpu->arch.apic_attention)
8252 kvm_lapic_sync_from_vapic(vcpu);
8254 vcpu->arch.gpa_available = false;
8255 r = kvm_x86_ops->handle_exit(vcpu);
8259 kvm_x86_ops->cancel_injection(vcpu);
8260 if (unlikely(vcpu->arch.apic_attention))
8261 kvm_lapic_sync_from_vapic(vcpu);
8266 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8268 if (!kvm_arch_vcpu_runnable(vcpu) &&
8269 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8270 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8271 kvm_vcpu_block(vcpu);
8272 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8274 if (kvm_x86_ops->post_block)
8275 kvm_x86_ops->post_block(vcpu);
8277 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8281 kvm_apic_accept_events(vcpu);
8282 switch(vcpu->arch.mp_state) {
8283 case KVM_MP_STATE_HALTED:
8284 vcpu->arch.pv.pv_unhalted = false;
8285 vcpu->arch.mp_state =
8286 KVM_MP_STATE_RUNNABLE;
8288 case KVM_MP_STATE_RUNNABLE:
8289 vcpu->arch.apf.halted = false;
8291 case KVM_MP_STATE_INIT_RECEIVED:
8300 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8302 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8303 kvm_x86_ops->check_nested_events(vcpu, false);
8305 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8306 !vcpu->arch.apf.halted);
8309 static int vcpu_run(struct kvm_vcpu *vcpu)
8312 struct kvm *kvm = vcpu->kvm;
8314 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8315 vcpu->arch.l1tf_flush_l1d = true;
8318 if (kvm_vcpu_running(vcpu)) {
8319 r = vcpu_enter_guest(vcpu);
8321 r = vcpu_block(kvm, vcpu);
8327 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8328 if (kvm_cpu_has_pending_timer(vcpu))
8329 kvm_inject_pending_timer_irqs(vcpu);
8331 if (dm_request_for_irq_injection(vcpu) &&
8332 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8334 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8335 ++vcpu->stat.request_irq_exits;
8339 kvm_check_async_pf_completion(vcpu);
8341 if (signal_pending(current)) {
8343 vcpu->run->exit_reason = KVM_EXIT_INTR;
8344 ++vcpu->stat.signal_exits;
8347 if (need_resched()) {
8348 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8350 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8354 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8359 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8363 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8364 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8365 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8369 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8371 BUG_ON(!vcpu->arch.pio.count);
8373 return complete_emulated_io(vcpu);
8377 * Implements the following, as a state machine:
8381 * for each mmio piece in the fragment
8389 * for each mmio piece in the fragment
8394 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8396 struct kvm_run *run = vcpu->run;
8397 struct kvm_mmio_fragment *frag;
8400 BUG_ON(!vcpu->mmio_needed);
8402 /* Complete previous fragment */
8403 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8404 len = min(8u, frag->len);
8405 if (!vcpu->mmio_is_write)
8406 memcpy(frag->data, run->mmio.data, len);
8408 if (frag->len <= 8) {
8409 /* Switch to the next fragment. */
8411 vcpu->mmio_cur_fragment++;
8413 /* Go forward to the next mmio piece. */
8419 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8420 vcpu->mmio_needed = 0;
8422 /* FIXME: return into emulator if single-stepping. */
8423 if (vcpu->mmio_is_write)
8425 vcpu->mmio_read_completed = 1;
8426 return complete_emulated_io(vcpu);
8429 run->exit_reason = KVM_EXIT_MMIO;
8430 run->mmio.phys_addr = frag->gpa;
8431 if (vcpu->mmio_is_write)
8432 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8433 run->mmio.len = min(8u, frag->len);
8434 run->mmio.is_write = vcpu->mmio_is_write;
8435 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8439 /* Swap (qemu) user FPU context for the guest FPU context. */
8440 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8444 copy_fpregs_to_fpstate(vcpu->arch.user_fpu);
8445 /* PKRU is separately restored in kvm_x86_ops->run. */
8446 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8447 ~XFEATURE_MASK_PKRU);
8449 fpregs_mark_activate();
8455 /* When vcpu_run ends, restore user space FPU context. */
8456 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8460 copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8461 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8463 fpregs_mark_activate();
8466 ++vcpu->stat.fpu_reload;
8470 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8475 kvm_sigset_activate(vcpu);
8476 kvm_load_guest_fpu(vcpu);
8478 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8479 if (kvm_run->immediate_exit) {
8483 kvm_vcpu_block(vcpu);
8484 kvm_apic_accept_events(vcpu);
8485 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8487 if (signal_pending(current)) {
8489 vcpu->run->exit_reason = KVM_EXIT_INTR;
8490 ++vcpu->stat.signal_exits;
8495 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8500 if (vcpu->run->kvm_dirty_regs) {
8501 r = sync_regs(vcpu);
8506 /* re-sync apic's tpr */
8507 if (!lapic_in_kernel(vcpu)) {
8508 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8514 if (unlikely(vcpu->arch.complete_userspace_io)) {
8515 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8516 vcpu->arch.complete_userspace_io = NULL;
8521 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8523 if (kvm_run->immediate_exit)
8529 kvm_put_guest_fpu(vcpu);
8530 if (vcpu->run->kvm_valid_regs)
8532 post_kvm_run_save(vcpu);
8533 kvm_sigset_deactivate(vcpu);
8539 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8541 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8543 * We are here if userspace calls get_regs() in the middle of
8544 * instruction emulation. Registers state needs to be copied
8545 * back from emulation context to vcpu. Userspace shouldn't do
8546 * that usually, but some bad designed PV devices (vmware
8547 * backdoor interface) need this to work
8549 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8550 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8552 regs->rax = kvm_rax_read(vcpu);
8553 regs->rbx = kvm_rbx_read(vcpu);
8554 regs->rcx = kvm_rcx_read(vcpu);
8555 regs->rdx = kvm_rdx_read(vcpu);
8556 regs->rsi = kvm_rsi_read(vcpu);
8557 regs->rdi = kvm_rdi_read(vcpu);
8558 regs->rsp = kvm_rsp_read(vcpu);
8559 regs->rbp = kvm_rbp_read(vcpu);
8560 #ifdef CONFIG_X86_64
8561 regs->r8 = kvm_r8_read(vcpu);
8562 regs->r9 = kvm_r9_read(vcpu);
8563 regs->r10 = kvm_r10_read(vcpu);
8564 regs->r11 = kvm_r11_read(vcpu);
8565 regs->r12 = kvm_r12_read(vcpu);
8566 regs->r13 = kvm_r13_read(vcpu);
8567 regs->r14 = kvm_r14_read(vcpu);
8568 regs->r15 = kvm_r15_read(vcpu);
8571 regs->rip = kvm_rip_read(vcpu);
8572 regs->rflags = kvm_get_rflags(vcpu);
8575 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8578 __get_regs(vcpu, regs);
8583 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8585 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8586 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8588 kvm_rax_write(vcpu, regs->rax);
8589 kvm_rbx_write(vcpu, regs->rbx);
8590 kvm_rcx_write(vcpu, regs->rcx);
8591 kvm_rdx_write(vcpu, regs->rdx);
8592 kvm_rsi_write(vcpu, regs->rsi);
8593 kvm_rdi_write(vcpu, regs->rdi);
8594 kvm_rsp_write(vcpu, regs->rsp);
8595 kvm_rbp_write(vcpu, regs->rbp);
8596 #ifdef CONFIG_X86_64
8597 kvm_r8_write(vcpu, regs->r8);
8598 kvm_r9_write(vcpu, regs->r9);
8599 kvm_r10_write(vcpu, regs->r10);
8600 kvm_r11_write(vcpu, regs->r11);
8601 kvm_r12_write(vcpu, regs->r12);
8602 kvm_r13_write(vcpu, regs->r13);
8603 kvm_r14_write(vcpu, regs->r14);
8604 kvm_r15_write(vcpu, regs->r15);
8607 kvm_rip_write(vcpu, regs->rip);
8608 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8610 vcpu->arch.exception.pending = false;
8612 kvm_make_request(KVM_REQ_EVENT, vcpu);
8615 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8618 __set_regs(vcpu, regs);
8623 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8625 struct kvm_segment cs;
8627 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8631 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8633 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8637 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8638 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8639 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8640 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8641 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8642 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8644 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8645 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8647 kvm_x86_ops->get_idt(vcpu, &dt);
8648 sregs->idt.limit = dt.size;
8649 sregs->idt.base = dt.address;
8650 kvm_x86_ops->get_gdt(vcpu, &dt);
8651 sregs->gdt.limit = dt.size;
8652 sregs->gdt.base = dt.address;
8654 sregs->cr0 = kvm_read_cr0(vcpu);
8655 sregs->cr2 = vcpu->arch.cr2;
8656 sregs->cr3 = kvm_read_cr3(vcpu);
8657 sregs->cr4 = kvm_read_cr4(vcpu);
8658 sregs->cr8 = kvm_get_cr8(vcpu);
8659 sregs->efer = vcpu->arch.efer;
8660 sregs->apic_base = kvm_get_apic_base(vcpu);
8662 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8664 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8665 set_bit(vcpu->arch.interrupt.nr,
8666 (unsigned long *)sregs->interrupt_bitmap);
8669 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8670 struct kvm_sregs *sregs)
8673 __get_sregs(vcpu, sregs);
8678 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8679 struct kvm_mp_state *mp_state)
8683 kvm_apic_accept_events(vcpu);
8684 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8685 vcpu->arch.pv.pv_unhalted)
8686 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8688 mp_state->mp_state = vcpu->arch.mp_state;
8694 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8695 struct kvm_mp_state *mp_state)
8701 if (!lapic_in_kernel(vcpu) &&
8702 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8705 /* INITs are latched while in SMM */
8706 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8707 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8708 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8711 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8712 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8713 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8715 vcpu->arch.mp_state = mp_state->mp_state;
8716 kvm_make_request(KVM_REQ_EVENT, vcpu);
8724 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8725 int reason, bool has_error_code, u32 error_code)
8727 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8730 init_emulate_ctxt(vcpu);
8732 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8733 has_error_code, error_code);
8735 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8736 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8737 vcpu->run->internal.ndata = 0;
8741 kvm_rip_write(vcpu, ctxt->eip);
8742 kvm_set_rflags(vcpu, ctxt->eflags);
8743 kvm_make_request(KVM_REQ_EVENT, vcpu);
8746 EXPORT_SYMBOL_GPL(kvm_task_switch);
8748 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8750 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8752 * When EFER.LME and CR0.PG are set, the processor is in
8753 * 64-bit mode (though maybe in a 32-bit code segment).
8754 * CR4.PAE and EFER.LMA must be set.
8756 if (!(sregs->cr4 & X86_CR4_PAE)
8757 || !(sregs->efer & EFER_LMA))
8761 * Not in 64-bit mode: EFER.LMA is clear and the code
8762 * segment cannot be 64-bit.
8764 if (sregs->efer & EFER_LMA || sregs->cs.l)
8768 return kvm_valid_cr4(vcpu, sregs->cr4);
8771 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8773 struct msr_data apic_base_msr;
8774 int mmu_reset_needed = 0;
8775 int cpuid_update_needed = 0;
8776 int pending_vec, max_bits, idx;
8780 if (kvm_valid_sregs(vcpu, sregs))
8783 apic_base_msr.data = sregs->apic_base;
8784 apic_base_msr.host_initiated = true;
8785 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8788 dt.size = sregs->idt.limit;
8789 dt.address = sregs->idt.base;
8790 kvm_x86_ops->set_idt(vcpu, &dt);
8791 dt.size = sregs->gdt.limit;
8792 dt.address = sregs->gdt.base;
8793 kvm_x86_ops->set_gdt(vcpu, &dt);
8795 vcpu->arch.cr2 = sregs->cr2;
8796 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8797 vcpu->arch.cr3 = sregs->cr3;
8798 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8800 kvm_set_cr8(vcpu, sregs->cr8);
8802 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8803 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8805 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8806 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8807 vcpu->arch.cr0 = sregs->cr0;
8809 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8810 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8811 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8812 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8813 if (cpuid_update_needed)
8814 kvm_update_cpuid(vcpu);
8816 idx = srcu_read_lock(&vcpu->kvm->srcu);
8817 if (is_pae_paging(vcpu)) {
8818 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8819 mmu_reset_needed = 1;
8821 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8823 if (mmu_reset_needed)
8824 kvm_mmu_reset_context(vcpu);
8826 max_bits = KVM_NR_INTERRUPTS;
8827 pending_vec = find_first_bit(
8828 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8829 if (pending_vec < max_bits) {
8830 kvm_queue_interrupt(vcpu, pending_vec, false);
8831 pr_debug("Set back pending irq %d\n", pending_vec);
8834 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8835 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8836 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8837 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8838 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8839 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8841 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8842 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8844 update_cr8_intercept(vcpu);
8846 /* Older userspace won't unhalt the vcpu on reset. */
8847 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8848 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8850 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8852 kvm_make_request(KVM_REQ_EVENT, vcpu);
8859 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8860 struct kvm_sregs *sregs)
8865 ret = __set_sregs(vcpu, sregs);
8870 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8871 struct kvm_guest_debug *dbg)
8873 unsigned long rflags;
8878 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8880 if (vcpu->arch.exception.pending)
8882 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8883 kvm_queue_exception(vcpu, DB_VECTOR);
8885 kvm_queue_exception(vcpu, BP_VECTOR);
8889 * Read rflags as long as potentially injected trace flags are still
8892 rflags = kvm_get_rflags(vcpu);
8894 vcpu->guest_debug = dbg->control;
8895 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8896 vcpu->guest_debug = 0;
8898 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8899 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8900 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8901 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8903 for (i = 0; i < KVM_NR_DB_REGS; i++)
8904 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8906 kvm_update_dr7(vcpu);
8908 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8909 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8910 get_segment_base(vcpu, VCPU_SREG_CS);
8913 * Trigger an rflags update that will inject or remove the trace
8916 kvm_set_rflags(vcpu, rflags);
8918 kvm_x86_ops->update_bp_intercept(vcpu);
8928 * Translate a guest virtual address to a guest physical address.
8930 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8931 struct kvm_translation *tr)
8933 unsigned long vaddr = tr->linear_address;
8939 idx = srcu_read_lock(&vcpu->kvm->srcu);
8940 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8941 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8942 tr->physical_address = gpa;
8943 tr->valid = gpa != UNMAPPED_GVA;
8951 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8953 struct fxregs_state *fxsave;
8957 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8958 memcpy(fpu->fpr, fxsave->st_space, 128);
8959 fpu->fcw = fxsave->cwd;
8960 fpu->fsw = fxsave->swd;
8961 fpu->ftwx = fxsave->twd;
8962 fpu->last_opcode = fxsave->fop;
8963 fpu->last_ip = fxsave->rip;
8964 fpu->last_dp = fxsave->rdp;
8965 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8971 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8973 struct fxregs_state *fxsave;
8977 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8979 memcpy(fxsave->st_space, fpu->fpr, 128);
8980 fxsave->cwd = fpu->fcw;
8981 fxsave->swd = fpu->fsw;
8982 fxsave->twd = fpu->ftwx;
8983 fxsave->fop = fpu->last_opcode;
8984 fxsave->rip = fpu->last_ip;
8985 fxsave->rdp = fpu->last_dp;
8986 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8992 static void store_regs(struct kvm_vcpu *vcpu)
8994 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8996 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8997 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8999 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
9000 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
9002 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
9003 kvm_vcpu_ioctl_x86_get_vcpu_events(
9004 vcpu, &vcpu->run->s.regs.events);
9007 static int sync_regs(struct kvm_vcpu *vcpu)
9009 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9012 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9013 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9014 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9016 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9017 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9019 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9021 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9022 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9023 vcpu, &vcpu->run->s.regs.events))
9025 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9031 static void fx_init(struct kvm_vcpu *vcpu)
9033 fpstate_init(&vcpu->arch.guest_fpu->state);
9034 if (boot_cpu_has(X86_FEATURE_XSAVES))
9035 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9036 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9039 * Ensure guest xcr0 is valid for loading
9041 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9043 vcpu->arch.cr0 |= X86_CR0_ET;
9046 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
9048 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
9050 kvmclock_reset(vcpu);
9052 kvm_x86_ops->vcpu_free(vcpu);
9053 free_cpumask_var(wbinvd_dirty_mask);
9056 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
9059 struct kvm_vcpu *vcpu;
9061 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9062 printk_once(KERN_WARNING
9063 "kvm: SMP vm created on host with unstable TSC; "
9064 "guest TSC will not be reliable\n");
9066 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
9071 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
9073 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9074 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9075 kvm_vcpu_mtrr_init(vcpu);
9077 kvm_vcpu_reset(vcpu, false);
9078 kvm_init_mmu(vcpu, false);
9083 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9085 struct msr_data msr;
9086 struct kvm *kvm = vcpu->kvm;
9088 kvm_hv_vcpu_postcreate(vcpu);
9090 if (mutex_lock_killable(&vcpu->mutex))
9094 msr.index = MSR_IA32_TSC;
9095 msr.host_initiated = true;
9096 kvm_write_tsc(vcpu, &msr);
9099 /* poll control enabled by default */
9100 vcpu->arch.msr_kvm_poll_control = 1;
9102 mutex_unlock(&vcpu->mutex);
9104 if (!kvmclock_periodic_sync)
9107 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9108 KVMCLOCK_SYNC_PERIOD);
9111 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9113 vcpu->arch.apf.msr_val = 0;
9116 kvm_mmu_unload(vcpu);
9119 kvm_x86_ops->vcpu_free(vcpu);
9122 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9124 kvm_lapic_reset(vcpu, init_event);
9126 vcpu->arch.hflags = 0;
9128 vcpu->arch.smi_pending = 0;
9129 vcpu->arch.smi_count = 0;
9130 atomic_set(&vcpu->arch.nmi_queued, 0);
9131 vcpu->arch.nmi_pending = 0;
9132 vcpu->arch.nmi_injected = false;
9133 kvm_clear_interrupt_queue(vcpu);
9134 kvm_clear_exception_queue(vcpu);
9135 vcpu->arch.exception.pending = false;
9137 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9138 kvm_update_dr0123(vcpu);
9139 vcpu->arch.dr6 = DR6_INIT;
9140 kvm_update_dr6(vcpu);
9141 vcpu->arch.dr7 = DR7_FIXED_1;
9142 kvm_update_dr7(vcpu);
9146 kvm_make_request(KVM_REQ_EVENT, vcpu);
9147 vcpu->arch.apf.msr_val = 0;
9148 vcpu->arch.st.msr_val = 0;
9150 kvmclock_reset(vcpu);
9152 kvm_clear_async_pf_completion_queue(vcpu);
9153 kvm_async_pf_hash_reset(vcpu);
9154 vcpu->arch.apf.halted = false;
9156 if (kvm_mpx_supported()) {
9157 void *mpx_state_buffer;
9160 * To avoid have the INIT path from kvm_apic_has_events() that be
9161 * called with loaded FPU and does not let userspace fix the state.
9164 kvm_put_guest_fpu(vcpu);
9165 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9167 if (mpx_state_buffer)
9168 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9169 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9171 if (mpx_state_buffer)
9172 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9174 kvm_load_guest_fpu(vcpu);
9178 kvm_pmu_reset(vcpu);
9179 vcpu->arch.smbase = 0x30000;
9181 vcpu->arch.msr_misc_features_enables = 0;
9183 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9186 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9187 vcpu->arch.regs_avail = ~0;
9188 vcpu->arch.regs_dirty = ~0;
9190 vcpu->arch.ia32_xss = 0;
9192 kvm_x86_ops->vcpu_reset(vcpu, init_event);
9195 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9197 struct kvm_segment cs;
9199 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9200 cs.selector = vector << 8;
9201 cs.base = vector << 12;
9202 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9203 kvm_rip_write(vcpu, 0);
9206 int kvm_arch_hardware_enable(void)
9209 struct kvm_vcpu *vcpu;
9214 bool stable, backwards_tsc = false;
9216 kvm_shared_msr_cpu_online();
9217 ret = kvm_x86_ops->hardware_enable();
9221 local_tsc = rdtsc();
9222 stable = !kvm_check_tsc_unstable();
9223 list_for_each_entry(kvm, &vm_list, vm_list) {
9224 kvm_for_each_vcpu(i, vcpu, kvm) {
9225 if (!stable && vcpu->cpu == smp_processor_id())
9226 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9227 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9228 backwards_tsc = true;
9229 if (vcpu->arch.last_host_tsc > max_tsc)
9230 max_tsc = vcpu->arch.last_host_tsc;
9236 * Sometimes, even reliable TSCs go backwards. This happens on
9237 * platforms that reset TSC during suspend or hibernate actions, but
9238 * maintain synchronization. We must compensate. Fortunately, we can
9239 * detect that condition here, which happens early in CPU bringup,
9240 * before any KVM threads can be running. Unfortunately, we can't
9241 * bring the TSCs fully up to date with real time, as we aren't yet far
9242 * enough into CPU bringup that we know how much real time has actually
9243 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9244 * variables that haven't been updated yet.
9246 * So we simply find the maximum observed TSC above, then record the
9247 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9248 * the adjustment will be applied. Note that we accumulate
9249 * adjustments, in case multiple suspend cycles happen before some VCPU
9250 * gets a chance to run again. In the event that no KVM threads get a
9251 * chance to run, we will miss the entire elapsed period, as we'll have
9252 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9253 * loose cycle time. This isn't too big a deal, since the loss will be
9254 * uniform across all VCPUs (not to mention the scenario is extremely
9255 * unlikely). It is possible that a second hibernate recovery happens
9256 * much faster than a first, causing the observed TSC here to be
9257 * smaller; this would require additional padding adjustment, which is
9258 * why we set last_host_tsc to the local tsc observed here.
9260 * N.B. - this code below runs only on platforms with reliable TSC,
9261 * as that is the only way backwards_tsc is set above. Also note
9262 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9263 * have the same delta_cyc adjustment applied if backwards_tsc
9264 * is detected. Note further, this adjustment is only done once,
9265 * as we reset last_host_tsc on all VCPUs to stop this from being
9266 * called multiple times (one for each physical CPU bringup).
9268 * Platforms with unreliable TSCs don't have to deal with this, they
9269 * will be compensated by the logic in vcpu_load, which sets the TSC to
9270 * catchup mode. This will catchup all VCPUs to real time, but cannot
9271 * guarantee that they stay in perfect synchronization.
9273 if (backwards_tsc) {
9274 u64 delta_cyc = max_tsc - local_tsc;
9275 list_for_each_entry(kvm, &vm_list, vm_list) {
9276 kvm->arch.backwards_tsc_observed = true;
9277 kvm_for_each_vcpu(i, vcpu, kvm) {
9278 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9279 vcpu->arch.last_host_tsc = local_tsc;
9280 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9284 * We have to disable TSC offset matching.. if you were
9285 * booting a VM while issuing an S4 host suspend....
9286 * you may have some problem. Solving this issue is
9287 * left as an exercise to the reader.
9289 kvm->arch.last_tsc_nsec = 0;
9290 kvm->arch.last_tsc_write = 0;
9297 void kvm_arch_hardware_disable(void)
9299 kvm_x86_ops->hardware_disable();
9300 drop_user_return_notifiers();
9303 int kvm_arch_hardware_setup(void)
9307 r = kvm_x86_ops->hardware_setup();
9311 if (kvm_has_tsc_control) {
9313 * Make sure the user can only configure tsc_khz values that
9314 * fit into a signed integer.
9315 * A min value is not calculated because it will always
9316 * be 1 on all machines.
9318 u64 max = min(0x7fffffffULL,
9319 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9320 kvm_max_guest_tsc_khz = max;
9322 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9325 kvm_init_msr_list();
9329 void kvm_arch_hardware_unsetup(void)
9331 kvm_x86_ops->hardware_unsetup();
9334 int kvm_arch_check_processor_compat(void)
9336 return kvm_x86_ops->check_processor_compatibility();
9339 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9341 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9343 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9345 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9347 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9350 struct static_key kvm_no_apic_vcpu __read_mostly;
9351 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9353 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9358 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9359 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9360 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9362 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9364 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9369 vcpu->arch.pio_data = page_address(page);
9371 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9373 r = kvm_mmu_create(vcpu);
9375 goto fail_free_pio_data;
9377 if (irqchip_in_kernel(vcpu->kvm)) {
9378 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9379 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9381 goto fail_mmu_destroy;
9383 static_key_slow_inc(&kvm_no_apic_vcpu);
9385 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9386 GFP_KERNEL_ACCOUNT);
9387 if (!vcpu->arch.mce_banks) {
9389 goto fail_free_lapic;
9391 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9393 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9394 GFP_KERNEL_ACCOUNT)) {
9396 goto fail_free_mce_banks;
9401 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9403 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9405 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9407 kvm_async_pf_hash_reset(vcpu);
9410 vcpu->arch.pending_external_vector = -1;
9411 vcpu->arch.preempted_in_kernel = false;
9413 kvm_hv_vcpu_init(vcpu);
9417 fail_free_mce_banks:
9418 kfree(vcpu->arch.mce_banks);
9420 kvm_free_lapic(vcpu);
9422 kvm_mmu_destroy(vcpu);
9424 free_page((unsigned long)vcpu->arch.pio_data);
9429 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9433 kvm_hv_vcpu_uninit(vcpu);
9434 kvm_pmu_destroy(vcpu);
9435 kfree(vcpu->arch.mce_banks);
9436 kvm_free_lapic(vcpu);
9437 idx = srcu_read_lock(&vcpu->kvm->srcu);
9438 kvm_mmu_destroy(vcpu);
9439 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9440 free_page((unsigned long)vcpu->arch.pio_data);
9441 if (!lapic_in_kernel(vcpu))
9442 static_key_slow_dec(&kvm_no_apic_vcpu);
9445 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9447 vcpu->arch.l1tf_flush_l1d = true;
9448 kvm_x86_ops->sched_in(vcpu, cpu);
9451 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9456 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9457 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9458 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9459 INIT_LIST_HEAD(&kvm->arch.lpage_disallowed_mmu_pages);
9460 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9461 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9463 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9464 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9465 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9466 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9467 &kvm->arch.irq_sources_bitmap);
9469 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9470 mutex_init(&kvm->arch.apic_map_lock);
9471 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9473 kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9474 pvclock_update_vm_gtod_copy(kvm);
9476 kvm->arch.guest_can_read_msr_platform_info = true;
9478 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9479 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9481 kvm_hv_init_vm(kvm);
9482 kvm_page_track_init(kvm);
9483 kvm_mmu_init_vm(kvm);
9485 return kvm_x86_ops->vm_init(kvm);
9488 int kvm_arch_post_init_vm(struct kvm *kvm)
9490 return kvm_mmu_post_init_vm(kvm);
9493 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9496 kvm_mmu_unload(vcpu);
9500 static void kvm_free_vcpus(struct kvm *kvm)
9503 struct kvm_vcpu *vcpu;
9506 * Unpin any mmu pages first.
9508 kvm_for_each_vcpu(i, vcpu, kvm) {
9509 kvm_clear_async_pf_completion_queue(vcpu);
9510 kvm_unload_vcpu_mmu(vcpu);
9512 kvm_for_each_vcpu(i, vcpu, kvm)
9513 kvm_arch_vcpu_free(vcpu);
9515 mutex_lock(&kvm->lock);
9516 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9517 kvm->vcpus[i] = NULL;
9519 atomic_set(&kvm->online_vcpus, 0);
9520 mutex_unlock(&kvm->lock);
9523 void kvm_arch_sync_events(struct kvm *kvm)
9525 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9526 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9530 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9534 struct kvm_memslots *slots = kvm_memslots(kvm);
9535 struct kvm_memory_slot *slot, old;
9537 /* Called with kvm->slots_lock held. */
9538 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9541 slot = id_to_memslot(slots, id);
9547 * MAP_SHARED to prevent internal slot pages from being moved
9550 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9551 MAP_SHARED | MAP_ANONYMOUS, 0);
9552 if (IS_ERR((void *)hva))
9553 return PTR_ERR((void *)hva);
9562 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9563 struct kvm_userspace_memory_region m;
9565 m.slot = id | (i << 16);
9567 m.guest_phys_addr = gpa;
9568 m.userspace_addr = hva;
9569 m.memory_size = size;
9570 r = __kvm_set_memory_region(kvm, &m);
9576 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9580 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9582 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9586 mutex_lock(&kvm->slots_lock);
9587 r = __x86_set_memory_region(kvm, id, gpa, size);
9588 mutex_unlock(&kvm->slots_lock);
9592 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9594 void kvm_arch_pre_destroy_vm(struct kvm *kvm)
9596 kvm_mmu_pre_destroy_vm(kvm);
9599 void kvm_arch_destroy_vm(struct kvm *kvm)
9601 if (current->mm == kvm->mm) {
9603 * Free memory regions allocated on behalf of userspace,
9604 * unless the the memory map has changed due to process exit
9607 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9608 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9609 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9611 if (kvm_x86_ops->vm_destroy)
9612 kvm_x86_ops->vm_destroy(kvm);
9613 kvm_pic_destroy(kvm);
9614 kvm_ioapic_destroy(kvm);
9615 kvm_free_vcpus(kvm);
9616 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9617 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9618 kvm_mmu_uninit_vm(kvm);
9619 kvm_page_track_cleanup(kvm);
9620 kvm_hv_destroy_vm(kvm);
9623 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9624 struct kvm_memory_slot *dont)
9628 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9629 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9630 kvfree(free->arch.rmap[i]);
9631 free->arch.rmap[i] = NULL;
9636 if (!dont || free->arch.lpage_info[i - 1] !=
9637 dont->arch.lpage_info[i - 1]) {
9638 kvfree(free->arch.lpage_info[i - 1]);
9639 free->arch.lpage_info[i - 1] = NULL;
9643 kvm_page_track_free_memslot(free, dont);
9646 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9647 unsigned long npages)
9651 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9652 struct kvm_lpage_info *linfo;
9657 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9658 slot->base_gfn, level) + 1;
9660 slot->arch.rmap[i] =
9661 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9662 GFP_KERNEL_ACCOUNT);
9663 if (!slot->arch.rmap[i])
9668 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9672 slot->arch.lpage_info[i - 1] = linfo;
9674 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9675 linfo[0].disallow_lpage = 1;
9676 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9677 linfo[lpages - 1].disallow_lpage = 1;
9678 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9680 * If the gfn and userspace address are not aligned wrt each
9681 * other, or if explicitly asked to, disable large page
9682 * support for this slot
9684 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9685 !kvm_largepages_enabled()) {
9688 for (j = 0; j < lpages; ++j)
9689 linfo[j].disallow_lpage = 1;
9693 if (kvm_page_track_create_memslot(slot, npages))
9699 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9700 kvfree(slot->arch.rmap[i]);
9701 slot->arch.rmap[i] = NULL;
9705 kvfree(slot->arch.lpage_info[i - 1]);
9706 slot->arch.lpage_info[i - 1] = NULL;
9711 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9714 * memslots->generation has been incremented.
9715 * mmio generation may have reached its maximum value.
9717 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9720 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9721 struct kvm_memory_slot *memslot,
9722 const struct kvm_userspace_memory_region *mem,
9723 enum kvm_mr_change change)
9728 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9729 struct kvm_memory_slot *new)
9731 /* Still write protect RO slot */
9732 if (new->flags & KVM_MEM_READONLY) {
9733 kvm_mmu_slot_remove_write_access(kvm, new);
9738 * Call kvm_x86_ops dirty logging hooks when they are valid.
9740 * kvm_x86_ops->slot_disable_log_dirty is called when:
9742 * - KVM_MR_CREATE with dirty logging is disabled
9743 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9745 * The reason is, in case of PML, we need to set D-bit for any slots
9746 * with dirty logging disabled in order to eliminate unnecessary GPA
9747 * logging in PML buffer (and potential PML buffer full VMEXT). This
9748 * guarantees leaving PML enabled during guest's lifetime won't have
9749 * any additional overhead from PML when guest is running with dirty
9750 * logging disabled for memory slots.
9752 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9753 * to dirty logging mode.
9755 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9757 * In case of write protect:
9759 * Write protect all pages for dirty logging.
9761 * All the sptes including the large sptes which point to this
9762 * slot are set to readonly. We can not create any new large
9763 * spte on this slot until the end of the logging.
9765 * See the comments in fast_page_fault().
9767 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9768 if (kvm_x86_ops->slot_enable_log_dirty)
9769 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9771 kvm_mmu_slot_remove_write_access(kvm, new);
9773 if (kvm_x86_ops->slot_disable_log_dirty)
9774 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9778 void kvm_arch_commit_memory_region(struct kvm *kvm,
9779 const struct kvm_userspace_memory_region *mem,
9780 const struct kvm_memory_slot *old,
9781 const struct kvm_memory_slot *new,
9782 enum kvm_mr_change change)
9784 if (!kvm->arch.n_requested_mmu_pages)
9785 kvm_mmu_change_mmu_pages(kvm,
9786 kvm_mmu_calculate_default_mmu_pages(kvm));
9789 * Dirty logging tracks sptes in 4k granularity, meaning that large
9790 * sptes have to be split. If live migration is successful, the guest
9791 * in the source machine will be destroyed and large sptes will be
9792 * created in the destination. However, if the guest continues to run
9793 * in the source machine (for example if live migration fails), small
9794 * sptes will remain around and cause bad performance.
9796 * Scan sptes if dirty logging has been stopped, dropping those
9797 * which can be collapsed into a single large-page spte. Later
9798 * page faults will create the large-page sptes.
9800 * There is no need to do this in any of the following cases:
9801 * CREATE: No dirty mappings will already exist.
9802 * MOVE/DELETE: The old mappings will already have been cleaned up by
9803 * kvm_arch_flush_shadow_memslot()
9805 if (change == KVM_MR_FLAGS_ONLY &&
9806 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9807 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9808 kvm_mmu_zap_collapsible_sptes(kvm, new);
9811 * Set up write protection and/or dirty logging for the new slot.
9813 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9814 * been zapped so no dirty logging staff is needed for old slot. For
9815 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9816 * new and it's also covered when dealing with the new slot.
9818 * FIXME: const-ify all uses of struct kvm_memory_slot.
9820 if (change != KVM_MR_DELETE)
9821 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9824 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9826 kvm_mmu_zap_all(kvm);
9829 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9830 struct kvm_memory_slot *slot)
9832 kvm_page_track_flush_slot(kvm, slot);
9835 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9837 return (is_guest_mode(vcpu) &&
9838 kvm_x86_ops->guest_apic_has_interrupt &&
9839 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9842 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9844 if (!list_empty_careful(&vcpu->async_pf.done))
9847 if (kvm_apic_has_events(vcpu))
9850 if (vcpu->arch.pv.pv_unhalted)
9853 if (vcpu->arch.exception.pending)
9856 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9857 (vcpu->arch.nmi_pending &&
9858 kvm_x86_ops->nmi_allowed(vcpu)))
9861 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9862 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9865 if (kvm_arch_interrupt_allowed(vcpu) &&
9866 (kvm_cpu_has_interrupt(vcpu) ||
9867 kvm_guest_apic_has_interrupt(vcpu)))
9870 if (kvm_hv_has_stimer_pending(vcpu))
9876 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9878 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9881 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
9883 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
9886 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9887 kvm_test_request(KVM_REQ_SMI, vcpu) ||
9888 kvm_test_request(KVM_REQ_EVENT, vcpu))
9891 if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
9897 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9899 return vcpu->arch.preempted_in_kernel;
9902 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9904 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9907 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9909 return kvm_x86_ops->interrupt_allowed(vcpu);
9912 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9914 if (is_64_bit_mode(vcpu))
9915 return kvm_rip_read(vcpu);
9916 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9917 kvm_rip_read(vcpu));
9919 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9921 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9923 return kvm_get_linear_rip(vcpu) == linear_rip;
9925 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9927 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9929 unsigned long rflags;
9931 rflags = kvm_x86_ops->get_rflags(vcpu);
9932 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9933 rflags &= ~X86_EFLAGS_TF;
9936 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9938 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9940 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9941 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9942 rflags |= X86_EFLAGS_TF;
9943 kvm_x86_ops->set_rflags(vcpu, rflags);
9946 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9948 __kvm_set_rflags(vcpu, rflags);
9949 kvm_make_request(KVM_REQ_EVENT, vcpu);
9951 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9953 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9957 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9961 r = kvm_mmu_reload(vcpu);
9965 if (!vcpu->arch.mmu->direct_map &&
9966 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9969 vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9972 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9974 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9977 static inline u32 kvm_async_pf_next_probe(u32 key)
9979 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9982 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9984 u32 key = kvm_async_pf_hash_fn(gfn);
9986 while (vcpu->arch.apf.gfns[key] != ~0)
9987 key = kvm_async_pf_next_probe(key);
9989 vcpu->arch.apf.gfns[key] = gfn;
9992 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9995 u32 key = kvm_async_pf_hash_fn(gfn);
9997 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9998 (vcpu->arch.apf.gfns[key] != gfn &&
9999 vcpu->arch.apf.gfns[key] != ~0); i++)
10000 key = kvm_async_pf_next_probe(key);
10005 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10007 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
10010 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
10014 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
10016 vcpu->arch.apf.gfns[i] = ~0;
10018 j = kvm_async_pf_next_probe(j);
10019 if (vcpu->arch.apf.gfns[j] == ~0)
10021 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10023 * k lies cyclically in ]i,j]
10025 * |....j i.k.| or |.k..j i...|
10027 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10028 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10033 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10036 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10040 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10043 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10047 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10049 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10052 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10053 (vcpu->arch.apf.send_user_only &&
10054 kvm_x86_ops->get_cpl(vcpu) == 0))
10060 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10062 if (unlikely(!lapic_in_kernel(vcpu) ||
10063 kvm_event_needs_reinjection(vcpu) ||
10064 vcpu->arch.exception.pending))
10067 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10071 * If interrupts are off we cannot even use an artificial
10074 return kvm_x86_ops->interrupt_allowed(vcpu);
10077 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10078 struct kvm_async_pf *work)
10080 struct x86_exception fault;
10082 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
10083 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10085 if (kvm_can_deliver_async_pf(vcpu) &&
10086 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10087 fault.vector = PF_VECTOR;
10088 fault.error_code_valid = true;
10089 fault.error_code = 0;
10090 fault.nested_page_fault = false;
10091 fault.address = work->arch.token;
10092 fault.async_page_fault = true;
10093 kvm_inject_page_fault(vcpu, &fault);
10096 * It is not possible to deliver a paravirtualized asynchronous
10097 * page fault, but putting the guest in an artificial halt state
10098 * can be beneficial nevertheless: if an interrupt arrives, we
10099 * can deliver it timely and perhaps the guest will schedule
10100 * another process. When the instruction that triggered a page
10101 * fault is retried, hopefully the page will be ready in the host.
10103 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10107 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10108 struct kvm_async_pf *work)
10110 struct x86_exception fault;
10113 if (work->wakeup_all)
10114 work->arch.token = ~0; /* broadcast wakeup */
10116 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10117 trace_kvm_async_pf_ready(work->arch.token, work->gva);
10119 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10120 !apf_get_user(vcpu, &val)) {
10121 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10122 vcpu->arch.exception.pending &&
10123 vcpu->arch.exception.nr == PF_VECTOR &&
10124 !apf_put_user(vcpu, 0)) {
10125 vcpu->arch.exception.injected = false;
10126 vcpu->arch.exception.pending = false;
10127 vcpu->arch.exception.nr = 0;
10128 vcpu->arch.exception.has_error_code = false;
10129 vcpu->arch.exception.error_code = 0;
10130 vcpu->arch.exception.has_payload = false;
10131 vcpu->arch.exception.payload = 0;
10132 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10133 fault.vector = PF_VECTOR;
10134 fault.error_code_valid = true;
10135 fault.error_code = 0;
10136 fault.nested_page_fault = false;
10137 fault.address = work->arch.token;
10138 fault.async_page_fault = true;
10139 kvm_inject_page_fault(vcpu, &fault);
10142 vcpu->arch.apf.halted = false;
10143 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10146 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10148 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10151 return kvm_can_do_async_pf(vcpu);
10154 void kvm_arch_start_assignment(struct kvm *kvm)
10156 atomic_inc(&kvm->arch.assigned_device_count);
10158 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10160 void kvm_arch_end_assignment(struct kvm *kvm)
10162 atomic_dec(&kvm->arch.assigned_device_count);
10164 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10166 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10168 return atomic_read(&kvm->arch.assigned_device_count);
10170 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10172 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10174 atomic_inc(&kvm->arch.noncoherent_dma_count);
10176 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10178 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10180 atomic_dec(&kvm->arch.noncoherent_dma_count);
10182 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10184 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10186 return atomic_read(&kvm->arch.noncoherent_dma_count);
10188 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10190 bool kvm_arch_has_irq_bypass(void)
10195 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10196 struct irq_bypass_producer *prod)
10198 struct kvm_kernel_irqfd *irqfd =
10199 container_of(cons, struct kvm_kernel_irqfd, consumer);
10201 irqfd->producer = prod;
10203 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10204 prod->irq, irqfd->gsi, 1);
10207 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10208 struct irq_bypass_producer *prod)
10211 struct kvm_kernel_irqfd *irqfd =
10212 container_of(cons, struct kvm_kernel_irqfd, consumer);
10214 WARN_ON(irqfd->producer != prod);
10215 irqfd->producer = NULL;
10218 * When producer of consumer is unregistered, we change back to
10219 * remapped mode, so we can re-use the current implementation
10220 * when the irq is masked/disabled or the consumer side (KVM
10221 * int this case doesn't want to receive the interrupts.
10223 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10225 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10226 " fails: %d\n", irqfd->consumer.token, ret);
10229 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10230 uint32_t guest_irq, bool set)
10232 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10235 bool kvm_vector_hashing_enabled(void)
10237 return vector_hashing;
10239 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10241 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10243 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10245 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10248 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10249 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10250 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10251 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10252 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10253 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10254 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10255 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10256 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10257 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10258 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10259 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10260 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10261 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10262 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10263 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10264 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10265 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10266 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10267 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);