1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * derived from drivers/kvm/kvm_main.c
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
19 #include <linux/kvm_host.h>
24 #include "kvm_cache_regs.h"
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
34 #include <linux/vmalloc.h>
35 #include <linux/export.h>
36 #include <linux/moduleparam.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <linux/kvm_irqfd.h>
52 #include <linux/irqbypass.h>
53 #include <linux/sched/stat.h>
54 #include <linux/sched/isolation.h>
55 #include <linux/mem_encrypt.h>
57 #include <trace/events/kvm.h>
59 #include <asm/debugreg.h>
63 #include <linux/kernel_stat.h>
64 #include <asm/fpu/internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
67 #include <asm/irq_remapping.h>
68 #include <asm/mshyperv.h>
69 #include <asm/hypervisor.h>
70 #include <asm/intel_pt.h>
71 #include <clocksource/hyperv_timer.h>
73 #define CREATE_TRACE_POINTS
76 #define MAX_IO_MSRS 256
77 #define KVM_MAX_MCE_BANKS 32
78 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
79 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
81 #define emul_to_vcpu(ctxt) \
82 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
85 * - enable syscall per default because its emulated by KVM
86 * - enable LME and LMA per default on 64 bit KVM
90 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
92 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
95 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
96 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
98 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
99 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
101 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
102 static void process_nmi(struct kvm_vcpu *vcpu);
103 static void enter_smm(struct kvm_vcpu *vcpu);
104 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
105 static void store_regs(struct kvm_vcpu *vcpu);
106 static int sync_regs(struct kvm_vcpu *vcpu);
108 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
109 EXPORT_SYMBOL_GPL(kvm_x86_ops);
111 static bool __read_mostly ignore_msrs = 0;
112 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
114 static bool __read_mostly report_ignored_msrs = true;
115 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
117 unsigned int min_timer_period_us = 200;
118 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
120 static bool __read_mostly kvmclock_periodic_sync = true;
121 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
123 bool __read_mostly kvm_has_tsc_control;
124 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
125 u32 __read_mostly kvm_max_guest_tsc_khz;
126 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
127 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
128 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
129 u64 __read_mostly kvm_max_tsc_scaling_ratio;
130 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
131 u64 __read_mostly kvm_default_tsc_scaling_ratio;
132 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
134 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
135 static u32 __read_mostly tsc_tolerance_ppm = 250;
136 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
139 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
140 * adaptive tuning starting from default advancment of 1000ns. '0' disables
141 * advancement entirely. Any other value is used as-is and disables adaptive
142 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
144 static int __read_mostly lapic_timer_advance_ns = -1;
145 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
147 static bool __read_mostly vector_hashing = true;
148 module_param(vector_hashing, bool, S_IRUGO);
150 bool __read_mostly enable_vmware_backdoor = false;
151 module_param(enable_vmware_backdoor, bool, S_IRUGO);
152 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
154 static bool __read_mostly force_emulation_prefix = false;
155 module_param(force_emulation_prefix, bool, S_IRUGO);
157 int __read_mostly pi_inject_timer = -1;
158 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
160 #define KVM_NR_SHARED_MSRS 16
162 struct kvm_shared_msrs_global {
164 u32 msrs[KVM_NR_SHARED_MSRS];
167 struct kvm_shared_msrs {
168 struct user_return_notifier urn;
170 struct kvm_shared_msr_values {
173 } values[KVM_NR_SHARED_MSRS];
176 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
177 static struct kvm_shared_msrs __percpu *shared_msrs;
179 struct kvm_stats_debugfs_item debugfs_entries[] = {
180 { "pf_fixed", VCPU_STAT(pf_fixed) },
181 { "pf_guest", VCPU_STAT(pf_guest) },
182 { "tlb_flush", VCPU_STAT(tlb_flush) },
183 { "invlpg", VCPU_STAT(invlpg) },
184 { "exits", VCPU_STAT(exits) },
185 { "io_exits", VCPU_STAT(io_exits) },
186 { "mmio_exits", VCPU_STAT(mmio_exits) },
187 { "signal_exits", VCPU_STAT(signal_exits) },
188 { "irq_window", VCPU_STAT(irq_window_exits) },
189 { "nmi_window", VCPU_STAT(nmi_window_exits) },
190 { "halt_exits", VCPU_STAT(halt_exits) },
191 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
192 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
193 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
194 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
195 { "hypercalls", VCPU_STAT(hypercalls) },
196 { "request_irq", VCPU_STAT(request_irq_exits) },
197 { "irq_exits", VCPU_STAT(irq_exits) },
198 { "host_state_reload", VCPU_STAT(host_state_reload) },
199 { "fpu_reload", VCPU_STAT(fpu_reload) },
200 { "insn_emulation", VCPU_STAT(insn_emulation) },
201 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
202 { "irq_injections", VCPU_STAT(irq_injections) },
203 { "nmi_injections", VCPU_STAT(nmi_injections) },
204 { "req_event", VCPU_STAT(req_event) },
205 { "l1d_flush", VCPU_STAT(l1d_flush) },
206 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
207 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
208 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
209 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
210 { "mmu_flooded", VM_STAT(mmu_flooded) },
211 { "mmu_recycled", VM_STAT(mmu_recycled) },
212 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
213 { "mmu_unsync", VM_STAT(mmu_unsync) },
214 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
215 { "largepages", VM_STAT(lpages) },
216 { "max_mmu_page_hash_collisions",
217 VM_STAT(max_mmu_page_hash_collisions) },
221 u64 __read_mostly host_xcr0;
223 struct kmem_cache *x86_fpu_cache;
224 EXPORT_SYMBOL_GPL(x86_fpu_cache);
226 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
228 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
231 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
232 vcpu->arch.apf.gfns[i] = ~0;
235 static void kvm_on_user_return(struct user_return_notifier *urn)
238 struct kvm_shared_msrs *locals
239 = container_of(urn, struct kvm_shared_msrs, urn);
240 struct kvm_shared_msr_values *values;
244 * Disabling irqs at this point since the following code could be
245 * interrupted and executed through kvm_arch_hardware_disable()
247 local_irq_save(flags);
248 if (locals->registered) {
249 locals->registered = false;
250 user_return_notifier_unregister(urn);
252 local_irq_restore(flags);
253 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
254 values = &locals->values[slot];
255 if (values->host != values->curr) {
256 wrmsrl(shared_msrs_global.msrs[slot], values->host);
257 values->curr = values->host;
262 static void shared_msr_update(unsigned slot, u32 msr)
265 unsigned int cpu = smp_processor_id();
266 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
268 /* only read, and nobody should modify it at this time,
269 * so don't need lock */
270 if (slot >= shared_msrs_global.nr) {
271 printk(KERN_ERR "kvm: invalid MSR slot!");
274 rdmsrl_safe(msr, &value);
275 smsr->values[slot].host = value;
276 smsr->values[slot].curr = value;
279 void kvm_define_shared_msr(unsigned slot, u32 msr)
281 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
282 shared_msrs_global.msrs[slot] = msr;
283 if (slot >= shared_msrs_global.nr)
284 shared_msrs_global.nr = slot + 1;
286 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
288 static void kvm_shared_msr_cpu_online(void)
292 for (i = 0; i < shared_msrs_global.nr; ++i)
293 shared_msr_update(i, shared_msrs_global.msrs[i]);
296 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
298 unsigned int cpu = smp_processor_id();
299 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
302 if (((value ^ smsr->values[slot].curr) & mask) == 0)
304 smsr->values[slot].curr = value;
305 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
309 if (!smsr->registered) {
310 smsr->urn.on_user_return = kvm_on_user_return;
311 user_return_notifier_register(&smsr->urn);
312 smsr->registered = true;
316 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
318 static void drop_user_return_notifiers(void)
320 unsigned int cpu = smp_processor_id();
321 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
323 if (smsr->registered)
324 kvm_on_user_return(&smsr->urn);
327 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
329 return vcpu->arch.apic_base;
331 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
333 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
335 return kvm_apic_mode(kvm_get_apic_base(vcpu));
337 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
339 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
341 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
342 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
343 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
344 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
346 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
348 if (!msr_info->host_initiated) {
349 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
351 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
355 kvm_lapic_set_base(vcpu, msr_info->data);
358 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
360 asmlinkage __visible void kvm_spurious_fault(void)
362 /* Fault while not rebooting. We want the trace. */
365 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
367 #define EXCPT_BENIGN 0
368 #define EXCPT_CONTRIBUTORY 1
371 static int exception_class(int vector)
381 return EXCPT_CONTRIBUTORY;
388 #define EXCPT_FAULT 0
390 #define EXCPT_ABORT 2
391 #define EXCPT_INTERRUPT 3
393 static int exception_type(int vector)
397 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
398 return EXCPT_INTERRUPT;
402 /* #DB is trap, as instruction watchpoints are handled elsewhere */
403 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
406 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
409 /* Reserved exceptions will result in fault */
413 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
415 unsigned nr = vcpu->arch.exception.nr;
416 bool has_payload = vcpu->arch.exception.has_payload;
417 unsigned long payload = vcpu->arch.exception.payload;
425 * "Certain debug exceptions may clear bit 0-3. The
426 * remaining contents of the DR6 register are never
427 * cleared by the processor".
429 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
431 * DR6.RTM is set by all #DB exceptions that don't clear it.
433 vcpu->arch.dr6 |= DR6_RTM;
434 vcpu->arch.dr6 |= payload;
436 * Bit 16 should be set in the payload whenever the #DB
437 * exception should clear DR6.RTM. This makes the payload
438 * compatible with the pending debug exceptions under VMX.
439 * Though not currently documented in the SDM, this also
440 * makes the payload compatible with the exit qualification
441 * for #DB exceptions under VMX.
443 vcpu->arch.dr6 ^= payload & DR6_RTM;
446 vcpu->arch.cr2 = payload;
450 vcpu->arch.exception.has_payload = false;
451 vcpu->arch.exception.payload = 0;
453 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
455 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
456 unsigned nr, bool has_error, u32 error_code,
457 bool has_payload, unsigned long payload, bool reinject)
462 kvm_make_request(KVM_REQ_EVENT, vcpu);
464 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
466 if (has_error && !is_protmode(vcpu))
470 * On vmentry, vcpu->arch.exception.pending is only
471 * true if an event injection was blocked by
472 * nested_run_pending. In that case, however,
473 * vcpu_enter_guest requests an immediate exit,
474 * and the guest shouldn't proceed far enough to
477 WARN_ON_ONCE(vcpu->arch.exception.pending);
478 vcpu->arch.exception.injected = true;
479 if (WARN_ON_ONCE(has_payload)) {
481 * A reinjected event has already
482 * delivered its payload.
488 vcpu->arch.exception.pending = true;
489 vcpu->arch.exception.injected = false;
491 vcpu->arch.exception.has_error_code = has_error;
492 vcpu->arch.exception.nr = nr;
493 vcpu->arch.exception.error_code = error_code;
494 vcpu->arch.exception.has_payload = has_payload;
495 vcpu->arch.exception.payload = payload;
497 * In guest mode, payload delivery should be deferred,
498 * so that the L1 hypervisor can intercept #PF before
499 * CR2 is modified (or intercept #DB before DR6 is
500 * modified under nVMX). However, for ABI
501 * compatibility with KVM_GET_VCPU_EVENTS and
502 * KVM_SET_VCPU_EVENTS, we can't delay payload
503 * delivery unless userspace has enabled this
504 * functionality via the per-VM capability,
505 * KVM_CAP_EXCEPTION_PAYLOAD.
507 if (!vcpu->kvm->arch.exception_payload_enabled ||
508 !is_guest_mode(vcpu))
509 kvm_deliver_exception_payload(vcpu);
513 /* to check exception */
514 prev_nr = vcpu->arch.exception.nr;
515 if (prev_nr == DF_VECTOR) {
516 /* triple fault -> shutdown */
517 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
520 class1 = exception_class(prev_nr);
521 class2 = exception_class(nr);
522 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
523 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
525 * Generate double fault per SDM Table 5-5. Set
526 * exception.pending = true so that the double fault
527 * can trigger a nested vmexit.
529 vcpu->arch.exception.pending = true;
530 vcpu->arch.exception.injected = false;
531 vcpu->arch.exception.has_error_code = true;
532 vcpu->arch.exception.nr = DF_VECTOR;
533 vcpu->arch.exception.error_code = 0;
534 vcpu->arch.exception.has_payload = false;
535 vcpu->arch.exception.payload = 0;
537 /* replace previous exception with a new one in a hope
538 that instruction re-execution will regenerate lost
543 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
545 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
547 EXPORT_SYMBOL_GPL(kvm_queue_exception);
549 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
551 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
553 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
555 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
556 unsigned long payload)
558 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
561 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
562 u32 error_code, unsigned long payload)
564 kvm_multiple_exception(vcpu, nr, true, error_code,
565 true, payload, false);
568 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
571 kvm_inject_gp(vcpu, 0);
573 return kvm_skip_emulated_instruction(vcpu);
577 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
579 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
581 ++vcpu->stat.pf_guest;
582 vcpu->arch.exception.nested_apf =
583 is_guest_mode(vcpu) && fault->async_page_fault;
584 if (vcpu->arch.exception.nested_apf) {
585 vcpu->arch.apf.nested_apf_token = fault->address;
586 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
588 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
592 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
594 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
596 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
597 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
599 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
601 return fault->nested_page_fault;
604 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
606 atomic_inc(&vcpu->arch.nmi_queued);
607 kvm_make_request(KVM_REQ_NMI, vcpu);
609 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
611 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
613 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
615 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
617 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
619 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
621 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
624 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
625 * a #GP and return false.
627 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
629 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
631 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
634 EXPORT_SYMBOL_GPL(kvm_require_cpl);
636 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
638 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
641 kvm_queue_exception(vcpu, UD_VECTOR);
644 EXPORT_SYMBOL_GPL(kvm_require_dr);
647 * This function will be used to read from the physical memory of the currently
648 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
649 * can read from guest physical or from the guest's guest physical memory.
651 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
652 gfn_t ngfn, void *data, int offset, int len,
655 struct x86_exception exception;
659 ngpa = gfn_to_gpa(ngfn);
660 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
661 if (real_gfn == UNMAPPED_GVA)
664 real_gfn = gpa_to_gfn(real_gfn);
666 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
668 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
670 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
671 void *data, int offset, int len, u32 access)
673 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
674 data, offset, len, access);
677 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
679 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
684 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
686 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
688 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
689 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
692 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
694 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
695 offset * sizeof(u64), sizeof(pdpte),
696 PFERR_USER_MASK|PFERR_WRITE_MASK);
701 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
702 if ((pdpte[i] & PT_PRESENT_MASK) &&
703 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
710 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
711 __set_bit(VCPU_EXREG_PDPTR,
712 (unsigned long *)&vcpu->arch.regs_avail);
713 __set_bit(VCPU_EXREG_PDPTR,
714 (unsigned long *)&vcpu->arch.regs_dirty);
719 EXPORT_SYMBOL_GPL(load_pdptrs);
721 bool pdptrs_changed(struct kvm_vcpu *vcpu)
723 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
729 if (!is_pae_paging(vcpu))
732 if (!test_bit(VCPU_EXREG_PDPTR,
733 (unsigned long *)&vcpu->arch.regs_avail))
736 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
737 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
738 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
739 PFERR_USER_MASK | PFERR_WRITE_MASK);
742 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
747 EXPORT_SYMBOL_GPL(pdptrs_changed);
749 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
751 unsigned long old_cr0 = kvm_read_cr0(vcpu);
752 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
757 if (cr0 & 0xffffffff00000000UL)
761 cr0 &= ~CR0_RESERVED_BITS;
763 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
766 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
769 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
771 if ((vcpu->arch.efer & EFER_LME)) {
776 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
781 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
786 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
789 kvm_x86_ops->set_cr0(vcpu, cr0);
791 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
792 kvm_clear_async_pf_completion_queue(vcpu);
793 kvm_async_pf_hash_reset(vcpu);
796 if ((cr0 ^ old_cr0) & update_bits)
797 kvm_mmu_reset_context(vcpu);
799 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
800 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
801 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
802 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
806 EXPORT_SYMBOL_GPL(kvm_set_cr0);
808 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
810 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
812 EXPORT_SYMBOL_GPL(kvm_lmsw);
814 void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
816 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
817 !vcpu->guest_xcr0_loaded) {
818 /* kvm_set_xcr() also depends on this */
819 if (vcpu->arch.xcr0 != host_xcr0)
820 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
821 vcpu->guest_xcr0_loaded = 1;
824 EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
826 void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
828 if (vcpu->guest_xcr0_loaded) {
829 if (vcpu->arch.xcr0 != host_xcr0)
830 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
831 vcpu->guest_xcr0_loaded = 0;
834 EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
836 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
839 u64 old_xcr0 = vcpu->arch.xcr0;
842 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
843 if (index != XCR_XFEATURE_ENABLED_MASK)
845 if (!(xcr0 & XFEATURE_MASK_FP))
847 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
851 * Do not allow the guest to set bits that we do not support
852 * saving. However, xcr0 bit 0 is always set, even if the
853 * emulated CPU does not support XSAVE (see fx_init).
855 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
856 if (xcr0 & ~valid_bits)
859 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
860 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
863 if (xcr0 & XFEATURE_MASK_AVX512) {
864 if (!(xcr0 & XFEATURE_MASK_YMM))
866 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
869 vcpu->arch.xcr0 = xcr0;
871 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
872 kvm_update_cpuid(vcpu);
876 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
878 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
879 __kvm_set_xcr(vcpu, index, xcr)) {
880 kvm_inject_gp(vcpu, 0);
885 EXPORT_SYMBOL_GPL(kvm_set_xcr);
887 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
889 unsigned long old_cr4 = kvm_read_cr4(vcpu);
890 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
891 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
893 if (cr4 & CR4_RESERVED_BITS)
896 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
899 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
902 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
905 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
908 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
911 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
914 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
917 if (is_long_mode(vcpu)) {
918 if (!(cr4 & X86_CR4_PAE))
920 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
921 && ((cr4 ^ old_cr4) & pdptr_bits)
922 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
926 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
927 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
930 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
931 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
935 if (kvm_x86_ops->set_cr4(vcpu, cr4))
938 if (((cr4 ^ old_cr4) & pdptr_bits) ||
939 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
940 kvm_mmu_reset_context(vcpu);
942 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
943 kvm_update_cpuid(vcpu);
947 EXPORT_SYMBOL_GPL(kvm_set_cr4);
949 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
951 bool skip_tlb_flush = false;
953 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
956 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
957 cr3 &= ~X86_CR3_PCID_NOFLUSH;
961 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
962 if (!skip_tlb_flush) {
963 kvm_mmu_sync_roots(vcpu);
964 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
969 if (is_long_mode(vcpu) &&
970 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
972 else if (is_pae_paging(vcpu) &&
973 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
976 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
977 vcpu->arch.cr3 = cr3;
978 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
982 EXPORT_SYMBOL_GPL(kvm_set_cr3);
984 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
986 if (cr8 & CR8_RESERVED_BITS)
988 if (lapic_in_kernel(vcpu))
989 kvm_lapic_set_tpr(vcpu, cr8);
991 vcpu->arch.cr8 = cr8;
994 EXPORT_SYMBOL_GPL(kvm_set_cr8);
996 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
998 if (lapic_in_kernel(vcpu))
999 return kvm_lapic_get_cr8(vcpu);
1001 return vcpu->arch.cr8;
1003 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1005 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1009 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1010 for (i = 0; i < KVM_NR_DB_REGS; i++)
1011 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1012 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1016 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1018 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1019 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1022 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1026 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1027 dr7 = vcpu->arch.guest_debug_dr7;
1029 dr7 = vcpu->arch.dr7;
1030 kvm_x86_ops->set_dr7(vcpu, dr7);
1031 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1032 if (dr7 & DR7_BP_EN_MASK)
1033 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1036 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1038 u64 fixed = DR6_FIXED_1;
1040 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1045 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1049 vcpu->arch.db[dr] = val;
1050 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1051 vcpu->arch.eff_db[dr] = val;
1056 if (val & 0xffffffff00000000ULL)
1057 return -1; /* #GP */
1058 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1059 kvm_update_dr6(vcpu);
1064 if (val & 0xffffffff00000000ULL)
1065 return -1; /* #GP */
1066 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1067 kvm_update_dr7(vcpu);
1074 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1076 if (__kvm_set_dr(vcpu, dr, val)) {
1077 kvm_inject_gp(vcpu, 0);
1082 EXPORT_SYMBOL_GPL(kvm_set_dr);
1084 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1088 *val = vcpu->arch.db[dr];
1093 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1094 *val = vcpu->arch.dr6;
1096 *val = kvm_x86_ops->get_dr6(vcpu);
1101 *val = vcpu->arch.dr7;
1106 EXPORT_SYMBOL_GPL(kvm_get_dr);
1108 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1110 u32 ecx = kvm_rcx_read(vcpu);
1114 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1117 kvm_rax_write(vcpu, (u32)data);
1118 kvm_rdx_write(vcpu, data >> 32);
1121 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1124 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1125 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1127 * This list is modified at module load time to reflect the
1128 * capabilities of the host cpu. This capabilities test skips MSRs that are
1129 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1130 * may depend on host virtualization features rather than host cpu features.
1133 static u32 msrs_to_save[] = {
1134 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1136 #ifdef CONFIG_X86_64
1137 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1139 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1140 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1142 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1143 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1144 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1145 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1146 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1147 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1148 MSR_IA32_UMWAIT_CONTROL,
1150 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1151 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1152 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1153 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1154 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1155 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1156 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1157 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1158 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1159 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1160 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1161 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1162 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1163 MSR_ARCH_PERFMON_PERFCTR0 + 18, MSR_ARCH_PERFMON_PERFCTR0 + 19,
1164 MSR_ARCH_PERFMON_PERFCTR0 + 20, MSR_ARCH_PERFMON_PERFCTR0 + 21,
1165 MSR_ARCH_PERFMON_PERFCTR0 + 22, MSR_ARCH_PERFMON_PERFCTR0 + 23,
1166 MSR_ARCH_PERFMON_PERFCTR0 + 24, MSR_ARCH_PERFMON_PERFCTR0 + 25,
1167 MSR_ARCH_PERFMON_PERFCTR0 + 26, MSR_ARCH_PERFMON_PERFCTR0 + 27,
1168 MSR_ARCH_PERFMON_PERFCTR0 + 28, MSR_ARCH_PERFMON_PERFCTR0 + 29,
1169 MSR_ARCH_PERFMON_PERFCTR0 + 30, MSR_ARCH_PERFMON_PERFCTR0 + 31,
1170 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1171 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1172 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1173 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1174 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1175 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1176 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1177 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1178 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1179 MSR_ARCH_PERFMON_EVENTSEL0 + 18, MSR_ARCH_PERFMON_EVENTSEL0 + 19,
1180 MSR_ARCH_PERFMON_EVENTSEL0 + 20, MSR_ARCH_PERFMON_EVENTSEL0 + 21,
1181 MSR_ARCH_PERFMON_EVENTSEL0 + 22, MSR_ARCH_PERFMON_EVENTSEL0 + 23,
1182 MSR_ARCH_PERFMON_EVENTSEL0 + 24, MSR_ARCH_PERFMON_EVENTSEL0 + 25,
1183 MSR_ARCH_PERFMON_EVENTSEL0 + 26, MSR_ARCH_PERFMON_EVENTSEL0 + 27,
1184 MSR_ARCH_PERFMON_EVENTSEL0 + 28, MSR_ARCH_PERFMON_EVENTSEL0 + 29,
1185 MSR_ARCH_PERFMON_EVENTSEL0 + 30, MSR_ARCH_PERFMON_EVENTSEL0 + 31,
1188 static unsigned num_msrs_to_save;
1190 static u32 emulated_msrs[] = {
1191 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1192 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1193 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1194 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1195 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1196 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1197 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1199 HV_X64_MSR_VP_INDEX,
1200 HV_X64_MSR_VP_RUNTIME,
1201 HV_X64_MSR_SCONTROL,
1202 HV_X64_MSR_STIMER0_CONFIG,
1203 HV_X64_MSR_VP_ASSIST_PAGE,
1204 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1205 HV_X64_MSR_TSC_EMULATION_STATUS,
1207 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1210 MSR_IA32_TSC_ADJUST,
1211 MSR_IA32_TSCDEADLINE,
1212 MSR_IA32_ARCH_CAPABILITIES,
1213 MSR_IA32_MISC_ENABLE,
1214 MSR_IA32_MCG_STATUS,
1216 MSR_IA32_MCG_EXT_CTL,
1220 MSR_MISC_FEATURES_ENABLES,
1221 MSR_AMD64_VIRT_SPEC_CTRL,
1225 * The following list leaves out MSRs whose values are determined
1226 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1227 * We always support the "true" VMX control MSRs, even if the host
1228 * processor does not, so I am putting these registers here rather
1229 * than in msrs_to_save.
1232 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1233 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1234 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1235 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1237 MSR_IA32_VMX_CR0_FIXED0,
1238 MSR_IA32_VMX_CR4_FIXED0,
1239 MSR_IA32_VMX_VMCS_ENUM,
1240 MSR_IA32_VMX_PROCBASED_CTLS2,
1241 MSR_IA32_VMX_EPT_VPID_CAP,
1242 MSR_IA32_VMX_VMFUNC,
1245 MSR_KVM_POLL_CONTROL,
1248 static unsigned num_emulated_msrs;
1251 * List of msr numbers which are used to expose MSR-based features that
1252 * can be used by a hypervisor to validate requested CPU features.
1254 static u32 msr_based_features[] = {
1256 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1257 MSR_IA32_VMX_PINBASED_CTLS,
1258 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1259 MSR_IA32_VMX_PROCBASED_CTLS,
1260 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1261 MSR_IA32_VMX_EXIT_CTLS,
1262 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1263 MSR_IA32_VMX_ENTRY_CTLS,
1265 MSR_IA32_VMX_CR0_FIXED0,
1266 MSR_IA32_VMX_CR0_FIXED1,
1267 MSR_IA32_VMX_CR4_FIXED0,
1268 MSR_IA32_VMX_CR4_FIXED1,
1269 MSR_IA32_VMX_VMCS_ENUM,
1270 MSR_IA32_VMX_PROCBASED_CTLS2,
1271 MSR_IA32_VMX_EPT_VPID_CAP,
1272 MSR_IA32_VMX_VMFUNC,
1276 MSR_IA32_ARCH_CAPABILITIES,
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 we're doing cache flushes (either "always" or "cond")
1290 * we will do one whenever the guest does a vmlaunch/vmresume.
1291 * If an outer hypervisor is doing the cache flush for us
1292 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1293 * capability to the guest too, and if EPT is disabled we're not
1294 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1295 * require a nested hypervisor to do a flush of its own.
1297 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1298 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1300 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1301 data |= ARCH_CAP_RDCL_NO;
1302 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1303 data |= ARCH_CAP_SSB_NO;
1304 if (!boot_cpu_has_bug(X86_BUG_MDS))
1305 data |= ARCH_CAP_MDS_NO;
1310 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1312 switch (msr->index) {
1313 case MSR_IA32_ARCH_CAPABILITIES:
1314 msr->data = kvm_get_arch_capabilities();
1316 case MSR_IA32_UCODE_REV:
1317 rdmsrl_safe(msr->index, &msr->data);
1320 if (kvm_x86_ops->get_msr_feature(msr))
1326 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1328 struct kvm_msr_entry msr;
1332 r = kvm_get_msr_feature(&msr);
1341 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1343 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1346 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1349 if (efer & (EFER_LME | EFER_LMA) &&
1350 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1353 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1359 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1361 if (efer & efer_reserved_bits)
1364 return __kvm_valid_efer(vcpu, efer);
1366 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1368 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1370 u64 old_efer = vcpu->arch.efer;
1371 u64 efer = msr_info->data;
1373 if (efer & efer_reserved_bits)
1376 if (!msr_info->host_initiated) {
1377 if (!__kvm_valid_efer(vcpu, efer))
1380 if (is_paging(vcpu) &&
1381 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1386 efer |= vcpu->arch.efer & EFER_LMA;
1388 kvm_x86_ops->set_efer(vcpu, efer);
1390 /* Update reserved bits */
1391 if ((efer ^ old_efer) & EFER_NX)
1392 kvm_mmu_reset_context(vcpu);
1397 void kvm_enable_efer_bits(u64 mask)
1399 efer_reserved_bits &= ~mask;
1401 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1404 * Write @data into the MSR specified by @index. Select MSR specific fault
1405 * checks are bypassed if @host_initiated is %true.
1406 * Returns 0 on success, non-0 otherwise.
1407 * Assumes vcpu_load() was already called.
1409 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1410 bool host_initiated)
1412 struct msr_data msr;
1417 case MSR_KERNEL_GS_BASE:
1420 if (is_noncanonical_address(data, vcpu))
1423 case MSR_IA32_SYSENTER_EIP:
1424 case MSR_IA32_SYSENTER_ESP:
1426 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1427 * non-canonical address is written on Intel but not on
1428 * AMD (which ignores the top 32-bits, because it does
1429 * not implement 64-bit SYSENTER).
1431 * 64-bit code should hence be able to write a non-canonical
1432 * value on AMD. Making the address canonical ensures that
1433 * vmentry does not fail on Intel after writing a non-canonical
1434 * value, and that something deterministic happens if the guest
1435 * invokes 64-bit SYSENTER.
1437 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1442 msr.host_initiated = host_initiated;
1444 return kvm_x86_ops->set_msr(vcpu, &msr);
1448 * Read the MSR specified by @index into @data. Select MSR specific fault
1449 * checks are bypassed if @host_initiated is %true.
1450 * Returns 0 on success, non-0 otherwise.
1451 * Assumes vcpu_load() was already called.
1453 static int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1454 bool host_initiated)
1456 struct msr_data msr;
1460 msr.host_initiated = host_initiated;
1462 ret = kvm_x86_ops->get_msr(vcpu, &msr);
1468 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1470 return __kvm_get_msr(vcpu, index, data, false);
1472 EXPORT_SYMBOL_GPL(kvm_get_msr);
1474 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1476 return __kvm_set_msr(vcpu, index, data, false);
1478 EXPORT_SYMBOL_GPL(kvm_set_msr);
1480 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1482 u32 ecx = kvm_rcx_read(vcpu);
1485 if (kvm_get_msr(vcpu, ecx, &data)) {
1486 trace_kvm_msr_read_ex(ecx);
1487 kvm_inject_gp(vcpu, 0);
1491 trace_kvm_msr_read(ecx, data);
1493 kvm_rax_write(vcpu, data & -1u);
1494 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1495 return kvm_skip_emulated_instruction(vcpu);
1497 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1499 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1501 u32 ecx = kvm_rcx_read(vcpu);
1502 u64 data = kvm_read_edx_eax(vcpu);
1504 if (kvm_set_msr(vcpu, ecx, data)) {
1505 trace_kvm_msr_write_ex(ecx, data);
1506 kvm_inject_gp(vcpu, 0);
1510 trace_kvm_msr_write(ecx, data);
1511 return kvm_skip_emulated_instruction(vcpu);
1513 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1516 * Adapt set_msr() to msr_io()'s calling convention
1518 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1520 return __kvm_get_msr(vcpu, index, data, true);
1523 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1525 return __kvm_set_msr(vcpu, index, *data, true);
1528 #ifdef CONFIG_X86_64
1529 struct pvclock_gtod_data {
1532 struct { /* extract of a clocksource struct */
1545 static struct pvclock_gtod_data pvclock_gtod_data;
1547 static void update_pvclock_gtod(struct timekeeper *tk)
1549 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1552 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1554 write_seqcount_begin(&vdata->seq);
1556 /* copy pvclock gtod data */
1557 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1558 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1559 vdata->clock.mask = tk->tkr_mono.mask;
1560 vdata->clock.mult = tk->tkr_mono.mult;
1561 vdata->clock.shift = tk->tkr_mono.shift;
1563 vdata->boot_ns = boot_ns;
1564 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1566 vdata->wall_time_sec = tk->xtime_sec;
1568 write_seqcount_end(&vdata->seq);
1572 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1574 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1575 kvm_vcpu_kick(vcpu);
1578 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1582 struct pvclock_wall_clock wc;
1583 struct timespec64 boot;
1588 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1593 ++version; /* first time write, random junk */
1597 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1601 * The guest calculates current wall clock time by adding
1602 * system time (updated by kvm_guest_time_update below) to the
1603 * wall clock specified here. guest system time equals host
1604 * system time for us, thus we must fill in host boot time here.
1606 getboottime64(&boot);
1608 if (kvm->arch.kvmclock_offset) {
1609 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1610 boot = timespec64_sub(boot, ts);
1612 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1613 wc.nsec = boot.tv_nsec;
1614 wc.version = version;
1616 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1619 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1622 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1624 do_shl32_div32(dividend, divisor);
1628 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1629 s8 *pshift, u32 *pmultiplier)
1637 scaled64 = scaled_hz;
1638 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1643 tps32 = (uint32_t)tps64;
1644 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1645 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1653 *pmultiplier = div_frac(scaled64, tps32);
1656 #ifdef CONFIG_X86_64
1657 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1660 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1661 static unsigned long max_tsc_khz;
1663 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1665 u64 v = (u64)khz * (1000000 + ppm);
1670 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1674 /* Guest TSC same frequency as host TSC? */
1676 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1680 /* TSC scaling supported? */
1681 if (!kvm_has_tsc_control) {
1682 if (user_tsc_khz > tsc_khz) {
1683 vcpu->arch.tsc_catchup = 1;
1684 vcpu->arch.tsc_always_catchup = 1;
1687 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1692 /* TSC scaling required - calculate ratio */
1693 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1694 user_tsc_khz, tsc_khz);
1696 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1697 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1702 vcpu->arch.tsc_scaling_ratio = ratio;
1706 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1708 u32 thresh_lo, thresh_hi;
1709 int use_scaling = 0;
1711 /* tsc_khz can be zero if TSC calibration fails */
1712 if (user_tsc_khz == 0) {
1713 /* set tsc_scaling_ratio to a safe value */
1714 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1718 /* Compute a scale to convert nanoseconds in TSC cycles */
1719 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1720 &vcpu->arch.virtual_tsc_shift,
1721 &vcpu->arch.virtual_tsc_mult);
1722 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1725 * Compute the variation in TSC rate which is acceptable
1726 * within the range of tolerance and decide if the
1727 * rate being applied is within that bounds of the hardware
1728 * rate. If so, no scaling or compensation need be done.
1730 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1731 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1732 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1733 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1736 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1739 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1741 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1742 vcpu->arch.virtual_tsc_mult,
1743 vcpu->arch.virtual_tsc_shift);
1744 tsc += vcpu->arch.this_tsc_write;
1748 static inline int gtod_is_based_on_tsc(int mode)
1750 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1753 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1755 #ifdef CONFIG_X86_64
1757 struct kvm_arch *ka = &vcpu->kvm->arch;
1758 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1760 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1761 atomic_read(&vcpu->kvm->online_vcpus));
1764 * Once the masterclock is enabled, always perform request in
1765 * order to update it.
1767 * In order to enable masterclock, the host clocksource must be TSC
1768 * and the vcpus need to have matched TSCs. When that happens,
1769 * perform request to enable masterclock.
1771 if (ka->use_master_clock ||
1772 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1773 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1775 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1776 atomic_read(&vcpu->kvm->online_vcpus),
1777 ka->use_master_clock, gtod->clock.vclock_mode);
1781 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1783 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1784 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1788 * Multiply tsc by a fixed point number represented by ratio.
1790 * The most significant 64-N bits (mult) of ratio represent the
1791 * integral part of the fixed point number; the remaining N bits
1792 * (frac) represent the fractional part, ie. ratio represents a fixed
1793 * point number (mult + frac * 2^(-N)).
1795 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1797 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1799 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1802 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1805 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1807 if (ratio != kvm_default_tsc_scaling_ratio)
1808 _tsc = __scale_tsc(ratio, tsc);
1812 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1814 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1818 tsc = kvm_scale_tsc(vcpu, rdtsc());
1820 return target_tsc - tsc;
1823 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1825 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1827 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1829 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1831 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1833 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1836 static inline bool kvm_check_tsc_unstable(void)
1838 #ifdef CONFIG_X86_64
1840 * TSC is marked unstable when we're running on Hyper-V,
1841 * 'TSC page' clocksource is good.
1843 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1846 return check_tsc_unstable();
1849 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1851 struct kvm *kvm = vcpu->kvm;
1852 u64 offset, ns, elapsed;
1853 unsigned long flags;
1855 bool already_matched;
1856 u64 data = msr->data;
1857 bool synchronizing = false;
1859 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1860 offset = kvm_compute_tsc_offset(vcpu, data);
1861 ns = ktime_get_boottime_ns();
1862 elapsed = ns - kvm->arch.last_tsc_nsec;
1864 if (vcpu->arch.virtual_tsc_khz) {
1865 if (data == 0 && msr->host_initiated) {
1867 * detection of vcpu initialization -- need to sync
1868 * with other vCPUs. This particularly helps to keep
1869 * kvm_clock stable after CPU hotplug
1871 synchronizing = true;
1873 u64 tsc_exp = kvm->arch.last_tsc_write +
1874 nsec_to_cycles(vcpu, elapsed);
1875 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1877 * Special case: TSC write with a small delta (1 second)
1878 * of virtual cycle time against real time is
1879 * interpreted as an attempt to synchronize the CPU.
1881 synchronizing = data < tsc_exp + tsc_hz &&
1882 data + tsc_hz > tsc_exp;
1887 * For a reliable TSC, we can match TSC offsets, and for an unstable
1888 * TSC, we add elapsed time in this computation. We could let the
1889 * compensation code attempt to catch up if we fall behind, but
1890 * it's better to try to match offsets from the beginning.
1892 if (synchronizing &&
1893 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1894 if (!kvm_check_tsc_unstable()) {
1895 offset = kvm->arch.cur_tsc_offset;
1897 u64 delta = nsec_to_cycles(vcpu, elapsed);
1899 offset = kvm_compute_tsc_offset(vcpu, data);
1902 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1905 * We split periods of matched TSC writes into generations.
1906 * For each generation, we track the original measured
1907 * nanosecond time, offset, and write, so if TSCs are in
1908 * sync, we can match exact offset, and if not, we can match
1909 * exact software computation in compute_guest_tsc()
1911 * These values are tracked in kvm->arch.cur_xxx variables.
1913 kvm->arch.cur_tsc_generation++;
1914 kvm->arch.cur_tsc_nsec = ns;
1915 kvm->arch.cur_tsc_write = data;
1916 kvm->arch.cur_tsc_offset = offset;
1921 * We also track th most recent recorded KHZ, write and time to
1922 * allow the matching interval to be extended at each write.
1924 kvm->arch.last_tsc_nsec = ns;
1925 kvm->arch.last_tsc_write = data;
1926 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1928 vcpu->arch.last_guest_tsc = data;
1930 /* Keep track of which generation this VCPU has synchronized to */
1931 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1932 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1933 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1935 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1936 update_ia32_tsc_adjust_msr(vcpu, offset);
1938 kvm_vcpu_write_tsc_offset(vcpu, offset);
1939 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1941 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1943 kvm->arch.nr_vcpus_matched_tsc = 0;
1944 } else if (!already_matched) {
1945 kvm->arch.nr_vcpus_matched_tsc++;
1948 kvm_track_tsc_matching(vcpu);
1949 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1952 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1954 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1957 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1958 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1961 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1963 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1964 WARN_ON(adjustment < 0);
1965 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1966 adjust_tsc_offset_guest(vcpu, adjustment);
1969 #ifdef CONFIG_X86_64
1971 static u64 read_tsc(void)
1973 u64 ret = (u64)rdtsc_ordered();
1974 u64 last = pvclock_gtod_data.clock.cycle_last;
1976 if (likely(ret >= last))
1980 * GCC likes to generate cmov here, but this branch is extremely
1981 * predictable (it's just a function of time and the likely is
1982 * very likely) and there's a data dependence, so force GCC
1983 * to generate a branch instead. I don't barrier() because
1984 * we don't actually need a barrier, and if this function
1985 * ever gets inlined it will generate worse code.
1991 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1994 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1997 switch (gtod->clock.vclock_mode) {
1998 case VCLOCK_HVCLOCK:
1999 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2001 if (tsc_pg_val != U64_MAX) {
2002 /* TSC page valid */
2003 *mode = VCLOCK_HVCLOCK;
2004 v = (tsc_pg_val - gtod->clock.cycle_last) &
2007 /* TSC page invalid */
2008 *mode = VCLOCK_NONE;
2013 *tsc_timestamp = read_tsc();
2014 v = (*tsc_timestamp - gtod->clock.cycle_last) &
2018 *mode = VCLOCK_NONE;
2021 if (*mode == VCLOCK_NONE)
2022 *tsc_timestamp = v = 0;
2024 return v * gtod->clock.mult;
2027 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
2029 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2035 seq = read_seqcount_begin(>od->seq);
2036 ns = gtod->nsec_base;
2037 ns += vgettsc(tsc_timestamp, &mode);
2038 ns >>= gtod->clock.shift;
2039 ns += gtod->boot_ns;
2040 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2046 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2048 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2054 seq = read_seqcount_begin(>od->seq);
2055 ts->tv_sec = gtod->wall_time_sec;
2056 ns = gtod->nsec_base;
2057 ns += vgettsc(tsc_timestamp, &mode);
2058 ns >>= gtod->clock.shift;
2059 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2061 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2067 /* returns true if host is using TSC based clocksource */
2068 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2070 /* checked again under seqlock below */
2071 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2074 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
2078 /* returns true if host is using TSC based clocksource */
2079 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2082 /* checked again under seqlock below */
2083 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2086 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2092 * Assuming a stable TSC across physical CPUS, and a stable TSC
2093 * across virtual CPUs, the following condition is possible.
2094 * Each numbered line represents an event visible to both
2095 * CPUs at the next numbered event.
2097 * "timespecX" represents host monotonic time. "tscX" represents
2100 * VCPU0 on CPU0 | VCPU1 on CPU1
2102 * 1. read timespec0,tsc0
2103 * 2. | timespec1 = timespec0 + N
2105 * 3. transition to guest | transition to guest
2106 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2107 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2108 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2110 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2113 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2115 * - 0 < N - M => M < N
2117 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2118 * always the case (the difference between two distinct xtime instances
2119 * might be smaller then the difference between corresponding TSC reads,
2120 * when updating guest vcpus pvclock areas).
2122 * To avoid that problem, do not allow visibility of distinct
2123 * system_timestamp/tsc_timestamp values simultaneously: use a master
2124 * copy of host monotonic time values. Update that master copy
2127 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2131 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2133 #ifdef CONFIG_X86_64
2134 struct kvm_arch *ka = &kvm->arch;
2136 bool host_tsc_clocksource, vcpus_matched;
2138 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2139 atomic_read(&kvm->online_vcpus));
2142 * If the host uses TSC clock, then passthrough TSC as stable
2145 host_tsc_clocksource = kvm_get_time_and_clockread(
2146 &ka->master_kernel_ns,
2147 &ka->master_cycle_now);
2149 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2150 && !ka->backwards_tsc_observed
2151 && !ka->boot_vcpu_runs_old_kvmclock;
2153 if (ka->use_master_clock)
2154 atomic_set(&kvm_guest_has_master_clock, 1);
2156 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2157 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2162 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2164 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2167 static void kvm_gen_update_masterclock(struct kvm *kvm)
2169 #ifdef CONFIG_X86_64
2171 struct kvm_vcpu *vcpu;
2172 struct kvm_arch *ka = &kvm->arch;
2174 spin_lock(&ka->pvclock_gtod_sync_lock);
2175 kvm_make_mclock_inprogress_request(kvm);
2176 /* no guest entries from this point */
2177 pvclock_update_vm_gtod_copy(kvm);
2179 kvm_for_each_vcpu(i, vcpu, kvm)
2180 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2182 /* guest entries allowed */
2183 kvm_for_each_vcpu(i, vcpu, kvm)
2184 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2186 spin_unlock(&ka->pvclock_gtod_sync_lock);
2190 u64 get_kvmclock_ns(struct kvm *kvm)
2192 struct kvm_arch *ka = &kvm->arch;
2193 struct pvclock_vcpu_time_info hv_clock;
2196 spin_lock(&ka->pvclock_gtod_sync_lock);
2197 if (!ka->use_master_clock) {
2198 spin_unlock(&ka->pvclock_gtod_sync_lock);
2199 return ktime_get_boottime_ns() + ka->kvmclock_offset;
2202 hv_clock.tsc_timestamp = ka->master_cycle_now;
2203 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2204 spin_unlock(&ka->pvclock_gtod_sync_lock);
2206 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2209 if (__this_cpu_read(cpu_tsc_khz)) {
2210 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2211 &hv_clock.tsc_shift,
2212 &hv_clock.tsc_to_system_mul);
2213 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2215 ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2222 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2224 struct kvm_vcpu_arch *vcpu = &v->arch;
2225 struct pvclock_vcpu_time_info guest_hv_clock;
2227 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2228 &guest_hv_clock, sizeof(guest_hv_clock))))
2231 /* This VCPU is paused, but it's legal for a guest to read another
2232 * VCPU's kvmclock, so we really have to follow the specification where
2233 * it says that version is odd if data is being modified, and even after
2236 * Version field updates must be kept separate. This is because
2237 * kvm_write_guest_cached might use a "rep movs" instruction, and
2238 * writes within a string instruction are weakly ordered. So there
2239 * are three writes overall.
2241 * As a small optimization, only write the version field in the first
2242 * and third write. The vcpu->pv_time cache is still valid, because the
2243 * version field is the first in the struct.
2245 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2247 if (guest_hv_clock.version & 1)
2248 ++guest_hv_clock.version; /* first time write, random junk */
2250 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2251 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2253 sizeof(vcpu->hv_clock.version));
2257 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2258 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2260 if (vcpu->pvclock_set_guest_stopped_request) {
2261 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2262 vcpu->pvclock_set_guest_stopped_request = false;
2265 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2267 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2269 sizeof(vcpu->hv_clock));
2273 vcpu->hv_clock.version++;
2274 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2276 sizeof(vcpu->hv_clock.version));
2279 static int kvm_guest_time_update(struct kvm_vcpu *v)
2281 unsigned long flags, tgt_tsc_khz;
2282 struct kvm_vcpu_arch *vcpu = &v->arch;
2283 struct kvm_arch *ka = &v->kvm->arch;
2285 u64 tsc_timestamp, host_tsc;
2287 bool use_master_clock;
2293 * If the host uses TSC clock, then passthrough TSC as stable
2296 spin_lock(&ka->pvclock_gtod_sync_lock);
2297 use_master_clock = ka->use_master_clock;
2298 if (use_master_clock) {
2299 host_tsc = ka->master_cycle_now;
2300 kernel_ns = ka->master_kernel_ns;
2302 spin_unlock(&ka->pvclock_gtod_sync_lock);
2304 /* Keep irq disabled to prevent changes to the clock */
2305 local_irq_save(flags);
2306 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2307 if (unlikely(tgt_tsc_khz == 0)) {
2308 local_irq_restore(flags);
2309 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2312 if (!use_master_clock) {
2314 kernel_ns = ktime_get_boottime_ns();
2317 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2320 * We may have to catch up the TSC to match elapsed wall clock
2321 * time for two reasons, even if kvmclock is used.
2322 * 1) CPU could have been running below the maximum TSC rate
2323 * 2) Broken TSC compensation resets the base at each VCPU
2324 * entry to avoid unknown leaps of TSC even when running
2325 * again on the same CPU. This may cause apparent elapsed
2326 * time to disappear, and the guest to stand still or run
2329 if (vcpu->tsc_catchup) {
2330 u64 tsc = compute_guest_tsc(v, kernel_ns);
2331 if (tsc > tsc_timestamp) {
2332 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2333 tsc_timestamp = tsc;
2337 local_irq_restore(flags);
2339 /* With all the info we got, fill in the values */
2341 if (kvm_has_tsc_control)
2342 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2344 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2345 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2346 &vcpu->hv_clock.tsc_shift,
2347 &vcpu->hv_clock.tsc_to_system_mul);
2348 vcpu->hw_tsc_khz = tgt_tsc_khz;
2351 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2352 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2353 vcpu->last_guest_tsc = tsc_timestamp;
2355 /* If the host uses TSC clocksource, then it is stable */
2357 if (use_master_clock)
2358 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2360 vcpu->hv_clock.flags = pvclock_flags;
2362 if (vcpu->pv_time_enabled)
2363 kvm_setup_pvclock_page(v);
2364 if (v == kvm_get_vcpu(v->kvm, 0))
2365 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2370 * kvmclock updates which are isolated to a given vcpu, such as
2371 * vcpu->cpu migration, should not allow system_timestamp from
2372 * the rest of the vcpus to remain static. Otherwise ntp frequency
2373 * correction applies to one vcpu's system_timestamp but not
2376 * So in those cases, request a kvmclock update for all vcpus.
2377 * We need to rate-limit these requests though, as they can
2378 * considerably slow guests that have a large number of vcpus.
2379 * The time for a remote vcpu to update its kvmclock is bound
2380 * by the delay we use to rate-limit the updates.
2383 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2385 static void kvmclock_update_fn(struct work_struct *work)
2388 struct delayed_work *dwork = to_delayed_work(work);
2389 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2390 kvmclock_update_work);
2391 struct kvm *kvm = container_of(ka, struct kvm, arch);
2392 struct kvm_vcpu *vcpu;
2394 kvm_for_each_vcpu(i, vcpu, kvm) {
2395 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2396 kvm_vcpu_kick(vcpu);
2400 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2402 struct kvm *kvm = v->kvm;
2404 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2405 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2406 KVMCLOCK_UPDATE_DELAY);
2409 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2411 static void kvmclock_sync_fn(struct work_struct *work)
2413 struct delayed_work *dwork = to_delayed_work(work);
2414 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2415 kvmclock_sync_work);
2416 struct kvm *kvm = container_of(ka, struct kvm, arch);
2418 if (!kvmclock_periodic_sync)
2421 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2422 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2423 KVMCLOCK_SYNC_PERIOD);
2427 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2429 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2431 /* McStatusWrEn enabled? */
2432 if (guest_cpuid_is_amd(vcpu))
2433 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2438 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2440 u64 mcg_cap = vcpu->arch.mcg_cap;
2441 unsigned bank_num = mcg_cap & 0xff;
2442 u32 msr = msr_info->index;
2443 u64 data = msr_info->data;
2446 case MSR_IA32_MCG_STATUS:
2447 vcpu->arch.mcg_status = data;
2449 case MSR_IA32_MCG_CTL:
2450 if (!(mcg_cap & MCG_CTL_P) &&
2451 (data || !msr_info->host_initiated))
2453 if (data != 0 && data != ~(u64)0)
2455 vcpu->arch.mcg_ctl = data;
2458 if (msr >= MSR_IA32_MC0_CTL &&
2459 msr < MSR_IA32_MCx_CTL(bank_num)) {
2460 u32 offset = msr - MSR_IA32_MC0_CTL;
2461 /* only 0 or all 1s can be written to IA32_MCi_CTL
2462 * some Linux kernels though clear bit 10 in bank 4 to
2463 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2464 * this to avoid an uncatched #GP in the guest
2466 if ((offset & 0x3) == 0 &&
2467 data != 0 && (data | (1 << 10)) != ~(u64)0)
2471 if (!msr_info->host_initiated &&
2472 (offset & 0x3) == 1 && data != 0) {
2473 if (!can_set_mci_status(vcpu))
2477 vcpu->arch.mce_banks[offset] = data;
2485 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2487 struct kvm *kvm = vcpu->kvm;
2488 int lm = is_long_mode(vcpu);
2489 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2490 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2491 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2492 : kvm->arch.xen_hvm_config.blob_size_32;
2493 u32 page_num = data & ~PAGE_MASK;
2494 u64 page_addr = data & PAGE_MASK;
2499 if (page_num >= blob_size)
2502 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2507 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2516 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2518 gpa_t gpa = data & ~0x3f;
2520 /* Bits 3:5 are reserved, Should be zero */
2524 vcpu->arch.apf.msr_val = data;
2526 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2527 kvm_clear_async_pf_completion_queue(vcpu);
2528 kvm_async_pf_hash_reset(vcpu);
2532 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2536 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2537 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2538 kvm_async_pf_wakeup_all(vcpu);
2542 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2544 vcpu->arch.pv_time_enabled = false;
2547 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2549 ++vcpu->stat.tlb_flush;
2550 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2553 static void record_steal_time(struct kvm_vcpu *vcpu)
2555 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2558 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2559 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2563 * Doing a TLB flush here, on the guest's behalf, can avoid
2566 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2567 vcpu->arch.st.steal.preempted & KVM_VCPU_FLUSH_TLB);
2568 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2569 kvm_vcpu_flush_tlb(vcpu, false);
2571 if (vcpu->arch.st.steal.version & 1)
2572 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2574 vcpu->arch.st.steal.version += 1;
2576 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2577 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2581 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2582 vcpu->arch.st.last_steal;
2583 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2585 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2586 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2590 vcpu->arch.st.steal.version += 1;
2592 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2593 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2596 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2599 u32 msr = msr_info->index;
2600 u64 data = msr_info->data;
2603 case MSR_AMD64_NB_CFG:
2604 case MSR_IA32_UCODE_WRITE:
2605 case MSR_VM_HSAVE_PA:
2606 case MSR_AMD64_PATCH_LOADER:
2607 case MSR_AMD64_BU_CFG2:
2608 case MSR_AMD64_DC_CFG:
2609 case MSR_F15H_EX_CFG:
2612 case MSR_IA32_UCODE_REV:
2613 if (msr_info->host_initiated)
2614 vcpu->arch.microcode_version = data;
2616 case MSR_IA32_ARCH_CAPABILITIES:
2617 if (!msr_info->host_initiated)
2619 vcpu->arch.arch_capabilities = data;
2622 return set_efer(vcpu, msr_info);
2624 data &= ~(u64)0x40; /* ignore flush filter disable */
2625 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2626 data &= ~(u64)0x8; /* ignore TLB cache disable */
2628 /* Handle McStatusWrEn */
2629 if (data == BIT_ULL(18)) {
2630 vcpu->arch.msr_hwcr = data;
2631 } else if (data != 0) {
2632 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2637 case MSR_FAM10H_MMIO_CONF_BASE:
2639 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2644 case MSR_IA32_DEBUGCTLMSR:
2646 /* We support the non-activated case already */
2648 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2649 /* Values other than LBR and BTF are vendor-specific,
2650 thus reserved and should throw a #GP */
2653 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2656 case 0x200 ... 0x2ff:
2657 return kvm_mtrr_set_msr(vcpu, msr, data);
2658 case MSR_IA32_APICBASE:
2659 return kvm_set_apic_base(vcpu, msr_info);
2660 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2661 return kvm_x2apic_msr_write(vcpu, msr, data);
2662 case MSR_IA32_TSCDEADLINE:
2663 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2665 case MSR_IA32_TSC_ADJUST:
2666 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2667 if (!msr_info->host_initiated) {
2668 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2669 adjust_tsc_offset_guest(vcpu, adj);
2671 vcpu->arch.ia32_tsc_adjust_msr = data;
2674 case MSR_IA32_MISC_ENABLE:
2675 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2676 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2677 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2679 vcpu->arch.ia32_misc_enable_msr = data;
2680 kvm_update_cpuid(vcpu);
2682 vcpu->arch.ia32_misc_enable_msr = data;
2685 case MSR_IA32_SMBASE:
2686 if (!msr_info->host_initiated)
2688 vcpu->arch.smbase = data;
2690 case MSR_IA32_POWER_CTL:
2691 vcpu->arch.msr_ia32_power_ctl = data;
2694 kvm_write_tsc(vcpu, msr_info);
2697 if (!msr_info->host_initiated)
2699 vcpu->arch.smi_count = data;
2701 case MSR_KVM_WALL_CLOCK_NEW:
2702 case MSR_KVM_WALL_CLOCK:
2703 vcpu->kvm->arch.wall_clock = data;
2704 kvm_write_wall_clock(vcpu->kvm, data);
2706 case MSR_KVM_SYSTEM_TIME_NEW:
2707 case MSR_KVM_SYSTEM_TIME: {
2708 struct kvm_arch *ka = &vcpu->kvm->arch;
2710 kvmclock_reset(vcpu);
2712 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2713 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2715 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2716 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2718 ka->boot_vcpu_runs_old_kvmclock = tmp;
2721 vcpu->arch.time = data;
2722 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2724 /* we verify if the enable bit is set... */
2728 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2729 &vcpu->arch.pv_time, data & ~1ULL,
2730 sizeof(struct pvclock_vcpu_time_info)))
2731 vcpu->arch.pv_time_enabled = false;
2733 vcpu->arch.pv_time_enabled = true;
2737 case MSR_KVM_ASYNC_PF_EN:
2738 if (kvm_pv_enable_async_pf(vcpu, data))
2741 case MSR_KVM_STEAL_TIME:
2743 if (unlikely(!sched_info_on()))
2746 if (data & KVM_STEAL_RESERVED_MASK)
2749 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2750 data & KVM_STEAL_VALID_BITS,
2751 sizeof(struct kvm_steal_time)))
2754 vcpu->arch.st.msr_val = data;
2756 if (!(data & KVM_MSR_ENABLED))
2759 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2762 case MSR_KVM_PV_EOI_EN:
2763 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2767 case MSR_KVM_POLL_CONTROL:
2768 /* only enable bit supported */
2769 if (data & (-1ULL << 1))
2772 vcpu->arch.msr_kvm_poll_control = data;
2775 case MSR_IA32_MCG_CTL:
2776 case MSR_IA32_MCG_STATUS:
2777 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2778 return set_msr_mce(vcpu, msr_info);
2780 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2781 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2782 pr = true; /* fall through */
2783 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2784 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2785 if (kvm_pmu_is_valid_msr(vcpu, msr))
2786 return kvm_pmu_set_msr(vcpu, msr_info);
2788 if (pr || data != 0)
2789 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2790 "0x%x data 0x%llx\n", msr, data);
2792 case MSR_K7_CLK_CTL:
2794 * Ignore all writes to this no longer documented MSR.
2795 * Writes are only relevant for old K7 processors,
2796 * all pre-dating SVM, but a recommended workaround from
2797 * AMD for these chips. It is possible to specify the
2798 * affected processor models on the command line, hence
2799 * the need to ignore the workaround.
2802 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2803 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2804 case HV_X64_MSR_CRASH_CTL:
2805 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2806 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2807 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2808 case HV_X64_MSR_TSC_EMULATION_STATUS:
2809 return kvm_hv_set_msr_common(vcpu, msr, data,
2810 msr_info->host_initiated);
2811 case MSR_IA32_BBL_CR_CTL3:
2812 /* Drop writes to this legacy MSR -- see rdmsr
2813 * counterpart for further detail.
2815 if (report_ignored_msrs)
2816 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2819 case MSR_AMD64_OSVW_ID_LENGTH:
2820 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2822 vcpu->arch.osvw.length = data;
2824 case MSR_AMD64_OSVW_STATUS:
2825 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2827 vcpu->arch.osvw.status = data;
2829 case MSR_PLATFORM_INFO:
2830 if (!msr_info->host_initiated ||
2831 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2832 cpuid_fault_enabled(vcpu)))
2834 vcpu->arch.msr_platform_info = data;
2836 case MSR_MISC_FEATURES_ENABLES:
2837 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2838 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2839 !supports_cpuid_fault(vcpu)))
2841 vcpu->arch.msr_misc_features_enables = data;
2844 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2845 return xen_hvm_config(vcpu, data);
2846 if (kvm_pmu_is_valid_msr(vcpu, msr))
2847 return kvm_pmu_set_msr(vcpu, msr_info);
2849 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2853 if (report_ignored_msrs)
2855 "ignored wrmsr: 0x%x data 0x%llx\n",
2862 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2864 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2867 u64 mcg_cap = vcpu->arch.mcg_cap;
2868 unsigned bank_num = mcg_cap & 0xff;
2871 case MSR_IA32_P5_MC_ADDR:
2872 case MSR_IA32_P5_MC_TYPE:
2875 case MSR_IA32_MCG_CAP:
2876 data = vcpu->arch.mcg_cap;
2878 case MSR_IA32_MCG_CTL:
2879 if (!(mcg_cap & MCG_CTL_P) && !host)
2881 data = vcpu->arch.mcg_ctl;
2883 case MSR_IA32_MCG_STATUS:
2884 data = vcpu->arch.mcg_status;
2887 if (msr >= MSR_IA32_MC0_CTL &&
2888 msr < MSR_IA32_MCx_CTL(bank_num)) {
2889 u32 offset = msr - MSR_IA32_MC0_CTL;
2890 data = vcpu->arch.mce_banks[offset];
2899 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2901 switch (msr_info->index) {
2902 case MSR_IA32_PLATFORM_ID:
2903 case MSR_IA32_EBL_CR_POWERON:
2904 case MSR_IA32_DEBUGCTLMSR:
2905 case MSR_IA32_LASTBRANCHFROMIP:
2906 case MSR_IA32_LASTBRANCHTOIP:
2907 case MSR_IA32_LASTINTFROMIP:
2908 case MSR_IA32_LASTINTTOIP:
2910 case MSR_K8_TSEG_ADDR:
2911 case MSR_K8_TSEG_MASK:
2912 case MSR_VM_HSAVE_PA:
2913 case MSR_K8_INT_PENDING_MSG:
2914 case MSR_AMD64_NB_CFG:
2915 case MSR_FAM10H_MMIO_CONF_BASE:
2916 case MSR_AMD64_BU_CFG2:
2917 case MSR_IA32_PERF_CTL:
2918 case MSR_AMD64_DC_CFG:
2919 case MSR_F15H_EX_CFG:
2922 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2923 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2924 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2925 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2926 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2927 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2928 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2931 case MSR_IA32_UCODE_REV:
2932 msr_info->data = vcpu->arch.microcode_version;
2934 case MSR_IA32_ARCH_CAPABILITIES:
2935 if (!msr_info->host_initiated &&
2936 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2938 msr_info->data = vcpu->arch.arch_capabilities;
2940 case MSR_IA32_POWER_CTL:
2941 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2944 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2947 case 0x200 ... 0x2ff:
2948 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2949 case 0xcd: /* fsb frequency */
2953 * MSR_EBC_FREQUENCY_ID
2954 * Conservative value valid for even the basic CPU models.
2955 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2956 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2957 * and 266MHz for model 3, or 4. Set Core Clock
2958 * Frequency to System Bus Frequency Ratio to 1 (bits
2959 * 31:24) even though these are only valid for CPU
2960 * models > 2, however guests may end up dividing or
2961 * multiplying by zero otherwise.
2963 case MSR_EBC_FREQUENCY_ID:
2964 msr_info->data = 1 << 24;
2966 case MSR_IA32_APICBASE:
2967 msr_info->data = kvm_get_apic_base(vcpu);
2969 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2970 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2972 case MSR_IA32_TSCDEADLINE:
2973 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2975 case MSR_IA32_TSC_ADJUST:
2976 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2978 case MSR_IA32_MISC_ENABLE:
2979 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2981 case MSR_IA32_SMBASE:
2982 if (!msr_info->host_initiated)
2984 msr_info->data = vcpu->arch.smbase;
2987 msr_info->data = vcpu->arch.smi_count;
2989 case MSR_IA32_PERF_STATUS:
2990 /* TSC increment by tick */
2991 msr_info->data = 1000ULL;
2992 /* CPU multiplier */
2993 msr_info->data |= (((uint64_t)4ULL) << 40);
2996 msr_info->data = vcpu->arch.efer;
2998 case MSR_KVM_WALL_CLOCK:
2999 case MSR_KVM_WALL_CLOCK_NEW:
3000 msr_info->data = vcpu->kvm->arch.wall_clock;
3002 case MSR_KVM_SYSTEM_TIME:
3003 case MSR_KVM_SYSTEM_TIME_NEW:
3004 msr_info->data = vcpu->arch.time;
3006 case MSR_KVM_ASYNC_PF_EN:
3007 msr_info->data = vcpu->arch.apf.msr_val;
3009 case MSR_KVM_STEAL_TIME:
3010 msr_info->data = vcpu->arch.st.msr_val;
3012 case MSR_KVM_PV_EOI_EN:
3013 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3015 case MSR_KVM_POLL_CONTROL:
3016 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3018 case MSR_IA32_P5_MC_ADDR:
3019 case MSR_IA32_P5_MC_TYPE:
3020 case MSR_IA32_MCG_CAP:
3021 case MSR_IA32_MCG_CTL:
3022 case MSR_IA32_MCG_STATUS:
3023 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3024 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3025 msr_info->host_initiated);
3026 case MSR_K7_CLK_CTL:
3028 * Provide expected ramp-up count for K7. All other
3029 * are set to zero, indicating minimum divisors for
3032 * This prevents guest kernels on AMD host with CPU
3033 * type 6, model 8 and higher from exploding due to
3034 * the rdmsr failing.
3036 msr_info->data = 0x20000000;
3038 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3039 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3040 case HV_X64_MSR_CRASH_CTL:
3041 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3042 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3043 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3044 case HV_X64_MSR_TSC_EMULATION_STATUS:
3045 return kvm_hv_get_msr_common(vcpu,
3046 msr_info->index, &msr_info->data,
3047 msr_info->host_initiated);
3049 case MSR_IA32_BBL_CR_CTL3:
3050 /* This legacy MSR exists but isn't fully documented in current
3051 * silicon. It is however accessed by winxp in very narrow
3052 * scenarios where it sets bit #19, itself documented as
3053 * a "reserved" bit. Best effort attempt to source coherent
3054 * read data here should the balance of the register be
3055 * interpreted by the guest:
3057 * L2 cache control register 3: 64GB range, 256KB size,
3058 * enabled, latency 0x1, configured
3060 msr_info->data = 0xbe702111;
3062 case MSR_AMD64_OSVW_ID_LENGTH:
3063 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3065 msr_info->data = vcpu->arch.osvw.length;
3067 case MSR_AMD64_OSVW_STATUS:
3068 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3070 msr_info->data = vcpu->arch.osvw.status;
3072 case MSR_PLATFORM_INFO:
3073 if (!msr_info->host_initiated &&
3074 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3076 msr_info->data = vcpu->arch.msr_platform_info;
3078 case MSR_MISC_FEATURES_ENABLES:
3079 msr_info->data = vcpu->arch.msr_misc_features_enables;
3082 msr_info->data = vcpu->arch.msr_hwcr;
3085 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3086 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3088 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3092 if (report_ignored_msrs)
3093 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3101 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3104 * Read or write a bunch of msrs. All parameters are kernel addresses.
3106 * @return number of msrs set successfully.
3108 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3109 struct kvm_msr_entry *entries,
3110 int (*do_msr)(struct kvm_vcpu *vcpu,
3111 unsigned index, u64 *data))
3115 for (i = 0; i < msrs->nmsrs; ++i)
3116 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3123 * Read or write a bunch of msrs. Parameters are user addresses.
3125 * @return number of msrs set successfully.
3127 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3128 int (*do_msr)(struct kvm_vcpu *vcpu,
3129 unsigned index, u64 *data),
3132 struct kvm_msrs msrs;
3133 struct kvm_msr_entry *entries;
3138 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3142 if (msrs.nmsrs >= MAX_IO_MSRS)
3145 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3146 entries = memdup_user(user_msrs->entries, size);
3147 if (IS_ERR(entries)) {
3148 r = PTR_ERR(entries);
3152 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3157 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3168 static inline bool kvm_can_mwait_in_guest(void)
3170 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3171 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3172 boot_cpu_has(X86_FEATURE_ARAT);
3175 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3180 case KVM_CAP_IRQCHIP:
3182 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3183 case KVM_CAP_SET_TSS_ADDR:
3184 case KVM_CAP_EXT_CPUID:
3185 case KVM_CAP_EXT_EMUL_CPUID:
3186 case KVM_CAP_CLOCKSOURCE:
3188 case KVM_CAP_NOP_IO_DELAY:
3189 case KVM_CAP_MP_STATE:
3190 case KVM_CAP_SYNC_MMU:
3191 case KVM_CAP_USER_NMI:
3192 case KVM_CAP_REINJECT_CONTROL:
3193 case KVM_CAP_IRQ_INJECT_STATUS:
3194 case KVM_CAP_IOEVENTFD:
3195 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3197 case KVM_CAP_PIT_STATE2:
3198 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3199 case KVM_CAP_XEN_HVM:
3200 case KVM_CAP_VCPU_EVENTS:
3201 case KVM_CAP_HYPERV:
3202 case KVM_CAP_HYPERV_VAPIC:
3203 case KVM_CAP_HYPERV_SPIN:
3204 case KVM_CAP_HYPERV_SYNIC:
3205 case KVM_CAP_HYPERV_SYNIC2:
3206 case KVM_CAP_HYPERV_VP_INDEX:
3207 case KVM_CAP_HYPERV_EVENTFD:
3208 case KVM_CAP_HYPERV_TLBFLUSH:
3209 case KVM_CAP_HYPERV_SEND_IPI:
3210 case KVM_CAP_HYPERV_CPUID:
3211 case KVM_CAP_PCI_SEGMENT:
3212 case KVM_CAP_DEBUGREGS:
3213 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3215 case KVM_CAP_ASYNC_PF:
3216 case KVM_CAP_GET_TSC_KHZ:
3217 case KVM_CAP_KVMCLOCK_CTRL:
3218 case KVM_CAP_READONLY_MEM:
3219 case KVM_CAP_HYPERV_TIME:
3220 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3221 case KVM_CAP_TSC_DEADLINE_TIMER:
3222 case KVM_CAP_DISABLE_QUIRKS:
3223 case KVM_CAP_SET_BOOT_CPU_ID:
3224 case KVM_CAP_SPLIT_IRQCHIP:
3225 case KVM_CAP_IMMEDIATE_EXIT:
3226 case KVM_CAP_PMU_EVENT_FILTER:
3227 case KVM_CAP_GET_MSR_FEATURES:
3228 case KVM_CAP_MSR_PLATFORM_INFO:
3229 case KVM_CAP_EXCEPTION_PAYLOAD:
3232 case KVM_CAP_SYNC_REGS:
3233 r = KVM_SYNC_X86_VALID_FIELDS;
3235 case KVM_CAP_ADJUST_CLOCK:
3236 r = KVM_CLOCK_TSC_STABLE;
3238 case KVM_CAP_X86_DISABLE_EXITS:
3239 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3240 KVM_X86_DISABLE_EXITS_CSTATE;
3241 if(kvm_can_mwait_in_guest())
3242 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3244 case KVM_CAP_X86_SMM:
3245 /* SMBASE is usually relocated above 1M on modern chipsets,
3246 * and SMM handlers might indeed rely on 4G segment limits,
3247 * so do not report SMM to be available if real mode is
3248 * emulated via vm86 mode. Still, do not go to great lengths
3249 * to avoid userspace's usage of the feature, because it is a
3250 * fringe case that is not enabled except via specific settings
3251 * of the module parameters.
3253 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3256 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3258 case KVM_CAP_NR_VCPUS:
3259 r = KVM_SOFT_MAX_VCPUS;
3261 case KVM_CAP_MAX_VCPUS:
3264 case KVM_CAP_MAX_VCPU_ID:
3265 r = KVM_MAX_VCPU_ID;
3267 case KVM_CAP_PV_MMU: /* obsolete */
3271 r = KVM_MAX_MCE_BANKS;
3274 r = boot_cpu_has(X86_FEATURE_XSAVE);
3276 case KVM_CAP_TSC_CONTROL:
3277 r = kvm_has_tsc_control;
3279 case KVM_CAP_X2APIC_API:
3280 r = KVM_X2APIC_API_VALID_FLAGS;
3282 case KVM_CAP_NESTED_STATE:
3283 r = kvm_x86_ops->get_nested_state ?
3284 kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3286 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3287 r = kvm_x86_ops->enable_direct_tlbflush != NULL;
3289 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3290 r = kvm_x86_ops->nested_enable_evmcs != NULL;
3299 long kvm_arch_dev_ioctl(struct file *filp,
3300 unsigned int ioctl, unsigned long arg)
3302 void __user *argp = (void __user *)arg;
3306 case KVM_GET_MSR_INDEX_LIST: {
3307 struct kvm_msr_list __user *user_msr_list = argp;
3308 struct kvm_msr_list msr_list;
3312 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3315 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3316 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3319 if (n < msr_list.nmsrs)
3322 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3323 num_msrs_to_save * sizeof(u32)))
3325 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3327 num_emulated_msrs * sizeof(u32)))
3332 case KVM_GET_SUPPORTED_CPUID:
3333 case KVM_GET_EMULATED_CPUID: {
3334 struct kvm_cpuid2 __user *cpuid_arg = argp;
3335 struct kvm_cpuid2 cpuid;
3338 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3341 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3347 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3352 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3354 if (copy_to_user(argp, &kvm_mce_cap_supported,
3355 sizeof(kvm_mce_cap_supported)))
3359 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3360 struct kvm_msr_list __user *user_msr_list = argp;
3361 struct kvm_msr_list msr_list;
3365 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3368 msr_list.nmsrs = num_msr_based_features;
3369 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3372 if (n < msr_list.nmsrs)
3375 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3376 num_msr_based_features * sizeof(u32)))
3382 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3392 static void wbinvd_ipi(void *garbage)
3397 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3399 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3402 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3404 /* Address WBINVD may be executed by guest */
3405 if (need_emulate_wbinvd(vcpu)) {
3406 if (kvm_x86_ops->has_wbinvd_exit())
3407 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3408 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3409 smp_call_function_single(vcpu->cpu,
3410 wbinvd_ipi, NULL, 1);
3413 kvm_x86_ops->vcpu_load(vcpu, cpu);
3415 fpregs_assert_state_consistent();
3416 if (test_thread_flag(TIF_NEED_FPU_LOAD))
3417 switch_fpu_return();
3419 /* Apply any externally detected TSC adjustments (due to suspend) */
3420 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3421 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3422 vcpu->arch.tsc_offset_adjustment = 0;
3423 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3426 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3427 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3428 rdtsc() - vcpu->arch.last_host_tsc;
3430 mark_tsc_unstable("KVM discovered backwards TSC");
3432 if (kvm_check_tsc_unstable()) {
3433 u64 offset = kvm_compute_tsc_offset(vcpu,
3434 vcpu->arch.last_guest_tsc);
3435 kvm_vcpu_write_tsc_offset(vcpu, offset);
3436 vcpu->arch.tsc_catchup = 1;
3439 if (kvm_lapic_hv_timer_in_use(vcpu))
3440 kvm_lapic_restart_hv_timer(vcpu);
3443 * On a host with synchronized TSC, there is no need to update
3444 * kvmclock on vcpu->cpu migration
3446 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3447 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3448 if (vcpu->cpu != cpu)
3449 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3453 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3456 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3458 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3461 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3463 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3464 &vcpu->arch.st.steal.preempted,
3465 offsetof(struct kvm_steal_time, preempted),
3466 sizeof(vcpu->arch.st.steal.preempted));
3469 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3473 if (vcpu->preempted)
3474 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3477 * Disable page faults because we're in atomic context here.
3478 * kvm_write_guest_offset_cached() would call might_fault()
3479 * that relies on pagefault_disable() to tell if there's a
3480 * bug. NOTE: the write to guest memory may not go through if
3481 * during postcopy live migration or if there's heavy guest
3484 pagefault_disable();
3486 * kvm_memslots() will be called by
3487 * kvm_write_guest_offset_cached() so take the srcu lock.
3489 idx = srcu_read_lock(&vcpu->kvm->srcu);
3490 kvm_steal_time_set_preempted(vcpu);
3491 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3493 kvm_x86_ops->vcpu_put(vcpu);
3494 vcpu->arch.last_host_tsc = rdtsc();
3496 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3497 * on every vmexit, but if not, we might have a stale dr6 from the
3498 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3503 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3504 struct kvm_lapic_state *s)
3506 if (vcpu->arch.apicv_active)
3507 kvm_x86_ops->sync_pir_to_irr(vcpu);
3509 return kvm_apic_get_state(vcpu, s);
3512 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3513 struct kvm_lapic_state *s)
3517 r = kvm_apic_set_state(vcpu, s);
3520 update_cr8_intercept(vcpu);
3525 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3527 return (!lapic_in_kernel(vcpu) ||
3528 kvm_apic_accept_pic_intr(vcpu));
3532 * if userspace requested an interrupt window, check that the
3533 * interrupt window is open.
3535 * No need to exit to userspace if we already have an interrupt queued.
3537 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3539 return kvm_arch_interrupt_allowed(vcpu) &&
3540 !kvm_cpu_has_interrupt(vcpu) &&
3541 !kvm_event_needs_reinjection(vcpu) &&
3542 kvm_cpu_accept_dm_intr(vcpu);
3545 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3546 struct kvm_interrupt *irq)
3548 if (irq->irq >= KVM_NR_INTERRUPTS)
3551 if (!irqchip_in_kernel(vcpu->kvm)) {
3552 kvm_queue_interrupt(vcpu, irq->irq, false);
3553 kvm_make_request(KVM_REQ_EVENT, vcpu);
3558 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3559 * fail for in-kernel 8259.
3561 if (pic_in_kernel(vcpu->kvm))
3564 if (vcpu->arch.pending_external_vector != -1)
3567 vcpu->arch.pending_external_vector = irq->irq;
3568 kvm_make_request(KVM_REQ_EVENT, vcpu);
3572 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3574 kvm_inject_nmi(vcpu);
3579 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3581 kvm_make_request(KVM_REQ_SMI, vcpu);
3586 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3587 struct kvm_tpr_access_ctl *tac)
3591 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3595 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3599 unsigned bank_num = mcg_cap & 0xff, bank;
3602 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3604 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3607 vcpu->arch.mcg_cap = mcg_cap;
3608 /* Init IA32_MCG_CTL to all 1s */
3609 if (mcg_cap & MCG_CTL_P)
3610 vcpu->arch.mcg_ctl = ~(u64)0;
3611 /* Init IA32_MCi_CTL to all 1s */
3612 for (bank = 0; bank < bank_num; bank++)
3613 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3615 kvm_x86_ops->setup_mce(vcpu);
3620 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3621 struct kvm_x86_mce *mce)
3623 u64 mcg_cap = vcpu->arch.mcg_cap;
3624 unsigned bank_num = mcg_cap & 0xff;
3625 u64 *banks = vcpu->arch.mce_banks;
3627 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3630 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3631 * reporting is disabled
3633 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3634 vcpu->arch.mcg_ctl != ~(u64)0)
3636 banks += 4 * mce->bank;
3638 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3639 * reporting is disabled for the bank
3641 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3643 if (mce->status & MCI_STATUS_UC) {
3644 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3645 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3646 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3649 if (banks[1] & MCI_STATUS_VAL)
3650 mce->status |= MCI_STATUS_OVER;
3651 banks[2] = mce->addr;
3652 banks[3] = mce->misc;
3653 vcpu->arch.mcg_status = mce->mcg_status;
3654 banks[1] = mce->status;
3655 kvm_queue_exception(vcpu, MC_VECTOR);
3656 } else if (!(banks[1] & MCI_STATUS_VAL)
3657 || !(banks[1] & MCI_STATUS_UC)) {
3658 if (banks[1] & MCI_STATUS_VAL)
3659 mce->status |= MCI_STATUS_OVER;
3660 banks[2] = mce->addr;
3661 banks[3] = mce->misc;
3662 banks[1] = mce->status;
3664 banks[1] |= MCI_STATUS_OVER;
3668 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3669 struct kvm_vcpu_events *events)
3674 * The API doesn't provide the instruction length for software
3675 * exceptions, so don't report them. As long as the guest RIP
3676 * isn't advanced, we should expect to encounter the exception
3679 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3680 events->exception.injected = 0;
3681 events->exception.pending = 0;
3683 events->exception.injected = vcpu->arch.exception.injected;
3684 events->exception.pending = vcpu->arch.exception.pending;
3686 * For ABI compatibility, deliberately conflate
3687 * pending and injected exceptions when
3688 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3690 if (!vcpu->kvm->arch.exception_payload_enabled)
3691 events->exception.injected |=
3692 vcpu->arch.exception.pending;
3694 events->exception.nr = vcpu->arch.exception.nr;
3695 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3696 events->exception.error_code = vcpu->arch.exception.error_code;
3697 events->exception_has_payload = vcpu->arch.exception.has_payload;
3698 events->exception_payload = vcpu->arch.exception.payload;
3700 events->interrupt.injected =
3701 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3702 events->interrupt.nr = vcpu->arch.interrupt.nr;
3703 events->interrupt.soft = 0;
3704 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3706 events->nmi.injected = vcpu->arch.nmi_injected;
3707 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3708 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3709 events->nmi.pad = 0;
3711 events->sipi_vector = 0; /* never valid when reporting to user space */
3713 events->smi.smm = is_smm(vcpu);
3714 events->smi.pending = vcpu->arch.smi_pending;
3715 events->smi.smm_inside_nmi =
3716 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3717 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3719 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3720 | KVM_VCPUEVENT_VALID_SHADOW
3721 | KVM_VCPUEVENT_VALID_SMM);
3722 if (vcpu->kvm->arch.exception_payload_enabled)
3723 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3725 memset(&events->reserved, 0, sizeof(events->reserved));
3728 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3730 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3731 struct kvm_vcpu_events *events)
3733 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3734 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3735 | KVM_VCPUEVENT_VALID_SHADOW
3736 | KVM_VCPUEVENT_VALID_SMM
3737 | KVM_VCPUEVENT_VALID_PAYLOAD))
3740 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3741 if (!vcpu->kvm->arch.exception_payload_enabled)
3743 if (events->exception.pending)
3744 events->exception.injected = 0;
3746 events->exception_has_payload = 0;
3748 events->exception.pending = 0;
3749 events->exception_has_payload = 0;
3752 if ((events->exception.injected || events->exception.pending) &&
3753 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3756 /* INITs are latched while in SMM */
3757 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3758 (events->smi.smm || events->smi.pending) &&
3759 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3763 vcpu->arch.exception.injected = events->exception.injected;
3764 vcpu->arch.exception.pending = events->exception.pending;
3765 vcpu->arch.exception.nr = events->exception.nr;
3766 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3767 vcpu->arch.exception.error_code = events->exception.error_code;
3768 vcpu->arch.exception.has_payload = events->exception_has_payload;
3769 vcpu->arch.exception.payload = events->exception_payload;
3771 vcpu->arch.interrupt.injected = events->interrupt.injected;
3772 vcpu->arch.interrupt.nr = events->interrupt.nr;
3773 vcpu->arch.interrupt.soft = events->interrupt.soft;
3774 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3775 kvm_x86_ops->set_interrupt_shadow(vcpu,
3776 events->interrupt.shadow);
3778 vcpu->arch.nmi_injected = events->nmi.injected;
3779 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3780 vcpu->arch.nmi_pending = events->nmi.pending;
3781 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3783 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3784 lapic_in_kernel(vcpu))
3785 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3787 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3788 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3789 if (events->smi.smm)
3790 vcpu->arch.hflags |= HF_SMM_MASK;
3792 vcpu->arch.hflags &= ~HF_SMM_MASK;
3793 kvm_smm_changed(vcpu);
3796 vcpu->arch.smi_pending = events->smi.pending;
3798 if (events->smi.smm) {
3799 if (events->smi.smm_inside_nmi)
3800 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3802 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3803 if (lapic_in_kernel(vcpu)) {
3804 if (events->smi.latched_init)
3805 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3807 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3812 kvm_make_request(KVM_REQ_EVENT, vcpu);
3817 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3818 struct kvm_debugregs *dbgregs)
3822 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3823 kvm_get_dr(vcpu, 6, &val);
3825 dbgregs->dr7 = vcpu->arch.dr7;
3827 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3830 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3831 struct kvm_debugregs *dbgregs)
3836 if (dbgregs->dr6 & ~0xffffffffull)
3838 if (dbgregs->dr7 & ~0xffffffffull)
3841 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3842 kvm_update_dr0123(vcpu);
3843 vcpu->arch.dr6 = dbgregs->dr6;
3844 kvm_update_dr6(vcpu);
3845 vcpu->arch.dr7 = dbgregs->dr7;
3846 kvm_update_dr7(vcpu);
3851 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3853 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3855 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3856 u64 xstate_bv = xsave->header.xfeatures;
3860 * Copy legacy XSAVE area, to avoid complications with CPUID
3861 * leaves 0 and 1 in the loop below.
3863 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3866 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3867 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3870 * Copy each region from the possibly compacted offset to the
3871 * non-compacted offset.
3873 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3875 u64 xfeature_mask = valid & -valid;
3876 int xfeature_nr = fls64(xfeature_mask) - 1;
3877 void *src = get_xsave_addr(xsave, xfeature_nr);
3880 u32 size, offset, ecx, edx;
3881 cpuid_count(XSTATE_CPUID, xfeature_nr,
3882 &size, &offset, &ecx, &edx);
3883 if (xfeature_nr == XFEATURE_PKRU)
3884 memcpy(dest + offset, &vcpu->arch.pkru,
3885 sizeof(vcpu->arch.pkru));
3887 memcpy(dest + offset, src, size);
3891 valid -= xfeature_mask;
3895 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3897 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3898 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3902 * Copy legacy XSAVE area, to avoid complications with CPUID
3903 * leaves 0 and 1 in the loop below.
3905 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3907 /* Set XSTATE_BV and possibly XCOMP_BV. */
3908 xsave->header.xfeatures = xstate_bv;
3909 if (boot_cpu_has(X86_FEATURE_XSAVES))
3910 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3913 * Copy each region from the non-compacted offset to the
3914 * possibly compacted offset.
3916 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3918 u64 xfeature_mask = valid & -valid;
3919 int xfeature_nr = fls64(xfeature_mask) - 1;
3920 void *dest = get_xsave_addr(xsave, xfeature_nr);
3923 u32 size, offset, ecx, edx;
3924 cpuid_count(XSTATE_CPUID, xfeature_nr,
3925 &size, &offset, &ecx, &edx);
3926 if (xfeature_nr == XFEATURE_PKRU)
3927 memcpy(&vcpu->arch.pkru, src + offset,
3928 sizeof(vcpu->arch.pkru));
3930 memcpy(dest, src + offset, size);
3933 valid -= xfeature_mask;
3937 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3938 struct kvm_xsave *guest_xsave)
3940 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3941 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3942 fill_xsave((u8 *) guest_xsave->region, vcpu);
3944 memcpy(guest_xsave->region,
3945 &vcpu->arch.guest_fpu->state.fxsave,
3946 sizeof(struct fxregs_state));
3947 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3948 XFEATURE_MASK_FPSSE;
3952 #define XSAVE_MXCSR_OFFSET 24
3954 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3955 struct kvm_xsave *guest_xsave)
3958 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3959 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3961 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3963 * Here we allow setting states that are not present in
3964 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3965 * with old userspace.
3967 if (xstate_bv & ~kvm_supported_xcr0() ||
3968 mxcsr & ~mxcsr_feature_mask)
3970 load_xsave(vcpu, (u8 *)guest_xsave->region);
3972 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3973 mxcsr & ~mxcsr_feature_mask)
3975 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3976 guest_xsave->region, sizeof(struct fxregs_state));
3981 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3982 struct kvm_xcrs *guest_xcrs)
3984 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3985 guest_xcrs->nr_xcrs = 0;
3989 guest_xcrs->nr_xcrs = 1;
3990 guest_xcrs->flags = 0;
3991 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3992 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3995 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3996 struct kvm_xcrs *guest_xcrs)
4000 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4003 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4006 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4007 /* Only support XCR0 currently */
4008 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4009 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4010 guest_xcrs->xcrs[i].value);
4019 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4020 * stopped by the hypervisor. This function will be called from the host only.
4021 * EINVAL is returned when the host attempts to set the flag for a guest that
4022 * does not support pv clocks.
4024 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4026 if (!vcpu->arch.pv_time_enabled)
4028 vcpu->arch.pvclock_set_guest_stopped_request = true;
4029 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4033 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4034 struct kvm_enable_cap *cap)
4037 uint16_t vmcs_version;
4038 void __user *user_ptr;
4044 case KVM_CAP_HYPERV_SYNIC2:
4049 case KVM_CAP_HYPERV_SYNIC:
4050 if (!irqchip_in_kernel(vcpu->kvm))
4052 return kvm_hv_activate_synic(vcpu, cap->cap ==
4053 KVM_CAP_HYPERV_SYNIC2);
4054 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4055 if (!kvm_x86_ops->nested_enable_evmcs)
4057 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4059 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4060 if (copy_to_user(user_ptr, &vmcs_version,
4061 sizeof(vmcs_version)))
4065 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4066 if (!kvm_x86_ops->enable_direct_tlbflush)
4069 return kvm_x86_ops->enable_direct_tlbflush(vcpu);
4076 long kvm_arch_vcpu_ioctl(struct file *filp,
4077 unsigned int ioctl, unsigned long arg)
4079 struct kvm_vcpu *vcpu = filp->private_data;
4080 void __user *argp = (void __user *)arg;
4083 struct kvm_lapic_state *lapic;
4084 struct kvm_xsave *xsave;
4085 struct kvm_xcrs *xcrs;
4093 case KVM_GET_LAPIC: {
4095 if (!lapic_in_kernel(vcpu))
4097 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4098 GFP_KERNEL_ACCOUNT);
4103 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4107 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4112 case KVM_SET_LAPIC: {
4114 if (!lapic_in_kernel(vcpu))
4116 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4117 if (IS_ERR(u.lapic)) {
4118 r = PTR_ERR(u.lapic);
4122 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4125 case KVM_INTERRUPT: {
4126 struct kvm_interrupt irq;
4129 if (copy_from_user(&irq, argp, sizeof(irq)))
4131 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4135 r = kvm_vcpu_ioctl_nmi(vcpu);
4139 r = kvm_vcpu_ioctl_smi(vcpu);
4142 case KVM_SET_CPUID: {
4143 struct kvm_cpuid __user *cpuid_arg = argp;
4144 struct kvm_cpuid cpuid;
4147 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4149 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4152 case KVM_SET_CPUID2: {
4153 struct kvm_cpuid2 __user *cpuid_arg = argp;
4154 struct kvm_cpuid2 cpuid;
4157 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4159 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4160 cpuid_arg->entries);
4163 case KVM_GET_CPUID2: {
4164 struct kvm_cpuid2 __user *cpuid_arg = argp;
4165 struct kvm_cpuid2 cpuid;
4168 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4170 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4171 cpuid_arg->entries);
4175 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4180 case KVM_GET_MSRS: {
4181 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4182 r = msr_io(vcpu, argp, do_get_msr, 1);
4183 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4186 case KVM_SET_MSRS: {
4187 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4188 r = msr_io(vcpu, argp, do_set_msr, 0);
4189 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4192 case KVM_TPR_ACCESS_REPORTING: {
4193 struct kvm_tpr_access_ctl tac;
4196 if (copy_from_user(&tac, argp, sizeof(tac)))
4198 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4202 if (copy_to_user(argp, &tac, sizeof(tac)))
4207 case KVM_SET_VAPIC_ADDR: {
4208 struct kvm_vapic_addr va;
4212 if (!lapic_in_kernel(vcpu))
4215 if (copy_from_user(&va, argp, sizeof(va)))
4217 idx = srcu_read_lock(&vcpu->kvm->srcu);
4218 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4219 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4222 case KVM_X86_SETUP_MCE: {
4226 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4228 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4231 case KVM_X86_SET_MCE: {
4232 struct kvm_x86_mce mce;
4235 if (copy_from_user(&mce, argp, sizeof(mce)))
4237 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4240 case KVM_GET_VCPU_EVENTS: {
4241 struct kvm_vcpu_events events;
4243 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4246 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4251 case KVM_SET_VCPU_EVENTS: {
4252 struct kvm_vcpu_events events;
4255 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4258 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4261 case KVM_GET_DEBUGREGS: {
4262 struct kvm_debugregs dbgregs;
4264 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4267 if (copy_to_user(argp, &dbgregs,
4268 sizeof(struct kvm_debugregs)))
4273 case KVM_SET_DEBUGREGS: {
4274 struct kvm_debugregs dbgregs;
4277 if (copy_from_user(&dbgregs, argp,
4278 sizeof(struct kvm_debugregs)))
4281 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4284 case KVM_GET_XSAVE: {
4285 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4290 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4293 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4298 case KVM_SET_XSAVE: {
4299 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4300 if (IS_ERR(u.xsave)) {
4301 r = PTR_ERR(u.xsave);
4305 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4308 case KVM_GET_XCRS: {
4309 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4314 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4317 if (copy_to_user(argp, u.xcrs,
4318 sizeof(struct kvm_xcrs)))
4323 case KVM_SET_XCRS: {
4324 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4325 if (IS_ERR(u.xcrs)) {
4326 r = PTR_ERR(u.xcrs);
4330 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4333 case KVM_SET_TSC_KHZ: {
4337 user_tsc_khz = (u32)arg;
4339 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4342 if (user_tsc_khz == 0)
4343 user_tsc_khz = tsc_khz;
4345 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4350 case KVM_GET_TSC_KHZ: {
4351 r = vcpu->arch.virtual_tsc_khz;
4354 case KVM_KVMCLOCK_CTRL: {
4355 r = kvm_set_guest_paused(vcpu);
4358 case KVM_ENABLE_CAP: {
4359 struct kvm_enable_cap cap;
4362 if (copy_from_user(&cap, argp, sizeof(cap)))
4364 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4367 case KVM_GET_NESTED_STATE: {
4368 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4372 if (!kvm_x86_ops->get_nested_state)
4375 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4377 if (get_user(user_data_size, &user_kvm_nested_state->size))
4380 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4385 if (r > user_data_size) {
4386 if (put_user(r, &user_kvm_nested_state->size))
4396 case KVM_SET_NESTED_STATE: {
4397 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4398 struct kvm_nested_state kvm_state;
4401 if (!kvm_x86_ops->set_nested_state)
4405 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4409 if (kvm_state.size < sizeof(kvm_state))
4412 if (kvm_state.flags &
4413 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4414 | KVM_STATE_NESTED_EVMCS))
4417 /* nested_run_pending implies guest_mode. */
4418 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4419 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4422 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4425 case KVM_GET_SUPPORTED_HV_CPUID: {
4426 struct kvm_cpuid2 __user *cpuid_arg = argp;
4427 struct kvm_cpuid2 cpuid;
4430 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4433 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4434 cpuid_arg->entries);
4439 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4454 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4456 return VM_FAULT_SIGBUS;
4459 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4463 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4465 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4469 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4472 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4475 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4476 unsigned long kvm_nr_mmu_pages)
4478 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4481 mutex_lock(&kvm->slots_lock);
4483 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4484 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4486 mutex_unlock(&kvm->slots_lock);
4490 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4492 return kvm->arch.n_max_mmu_pages;
4495 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4497 struct kvm_pic *pic = kvm->arch.vpic;
4501 switch (chip->chip_id) {
4502 case KVM_IRQCHIP_PIC_MASTER:
4503 memcpy(&chip->chip.pic, &pic->pics[0],
4504 sizeof(struct kvm_pic_state));
4506 case KVM_IRQCHIP_PIC_SLAVE:
4507 memcpy(&chip->chip.pic, &pic->pics[1],
4508 sizeof(struct kvm_pic_state));
4510 case KVM_IRQCHIP_IOAPIC:
4511 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4520 static int kvm_vm_ioctl_set_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 spin_lock(&pic->lock);
4529 memcpy(&pic->pics[0], &chip->chip.pic,
4530 sizeof(struct kvm_pic_state));
4531 spin_unlock(&pic->lock);
4533 case KVM_IRQCHIP_PIC_SLAVE:
4534 spin_lock(&pic->lock);
4535 memcpy(&pic->pics[1], &chip->chip.pic,
4536 sizeof(struct kvm_pic_state));
4537 spin_unlock(&pic->lock);
4539 case KVM_IRQCHIP_IOAPIC:
4540 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4546 kvm_pic_update_irq(pic);
4550 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4552 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4554 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4556 mutex_lock(&kps->lock);
4557 memcpy(ps, &kps->channels, sizeof(*ps));
4558 mutex_unlock(&kps->lock);
4562 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4565 struct kvm_pit *pit = kvm->arch.vpit;
4567 mutex_lock(&pit->pit_state.lock);
4568 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4569 for (i = 0; i < 3; i++)
4570 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4571 mutex_unlock(&pit->pit_state.lock);
4575 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4577 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4578 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4579 sizeof(ps->channels));
4580 ps->flags = kvm->arch.vpit->pit_state.flags;
4581 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4582 memset(&ps->reserved, 0, sizeof(ps->reserved));
4586 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4590 u32 prev_legacy, cur_legacy;
4591 struct kvm_pit *pit = kvm->arch.vpit;
4593 mutex_lock(&pit->pit_state.lock);
4594 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4595 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4596 if (!prev_legacy && cur_legacy)
4598 memcpy(&pit->pit_state.channels, &ps->channels,
4599 sizeof(pit->pit_state.channels));
4600 pit->pit_state.flags = ps->flags;
4601 for (i = 0; i < 3; i++)
4602 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4604 mutex_unlock(&pit->pit_state.lock);
4608 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4609 struct kvm_reinject_control *control)
4611 struct kvm_pit *pit = kvm->arch.vpit;
4616 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4617 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4618 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4620 mutex_lock(&pit->pit_state.lock);
4621 kvm_pit_set_reinject(pit, control->pit_reinject);
4622 mutex_unlock(&pit->pit_state.lock);
4628 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4629 * @kvm: kvm instance
4630 * @log: slot id and address to which we copy the log
4632 * Steps 1-4 below provide general overview of dirty page logging. See
4633 * kvm_get_dirty_log_protect() function description for additional details.
4635 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4636 * always flush the TLB (step 4) even if previous step failed and the dirty
4637 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4638 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4639 * writes will be marked dirty for next log read.
4641 * 1. Take a snapshot of the bit and clear it if needed.
4642 * 2. Write protect the corresponding page.
4643 * 3. Copy the snapshot to the userspace.
4644 * 4. Flush TLB's if needed.
4646 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4651 mutex_lock(&kvm->slots_lock);
4654 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4656 if (kvm_x86_ops->flush_log_dirty)
4657 kvm_x86_ops->flush_log_dirty(kvm);
4659 r = kvm_get_dirty_log_protect(kvm, log, &flush);
4662 * All the TLBs can be flushed out of mmu lock, see the comments in
4663 * kvm_mmu_slot_remove_write_access().
4665 lockdep_assert_held(&kvm->slots_lock);
4667 kvm_flush_remote_tlbs(kvm);
4669 mutex_unlock(&kvm->slots_lock);
4673 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4678 mutex_lock(&kvm->slots_lock);
4681 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4683 if (kvm_x86_ops->flush_log_dirty)
4684 kvm_x86_ops->flush_log_dirty(kvm);
4686 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4689 * All the TLBs can be flushed out of mmu lock, see the comments in
4690 * kvm_mmu_slot_remove_write_access().
4692 lockdep_assert_held(&kvm->slots_lock);
4694 kvm_flush_remote_tlbs(kvm);
4696 mutex_unlock(&kvm->slots_lock);
4700 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4703 if (!irqchip_in_kernel(kvm))
4706 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4707 irq_event->irq, irq_event->level,
4712 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4713 struct kvm_enable_cap *cap)
4721 case KVM_CAP_DISABLE_QUIRKS:
4722 kvm->arch.disabled_quirks = cap->args[0];
4725 case KVM_CAP_SPLIT_IRQCHIP: {
4726 mutex_lock(&kvm->lock);
4728 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4729 goto split_irqchip_unlock;
4731 if (irqchip_in_kernel(kvm))
4732 goto split_irqchip_unlock;
4733 if (kvm->created_vcpus)
4734 goto split_irqchip_unlock;
4735 r = kvm_setup_empty_irq_routing(kvm);
4737 goto split_irqchip_unlock;
4738 /* Pairs with irqchip_in_kernel. */
4740 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4741 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4743 split_irqchip_unlock:
4744 mutex_unlock(&kvm->lock);
4747 case KVM_CAP_X2APIC_API:
4749 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4752 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4753 kvm->arch.x2apic_format = true;
4754 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4755 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4759 case KVM_CAP_X86_DISABLE_EXITS:
4761 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4764 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4765 kvm_can_mwait_in_guest())
4766 kvm->arch.mwait_in_guest = true;
4767 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4768 kvm->arch.hlt_in_guest = true;
4769 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4770 kvm->arch.pause_in_guest = true;
4771 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4772 kvm->arch.cstate_in_guest = true;
4775 case KVM_CAP_MSR_PLATFORM_INFO:
4776 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4779 case KVM_CAP_EXCEPTION_PAYLOAD:
4780 kvm->arch.exception_payload_enabled = cap->args[0];
4790 long kvm_arch_vm_ioctl(struct file *filp,
4791 unsigned int ioctl, unsigned long arg)
4793 struct kvm *kvm = filp->private_data;
4794 void __user *argp = (void __user *)arg;
4797 * This union makes it completely explicit to gcc-3.x
4798 * that these two variables' stack usage should be
4799 * combined, not added together.
4802 struct kvm_pit_state ps;
4803 struct kvm_pit_state2 ps2;
4804 struct kvm_pit_config pit_config;
4808 case KVM_SET_TSS_ADDR:
4809 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4811 case KVM_SET_IDENTITY_MAP_ADDR: {
4814 mutex_lock(&kvm->lock);
4816 if (kvm->created_vcpus)
4817 goto set_identity_unlock;
4819 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4820 goto set_identity_unlock;
4821 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4822 set_identity_unlock:
4823 mutex_unlock(&kvm->lock);
4826 case KVM_SET_NR_MMU_PAGES:
4827 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4829 case KVM_GET_NR_MMU_PAGES:
4830 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4832 case KVM_CREATE_IRQCHIP: {
4833 mutex_lock(&kvm->lock);
4836 if (irqchip_in_kernel(kvm))
4837 goto create_irqchip_unlock;
4840 if (kvm->created_vcpus)
4841 goto create_irqchip_unlock;
4843 r = kvm_pic_init(kvm);
4845 goto create_irqchip_unlock;
4847 r = kvm_ioapic_init(kvm);
4849 kvm_pic_destroy(kvm);
4850 goto create_irqchip_unlock;
4853 r = kvm_setup_default_irq_routing(kvm);
4855 kvm_ioapic_destroy(kvm);
4856 kvm_pic_destroy(kvm);
4857 goto create_irqchip_unlock;
4859 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4861 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4862 create_irqchip_unlock:
4863 mutex_unlock(&kvm->lock);
4866 case KVM_CREATE_PIT:
4867 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4869 case KVM_CREATE_PIT2:
4871 if (copy_from_user(&u.pit_config, argp,
4872 sizeof(struct kvm_pit_config)))
4875 mutex_lock(&kvm->lock);
4878 goto create_pit_unlock;
4880 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4884 mutex_unlock(&kvm->lock);
4886 case KVM_GET_IRQCHIP: {
4887 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4888 struct kvm_irqchip *chip;
4890 chip = memdup_user(argp, sizeof(*chip));
4897 if (!irqchip_kernel(kvm))
4898 goto get_irqchip_out;
4899 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4901 goto get_irqchip_out;
4903 if (copy_to_user(argp, chip, sizeof(*chip)))
4904 goto get_irqchip_out;
4910 case KVM_SET_IRQCHIP: {
4911 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4912 struct kvm_irqchip *chip;
4914 chip = memdup_user(argp, sizeof(*chip));
4921 if (!irqchip_kernel(kvm))
4922 goto set_irqchip_out;
4923 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4925 goto set_irqchip_out;
4933 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4936 if (!kvm->arch.vpit)
4938 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4942 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4949 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4952 if (!kvm->arch.vpit)
4954 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4957 case KVM_GET_PIT2: {
4959 if (!kvm->arch.vpit)
4961 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4965 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4970 case KVM_SET_PIT2: {
4972 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4975 if (!kvm->arch.vpit)
4977 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4980 case KVM_REINJECT_CONTROL: {
4981 struct kvm_reinject_control control;
4983 if (copy_from_user(&control, argp, sizeof(control)))
4985 r = kvm_vm_ioctl_reinject(kvm, &control);
4988 case KVM_SET_BOOT_CPU_ID:
4990 mutex_lock(&kvm->lock);
4991 if (kvm->created_vcpus)
4994 kvm->arch.bsp_vcpu_id = arg;
4995 mutex_unlock(&kvm->lock);
4997 case KVM_XEN_HVM_CONFIG: {
4998 struct kvm_xen_hvm_config xhc;
5000 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5005 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5009 case KVM_SET_CLOCK: {
5010 struct kvm_clock_data user_ns;
5014 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5023 * TODO: userspace has to take care of races with VCPU_RUN, so
5024 * kvm_gen_update_masterclock() can be cut down to locked
5025 * pvclock_update_vm_gtod_copy().
5027 kvm_gen_update_masterclock(kvm);
5028 now_ns = get_kvmclock_ns(kvm);
5029 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5030 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5033 case KVM_GET_CLOCK: {
5034 struct kvm_clock_data user_ns;
5037 now_ns = get_kvmclock_ns(kvm);
5038 user_ns.clock = now_ns;
5039 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5040 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5043 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5048 case KVM_MEMORY_ENCRYPT_OP: {
5050 if (kvm_x86_ops->mem_enc_op)
5051 r = kvm_x86_ops->mem_enc_op(kvm, argp);
5054 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5055 struct kvm_enc_region region;
5058 if (copy_from_user(®ion, argp, sizeof(region)))
5062 if (kvm_x86_ops->mem_enc_reg_region)
5063 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
5066 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5067 struct kvm_enc_region region;
5070 if (copy_from_user(®ion, argp, sizeof(region)))
5074 if (kvm_x86_ops->mem_enc_unreg_region)
5075 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
5078 case KVM_HYPERV_EVENTFD: {
5079 struct kvm_hyperv_eventfd hvevfd;
5082 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5084 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5087 case KVM_SET_PMU_EVENT_FILTER:
5088 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5097 static void kvm_init_msr_list(void)
5102 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5103 "Please update the fixed PMCs in msrs_to_save[]");
5104 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_GENERIC != 32,
5105 "Please update the generic perfctr/eventsel MSRs in msrs_to_save[]");
5107 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
5108 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
5112 * Even MSRs that are valid in the host may not be exposed
5113 * to the guests in some cases.
5115 switch (msrs_to_save[i]) {
5116 case MSR_IA32_BNDCFGS:
5117 if (!kvm_mpx_supported())
5121 if (!kvm_x86_ops->rdtscp_supported())
5124 case MSR_IA32_RTIT_CTL:
5125 case MSR_IA32_RTIT_STATUS:
5126 if (!kvm_x86_ops->pt_supported())
5129 case MSR_IA32_RTIT_CR3_MATCH:
5130 if (!kvm_x86_ops->pt_supported() ||
5131 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5134 case MSR_IA32_RTIT_OUTPUT_BASE:
5135 case MSR_IA32_RTIT_OUTPUT_MASK:
5136 if (!kvm_x86_ops->pt_supported() ||
5137 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5138 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5141 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5142 if (!kvm_x86_ops->pt_supported() ||
5143 msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
5144 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5153 msrs_to_save[j] = msrs_to_save[i];
5156 num_msrs_to_save = j;
5158 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
5159 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
5163 emulated_msrs[j] = emulated_msrs[i];
5166 num_emulated_msrs = j;
5168 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
5169 struct kvm_msr_entry msr;
5171 msr.index = msr_based_features[i];
5172 if (kvm_get_msr_feature(&msr))
5176 msr_based_features[j] = msr_based_features[i];
5179 num_msr_based_features = j;
5182 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5190 if (!(lapic_in_kernel(vcpu) &&
5191 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5192 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5203 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5210 if (!(lapic_in_kernel(vcpu) &&
5211 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5213 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5215 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5225 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5226 struct kvm_segment *var, int seg)
5228 kvm_x86_ops->set_segment(vcpu, var, seg);
5231 void kvm_get_segment(struct kvm_vcpu *vcpu,
5232 struct kvm_segment *var, int seg)
5234 kvm_x86_ops->get_segment(vcpu, var, seg);
5237 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5238 struct x86_exception *exception)
5242 BUG_ON(!mmu_is_nested(vcpu));
5244 /* NPT walks are always user-walks */
5245 access |= PFERR_USER_MASK;
5246 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5251 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5252 struct x86_exception *exception)
5254 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5255 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5258 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5259 struct x86_exception *exception)
5261 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5262 access |= PFERR_FETCH_MASK;
5263 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5266 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5267 struct x86_exception *exception)
5269 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5270 access |= PFERR_WRITE_MASK;
5271 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5274 /* uses this to access any guest's mapped memory without checking CPL */
5275 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5276 struct x86_exception *exception)
5278 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5281 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5282 struct kvm_vcpu *vcpu, u32 access,
5283 struct x86_exception *exception)
5286 int r = X86EMUL_CONTINUE;
5289 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5291 unsigned offset = addr & (PAGE_SIZE-1);
5292 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5295 if (gpa == UNMAPPED_GVA)
5296 return X86EMUL_PROPAGATE_FAULT;
5297 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5300 r = X86EMUL_IO_NEEDED;
5312 /* used for instruction fetching */
5313 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5314 gva_t addr, void *val, unsigned int bytes,
5315 struct x86_exception *exception)
5317 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5318 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5322 /* Inline kvm_read_guest_virt_helper for speed. */
5323 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5325 if (unlikely(gpa == UNMAPPED_GVA))
5326 return X86EMUL_PROPAGATE_FAULT;
5328 offset = addr & (PAGE_SIZE-1);
5329 if (WARN_ON(offset + bytes > PAGE_SIZE))
5330 bytes = (unsigned)PAGE_SIZE - offset;
5331 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5333 if (unlikely(ret < 0))
5334 return X86EMUL_IO_NEEDED;
5336 return X86EMUL_CONTINUE;
5339 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5340 gva_t addr, void *val, unsigned int bytes,
5341 struct x86_exception *exception)
5343 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5346 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5347 * is returned, but our callers are not ready for that and they blindly
5348 * call kvm_inject_page_fault. Ensure that they at least do not leak
5349 * uninitialized kernel stack memory into cr2 and error code.
5351 memset(exception, 0, sizeof(*exception));
5352 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5355 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5357 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5358 gva_t addr, void *val, unsigned int bytes,
5359 struct x86_exception *exception, bool system)
5361 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5364 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5365 access |= PFERR_USER_MASK;
5367 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5370 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5371 unsigned long addr, void *val, unsigned int bytes)
5373 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5374 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5376 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5379 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5380 struct kvm_vcpu *vcpu, u32 access,
5381 struct x86_exception *exception)
5384 int r = X86EMUL_CONTINUE;
5387 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5390 unsigned offset = addr & (PAGE_SIZE-1);
5391 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5394 if (gpa == UNMAPPED_GVA)
5395 return X86EMUL_PROPAGATE_FAULT;
5396 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5398 r = X86EMUL_IO_NEEDED;
5410 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5411 unsigned int bytes, struct x86_exception *exception,
5414 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5415 u32 access = PFERR_WRITE_MASK;
5417 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5418 access |= PFERR_USER_MASK;
5420 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5424 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5425 unsigned int bytes, struct x86_exception *exception)
5427 /* kvm_write_guest_virt_system can pull in tons of pages. */
5428 vcpu->arch.l1tf_flush_l1d = true;
5431 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5432 * is returned, but our callers are not ready for that and they blindly
5433 * call kvm_inject_page_fault. Ensure that they at least do not leak
5434 * uninitialized kernel stack memory into cr2 and error code.
5436 memset(exception, 0, sizeof(*exception));
5437 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5438 PFERR_WRITE_MASK, exception);
5440 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5442 int handle_ud(struct kvm_vcpu *vcpu)
5444 int emul_type = EMULTYPE_TRAP_UD;
5445 char sig[5]; /* ud2; .ascii "kvm" */
5446 struct x86_exception e;
5448 if (force_emulation_prefix &&
5449 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5450 sig, sizeof(sig), &e) == 0 &&
5451 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5452 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5453 emul_type = EMULTYPE_TRAP_UD_FORCED;
5456 return kvm_emulate_instruction(vcpu, emul_type);
5458 EXPORT_SYMBOL_GPL(handle_ud);
5460 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5461 gpa_t gpa, bool write)
5463 /* For APIC access vmexit */
5464 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5467 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5468 trace_vcpu_match_mmio(gva, gpa, write, true);
5475 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5476 gpa_t *gpa, struct x86_exception *exception,
5479 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5480 | (write ? PFERR_WRITE_MASK : 0);
5483 * currently PKRU is only applied to ept enabled guest so
5484 * there is no pkey in EPT page table for L1 guest or EPT
5485 * shadow page table for L2 guest.
5487 if (vcpu_match_mmio_gva(vcpu, gva)
5488 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5489 vcpu->arch.mmio_access, 0, access)) {
5490 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5491 (gva & (PAGE_SIZE - 1));
5492 trace_vcpu_match_mmio(gva, *gpa, write, false);
5496 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5498 if (*gpa == UNMAPPED_GVA)
5501 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5504 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5505 const void *val, int bytes)
5509 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5512 kvm_page_track_write(vcpu, gpa, val, bytes);
5516 struct read_write_emulator_ops {
5517 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5519 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5520 void *val, int bytes);
5521 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5522 int bytes, void *val);
5523 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5524 void *val, int bytes);
5528 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5530 if (vcpu->mmio_read_completed) {
5531 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5532 vcpu->mmio_fragments[0].gpa, val);
5533 vcpu->mmio_read_completed = 0;
5540 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5541 void *val, int bytes)
5543 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5546 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5547 void *val, int bytes)
5549 return emulator_write_phys(vcpu, gpa, val, bytes);
5552 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5554 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5555 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5558 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5559 void *val, int bytes)
5561 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5562 return X86EMUL_IO_NEEDED;
5565 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5566 void *val, int bytes)
5568 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5570 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5571 return X86EMUL_CONTINUE;
5574 static const struct read_write_emulator_ops read_emultor = {
5575 .read_write_prepare = read_prepare,
5576 .read_write_emulate = read_emulate,
5577 .read_write_mmio = vcpu_mmio_read,
5578 .read_write_exit_mmio = read_exit_mmio,
5581 static const struct read_write_emulator_ops write_emultor = {
5582 .read_write_emulate = write_emulate,
5583 .read_write_mmio = write_mmio,
5584 .read_write_exit_mmio = write_exit_mmio,
5588 static int emulator_read_write_onepage(unsigned long addr, void *val,
5590 struct x86_exception *exception,
5591 struct kvm_vcpu *vcpu,
5592 const struct read_write_emulator_ops *ops)
5596 bool write = ops->write;
5597 struct kvm_mmio_fragment *frag;
5598 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5601 * If the exit was due to a NPF we may already have a GPA.
5602 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5603 * Note, this cannot be used on string operations since string
5604 * operation using rep will only have the initial GPA from the NPF
5607 if (vcpu->arch.gpa_available &&
5608 emulator_can_use_gpa(ctxt) &&
5609 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5610 gpa = vcpu->arch.gpa_val;
5611 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5613 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5615 return X86EMUL_PROPAGATE_FAULT;
5618 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5619 return X86EMUL_CONTINUE;
5622 * Is this MMIO handled locally?
5624 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5625 if (handled == bytes)
5626 return X86EMUL_CONTINUE;
5632 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5633 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5637 return X86EMUL_CONTINUE;
5640 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5642 void *val, unsigned int bytes,
5643 struct x86_exception *exception,
5644 const struct read_write_emulator_ops *ops)
5646 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5650 if (ops->read_write_prepare &&
5651 ops->read_write_prepare(vcpu, val, bytes))
5652 return X86EMUL_CONTINUE;
5654 vcpu->mmio_nr_fragments = 0;
5656 /* Crossing a page boundary? */
5657 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5660 now = -addr & ~PAGE_MASK;
5661 rc = emulator_read_write_onepage(addr, val, now, exception,
5664 if (rc != X86EMUL_CONTINUE)
5667 if (ctxt->mode != X86EMUL_MODE_PROT64)
5673 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5675 if (rc != X86EMUL_CONTINUE)
5678 if (!vcpu->mmio_nr_fragments)
5681 gpa = vcpu->mmio_fragments[0].gpa;
5683 vcpu->mmio_needed = 1;
5684 vcpu->mmio_cur_fragment = 0;
5686 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5687 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5688 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5689 vcpu->run->mmio.phys_addr = gpa;
5691 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5694 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5698 struct x86_exception *exception)
5700 return emulator_read_write(ctxt, addr, val, bytes,
5701 exception, &read_emultor);
5704 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5708 struct x86_exception *exception)
5710 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5711 exception, &write_emultor);
5714 #define CMPXCHG_TYPE(t, ptr, old, new) \
5715 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5717 #ifdef CONFIG_X86_64
5718 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5720 # define CMPXCHG64(ptr, old, new) \
5721 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5724 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5729 struct x86_exception *exception)
5731 struct kvm_host_map map;
5732 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5737 /* guests cmpxchg8b have to be emulated atomically */
5738 if (bytes > 8 || (bytes & (bytes - 1)))
5741 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5743 if (gpa == UNMAPPED_GVA ||
5744 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5747 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5750 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5753 kaddr = map.hva + offset_in_page(gpa);
5757 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5760 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5763 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5766 exchanged = CMPXCHG64(kaddr, old, new);
5772 kvm_vcpu_unmap(vcpu, &map, true);
5775 return X86EMUL_CMPXCHG_FAILED;
5777 kvm_page_track_write(vcpu, gpa, new, bytes);
5779 return X86EMUL_CONTINUE;
5782 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5784 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5787 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5791 for (i = 0; i < vcpu->arch.pio.count; i++) {
5792 if (vcpu->arch.pio.in)
5793 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5794 vcpu->arch.pio.size, pd);
5796 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5797 vcpu->arch.pio.port, vcpu->arch.pio.size,
5801 pd += vcpu->arch.pio.size;
5806 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5807 unsigned short port, void *val,
5808 unsigned int count, bool in)
5810 vcpu->arch.pio.port = port;
5811 vcpu->arch.pio.in = in;
5812 vcpu->arch.pio.count = count;
5813 vcpu->arch.pio.size = size;
5815 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5816 vcpu->arch.pio.count = 0;
5820 vcpu->run->exit_reason = KVM_EXIT_IO;
5821 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5822 vcpu->run->io.size = size;
5823 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5824 vcpu->run->io.count = count;
5825 vcpu->run->io.port = port;
5830 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5831 int size, unsigned short port, void *val,
5834 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5837 if (vcpu->arch.pio.count)
5840 memset(vcpu->arch.pio_data, 0, size * count);
5842 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5845 memcpy(val, vcpu->arch.pio_data, size * count);
5846 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5847 vcpu->arch.pio.count = 0;
5854 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5855 int size, unsigned short port,
5856 const void *val, unsigned int count)
5858 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5860 memcpy(vcpu->arch.pio_data, val, size * count);
5861 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5862 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5865 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5867 return kvm_x86_ops->get_segment_base(vcpu, seg);
5870 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5872 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5875 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5877 if (!need_emulate_wbinvd(vcpu))
5878 return X86EMUL_CONTINUE;
5880 if (kvm_x86_ops->has_wbinvd_exit()) {
5881 int cpu = get_cpu();
5883 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5884 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5885 wbinvd_ipi, NULL, 1);
5887 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5890 return X86EMUL_CONTINUE;
5893 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5895 kvm_emulate_wbinvd_noskip(vcpu);
5896 return kvm_skip_emulated_instruction(vcpu);
5898 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5902 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5904 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5907 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5908 unsigned long *dest)
5910 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5913 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5914 unsigned long value)
5917 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5920 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5922 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5925 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5927 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5928 unsigned long value;
5932 value = kvm_read_cr0(vcpu);
5935 value = vcpu->arch.cr2;
5938 value = kvm_read_cr3(vcpu);
5941 value = kvm_read_cr4(vcpu);
5944 value = kvm_get_cr8(vcpu);
5947 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5954 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5956 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5961 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5964 vcpu->arch.cr2 = val;
5967 res = kvm_set_cr3(vcpu, val);
5970 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5973 res = kvm_set_cr8(vcpu, val);
5976 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5983 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5985 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5988 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5990 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5993 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5995 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5998 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6000 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
6003 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6005 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
6008 static unsigned long emulator_get_cached_segment_base(
6009 struct x86_emulate_ctxt *ctxt, int seg)
6011 return get_segment_base(emul_to_vcpu(ctxt), seg);
6014 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6015 struct desc_struct *desc, u32 *base3,
6018 struct kvm_segment var;
6020 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6021 *selector = var.selector;
6024 memset(desc, 0, sizeof(*desc));
6032 set_desc_limit(desc, var.limit);
6033 set_desc_base(desc, (unsigned long)var.base);
6034 #ifdef CONFIG_X86_64
6036 *base3 = var.base >> 32;
6038 desc->type = var.type;
6040 desc->dpl = var.dpl;
6041 desc->p = var.present;
6042 desc->avl = var.avl;
6050 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6051 struct desc_struct *desc, u32 base3,
6054 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6055 struct kvm_segment var;
6057 var.selector = selector;
6058 var.base = get_desc_base(desc);
6059 #ifdef CONFIG_X86_64
6060 var.base |= ((u64)base3) << 32;
6062 var.limit = get_desc_limit(desc);
6064 var.limit = (var.limit << 12) | 0xfff;
6065 var.type = desc->type;
6066 var.dpl = desc->dpl;
6071 var.avl = desc->avl;
6072 var.present = desc->p;
6073 var.unusable = !var.present;
6076 kvm_set_segment(vcpu, &var, seg);
6080 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6081 u32 msr_index, u64 *pdata)
6083 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6086 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6087 u32 msr_index, u64 data)
6089 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6092 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6094 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6096 return vcpu->arch.smbase;
6099 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6101 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6103 vcpu->arch.smbase = smbase;
6106 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6109 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
6112 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6113 u32 pmc, u64 *pdata)
6115 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6118 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6120 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6123 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6124 struct x86_instruction_info *info,
6125 enum x86_intercept_stage stage)
6127 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6130 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6131 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6133 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6136 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6138 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6141 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6143 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6146 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6148 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6151 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6153 return emul_to_vcpu(ctxt)->arch.hflags;
6156 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6158 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6161 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6162 const char *smstate)
6164 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6167 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6169 kvm_smm_changed(emul_to_vcpu(ctxt));
6172 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6174 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6177 static const struct x86_emulate_ops emulate_ops = {
6178 .read_gpr = emulator_read_gpr,
6179 .write_gpr = emulator_write_gpr,
6180 .read_std = emulator_read_std,
6181 .write_std = emulator_write_std,
6182 .read_phys = kvm_read_guest_phys_system,
6183 .fetch = kvm_fetch_guest_virt,
6184 .read_emulated = emulator_read_emulated,
6185 .write_emulated = emulator_write_emulated,
6186 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6187 .invlpg = emulator_invlpg,
6188 .pio_in_emulated = emulator_pio_in_emulated,
6189 .pio_out_emulated = emulator_pio_out_emulated,
6190 .get_segment = emulator_get_segment,
6191 .set_segment = emulator_set_segment,
6192 .get_cached_segment_base = emulator_get_cached_segment_base,
6193 .get_gdt = emulator_get_gdt,
6194 .get_idt = emulator_get_idt,
6195 .set_gdt = emulator_set_gdt,
6196 .set_idt = emulator_set_idt,
6197 .get_cr = emulator_get_cr,
6198 .set_cr = emulator_set_cr,
6199 .cpl = emulator_get_cpl,
6200 .get_dr = emulator_get_dr,
6201 .set_dr = emulator_set_dr,
6202 .get_smbase = emulator_get_smbase,
6203 .set_smbase = emulator_set_smbase,
6204 .set_msr = emulator_set_msr,
6205 .get_msr = emulator_get_msr,
6206 .check_pmc = emulator_check_pmc,
6207 .read_pmc = emulator_read_pmc,
6208 .halt = emulator_halt,
6209 .wbinvd = emulator_wbinvd,
6210 .fix_hypercall = emulator_fix_hypercall,
6211 .intercept = emulator_intercept,
6212 .get_cpuid = emulator_get_cpuid,
6213 .set_nmi_mask = emulator_set_nmi_mask,
6214 .get_hflags = emulator_get_hflags,
6215 .set_hflags = emulator_set_hflags,
6216 .pre_leave_smm = emulator_pre_leave_smm,
6217 .post_leave_smm = emulator_post_leave_smm,
6218 .set_xcr = emulator_set_xcr,
6221 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6223 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6225 * an sti; sti; sequence only disable interrupts for the first
6226 * instruction. So, if the last instruction, be it emulated or
6227 * not, left the system with the INT_STI flag enabled, it
6228 * means that the last instruction is an sti. We should not
6229 * leave the flag on in this case. The same goes for mov ss
6231 if (int_shadow & mask)
6233 if (unlikely(int_shadow || mask)) {
6234 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6236 kvm_make_request(KVM_REQ_EVENT, vcpu);
6240 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6242 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6243 if (ctxt->exception.vector == PF_VECTOR)
6244 return kvm_propagate_fault(vcpu, &ctxt->exception);
6246 if (ctxt->exception.error_code_valid)
6247 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6248 ctxt->exception.error_code);
6250 kvm_queue_exception(vcpu, ctxt->exception.vector);
6254 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6256 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6259 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6261 ctxt->eflags = kvm_get_rflags(vcpu);
6262 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6264 ctxt->eip = kvm_rip_read(vcpu);
6265 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6266 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6267 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6268 cs_db ? X86EMUL_MODE_PROT32 :
6269 X86EMUL_MODE_PROT16;
6270 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6271 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6272 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6274 init_decode_cache(ctxt);
6275 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6278 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6280 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6283 init_emulate_ctxt(vcpu);
6287 ctxt->_eip = ctxt->eip + inc_eip;
6288 ret = emulate_int_real(ctxt, irq);
6290 if (ret != X86EMUL_CONTINUE) {
6291 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6293 ctxt->eip = ctxt->_eip;
6294 kvm_rip_write(vcpu, ctxt->eip);
6295 kvm_set_rflags(vcpu, ctxt->eflags);
6298 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6300 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6302 ++vcpu->stat.insn_emulation_fail;
6303 trace_kvm_emulate_insn_failed(vcpu);
6305 if (emulation_type & EMULTYPE_VMWARE_GP) {
6306 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6310 if (emulation_type & EMULTYPE_SKIP) {
6311 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6312 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6313 vcpu->run->internal.ndata = 0;
6317 kvm_queue_exception(vcpu, UD_VECTOR);
6319 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6320 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6321 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6322 vcpu->run->internal.ndata = 0;
6329 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6330 bool write_fault_to_shadow_pgtable,
6336 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6339 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6342 if (!vcpu->arch.mmu->direct_map) {
6344 * Write permission should be allowed since only
6345 * write access need to be emulated.
6347 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6350 * If the mapping is invalid in guest, let cpu retry
6351 * it to generate fault.
6353 if (gpa == UNMAPPED_GVA)
6358 * Do not retry the unhandleable instruction if it faults on the
6359 * readonly host memory, otherwise it will goto a infinite loop:
6360 * retry instruction -> write #PF -> emulation fail -> retry
6361 * instruction -> ...
6363 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6366 * If the instruction failed on the error pfn, it can not be fixed,
6367 * report the error to userspace.
6369 if (is_error_noslot_pfn(pfn))
6372 kvm_release_pfn_clean(pfn);
6374 /* The instructions are well-emulated on direct mmu. */
6375 if (vcpu->arch.mmu->direct_map) {
6376 unsigned int indirect_shadow_pages;
6378 spin_lock(&vcpu->kvm->mmu_lock);
6379 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6380 spin_unlock(&vcpu->kvm->mmu_lock);
6382 if (indirect_shadow_pages)
6383 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6389 * if emulation was due to access to shadowed page table
6390 * and it failed try to unshadow page and re-enter the
6391 * guest to let CPU execute the instruction.
6393 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6396 * If the access faults on its page table, it can not
6397 * be fixed by unprotecting shadow page and it should
6398 * be reported to userspace.
6400 return !write_fault_to_shadow_pgtable;
6403 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6404 unsigned long cr2, int emulation_type)
6406 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6407 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6409 last_retry_eip = vcpu->arch.last_retry_eip;
6410 last_retry_addr = vcpu->arch.last_retry_addr;
6413 * If the emulation is caused by #PF and it is non-page_table
6414 * writing instruction, it means the VM-EXIT is caused by shadow
6415 * page protected, we can zap the shadow page and retry this
6416 * instruction directly.
6418 * Note: if the guest uses a non-page-table modifying instruction
6419 * on the PDE that points to the instruction, then we will unmap
6420 * the instruction and go to an infinite loop. So, we cache the
6421 * last retried eip and the last fault address, if we meet the eip
6422 * and the address again, we can break out of the potential infinite
6425 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6427 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6430 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6433 if (x86_page_table_writing_insn(ctxt))
6436 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6439 vcpu->arch.last_retry_eip = ctxt->eip;
6440 vcpu->arch.last_retry_addr = cr2;
6442 if (!vcpu->arch.mmu->direct_map)
6443 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6445 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6450 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6451 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6453 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6455 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6456 /* This is a good place to trace that we are exiting SMM. */
6457 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6459 /* Process a latched INIT or SMI, if any. */
6460 kvm_make_request(KVM_REQ_EVENT, vcpu);
6463 kvm_mmu_reset_context(vcpu);
6466 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6475 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6476 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6481 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6483 struct kvm_run *kvm_run = vcpu->run;
6485 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6486 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6487 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6488 kvm_run->debug.arch.exception = DB_VECTOR;
6489 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6492 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6496 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6498 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6501 r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6506 * rflags is the old, "raw" value of the flags. The new value has
6507 * not been saved yet.
6509 * This is correct even for TF set by the guest, because "the
6510 * processor will not generate this exception after the instruction
6511 * that sets the TF flag".
6513 if (unlikely(rflags & X86_EFLAGS_TF))
6514 r = kvm_vcpu_do_singlestep(vcpu);
6517 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6519 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6521 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6522 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6523 struct kvm_run *kvm_run = vcpu->run;
6524 unsigned long eip = kvm_get_linear_rip(vcpu);
6525 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6526 vcpu->arch.guest_debug_dr7,
6530 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6531 kvm_run->debug.arch.pc = eip;
6532 kvm_run->debug.arch.exception = DB_VECTOR;
6533 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6539 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6540 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6541 unsigned long eip = kvm_get_linear_rip(vcpu);
6542 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6547 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6548 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6549 kvm_queue_exception(vcpu, DB_VECTOR);
6558 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6560 switch (ctxt->opcode_len) {
6567 case 0xe6: /* OUT */
6571 case 0x6c: /* INS */
6573 case 0x6e: /* OUTS */
6580 case 0x33: /* RDPMC */
6589 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6596 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6597 bool writeback = true;
6598 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6600 vcpu->arch.l1tf_flush_l1d = true;
6603 * Clear write_fault_to_shadow_pgtable here to ensure it is
6606 vcpu->arch.write_fault_to_shadow_pgtable = false;
6607 kvm_clear_exception_queue(vcpu);
6609 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6610 init_emulate_ctxt(vcpu);
6613 * We will reenter on the same instruction since
6614 * we do not set complete_userspace_io. This does not
6615 * handle watchpoints yet, those would be handled in
6618 if (!(emulation_type & EMULTYPE_SKIP) &&
6619 kvm_vcpu_check_breakpoint(vcpu, &r))
6622 ctxt->interruptibility = 0;
6623 ctxt->have_exception = false;
6624 ctxt->exception.vector = -1;
6625 ctxt->perm_ok = false;
6627 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6629 r = x86_decode_insn(ctxt, insn, insn_len);
6631 trace_kvm_emulate_insn_start(vcpu);
6632 ++vcpu->stat.insn_emulation;
6633 if (r != EMULATION_OK) {
6634 if ((emulation_type & EMULTYPE_TRAP_UD) ||
6635 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6636 kvm_queue_exception(vcpu, UD_VECTOR);
6639 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6642 if (ctxt->have_exception) {
6644 * #UD should result in just EMULATION_FAILED, and trap-like
6645 * exception should not be encountered during decode.
6647 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6648 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6649 inject_emulated_exception(vcpu);
6652 return handle_emulation_failure(vcpu, emulation_type);
6656 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6657 !is_vmware_backdoor_opcode(ctxt)) {
6658 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6663 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6664 * for kvm_skip_emulated_instruction(). The caller is responsible for
6665 * updating interruptibility state and injecting single-step #DBs.
6667 if (emulation_type & EMULTYPE_SKIP) {
6668 kvm_rip_write(vcpu, ctxt->_eip);
6669 if (ctxt->eflags & X86_EFLAGS_RF)
6670 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6674 if (retry_instruction(ctxt, cr2, emulation_type))
6677 /* this is needed for vmware backdoor interface to work since it
6678 changes registers values during IO operation */
6679 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6680 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6681 emulator_invalidate_register_cache(ctxt);
6685 /* Save the faulting GPA (cr2) in the address field */
6686 ctxt->exception.address = cr2;
6688 r = x86_emulate_insn(ctxt);
6690 if (r == EMULATION_INTERCEPTED)
6693 if (r == EMULATION_FAILED) {
6694 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6698 return handle_emulation_failure(vcpu, emulation_type);
6701 if (ctxt->have_exception) {
6703 if (inject_emulated_exception(vcpu))
6705 } else if (vcpu->arch.pio.count) {
6706 if (!vcpu->arch.pio.in) {
6707 /* FIXME: return into emulator if single-stepping. */
6708 vcpu->arch.pio.count = 0;
6711 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6714 } else if (vcpu->mmio_needed) {
6715 ++vcpu->stat.mmio_exits;
6717 if (!vcpu->mmio_is_write)
6720 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6721 } else if (r == EMULATION_RESTART)
6727 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6728 toggle_interruptibility(vcpu, ctxt->interruptibility);
6729 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6730 if (!ctxt->have_exception ||
6731 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6732 kvm_rip_write(vcpu, ctxt->eip);
6734 r = kvm_vcpu_do_singlestep(vcpu);
6735 __kvm_set_rflags(vcpu, ctxt->eflags);
6739 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6740 * do nothing, and it will be requested again as soon as
6741 * the shadow expires. But we still need to check here,
6742 * because POPF has no interrupt shadow.
6744 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6745 kvm_make_request(KVM_REQ_EVENT, vcpu);
6747 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6752 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6754 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6756 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6758 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6759 void *insn, int insn_len)
6761 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6763 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6765 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6767 vcpu->arch.pio.count = 0;
6771 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6773 vcpu->arch.pio.count = 0;
6775 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6778 return kvm_skip_emulated_instruction(vcpu);
6781 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6782 unsigned short port)
6784 unsigned long val = kvm_rax_read(vcpu);
6785 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6786 size, port, &val, 1);
6791 * Workaround userspace that relies on old KVM behavior of %rip being
6792 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6795 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6796 vcpu->arch.complete_userspace_io =
6797 complete_fast_pio_out_port_0x7e;
6798 kvm_skip_emulated_instruction(vcpu);
6800 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6801 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6806 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6810 /* We should only ever be called with arch.pio.count equal to 1 */
6811 BUG_ON(vcpu->arch.pio.count != 1);
6813 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6814 vcpu->arch.pio.count = 0;
6818 /* For size less than 4 we merge, else we zero extend */
6819 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6822 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6823 * the copy and tracing
6825 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6826 vcpu->arch.pio.port, &val, 1);
6827 kvm_rax_write(vcpu, val);
6829 return kvm_skip_emulated_instruction(vcpu);
6832 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6833 unsigned short port)
6838 /* For size less than 4 we merge, else we zero extend */
6839 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6841 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6844 kvm_rax_write(vcpu, val);
6848 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6849 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6854 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6859 ret = kvm_fast_pio_in(vcpu, size, port);
6861 ret = kvm_fast_pio_out(vcpu, size, port);
6862 return ret && kvm_skip_emulated_instruction(vcpu);
6864 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6866 static int kvmclock_cpu_down_prep(unsigned int cpu)
6868 __this_cpu_write(cpu_tsc_khz, 0);
6872 static void tsc_khz_changed(void *data)
6874 struct cpufreq_freqs *freq = data;
6875 unsigned long khz = 0;
6879 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6880 khz = cpufreq_quick_get(raw_smp_processor_id());
6883 __this_cpu_write(cpu_tsc_khz, khz);
6886 #ifdef CONFIG_X86_64
6887 static void kvm_hyperv_tsc_notifier(void)
6890 struct kvm_vcpu *vcpu;
6893 mutex_lock(&kvm_lock);
6894 list_for_each_entry(kvm, &vm_list, vm_list)
6895 kvm_make_mclock_inprogress_request(kvm);
6897 hyperv_stop_tsc_emulation();
6899 /* TSC frequency always matches when on Hyper-V */
6900 for_each_present_cpu(cpu)
6901 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6902 kvm_max_guest_tsc_khz = tsc_khz;
6904 list_for_each_entry(kvm, &vm_list, vm_list) {
6905 struct kvm_arch *ka = &kvm->arch;
6907 spin_lock(&ka->pvclock_gtod_sync_lock);
6909 pvclock_update_vm_gtod_copy(kvm);
6911 kvm_for_each_vcpu(cpu, vcpu, kvm)
6912 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6914 kvm_for_each_vcpu(cpu, vcpu, kvm)
6915 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6917 spin_unlock(&ka->pvclock_gtod_sync_lock);
6919 mutex_unlock(&kvm_lock);
6923 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6926 struct kvm_vcpu *vcpu;
6927 int i, send_ipi = 0;
6930 * We allow guests to temporarily run on slowing clocks,
6931 * provided we notify them after, or to run on accelerating
6932 * clocks, provided we notify them before. Thus time never
6935 * However, we have a problem. We can't atomically update
6936 * the frequency of a given CPU from this function; it is
6937 * merely a notifier, which can be called from any CPU.
6938 * Changing the TSC frequency at arbitrary points in time
6939 * requires a recomputation of local variables related to
6940 * the TSC for each VCPU. We must flag these local variables
6941 * to be updated and be sure the update takes place with the
6942 * new frequency before any guests proceed.
6944 * Unfortunately, the combination of hotplug CPU and frequency
6945 * change creates an intractable locking scenario; the order
6946 * of when these callouts happen is undefined with respect to
6947 * CPU hotplug, and they can race with each other. As such,
6948 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6949 * undefined; you can actually have a CPU frequency change take
6950 * place in between the computation of X and the setting of the
6951 * variable. To protect against this problem, all updates of
6952 * the per_cpu tsc_khz variable are done in an interrupt
6953 * protected IPI, and all callers wishing to update the value
6954 * must wait for a synchronous IPI to complete (which is trivial
6955 * if the caller is on the CPU already). This establishes the
6956 * necessary total order on variable updates.
6958 * Note that because a guest time update may take place
6959 * anytime after the setting of the VCPU's request bit, the
6960 * correct TSC value must be set before the request. However,
6961 * to ensure the update actually makes it to any guest which
6962 * starts running in hardware virtualization between the set
6963 * and the acquisition of the spinlock, we must also ping the
6964 * CPU after setting the request bit.
6968 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6970 mutex_lock(&kvm_lock);
6971 list_for_each_entry(kvm, &vm_list, vm_list) {
6972 kvm_for_each_vcpu(i, vcpu, kvm) {
6973 if (vcpu->cpu != cpu)
6975 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6976 if (vcpu->cpu != raw_smp_processor_id())
6980 mutex_unlock(&kvm_lock);
6982 if (freq->old < freq->new && send_ipi) {
6984 * We upscale the frequency. Must make the guest
6985 * doesn't see old kvmclock values while running with
6986 * the new frequency, otherwise we risk the guest sees
6987 * time go backwards.
6989 * In case we update the frequency for another cpu
6990 * (which might be in guest context) send an interrupt
6991 * to kick the cpu out of guest context. Next time
6992 * guest context is entered kvmclock will be updated,
6993 * so the guest will not see stale values.
6995 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6999 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7002 struct cpufreq_freqs *freq = data;
7005 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7007 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7010 for_each_cpu(cpu, freq->policy->cpus)
7011 __kvmclock_cpufreq_notifier(freq, cpu);
7016 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7017 .notifier_call = kvmclock_cpufreq_notifier
7020 static int kvmclock_cpu_online(unsigned int cpu)
7022 tsc_khz_changed(NULL);
7026 static void kvm_timer_init(void)
7028 max_tsc_khz = tsc_khz;
7030 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7031 #ifdef CONFIG_CPU_FREQ
7032 struct cpufreq_policy policy;
7035 memset(&policy, 0, sizeof(policy));
7037 cpufreq_get_policy(&policy, cpu);
7038 if (policy.cpuinfo.max_freq)
7039 max_tsc_khz = policy.cpuinfo.max_freq;
7042 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7043 CPUFREQ_TRANSITION_NOTIFIER);
7046 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7047 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7050 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7051 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7053 int kvm_is_in_guest(void)
7055 return __this_cpu_read(current_vcpu) != NULL;
7058 static int kvm_is_user_mode(void)
7062 if (__this_cpu_read(current_vcpu))
7063 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7065 return user_mode != 0;
7068 static unsigned long kvm_get_guest_ip(void)
7070 unsigned long ip = 0;
7072 if (__this_cpu_read(current_vcpu))
7073 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7078 static void kvm_handle_intel_pt_intr(void)
7080 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7082 kvm_make_request(KVM_REQ_PMI, vcpu);
7083 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7084 (unsigned long *)&vcpu->arch.pmu.global_status);
7087 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7088 .is_in_guest = kvm_is_in_guest,
7089 .is_user_mode = kvm_is_user_mode,
7090 .get_guest_ip = kvm_get_guest_ip,
7091 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7094 #ifdef CONFIG_X86_64
7095 static void pvclock_gtod_update_fn(struct work_struct *work)
7099 struct kvm_vcpu *vcpu;
7102 mutex_lock(&kvm_lock);
7103 list_for_each_entry(kvm, &vm_list, vm_list)
7104 kvm_for_each_vcpu(i, vcpu, kvm)
7105 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7106 atomic_set(&kvm_guest_has_master_clock, 0);
7107 mutex_unlock(&kvm_lock);
7110 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7113 * Notification about pvclock gtod data update.
7115 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7118 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7119 struct timekeeper *tk = priv;
7121 update_pvclock_gtod(tk);
7123 /* disable master clock if host does not trust, or does not
7124 * use, TSC based clocksource.
7126 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7127 atomic_read(&kvm_guest_has_master_clock) != 0)
7128 queue_work(system_long_wq, &pvclock_gtod_work);
7133 static struct notifier_block pvclock_gtod_notifier = {
7134 .notifier_call = pvclock_gtod_notify,
7138 int kvm_arch_init(void *opaque)
7141 struct kvm_x86_ops *ops = opaque;
7144 printk(KERN_ERR "kvm: already loaded the other module\n");
7149 if (!ops->cpu_has_kvm_support()) {
7150 printk(KERN_ERR "kvm: no hardware support\n");
7154 if (ops->disabled_by_bios()) {
7155 printk(KERN_ERR "kvm: disabled by bios\n");
7161 * KVM explicitly assumes that the guest has an FPU and
7162 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7163 * vCPU's FPU state as a fxregs_state struct.
7165 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7166 printk(KERN_ERR "kvm: inadequate fpu\n");
7172 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7173 __alignof__(struct fpu), SLAB_ACCOUNT,
7175 if (!x86_fpu_cache) {
7176 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7180 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7182 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7183 goto out_free_x86_fpu_cache;
7186 r = kvm_mmu_module_init();
7188 goto out_free_percpu;
7192 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7193 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7194 PT_PRESENT_MASK, 0, sme_me_mask);
7197 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7199 if (boot_cpu_has(X86_FEATURE_XSAVE))
7200 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7203 if (pi_inject_timer == -1)
7204 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7205 #ifdef CONFIG_X86_64
7206 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7208 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7209 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7215 free_percpu(shared_msrs);
7216 out_free_x86_fpu_cache:
7217 kmem_cache_destroy(x86_fpu_cache);
7222 void kvm_arch_exit(void)
7224 #ifdef CONFIG_X86_64
7225 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7226 clear_hv_tscchange_cb();
7229 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7231 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7232 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7233 CPUFREQ_TRANSITION_NOTIFIER);
7234 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7235 #ifdef CONFIG_X86_64
7236 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7239 kvm_mmu_module_exit();
7240 free_percpu(shared_msrs);
7241 kmem_cache_destroy(x86_fpu_cache);
7244 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7246 ++vcpu->stat.halt_exits;
7247 if (lapic_in_kernel(vcpu)) {
7248 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7251 vcpu->run->exit_reason = KVM_EXIT_HLT;
7255 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7257 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7259 int ret = kvm_skip_emulated_instruction(vcpu);
7261 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7262 * KVM_EXIT_DEBUG here.
7264 return kvm_vcpu_halt(vcpu) && ret;
7266 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7268 #ifdef CONFIG_X86_64
7269 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7270 unsigned long clock_type)
7272 struct kvm_clock_pairing clock_pairing;
7273 struct timespec64 ts;
7277 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7278 return -KVM_EOPNOTSUPP;
7280 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7281 return -KVM_EOPNOTSUPP;
7283 clock_pairing.sec = ts.tv_sec;
7284 clock_pairing.nsec = ts.tv_nsec;
7285 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7286 clock_pairing.flags = 0;
7287 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7290 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7291 sizeof(struct kvm_clock_pairing)))
7299 * kvm_pv_kick_cpu_op: Kick a vcpu.
7301 * @apicid - apicid of vcpu to be kicked.
7303 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7305 struct kvm_lapic_irq lapic_irq;
7307 lapic_irq.shorthand = 0;
7308 lapic_irq.dest_mode = 0;
7309 lapic_irq.level = 0;
7310 lapic_irq.dest_id = apicid;
7311 lapic_irq.msi_redir_hint = false;
7313 lapic_irq.delivery_mode = APIC_DM_REMRD;
7314 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7317 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7319 if (!lapic_in_kernel(vcpu)) {
7320 WARN_ON_ONCE(vcpu->arch.apicv_active);
7323 if (!vcpu->arch.apicv_active)
7326 vcpu->arch.apicv_active = false;
7327 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7330 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7332 struct kvm_vcpu *target = NULL;
7333 struct kvm_apic_map *map;
7336 map = rcu_dereference(kvm->arch.apic_map);
7338 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7339 target = map->phys_map[dest_id]->vcpu;
7343 if (target && READ_ONCE(target->ready))
7344 kvm_vcpu_yield_to(target);
7347 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7349 unsigned long nr, a0, a1, a2, a3, ret;
7352 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7353 return kvm_hv_hypercall(vcpu);
7355 nr = kvm_rax_read(vcpu);
7356 a0 = kvm_rbx_read(vcpu);
7357 a1 = kvm_rcx_read(vcpu);
7358 a2 = kvm_rdx_read(vcpu);
7359 a3 = kvm_rsi_read(vcpu);
7361 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7363 op_64_bit = is_64_bit_mode(vcpu);
7372 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7378 case KVM_HC_VAPIC_POLL_IRQ:
7381 case KVM_HC_KICK_CPU:
7382 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7383 kvm_sched_yield(vcpu->kvm, a1);
7386 #ifdef CONFIG_X86_64
7387 case KVM_HC_CLOCK_PAIRING:
7388 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7391 case KVM_HC_SEND_IPI:
7392 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7394 case KVM_HC_SCHED_YIELD:
7395 kvm_sched_yield(vcpu->kvm, a0);
7405 kvm_rax_write(vcpu, ret);
7407 ++vcpu->stat.hypercalls;
7408 return kvm_skip_emulated_instruction(vcpu);
7410 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7412 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7414 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7415 char instruction[3];
7416 unsigned long rip = kvm_rip_read(vcpu);
7418 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7420 return emulator_write_emulated(ctxt, rip, instruction, 3,
7424 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7426 return vcpu->run->request_interrupt_window &&
7427 likely(!pic_in_kernel(vcpu->kvm));
7430 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7432 struct kvm_run *kvm_run = vcpu->run;
7434 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7435 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7436 kvm_run->cr8 = kvm_get_cr8(vcpu);
7437 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7438 kvm_run->ready_for_interrupt_injection =
7439 pic_in_kernel(vcpu->kvm) ||
7440 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7443 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7447 if (!kvm_x86_ops->update_cr8_intercept)
7450 if (!lapic_in_kernel(vcpu))
7453 if (vcpu->arch.apicv_active)
7456 if (!vcpu->arch.apic->vapic_addr)
7457 max_irr = kvm_lapic_find_highest_irr(vcpu);
7464 tpr = kvm_lapic_get_cr8(vcpu);
7466 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7469 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7473 /* try to reinject previous events if any */
7475 if (vcpu->arch.exception.injected)
7476 kvm_x86_ops->queue_exception(vcpu);
7478 * Do not inject an NMI or interrupt if there is a pending
7479 * exception. Exceptions and interrupts are recognized at
7480 * instruction boundaries, i.e. the start of an instruction.
7481 * Trap-like exceptions, e.g. #DB, have higher priority than
7482 * NMIs and interrupts, i.e. traps are recognized before an
7483 * NMI/interrupt that's pending on the same instruction.
7484 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7485 * priority, but are only generated (pended) during instruction
7486 * execution, i.e. a pending fault-like exception means the
7487 * fault occurred on the *previous* instruction and must be
7488 * serviced prior to recognizing any new events in order to
7489 * fully complete the previous instruction.
7491 else if (!vcpu->arch.exception.pending) {
7492 if (vcpu->arch.nmi_injected)
7493 kvm_x86_ops->set_nmi(vcpu);
7494 else if (vcpu->arch.interrupt.injected)
7495 kvm_x86_ops->set_irq(vcpu);
7499 * Call check_nested_events() even if we reinjected a previous event
7500 * in order for caller to determine if it should require immediate-exit
7501 * from L2 to L1 due to pending L1 events which require exit
7504 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7505 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7510 /* try to inject new event if pending */
7511 if (vcpu->arch.exception.pending) {
7512 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7513 vcpu->arch.exception.has_error_code,
7514 vcpu->arch.exception.error_code);
7516 WARN_ON_ONCE(vcpu->arch.exception.injected);
7517 vcpu->arch.exception.pending = false;
7518 vcpu->arch.exception.injected = true;
7520 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7521 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7524 if (vcpu->arch.exception.nr == DB_VECTOR) {
7526 * This code assumes that nSVM doesn't use
7527 * check_nested_events(). If it does, the
7528 * DR6/DR7 changes should happen before L1
7529 * gets a #VMEXIT for an intercepted #DB in
7530 * L2. (Under VMX, on the other hand, the
7531 * DR6/DR7 changes should not happen in the
7532 * event of a VM-exit to L1 for an intercepted
7535 kvm_deliver_exception_payload(vcpu);
7536 if (vcpu->arch.dr7 & DR7_GD) {
7537 vcpu->arch.dr7 &= ~DR7_GD;
7538 kvm_update_dr7(vcpu);
7542 kvm_x86_ops->queue_exception(vcpu);
7545 /* Don't consider new event if we re-injected an event */
7546 if (kvm_event_needs_reinjection(vcpu))
7549 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7550 kvm_x86_ops->smi_allowed(vcpu)) {
7551 vcpu->arch.smi_pending = false;
7552 ++vcpu->arch.smi_count;
7554 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7555 --vcpu->arch.nmi_pending;
7556 vcpu->arch.nmi_injected = true;
7557 kvm_x86_ops->set_nmi(vcpu);
7558 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7560 * Because interrupts can be injected asynchronously, we are
7561 * calling check_nested_events again here to avoid a race condition.
7562 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7563 * proposal and current concerns. Perhaps we should be setting
7564 * KVM_REQ_EVENT only on certain events and not unconditionally?
7566 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7567 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7571 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7572 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7574 kvm_x86_ops->set_irq(vcpu);
7581 static void process_nmi(struct kvm_vcpu *vcpu)
7586 * x86 is limited to one NMI running, and one NMI pending after it.
7587 * If an NMI is already in progress, limit further NMIs to just one.
7588 * Otherwise, allow two (and we'll inject the first one immediately).
7590 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7593 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7594 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7595 kvm_make_request(KVM_REQ_EVENT, vcpu);
7598 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7601 flags |= seg->g << 23;
7602 flags |= seg->db << 22;
7603 flags |= seg->l << 21;
7604 flags |= seg->avl << 20;
7605 flags |= seg->present << 15;
7606 flags |= seg->dpl << 13;
7607 flags |= seg->s << 12;
7608 flags |= seg->type << 8;
7612 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7614 struct kvm_segment seg;
7617 kvm_get_segment(vcpu, &seg, n);
7618 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7621 offset = 0x7f84 + n * 12;
7623 offset = 0x7f2c + (n - 3) * 12;
7625 put_smstate(u32, buf, offset + 8, seg.base);
7626 put_smstate(u32, buf, offset + 4, seg.limit);
7627 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7630 #ifdef CONFIG_X86_64
7631 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7633 struct kvm_segment seg;
7637 kvm_get_segment(vcpu, &seg, n);
7638 offset = 0x7e00 + n * 16;
7640 flags = enter_smm_get_segment_flags(&seg) >> 8;
7641 put_smstate(u16, buf, offset, seg.selector);
7642 put_smstate(u16, buf, offset + 2, flags);
7643 put_smstate(u32, buf, offset + 4, seg.limit);
7644 put_smstate(u64, buf, offset + 8, seg.base);
7648 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7651 struct kvm_segment seg;
7655 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7656 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7657 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7658 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7660 for (i = 0; i < 8; i++)
7661 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7663 kvm_get_dr(vcpu, 6, &val);
7664 put_smstate(u32, buf, 0x7fcc, (u32)val);
7665 kvm_get_dr(vcpu, 7, &val);
7666 put_smstate(u32, buf, 0x7fc8, (u32)val);
7668 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7669 put_smstate(u32, buf, 0x7fc4, seg.selector);
7670 put_smstate(u32, buf, 0x7f64, seg.base);
7671 put_smstate(u32, buf, 0x7f60, seg.limit);
7672 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7674 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7675 put_smstate(u32, buf, 0x7fc0, seg.selector);
7676 put_smstate(u32, buf, 0x7f80, seg.base);
7677 put_smstate(u32, buf, 0x7f7c, seg.limit);
7678 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7680 kvm_x86_ops->get_gdt(vcpu, &dt);
7681 put_smstate(u32, buf, 0x7f74, dt.address);
7682 put_smstate(u32, buf, 0x7f70, dt.size);
7684 kvm_x86_ops->get_idt(vcpu, &dt);
7685 put_smstate(u32, buf, 0x7f58, dt.address);
7686 put_smstate(u32, buf, 0x7f54, dt.size);
7688 for (i = 0; i < 6; i++)
7689 enter_smm_save_seg_32(vcpu, buf, i);
7691 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7694 put_smstate(u32, buf, 0x7efc, 0x00020000);
7695 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7698 #ifdef CONFIG_X86_64
7699 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7702 struct kvm_segment seg;
7706 for (i = 0; i < 16; i++)
7707 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7709 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7710 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7712 kvm_get_dr(vcpu, 6, &val);
7713 put_smstate(u64, buf, 0x7f68, val);
7714 kvm_get_dr(vcpu, 7, &val);
7715 put_smstate(u64, buf, 0x7f60, val);
7717 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7718 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7719 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7721 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7724 put_smstate(u32, buf, 0x7efc, 0x00020064);
7726 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7728 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7729 put_smstate(u16, buf, 0x7e90, seg.selector);
7730 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7731 put_smstate(u32, buf, 0x7e94, seg.limit);
7732 put_smstate(u64, buf, 0x7e98, seg.base);
7734 kvm_x86_ops->get_idt(vcpu, &dt);
7735 put_smstate(u32, buf, 0x7e84, dt.size);
7736 put_smstate(u64, buf, 0x7e88, dt.address);
7738 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7739 put_smstate(u16, buf, 0x7e70, seg.selector);
7740 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7741 put_smstate(u32, buf, 0x7e74, seg.limit);
7742 put_smstate(u64, buf, 0x7e78, seg.base);
7744 kvm_x86_ops->get_gdt(vcpu, &dt);
7745 put_smstate(u32, buf, 0x7e64, dt.size);
7746 put_smstate(u64, buf, 0x7e68, dt.address);
7748 for (i = 0; i < 6; i++)
7749 enter_smm_save_seg_64(vcpu, buf, i);
7753 static void enter_smm(struct kvm_vcpu *vcpu)
7755 struct kvm_segment cs, ds;
7760 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7761 memset(buf, 0, 512);
7762 #ifdef CONFIG_X86_64
7763 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7764 enter_smm_save_state_64(vcpu, buf);
7767 enter_smm_save_state_32(vcpu, buf);
7770 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7771 * vCPU state (e.g. leave guest mode) after we've saved the state into
7772 * the SMM state-save area.
7774 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7776 vcpu->arch.hflags |= HF_SMM_MASK;
7777 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7779 if (kvm_x86_ops->get_nmi_mask(vcpu))
7780 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7782 kvm_x86_ops->set_nmi_mask(vcpu, true);
7784 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7785 kvm_rip_write(vcpu, 0x8000);
7787 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7788 kvm_x86_ops->set_cr0(vcpu, cr0);
7789 vcpu->arch.cr0 = cr0;
7791 kvm_x86_ops->set_cr4(vcpu, 0);
7793 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7794 dt.address = dt.size = 0;
7795 kvm_x86_ops->set_idt(vcpu, &dt);
7797 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7799 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7800 cs.base = vcpu->arch.smbase;
7805 cs.limit = ds.limit = 0xffffffff;
7806 cs.type = ds.type = 0x3;
7807 cs.dpl = ds.dpl = 0;
7812 cs.avl = ds.avl = 0;
7813 cs.present = ds.present = 1;
7814 cs.unusable = ds.unusable = 0;
7815 cs.padding = ds.padding = 0;
7817 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7818 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7819 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7820 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7821 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7822 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7824 #ifdef CONFIG_X86_64
7825 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7826 kvm_x86_ops->set_efer(vcpu, 0);
7829 kvm_update_cpuid(vcpu);
7830 kvm_mmu_reset_context(vcpu);
7833 static void process_smi(struct kvm_vcpu *vcpu)
7835 vcpu->arch.smi_pending = true;
7836 kvm_make_request(KVM_REQ_EVENT, vcpu);
7839 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7841 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7844 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7846 if (!kvm_apic_present(vcpu))
7849 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7851 if (irqchip_split(vcpu->kvm))
7852 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7854 if (vcpu->arch.apicv_active)
7855 kvm_x86_ops->sync_pir_to_irr(vcpu);
7856 if (ioapic_in_kernel(vcpu->kvm))
7857 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7860 if (is_guest_mode(vcpu))
7861 vcpu->arch.load_eoi_exitmap_pending = true;
7863 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7866 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7868 u64 eoi_exit_bitmap[4];
7870 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7873 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7874 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7875 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7878 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7879 unsigned long start, unsigned long end,
7882 unsigned long apic_address;
7885 * The physical address of apic access page is stored in the VMCS.
7886 * Update it when it becomes invalid.
7888 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7889 if (start <= apic_address && apic_address < end)
7890 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7895 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7897 struct page *page = NULL;
7899 if (!lapic_in_kernel(vcpu))
7902 if (!kvm_x86_ops->set_apic_access_page_addr)
7905 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7906 if (is_error_page(page))
7908 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7911 * Do not pin apic access page in memory, the MMU notifier
7912 * will call us again if it is migrated or swapped out.
7916 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7918 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7920 smp_send_reschedule(vcpu->cpu);
7922 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7925 * Returns 1 to let vcpu_run() continue the guest execution loop without
7926 * exiting to the userspace. Otherwise, the value will be returned to the
7929 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7933 dm_request_for_irq_injection(vcpu) &&
7934 kvm_cpu_accept_dm_intr(vcpu);
7936 bool req_immediate_exit = false;
7938 if (kvm_request_pending(vcpu)) {
7939 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7940 kvm_x86_ops->get_vmcs12_pages(vcpu);
7941 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7942 kvm_mmu_unload(vcpu);
7943 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7944 __kvm_migrate_timers(vcpu);
7945 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7946 kvm_gen_update_masterclock(vcpu->kvm);
7947 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7948 kvm_gen_kvmclock_update(vcpu);
7949 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7950 r = kvm_guest_time_update(vcpu);
7954 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7955 kvm_mmu_sync_roots(vcpu);
7956 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7957 kvm_mmu_load_cr3(vcpu);
7958 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7959 kvm_vcpu_flush_tlb(vcpu, true);
7960 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7961 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7965 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7966 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7967 vcpu->mmio_needed = 0;
7971 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7972 /* Page is swapped out. Do synthetic halt */
7973 vcpu->arch.apf.halted = true;
7977 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7978 record_steal_time(vcpu);
7979 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7981 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7983 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7984 kvm_pmu_handle_event(vcpu);
7985 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7986 kvm_pmu_deliver_pmi(vcpu);
7987 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7988 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7989 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7990 vcpu->arch.ioapic_handled_vectors)) {
7991 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7992 vcpu->run->eoi.vector =
7993 vcpu->arch.pending_ioapic_eoi;
7998 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7999 vcpu_scan_ioapic(vcpu);
8000 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8001 vcpu_load_eoi_exitmap(vcpu);
8002 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8003 kvm_vcpu_reload_apic_access_page(vcpu);
8004 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8005 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8006 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8010 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8011 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8012 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8016 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8017 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8018 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8024 * KVM_REQ_HV_STIMER has to be processed after
8025 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8026 * depend on the guest clock being up-to-date
8028 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8029 kvm_hv_process_stimers(vcpu);
8032 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8033 ++vcpu->stat.req_event;
8034 kvm_apic_accept_events(vcpu);
8035 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8040 if (inject_pending_event(vcpu, req_int_win) != 0)
8041 req_immediate_exit = true;
8043 /* Enable SMI/NMI/IRQ window open exits if needed.
8045 * SMIs have three cases:
8046 * 1) They can be nested, and then there is nothing to
8047 * do here because RSM will cause a vmexit anyway.
8048 * 2) There is an ISA-specific reason why SMI cannot be
8049 * injected, and the moment when this changes can be
8051 * 3) Or the SMI can be pending because
8052 * inject_pending_event has completed the injection
8053 * of an IRQ or NMI from the previous vmexit, and
8054 * then we request an immediate exit to inject the
8057 if (vcpu->arch.smi_pending && !is_smm(vcpu))
8058 if (!kvm_x86_ops->enable_smi_window(vcpu))
8059 req_immediate_exit = true;
8060 if (vcpu->arch.nmi_pending)
8061 kvm_x86_ops->enable_nmi_window(vcpu);
8062 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8063 kvm_x86_ops->enable_irq_window(vcpu);
8064 WARN_ON(vcpu->arch.exception.pending);
8067 if (kvm_lapic_enabled(vcpu)) {
8068 update_cr8_intercept(vcpu);
8069 kvm_lapic_sync_to_vapic(vcpu);
8073 r = kvm_mmu_reload(vcpu);
8075 goto cancel_injection;
8080 kvm_x86_ops->prepare_guest_switch(vcpu);
8083 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8084 * IPI are then delayed after guest entry, which ensures that they
8085 * result in virtual interrupt delivery.
8087 local_irq_disable();
8088 vcpu->mode = IN_GUEST_MODE;
8090 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8093 * 1) We should set ->mode before checking ->requests. Please see
8094 * the comment in kvm_vcpu_exiting_guest_mode().
8096 * 2) For APICv, we should set ->mode before checking PID.ON. This
8097 * pairs with the memory barrier implicit in pi_test_and_set_on
8098 * (see vmx_deliver_posted_interrupt).
8100 * 3) This also orders the write to mode from any reads to the page
8101 * tables done while the VCPU is running. Please see the comment
8102 * in kvm_flush_remote_tlbs.
8104 smp_mb__after_srcu_read_unlock();
8107 * This handles the case where a posted interrupt was
8108 * notified with kvm_vcpu_kick.
8110 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8111 kvm_x86_ops->sync_pir_to_irr(vcpu);
8113 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8114 || need_resched() || signal_pending(current)) {
8115 vcpu->mode = OUTSIDE_GUEST_MODE;
8119 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8121 goto cancel_injection;
8124 if (req_immediate_exit) {
8125 kvm_make_request(KVM_REQ_EVENT, vcpu);
8126 kvm_x86_ops->request_immediate_exit(vcpu);
8129 trace_kvm_entry(vcpu->vcpu_id);
8130 guest_enter_irqoff();
8132 /* The preempt notifier should have taken care of the FPU already. */
8133 WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8135 if (unlikely(vcpu->arch.switch_db_regs)) {
8137 set_debugreg(vcpu->arch.eff_db[0], 0);
8138 set_debugreg(vcpu->arch.eff_db[1], 1);
8139 set_debugreg(vcpu->arch.eff_db[2], 2);
8140 set_debugreg(vcpu->arch.eff_db[3], 3);
8141 set_debugreg(vcpu->arch.dr6, 6);
8142 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8145 kvm_x86_ops->run(vcpu);
8148 * Do this here before restoring debug registers on the host. And
8149 * since we do this before handling the vmexit, a DR access vmexit
8150 * can (a) read the correct value of the debug registers, (b) set
8151 * KVM_DEBUGREG_WONT_EXIT again.
8153 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8154 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8155 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8156 kvm_update_dr0123(vcpu);
8157 kvm_update_dr6(vcpu);
8158 kvm_update_dr7(vcpu);
8159 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8163 * If the guest has used debug registers, at least dr7
8164 * will be disabled while returning to the host.
8165 * If we don't have active breakpoints in the host, we don't
8166 * care about the messed up debug address registers. But if
8167 * we have some of them active, restore the old state.
8169 if (hw_breakpoint_active())
8170 hw_breakpoint_restore();
8172 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8174 vcpu->mode = OUTSIDE_GUEST_MODE;
8177 kvm_x86_ops->handle_exit_irqoff(vcpu);
8180 * Consume any pending interrupts, including the possible source of
8181 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8182 * An instruction is required after local_irq_enable() to fully unblock
8183 * interrupts on processors that implement an interrupt shadow, the
8184 * stat.exits increment will do nicely.
8186 kvm_before_interrupt(vcpu);
8189 local_irq_disable();
8190 kvm_after_interrupt(vcpu);
8192 guest_exit_irqoff();
8193 if (lapic_in_kernel(vcpu)) {
8194 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8195 if (delta != S64_MIN) {
8196 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8197 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8204 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8207 * Profile KVM exit RIPs:
8209 if (unlikely(prof_on == KVM_PROFILING)) {
8210 unsigned long rip = kvm_rip_read(vcpu);
8211 profile_hit(KVM_PROFILING, (void *)rip);
8214 if (unlikely(vcpu->arch.tsc_always_catchup))
8215 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8217 if (vcpu->arch.apic_attention)
8218 kvm_lapic_sync_from_vapic(vcpu);
8220 vcpu->arch.gpa_available = false;
8221 r = kvm_x86_ops->handle_exit(vcpu);
8225 kvm_x86_ops->cancel_injection(vcpu);
8226 if (unlikely(vcpu->arch.apic_attention))
8227 kvm_lapic_sync_from_vapic(vcpu);
8232 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8234 if (!kvm_arch_vcpu_runnable(vcpu) &&
8235 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8236 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8237 kvm_vcpu_block(vcpu);
8238 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8240 if (kvm_x86_ops->post_block)
8241 kvm_x86_ops->post_block(vcpu);
8243 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8247 kvm_apic_accept_events(vcpu);
8248 switch(vcpu->arch.mp_state) {
8249 case KVM_MP_STATE_HALTED:
8250 vcpu->arch.pv.pv_unhalted = false;
8251 vcpu->arch.mp_state =
8252 KVM_MP_STATE_RUNNABLE;
8254 case KVM_MP_STATE_RUNNABLE:
8255 vcpu->arch.apf.halted = false;
8257 case KVM_MP_STATE_INIT_RECEIVED:
8266 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8268 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8269 kvm_x86_ops->check_nested_events(vcpu, false);
8271 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8272 !vcpu->arch.apf.halted);
8275 static int vcpu_run(struct kvm_vcpu *vcpu)
8278 struct kvm *kvm = vcpu->kvm;
8280 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8281 vcpu->arch.l1tf_flush_l1d = true;
8284 if (kvm_vcpu_running(vcpu)) {
8285 r = vcpu_enter_guest(vcpu);
8287 r = vcpu_block(kvm, vcpu);
8293 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8294 if (kvm_cpu_has_pending_timer(vcpu))
8295 kvm_inject_pending_timer_irqs(vcpu);
8297 if (dm_request_for_irq_injection(vcpu) &&
8298 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8300 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8301 ++vcpu->stat.request_irq_exits;
8305 kvm_check_async_pf_completion(vcpu);
8307 if (signal_pending(current)) {
8309 vcpu->run->exit_reason = KVM_EXIT_INTR;
8310 ++vcpu->stat.signal_exits;
8313 if (need_resched()) {
8314 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8316 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8320 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8325 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8329 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8330 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8331 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8335 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8337 BUG_ON(!vcpu->arch.pio.count);
8339 return complete_emulated_io(vcpu);
8343 * Implements the following, as a state machine:
8347 * for each mmio piece in the fragment
8355 * for each mmio piece in the fragment
8360 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8362 struct kvm_run *run = vcpu->run;
8363 struct kvm_mmio_fragment *frag;
8366 BUG_ON(!vcpu->mmio_needed);
8368 /* Complete previous fragment */
8369 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8370 len = min(8u, frag->len);
8371 if (!vcpu->mmio_is_write)
8372 memcpy(frag->data, run->mmio.data, len);
8374 if (frag->len <= 8) {
8375 /* Switch to the next fragment. */
8377 vcpu->mmio_cur_fragment++;
8379 /* Go forward to the next mmio piece. */
8385 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8386 vcpu->mmio_needed = 0;
8388 /* FIXME: return into emulator if single-stepping. */
8389 if (vcpu->mmio_is_write)
8391 vcpu->mmio_read_completed = 1;
8392 return complete_emulated_io(vcpu);
8395 run->exit_reason = KVM_EXIT_MMIO;
8396 run->mmio.phys_addr = frag->gpa;
8397 if (vcpu->mmio_is_write)
8398 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8399 run->mmio.len = min(8u, frag->len);
8400 run->mmio.is_write = vcpu->mmio_is_write;
8401 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8405 /* Swap (qemu) user FPU context for the guest FPU context. */
8406 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8410 copy_fpregs_to_fpstate(vcpu->arch.user_fpu);
8411 /* PKRU is separately restored in kvm_x86_ops->run. */
8412 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8413 ~XFEATURE_MASK_PKRU);
8415 fpregs_mark_activate();
8421 /* When vcpu_run ends, restore user space FPU context. */
8422 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8426 copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8427 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8429 fpregs_mark_activate();
8432 ++vcpu->stat.fpu_reload;
8436 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8441 kvm_sigset_activate(vcpu);
8442 kvm_load_guest_fpu(vcpu);
8444 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8445 if (kvm_run->immediate_exit) {
8449 kvm_vcpu_block(vcpu);
8450 kvm_apic_accept_events(vcpu);
8451 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8453 if (signal_pending(current)) {
8455 vcpu->run->exit_reason = KVM_EXIT_INTR;
8456 ++vcpu->stat.signal_exits;
8461 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8466 if (vcpu->run->kvm_dirty_regs) {
8467 r = sync_regs(vcpu);
8472 /* re-sync apic's tpr */
8473 if (!lapic_in_kernel(vcpu)) {
8474 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8480 if (unlikely(vcpu->arch.complete_userspace_io)) {
8481 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8482 vcpu->arch.complete_userspace_io = NULL;
8487 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8489 if (kvm_run->immediate_exit)
8495 kvm_put_guest_fpu(vcpu);
8496 if (vcpu->run->kvm_valid_regs)
8498 post_kvm_run_save(vcpu);
8499 kvm_sigset_deactivate(vcpu);
8505 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8507 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8509 * We are here if userspace calls get_regs() in the middle of
8510 * instruction emulation. Registers state needs to be copied
8511 * back from emulation context to vcpu. Userspace shouldn't do
8512 * that usually, but some bad designed PV devices (vmware
8513 * backdoor interface) need this to work
8515 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8516 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8518 regs->rax = kvm_rax_read(vcpu);
8519 regs->rbx = kvm_rbx_read(vcpu);
8520 regs->rcx = kvm_rcx_read(vcpu);
8521 regs->rdx = kvm_rdx_read(vcpu);
8522 regs->rsi = kvm_rsi_read(vcpu);
8523 regs->rdi = kvm_rdi_read(vcpu);
8524 regs->rsp = kvm_rsp_read(vcpu);
8525 regs->rbp = kvm_rbp_read(vcpu);
8526 #ifdef CONFIG_X86_64
8527 regs->r8 = kvm_r8_read(vcpu);
8528 regs->r9 = kvm_r9_read(vcpu);
8529 regs->r10 = kvm_r10_read(vcpu);
8530 regs->r11 = kvm_r11_read(vcpu);
8531 regs->r12 = kvm_r12_read(vcpu);
8532 regs->r13 = kvm_r13_read(vcpu);
8533 regs->r14 = kvm_r14_read(vcpu);
8534 regs->r15 = kvm_r15_read(vcpu);
8537 regs->rip = kvm_rip_read(vcpu);
8538 regs->rflags = kvm_get_rflags(vcpu);
8541 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8544 __get_regs(vcpu, regs);
8549 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8551 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8552 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8554 kvm_rax_write(vcpu, regs->rax);
8555 kvm_rbx_write(vcpu, regs->rbx);
8556 kvm_rcx_write(vcpu, regs->rcx);
8557 kvm_rdx_write(vcpu, regs->rdx);
8558 kvm_rsi_write(vcpu, regs->rsi);
8559 kvm_rdi_write(vcpu, regs->rdi);
8560 kvm_rsp_write(vcpu, regs->rsp);
8561 kvm_rbp_write(vcpu, regs->rbp);
8562 #ifdef CONFIG_X86_64
8563 kvm_r8_write(vcpu, regs->r8);
8564 kvm_r9_write(vcpu, regs->r9);
8565 kvm_r10_write(vcpu, regs->r10);
8566 kvm_r11_write(vcpu, regs->r11);
8567 kvm_r12_write(vcpu, regs->r12);
8568 kvm_r13_write(vcpu, regs->r13);
8569 kvm_r14_write(vcpu, regs->r14);
8570 kvm_r15_write(vcpu, regs->r15);
8573 kvm_rip_write(vcpu, regs->rip);
8574 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8576 vcpu->arch.exception.pending = false;
8578 kvm_make_request(KVM_REQ_EVENT, vcpu);
8581 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8584 __set_regs(vcpu, regs);
8589 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8591 struct kvm_segment cs;
8593 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8597 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8599 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8603 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8604 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8605 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8606 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8607 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8608 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8610 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8611 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8613 kvm_x86_ops->get_idt(vcpu, &dt);
8614 sregs->idt.limit = dt.size;
8615 sregs->idt.base = dt.address;
8616 kvm_x86_ops->get_gdt(vcpu, &dt);
8617 sregs->gdt.limit = dt.size;
8618 sregs->gdt.base = dt.address;
8620 sregs->cr0 = kvm_read_cr0(vcpu);
8621 sregs->cr2 = vcpu->arch.cr2;
8622 sregs->cr3 = kvm_read_cr3(vcpu);
8623 sregs->cr4 = kvm_read_cr4(vcpu);
8624 sregs->cr8 = kvm_get_cr8(vcpu);
8625 sregs->efer = vcpu->arch.efer;
8626 sregs->apic_base = kvm_get_apic_base(vcpu);
8628 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8630 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8631 set_bit(vcpu->arch.interrupt.nr,
8632 (unsigned long *)sregs->interrupt_bitmap);
8635 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8636 struct kvm_sregs *sregs)
8639 __get_sregs(vcpu, sregs);
8644 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8645 struct kvm_mp_state *mp_state)
8649 kvm_apic_accept_events(vcpu);
8650 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8651 vcpu->arch.pv.pv_unhalted)
8652 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8654 mp_state->mp_state = vcpu->arch.mp_state;
8660 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8661 struct kvm_mp_state *mp_state)
8667 if (!lapic_in_kernel(vcpu) &&
8668 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8671 /* INITs are latched while in SMM */
8672 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8673 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8674 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8677 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8678 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8679 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8681 vcpu->arch.mp_state = mp_state->mp_state;
8682 kvm_make_request(KVM_REQ_EVENT, vcpu);
8690 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8691 int reason, bool has_error_code, u32 error_code)
8693 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8696 init_emulate_ctxt(vcpu);
8698 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8699 has_error_code, error_code);
8701 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8702 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8703 vcpu->run->internal.ndata = 0;
8707 kvm_rip_write(vcpu, ctxt->eip);
8708 kvm_set_rflags(vcpu, ctxt->eflags);
8709 kvm_make_request(KVM_REQ_EVENT, vcpu);
8712 EXPORT_SYMBOL_GPL(kvm_task_switch);
8714 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8716 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8717 (sregs->cr4 & X86_CR4_OSXSAVE))
8720 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8722 * When EFER.LME and CR0.PG are set, the processor is in
8723 * 64-bit mode (though maybe in a 32-bit code segment).
8724 * CR4.PAE and EFER.LMA must be set.
8726 if (!(sregs->cr4 & X86_CR4_PAE)
8727 || !(sregs->efer & EFER_LMA))
8731 * Not in 64-bit mode: EFER.LMA is clear and the code
8732 * segment cannot be 64-bit.
8734 if (sregs->efer & EFER_LMA || sregs->cs.l)
8741 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8743 struct msr_data apic_base_msr;
8744 int mmu_reset_needed = 0;
8745 int cpuid_update_needed = 0;
8746 int pending_vec, max_bits, idx;
8750 if (kvm_valid_sregs(vcpu, sregs))
8753 apic_base_msr.data = sregs->apic_base;
8754 apic_base_msr.host_initiated = true;
8755 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8758 dt.size = sregs->idt.limit;
8759 dt.address = sregs->idt.base;
8760 kvm_x86_ops->set_idt(vcpu, &dt);
8761 dt.size = sregs->gdt.limit;
8762 dt.address = sregs->gdt.base;
8763 kvm_x86_ops->set_gdt(vcpu, &dt);
8765 vcpu->arch.cr2 = sregs->cr2;
8766 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8767 vcpu->arch.cr3 = sregs->cr3;
8768 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8770 kvm_set_cr8(vcpu, sregs->cr8);
8772 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8773 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8775 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8776 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8777 vcpu->arch.cr0 = sregs->cr0;
8779 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8780 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8781 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8782 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8783 if (cpuid_update_needed)
8784 kvm_update_cpuid(vcpu);
8786 idx = srcu_read_lock(&vcpu->kvm->srcu);
8787 if (is_pae_paging(vcpu)) {
8788 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8789 mmu_reset_needed = 1;
8791 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8793 if (mmu_reset_needed)
8794 kvm_mmu_reset_context(vcpu);
8796 max_bits = KVM_NR_INTERRUPTS;
8797 pending_vec = find_first_bit(
8798 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8799 if (pending_vec < max_bits) {
8800 kvm_queue_interrupt(vcpu, pending_vec, false);
8801 pr_debug("Set back pending irq %d\n", pending_vec);
8804 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8805 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8806 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8807 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8808 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8809 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8811 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8812 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8814 update_cr8_intercept(vcpu);
8816 /* Older userspace won't unhalt the vcpu on reset. */
8817 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8818 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8820 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8822 kvm_make_request(KVM_REQ_EVENT, vcpu);
8829 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8830 struct kvm_sregs *sregs)
8835 ret = __set_sregs(vcpu, sregs);
8840 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8841 struct kvm_guest_debug *dbg)
8843 unsigned long rflags;
8848 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8850 if (vcpu->arch.exception.pending)
8852 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8853 kvm_queue_exception(vcpu, DB_VECTOR);
8855 kvm_queue_exception(vcpu, BP_VECTOR);
8859 * Read rflags as long as potentially injected trace flags are still
8862 rflags = kvm_get_rflags(vcpu);
8864 vcpu->guest_debug = dbg->control;
8865 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8866 vcpu->guest_debug = 0;
8868 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8869 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8870 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8871 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8873 for (i = 0; i < KVM_NR_DB_REGS; i++)
8874 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8876 kvm_update_dr7(vcpu);
8878 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8879 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8880 get_segment_base(vcpu, VCPU_SREG_CS);
8883 * Trigger an rflags update that will inject or remove the trace
8886 kvm_set_rflags(vcpu, rflags);
8888 kvm_x86_ops->update_bp_intercept(vcpu);
8898 * Translate a guest virtual address to a guest physical address.
8900 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8901 struct kvm_translation *tr)
8903 unsigned long vaddr = tr->linear_address;
8909 idx = srcu_read_lock(&vcpu->kvm->srcu);
8910 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8911 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8912 tr->physical_address = gpa;
8913 tr->valid = gpa != UNMAPPED_GVA;
8921 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8923 struct fxregs_state *fxsave;
8927 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8928 memcpy(fpu->fpr, fxsave->st_space, 128);
8929 fpu->fcw = fxsave->cwd;
8930 fpu->fsw = fxsave->swd;
8931 fpu->ftwx = fxsave->twd;
8932 fpu->last_opcode = fxsave->fop;
8933 fpu->last_ip = fxsave->rip;
8934 fpu->last_dp = fxsave->rdp;
8935 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8941 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8943 struct fxregs_state *fxsave;
8947 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8949 memcpy(fxsave->st_space, fpu->fpr, 128);
8950 fxsave->cwd = fpu->fcw;
8951 fxsave->swd = fpu->fsw;
8952 fxsave->twd = fpu->ftwx;
8953 fxsave->fop = fpu->last_opcode;
8954 fxsave->rip = fpu->last_ip;
8955 fxsave->rdp = fpu->last_dp;
8956 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8962 static void store_regs(struct kvm_vcpu *vcpu)
8964 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8966 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8967 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8969 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8970 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8972 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8973 kvm_vcpu_ioctl_x86_get_vcpu_events(
8974 vcpu, &vcpu->run->s.regs.events);
8977 static int sync_regs(struct kvm_vcpu *vcpu)
8979 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8982 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8983 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8984 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8986 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8987 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8989 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8991 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8992 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8993 vcpu, &vcpu->run->s.regs.events))
8995 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9001 static void fx_init(struct kvm_vcpu *vcpu)
9003 fpstate_init(&vcpu->arch.guest_fpu->state);
9004 if (boot_cpu_has(X86_FEATURE_XSAVES))
9005 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9006 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9009 * Ensure guest xcr0 is valid for loading
9011 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9013 vcpu->arch.cr0 |= X86_CR0_ET;
9016 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
9018 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
9020 kvmclock_reset(vcpu);
9022 kvm_x86_ops->vcpu_free(vcpu);
9023 free_cpumask_var(wbinvd_dirty_mask);
9026 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
9029 struct kvm_vcpu *vcpu;
9031 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9032 printk_once(KERN_WARNING
9033 "kvm: SMP vm created on host with unstable TSC; "
9034 "guest TSC will not be reliable\n");
9036 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
9041 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
9043 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9044 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9045 kvm_vcpu_mtrr_init(vcpu);
9047 kvm_vcpu_reset(vcpu, false);
9048 kvm_init_mmu(vcpu, false);
9053 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9055 struct msr_data msr;
9056 struct kvm *kvm = vcpu->kvm;
9058 kvm_hv_vcpu_postcreate(vcpu);
9060 if (mutex_lock_killable(&vcpu->mutex))
9064 msr.index = MSR_IA32_TSC;
9065 msr.host_initiated = true;
9066 kvm_write_tsc(vcpu, &msr);
9069 /* poll control enabled by default */
9070 vcpu->arch.msr_kvm_poll_control = 1;
9072 mutex_unlock(&vcpu->mutex);
9074 if (!kvmclock_periodic_sync)
9077 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9078 KVMCLOCK_SYNC_PERIOD);
9081 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9083 vcpu->arch.apf.msr_val = 0;
9086 kvm_mmu_unload(vcpu);
9089 kvm_x86_ops->vcpu_free(vcpu);
9092 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9094 kvm_lapic_reset(vcpu, init_event);
9096 vcpu->arch.hflags = 0;
9098 vcpu->arch.smi_pending = 0;
9099 vcpu->arch.smi_count = 0;
9100 atomic_set(&vcpu->arch.nmi_queued, 0);
9101 vcpu->arch.nmi_pending = 0;
9102 vcpu->arch.nmi_injected = false;
9103 kvm_clear_interrupt_queue(vcpu);
9104 kvm_clear_exception_queue(vcpu);
9105 vcpu->arch.exception.pending = false;
9107 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9108 kvm_update_dr0123(vcpu);
9109 vcpu->arch.dr6 = DR6_INIT;
9110 kvm_update_dr6(vcpu);
9111 vcpu->arch.dr7 = DR7_FIXED_1;
9112 kvm_update_dr7(vcpu);
9116 kvm_make_request(KVM_REQ_EVENT, vcpu);
9117 vcpu->arch.apf.msr_val = 0;
9118 vcpu->arch.st.msr_val = 0;
9120 kvmclock_reset(vcpu);
9122 kvm_clear_async_pf_completion_queue(vcpu);
9123 kvm_async_pf_hash_reset(vcpu);
9124 vcpu->arch.apf.halted = false;
9126 if (kvm_mpx_supported()) {
9127 void *mpx_state_buffer;
9130 * To avoid have the INIT path from kvm_apic_has_events() that be
9131 * called with loaded FPU and does not let userspace fix the state.
9134 kvm_put_guest_fpu(vcpu);
9135 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9137 if (mpx_state_buffer)
9138 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9139 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9141 if (mpx_state_buffer)
9142 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9144 kvm_load_guest_fpu(vcpu);
9148 kvm_pmu_reset(vcpu);
9149 vcpu->arch.smbase = 0x30000;
9151 vcpu->arch.msr_misc_features_enables = 0;
9153 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9156 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9157 vcpu->arch.regs_avail = ~0;
9158 vcpu->arch.regs_dirty = ~0;
9160 vcpu->arch.ia32_xss = 0;
9162 kvm_x86_ops->vcpu_reset(vcpu, init_event);
9165 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9167 struct kvm_segment cs;
9169 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9170 cs.selector = vector << 8;
9171 cs.base = vector << 12;
9172 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9173 kvm_rip_write(vcpu, 0);
9176 int kvm_arch_hardware_enable(void)
9179 struct kvm_vcpu *vcpu;
9184 bool stable, backwards_tsc = false;
9186 kvm_shared_msr_cpu_online();
9187 ret = kvm_x86_ops->hardware_enable();
9191 local_tsc = rdtsc();
9192 stable = !kvm_check_tsc_unstable();
9193 list_for_each_entry(kvm, &vm_list, vm_list) {
9194 kvm_for_each_vcpu(i, vcpu, kvm) {
9195 if (!stable && vcpu->cpu == smp_processor_id())
9196 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9197 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9198 backwards_tsc = true;
9199 if (vcpu->arch.last_host_tsc > max_tsc)
9200 max_tsc = vcpu->arch.last_host_tsc;
9206 * Sometimes, even reliable TSCs go backwards. This happens on
9207 * platforms that reset TSC during suspend or hibernate actions, but
9208 * maintain synchronization. We must compensate. Fortunately, we can
9209 * detect that condition here, which happens early in CPU bringup,
9210 * before any KVM threads can be running. Unfortunately, we can't
9211 * bring the TSCs fully up to date with real time, as we aren't yet far
9212 * enough into CPU bringup that we know how much real time has actually
9213 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9214 * variables that haven't been updated yet.
9216 * So we simply find the maximum observed TSC above, then record the
9217 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9218 * the adjustment will be applied. Note that we accumulate
9219 * adjustments, in case multiple suspend cycles happen before some VCPU
9220 * gets a chance to run again. In the event that no KVM threads get a
9221 * chance to run, we will miss the entire elapsed period, as we'll have
9222 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9223 * loose cycle time. This isn't too big a deal, since the loss will be
9224 * uniform across all VCPUs (not to mention the scenario is extremely
9225 * unlikely). It is possible that a second hibernate recovery happens
9226 * much faster than a first, causing the observed TSC here to be
9227 * smaller; this would require additional padding adjustment, which is
9228 * why we set last_host_tsc to the local tsc observed here.
9230 * N.B. - this code below runs only on platforms with reliable TSC,
9231 * as that is the only way backwards_tsc is set above. Also note
9232 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9233 * have the same delta_cyc adjustment applied if backwards_tsc
9234 * is detected. Note further, this adjustment is only done once,
9235 * as we reset last_host_tsc on all VCPUs to stop this from being
9236 * called multiple times (one for each physical CPU bringup).
9238 * Platforms with unreliable TSCs don't have to deal with this, they
9239 * will be compensated by the logic in vcpu_load, which sets the TSC to
9240 * catchup mode. This will catchup all VCPUs to real time, but cannot
9241 * guarantee that they stay in perfect synchronization.
9243 if (backwards_tsc) {
9244 u64 delta_cyc = max_tsc - local_tsc;
9245 list_for_each_entry(kvm, &vm_list, vm_list) {
9246 kvm->arch.backwards_tsc_observed = true;
9247 kvm_for_each_vcpu(i, vcpu, kvm) {
9248 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9249 vcpu->arch.last_host_tsc = local_tsc;
9250 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9254 * We have to disable TSC offset matching.. if you were
9255 * booting a VM while issuing an S4 host suspend....
9256 * you may have some problem. Solving this issue is
9257 * left as an exercise to the reader.
9259 kvm->arch.last_tsc_nsec = 0;
9260 kvm->arch.last_tsc_write = 0;
9267 void kvm_arch_hardware_disable(void)
9269 kvm_x86_ops->hardware_disable();
9270 drop_user_return_notifiers();
9273 int kvm_arch_hardware_setup(void)
9277 r = kvm_x86_ops->hardware_setup();
9281 if (kvm_has_tsc_control) {
9283 * Make sure the user can only configure tsc_khz values that
9284 * fit into a signed integer.
9285 * A min value is not calculated because it will always
9286 * be 1 on all machines.
9288 u64 max = min(0x7fffffffULL,
9289 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9290 kvm_max_guest_tsc_khz = max;
9292 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9295 kvm_init_msr_list();
9299 void kvm_arch_hardware_unsetup(void)
9301 kvm_x86_ops->hardware_unsetup();
9304 int kvm_arch_check_processor_compat(void)
9306 return kvm_x86_ops->check_processor_compatibility();
9309 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9311 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9313 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9315 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9317 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9320 struct static_key kvm_no_apic_vcpu __read_mostly;
9321 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9323 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9328 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9329 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9330 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9332 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9334 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9339 vcpu->arch.pio_data = page_address(page);
9341 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9343 r = kvm_mmu_create(vcpu);
9345 goto fail_free_pio_data;
9347 if (irqchip_in_kernel(vcpu->kvm)) {
9348 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9349 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9351 goto fail_mmu_destroy;
9353 static_key_slow_inc(&kvm_no_apic_vcpu);
9355 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9356 GFP_KERNEL_ACCOUNT);
9357 if (!vcpu->arch.mce_banks) {
9359 goto fail_free_lapic;
9361 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9363 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9364 GFP_KERNEL_ACCOUNT)) {
9366 goto fail_free_mce_banks;
9371 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9373 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9375 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9377 kvm_async_pf_hash_reset(vcpu);
9380 vcpu->arch.pending_external_vector = -1;
9381 vcpu->arch.preempted_in_kernel = false;
9383 kvm_hv_vcpu_init(vcpu);
9387 fail_free_mce_banks:
9388 kfree(vcpu->arch.mce_banks);
9390 kvm_free_lapic(vcpu);
9392 kvm_mmu_destroy(vcpu);
9394 free_page((unsigned long)vcpu->arch.pio_data);
9399 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9403 kvm_hv_vcpu_uninit(vcpu);
9404 kvm_pmu_destroy(vcpu);
9405 kfree(vcpu->arch.mce_banks);
9406 kvm_free_lapic(vcpu);
9407 idx = srcu_read_lock(&vcpu->kvm->srcu);
9408 kvm_mmu_destroy(vcpu);
9409 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9410 free_page((unsigned long)vcpu->arch.pio_data);
9411 if (!lapic_in_kernel(vcpu))
9412 static_key_slow_dec(&kvm_no_apic_vcpu);
9415 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9417 vcpu->arch.l1tf_flush_l1d = true;
9418 kvm_x86_ops->sched_in(vcpu, cpu);
9421 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9426 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9427 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9428 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9429 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9431 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9432 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9433 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9434 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9435 &kvm->arch.irq_sources_bitmap);
9437 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9438 mutex_init(&kvm->arch.apic_map_lock);
9439 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9441 kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9442 pvclock_update_vm_gtod_copy(kvm);
9444 kvm->arch.guest_can_read_msr_platform_info = true;
9446 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9447 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9449 kvm_hv_init_vm(kvm);
9450 kvm_page_track_init(kvm);
9451 kvm_mmu_init_vm(kvm);
9453 return kvm_x86_ops->vm_init(kvm);
9456 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9459 kvm_mmu_unload(vcpu);
9463 static void kvm_free_vcpus(struct kvm *kvm)
9466 struct kvm_vcpu *vcpu;
9469 * Unpin any mmu pages first.
9471 kvm_for_each_vcpu(i, vcpu, kvm) {
9472 kvm_clear_async_pf_completion_queue(vcpu);
9473 kvm_unload_vcpu_mmu(vcpu);
9475 kvm_for_each_vcpu(i, vcpu, kvm)
9476 kvm_arch_vcpu_free(vcpu);
9478 mutex_lock(&kvm->lock);
9479 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9480 kvm->vcpus[i] = NULL;
9482 atomic_set(&kvm->online_vcpus, 0);
9483 mutex_unlock(&kvm->lock);
9486 void kvm_arch_sync_events(struct kvm *kvm)
9488 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9489 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9493 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9497 struct kvm_memslots *slots = kvm_memslots(kvm);
9498 struct kvm_memory_slot *slot, old;
9500 /* Called with kvm->slots_lock held. */
9501 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9504 slot = id_to_memslot(slots, id);
9510 * MAP_SHARED to prevent internal slot pages from being moved
9513 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9514 MAP_SHARED | MAP_ANONYMOUS, 0);
9515 if (IS_ERR((void *)hva))
9516 return PTR_ERR((void *)hva);
9525 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9526 struct kvm_userspace_memory_region m;
9528 m.slot = id | (i << 16);
9530 m.guest_phys_addr = gpa;
9531 m.userspace_addr = hva;
9532 m.memory_size = size;
9533 r = __kvm_set_memory_region(kvm, &m);
9539 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9543 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9545 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9549 mutex_lock(&kvm->slots_lock);
9550 r = __x86_set_memory_region(kvm, id, gpa, size);
9551 mutex_unlock(&kvm->slots_lock);
9555 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9557 void kvm_arch_destroy_vm(struct kvm *kvm)
9559 if (current->mm == kvm->mm) {
9561 * Free memory regions allocated on behalf of userspace,
9562 * unless the the memory map has changed due to process exit
9565 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9566 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9567 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9569 if (kvm_x86_ops->vm_destroy)
9570 kvm_x86_ops->vm_destroy(kvm);
9571 kvm_pic_destroy(kvm);
9572 kvm_ioapic_destroy(kvm);
9573 kvm_free_vcpus(kvm);
9574 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9575 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9576 kvm_mmu_uninit_vm(kvm);
9577 kvm_page_track_cleanup(kvm);
9578 kvm_hv_destroy_vm(kvm);
9581 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9582 struct kvm_memory_slot *dont)
9586 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9587 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9588 kvfree(free->arch.rmap[i]);
9589 free->arch.rmap[i] = NULL;
9594 if (!dont || free->arch.lpage_info[i - 1] !=
9595 dont->arch.lpage_info[i - 1]) {
9596 kvfree(free->arch.lpage_info[i - 1]);
9597 free->arch.lpage_info[i - 1] = NULL;
9601 kvm_page_track_free_memslot(free, dont);
9604 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9605 unsigned long npages)
9609 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9610 struct kvm_lpage_info *linfo;
9615 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9616 slot->base_gfn, level) + 1;
9618 slot->arch.rmap[i] =
9619 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9620 GFP_KERNEL_ACCOUNT);
9621 if (!slot->arch.rmap[i])
9626 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9630 slot->arch.lpage_info[i - 1] = linfo;
9632 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9633 linfo[0].disallow_lpage = 1;
9634 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9635 linfo[lpages - 1].disallow_lpage = 1;
9636 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9638 * If the gfn and userspace address are not aligned wrt each
9639 * other, or if explicitly asked to, disable large page
9640 * support for this slot
9642 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9643 !kvm_largepages_enabled()) {
9646 for (j = 0; j < lpages; ++j)
9647 linfo[j].disallow_lpage = 1;
9651 if (kvm_page_track_create_memslot(slot, npages))
9657 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9658 kvfree(slot->arch.rmap[i]);
9659 slot->arch.rmap[i] = NULL;
9663 kvfree(slot->arch.lpage_info[i - 1]);
9664 slot->arch.lpage_info[i - 1] = NULL;
9669 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9672 * memslots->generation has been incremented.
9673 * mmio generation may have reached its maximum value.
9675 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9678 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9679 struct kvm_memory_slot *memslot,
9680 const struct kvm_userspace_memory_region *mem,
9681 enum kvm_mr_change change)
9686 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9687 struct kvm_memory_slot *new)
9689 /* Still write protect RO slot */
9690 if (new->flags & KVM_MEM_READONLY) {
9691 kvm_mmu_slot_remove_write_access(kvm, new);
9696 * Call kvm_x86_ops dirty logging hooks when they are valid.
9698 * kvm_x86_ops->slot_disable_log_dirty is called when:
9700 * - KVM_MR_CREATE with dirty logging is disabled
9701 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9703 * The reason is, in case of PML, we need to set D-bit for any slots
9704 * with dirty logging disabled in order to eliminate unnecessary GPA
9705 * logging in PML buffer (and potential PML buffer full VMEXT). This
9706 * guarantees leaving PML enabled during guest's lifetime won't have
9707 * any additional overhead from PML when guest is running with dirty
9708 * logging disabled for memory slots.
9710 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9711 * to dirty logging mode.
9713 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9715 * In case of write protect:
9717 * Write protect all pages for dirty logging.
9719 * All the sptes including the large sptes which point to this
9720 * slot are set to readonly. We can not create any new large
9721 * spte on this slot until the end of the logging.
9723 * See the comments in fast_page_fault().
9725 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9726 if (kvm_x86_ops->slot_enable_log_dirty)
9727 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9729 kvm_mmu_slot_remove_write_access(kvm, new);
9731 if (kvm_x86_ops->slot_disable_log_dirty)
9732 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9736 void kvm_arch_commit_memory_region(struct kvm *kvm,
9737 const struct kvm_userspace_memory_region *mem,
9738 const struct kvm_memory_slot *old,
9739 const struct kvm_memory_slot *new,
9740 enum kvm_mr_change change)
9742 if (!kvm->arch.n_requested_mmu_pages)
9743 kvm_mmu_change_mmu_pages(kvm,
9744 kvm_mmu_calculate_default_mmu_pages(kvm));
9747 * Dirty logging tracks sptes in 4k granularity, meaning that large
9748 * sptes have to be split. If live migration is successful, the guest
9749 * in the source machine will be destroyed and large sptes will be
9750 * created in the destination. However, if the guest continues to run
9751 * in the source machine (for example if live migration fails), small
9752 * sptes will remain around and cause bad performance.
9754 * Scan sptes if dirty logging has been stopped, dropping those
9755 * which can be collapsed into a single large-page spte. Later
9756 * page faults will create the large-page sptes.
9758 * There is no need to do this in any of the following cases:
9759 * CREATE: No dirty mappings will already exist.
9760 * MOVE/DELETE: The old mappings will already have been cleaned up by
9761 * kvm_arch_flush_shadow_memslot()
9763 if (change == KVM_MR_FLAGS_ONLY &&
9764 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9765 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9766 kvm_mmu_zap_collapsible_sptes(kvm, new);
9769 * Set up write protection and/or dirty logging for the new slot.
9771 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9772 * been zapped so no dirty logging staff is needed for old slot. For
9773 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9774 * new and it's also covered when dealing with the new slot.
9776 * FIXME: const-ify all uses of struct kvm_memory_slot.
9778 if (change != KVM_MR_DELETE)
9779 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9782 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9784 kvm_mmu_zap_all(kvm);
9787 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9788 struct kvm_memory_slot *slot)
9790 kvm_page_track_flush_slot(kvm, slot);
9793 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9795 return (is_guest_mode(vcpu) &&
9796 kvm_x86_ops->guest_apic_has_interrupt &&
9797 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9800 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9802 if (!list_empty_careful(&vcpu->async_pf.done))
9805 if (kvm_apic_has_events(vcpu))
9808 if (vcpu->arch.pv.pv_unhalted)
9811 if (vcpu->arch.exception.pending)
9814 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9815 (vcpu->arch.nmi_pending &&
9816 kvm_x86_ops->nmi_allowed(vcpu)))
9819 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9820 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9823 if (kvm_arch_interrupt_allowed(vcpu) &&
9824 (kvm_cpu_has_interrupt(vcpu) ||
9825 kvm_guest_apic_has_interrupt(vcpu)))
9828 if (kvm_hv_has_stimer_pending(vcpu))
9834 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9836 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9839 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
9841 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
9844 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9845 kvm_test_request(KVM_REQ_SMI, vcpu) ||
9846 kvm_test_request(KVM_REQ_EVENT, vcpu))
9849 if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
9855 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9857 return vcpu->arch.preempted_in_kernel;
9860 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9862 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9865 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9867 return kvm_x86_ops->interrupt_allowed(vcpu);
9870 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9872 if (is_64_bit_mode(vcpu))
9873 return kvm_rip_read(vcpu);
9874 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9875 kvm_rip_read(vcpu));
9877 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9879 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9881 return kvm_get_linear_rip(vcpu) == linear_rip;
9883 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9885 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9887 unsigned long rflags;
9889 rflags = kvm_x86_ops->get_rflags(vcpu);
9890 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9891 rflags &= ~X86_EFLAGS_TF;
9894 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9896 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9898 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9899 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9900 rflags |= X86_EFLAGS_TF;
9901 kvm_x86_ops->set_rflags(vcpu, rflags);
9904 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9906 __kvm_set_rflags(vcpu, rflags);
9907 kvm_make_request(KVM_REQ_EVENT, vcpu);
9909 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9911 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9915 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9919 r = kvm_mmu_reload(vcpu);
9923 if (!vcpu->arch.mmu->direct_map &&
9924 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9927 vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9930 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9932 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9935 static inline u32 kvm_async_pf_next_probe(u32 key)
9937 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9940 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9942 u32 key = kvm_async_pf_hash_fn(gfn);
9944 while (vcpu->arch.apf.gfns[key] != ~0)
9945 key = kvm_async_pf_next_probe(key);
9947 vcpu->arch.apf.gfns[key] = gfn;
9950 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9953 u32 key = kvm_async_pf_hash_fn(gfn);
9955 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9956 (vcpu->arch.apf.gfns[key] != gfn &&
9957 vcpu->arch.apf.gfns[key] != ~0); i++)
9958 key = kvm_async_pf_next_probe(key);
9963 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9965 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9968 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9972 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9974 vcpu->arch.apf.gfns[i] = ~0;
9976 j = kvm_async_pf_next_probe(j);
9977 if (vcpu->arch.apf.gfns[j] == ~0)
9979 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9981 * k lies cyclically in ]i,j]
9983 * |....j i.k.| or |.k..j i...|
9985 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9986 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9991 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9994 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9998 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10001 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10005 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10007 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10010 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10011 (vcpu->arch.apf.send_user_only &&
10012 kvm_x86_ops->get_cpl(vcpu) == 0))
10018 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10020 if (unlikely(!lapic_in_kernel(vcpu) ||
10021 kvm_event_needs_reinjection(vcpu) ||
10022 vcpu->arch.exception.pending))
10025 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10029 * If interrupts are off we cannot even use an artificial
10032 return kvm_x86_ops->interrupt_allowed(vcpu);
10035 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10036 struct kvm_async_pf *work)
10038 struct x86_exception fault;
10040 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
10041 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10043 if (kvm_can_deliver_async_pf(vcpu) &&
10044 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10045 fault.vector = PF_VECTOR;
10046 fault.error_code_valid = true;
10047 fault.error_code = 0;
10048 fault.nested_page_fault = false;
10049 fault.address = work->arch.token;
10050 fault.async_page_fault = true;
10051 kvm_inject_page_fault(vcpu, &fault);
10054 * It is not possible to deliver a paravirtualized asynchronous
10055 * page fault, but putting the guest in an artificial halt state
10056 * can be beneficial nevertheless: if an interrupt arrives, we
10057 * can deliver it timely and perhaps the guest will schedule
10058 * another process. When the instruction that triggered a page
10059 * fault is retried, hopefully the page will be ready in the host.
10061 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10065 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10066 struct kvm_async_pf *work)
10068 struct x86_exception fault;
10071 if (work->wakeup_all)
10072 work->arch.token = ~0; /* broadcast wakeup */
10074 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10075 trace_kvm_async_pf_ready(work->arch.token, work->gva);
10077 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10078 !apf_get_user(vcpu, &val)) {
10079 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10080 vcpu->arch.exception.pending &&
10081 vcpu->arch.exception.nr == PF_VECTOR &&
10082 !apf_put_user(vcpu, 0)) {
10083 vcpu->arch.exception.injected = false;
10084 vcpu->arch.exception.pending = false;
10085 vcpu->arch.exception.nr = 0;
10086 vcpu->arch.exception.has_error_code = false;
10087 vcpu->arch.exception.error_code = 0;
10088 vcpu->arch.exception.has_payload = false;
10089 vcpu->arch.exception.payload = 0;
10090 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10091 fault.vector = PF_VECTOR;
10092 fault.error_code_valid = true;
10093 fault.error_code = 0;
10094 fault.nested_page_fault = false;
10095 fault.address = work->arch.token;
10096 fault.async_page_fault = true;
10097 kvm_inject_page_fault(vcpu, &fault);
10100 vcpu->arch.apf.halted = false;
10101 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10104 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10106 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10109 return kvm_can_do_async_pf(vcpu);
10112 void kvm_arch_start_assignment(struct kvm *kvm)
10114 atomic_inc(&kvm->arch.assigned_device_count);
10116 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10118 void kvm_arch_end_assignment(struct kvm *kvm)
10120 atomic_dec(&kvm->arch.assigned_device_count);
10122 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10124 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10126 return atomic_read(&kvm->arch.assigned_device_count);
10128 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10130 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10132 atomic_inc(&kvm->arch.noncoherent_dma_count);
10134 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10136 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10138 atomic_dec(&kvm->arch.noncoherent_dma_count);
10140 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10142 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10144 return atomic_read(&kvm->arch.noncoherent_dma_count);
10146 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10148 bool kvm_arch_has_irq_bypass(void)
10153 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10154 struct irq_bypass_producer *prod)
10156 struct kvm_kernel_irqfd *irqfd =
10157 container_of(cons, struct kvm_kernel_irqfd, consumer);
10159 irqfd->producer = prod;
10161 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10162 prod->irq, irqfd->gsi, 1);
10165 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10166 struct irq_bypass_producer *prod)
10169 struct kvm_kernel_irqfd *irqfd =
10170 container_of(cons, struct kvm_kernel_irqfd, consumer);
10172 WARN_ON(irqfd->producer != prod);
10173 irqfd->producer = NULL;
10176 * When producer of consumer is unregistered, we change back to
10177 * remapped mode, so we can re-use the current implementation
10178 * when the irq is masked/disabled or the consumer side (KVM
10179 * int this case doesn't want to receive the interrupts.
10181 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10183 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10184 " fails: %d\n", irqfd->consumer.token, ret);
10187 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10188 uint32_t guest_irq, bool set)
10190 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10193 bool kvm_vector_hashing_enabled(void)
10195 return vector_hashing;
10197 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10199 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10201 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10203 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10206 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10207 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10208 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10209 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10210 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10211 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10212 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10213 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10214 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10215 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10216 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10217 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10218 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10219 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10220 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10221 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10222 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10223 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10224 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10225 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);