2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/mem_encrypt.h>
59 #include <trace/events/kvm.h>
61 #include <asm/debugreg.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.h>
72 #include <asm/intel_pt.h>
74 #define CREATE_TRACE_POINTS
77 #define MAX_IO_MSRS 256
78 #define KVM_MAX_MCE_BANKS 32
79 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
80 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
82 #define emul_to_vcpu(ctxt) \
83 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
86 * - enable syscall per default because its emulated by KVM
87 * - enable LME and LMA per default on 64 bit KVM
91 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
93 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
96 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
97 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
99 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
100 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
102 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
103 static void process_nmi(struct kvm_vcpu *vcpu);
104 static void enter_smm(struct kvm_vcpu *vcpu);
105 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
106 static void store_regs(struct kvm_vcpu *vcpu);
107 static int sync_regs(struct kvm_vcpu *vcpu);
109 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
110 EXPORT_SYMBOL_GPL(kvm_x86_ops);
112 static bool __read_mostly ignore_msrs = 0;
113 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
115 static bool __read_mostly report_ignored_msrs = true;
116 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
118 unsigned int min_timer_period_us = 200;
119 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
121 static bool __read_mostly kvmclock_periodic_sync = true;
122 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
124 bool __read_mostly kvm_has_tsc_control;
125 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
126 u32 __read_mostly kvm_max_guest_tsc_khz;
127 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
128 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
129 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
130 u64 __read_mostly kvm_max_tsc_scaling_ratio;
131 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
132 u64 __read_mostly kvm_default_tsc_scaling_ratio;
133 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
135 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
136 static u32 __read_mostly tsc_tolerance_ppm = 250;
137 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
140 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
141 * adaptive tuning starting from default advancment of 1000ns. '0' disables
142 * advancement entirely. Any other value is used as-is and disables adaptive
143 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
145 static int __read_mostly lapic_timer_advance_ns = -1;
146 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
148 static bool __read_mostly vector_hashing = true;
149 module_param(vector_hashing, bool, S_IRUGO);
151 bool __read_mostly enable_vmware_backdoor = false;
152 module_param(enable_vmware_backdoor, bool, S_IRUGO);
153 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
155 static bool __read_mostly force_emulation_prefix = false;
156 module_param(force_emulation_prefix, bool, S_IRUGO);
158 #define KVM_NR_SHARED_MSRS 16
160 struct kvm_shared_msrs_global {
162 u32 msrs[KVM_NR_SHARED_MSRS];
165 struct kvm_shared_msrs {
166 struct user_return_notifier urn;
168 struct kvm_shared_msr_values {
171 } values[KVM_NR_SHARED_MSRS];
174 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
175 static struct kvm_shared_msrs __percpu *shared_msrs;
177 struct kvm_stats_debugfs_item debugfs_entries[] = {
178 { "pf_fixed", VCPU_STAT(pf_fixed) },
179 { "pf_guest", VCPU_STAT(pf_guest) },
180 { "tlb_flush", VCPU_STAT(tlb_flush) },
181 { "invlpg", VCPU_STAT(invlpg) },
182 { "exits", VCPU_STAT(exits) },
183 { "io_exits", VCPU_STAT(io_exits) },
184 { "mmio_exits", VCPU_STAT(mmio_exits) },
185 { "signal_exits", VCPU_STAT(signal_exits) },
186 { "irq_window", VCPU_STAT(irq_window_exits) },
187 { "nmi_window", VCPU_STAT(nmi_window_exits) },
188 { "halt_exits", VCPU_STAT(halt_exits) },
189 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
190 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
191 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
192 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
193 { "hypercalls", VCPU_STAT(hypercalls) },
194 { "request_irq", VCPU_STAT(request_irq_exits) },
195 { "irq_exits", VCPU_STAT(irq_exits) },
196 { "host_state_reload", VCPU_STAT(host_state_reload) },
197 { "fpu_reload", VCPU_STAT(fpu_reload) },
198 { "insn_emulation", VCPU_STAT(insn_emulation) },
199 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
200 { "irq_injections", VCPU_STAT(irq_injections) },
201 { "nmi_injections", VCPU_STAT(nmi_injections) },
202 { "req_event", VCPU_STAT(req_event) },
203 { "l1d_flush", VCPU_STAT(l1d_flush) },
204 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
205 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
206 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
207 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
208 { "mmu_flooded", VM_STAT(mmu_flooded) },
209 { "mmu_recycled", VM_STAT(mmu_recycled) },
210 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
211 { "mmu_unsync", VM_STAT(mmu_unsync) },
212 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
213 { "largepages", VM_STAT(lpages) },
214 { "max_mmu_page_hash_collisions",
215 VM_STAT(max_mmu_page_hash_collisions) },
219 u64 __read_mostly host_xcr0;
221 struct kmem_cache *x86_fpu_cache;
222 EXPORT_SYMBOL_GPL(x86_fpu_cache);
224 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
226 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
229 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
230 vcpu->arch.apf.gfns[i] = ~0;
233 static void kvm_on_user_return(struct user_return_notifier *urn)
236 struct kvm_shared_msrs *locals
237 = container_of(urn, struct kvm_shared_msrs, urn);
238 struct kvm_shared_msr_values *values;
242 * Disabling irqs at this point since the following code could be
243 * interrupted and executed through kvm_arch_hardware_disable()
245 local_irq_save(flags);
246 if (locals->registered) {
247 locals->registered = false;
248 user_return_notifier_unregister(urn);
250 local_irq_restore(flags);
251 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
252 values = &locals->values[slot];
253 if (values->host != values->curr) {
254 wrmsrl(shared_msrs_global.msrs[slot], values->host);
255 values->curr = values->host;
260 static void shared_msr_update(unsigned slot, u32 msr)
263 unsigned int cpu = smp_processor_id();
264 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
266 /* only read, and nobody should modify it at this time,
267 * so don't need lock */
268 if (slot >= shared_msrs_global.nr) {
269 printk(KERN_ERR "kvm: invalid MSR slot!");
272 rdmsrl_safe(msr, &value);
273 smsr->values[slot].host = value;
274 smsr->values[slot].curr = value;
277 void kvm_define_shared_msr(unsigned slot, u32 msr)
279 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
280 shared_msrs_global.msrs[slot] = msr;
281 if (slot >= shared_msrs_global.nr)
282 shared_msrs_global.nr = slot + 1;
284 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
286 static void kvm_shared_msr_cpu_online(void)
290 for (i = 0; i < shared_msrs_global.nr; ++i)
291 shared_msr_update(i, shared_msrs_global.msrs[i]);
294 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
296 unsigned int cpu = smp_processor_id();
297 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
300 if (((value ^ smsr->values[slot].curr) & mask) == 0)
302 smsr->values[slot].curr = value;
303 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
307 if (!smsr->registered) {
308 smsr->urn.on_user_return = kvm_on_user_return;
309 user_return_notifier_register(&smsr->urn);
310 smsr->registered = true;
314 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
316 static void drop_user_return_notifiers(void)
318 unsigned int cpu = smp_processor_id();
319 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
321 if (smsr->registered)
322 kvm_on_user_return(&smsr->urn);
325 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
327 return vcpu->arch.apic_base;
329 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
331 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
333 return kvm_apic_mode(kvm_get_apic_base(vcpu));
335 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
337 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
339 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
340 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
341 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
342 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
344 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
346 if (!msr_info->host_initiated) {
347 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
349 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
353 kvm_lapic_set_base(vcpu, msr_info->data);
356 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
358 asmlinkage __visible void kvm_spurious_fault(void)
360 /* Fault while not rebooting. We want the trace. */
363 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
365 #define EXCPT_BENIGN 0
366 #define EXCPT_CONTRIBUTORY 1
369 static int exception_class(int vector)
379 return EXCPT_CONTRIBUTORY;
386 #define EXCPT_FAULT 0
388 #define EXCPT_ABORT 2
389 #define EXCPT_INTERRUPT 3
391 static int exception_type(int vector)
395 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
396 return EXCPT_INTERRUPT;
400 /* #DB is trap, as instruction watchpoints are handled elsewhere */
401 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
404 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
407 /* Reserved exceptions will result in fault */
411 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
413 unsigned nr = vcpu->arch.exception.nr;
414 bool has_payload = vcpu->arch.exception.has_payload;
415 unsigned long payload = vcpu->arch.exception.payload;
423 * "Certain debug exceptions may clear bit 0-3. The
424 * remaining contents of the DR6 register are never
425 * cleared by the processor".
427 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
429 * DR6.RTM is set by all #DB exceptions that don't clear it.
431 vcpu->arch.dr6 |= DR6_RTM;
432 vcpu->arch.dr6 |= payload;
434 * Bit 16 should be set in the payload whenever the #DB
435 * exception should clear DR6.RTM. This makes the payload
436 * compatible with the pending debug exceptions under VMX.
437 * Though not currently documented in the SDM, this also
438 * makes the payload compatible with the exit qualification
439 * for #DB exceptions under VMX.
441 vcpu->arch.dr6 ^= payload & DR6_RTM;
444 vcpu->arch.cr2 = payload;
448 vcpu->arch.exception.has_payload = false;
449 vcpu->arch.exception.payload = 0;
451 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
453 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
454 unsigned nr, bool has_error, u32 error_code,
455 bool has_payload, unsigned long payload, bool reinject)
460 kvm_make_request(KVM_REQ_EVENT, vcpu);
462 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
464 if (has_error && !is_protmode(vcpu))
468 * On vmentry, vcpu->arch.exception.pending is only
469 * true if an event injection was blocked by
470 * nested_run_pending. In that case, however,
471 * vcpu_enter_guest requests an immediate exit,
472 * and the guest shouldn't proceed far enough to
475 WARN_ON_ONCE(vcpu->arch.exception.pending);
476 vcpu->arch.exception.injected = true;
477 if (WARN_ON_ONCE(has_payload)) {
479 * A reinjected event has already
480 * delivered its payload.
486 vcpu->arch.exception.pending = true;
487 vcpu->arch.exception.injected = false;
489 vcpu->arch.exception.has_error_code = has_error;
490 vcpu->arch.exception.nr = nr;
491 vcpu->arch.exception.error_code = error_code;
492 vcpu->arch.exception.has_payload = has_payload;
493 vcpu->arch.exception.payload = payload;
495 * In guest mode, payload delivery should be deferred,
496 * so that the L1 hypervisor can intercept #PF before
497 * CR2 is modified (or intercept #DB before DR6 is
498 * modified under nVMX). However, for ABI
499 * compatibility with KVM_GET_VCPU_EVENTS and
500 * KVM_SET_VCPU_EVENTS, we can't delay payload
501 * delivery unless userspace has enabled this
502 * functionality via the per-VM capability,
503 * KVM_CAP_EXCEPTION_PAYLOAD.
505 if (!vcpu->kvm->arch.exception_payload_enabled ||
506 !is_guest_mode(vcpu))
507 kvm_deliver_exception_payload(vcpu);
511 /* to check exception */
512 prev_nr = vcpu->arch.exception.nr;
513 if (prev_nr == DF_VECTOR) {
514 /* triple fault -> shutdown */
515 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
518 class1 = exception_class(prev_nr);
519 class2 = exception_class(nr);
520 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
521 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
523 * Generate double fault per SDM Table 5-5. Set
524 * exception.pending = true so that the double fault
525 * can trigger a nested vmexit.
527 vcpu->arch.exception.pending = true;
528 vcpu->arch.exception.injected = false;
529 vcpu->arch.exception.has_error_code = true;
530 vcpu->arch.exception.nr = DF_VECTOR;
531 vcpu->arch.exception.error_code = 0;
532 vcpu->arch.exception.has_payload = false;
533 vcpu->arch.exception.payload = 0;
535 /* replace previous exception with a new one in a hope
536 that instruction re-execution will regenerate lost
541 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
543 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
545 EXPORT_SYMBOL_GPL(kvm_queue_exception);
547 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
549 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
551 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
553 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
554 unsigned long payload)
556 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
559 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
560 u32 error_code, unsigned long payload)
562 kvm_multiple_exception(vcpu, nr, true, error_code,
563 true, payload, false);
566 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
569 kvm_inject_gp(vcpu, 0);
571 return kvm_skip_emulated_instruction(vcpu);
575 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
577 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
579 ++vcpu->stat.pf_guest;
580 vcpu->arch.exception.nested_apf =
581 is_guest_mode(vcpu) && fault->async_page_fault;
582 if (vcpu->arch.exception.nested_apf) {
583 vcpu->arch.apf.nested_apf_token = fault->address;
584 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
586 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
590 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
592 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
594 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
595 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
597 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
599 return fault->nested_page_fault;
602 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
604 atomic_inc(&vcpu->arch.nmi_queued);
605 kvm_make_request(KVM_REQ_NMI, vcpu);
607 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
609 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
611 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
613 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
615 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
617 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
619 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
622 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
623 * a #GP and return false.
625 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
627 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
629 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
632 EXPORT_SYMBOL_GPL(kvm_require_cpl);
634 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
636 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
639 kvm_queue_exception(vcpu, UD_VECTOR);
642 EXPORT_SYMBOL_GPL(kvm_require_dr);
645 * This function will be used to read from the physical memory of the currently
646 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
647 * can read from guest physical or from the guest's guest physical memory.
649 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
650 gfn_t ngfn, void *data, int offset, int len,
653 struct x86_exception exception;
657 ngpa = gfn_to_gpa(ngfn);
658 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
659 if (real_gfn == UNMAPPED_GVA)
662 real_gfn = gpa_to_gfn(real_gfn);
664 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
666 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
668 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
669 void *data, int offset, int len, u32 access)
671 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
672 data, offset, len, access);
676 * Load the pae pdptrs. Return true is they are all valid.
678 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
680 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
681 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
684 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
686 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
687 offset * sizeof(u64), sizeof(pdpte),
688 PFERR_USER_MASK|PFERR_WRITE_MASK);
693 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
694 if ((pdpte[i] & PT_PRESENT_MASK) &&
696 vcpu->arch.mmu->guest_rsvd_check.rsvd_bits_mask[0][2])) {
703 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
704 __set_bit(VCPU_EXREG_PDPTR,
705 (unsigned long *)&vcpu->arch.regs_avail);
706 __set_bit(VCPU_EXREG_PDPTR,
707 (unsigned long *)&vcpu->arch.regs_dirty);
712 EXPORT_SYMBOL_GPL(load_pdptrs);
714 bool pdptrs_changed(struct kvm_vcpu *vcpu)
716 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
722 if (!is_pae_paging(vcpu))
725 if (!test_bit(VCPU_EXREG_PDPTR,
726 (unsigned long *)&vcpu->arch.regs_avail))
729 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
730 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
731 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
732 PFERR_USER_MASK | PFERR_WRITE_MASK);
735 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
740 EXPORT_SYMBOL_GPL(pdptrs_changed);
742 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
744 unsigned long old_cr0 = kvm_read_cr0(vcpu);
745 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
750 if (cr0 & 0xffffffff00000000UL)
754 cr0 &= ~CR0_RESERVED_BITS;
756 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
759 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
762 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
764 if ((vcpu->arch.efer & EFER_LME)) {
769 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
774 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
779 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
782 kvm_x86_ops->set_cr0(vcpu, cr0);
784 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
785 kvm_clear_async_pf_completion_queue(vcpu);
786 kvm_async_pf_hash_reset(vcpu);
789 if ((cr0 ^ old_cr0) & update_bits)
790 kvm_mmu_reset_context(vcpu);
792 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
793 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
794 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
795 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
799 EXPORT_SYMBOL_GPL(kvm_set_cr0);
801 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
803 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
805 EXPORT_SYMBOL_GPL(kvm_lmsw);
807 void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
809 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
810 !vcpu->guest_xcr0_loaded) {
811 /* kvm_set_xcr() also depends on this */
812 if (vcpu->arch.xcr0 != host_xcr0)
813 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
814 vcpu->guest_xcr0_loaded = 1;
817 EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
819 void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
821 if (vcpu->guest_xcr0_loaded) {
822 if (vcpu->arch.xcr0 != host_xcr0)
823 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
824 vcpu->guest_xcr0_loaded = 0;
827 EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
829 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
832 u64 old_xcr0 = vcpu->arch.xcr0;
835 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
836 if (index != XCR_XFEATURE_ENABLED_MASK)
838 if (!(xcr0 & XFEATURE_MASK_FP))
840 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
844 * Do not allow the guest to set bits that we do not support
845 * saving. However, xcr0 bit 0 is always set, even if the
846 * emulated CPU does not support XSAVE (see fx_init).
848 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
849 if (xcr0 & ~valid_bits)
852 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
853 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
856 if (xcr0 & XFEATURE_MASK_AVX512) {
857 if (!(xcr0 & XFEATURE_MASK_YMM))
859 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
862 vcpu->arch.xcr0 = xcr0;
864 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
865 kvm_update_cpuid(vcpu);
869 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
871 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
872 __kvm_set_xcr(vcpu, index, xcr)) {
873 kvm_inject_gp(vcpu, 0);
878 EXPORT_SYMBOL_GPL(kvm_set_xcr);
880 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
882 unsigned long old_cr4 = kvm_read_cr4(vcpu);
883 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
884 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
886 if (cr4 & CR4_RESERVED_BITS)
889 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
892 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
895 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
898 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
901 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
904 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
907 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
910 if (is_long_mode(vcpu)) {
911 if (!(cr4 & X86_CR4_PAE))
913 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
914 && ((cr4 ^ old_cr4) & pdptr_bits)
915 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
919 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
920 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
923 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
924 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
928 if (kvm_x86_ops->set_cr4(vcpu, cr4))
931 if (((cr4 ^ old_cr4) & pdptr_bits) ||
932 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
933 kvm_mmu_reset_context(vcpu);
935 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
936 kvm_update_cpuid(vcpu);
940 EXPORT_SYMBOL_GPL(kvm_set_cr4);
942 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
944 bool skip_tlb_flush = false;
946 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
949 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
950 cr3 &= ~X86_CR3_PCID_NOFLUSH;
954 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
955 if (!skip_tlb_flush) {
956 kvm_mmu_sync_roots(vcpu);
957 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
962 if (is_long_mode(vcpu) &&
963 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
965 else if (is_pae_paging(vcpu) &&
966 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
969 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
970 vcpu->arch.cr3 = cr3;
971 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
975 EXPORT_SYMBOL_GPL(kvm_set_cr3);
977 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
979 if (cr8 & CR8_RESERVED_BITS)
981 if (lapic_in_kernel(vcpu))
982 kvm_lapic_set_tpr(vcpu, cr8);
984 vcpu->arch.cr8 = cr8;
987 EXPORT_SYMBOL_GPL(kvm_set_cr8);
989 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
991 if (lapic_in_kernel(vcpu))
992 return kvm_lapic_get_cr8(vcpu);
994 return vcpu->arch.cr8;
996 EXPORT_SYMBOL_GPL(kvm_get_cr8);
998 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1002 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1003 for (i = 0; i < KVM_NR_DB_REGS; i++)
1004 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1005 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1009 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1011 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1012 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1015 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1019 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1020 dr7 = vcpu->arch.guest_debug_dr7;
1022 dr7 = vcpu->arch.dr7;
1023 kvm_x86_ops->set_dr7(vcpu, dr7);
1024 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1025 if (dr7 & DR7_BP_EN_MASK)
1026 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1029 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1031 u64 fixed = DR6_FIXED_1;
1033 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1038 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1042 vcpu->arch.db[dr] = val;
1043 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1044 vcpu->arch.eff_db[dr] = val;
1049 if (val & 0xffffffff00000000ULL)
1050 return -1; /* #GP */
1051 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1052 kvm_update_dr6(vcpu);
1057 if (val & 0xffffffff00000000ULL)
1058 return -1; /* #GP */
1059 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1060 kvm_update_dr7(vcpu);
1067 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1069 if (__kvm_set_dr(vcpu, dr, val)) {
1070 kvm_inject_gp(vcpu, 0);
1075 EXPORT_SYMBOL_GPL(kvm_set_dr);
1077 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1081 *val = vcpu->arch.db[dr];
1086 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1087 *val = vcpu->arch.dr6;
1089 *val = kvm_x86_ops->get_dr6(vcpu);
1094 *val = vcpu->arch.dr7;
1099 EXPORT_SYMBOL_GPL(kvm_get_dr);
1101 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1103 u32 ecx = kvm_rcx_read(vcpu);
1107 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1110 kvm_rax_write(vcpu, (u32)data);
1111 kvm_rdx_write(vcpu, data >> 32);
1114 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1117 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1118 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1120 * This list is modified at module load time to reflect the
1121 * capabilities of the host cpu. This capabilities test skips MSRs that are
1122 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1123 * may depend on host virtualization features rather than host cpu features.
1126 static u32 msrs_to_save[] = {
1127 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1129 #ifdef CONFIG_X86_64
1130 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1132 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1133 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1135 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1136 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1137 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1138 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1139 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1140 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1143 static unsigned num_msrs_to_save;
1145 static u32 emulated_msrs[] = {
1146 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1147 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1148 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1149 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1150 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1151 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1152 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1154 HV_X64_MSR_VP_INDEX,
1155 HV_X64_MSR_VP_RUNTIME,
1156 HV_X64_MSR_SCONTROL,
1157 HV_X64_MSR_STIMER0_CONFIG,
1158 HV_X64_MSR_VP_ASSIST_PAGE,
1159 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1160 HV_X64_MSR_TSC_EMULATION_STATUS,
1162 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1165 MSR_IA32_TSC_ADJUST,
1166 MSR_IA32_TSCDEADLINE,
1167 MSR_IA32_ARCH_CAPABILITIES,
1168 MSR_IA32_MISC_ENABLE,
1169 MSR_IA32_MCG_STATUS,
1171 MSR_IA32_MCG_EXT_CTL,
1175 MSR_MISC_FEATURES_ENABLES,
1176 MSR_AMD64_VIRT_SPEC_CTRL,
1180 * The following list leaves out MSRs whose values are determined
1181 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1182 * We always support the "true" VMX control MSRs, even if the host
1183 * processor does not, so I am putting these registers here rather
1184 * than in msrs_to_save.
1187 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1188 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1189 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1190 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1192 MSR_IA32_VMX_CR0_FIXED0,
1193 MSR_IA32_VMX_CR4_FIXED0,
1194 MSR_IA32_VMX_VMCS_ENUM,
1195 MSR_IA32_VMX_PROCBASED_CTLS2,
1196 MSR_IA32_VMX_EPT_VPID_CAP,
1197 MSR_IA32_VMX_VMFUNC,
1200 MSR_KVM_POLL_CONTROL,
1203 static unsigned num_emulated_msrs;
1206 * List of msr numbers which are used to expose MSR-based features that
1207 * can be used by a hypervisor to validate requested CPU features.
1209 static u32 msr_based_features[] = {
1211 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1212 MSR_IA32_VMX_PINBASED_CTLS,
1213 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1214 MSR_IA32_VMX_PROCBASED_CTLS,
1215 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1216 MSR_IA32_VMX_EXIT_CTLS,
1217 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1218 MSR_IA32_VMX_ENTRY_CTLS,
1220 MSR_IA32_VMX_CR0_FIXED0,
1221 MSR_IA32_VMX_CR0_FIXED1,
1222 MSR_IA32_VMX_CR4_FIXED0,
1223 MSR_IA32_VMX_CR4_FIXED1,
1224 MSR_IA32_VMX_VMCS_ENUM,
1225 MSR_IA32_VMX_PROCBASED_CTLS2,
1226 MSR_IA32_VMX_EPT_VPID_CAP,
1227 MSR_IA32_VMX_VMFUNC,
1231 MSR_IA32_ARCH_CAPABILITIES,
1234 static unsigned int num_msr_based_features;
1236 static u64 kvm_get_arch_capabilities(void)
1240 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1241 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1244 * If we're doing cache flushes (either "always" or "cond")
1245 * we will do one whenever the guest does a vmlaunch/vmresume.
1246 * If an outer hypervisor is doing the cache flush for us
1247 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1248 * capability to the guest too, and if EPT is disabled we're not
1249 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1250 * require a nested hypervisor to do a flush of its own.
1252 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1253 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1258 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1260 switch (msr->index) {
1261 case MSR_IA32_ARCH_CAPABILITIES:
1262 msr->data = kvm_get_arch_capabilities();
1264 case MSR_IA32_UCODE_REV:
1265 rdmsrl_safe(msr->index, &msr->data);
1268 if (kvm_x86_ops->get_msr_feature(msr))
1274 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1276 struct kvm_msr_entry msr;
1280 r = kvm_get_msr_feature(&msr);
1289 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1291 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1294 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1297 if (efer & (EFER_LME | EFER_LMA) &&
1298 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1301 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1307 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1309 if (efer & efer_reserved_bits)
1312 return __kvm_valid_efer(vcpu, efer);
1314 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1316 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1318 u64 old_efer = vcpu->arch.efer;
1319 u64 efer = msr_info->data;
1321 if (efer & efer_reserved_bits)
1324 if (!msr_info->host_initiated) {
1325 if (!__kvm_valid_efer(vcpu, efer))
1328 if (is_paging(vcpu) &&
1329 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1334 efer |= vcpu->arch.efer & EFER_LMA;
1336 kvm_x86_ops->set_efer(vcpu, efer);
1338 /* Update reserved bits */
1339 if ((efer ^ old_efer) & EFER_NX)
1340 kvm_mmu_reset_context(vcpu);
1345 void kvm_enable_efer_bits(u64 mask)
1347 efer_reserved_bits &= ~mask;
1349 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1352 * Writes msr value into into the appropriate "register".
1353 * Returns 0 on success, non-0 otherwise.
1354 * Assumes vcpu_load() was already called.
1356 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1358 switch (msr->index) {
1361 case MSR_KERNEL_GS_BASE:
1364 if (is_noncanonical_address(msr->data, vcpu))
1367 case MSR_IA32_SYSENTER_EIP:
1368 case MSR_IA32_SYSENTER_ESP:
1370 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1371 * non-canonical address is written on Intel but not on
1372 * AMD (which ignores the top 32-bits, because it does
1373 * not implement 64-bit SYSENTER).
1375 * 64-bit code should hence be able to write a non-canonical
1376 * value on AMD. Making the address canonical ensures that
1377 * vmentry does not fail on Intel after writing a non-canonical
1378 * value, and that something deterministic happens if the guest
1379 * invokes 64-bit SYSENTER.
1381 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1383 return kvm_x86_ops->set_msr(vcpu, msr);
1385 EXPORT_SYMBOL_GPL(kvm_set_msr);
1388 * Adapt set_msr() to msr_io()'s calling convention
1390 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1392 struct msr_data msr;
1396 msr.host_initiated = true;
1397 r = kvm_get_msr(vcpu, &msr);
1405 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1407 struct msr_data msr;
1411 msr.host_initiated = true;
1412 return kvm_set_msr(vcpu, &msr);
1415 #ifdef CONFIG_X86_64
1416 struct pvclock_gtod_data {
1419 struct { /* extract of a clocksource struct */
1432 static struct pvclock_gtod_data pvclock_gtod_data;
1434 static void update_pvclock_gtod(struct timekeeper *tk)
1436 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1439 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1441 write_seqcount_begin(&vdata->seq);
1443 /* copy pvclock gtod data */
1444 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1445 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1446 vdata->clock.mask = tk->tkr_mono.mask;
1447 vdata->clock.mult = tk->tkr_mono.mult;
1448 vdata->clock.shift = tk->tkr_mono.shift;
1450 vdata->boot_ns = boot_ns;
1451 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1453 vdata->wall_time_sec = tk->xtime_sec;
1455 write_seqcount_end(&vdata->seq);
1459 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1462 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1463 * vcpu_enter_guest. This function is only called from
1464 * the physical CPU that is running vcpu.
1466 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1469 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1473 struct pvclock_wall_clock wc;
1474 struct timespec64 boot;
1479 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1484 ++version; /* first time write, random junk */
1488 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1492 * The guest calculates current wall clock time by adding
1493 * system time (updated by kvm_guest_time_update below) to the
1494 * wall clock specified here. guest system time equals host
1495 * system time for us, thus we must fill in host boot time here.
1497 getboottime64(&boot);
1499 if (kvm->arch.kvmclock_offset) {
1500 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1501 boot = timespec64_sub(boot, ts);
1503 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1504 wc.nsec = boot.tv_nsec;
1505 wc.version = version;
1507 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1510 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1513 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1515 do_shl32_div32(dividend, divisor);
1519 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1520 s8 *pshift, u32 *pmultiplier)
1528 scaled64 = scaled_hz;
1529 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1534 tps32 = (uint32_t)tps64;
1535 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1536 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1544 *pmultiplier = div_frac(scaled64, tps32);
1546 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1547 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1550 #ifdef CONFIG_X86_64
1551 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1554 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1555 static unsigned long max_tsc_khz;
1557 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1559 u64 v = (u64)khz * (1000000 + ppm);
1564 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1568 /* Guest TSC same frequency as host TSC? */
1570 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1574 /* TSC scaling supported? */
1575 if (!kvm_has_tsc_control) {
1576 if (user_tsc_khz > tsc_khz) {
1577 vcpu->arch.tsc_catchup = 1;
1578 vcpu->arch.tsc_always_catchup = 1;
1581 WARN(1, "user requested TSC rate below hardware speed\n");
1586 /* TSC scaling required - calculate ratio */
1587 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1588 user_tsc_khz, tsc_khz);
1590 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1591 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1596 vcpu->arch.tsc_scaling_ratio = ratio;
1600 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1602 u32 thresh_lo, thresh_hi;
1603 int use_scaling = 0;
1605 /* tsc_khz can be zero if TSC calibration fails */
1606 if (user_tsc_khz == 0) {
1607 /* set tsc_scaling_ratio to a safe value */
1608 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1612 /* Compute a scale to convert nanoseconds in TSC cycles */
1613 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1614 &vcpu->arch.virtual_tsc_shift,
1615 &vcpu->arch.virtual_tsc_mult);
1616 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1619 * Compute the variation in TSC rate which is acceptable
1620 * within the range of tolerance and decide if the
1621 * rate being applied is within that bounds of the hardware
1622 * rate. If so, no scaling or compensation need be done.
1624 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1625 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1626 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1627 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1630 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1633 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1635 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1636 vcpu->arch.virtual_tsc_mult,
1637 vcpu->arch.virtual_tsc_shift);
1638 tsc += vcpu->arch.this_tsc_write;
1642 static inline int gtod_is_based_on_tsc(int mode)
1644 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1647 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1649 #ifdef CONFIG_X86_64
1651 struct kvm_arch *ka = &vcpu->kvm->arch;
1652 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1654 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1655 atomic_read(&vcpu->kvm->online_vcpus));
1658 * Once the masterclock is enabled, always perform request in
1659 * order to update it.
1661 * In order to enable masterclock, the host clocksource must be TSC
1662 * and the vcpus need to have matched TSCs. When that happens,
1663 * perform request to enable masterclock.
1665 if (ka->use_master_clock ||
1666 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1667 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1669 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1670 atomic_read(&vcpu->kvm->online_vcpus),
1671 ka->use_master_clock, gtod->clock.vclock_mode);
1675 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1677 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1678 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1682 * Multiply tsc by a fixed point number represented by ratio.
1684 * The most significant 64-N bits (mult) of ratio represent the
1685 * integral part of the fixed point number; the remaining N bits
1686 * (frac) represent the fractional part, ie. ratio represents a fixed
1687 * point number (mult + frac * 2^(-N)).
1689 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1691 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1693 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1696 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1699 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1701 if (ratio != kvm_default_tsc_scaling_ratio)
1702 _tsc = __scale_tsc(ratio, tsc);
1706 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1708 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1712 tsc = kvm_scale_tsc(vcpu, rdtsc());
1714 return target_tsc - tsc;
1717 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1719 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1721 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1723 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1725 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1727 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1730 static inline bool kvm_check_tsc_unstable(void)
1732 #ifdef CONFIG_X86_64
1734 * TSC is marked unstable when we're running on Hyper-V,
1735 * 'TSC page' clocksource is good.
1737 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1740 return check_tsc_unstable();
1743 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1745 struct kvm *kvm = vcpu->kvm;
1746 u64 offset, ns, elapsed;
1747 unsigned long flags;
1749 bool already_matched;
1750 u64 data = msr->data;
1751 bool synchronizing = false;
1753 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1754 offset = kvm_compute_tsc_offset(vcpu, data);
1755 ns = ktime_get_boot_ns();
1756 elapsed = ns - kvm->arch.last_tsc_nsec;
1758 if (vcpu->arch.virtual_tsc_khz) {
1759 if (data == 0 && msr->host_initiated) {
1761 * detection of vcpu initialization -- need to sync
1762 * with other vCPUs. This particularly helps to keep
1763 * kvm_clock stable after CPU hotplug
1765 synchronizing = true;
1767 u64 tsc_exp = kvm->arch.last_tsc_write +
1768 nsec_to_cycles(vcpu, elapsed);
1769 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1771 * Special case: TSC write with a small delta (1 second)
1772 * of virtual cycle time against real time is
1773 * interpreted as an attempt to synchronize the CPU.
1775 synchronizing = data < tsc_exp + tsc_hz &&
1776 data + tsc_hz > tsc_exp;
1781 * For a reliable TSC, we can match TSC offsets, and for an unstable
1782 * TSC, we add elapsed time in this computation. We could let the
1783 * compensation code attempt to catch up if we fall behind, but
1784 * it's better to try to match offsets from the beginning.
1786 if (synchronizing &&
1787 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1788 if (!kvm_check_tsc_unstable()) {
1789 offset = kvm->arch.cur_tsc_offset;
1790 pr_debug("kvm: matched tsc offset for %llu\n", data);
1792 u64 delta = nsec_to_cycles(vcpu, elapsed);
1794 offset = kvm_compute_tsc_offset(vcpu, data);
1795 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1798 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1801 * We split periods of matched TSC writes into generations.
1802 * For each generation, we track the original measured
1803 * nanosecond time, offset, and write, so if TSCs are in
1804 * sync, we can match exact offset, and if not, we can match
1805 * exact software computation in compute_guest_tsc()
1807 * These values are tracked in kvm->arch.cur_xxx variables.
1809 kvm->arch.cur_tsc_generation++;
1810 kvm->arch.cur_tsc_nsec = ns;
1811 kvm->arch.cur_tsc_write = data;
1812 kvm->arch.cur_tsc_offset = offset;
1814 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1815 kvm->arch.cur_tsc_generation, data);
1819 * We also track th most recent recorded KHZ, write and time to
1820 * allow the matching interval to be extended at each write.
1822 kvm->arch.last_tsc_nsec = ns;
1823 kvm->arch.last_tsc_write = data;
1824 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1826 vcpu->arch.last_guest_tsc = data;
1828 /* Keep track of which generation this VCPU has synchronized to */
1829 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1830 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1831 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1833 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1834 update_ia32_tsc_adjust_msr(vcpu, offset);
1836 kvm_vcpu_write_tsc_offset(vcpu, offset);
1837 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1839 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1841 kvm->arch.nr_vcpus_matched_tsc = 0;
1842 } else if (!already_matched) {
1843 kvm->arch.nr_vcpus_matched_tsc++;
1846 kvm_track_tsc_matching(vcpu);
1847 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1850 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1852 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1855 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1856 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1859 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1861 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1862 WARN_ON(adjustment < 0);
1863 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1864 adjust_tsc_offset_guest(vcpu, adjustment);
1867 #ifdef CONFIG_X86_64
1869 static u64 read_tsc(void)
1871 u64 ret = (u64)rdtsc_ordered();
1872 u64 last = pvclock_gtod_data.clock.cycle_last;
1874 if (likely(ret >= last))
1878 * GCC likes to generate cmov here, but this branch is extremely
1879 * predictable (it's just a function of time and the likely is
1880 * very likely) and there's a data dependence, so force GCC
1881 * to generate a branch instead. I don't barrier() because
1882 * we don't actually need a barrier, and if this function
1883 * ever gets inlined it will generate worse code.
1889 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1892 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1895 switch (gtod->clock.vclock_mode) {
1896 case VCLOCK_HVCLOCK:
1897 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1899 if (tsc_pg_val != U64_MAX) {
1900 /* TSC page valid */
1901 *mode = VCLOCK_HVCLOCK;
1902 v = (tsc_pg_val - gtod->clock.cycle_last) &
1905 /* TSC page invalid */
1906 *mode = VCLOCK_NONE;
1911 *tsc_timestamp = read_tsc();
1912 v = (*tsc_timestamp - gtod->clock.cycle_last) &
1916 *mode = VCLOCK_NONE;
1919 if (*mode == VCLOCK_NONE)
1920 *tsc_timestamp = v = 0;
1922 return v * gtod->clock.mult;
1925 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1927 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1933 seq = read_seqcount_begin(>od->seq);
1934 ns = gtod->nsec_base;
1935 ns += vgettsc(tsc_timestamp, &mode);
1936 ns >>= gtod->clock.shift;
1937 ns += gtod->boot_ns;
1938 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1944 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1946 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1952 seq = read_seqcount_begin(>od->seq);
1953 ts->tv_sec = gtod->wall_time_sec;
1954 ns = gtod->nsec_base;
1955 ns += vgettsc(tsc_timestamp, &mode);
1956 ns >>= gtod->clock.shift;
1957 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1959 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1965 /* returns true if host is using TSC based clocksource */
1966 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1968 /* checked again under seqlock below */
1969 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1972 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1976 /* returns true if host is using TSC based clocksource */
1977 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1980 /* checked again under seqlock below */
1981 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1984 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1990 * Assuming a stable TSC across physical CPUS, and a stable TSC
1991 * across virtual CPUs, the following condition is possible.
1992 * Each numbered line represents an event visible to both
1993 * CPUs at the next numbered event.
1995 * "timespecX" represents host monotonic time. "tscX" represents
1998 * VCPU0 on CPU0 | VCPU1 on CPU1
2000 * 1. read timespec0,tsc0
2001 * 2. | timespec1 = timespec0 + N
2003 * 3. transition to guest | transition to guest
2004 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2005 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2006 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2008 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2011 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2013 * - 0 < N - M => M < N
2015 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2016 * always the case (the difference between two distinct xtime instances
2017 * might be smaller then the difference between corresponding TSC reads,
2018 * when updating guest vcpus pvclock areas).
2020 * To avoid that problem, do not allow visibility of distinct
2021 * system_timestamp/tsc_timestamp values simultaneously: use a master
2022 * copy of host monotonic time values. Update that master copy
2025 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2029 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2031 #ifdef CONFIG_X86_64
2032 struct kvm_arch *ka = &kvm->arch;
2034 bool host_tsc_clocksource, vcpus_matched;
2036 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2037 atomic_read(&kvm->online_vcpus));
2040 * If the host uses TSC clock, then passthrough TSC as stable
2043 host_tsc_clocksource = kvm_get_time_and_clockread(
2044 &ka->master_kernel_ns,
2045 &ka->master_cycle_now);
2047 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2048 && !ka->backwards_tsc_observed
2049 && !ka->boot_vcpu_runs_old_kvmclock;
2051 if (ka->use_master_clock)
2052 atomic_set(&kvm_guest_has_master_clock, 1);
2054 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2055 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2060 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2062 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2065 static void kvm_gen_update_masterclock(struct kvm *kvm)
2067 #ifdef CONFIG_X86_64
2069 struct kvm_vcpu *vcpu;
2070 struct kvm_arch *ka = &kvm->arch;
2072 spin_lock(&ka->pvclock_gtod_sync_lock);
2073 kvm_make_mclock_inprogress_request(kvm);
2074 /* no guest entries from this point */
2075 pvclock_update_vm_gtod_copy(kvm);
2077 kvm_for_each_vcpu(i, vcpu, kvm)
2078 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2080 /* guest entries allowed */
2081 kvm_for_each_vcpu(i, vcpu, kvm)
2082 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2084 spin_unlock(&ka->pvclock_gtod_sync_lock);
2088 u64 get_kvmclock_ns(struct kvm *kvm)
2090 struct kvm_arch *ka = &kvm->arch;
2091 struct pvclock_vcpu_time_info hv_clock;
2094 spin_lock(&ka->pvclock_gtod_sync_lock);
2095 if (!ka->use_master_clock) {
2096 spin_unlock(&ka->pvclock_gtod_sync_lock);
2097 return ktime_get_boot_ns() + ka->kvmclock_offset;
2100 hv_clock.tsc_timestamp = ka->master_cycle_now;
2101 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2102 spin_unlock(&ka->pvclock_gtod_sync_lock);
2104 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2107 if (__this_cpu_read(cpu_tsc_khz)) {
2108 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2109 &hv_clock.tsc_shift,
2110 &hv_clock.tsc_to_system_mul);
2111 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2113 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
2120 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2122 struct kvm_vcpu_arch *vcpu = &v->arch;
2123 struct pvclock_vcpu_time_info guest_hv_clock;
2125 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2126 &guest_hv_clock, sizeof(guest_hv_clock))))
2129 /* This VCPU is paused, but it's legal for a guest to read another
2130 * VCPU's kvmclock, so we really have to follow the specification where
2131 * it says that version is odd if data is being modified, and even after
2134 * Version field updates must be kept separate. This is because
2135 * kvm_write_guest_cached might use a "rep movs" instruction, and
2136 * writes within a string instruction are weakly ordered. So there
2137 * are three writes overall.
2139 * As a small optimization, only write the version field in the first
2140 * and third write. The vcpu->pv_time cache is still valid, because the
2141 * version field is the first in the struct.
2143 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2145 if (guest_hv_clock.version & 1)
2146 ++guest_hv_clock.version; /* first time write, random junk */
2148 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2149 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2151 sizeof(vcpu->hv_clock.version));
2155 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2156 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2158 if (vcpu->pvclock_set_guest_stopped_request) {
2159 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2160 vcpu->pvclock_set_guest_stopped_request = false;
2163 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2165 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2167 sizeof(vcpu->hv_clock));
2171 vcpu->hv_clock.version++;
2172 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2174 sizeof(vcpu->hv_clock.version));
2177 static int kvm_guest_time_update(struct kvm_vcpu *v)
2179 unsigned long flags, tgt_tsc_khz;
2180 struct kvm_vcpu_arch *vcpu = &v->arch;
2181 struct kvm_arch *ka = &v->kvm->arch;
2183 u64 tsc_timestamp, host_tsc;
2185 bool use_master_clock;
2191 * If the host uses TSC clock, then passthrough TSC as stable
2194 spin_lock(&ka->pvclock_gtod_sync_lock);
2195 use_master_clock = ka->use_master_clock;
2196 if (use_master_clock) {
2197 host_tsc = ka->master_cycle_now;
2198 kernel_ns = ka->master_kernel_ns;
2200 spin_unlock(&ka->pvclock_gtod_sync_lock);
2202 /* Keep irq disabled to prevent changes to the clock */
2203 local_irq_save(flags);
2204 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2205 if (unlikely(tgt_tsc_khz == 0)) {
2206 local_irq_restore(flags);
2207 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2210 if (!use_master_clock) {
2212 kernel_ns = ktime_get_boot_ns();
2215 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2218 * We may have to catch up the TSC to match elapsed wall clock
2219 * time for two reasons, even if kvmclock is used.
2220 * 1) CPU could have been running below the maximum TSC rate
2221 * 2) Broken TSC compensation resets the base at each VCPU
2222 * entry to avoid unknown leaps of TSC even when running
2223 * again on the same CPU. This may cause apparent elapsed
2224 * time to disappear, and the guest to stand still or run
2227 if (vcpu->tsc_catchup) {
2228 u64 tsc = compute_guest_tsc(v, kernel_ns);
2229 if (tsc > tsc_timestamp) {
2230 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2231 tsc_timestamp = tsc;
2235 local_irq_restore(flags);
2237 /* With all the info we got, fill in the values */
2239 if (kvm_has_tsc_control)
2240 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2242 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2243 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2244 &vcpu->hv_clock.tsc_shift,
2245 &vcpu->hv_clock.tsc_to_system_mul);
2246 vcpu->hw_tsc_khz = tgt_tsc_khz;
2249 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2250 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2251 vcpu->last_guest_tsc = tsc_timestamp;
2253 /* If the host uses TSC clocksource, then it is stable */
2255 if (use_master_clock)
2256 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2258 vcpu->hv_clock.flags = pvclock_flags;
2260 if (vcpu->pv_time_enabled)
2261 kvm_setup_pvclock_page(v);
2262 if (v == kvm_get_vcpu(v->kvm, 0))
2263 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2268 * kvmclock updates which are isolated to a given vcpu, such as
2269 * vcpu->cpu migration, should not allow system_timestamp from
2270 * the rest of the vcpus to remain static. Otherwise ntp frequency
2271 * correction applies to one vcpu's system_timestamp but not
2274 * So in those cases, request a kvmclock update for all vcpus.
2275 * We need to rate-limit these requests though, as they can
2276 * considerably slow guests that have a large number of vcpus.
2277 * The time for a remote vcpu to update its kvmclock is bound
2278 * by the delay we use to rate-limit the updates.
2281 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2283 static void kvmclock_update_fn(struct work_struct *work)
2286 struct delayed_work *dwork = to_delayed_work(work);
2287 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2288 kvmclock_update_work);
2289 struct kvm *kvm = container_of(ka, struct kvm, arch);
2290 struct kvm_vcpu *vcpu;
2292 kvm_for_each_vcpu(i, vcpu, kvm) {
2293 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2294 kvm_vcpu_kick(vcpu);
2298 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2300 struct kvm *kvm = v->kvm;
2302 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2303 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2304 KVMCLOCK_UPDATE_DELAY);
2307 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2309 static void kvmclock_sync_fn(struct work_struct *work)
2311 struct delayed_work *dwork = to_delayed_work(work);
2312 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2313 kvmclock_sync_work);
2314 struct kvm *kvm = container_of(ka, struct kvm, arch);
2316 if (!kvmclock_periodic_sync)
2319 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2320 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2321 KVMCLOCK_SYNC_PERIOD);
2325 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2327 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2329 /* McStatusWrEn enabled? */
2330 if (guest_cpuid_is_amd(vcpu))
2331 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2336 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2338 u64 mcg_cap = vcpu->arch.mcg_cap;
2339 unsigned bank_num = mcg_cap & 0xff;
2340 u32 msr = msr_info->index;
2341 u64 data = msr_info->data;
2344 case MSR_IA32_MCG_STATUS:
2345 vcpu->arch.mcg_status = data;
2347 case MSR_IA32_MCG_CTL:
2348 if (!(mcg_cap & MCG_CTL_P) &&
2349 (data || !msr_info->host_initiated))
2351 if (data != 0 && data != ~(u64)0)
2353 vcpu->arch.mcg_ctl = data;
2356 if (msr >= MSR_IA32_MC0_CTL &&
2357 msr < MSR_IA32_MCx_CTL(bank_num)) {
2358 u32 offset = msr - MSR_IA32_MC0_CTL;
2359 /* only 0 or all 1s can be written to IA32_MCi_CTL
2360 * some Linux kernels though clear bit 10 in bank 4 to
2361 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2362 * this to avoid an uncatched #GP in the guest
2364 if ((offset & 0x3) == 0 &&
2365 data != 0 && (data | (1 << 10)) != ~(u64)0)
2369 if (!msr_info->host_initiated &&
2370 (offset & 0x3) == 1 && data != 0) {
2371 if (!can_set_mci_status(vcpu))
2375 vcpu->arch.mce_banks[offset] = data;
2383 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2385 struct kvm *kvm = vcpu->kvm;
2386 int lm = is_long_mode(vcpu);
2387 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2388 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2389 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2390 : kvm->arch.xen_hvm_config.blob_size_32;
2391 u32 page_num = data & ~PAGE_MASK;
2392 u64 page_addr = data & PAGE_MASK;
2397 if (page_num >= blob_size)
2400 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2405 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2414 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2416 gpa_t gpa = data & ~0x3f;
2418 /* Bits 3:5 are reserved, Should be zero */
2422 vcpu->arch.apf.msr_val = data;
2424 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2425 kvm_clear_async_pf_completion_queue(vcpu);
2426 kvm_async_pf_hash_reset(vcpu);
2430 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2434 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2435 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2436 kvm_async_pf_wakeup_all(vcpu);
2440 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2442 vcpu->arch.pv_time_enabled = false;
2445 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2447 ++vcpu->stat.tlb_flush;
2448 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2451 static void record_steal_time(struct kvm_vcpu *vcpu)
2453 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2456 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2457 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2461 * Doing a TLB flush here, on the guest's behalf, can avoid
2464 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2465 kvm_vcpu_flush_tlb(vcpu, false);
2467 if (vcpu->arch.st.steal.version & 1)
2468 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2470 vcpu->arch.st.steal.version += 1;
2472 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2473 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2477 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2478 vcpu->arch.st.last_steal;
2479 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2481 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2482 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2486 vcpu->arch.st.steal.version += 1;
2488 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2489 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2492 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2495 u32 msr = msr_info->index;
2496 u64 data = msr_info->data;
2499 case MSR_AMD64_NB_CFG:
2500 case MSR_IA32_UCODE_WRITE:
2501 case MSR_VM_HSAVE_PA:
2502 case MSR_AMD64_PATCH_LOADER:
2503 case MSR_AMD64_BU_CFG2:
2504 case MSR_AMD64_DC_CFG:
2505 case MSR_F15H_EX_CFG:
2508 case MSR_IA32_UCODE_REV:
2509 if (msr_info->host_initiated)
2510 vcpu->arch.microcode_version = data;
2512 case MSR_IA32_ARCH_CAPABILITIES:
2513 if (!msr_info->host_initiated)
2515 vcpu->arch.arch_capabilities = data;
2518 return set_efer(vcpu, msr_info);
2520 data &= ~(u64)0x40; /* ignore flush filter disable */
2521 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2522 data &= ~(u64)0x8; /* ignore TLB cache disable */
2524 /* Handle McStatusWrEn */
2525 if (data == BIT_ULL(18)) {
2526 vcpu->arch.msr_hwcr = data;
2527 } else if (data != 0) {
2528 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2533 case MSR_FAM10H_MMIO_CONF_BASE:
2535 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2540 case MSR_IA32_DEBUGCTLMSR:
2542 /* We support the non-activated case already */
2544 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2545 /* Values other than LBR and BTF are vendor-specific,
2546 thus reserved and should throw a #GP */
2549 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2552 case 0x200 ... 0x2ff:
2553 return kvm_mtrr_set_msr(vcpu, msr, data);
2554 case MSR_IA32_APICBASE:
2555 return kvm_set_apic_base(vcpu, msr_info);
2556 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2557 return kvm_x2apic_msr_write(vcpu, msr, data);
2558 case MSR_IA32_TSCDEADLINE:
2559 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2561 case MSR_IA32_TSC_ADJUST:
2562 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2563 if (!msr_info->host_initiated) {
2564 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2565 adjust_tsc_offset_guest(vcpu, adj);
2567 vcpu->arch.ia32_tsc_adjust_msr = data;
2570 case MSR_IA32_MISC_ENABLE:
2571 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2572 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2573 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2575 vcpu->arch.ia32_misc_enable_msr = data;
2576 kvm_update_cpuid(vcpu);
2578 vcpu->arch.ia32_misc_enable_msr = data;
2581 case MSR_IA32_SMBASE:
2582 if (!msr_info->host_initiated)
2584 vcpu->arch.smbase = data;
2586 case MSR_IA32_POWER_CTL:
2587 vcpu->arch.msr_ia32_power_ctl = data;
2590 kvm_write_tsc(vcpu, msr_info);
2593 if (!msr_info->host_initiated)
2595 vcpu->arch.smi_count = data;
2597 case MSR_KVM_WALL_CLOCK_NEW:
2598 case MSR_KVM_WALL_CLOCK:
2599 vcpu->kvm->arch.wall_clock = data;
2600 kvm_write_wall_clock(vcpu->kvm, data);
2602 case MSR_KVM_SYSTEM_TIME_NEW:
2603 case MSR_KVM_SYSTEM_TIME: {
2604 struct kvm_arch *ka = &vcpu->kvm->arch;
2606 kvmclock_reset(vcpu);
2608 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2609 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2611 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2612 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2614 ka->boot_vcpu_runs_old_kvmclock = tmp;
2617 vcpu->arch.time = data;
2618 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2620 /* we verify if the enable bit is set... */
2624 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2625 &vcpu->arch.pv_time, data & ~1ULL,
2626 sizeof(struct pvclock_vcpu_time_info)))
2627 vcpu->arch.pv_time_enabled = false;
2629 vcpu->arch.pv_time_enabled = true;
2633 case MSR_KVM_ASYNC_PF_EN:
2634 if (kvm_pv_enable_async_pf(vcpu, data))
2637 case MSR_KVM_STEAL_TIME:
2639 if (unlikely(!sched_info_on()))
2642 if (data & KVM_STEAL_RESERVED_MASK)
2645 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2646 data & KVM_STEAL_VALID_BITS,
2647 sizeof(struct kvm_steal_time)))
2650 vcpu->arch.st.msr_val = data;
2652 if (!(data & KVM_MSR_ENABLED))
2655 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2658 case MSR_KVM_PV_EOI_EN:
2659 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2663 case MSR_KVM_POLL_CONTROL:
2664 /* only enable bit supported */
2665 if (data & (-1ULL << 1))
2668 vcpu->arch.msr_kvm_poll_control = data;
2671 case MSR_IA32_MCG_CTL:
2672 case MSR_IA32_MCG_STATUS:
2673 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2674 return set_msr_mce(vcpu, msr_info);
2676 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2677 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2678 pr = true; /* fall through */
2679 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2680 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2681 if (kvm_pmu_is_valid_msr(vcpu, msr))
2682 return kvm_pmu_set_msr(vcpu, msr_info);
2684 if (pr || data != 0)
2685 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2686 "0x%x data 0x%llx\n", msr, data);
2688 case MSR_K7_CLK_CTL:
2690 * Ignore all writes to this no longer documented MSR.
2691 * Writes are only relevant for old K7 processors,
2692 * all pre-dating SVM, but a recommended workaround from
2693 * AMD for these chips. It is possible to specify the
2694 * affected processor models on the command line, hence
2695 * the need to ignore the workaround.
2698 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2699 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2700 case HV_X64_MSR_CRASH_CTL:
2701 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2702 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2703 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2704 case HV_X64_MSR_TSC_EMULATION_STATUS:
2705 return kvm_hv_set_msr_common(vcpu, msr, data,
2706 msr_info->host_initiated);
2707 case MSR_IA32_BBL_CR_CTL3:
2708 /* Drop writes to this legacy MSR -- see rdmsr
2709 * counterpart for further detail.
2711 if (report_ignored_msrs)
2712 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2715 case MSR_AMD64_OSVW_ID_LENGTH:
2716 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2718 vcpu->arch.osvw.length = data;
2720 case MSR_AMD64_OSVW_STATUS:
2721 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2723 vcpu->arch.osvw.status = data;
2725 case MSR_PLATFORM_INFO:
2726 if (!msr_info->host_initiated ||
2727 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2728 cpuid_fault_enabled(vcpu)))
2730 vcpu->arch.msr_platform_info = data;
2732 case MSR_MISC_FEATURES_ENABLES:
2733 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2734 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2735 !supports_cpuid_fault(vcpu)))
2737 vcpu->arch.msr_misc_features_enables = data;
2740 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2741 return xen_hvm_config(vcpu, data);
2742 if (kvm_pmu_is_valid_msr(vcpu, msr))
2743 return kvm_pmu_set_msr(vcpu, msr_info);
2745 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2749 if (report_ignored_msrs)
2751 "ignored wrmsr: 0x%x data 0x%llx\n",
2758 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2762 * Reads an msr value (of 'msr_index') into 'pdata'.
2763 * Returns 0 on success, non-0 otherwise.
2764 * Assumes vcpu_load() was already called.
2766 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2768 return kvm_x86_ops->get_msr(vcpu, msr);
2770 EXPORT_SYMBOL_GPL(kvm_get_msr);
2772 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2775 u64 mcg_cap = vcpu->arch.mcg_cap;
2776 unsigned bank_num = mcg_cap & 0xff;
2779 case MSR_IA32_P5_MC_ADDR:
2780 case MSR_IA32_P5_MC_TYPE:
2783 case MSR_IA32_MCG_CAP:
2784 data = vcpu->arch.mcg_cap;
2786 case MSR_IA32_MCG_CTL:
2787 if (!(mcg_cap & MCG_CTL_P) && !host)
2789 data = vcpu->arch.mcg_ctl;
2791 case MSR_IA32_MCG_STATUS:
2792 data = vcpu->arch.mcg_status;
2795 if (msr >= MSR_IA32_MC0_CTL &&
2796 msr < MSR_IA32_MCx_CTL(bank_num)) {
2797 u32 offset = msr - MSR_IA32_MC0_CTL;
2798 data = vcpu->arch.mce_banks[offset];
2807 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2809 switch (msr_info->index) {
2810 case MSR_IA32_PLATFORM_ID:
2811 case MSR_IA32_EBL_CR_POWERON:
2812 case MSR_IA32_DEBUGCTLMSR:
2813 case MSR_IA32_LASTBRANCHFROMIP:
2814 case MSR_IA32_LASTBRANCHTOIP:
2815 case MSR_IA32_LASTINTFROMIP:
2816 case MSR_IA32_LASTINTTOIP:
2818 case MSR_K8_TSEG_ADDR:
2819 case MSR_K8_TSEG_MASK:
2820 case MSR_VM_HSAVE_PA:
2821 case MSR_K8_INT_PENDING_MSG:
2822 case MSR_AMD64_NB_CFG:
2823 case MSR_FAM10H_MMIO_CONF_BASE:
2824 case MSR_AMD64_BU_CFG2:
2825 case MSR_IA32_PERF_CTL:
2826 case MSR_AMD64_DC_CFG:
2827 case MSR_F15H_EX_CFG:
2830 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2831 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2832 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2833 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2834 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2835 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2836 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2839 case MSR_IA32_UCODE_REV:
2840 msr_info->data = vcpu->arch.microcode_version;
2842 case MSR_IA32_ARCH_CAPABILITIES:
2843 if (!msr_info->host_initiated &&
2844 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2846 msr_info->data = vcpu->arch.arch_capabilities;
2848 case MSR_IA32_POWER_CTL:
2849 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2852 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2855 case 0x200 ... 0x2ff:
2856 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2857 case 0xcd: /* fsb frequency */
2861 * MSR_EBC_FREQUENCY_ID
2862 * Conservative value valid for even the basic CPU models.
2863 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2864 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2865 * and 266MHz for model 3, or 4. Set Core Clock
2866 * Frequency to System Bus Frequency Ratio to 1 (bits
2867 * 31:24) even though these are only valid for CPU
2868 * models > 2, however guests may end up dividing or
2869 * multiplying by zero otherwise.
2871 case MSR_EBC_FREQUENCY_ID:
2872 msr_info->data = 1 << 24;
2874 case MSR_IA32_APICBASE:
2875 msr_info->data = kvm_get_apic_base(vcpu);
2877 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2878 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2880 case MSR_IA32_TSCDEADLINE:
2881 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2883 case MSR_IA32_TSC_ADJUST:
2884 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2886 case MSR_IA32_MISC_ENABLE:
2887 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2889 case MSR_IA32_SMBASE:
2890 if (!msr_info->host_initiated)
2892 msr_info->data = vcpu->arch.smbase;
2895 msr_info->data = vcpu->arch.smi_count;
2897 case MSR_IA32_PERF_STATUS:
2898 /* TSC increment by tick */
2899 msr_info->data = 1000ULL;
2900 /* CPU multiplier */
2901 msr_info->data |= (((uint64_t)4ULL) << 40);
2904 msr_info->data = vcpu->arch.efer;
2906 case MSR_KVM_WALL_CLOCK:
2907 case MSR_KVM_WALL_CLOCK_NEW:
2908 msr_info->data = vcpu->kvm->arch.wall_clock;
2910 case MSR_KVM_SYSTEM_TIME:
2911 case MSR_KVM_SYSTEM_TIME_NEW:
2912 msr_info->data = vcpu->arch.time;
2914 case MSR_KVM_ASYNC_PF_EN:
2915 msr_info->data = vcpu->arch.apf.msr_val;
2917 case MSR_KVM_STEAL_TIME:
2918 msr_info->data = vcpu->arch.st.msr_val;
2920 case MSR_KVM_PV_EOI_EN:
2921 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2923 case MSR_KVM_POLL_CONTROL:
2924 msr_info->data = vcpu->arch.msr_kvm_poll_control;
2926 case MSR_IA32_P5_MC_ADDR:
2927 case MSR_IA32_P5_MC_TYPE:
2928 case MSR_IA32_MCG_CAP:
2929 case MSR_IA32_MCG_CTL:
2930 case MSR_IA32_MCG_STATUS:
2931 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2932 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2933 msr_info->host_initiated);
2934 case MSR_K7_CLK_CTL:
2936 * Provide expected ramp-up count for K7. All other
2937 * are set to zero, indicating minimum divisors for
2940 * This prevents guest kernels on AMD host with CPU
2941 * type 6, model 8 and higher from exploding due to
2942 * the rdmsr failing.
2944 msr_info->data = 0x20000000;
2946 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2947 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2948 case HV_X64_MSR_CRASH_CTL:
2949 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2950 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2951 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2952 case HV_X64_MSR_TSC_EMULATION_STATUS:
2953 return kvm_hv_get_msr_common(vcpu,
2954 msr_info->index, &msr_info->data,
2955 msr_info->host_initiated);
2957 case MSR_IA32_BBL_CR_CTL3:
2958 /* This legacy MSR exists but isn't fully documented in current
2959 * silicon. It is however accessed by winxp in very narrow
2960 * scenarios where it sets bit #19, itself documented as
2961 * a "reserved" bit. Best effort attempt to source coherent
2962 * read data here should the balance of the register be
2963 * interpreted by the guest:
2965 * L2 cache control register 3: 64GB range, 256KB size,
2966 * enabled, latency 0x1, configured
2968 msr_info->data = 0xbe702111;
2970 case MSR_AMD64_OSVW_ID_LENGTH:
2971 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2973 msr_info->data = vcpu->arch.osvw.length;
2975 case MSR_AMD64_OSVW_STATUS:
2976 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2978 msr_info->data = vcpu->arch.osvw.status;
2980 case MSR_PLATFORM_INFO:
2981 if (!msr_info->host_initiated &&
2982 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2984 msr_info->data = vcpu->arch.msr_platform_info;
2986 case MSR_MISC_FEATURES_ENABLES:
2987 msr_info->data = vcpu->arch.msr_misc_features_enables;
2990 msr_info->data = vcpu->arch.msr_hwcr;
2993 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2994 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2996 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3000 if (report_ignored_msrs)
3001 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3009 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3012 * Read or write a bunch of msrs. All parameters are kernel addresses.
3014 * @return number of msrs set successfully.
3016 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3017 struct kvm_msr_entry *entries,
3018 int (*do_msr)(struct kvm_vcpu *vcpu,
3019 unsigned index, u64 *data))
3023 for (i = 0; i < msrs->nmsrs; ++i)
3024 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3031 * Read or write a bunch of msrs. Parameters are user addresses.
3033 * @return number of msrs set successfully.
3035 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3036 int (*do_msr)(struct kvm_vcpu *vcpu,
3037 unsigned index, u64 *data),
3040 struct kvm_msrs msrs;
3041 struct kvm_msr_entry *entries;
3046 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3050 if (msrs.nmsrs >= MAX_IO_MSRS)
3053 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3054 entries = memdup_user(user_msrs->entries, size);
3055 if (IS_ERR(entries)) {
3056 r = PTR_ERR(entries);
3060 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3065 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3076 static inline bool kvm_can_mwait_in_guest(void)
3078 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3079 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3080 boot_cpu_has(X86_FEATURE_ARAT);
3083 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3088 case KVM_CAP_IRQCHIP:
3090 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3091 case KVM_CAP_SET_TSS_ADDR:
3092 case KVM_CAP_EXT_CPUID:
3093 case KVM_CAP_EXT_EMUL_CPUID:
3094 case KVM_CAP_CLOCKSOURCE:
3096 case KVM_CAP_NOP_IO_DELAY:
3097 case KVM_CAP_MP_STATE:
3098 case KVM_CAP_SYNC_MMU:
3099 case KVM_CAP_USER_NMI:
3100 case KVM_CAP_REINJECT_CONTROL:
3101 case KVM_CAP_IRQ_INJECT_STATUS:
3102 case KVM_CAP_IOEVENTFD:
3103 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3105 case KVM_CAP_PIT_STATE2:
3106 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3107 case KVM_CAP_XEN_HVM:
3108 case KVM_CAP_VCPU_EVENTS:
3109 case KVM_CAP_HYPERV:
3110 case KVM_CAP_HYPERV_VAPIC:
3111 case KVM_CAP_HYPERV_SPIN:
3112 case KVM_CAP_HYPERV_SYNIC:
3113 case KVM_CAP_HYPERV_SYNIC2:
3114 case KVM_CAP_HYPERV_VP_INDEX:
3115 case KVM_CAP_HYPERV_EVENTFD:
3116 case KVM_CAP_HYPERV_TLBFLUSH:
3117 case KVM_CAP_HYPERV_SEND_IPI:
3118 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3119 case KVM_CAP_HYPERV_CPUID:
3120 case KVM_CAP_PCI_SEGMENT:
3121 case KVM_CAP_DEBUGREGS:
3122 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3124 case KVM_CAP_ASYNC_PF:
3125 case KVM_CAP_GET_TSC_KHZ:
3126 case KVM_CAP_KVMCLOCK_CTRL:
3127 case KVM_CAP_READONLY_MEM:
3128 case KVM_CAP_HYPERV_TIME:
3129 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3130 case KVM_CAP_TSC_DEADLINE_TIMER:
3131 case KVM_CAP_DISABLE_QUIRKS:
3132 case KVM_CAP_SET_BOOT_CPU_ID:
3133 case KVM_CAP_SPLIT_IRQCHIP:
3134 case KVM_CAP_IMMEDIATE_EXIT:
3135 case KVM_CAP_PMU_EVENT_FILTER:
3136 case KVM_CAP_GET_MSR_FEATURES:
3137 case KVM_CAP_MSR_PLATFORM_INFO:
3138 case KVM_CAP_EXCEPTION_PAYLOAD:
3141 case KVM_CAP_SYNC_REGS:
3142 r = KVM_SYNC_X86_VALID_FIELDS;
3144 case KVM_CAP_ADJUST_CLOCK:
3145 r = KVM_CLOCK_TSC_STABLE;
3147 case KVM_CAP_X86_DISABLE_EXITS:
3148 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3149 KVM_X86_DISABLE_EXITS_CSTATE;
3150 if(kvm_can_mwait_in_guest())
3151 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3153 case KVM_CAP_X86_SMM:
3154 /* SMBASE is usually relocated above 1M on modern chipsets,
3155 * and SMM handlers might indeed rely on 4G segment limits,
3156 * so do not report SMM to be available if real mode is
3157 * emulated via vm86 mode. Still, do not go to great lengths
3158 * to avoid userspace's usage of the feature, because it is a
3159 * fringe case that is not enabled except via specific settings
3160 * of the module parameters.
3162 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3165 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3167 case KVM_CAP_NR_VCPUS:
3168 r = KVM_SOFT_MAX_VCPUS;
3170 case KVM_CAP_MAX_VCPUS:
3173 case KVM_CAP_MAX_VCPU_ID:
3174 r = KVM_MAX_VCPU_ID;
3176 case KVM_CAP_PV_MMU: /* obsolete */
3180 r = KVM_MAX_MCE_BANKS;
3183 r = boot_cpu_has(X86_FEATURE_XSAVE);
3185 case KVM_CAP_TSC_CONTROL:
3186 r = kvm_has_tsc_control;
3188 case KVM_CAP_X2APIC_API:
3189 r = KVM_X2APIC_API_VALID_FLAGS;
3191 case KVM_CAP_NESTED_STATE:
3192 r = kvm_x86_ops->get_nested_state ?
3193 kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3202 long kvm_arch_dev_ioctl(struct file *filp,
3203 unsigned int ioctl, unsigned long arg)
3205 void __user *argp = (void __user *)arg;
3209 case KVM_GET_MSR_INDEX_LIST: {
3210 struct kvm_msr_list __user *user_msr_list = argp;
3211 struct kvm_msr_list msr_list;
3215 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3218 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3219 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3222 if (n < msr_list.nmsrs)
3225 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3226 num_msrs_to_save * sizeof(u32)))
3228 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3230 num_emulated_msrs * sizeof(u32)))
3235 case KVM_GET_SUPPORTED_CPUID:
3236 case KVM_GET_EMULATED_CPUID: {
3237 struct kvm_cpuid2 __user *cpuid_arg = argp;
3238 struct kvm_cpuid2 cpuid;
3241 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3244 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3250 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3255 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3257 if (copy_to_user(argp, &kvm_mce_cap_supported,
3258 sizeof(kvm_mce_cap_supported)))
3262 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3263 struct kvm_msr_list __user *user_msr_list = argp;
3264 struct kvm_msr_list msr_list;
3268 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3271 msr_list.nmsrs = num_msr_based_features;
3272 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3275 if (n < msr_list.nmsrs)
3278 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3279 num_msr_based_features * sizeof(u32)))
3285 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3295 static void wbinvd_ipi(void *garbage)
3300 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3302 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3305 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3307 /* Address WBINVD may be executed by guest */
3308 if (need_emulate_wbinvd(vcpu)) {
3309 if (kvm_x86_ops->has_wbinvd_exit())
3310 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3311 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3312 smp_call_function_single(vcpu->cpu,
3313 wbinvd_ipi, NULL, 1);
3316 kvm_x86_ops->vcpu_load(vcpu, cpu);
3318 /* Apply any externally detected TSC adjustments (due to suspend) */
3319 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3320 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3321 vcpu->arch.tsc_offset_adjustment = 0;
3322 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3325 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3326 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3327 rdtsc() - vcpu->arch.last_host_tsc;
3329 mark_tsc_unstable("KVM discovered backwards TSC");
3331 if (kvm_check_tsc_unstable()) {
3332 u64 offset = kvm_compute_tsc_offset(vcpu,
3333 vcpu->arch.last_guest_tsc);
3334 kvm_vcpu_write_tsc_offset(vcpu, offset);
3335 vcpu->arch.tsc_catchup = 1;
3338 if (kvm_lapic_hv_timer_in_use(vcpu))
3339 kvm_lapic_restart_hv_timer(vcpu);
3342 * On a host with synchronized TSC, there is no need to update
3343 * kvmclock on vcpu->cpu migration
3345 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3346 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3347 if (vcpu->cpu != cpu)
3348 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3352 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3355 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3357 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3360 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3362 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3363 &vcpu->arch.st.steal.preempted,
3364 offsetof(struct kvm_steal_time, preempted),
3365 sizeof(vcpu->arch.st.steal.preempted));
3368 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3372 if (vcpu->preempted)
3373 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3376 * Disable page faults because we're in atomic context here.
3377 * kvm_write_guest_offset_cached() would call might_fault()
3378 * that relies on pagefault_disable() to tell if there's a
3379 * bug. NOTE: the write to guest memory may not go through if
3380 * during postcopy live migration or if there's heavy guest
3383 pagefault_disable();
3385 * kvm_memslots() will be called by
3386 * kvm_write_guest_offset_cached() so take the srcu lock.
3388 idx = srcu_read_lock(&vcpu->kvm->srcu);
3389 kvm_steal_time_set_preempted(vcpu);
3390 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3392 kvm_x86_ops->vcpu_put(vcpu);
3393 vcpu->arch.last_host_tsc = rdtsc();
3395 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3396 * on every vmexit, but if not, we might have a stale dr6 from the
3397 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3402 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3403 struct kvm_lapic_state *s)
3405 if (vcpu->arch.apicv_active)
3406 kvm_x86_ops->sync_pir_to_irr(vcpu);
3408 return kvm_apic_get_state(vcpu, s);
3411 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3412 struct kvm_lapic_state *s)
3416 r = kvm_apic_set_state(vcpu, s);
3419 update_cr8_intercept(vcpu);
3424 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3426 return (!lapic_in_kernel(vcpu) ||
3427 kvm_apic_accept_pic_intr(vcpu));
3431 * if userspace requested an interrupt window, check that the
3432 * interrupt window is open.
3434 * No need to exit to userspace if we already have an interrupt queued.
3436 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3438 return kvm_arch_interrupt_allowed(vcpu) &&
3439 !kvm_cpu_has_interrupt(vcpu) &&
3440 !kvm_event_needs_reinjection(vcpu) &&
3441 kvm_cpu_accept_dm_intr(vcpu);
3444 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3445 struct kvm_interrupt *irq)
3447 if (irq->irq >= KVM_NR_INTERRUPTS)
3450 if (!irqchip_in_kernel(vcpu->kvm)) {
3451 kvm_queue_interrupt(vcpu, irq->irq, false);
3452 kvm_make_request(KVM_REQ_EVENT, vcpu);
3457 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3458 * fail for in-kernel 8259.
3460 if (pic_in_kernel(vcpu->kvm))
3463 if (vcpu->arch.pending_external_vector != -1)
3466 vcpu->arch.pending_external_vector = irq->irq;
3467 kvm_make_request(KVM_REQ_EVENT, vcpu);
3471 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3473 kvm_inject_nmi(vcpu);
3478 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3480 kvm_make_request(KVM_REQ_SMI, vcpu);
3485 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3486 struct kvm_tpr_access_ctl *tac)
3490 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3494 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3498 unsigned bank_num = mcg_cap & 0xff, bank;
3501 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3503 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3506 vcpu->arch.mcg_cap = mcg_cap;
3507 /* Init IA32_MCG_CTL to all 1s */
3508 if (mcg_cap & MCG_CTL_P)
3509 vcpu->arch.mcg_ctl = ~(u64)0;
3510 /* Init IA32_MCi_CTL to all 1s */
3511 for (bank = 0; bank < bank_num; bank++)
3512 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3514 if (kvm_x86_ops->setup_mce)
3515 kvm_x86_ops->setup_mce(vcpu);
3520 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3521 struct kvm_x86_mce *mce)
3523 u64 mcg_cap = vcpu->arch.mcg_cap;
3524 unsigned bank_num = mcg_cap & 0xff;
3525 u64 *banks = vcpu->arch.mce_banks;
3527 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3530 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3531 * reporting is disabled
3533 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3534 vcpu->arch.mcg_ctl != ~(u64)0)
3536 banks += 4 * mce->bank;
3538 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3539 * reporting is disabled for the bank
3541 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3543 if (mce->status & MCI_STATUS_UC) {
3544 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3545 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3546 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3549 if (banks[1] & MCI_STATUS_VAL)
3550 mce->status |= MCI_STATUS_OVER;
3551 banks[2] = mce->addr;
3552 banks[3] = mce->misc;
3553 vcpu->arch.mcg_status = mce->mcg_status;
3554 banks[1] = mce->status;
3555 kvm_queue_exception(vcpu, MC_VECTOR);
3556 } else if (!(banks[1] & MCI_STATUS_VAL)
3557 || !(banks[1] & MCI_STATUS_UC)) {
3558 if (banks[1] & MCI_STATUS_VAL)
3559 mce->status |= MCI_STATUS_OVER;
3560 banks[2] = mce->addr;
3561 banks[3] = mce->misc;
3562 banks[1] = mce->status;
3564 banks[1] |= MCI_STATUS_OVER;
3568 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3569 struct kvm_vcpu_events *events)
3574 * The API doesn't provide the instruction length for software
3575 * exceptions, so don't report them. As long as the guest RIP
3576 * isn't advanced, we should expect to encounter the exception
3579 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3580 events->exception.injected = 0;
3581 events->exception.pending = 0;
3583 events->exception.injected = vcpu->arch.exception.injected;
3584 events->exception.pending = vcpu->arch.exception.pending;
3586 * For ABI compatibility, deliberately conflate
3587 * pending and injected exceptions when
3588 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3590 if (!vcpu->kvm->arch.exception_payload_enabled)
3591 events->exception.injected |=
3592 vcpu->arch.exception.pending;
3594 events->exception.nr = vcpu->arch.exception.nr;
3595 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3596 events->exception.error_code = vcpu->arch.exception.error_code;
3597 events->exception_has_payload = vcpu->arch.exception.has_payload;
3598 events->exception_payload = vcpu->arch.exception.payload;
3600 events->interrupt.injected =
3601 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3602 events->interrupt.nr = vcpu->arch.interrupt.nr;
3603 events->interrupt.soft = 0;
3604 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3606 events->nmi.injected = vcpu->arch.nmi_injected;
3607 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3608 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3609 events->nmi.pad = 0;
3611 events->sipi_vector = 0; /* never valid when reporting to user space */
3613 events->smi.smm = is_smm(vcpu);
3614 events->smi.pending = vcpu->arch.smi_pending;
3615 events->smi.smm_inside_nmi =
3616 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3617 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3619 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3620 | KVM_VCPUEVENT_VALID_SHADOW
3621 | KVM_VCPUEVENT_VALID_SMM);
3622 if (vcpu->kvm->arch.exception_payload_enabled)
3623 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3625 memset(&events->reserved, 0, sizeof(events->reserved));
3628 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3630 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3631 struct kvm_vcpu_events *events)
3633 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3634 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3635 | KVM_VCPUEVENT_VALID_SHADOW
3636 | KVM_VCPUEVENT_VALID_SMM
3637 | KVM_VCPUEVENT_VALID_PAYLOAD))
3640 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3641 if (!vcpu->kvm->arch.exception_payload_enabled)
3643 if (events->exception.pending)
3644 events->exception.injected = 0;
3646 events->exception_has_payload = 0;
3648 events->exception.pending = 0;
3649 events->exception_has_payload = 0;
3652 if ((events->exception.injected || events->exception.pending) &&
3653 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3656 /* INITs are latched while in SMM */
3657 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3658 (events->smi.smm || events->smi.pending) &&
3659 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3663 vcpu->arch.exception.injected = events->exception.injected;
3664 vcpu->arch.exception.pending = events->exception.pending;
3665 vcpu->arch.exception.nr = events->exception.nr;
3666 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3667 vcpu->arch.exception.error_code = events->exception.error_code;
3668 vcpu->arch.exception.has_payload = events->exception_has_payload;
3669 vcpu->arch.exception.payload = events->exception_payload;
3671 vcpu->arch.interrupt.injected = events->interrupt.injected;
3672 vcpu->arch.interrupt.nr = events->interrupt.nr;
3673 vcpu->arch.interrupt.soft = events->interrupt.soft;
3674 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3675 kvm_x86_ops->set_interrupt_shadow(vcpu,
3676 events->interrupt.shadow);
3678 vcpu->arch.nmi_injected = events->nmi.injected;
3679 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3680 vcpu->arch.nmi_pending = events->nmi.pending;
3681 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3683 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3684 lapic_in_kernel(vcpu))
3685 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3687 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3688 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3689 if (events->smi.smm)
3690 vcpu->arch.hflags |= HF_SMM_MASK;
3692 vcpu->arch.hflags &= ~HF_SMM_MASK;
3693 kvm_smm_changed(vcpu);
3696 vcpu->arch.smi_pending = events->smi.pending;
3698 if (events->smi.smm) {
3699 if (events->smi.smm_inside_nmi)
3700 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3702 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3703 if (lapic_in_kernel(vcpu)) {
3704 if (events->smi.latched_init)
3705 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3707 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3712 kvm_make_request(KVM_REQ_EVENT, vcpu);
3717 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3718 struct kvm_debugregs *dbgregs)
3722 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3723 kvm_get_dr(vcpu, 6, &val);
3725 dbgregs->dr7 = vcpu->arch.dr7;
3727 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3730 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3731 struct kvm_debugregs *dbgregs)
3736 if (dbgregs->dr6 & ~0xffffffffull)
3738 if (dbgregs->dr7 & ~0xffffffffull)
3741 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3742 kvm_update_dr0123(vcpu);
3743 vcpu->arch.dr6 = dbgregs->dr6;
3744 kvm_update_dr6(vcpu);
3745 vcpu->arch.dr7 = dbgregs->dr7;
3746 kvm_update_dr7(vcpu);
3751 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3753 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3755 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3756 u64 xstate_bv = xsave->header.xfeatures;
3760 * Copy legacy XSAVE area, to avoid complications with CPUID
3761 * leaves 0 and 1 in the loop below.
3763 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3766 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3767 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3770 * Copy each region from the possibly compacted offset to the
3771 * non-compacted offset.
3773 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3775 u64 xfeature_mask = valid & -valid;
3776 int xfeature_nr = fls64(xfeature_mask) - 1;
3777 void *src = get_xsave_addr(xsave, xfeature_nr);
3780 u32 size, offset, ecx, edx;
3781 cpuid_count(XSTATE_CPUID, xfeature_nr,
3782 &size, &offset, &ecx, &edx);
3783 if (xfeature_nr == XFEATURE_PKRU)
3784 memcpy(dest + offset, &vcpu->arch.pkru,
3785 sizeof(vcpu->arch.pkru));
3787 memcpy(dest + offset, src, size);
3791 valid -= xfeature_mask;
3795 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3797 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3798 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3802 * Copy legacy XSAVE area, to avoid complications with CPUID
3803 * leaves 0 and 1 in the loop below.
3805 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3807 /* Set XSTATE_BV and possibly XCOMP_BV. */
3808 xsave->header.xfeatures = xstate_bv;
3809 if (boot_cpu_has(X86_FEATURE_XSAVES))
3810 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3813 * Copy each region from the non-compacted offset to the
3814 * possibly compacted offset.
3816 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3818 u64 xfeature_mask = valid & -valid;
3819 int xfeature_nr = fls64(xfeature_mask) - 1;
3820 void *dest = get_xsave_addr(xsave, xfeature_nr);
3823 u32 size, offset, ecx, edx;
3824 cpuid_count(XSTATE_CPUID, xfeature_nr,
3825 &size, &offset, &ecx, &edx);
3826 if (xfeature_nr == XFEATURE_PKRU)
3827 memcpy(&vcpu->arch.pkru, src + offset,
3828 sizeof(vcpu->arch.pkru));
3830 memcpy(dest, src + offset, size);
3833 valid -= xfeature_mask;
3837 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3838 struct kvm_xsave *guest_xsave)
3840 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3841 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3842 fill_xsave((u8 *) guest_xsave->region, vcpu);
3844 memcpy(guest_xsave->region,
3845 &vcpu->arch.guest_fpu->state.fxsave,
3846 sizeof(struct fxregs_state));
3847 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3848 XFEATURE_MASK_FPSSE;
3852 #define XSAVE_MXCSR_OFFSET 24
3854 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3855 struct kvm_xsave *guest_xsave)
3858 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3859 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3861 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3863 * Here we allow setting states that are not present in
3864 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3865 * with old userspace.
3867 if (xstate_bv & ~kvm_supported_xcr0() ||
3868 mxcsr & ~mxcsr_feature_mask)
3870 load_xsave(vcpu, (u8 *)guest_xsave->region);
3872 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3873 mxcsr & ~mxcsr_feature_mask)
3875 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3876 guest_xsave->region, sizeof(struct fxregs_state));
3881 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3882 struct kvm_xcrs *guest_xcrs)
3884 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3885 guest_xcrs->nr_xcrs = 0;
3889 guest_xcrs->nr_xcrs = 1;
3890 guest_xcrs->flags = 0;
3891 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3892 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3895 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3896 struct kvm_xcrs *guest_xcrs)
3900 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3903 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3906 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3907 /* Only support XCR0 currently */
3908 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3909 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3910 guest_xcrs->xcrs[i].value);
3919 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3920 * stopped by the hypervisor. This function will be called from the host only.
3921 * EINVAL is returned when the host attempts to set the flag for a guest that
3922 * does not support pv clocks.
3924 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3926 if (!vcpu->arch.pv_time_enabled)
3928 vcpu->arch.pvclock_set_guest_stopped_request = true;
3929 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3933 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3934 struct kvm_enable_cap *cap)
3937 uint16_t vmcs_version;
3938 void __user *user_ptr;
3944 case KVM_CAP_HYPERV_SYNIC2:
3949 case KVM_CAP_HYPERV_SYNIC:
3950 if (!irqchip_in_kernel(vcpu->kvm))
3952 return kvm_hv_activate_synic(vcpu, cap->cap ==
3953 KVM_CAP_HYPERV_SYNIC2);
3954 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3955 if (!kvm_x86_ops->nested_enable_evmcs)
3957 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
3959 user_ptr = (void __user *)(uintptr_t)cap->args[0];
3960 if (copy_to_user(user_ptr, &vmcs_version,
3961 sizeof(vmcs_version)))
3971 long kvm_arch_vcpu_ioctl(struct file *filp,
3972 unsigned int ioctl, unsigned long arg)
3974 struct kvm_vcpu *vcpu = filp->private_data;
3975 void __user *argp = (void __user *)arg;
3978 struct kvm_lapic_state *lapic;
3979 struct kvm_xsave *xsave;
3980 struct kvm_xcrs *xcrs;
3988 case KVM_GET_LAPIC: {
3990 if (!lapic_in_kernel(vcpu))
3992 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
3993 GFP_KERNEL_ACCOUNT);
3998 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4002 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4007 case KVM_SET_LAPIC: {
4009 if (!lapic_in_kernel(vcpu))
4011 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4012 if (IS_ERR(u.lapic)) {
4013 r = PTR_ERR(u.lapic);
4017 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4020 case KVM_INTERRUPT: {
4021 struct kvm_interrupt irq;
4024 if (copy_from_user(&irq, argp, sizeof(irq)))
4026 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4030 r = kvm_vcpu_ioctl_nmi(vcpu);
4034 r = kvm_vcpu_ioctl_smi(vcpu);
4037 case KVM_SET_CPUID: {
4038 struct kvm_cpuid __user *cpuid_arg = argp;
4039 struct kvm_cpuid cpuid;
4042 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4044 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4047 case KVM_SET_CPUID2: {
4048 struct kvm_cpuid2 __user *cpuid_arg = argp;
4049 struct kvm_cpuid2 cpuid;
4052 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4054 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4055 cpuid_arg->entries);
4058 case KVM_GET_CPUID2: {
4059 struct kvm_cpuid2 __user *cpuid_arg = argp;
4060 struct kvm_cpuid2 cpuid;
4063 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4065 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4066 cpuid_arg->entries);
4070 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4075 case KVM_GET_MSRS: {
4076 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4077 r = msr_io(vcpu, argp, do_get_msr, 1);
4078 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4081 case KVM_SET_MSRS: {
4082 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4083 r = msr_io(vcpu, argp, do_set_msr, 0);
4084 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4087 case KVM_TPR_ACCESS_REPORTING: {
4088 struct kvm_tpr_access_ctl tac;
4091 if (copy_from_user(&tac, argp, sizeof(tac)))
4093 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4097 if (copy_to_user(argp, &tac, sizeof(tac)))
4102 case KVM_SET_VAPIC_ADDR: {
4103 struct kvm_vapic_addr va;
4107 if (!lapic_in_kernel(vcpu))
4110 if (copy_from_user(&va, argp, sizeof(va)))
4112 idx = srcu_read_lock(&vcpu->kvm->srcu);
4113 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4114 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4117 case KVM_X86_SETUP_MCE: {
4121 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4123 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4126 case KVM_X86_SET_MCE: {
4127 struct kvm_x86_mce mce;
4130 if (copy_from_user(&mce, argp, sizeof(mce)))
4132 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4135 case KVM_GET_VCPU_EVENTS: {
4136 struct kvm_vcpu_events events;
4138 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4141 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4146 case KVM_SET_VCPU_EVENTS: {
4147 struct kvm_vcpu_events events;
4150 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4153 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4156 case KVM_GET_DEBUGREGS: {
4157 struct kvm_debugregs dbgregs;
4159 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4162 if (copy_to_user(argp, &dbgregs,
4163 sizeof(struct kvm_debugregs)))
4168 case KVM_SET_DEBUGREGS: {
4169 struct kvm_debugregs dbgregs;
4172 if (copy_from_user(&dbgregs, argp,
4173 sizeof(struct kvm_debugregs)))
4176 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4179 case KVM_GET_XSAVE: {
4180 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4185 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4188 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4193 case KVM_SET_XSAVE: {
4194 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4195 if (IS_ERR(u.xsave)) {
4196 r = PTR_ERR(u.xsave);
4200 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4203 case KVM_GET_XCRS: {
4204 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4209 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4212 if (copy_to_user(argp, u.xcrs,
4213 sizeof(struct kvm_xcrs)))
4218 case KVM_SET_XCRS: {
4219 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4220 if (IS_ERR(u.xcrs)) {
4221 r = PTR_ERR(u.xcrs);
4225 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4228 case KVM_SET_TSC_KHZ: {
4232 user_tsc_khz = (u32)arg;
4234 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4237 if (user_tsc_khz == 0)
4238 user_tsc_khz = tsc_khz;
4240 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4245 case KVM_GET_TSC_KHZ: {
4246 r = vcpu->arch.virtual_tsc_khz;
4249 case KVM_KVMCLOCK_CTRL: {
4250 r = kvm_set_guest_paused(vcpu);
4253 case KVM_ENABLE_CAP: {
4254 struct kvm_enable_cap cap;
4257 if (copy_from_user(&cap, argp, sizeof(cap)))
4259 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4262 case KVM_GET_NESTED_STATE: {
4263 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4267 if (!kvm_x86_ops->get_nested_state)
4270 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4272 if (get_user(user_data_size, &user_kvm_nested_state->size))
4275 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4280 if (r > user_data_size) {
4281 if (put_user(r, &user_kvm_nested_state->size))
4291 case KVM_SET_NESTED_STATE: {
4292 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4293 struct kvm_nested_state kvm_state;
4296 if (!kvm_x86_ops->set_nested_state)
4300 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4304 if (kvm_state.size < sizeof(kvm_state))
4307 if (kvm_state.flags &
4308 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4309 | KVM_STATE_NESTED_EVMCS))
4312 /* nested_run_pending implies guest_mode. */
4313 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4314 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4317 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4320 case KVM_GET_SUPPORTED_HV_CPUID: {
4321 struct kvm_cpuid2 __user *cpuid_arg = argp;
4322 struct kvm_cpuid2 cpuid;
4325 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4328 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4329 cpuid_arg->entries);
4334 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4349 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4351 return VM_FAULT_SIGBUS;
4354 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4358 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4360 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4364 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4367 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4370 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4371 unsigned long kvm_nr_mmu_pages)
4373 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4376 mutex_lock(&kvm->slots_lock);
4378 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4379 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4381 mutex_unlock(&kvm->slots_lock);
4385 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4387 return kvm->arch.n_max_mmu_pages;
4390 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4392 struct kvm_pic *pic = kvm->arch.vpic;
4396 switch (chip->chip_id) {
4397 case KVM_IRQCHIP_PIC_MASTER:
4398 memcpy(&chip->chip.pic, &pic->pics[0],
4399 sizeof(struct kvm_pic_state));
4401 case KVM_IRQCHIP_PIC_SLAVE:
4402 memcpy(&chip->chip.pic, &pic->pics[1],
4403 sizeof(struct kvm_pic_state));
4405 case KVM_IRQCHIP_IOAPIC:
4406 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4415 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4417 struct kvm_pic *pic = kvm->arch.vpic;
4421 switch (chip->chip_id) {
4422 case KVM_IRQCHIP_PIC_MASTER:
4423 spin_lock(&pic->lock);
4424 memcpy(&pic->pics[0], &chip->chip.pic,
4425 sizeof(struct kvm_pic_state));
4426 spin_unlock(&pic->lock);
4428 case KVM_IRQCHIP_PIC_SLAVE:
4429 spin_lock(&pic->lock);
4430 memcpy(&pic->pics[1], &chip->chip.pic,
4431 sizeof(struct kvm_pic_state));
4432 spin_unlock(&pic->lock);
4434 case KVM_IRQCHIP_IOAPIC:
4435 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4441 kvm_pic_update_irq(pic);
4445 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4447 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4449 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4451 mutex_lock(&kps->lock);
4452 memcpy(ps, &kps->channels, sizeof(*ps));
4453 mutex_unlock(&kps->lock);
4457 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4460 struct kvm_pit *pit = kvm->arch.vpit;
4462 mutex_lock(&pit->pit_state.lock);
4463 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4464 for (i = 0; i < 3; i++)
4465 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4466 mutex_unlock(&pit->pit_state.lock);
4470 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4472 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4473 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4474 sizeof(ps->channels));
4475 ps->flags = kvm->arch.vpit->pit_state.flags;
4476 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4477 memset(&ps->reserved, 0, sizeof(ps->reserved));
4481 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4485 u32 prev_legacy, cur_legacy;
4486 struct kvm_pit *pit = kvm->arch.vpit;
4488 mutex_lock(&pit->pit_state.lock);
4489 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4490 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4491 if (!prev_legacy && cur_legacy)
4493 memcpy(&pit->pit_state.channels, &ps->channels,
4494 sizeof(pit->pit_state.channels));
4495 pit->pit_state.flags = ps->flags;
4496 for (i = 0; i < 3; i++)
4497 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4499 mutex_unlock(&pit->pit_state.lock);
4503 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4504 struct kvm_reinject_control *control)
4506 struct kvm_pit *pit = kvm->arch.vpit;
4511 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4512 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4513 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4515 mutex_lock(&pit->pit_state.lock);
4516 kvm_pit_set_reinject(pit, control->pit_reinject);
4517 mutex_unlock(&pit->pit_state.lock);
4523 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4524 * @kvm: kvm instance
4525 * @log: slot id and address to which we copy the log
4527 * Steps 1-4 below provide general overview of dirty page logging. See
4528 * kvm_get_dirty_log_protect() function description for additional details.
4530 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4531 * always flush the TLB (step 4) even if previous step failed and the dirty
4532 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4533 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4534 * writes will be marked dirty for next log read.
4536 * 1. Take a snapshot of the bit and clear it if needed.
4537 * 2. Write protect the corresponding page.
4538 * 3. Copy the snapshot to the userspace.
4539 * 4. Flush TLB's if needed.
4541 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4546 mutex_lock(&kvm->slots_lock);
4549 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4551 if (kvm_x86_ops->flush_log_dirty)
4552 kvm_x86_ops->flush_log_dirty(kvm);
4554 r = kvm_get_dirty_log_protect(kvm, log, &flush);
4557 * All the TLBs can be flushed out of mmu lock, see the comments in
4558 * kvm_mmu_slot_remove_write_access().
4560 lockdep_assert_held(&kvm->slots_lock);
4562 kvm_flush_remote_tlbs(kvm);
4564 mutex_unlock(&kvm->slots_lock);
4568 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4573 mutex_lock(&kvm->slots_lock);
4576 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4578 if (kvm_x86_ops->flush_log_dirty)
4579 kvm_x86_ops->flush_log_dirty(kvm);
4581 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4584 * All the TLBs can be flushed out of mmu lock, see the comments in
4585 * kvm_mmu_slot_remove_write_access().
4587 lockdep_assert_held(&kvm->slots_lock);
4589 kvm_flush_remote_tlbs(kvm);
4591 mutex_unlock(&kvm->slots_lock);
4595 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4598 if (!irqchip_in_kernel(kvm))
4601 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4602 irq_event->irq, irq_event->level,
4607 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4608 struct kvm_enable_cap *cap)
4616 case KVM_CAP_DISABLE_QUIRKS:
4617 kvm->arch.disabled_quirks = cap->args[0];
4620 case KVM_CAP_SPLIT_IRQCHIP: {
4621 mutex_lock(&kvm->lock);
4623 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4624 goto split_irqchip_unlock;
4626 if (irqchip_in_kernel(kvm))
4627 goto split_irqchip_unlock;
4628 if (kvm->created_vcpus)
4629 goto split_irqchip_unlock;
4630 r = kvm_setup_empty_irq_routing(kvm);
4632 goto split_irqchip_unlock;
4633 /* Pairs with irqchip_in_kernel. */
4635 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4636 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4638 split_irqchip_unlock:
4639 mutex_unlock(&kvm->lock);
4642 case KVM_CAP_X2APIC_API:
4644 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4647 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4648 kvm->arch.x2apic_format = true;
4649 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4650 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4654 case KVM_CAP_X86_DISABLE_EXITS:
4656 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4659 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4660 kvm_can_mwait_in_guest())
4661 kvm->arch.mwait_in_guest = true;
4662 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4663 kvm->arch.hlt_in_guest = true;
4664 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4665 kvm->arch.pause_in_guest = true;
4666 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4667 kvm->arch.cstate_in_guest = true;
4670 case KVM_CAP_MSR_PLATFORM_INFO:
4671 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4674 case KVM_CAP_EXCEPTION_PAYLOAD:
4675 kvm->arch.exception_payload_enabled = cap->args[0];
4685 long kvm_arch_vm_ioctl(struct file *filp,
4686 unsigned int ioctl, unsigned long arg)
4688 struct kvm *kvm = filp->private_data;
4689 void __user *argp = (void __user *)arg;
4692 * This union makes it completely explicit to gcc-3.x
4693 * that these two variables' stack usage should be
4694 * combined, not added together.
4697 struct kvm_pit_state ps;
4698 struct kvm_pit_state2 ps2;
4699 struct kvm_pit_config pit_config;
4703 case KVM_SET_TSS_ADDR:
4704 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4706 case KVM_SET_IDENTITY_MAP_ADDR: {
4709 mutex_lock(&kvm->lock);
4711 if (kvm->created_vcpus)
4712 goto set_identity_unlock;
4714 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4715 goto set_identity_unlock;
4716 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4717 set_identity_unlock:
4718 mutex_unlock(&kvm->lock);
4721 case KVM_SET_NR_MMU_PAGES:
4722 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4724 case KVM_GET_NR_MMU_PAGES:
4725 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4727 case KVM_CREATE_IRQCHIP: {
4728 mutex_lock(&kvm->lock);
4731 if (irqchip_in_kernel(kvm))
4732 goto create_irqchip_unlock;
4735 if (kvm->created_vcpus)
4736 goto create_irqchip_unlock;
4738 r = kvm_pic_init(kvm);
4740 goto create_irqchip_unlock;
4742 r = kvm_ioapic_init(kvm);
4744 kvm_pic_destroy(kvm);
4745 goto create_irqchip_unlock;
4748 r = kvm_setup_default_irq_routing(kvm);
4750 kvm_ioapic_destroy(kvm);
4751 kvm_pic_destroy(kvm);
4752 goto create_irqchip_unlock;
4754 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4756 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4757 create_irqchip_unlock:
4758 mutex_unlock(&kvm->lock);
4761 case KVM_CREATE_PIT:
4762 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4764 case KVM_CREATE_PIT2:
4766 if (copy_from_user(&u.pit_config, argp,
4767 sizeof(struct kvm_pit_config)))
4770 mutex_lock(&kvm->lock);
4773 goto create_pit_unlock;
4775 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4779 mutex_unlock(&kvm->lock);
4781 case KVM_GET_IRQCHIP: {
4782 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4783 struct kvm_irqchip *chip;
4785 chip = memdup_user(argp, sizeof(*chip));
4792 if (!irqchip_kernel(kvm))
4793 goto get_irqchip_out;
4794 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4796 goto get_irqchip_out;
4798 if (copy_to_user(argp, chip, sizeof(*chip)))
4799 goto get_irqchip_out;
4805 case KVM_SET_IRQCHIP: {
4806 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4807 struct kvm_irqchip *chip;
4809 chip = memdup_user(argp, sizeof(*chip));
4816 if (!irqchip_kernel(kvm))
4817 goto set_irqchip_out;
4818 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4820 goto set_irqchip_out;
4828 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4831 if (!kvm->arch.vpit)
4833 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4837 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4844 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4847 if (!kvm->arch.vpit)
4849 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4852 case KVM_GET_PIT2: {
4854 if (!kvm->arch.vpit)
4856 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4860 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4865 case KVM_SET_PIT2: {
4867 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4870 if (!kvm->arch.vpit)
4872 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4875 case KVM_REINJECT_CONTROL: {
4876 struct kvm_reinject_control control;
4878 if (copy_from_user(&control, argp, sizeof(control)))
4880 r = kvm_vm_ioctl_reinject(kvm, &control);
4883 case KVM_SET_BOOT_CPU_ID:
4885 mutex_lock(&kvm->lock);
4886 if (kvm->created_vcpus)
4889 kvm->arch.bsp_vcpu_id = arg;
4890 mutex_unlock(&kvm->lock);
4892 case KVM_XEN_HVM_CONFIG: {
4893 struct kvm_xen_hvm_config xhc;
4895 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4900 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4904 case KVM_SET_CLOCK: {
4905 struct kvm_clock_data user_ns;
4909 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4918 * TODO: userspace has to take care of races with VCPU_RUN, so
4919 * kvm_gen_update_masterclock() can be cut down to locked
4920 * pvclock_update_vm_gtod_copy().
4922 kvm_gen_update_masterclock(kvm);
4923 now_ns = get_kvmclock_ns(kvm);
4924 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4925 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4928 case KVM_GET_CLOCK: {
4929 struct kvm_clock_data user_ns;
4932 now_ns = get_kvmclock_ns(kvm);
4933 user_ns.clock = now_ns;
4934 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4935 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4938 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4943 case KVM_MEMORY_ENCRYPT_OP: {
4945 if (kvm_x86_ops->mem_enc_op)
4946 r = kvm_x86_ops->mem_enc_op(kvm, argp);
4949 case KVM_MEMORY_ENCRYPT_REG_REGION: {
4950 struct kvm_enc_region region;
4953 if (copy_from_user(®ion, argp, sizeof(region)))
4957 if (kvm_x86_ops->mem_enc_reg_region)
4958 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
4961 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4962 struct kvm_enc_region region;
4965 if (copy_from_user(®ion, argp, sizeof(region)))
4969 if (kvm_x86_ops->mem_enc_unreg_region)
4970 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
4973 case KVM_HYPERV_EVENTFD: {
4974 struct kvm_hyperv_eventfd hvevfd;
4977 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4979 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4982 case KVM_SET_PMU_EVENT_FILTER:
4983 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
4992 static void kvm_init_msr_list(void)
4997 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4998 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
5002 * Even MSRs that are valid in the host may not be exposed
5003 * to the guests in some cases.
5005 switch (msrs_to_save[i]) {
5006 case MSR_IA32_BNDCFGS:
5007 if (!kvm_mpx_supported())
5011 if (!kvm_x86_ops->rdtscp_supported())
5014 case MSR_IA32_RTIT_CTL:
5015 case MSR_IA32_RTIT_STATUS:
5016 if (!kvm_x86_ops->pt_supported())
5019 case MSR_IA32_RTIT_CR3_MATCH:
5020 if (!kvm_x86_ops->pt_supported() ||
5021 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5024 case MSR_IA32_RTIT_OUTPUT_BASE:
5025 case MSR_IA32_RTIT_OUTPUT_MASK:
5026 if (!kvm_x86_ops->pt_supported() ||
5027 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5028 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5031 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5032 if (!kvm_x86_ops->pt_supported() ||
5033 msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
5034 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5043 msrs_to_save[j] = msrs_to_save[i];
5046 num_msrs_to_save = j;
5048 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
5049 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
5053 emulated_msrs[j] = emulated_msrs[i];
5056 num_emulated_msrs = j;
5058 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
5059 struct kvm_msr_entry msr;
5061 msr.index = msr_based_features[i];
5062 if (kvm_get_msr_feature(&msr))
5066 msr_based_features[j] = msr_based_features[i];
5069 num_msr_based_features = j;
5072 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5080 if (!(lapic_in_kernel(vcpu) &&
5081 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5082 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5093 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5100 if (!(lapic_in_kernel(vcpu) &&
5101 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5103 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5105 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5115 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5116 struct kvm_segment *var, int seg)
5118 kvm_x86_ops->set_segment(vcpu, var, seg);
5121 void kvm_get_segment(struct kvm_vcpu *vcpu,
5122 struct kvm_segment *var, int seg)
5124 kvm_x86_ops->get_segment(vcpu, var, seg);
5127 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5128 struct x86_exception *exception)
5132 BUG_ON(!mmu_is_nested(vcpu));
5134 /* NPT walks are always user-walks */
5135 access |= PFERR_USER_MASK;
5136 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5141 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5142 struct x86_exception *exception)
5144 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5145 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5148 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5149 struct x86_exception *exception)
5151 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5152 access |= PFERR_FETCH_MASK;
5153 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5156 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5157 struct x86_exception *exception)
5159 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5160 access |= PFERR_WRITE_MASK;
5161 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5164 /* uses this to access any guest's mapped memory without checking CPL */
5165 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5166 struct x86_exception *exception)
5168 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5171 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5172 struct kvm_vcpu *vcpu, u32 access,
5173 struct x86_exception *exception)
5176 int r = X86EMUL_CONTINUE;
5179 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5181 unsigned offset = addr & (PAGE_SIZE-1);
5182 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5185 if (gpa == UNMAPPED_GVA)
5186 return X86EMUL_PROPAGATE_FAULT;
5187 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5190 r = X86EMUL_IO_NEEDED;
5202 /* used for instruction fetching */
5203 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5204 gva_t addr, void *val, unsigned int bytes,
5205 struct x86_exception *exception)
5207 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5208 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5212 /* Inline kvm_read_guest_virt_helper for speed. */
5213 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5215 if (unlikely(gpa == UNMAPPED_GVA))
5216 return X86EMUL_PROPAGATE_FAULT;
5218 offset = addr & (PAGE_SIZE-1);
5219 if (WARN_ON(offset + bytes > PAGE_SIZE))
5220 bytes = (unsigned)PAGE_SIZE - offset;
5221 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5223 if (unlikely(ret < 0))
5224 return X86EMUL_IO_NEEDED;
5226 return X86EMUL_CONTINUE;
5229 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5230 gva_t addr, void *val, unsigned int bytes,
5231 struct x86_exception *exception)
5233 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5236 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5237 * is returned, but our callers are not ready for that and they blindly
5238 * call kvm_inject_page_fault. Ensure that they at least do not leak
5239 * uninitialized kernel stack memory into cr2 and error code.
5241 memset(exception, 0, sizeof(*exception));
5242 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5245 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5247 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5248 gva_t addr, void *val, unsigned int bytes,
5249 struct x86_exception *exception, bool system)
5251 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5254 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5255 access |= PFERR_USER_MASK;
5257 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5260 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5261 unsigned long addr, void *val, unsigned int bytes)
5263 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5264 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5266 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5269 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5270 struct kvm_vcpu *vcpu, u32 access,
5271 struct x86_exception *exception)
5274 int r = X86EMUL_CONTINUE;
5277 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5280 unsigned offset = addr & (PAGE_SIZE-1);
5281 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5284 if (gpa == UNMAPPED_GVA)
5285 return X86EMUL_PROPAGATE_FAULT;
5286 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5288 r = X86EMUL_IO_NEEDED;
5300 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5301 unsigned int bytes, struct x86_exception *exception,
5304 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5305 u32 access = PFERR_WRITE_MASK;
5307 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5308 access |= PFERR_USER_MASK;
5310 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5314 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5315 unsigned int bytes, struct x86_exception *exception)
5317 /* kvm_write_guest_virt_system can pull in tons of pages. */
5318 vcpu->arch.l1tf_flush_l1d = true;
5320 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5321 PFERR_WRITE_MASK, exception);
5323 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5325 int handle_ud(struct kvm_vcpu *vcpu)
5327 int emul_type = EMULTYPE_TRAP_UD;
5328 enum emulation_result er;
5329 char sig[5]; /* ud2; .ascii "kvm" */
5330 struct x86_exception e;
5332 if (force_emulation_prefix &&
5333 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5334 sig, sizeof(sig), &e) == 0 &&
5335 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5336 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5340 er = kvm_emulate_instruction(vcpu, emul_type);
5341 if (er == EMULATE_USER_EXIT)
5343 if (er != EMULATE_DONE)
5344 kvm_queue_exception(vcpu, UD_VECTOR);
5347 EXPORT_SYMBOL_GPL(handle_ud);
5349 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5350 gpa_t gpa, bool write)
5352 /* For APIC access vmexit */
5353 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5356 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5357 trace_vcpu_match_mmio(gva, gpa, write, true);
5364 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5365 gpa_t *gpa, struct x86_exception *exception,
5368 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5369 | (write ? PFERR_WRITE_MASK : 0);
5372 * currently PKRU is only applied to ept enabled guest so
5373 * there is no pkey in EPT page table for L1 guest or EPT
5374 * shadow page table for L2 guest.
5376 if (vcpu_match_mmio_gva(vcpu, gva)
5377 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5378 vcpu->arch.access, 0, access)) {
5379 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5380 (gva & (PAGE_SIZE - 1));
5381 trace_vcpu_match_mmio(gva, *gpa, write, false);
5385 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5387 if (*gpa == UNMAPPED_GVA)
5390 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5393 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5394 const void *val, int bytes)
5398 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5401 kvm_page_track_write(vcpu, gpa, val, bytes);
5405 struct read_write_emulator_ops {
5406 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5408 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5409 void *val, int bytes);
5410 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5411 int bytes, void *val);
5412 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5413 void *val, int bytes);
5417 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5419 if (vcpu->mmio_read_completed) {
5420 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5421 vcpu->mmio_fragments[0].gpa, val);
5422 vcpu->mmio_read_completed = 0;
5429 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5430 void *val, int bytes)
5432 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5435 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5436 void *val, int bytes)
5438 return emulator_write_phys(vcpu, gpa, val, bytes);
5441 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5443 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5444 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5447 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5448 void *val, int bytes)
5450 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5451 return X86EMUL_IO_NEEDED;
5454 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5455 void *val, int bytes)
5457 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5459 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5460 return X86EMUL_CONTINUE;
5463 static const struct read_write_emulator_ops read_emultor = {
5464 .read_write_prepare = read_prepare,
5465 .read_write_emulate = read_emulate,
5466 .read_write_mmio = vcpu_mmio_read,
5467 .read_write_exit_mmio = read_exit_mmio,
5470 static const struct read_write_emulator_ops write_emultor = {
5471 .read_write_emulate = write_emulate,
5472 .read_write_mmio = write_mmio,
5473 .read_write_exit_mmio = write_exit_mmio,
5477 static int emulator_read_write_onepage(unsigned long addr, void *val,
5479 struct x86_exception *exception,
5480 struct kvm_vcpu *vcpu,
5481 const struct read_write_emulator_ops *ops)
5485 bool write = ops->write;
5486 struct kvm_mmio_fragment *frag;
5487 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5490 * If the exit was due to a NPF we may already have a GPA.
5491 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5492 * Note, this cannot be used on string operations since string
5493 * operation using rep will only have the initial GPA from the NPF
5496 if (vcpu->arch.gpa_available &&
5497 emulator_can_use_gpa(ctxt) &&
5498 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5499 gpa = vcpu->arch.gpa_val;
5500 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5502 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5504 return X86EMUL_PROPAGATE_FAULT;
5507 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5508 return X86EMUL_CONTINUE;
5511 * Is this MMIO handled locally?
5513 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5514 if (handled == bytes)
5515 return X86EMUL_CONTINUE;
5521 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5522 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5526 return X86EMUL_CONTINUE;
5529 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5531 void *val, unsigned int bytes,
5532 struct x86_exception *exception,
5533 const struct read_write_emulator_ops *ops)
5535 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5539 if (ops->read_write_prepare &&
5540 ops->read_write_prepare(vcpu, val, bytes))
5541 return X86EMUL_CONTINUE;
5543 vcpu->mmio_nr_fragments = 0;
5545 /* Crossing a page boundary? */
5546 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5549 now = -addr & ~PAGE_MASK;
5550 rc = emulator_read_write_onepage(addr, val, now, exception,
5553 if (rc != X86EMUL_CONTINUE)
5556 if (ctxt->mode != X86EMUL_MODE_PROT64)
5562 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5564 if (rc != X86EMUL_CONTINUE)
5567 if (!vcpu->mmio_nr_fragments)
5570 gpa = vcpu->mmio_fragments[0].gpa;
5572 vcpu->mmio_needed = 1;
5573 vcpu->mmio_cur_fragment = 0;
5575 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5576 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5577 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5578 vcpu->run->mmio.phys_addr = gpa;
5580 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5583 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5587 struct x86_exception *exception)
5589 return emulator_read_write(ctxt, addr, val, bytes,
5590 exception, &read_emultor);
5593 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5597 struct x86_exception *exception)
5599 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5600 exception, &write_emultor);
5603 #define CMPXCHG_TYPE(t, ptr, old, new) \
5604 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5606 #ifdef CONFIG_X86_64
5607 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5609 # define CMPXCHG64(ptr, old, new) \
5610 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5613 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5618 struct x86_exception *exception)
5620 struct kvm_host_map map;
5621 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5626 /* guests cmpxchg8b have to be emulated atomically */
5627 if (bytes > 8 || (bytes & (bytes - 1)))
5630 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5632 if (gpa == UNMAPPED_GVA ||
5633 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5636 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5639 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5642 kaddr = map.hva + offset_in_page(gpa);
5646 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5649 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5652 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5655 exchanged = CMPXCHG64(kaddr, old, new);
5661 kvm_vcpu_unmap(vcpu, &map, true);
5664 return X86EMUL_CMPXCHG_FAILED;
5666 kvm_page_track_write(vcpu, gpa, new, bytes);
5668 return X86EMUL_CONTINUE;
5671 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5673 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5676 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5680 for (i = 0; i < vcpu->arch.pio.count; i++) {
5681 if (vcpu->arch.pio.in)
5682 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5683 vcpu->arch.pio.size, pd);
5685 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5686 vcpu->arch.pio.port, vcpu->arch.pio.size,
5690 pd += vcpu->arch.pio.size;
5695 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5696 unsigned short port, void *val,
5697 unsigned int count, bool in)
5699 vcpu->arch.pio.port = port;
5700 vcpu->arch.pio.in = in;
5701 vcpu->arch.pio.count = count;
5702 vcpu->arch.pio.size = size;
5704 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5705 vcpu->arch.pio.count = 0;
5709 vcpu->run->exit_reason = KVM_EXIT_IO;
5710 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5711 vcpu->run->io.size = size;
5712 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5713 vcpu->run->io.count = count;
5714 vcpu->run->io.port = port;
5719 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5720 int size, unsigned short port, void *val,
5723 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5726 if (vcpu->arch.pio.count)
5729 memset(vcpu->arch.pio_data, 0, size * count);
5731 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5734 memcpy(val, vcpu->arch.pio_data, size * count);
5735 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5736 vcpu->arch.pio.count = 0;
5743 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5744 int size, unsigned short port,
5745 const void *val, unsigned int count)
5747 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5749 memcpy(vcpu->arch.pio_data, val, size * count);
5750 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5751 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5754 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5756 return kvm_x86_ops->get_segment_base(vcpu, seg);
5759 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5761 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5764 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5766 if (!need_emulate_wbinvd(vcpu))
5767 return X86EMUL_CONTINUE;
5769 if (kvm_x86_ops->has_wbinvd_exit()) {
5770 int cpu = get_cpu();
5772 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5773 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5774 wbinvd_ipi, NULL, 1);
5776 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5779 return X86EMUL_CONTINUE;
5782 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5784 kvm_emulate_wbinvd_noskip(vcpu);
5785 return kvm_skip_emulated_instruction(vcpu);
5787 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5791 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5793 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5796 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5797 unsigned long *dest)
5799 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5802 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5803 unsigned long value)
5806 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5809 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5811 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5814 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5816 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5817 unsigned long value;
5821 value = kvm_read_cr0(vcpu);
5824 value = vcpu->arch.cr2;
5827 value = kvm_read_cr3(vcpu);
5830 value = kvm_read_cr4(vcpu);
5833 value = kvm_get_cr8(vcpu);
5836 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5843 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5845 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5850 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5853 vcpu->arch.cr2 = val;
5856 res = kvm_set_cr3(vcpu, val);
5859 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5862 res = kvm_set_cr8(vcpu, val);
5865 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5872 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5874 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5877 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5879 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5882 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5884 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5887 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5889 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5892 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5894 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5897 static unsigned long emulator_get_cached_segment_base(
5898 struct x86_emulate_ctxt *ctxt, int seg)
5900 return get_segment_base(emul_to_vcpu(ctxt), seg);
5903 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5904 struct desc_struct *desc, u32 *base3,
5907 struct kvm_segment var;
5909 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5910 *selector = var.selector;
5913 memset(desc, 0, sizeof(*desc));
5921 set_desc_limit(desc, var.limit);
5922 set_desc_base(desc, (unsigned long)var.base);
5923 #ifdef CONFIG_X86_64
5925 *base3 = var.base >> 32;
5927 desc->type = var.type;
5929 desc->dpl = var.dpl;
5930 desc->p = var.present;
5931 desc->avl = var.avl;
5939 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5940 struct desc_struct *desc, u32 base3,
5943 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5944 struct kvm_segment var;
5946 var.selector = selector;
5947 var.base = get_desc_base(desc);
5948 #ifdef CONFIG_X86_64
5949 var.base |= ((u64)base3) << 32;
5951 var.limit = get_desc_limit(desc);
5953 var.limit = (var.limit << 12) | 0xfff;
5954 var.type = desc->type;
5955 var.dpl = desc->dpl;
5960 var.avl = desc->avl;
5961 var.present = desc->p;
5962 var.unusable = !var.present;
5965 kvm_set_segment(vcpu, &var, seg);
5969 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5970 u32 msr_index, u64 *pdata)
5972 struct msr_data msr;
5975 msr.index = msr_index;
5976 msr.host_initiated = false;
5977 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5985 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5986 u32 msr_index, u64 data)
5988 struct msr_data msr;
5991 msr.index = msr_index;
5992 msr.host_initiated = false;
5993 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5996 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5998 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6000 return vcpu->arch.smbase;
6003 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6005 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6007 vcpu->arch.smbase = smbase;
6010 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6013 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
6016 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6017 u32 pmc, u64 *pdata)
6019 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6022 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6024 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6027 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6028 struct x86_instruction_info *info,
6029 enum x86_intercept_stage stage)
6031 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6034 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6035 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6037 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6040 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6042 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6045 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6047 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6050 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6052 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6055 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6057 return emul_to_vcpu(ctxt)->arch.hflags;
6060 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6062 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6065 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6066 const char *smstate)
6068 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6071 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6073 kvm_smm_changed(emul_to_vcpu(ctxt));
6076 static const struct x86_emulate_ops emulate_ops = {
6077 .read_gpr = emulator_read_gpr,
6078 .write_gpr = emulator_write_gpr,
6079 .read_std = emulator_read_std,
6080 .write_std = emulator_write_std,
6081 .read_phys = kvm_read_guest_phys_system,
6082 .fetch = kvm_fetch_guest_virt,
6083 .read_emulated = emulator_read_emulated,
6084 .write_emulated = emulator_write_emulated,
6085 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6086 .invlpg = emulator_invlpg,
6087 .pio_in_emulated = emulator_pio_in_emulated,
6088 .pio_out_emulated = emulator_pio_out_emulated,
6089 .get_segment = emulator_get_segment,
6090 .set_segment = emulator_set_segment,
6091 .get_cached_segment_base = emulator_get_cached_segment_base,
6092 .get_gdt = emulator_get_gdt,
6093 .get_idt = emulator_get_idt,
6094 .set_gdt = emulator_set_gdt,
6095 .set_idt = emulator_set_idt,
6096 .get_cr = emulator_get_cr,
6097 .set_cr = emulator_set_cr,
6098 .cpl = emulator_get_cpl,
6099 .get_dr = emulator_get_dr,
6100 .set_dr = emulator_set_dr,
6101 .get_smbase = emulator_get_smbase,
6102 .set_smbase = emulator_set_smbase,
6103 .set_msr = emulator_set_msr,
6104 .get_msr = emulator_get_msr,
6105 .check_pmc = emulator_check_pmc,
6106 .read_pmc = emulator_read_pmc,
6107 .halt = emulator_halt,
6108 .wbinvd = emulator_wbinvd,
6109 .fix_hypercall = emulator_fix_hypercall,
6110 .intercept = emulator_intercept,
6111 .get_cpuid = emulator_get_cpuid,
6112 .set_nmi_mask = emulator_set_nmi_mask,
6113 .get_hflags = emulator_get_hflags,
6114 .set_hflags = emulator_set_hflags,
6115 .pre_leave_smm = emulator_pre_leave_smm,
6116 .post_leave_smm = emulator_post_leave_smm,
6119 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6121 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6123 * an sti; sti; sequence only disable interrupts for the first
6124 * instruction. So, if the last instruction, be it emulated or
6125 * not, left the system with the INT_STI flag enabled, it
6126 * means that the last instruction is an sti. We should not
6127 * leave the flag on in this case. The same goes for mov ss
6129 if (int_shadow & mask)
6131 if (unlikely(int_shadow || mask)) {
6132 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6134 kvm_make_request(KVM_REQ_EVENT, vcpu);
6138 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6140 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6141 if (ctxt->exception.vector == PF_VECTOR)
6142 return kvm_propagate_fault(vcpu, &ctxt->exception);
6144 if (ctxt->exception.error_code_valid)
6145 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6146 ctxt->exception.error_code);
6148 kvm_queue_exception(vcpu, ctxt->exception.vector);
6152 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6154 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6157 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6159 ctxt->eflags = kvm_get_rflags(vcpu);
6160 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6162 ctxt->eip = kvm_rip_read(vcpu);
6163 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6164 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6165 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6166 cs_db ? X86EMUL_MODE_PROT32 :
6167 X86EMUL_MODE_PROT16;
6168 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6169 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6170 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6172 init_decode_cache(ctxt);
6173 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6176 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6178 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6181 init_emulate_ctxt(vcpu);
6185 ctxt->_eip = ctxt->eip + inc_eip;
6186 ret = emulate_int_real(ctxt, irq);
6188 if (ret != X86EMUL_CONTINUE)
6189 return EMULATE_FAIL;
6191 ctxt->eip = ctxt->_eip;
6192 kvm_rip_write(vcpu, ctxt->eip);
6193 kvm_set_rflags(vcpu, ctxt->eflags);
6195 return EMULATE_DONE;
6197 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6199 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6201 int r = EMULATE_DONE;
6203 ++vcpu->stat.insn_emulation_fail;
6204 trace_kvm_emulate_insn_failed(vcpu);
6206 if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6207 return EMULATE_FAIL;
6209 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6210 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6211 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6212 vcpu->run->internal.ndata = 0;
6213 r = EMULATE_USER_EXIT;
6216 kvm_queue_exception(vcpu, UD_VECTOR);
6221 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6222 bool write_fault_to_shadow_pgtable,
6228 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6231 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6234 if (!vcpu->arch.mmu->direct_map) {
6236 * Write permission should be allowed since only
6237 * write access need to be emulated.
6239 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6242 * If the mapping is invalid in guest, let cpu retry
6243 * it to generate fault.
6245 if (gpa == UNMAPPED_GVA)
6250 * Do not retry the unhandleable instruction if it faults on the
6251 * readonly host memory, otherwise it will goto a infinite loop:
6252 * retry instruction -> write #PF -> emulation fail -> retry
6253 * instruction -> ...
6255 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6258 * If the instruction failed on the error pfn, it can not be fixed,
6259 * report the error to userspace.
6261 if (is_error_noslot_pfn(pfn))
6264 kvm_release_pfn_clean(pfn);
6266 /* The instructions are well-emulated on direct mmu. */
6267 if (vcpu->arch.mmu->direct_map) {
6268 unsigned int indirect_shadow_pages;
6270 spin_lock(&vcpu->kvm->mmu_lock);
6271 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6272 spin_unlock(&vcpu->kvm->mmu_lock);
6274 if (indirect_shadow_pages)
6275 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6281 * if emulation was due to access to shadowed page table
6282 * and it failed try to unshadow page and re-enter the
6283 * guest to let CPU execute the instruction.
6285 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6288 * If the access faults on its page table, it can not
6289 * be fixed by unprotecting shadow page and it should
6290 * be reported to userspace.
6292 return !write_fault_to_shadow_pgtable;
6295 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6296 unsigned long cr2, int emulation_type)
6298 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6299 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6301 last_retry_eip = vcpu->arch.last_retry_eip;
6302 last_retry_addr = vcpu->arch.last_retry_addr;
6305 * If the emulation is caused by #PF and it is non-page_table
6306 * writing instruction, it means the VM-EXIT is caused by shadow
6307 * page protected, we can zap the shadow page and retry this
6308 * instruction directly.
6310 * Note: if the guest uses a non-page-table modifying instruction
6311 * on the PDE that points to the instruction, then we will unmap
6312 * the instruction and go to an infinite loop. So, we cache the
6313 * last retried eip and the last fault address, if we meet the eip
6314 * and the address again, we can break out of the potential infinite
6317 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6319 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6322 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6325 if (x86_page_table_writing_insn(ctxt))
6328 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6331 vcpu->arch.last_retry_eip = ctxt->eip;
6332 vcpu->arch.last_retry_addr = cr2;
6334 if (!vcpu->arch.mmu->direct_map)
6335 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6337 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6342 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6343 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6345 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6347 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6348 /* This is a good place to trace that we are exiting SMM. */
6349 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6351 /* Process a latched INIT or SMI, if any. */
6352 kvm_make_request(KVM_REQ_EVENT, vcpu);
6355 kvm_mmu_reset_context(vcpu);
6358 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6367 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6368 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6373 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6375 struct kvm_run *kvm_run = vcpu->run;
6377 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6378 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6379 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6380 kvm_run->debug.arch.exception = DB_VECTOR;
6381 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6382 *r = EMULATE_USER_EXIT;
6384 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6388 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6390 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6391 int r = EMULATE_DONE;
6393 kvm_x86_ops->skip_emulated_instruction(vcpu);
6396 * rflags is the old, "raw" value of the flags. The new value has
6397 * not been saved yet.
6399 * This is correct even for TF set by the guest, because "the
6400 * processor will not generate this exception after the instruction
6401 * that sets the TF flag".
6403 if (unlikely(rflags & X86_EFLAGS_TF))
6404 kvm_vcpu_do_singlestep(vcpu, &r);
6405 return r == EMULATE_DONE;
6407 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6409 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6411 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6412 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6413 struct kvm_run *kvm_run = vcpu->run;
6414 unsigned long eip = kvm_get_linear_rip(vcpu);
6415 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6416 vcpu->arch.guest_debug_dr7,
6420 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6421 kvm_run->debug.arch.pc = eip;
6422 kvm_run->debug.arch.exception = DB_VECTOR;
6423 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6424 *r = EMULATE_USER_EXIT;
6429 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6430 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6431 unsigned long eip = kvm_get_linear_rip(vcpu);
6432 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6437 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6438 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6439 kvm_queue_exception(vcpu, DB_VECTOR);
6448 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6450 switch (ctxt->opcode_len) {
6457 case 0xe6: /* OUT */
6461 case 0x6c: /* INS */
6463 case 0x6e: /* OUTS */
6470 case 0x33: /* RDPMC */
6479 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6486 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6487 bool writeback = true;
6488 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6490 vcpu->arch.l1tf_flush_l1d = true;
6493 * Clear write_fault_to_shadow_pgtable here to ensure it is
6496 vcpu->arch.write_fault_to_shadow_pgtable = false;
6497 kvm_clear_exception_queue(vcpu);
6499 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6500 init_emulate_ctxt(vcpu);
6503 * We will reenter on the same instruction since
6504 * we do not set complete_userspace_io. This does not
6505 * handle watchpoints yet, those would be handled in
6508 if (!(emulation_type & EMULTYPE_SKIP) &&
6509 kvm_vcpu_check_breakpoint(vcpu, &r))
6512 ctxt->interruptibility = 0;
6513 ctxt->have_exception = false;
6514 ctxt->exception.vector = -1;
6515 ctxt->perm_ok = false;
6517 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6519 r = x86_decode_insn(ctxt, insn, insn_len);
6521 trace_kvm_emulate_insn_start(vcpu);
6522 ++vcpu->stat.insn_emulation;
6523 if (r != EMULATION_OK) {
6524 if (emulation_type & EMULTYPE_TRAP_UD)
6525 return EMULATE_FAIL;
6526 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6528 return EMULATE_DONE;
6529 if (ctxt->have_exception && inject_emulated_exception(vcpu))
6530 return EMULATE_DONE;
6531 if (emulation_type & EMULTYPE_SKIP)
6532 return EMULATE_FAIL;
6533 return handle_emulation_failure(vcpu, emulation_type);
6537 if ((emulation_type & EMULTYPE_VMWARE) &&
6538 !is_vmware_backdoor_opcode(ctxt))
6539 return EMULATE_FAIL;
6541 if (emulation_type & EMULTYPE_SKIP) {
6542 kvm_rip_write(vcpu, ctxt->_eip);
6543 if (ctxt->eflags & X86_EFLAGS_RF)
6544 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6545 return EMULATE_DONE;
6548 if (retry_instruction(ctxt, cr2, emulation_type))
6549 return EMULATE_DONE;
6551 /* this is needed for vmware backdoor interface to work since it
6552 changes registers values during IO operation */
6553 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6554 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6555 emulator_invalidate_register_cache(ctxt);
6559 /* Save the faulting GPA (cr2) in the address field */
6560 ctxt->exception.address = cr2;
6562 r = x86_emulate_insn(ctxt);
6564 if (r == EMULATION_INTERCEPTED)
6565 return EMULATE_DONE;
6567 if (r == EMULATION_FAILED) {
6568 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6570 return EMULATE_DONE;
6572 return handle_emulation_failure(vcpu, emulation_type);
6575 if (ctxt->have_exception) {
6577 if (inject_emulated_exception(vcpu))
6579 } else if (vcpu->arch.pio.count) {
6580 if (!vcpu->arch.pio.in) {
6581 /* FIXME: return into emulator if single-stepping. */
6582 vcpu->arch.pio.count = 0;
6585 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6587 r = EMULATE_USER_EXIT;
6588 } else if (vcpu->mmio_needed) {
6589 if (!vcpu->mmio_is_write)
6591 r = EMULATE_USER_EXIT;
6592 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6593 } else if (r == EMULATION_RESTART)
6599 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6600 toggle_interruptibility(vcpu, ctxt->interruptibility);
6601 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6602 kvm_rip_write(vcpu, ctxt->eip);
6603 if (r == EMULATE_DONE && ctxt->tf)
6604 kvm_vcpu_do_singlestep(vcpu, &r);
6605 if (!ctxt->have_exception ||
6606 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6607 __kvm_set_rflags(vcpu, ctxt->eflags);
6610 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6611 * do nothing, and it will be requested again as soon as
6612 * the shadow expires. But we still need to check here,
6613 * because POPF has no interrupt shadow.
6615 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6616 kvm_make_request(KVM_REQ_EVENT, vcpu);
6618 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6623 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6625 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6627 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6629 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6630 void *insn, int insn_len)
6632 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6634 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6636 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6638 vcpu->arch.pio.count = 0;
6642 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6644 vcpu->arch.pio.count = 0;
6646 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6649 return kvm_skip_emulated_instruction(vcpu);
6652 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6653 unsigned short port)
6655 unsigned long val = kvm_rax_read(vcpu);
6656 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6657 size, port, &val, 1);
6662 * Workaround userspace that relies on old KVM behavior of %rip being
6663 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6666 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6667 vcpu->arch.complete_userspace_io =
6668 complete_fast_pio_out_port_0x7e;
6669 kvm_skip_emulated_instruction(vcpu);
6671 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6672 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6677 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6681 /* We should only ever be called with arch.pio.count equal to 1 */
6682 BUG_ON(vcpu->arch.pio.count != 1);
6684 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6685 vcpu->arch.pio.count = 0;
6689 /* For size less than 4 we merge, else we zero extend */
6690 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6693 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6694 * the copy and tracing
6696 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6697 vcpu->arch.pio.port, &val, 1);
6698 kvm_rax_write(vcpu, val);
6700 return kvm_skip_emulated_instruction(vcpu);
6703 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6704 unsigned short port)
6709 /* For size less than 4 we merge, else we zero extend */
6710 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6712 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6715 kvm_rax_write(vcpu, val);
6719 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6720 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6725 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6730 ret = kvm_fast_pio_in(vcpu, size, port);
6732 ret = kvm_fast_pio_out(vcpu, size, port);
6733 return ret && kvm_skip_emulated_instruction(vcpu);
6735 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6737 static int kvmclock_cpu_down_prep(unsigned int cpu)
6739 __this_cpu_write(cpu_tsc_khz, 0);
6743 static void tsc_khz_changed(void *data)
6745 struct cpufreq_freqs *freq = data;
6746 unsigned long khz = 0;
6750 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6751 khz = cpufreq_quick_get(raw_smp_processor_id());
6754 __this_cpu_write(cpu_tsc_khz, khz);
6757 #ifdef CONFIG_X86_64
6758 static void kvm_hyperv_tsc_notifier(void)
6761 struct kvm_vcpu *vcpu;
6764 mutex_lock(&kvm_lock);
6765 list_for_each_entry(kvm, &vm_list, vm_list)
6766 kvm_make_mclock_inprogress_request(kvm);
6768 hyperv_stop_tsc_emulation();
6770 /* TSC frequency always matches when on Hyper-V */
6771 for_each_present_cpu(cpu)
6772 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6773 kvm_max_guest_tsc_khz = tsc_khz;
6775 list_for_each_entry(kvm, &vm_list, vm_list) {
6776 struct kvm_arch *ka = &kvm->arch;
6778 spin_lock(&ka->pvclock_gtod_sync_lock);
6780 pvclock_update_vm_gtod_copy(kvm);
6782 kvm_for_each_vcpu(cpu, vcpu, kvm)
6783 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6785 kvm_for_each_vcpu(cpu, vcpu, kvm)
6786 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6788 spin_unlock(&ka->pvclock_gtod_sync_lock);
6790 mutex_unlock(&kvm_lock);
6794 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6797 struct kvm_vcpu *vcpu;
6798 int i, send_ipi = 0;
6801 * We allow guests to temporarily run on slowing clocks,
6802 * provided we notify them after, or to run on accelerating
6803 * clocks, provided we notify them before. Thus time never
6806 * However, we have a problem. We can't atomically update
6807 * the frequency of a given CPU from this function; it is
6808 * merely a notifier, which can be called from any CPU.
6809 * Changing the TSC frequency at arbitrary points in time
6810 * requires a recomputation of local variables related to
6811 * the TSC for each VCPU. We must flag these local variables
6812 * to be updated and be sure the update takes place with the
6813 * new frequency before any guests proceed.
6815 * Unfortunately, the combination of hotplug CPU and frequency
6816 * change creates an intractable locking scenario; the order
6817 * of when these callouts happen is undefined with respect to
6818 * CPU hotplug, and they can race with each other. As such,
6819 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6820 * undefined; you can actually have a CPU frequency change take
6821 * place in between the computation of X and the setting of the
6822 * variable. To protect against this problem, all updates of
6823 * the per_cpu tsc_khz variable are done in an interrupt
6824 * protected IPI, and all callers wishing to update the value
6825 * must wait for a synchronous IPI to complete (which is trivial
6826 * if the caller is on the CPU already). This establishes the
6827 * necessary total order on variable updates.
6829 * Note that because a guest time update may take place
6830 * anytime after the setting of the VCPU's request bit, the
6831 * correct TSC value must be set before the request. However,
6832 * to ensure the update actually makes it to any guest which
6833 * starts running in hardware virtualization between the set
6834 * and the acquisition of the spinlock, we must also ping the
6835 * CPU after setting the request bit.
6839 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6841 mutex_lock(&kvm_lock);
6842 list_for_each_entry(kvm, &vm_list, vm_list) {
6843 kvm_for_each_vcpu(i, vcpu, kvm) {
6844 if (vcpu->cpu != cpu)
6846 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6847 if (vcpu->cpu != raw_smp_processor_id())
6851 mutex_unlock(&kvm_lock);
6853 if (freq->old < freq->new && send_ipi) {
6855 * We upscale the frequency. Must make the guest
6856 * doesn't see old kvmclock values while running with
6857 * the new frequency, otherwise we risk the guest sees
6858 * time go backwards.
6860 * In case we update the frequency for another cpu
6861 * (which might be in guest context) send an interrupt
6862 * to kick the cpu out of guest context. Next time
6863 * guest context is entered kvmclock will be updated,
6864 * so the guest will not see stale values.
6866 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6870 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6873 struct cpufreq_freqs *freq = data;
6876 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6878 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6881 for_each_cpu(cpu, freq->policy->cpus)
6882 __kvmclock_cpufreq_notifier(freq, cpu);
6887 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6888 .notifier_call = kvmclock_cpufreq_notifier
6891 static int kvmclock_cpu_online(unsigned int cpu)
6893 tsc_khz_changed(NULL);
6897 static void kvm_timer_init(void)
6899 max_tsc_khz = tsc_khz;
6901 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6902 #ifdef CONFIG_CPU_FREQ
6903 struct cpufreq_policy policy;
6906 memset(&policy, 0, sizeof(policy));
6908 cpufreq_get_policy(&policy, cpu);
6909 if (policy.cpuinfo.max_freq)
6910 max_tsc_khz = policy.cpuinfo.max_freq;
6913 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6914 CPUFREQ_TRANSITION_NOTIFIER);
6916 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6918 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6919 kvmclock_cpu_online, kvmclock_cpu_down_prep);
6922 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6923 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6925 int kvm_is_in_guest(void)
6927 return __this_cpu_read(current_vcpu) != NULL;
6930 static int kvm_is_user_mode(void)
6934 if (__this_cpu_read(current_vcpu))
6935 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6937 return user_mode != 0;
6940 static unsigned long kvm_get_guest_ip(void)
6942 unsigned long ip = 0;
6944 if (__this_cpu_read(current_vcpu))
6945 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6950 static void kvm_handle_intel_pt_intr(void)
6952 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
6954 kvm_make_request(KVM_REQ_PMI, vcpu);
6955 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
6956 (unsigned long *)&vcpu->arch.pmu.global_status);
6959 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6960 .is_in_guest = kvm_is_in_guest,
6961 .is_user_mode = kvm_is_user_mode,
6962 .get_guest_ip = kvm_get_guest_ip,
6963 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
6966 #ifdef CONFIG_X86_64
6967 static void pvclock_gtod_update_fn(struct work_struct *work)
6971 struct kvm_vcpu *vcpu;
6974 mutex_lock(&kvm_lock);
6975 list_for_each_entry(kvm, &vm_list, vm_list)
6976 kvm_for_each_vcpu(i, vcpu, kvm)
6977 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6978 atomic_set(&kvm_guest_has_master_clock, 0);
6979 mutex_unlock(&kvm_lock);
6982 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6985 * Notification about pvclock gtod data update.
6987 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6990 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6991 struct timekeeper *tk = priv;
6993 update_pvclock_gtod(tk);
6995 /* disable master clock if host does not trust, or does not
6996 * use, TSC based clocksource.
6998 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6999 atomic_read(&kvm_guest_has_master_clock) != 0)
7000 queue_work(system_long_wq, &pvclock_gtod_work);
7005 static struct notifier_block pvclock_gtod_notifier = {
7006 .notifier_call = pvclock_gtod_notify,
7010 int kvm_arch_init(void *opaque)
7013 struct kvm_x86_ops *ops = opaque;
7016 printk(KERN_ERR "kvm: already loaded the other module\n");
7021 if (!ops->cpu_has_kvm_support()) {
7022 printk(KERN_ERR "kvm: no hardware support\n");
7026 if (ops->disabled_by_bios()) {
7027 printk(KERN_ERR "kvm: disabled by bios\n");
7033 * KVM explicitly assumes that the guest has an FPU and
7034 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7035 * vCPU's FPU state as a fxregs_state struct.
7037 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7038 printk(KERN_ERR "kvm: inadequate fpu\n");
7044 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7045 __alignof__(struct fpu), SLAB_ACCOUNT,
7047 if (!x86_fpu_cache) {
7048 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7052 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7054 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7055 goto out_free_x86_fpu_cache;
7058 r = kvm_mmu_module_init();
7060 goto out_free_percpu;
7064 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7065 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7066 PT_PRESENT_MASK, 0, sme_me_mask);
7069 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7071 if (boot_cpu_has(X86_FEATURE_XSAVE))
7072 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7075 #ifdef CONFIG_X86_64
7076 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7078 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7079 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7085 free_percpu(shared_msrs);
7086 out_free_x86_fpu_cache:
7087 kmem_cache_destroy(x86_fpu_cache);
7092 void kvm_arch_exit(void)
7094 #ifdef CONFIG_X86_64
7095 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7096 clear_hv_tscchange_cb();
7099 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7101 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7102 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7103 CPUFREQ_TRANSITION_NOTIFIER);
7104 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7105 #ifdef CONFIG_X86_64
7106 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7109 kvm_mmu_module_exit();
7110 free_percpu(shared_msrs);
7111 kmem_cache_destroy(x86_fpu_cache);
7114 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7116 ++vcpu->stat.halt_exits;
7117 if (lapic_in_kernel(vcpu)) {
7118 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7121 vcpu->run->exit_reason = KVM_EXIT_HLT;
7125 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7127 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7129 int ret = kvm_skip_emulated_instruction(vcpu);
7131 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7132 * KVM_EXIT_DEBUG here.
7134 return kvm_vcpu_halt(vcpu) && ret;
7136 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7138 #ifdef CONFIG_X86_64
7139 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7140 unsigned long clock_type)
7142 struct kvm_clock_pairing clock_pairing;
7143 struct timespec64 ts;
7147 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7148 return -KVM_EOPNOTSUPP;
7150 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7151 return -KVM_EOPNOTSUPP;
7153 clock_pairing.sec = ts.tv_sec;
7154 clock_pairing.nsec = ts.tv_nsec;
7155 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7156 clock_pairing.flags = 0;
7157 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7160 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7161 sizeof(struct kvm_clock_pairing)))
7169 * kvm_pv_kick_cpu_op: Kick a vcpu.
7171 * @apicid - apicid of vcpu to be kicked.
7173 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7175 struct kvm_lapic_irq lapic_irq;
7177 lapic_irq.shorthand = 0;
7178 lapic_irq.dest_mode = 0;
7179 lapic_irq.level = 0;
7180 lapic_irq.dest_id = apicid;
7181 lapic_irq.msi_redir_hint = false;
7183 lapic_irq.delivery_mode = APIC_DM_REMRD;
7184 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7187 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7189 if (!lapic_in_kernel(vcpu)) {
7190 WARN_ON_ONCE(vcpu->arch.apicv_active);
7193 if (!vcpu->arch.apicv_active)
7196 vcpu->arch.apicv_active = false;
7197 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7200 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7202 struct kvm_vcpu *target = NULL;
7203 struct kvm_apic_map *map;
7206 map = rcu_dereference(kvm->arch.apic_map);
7208 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7209 target = map->phys_map[dest_id]->vcpu;
7214 kvm_vcpu_yield_to(target);
7217 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7219 unsigned long nr, a0, a1, a2, a3, ret;
7222 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7223 return kvm_hv_hypercall(vcpu);
7225 nr = kvm_rax_read(vcpu);
7226 a0 = kvm_rbx_read(vcpu);
7227 a1 = kvm_rcx_read(vcpu);
7228 a2 = kvm_rdx_read(vcpu);
7229 a3 = kvm_rsi_read(vcpu);
7231 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7233 op_64_bit = is_64_bit_mode(vcpu);
7242 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7248 case KVM_HC_VAPIC_POLL_IRQ:
7251 case KVM_HC_KICK_CPU:
7252 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7255 #ifdef CONFIG_X86_64
7256 case KVM_HC_CLOCK_PAIRING:
7257 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7260 case KVM_HC_SEND_IPI:
7261 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7263 case KVM_HC_SCHED_YIELD:
7264 kvm_sched_yield(vcpu->kvm, a0);
7274 kvm_rax_write(vcpu, ret);
7276 ++vcpu->stat.hypercalls;
7277 return kvm_skip_emulated_instruction(vcpu);
7279 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7281 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7283 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7284 char instruction[3];
7285 unsigned long rip = kvm_rip_read(vcpu);
7287 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7289 return emulator_write_emulated(ctxt, rip, instruction, 3,
7293 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7295 return vcpu->run->request_interrupt_window &&
7296 likely(!pic_in_kernel(vcpu->kvm));
7299 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7301 struct kvm_run *kvm_run = vcpu->run;
7303 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7304 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7305 kvm_run->cr8 = kvm_get_cr8(vcpu);
7306 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7307 kvm_run->ready_for_interrupt_injection =
7308 pic_in_kernel(vcpu->kvm) ||
7309 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7312 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7316 if (!kvm_x86_ops->update_cr8_intercept)
7319 if (!lapic_in_kernel(vcpu))
7322 if (vcpu->arch.apicv_active)
7325 if (!vcpu->arch.apic->vapic_addr)
7326 max_irr = kvm_lapic_find_highest_irr(vcpu);
7333 tpr = kvm_lapic_get_cr8(vcpu);
7335 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7338 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7342 /* try to reinject previous events if any */
7344 if (vcpu->arch.exception.injected)
7345 kvm_x86_ops->queue_exception(vcpu);
7347 * Do not inject an NMI or interrupt if there is a pending
7348 * exception. Exceptions and interrupts are recognized at
7349 * instruction boundaries, i.e. the start of an instruction.
7350 * Trap-like exceptions, e.g. #DB, have higher priority than
7351 * NMIs and interrupts, i.e. traps are recognized before an
7352 * NMI/interrupt that's pending on the same instruction.
7353 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7354 * priority, but are only generated (pended) during instruction
7355 * execution, i.e. a pending fault-like exception means the
7356 * fault occurred on the *previous* instruction and must be
7357 * serviced prior to recognizing any new events in order to
7358 * fully complete the previous instruction.
7360 else if (!vcpu->arch.exception.pending) {
7361 if (vcpu->arch.nmi_injected)
7362 kvm_x86_ops->set_nmi(vcpu);
7363 else if (vcpu->arch.interrupt.injected)
7364 kvm_x86_ops->set_irq(vcpu);
7368 * Call check_nested_events() even if we reinjected a previous event
7369 * in order for caller to determine if it should require immediate-exit
7370 * from L2 to L1 due to pending L1 events which require exit
7373 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7374 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7379 /* try to inject new event if pending */
7380 if (vcpu->arch.exception.pending) {
7381 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7382 vcpu->arch.exception.has_error_code,
7383 vcpu->arch.exception.error_code);
7385 WARN_ON_ONCE(vcpu->arch.exception.injected);
7386 vcpu->arch.exception.pending = false;
7387 vcpu->arch.exception.injected = true;
7389 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7390 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7393 if (vcpu->arch.exception.nr == DB_VECTOR) {
7395 * This code assumes that nSVM doesn't use
7396 * check_nested_events(). If it does, the
7397 * DR6/DR7 changes should happen before L1
7398 * gets a #VMEXIT for an intercepted #DB in
7399 * L2. (Under VMX, on the other hand, the
7400 * DR6/DR7 changes should not happen in the
7401 * event of a VM-exit to L1 for an intercepted
7404 kvm_deliver_exception_payload(vcpu);
7405 if (vcpu->arch.dr7 & DR7_GD) {
7406 vcpu->arch.dr7 &= ~DR7_GD;
7407 kvm_update_dr7(vcpu);
7411 kvm_x86_ops->queue_exception(vcpu);
7414 /* Don't consider new event if we re-injected an event */
7415 if (kvm_event_needs_reinjection(vcpu))
7418 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7419 kvm_x86_ops->smi_allowed(vcpu)) {
7420 vcpu->arch.smi_pending = false;
7421 ++vcpu->arch.smi_count;
7423 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7424 --vcpu->arch.nmi_pending;
7425 vcpu->arch.nmi_injected = true;
7426 kvm_x86_ops->set_nmi(vcpu);
7427 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7429 * Because interrupts can be injected asynchronously, we are
7430 * calling check_nested_events again here to avoid a race condition.
7431 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7432 * proposal and current concerns. Perhaps we should be setting
7433 * KVM_REQ_EVENT only on certain events and not unconditionally?
7435 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7436 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7440 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7441 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7443 kvm_x86_ops->set_irq(vcpu);
7450 static void process_nmi(struct kvm_vcpu *vcpu)
7455 * x86 is limited to one NMI running, and one NMI pending after it.
7456 * If an NMI is already in progress, limit further NMIs to just one.
7457 * Otherwise, allow two (and we'll inject the first one immediately).
7459 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7462 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7463 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7464 kvm_make_request(KVM_REQ_EVENT, vcpu);
7467 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7470 flags |= seg->g << 23;
7471 flags |= seg->db << 22;
7472 flags |= seg->l << 21;
7473 flags |= seg->avl << 20;
7474 flags |= seg->present << 15;
7475 flags |= seg->dpl << 13;
7476 flags |= seg->s << 12;
7477 flags |= seg->type << 8;
7481 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7483 struct kvm_segment seg;
7486 kvm_get_segment(vcpu, &seg, n);
7487 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7490 offset = 0x7f84 + n * 12;
7492 offset = 0x7f2c + (n - 3) * 12;
7494 put_smstate(u32, buf, offset + 8, seg.base);
7495 put_smstate(u32, buf, offset + 4, seg.limit);
7496 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7499 #ifdef CONFIG_X86_64
7500 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7502 struct kvm_segment seg;
7506 kvm_get_segment(vcpu, &seg, n);
7507 offset = 0x7e00 + n * 16;
7509 flags = enter_smm_get_segment_flags(&seg) >> 8;
7510 put_smstate(u16, buf, offset, seg.selector);
7511 put_smstate(u16, buf, offset + 2, flags);
7512 put_smstate(u32, buf, offset + 4, seg.limit);
7513 put_smstate(u64, buf, offset + 8, seg.base);
7517 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7520 struct kvm_segment seg;
7524 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7525 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7526 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7527 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7529 for (i = 0; i < 8; i++)
7530 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7532 kvm_get_dr(vcpu, 6, &val);
7533 put_smstate(u32, buf, 0x7fcc, (u32)val);
7534 kvm_get_dr(vcpu, 7, &val);
7535 put_smstate(u32, buf, 0x7fc8, (u32)val);
7537 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7538 put_smstate(u32, buf, 0x7fc4, seg.selector);
7539 put_smstate(u32, buf, 0x7f64, seg.base);
7540 put_smstate(u32, buf, 0x7f60, seg.limit);
7541 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7543 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7544 put_smstate(u32, buf, 0x7fc0, seg.selector);
7545 put_smstate(u32, buf, 0x7f80, seg.base);
7546 put_smstate(u32, buf, 0x7f7c, seg.limit);
7547 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7549 kvm_x86_ops->get_gdt(vcpu, &dt);
7550 put_smstate(u32, buf, 0x7f74, dt.address);
7551 put_smstate(u32, buf, 0x7f70, dt.size);
7553 kvm_x86_ops->get_idt(vcpu, &dt);
7554 put_smstate(u32, buf, 0x7f58, dt.address);
7555 put_smstate(u32, buf, 0x7f54, dt.size);
7557 for (i = 0; i < 6; i++)
7558 enter_smm_save_seg_32(vcpu, buf, i);
7560 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7563 put_smstate(u32, buf, 0x7efc, 0x00020000);
7564 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7567 #ifdef CONFIG_X86_64
7568 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7571 struct kvm_segment seg;
7575 for (i = 0; i < 16; i++)
7576 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7578 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7579 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7581 kvm_get_dr(vcpu, 6, &val);
7582 put_smstate(u64, buf, 0x7f68, val);
7583 kvm_get_dr(vcpu, 7, &val);
7584 put_smstate(u64, buf, 0x7f60, val);
7586 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7587 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7588 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7590 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7593 put_smstate(u32, buf, 0x7efc, 0x00020064);
7595 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7597 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7598 put_smstate(u16, buf, 0x7e90, seg.selector);
7599 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7600 put_smstate(u32, buf, 0x7e94, seg.limit);
7601 put_smstate(u64, buf, 0x7e98, seg.base);
7603 kvm_x86_ops->get_idt(vcpu, &dt);
7604 put_smstate(u32, buf, 0x7e84, dt.size);
7605 put_smstate(u64, buf, 0x7e88, dt.address);
7607 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7608 put_smstate(u16, buf, 0x7e70, seg.selector);
7609 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7610 put_smstate(u32, buf, 0x7e74, seg.limit);
7611 put_smstate(u64, buf, 0x7e78, seg.base);
7613 kvm_x86_ops->get_gdt(vcpu, &dt);
7614 put_smstate(u32, buf, 0x7e64, dt.size);
7615 put_smstate(u64, buf, 0x7e68, dt.address);
7617 for (i = 0; i < 6; i++)
7618 enter_smm_save_seg_64(vcpu, buf, i);
7622 static void enter_smm(struct kvm_vcpu *vcpu)
7624 struct kvm_segment cs, ds;
7629 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7630 memset(buf, 0, 512);
7631 #ifdef CONFIG_X86_64
7632 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7633 enter_smm_save_state_64(vcpu, buf);
7636 enter_smm_save_state_32(vcpu, buf);
7639 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7640 * vCPU state (e.g. leave guest mode) after we've saved the state into
7641 * the SMM state-save area.
7643 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7645 vcpu->arch.hflags |= HF_SMM_MASK;
7646 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7648 if (kvm_x86_ops->get_nmi_mask(vcpu))
7649 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7651 kvm_x86_ops->set_nmi_mask(vcpu, true);
7653 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7654 kvm_rip_write(vcpu, 0x8000);
7656 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7657 kvm_x86_ops->set_cr0(vcpu, cr0);
7658 vcpu->arch.cr0 = cr0;
7660 kvm_x86_ops->set_cr4(vcpu, 0);
7662 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7663 dt.address = dt.size = 0;
7664 kvm_x86_ops->set_idt(vcpu, &dt);
7666 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7668 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7669 cs.base = vcpu->arch.smbase;
7674 cs.limit = ds.limit = 0xffffffff;
7675 cs.type = ds.type = 0x3;
7676 cs.dpl = ds.dpl = 0;
7681 cs.avl = ds.avl = 0;
7682 cs.present = ds.present = 1;
7683 cs.unusable = ds.unusable = 0;
7684 cs.padding = ds.padding = 0;
7686 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7687 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7688 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7689 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7690 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7691 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7693 #ifdef CONFIG_X86_64
7694 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7695 kvm_x86_ops->set_efer(vcpu, 0);
7698 kvm_update_cpuid(vcpu);
7699 kvm_mmu_reset_context(vcpu);
7702 static void process_smi(struct kvm_vcpu *vcpu)
7704 vcpu->arch.smi_pending = true;
7705 kvm_make_request(KVM_REQ_EVENT, vcpu);
7708 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7710 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7713 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7715 if (!kvm_apic_present(vcpu))
7718 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7720 if (irqchip_split(vcpu->kvm))
7721 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7723 if (vcpu->arch.apicv_active)
7724 kvm_x86_ops->sync_pir_to_irr(vcpu);
7725 if (ioapic_in_kernel(vcpu->kvm))
7726 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7729 if (is_guest_mode(vcpu))
7730 vcpu->arch.load_eoi_exitmap_pending = true;
7732 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7735 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7737 u64 eoi_exit_bitmap[4];
7739 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7742 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7743 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7744 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7747 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7748 unsigned long start, unsigned long end,
7751 unsigned long apic_address;
7754 * The physical address of apic access page is stored in the VMCS.
7755 * Update it when it becomes invalid.
7757 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7758 if (start <= apic_address && apic_address < end)
7759 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7764 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7766 struct page *page = NULL;
7768 if (!lapic_in_kernel(vcpu))
7771 if (!kvm_x86_ops->set_apic_access_page_addr)
7774 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7775 if (is_error_page(page))
7777 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7780 * Do not pin apic access page in memory, the MMU notifier
7781 * will call us again if it is migrated or swapped out.
7785 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7787 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7789 smp_send_reschedule(vcpu->cpu);
7791 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7794 * Returns 1 to let vcpu_run() continue the guest execution loop without
7795 * exiting to the userspace. Otherwise, the value will be returned to the
7798 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7802 dm_request_for_irq_injection(vcpu) &&
7803 kvm_cpu_accept_dm_intr(vcpu);
7805 bool req_immediate_exit = false;
7807 if (kvm_request_pending(vcpu)) {
7808 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7809 kvm_x86_ops->get_vmcs12_pages(vcpu);
7810 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7811 kvm_mmu_unload(vcpu);
7812 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7813 __kvm_migrate_timers(vcpu);
7814 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7815 kvm_gen_update_masterclock(vcpu->kvm);
7816 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7817 kvm_gen_kvmclock_update(vcpu);
7818 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7819 r = kvm_guest_time_update(vcpu);
7823 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7824 kvm_mmu_sync_roots(vcpu);
7825 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7826 kvm_mmu_load_cr3(vcpu);
7827 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7828 kvm_vcpu_flush_tlb(vcpu, true);
7829 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7830 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7834 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7835 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7836 vcpu->mmio_needed = 0;
7840 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7841 /* Page is swapped out. Do synthetic halt */
7842 vcpu->arch.apf.halted = true;
7846 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7847 record_steal_time(vcpu);
7848 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7850 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7852 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7853 kvm_pmu_handle_event(vcpu);
7854 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7855 kvm_pmu_deliver_pmi(vcpu);
7856 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7857 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7858 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7859 vcpu->arch.ioapic_handled_vectors)) {
7860 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7861 vcpu->run->eoi.vector =
7862 vcpu->arch.pending_ioapic_eoi;
7867 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7868 vcpu_scan_ioapic(vcpu);
7869 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7870 vcpu_load_eoi_exitmap(vcpu);
7871 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7872 kvm_vcpu_reload_apic_access_page(vcpu);
7873 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7874 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7875 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7879 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7880 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7881 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7885 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7886 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7887 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7893 * KVM_REQ_HV_STIMER has to be processed after
7894 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7895 * depend on the guest clock being up-to-date
7897 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7898 kvm_hv_process_stimers(vcpu);
7901 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7902 ++vcpu->stat.req_event;
7903 kvm_apic_accept_events(vcpu);
7904 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7909 if (inject_pending_event(vcpu, req_int_win) != 0)
7910 req_immediate_exit = true;
7912 /* Enable SMI/NMI/IRQ window open exits if needed.
7914 * SMIs have three cases:
7915 * 1) They can be nested, and then there is nothing to
7916 * do here because RSM will cause a vmexit anyway.
7917 * 2) There is an ISA-specific reason why SMI cannot be
7918 * injected, and the moment when this changes can be
7920 * 3) Or the SMI can be pending because
7921 * inject_pending_event has completed the injection
7922 * of an IRQ or NMI from the previous vmexit, and
7923 * then we request an immediate exit to inject the
7926 if (vcpu->arch.smi_pending && !is_smm(vcpu))
7927 if (!kvm_x86_ops->enable_smi_window(vcpu))
7928 req_immediate_exit = true;
7929 if (vcpu->arch.nmi_pending)
7930 kvm_x86_ops->enable_nmi_window(vcpu);
7931 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7932 kvm_x86_ops->enable_irq_window(vcpu);
7933 WARN_ON(vcpu->arch.exception.pending);
7936 if (kvm_lapic_enabled(vcpu)) {
7937 update_cr8_intercept(vcpu);
7938 kvm_lapic_sync_to_vapic(vcpu);
7942 r = kvm_mmu_reload(vcpu);
7944 goto cancel_injection;
7949 kvm_x86_ops->prepare_guest_switch(vcpu);
7952 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7953 * IPI are then delayed after guest entry, which ensures that they
7954 * result in virtual interrupt delivery.
7956 local_irq_disable();
7957 vcpu->mode = IN_GUEST_MODE;
7959 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7962 * 1) We should set ->mode before checking ->requests. Please see
7963 * the comment in kvm_vcpu_exiting_guest_mode().
7965 * 2) For APICv, we should set ->mode before checking PID.ON. This
7966 * pairs with the memory barrier implicit in pi_test_and_set_on
7967 * (see vmx_deliver_posted_interrupt).
7969 * 3) This also orders the write to mode from any reads to the page
7970 * tables done while the VCPU is running. Please see the comment
7971 * in kvm_flush_remote_tlbs.
7973 smp_mb__after_srcu_read_unlock();
7976 * This handles the case where a posted interrupt was
7977 * notified with kvm_vcpu_kick.
7979 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7980 kvm_x86_ops->sync_pir_to_irr(vcpu);
7982 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7983 || need_resched() || signal_pending(current)) {
7984 vcpu->mode = OUTSIDE_GUEST_MODE;
7988 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7990 goto cancel_injection;
7993 if (req_immediate_exit) {
7994 kvm_make_request(KVM_REQ_EVENT, vcpu);
7995 kvm_x86_ops->request_immediate_exit(vcpu);
7998 trace_kvm_entry(vcpu->vcpu_id);
7999 guest_enter_irqoff();
8001 fpregs_assert_state_consistent();
8002 if (test_thread_flag(TIF_NEED_FPU_LOAD))
8003 switch_fpu_return();
8005 if (unlikely(vcpu->arch.switch_db_regs)) {
8007 set_debugreg(vcpu->arch.eff_db[0], 0);
8008 set_debugreg(vcpu->arch.eff_db[1], 1);
8009 set_debugreg(vcpu->arch.eff_db[2], 2);
8010 set_debugreg(vcpu->arch.eff_db[3], 3);
8011 set_debugreg(vcpu->arch.dr6, 6);
8012 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8015 kvm_x86_ops->run(vcpu);
8018 * Do this here before restoring debug registers on the host. And
8019 * since we do this before handling the vmexit, a DR access vmexit
8020 * can (a) read the correct value of the debug registers, (b) set
8021 * KVM_DEBUGREG_WONT_EXIT again.
8023 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8024 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8025 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8026 kvm_update_dr0123(vcpu);
8027 kvm_update_dr6(vcpu);
8028 kvm_update_dr7(vcpu);
8029 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8033 * If the guest has used debug registers, at least dr7
8034 * will be disabled while returning to the host.
8035 * If we don't have active breakpoints in the host, we don't
8036 * care about the messed up debug address registers. But if
8037 * we have some of them active, restore the old state.
8039 if (hw_breakpoint_active())
8040 hw_breakpoint_restore();
8042 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8044 vcpu->mode = OUTSIDE_GUEST_MODE;
8047 kvm_x86_ops->handle_exit_irqoff(vcpu);
8051 guest_exit_irqoff();
8052 if (lapic_in_kernel(vcpu)) {
8053 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8054 if (delta != S64_MIN) {
8055 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8056 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8063 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8066 * Profile KVM exit RIPs:
8068 if (unlikely(prof_on == KVM_PROFILING)) {
8069 unsigned long rip = kvm_rip_read(vcpu);
8070 profile_hit(KVM_PROFILING, (void *)rip);
8073 if (unlikely(vcpu->arch.tsc_always_catchup))
8074 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8076 if (vcpu->arch.apic_attention)
8077 kvm_lapic_sync_from_vapic(vcpu);
8079 vcpu->arch.gpa_available = false;
8080 r = kvm_x86_ops->handle_exit(vcpu);
8084 kvm_x86_ops->cancel_injection(vcpu);
8085 if (unlikely(vcpu->arch.apic_attention))
8086 kvm_lapic_sync_from_vapic(vcpu);
8091 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8093 if (!kvm_arch_vcpu_runnable(vcpu) &&
8094 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8095 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8096 kvm_vcpu_block(vcpu);
8097 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8099 if (kvm_x86_ops->post_block)
8100 kvm_x86_ops->post_block(vcpu);
8102 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8106 kvm_apic_accept_events(vcpu);
8107 switch(vcpu->arch.mp_state) {
8108 case KVM_MP_STATE_HALTED:
8109 vcpu->arch.pv.pv_unhalted = false;
8110 vcpu->arch.mp_state =
8111 KVM_MP_STATE_RUNNABLE;
8113 case KVM_MP_STATE_RUNNABLE:
8114 vcpu->arch.apf.halted = false;
8116 case KVM_MP_STATE_INIT_RECEIVED:
8125 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8127 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8128 kvm_x86_ops->check_nested_events(vcpu, false);
8130 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8131 !vcpu->arch.apf.halted);
8134 static int vcpu_run(struct kvm_vcpu *vcpu)
8137 struct kvm *kvm = vcpu->kvm;
8139 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8140 vcpu->arch.l1tf_flush_l1d = true;
8143 if (kvm_vcpu_running(vcpu)) {
8144 r = vcpu_enter_guest(vcpu);
8146 r = vcpu_block(kvm, vcpu);
8152 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8153 if (kvm_cpu_has_pending_timer(vcpu))
8154 kvm_inject_pending_timer_irqs(vcpu);
8156 if (dm_request_for_irq_injection(vcpu) &&
8157 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8159 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8160 ++vcpu->stat.request_irq_exits;
8164 kvm_check_async_pf_completion(vcpu);
8166 if (signal_pending(current)) {
8168 vcpu->run->exit_reason = KVM_EXIT_INTR;
8169 ++vcpu->stat.signal_exits;
8172 if (need_resched()) {
8173 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8175 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8179 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8184 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8187 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8188 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8189 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8190 if (r != EMULATE_DONE)
8195 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8197 BUG_ON(!vcpu->arch.pio.count);
8199 return complete_emulated_io(vcpu);
8203 * Implements the following, as a state machine:
8207 * for each mmio piece in the fragment
8215 * for each mmio piece in the fragment
8220 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8222 struct kvm_run *run = vcpu->run;
8223 struct kvm_mmio_fragment *frag;
8226 BUG_ON(!vcpu->mmio_needed);
8228 /* Complete previous fragment */
8229 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8230 len = min(8u, frag->len);
8231 if (!vcpu->mmio_is_write)
8232 memcpy(frag->data, run->mmio.data, len);
8234 if (frag->len <= 8) {
8235 /* Switch to the next fragment. */
8237 vcpu->mmio_cur_fragment++;
8239 /* Go forward to the next mmio piece. */
8245 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8246 vcpu->mmio_needed = 0;
8248 /* FIXME: return into emulator if single-stepping. */
8249 if (vcpu->mmio_is_write)
8251 vcpu->mmio_read_completed = 1;
8252 return complete_emulated_io(vcpu);
8255 run->exit_reason = KVM_EXIT_MMIO;
8256 run->mmio.phys_addr = frag->gpa;
8257 if (vcpu->mmio_is_write)
8258 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8259 run->mmio.len = min(8u, frag->len);
8260 run->mmio.is_write = vcpu->mmio_is_write;
8261 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8265 /* Swap (qemu) user FPU context for the guest FPU context. */
8266 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8270 copy_fpregs_to_fpstate(¤t->thread.fpu);
8271 /* PKRU is separately restored in kvm_x86_ops->run. */
8272 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8273 ~XFEATURE_MASK_PKRU);
8275 fpregs_mark_activate();
8281 /* When vcpu_run ends, restore user space FPU context. */
8282 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8286 copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8287 copy_kernel_to_fpregs(¤t->thread.fpu.state);
8289 fpregs_mark_activate();
8292 ++vcpu->stat.fpu_reload;
8296 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8301 kvm_sigset_activate(vcpu);
8302 kvm_load_guest_fpu(vcpu);
8304 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8305 if (kvm_run->immediate_exit) {
8309 kvm_vcpu_block(vcpu);
8310 kvm_apic_accept_events(vcpu);
8311 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8313 if (signal_pending(current)) {
8315 vcpu->run->exit_reason = KVM_EXIT_INTR;
8316 ++vcpu->stat.signal_exits;
8321 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8326 if (vcpu->run->kvm_dirty_regs) {
8327 r = sync_regs(vcpu);
8332 /* re-sync apic's tpr */
8333 if (!lapic_in_kernel(vcpu)) {
8334 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8340 if (unlikely(vcpu->arch.complete_userspace_io)) {
8341 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8342 vcpu->arch.complete_userspace_io = NULL;
8347 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8349 if (kvm_run->immediate_exit)
8355 kvm_put_guest_fpu(vcpu);
8356 if (vcpu->run->kvm_valid_regs)
8358 post_kvm_run_save(vcpu);
8359 kvm_sigset_deactivate(vcpu);
8365 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8367 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8369 * We are here if userspace calls get_regs() in the middle of
8370 * instruction emulation. Registers state needs to be copied
8371 * back from emulation context to vcpu. Userspace shouldn't do
8372 * that usually, but some bad designed PV devices (vmware
8373 * backdoor interface) need this to work
8375 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8376 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8378 regs->rax = kvm_rax_read(vcpu);
8379 regs->rbx = kvm_rbx_read(vcpu);
8380 regs->rcx = kvm_rcx_read(vcpu);
8381 regs->rdx = kvm_rdx_read(vcpu);
8382 regs->rsi = kvm_rsi_read(vcpu);
8383 regs->rdi = kvm_rdi_read(vcpu);
8384 regs->rsp = kvm_rsp_read(vcpu);
8385 regs->rbp = kvm_rbp_read(vcpu);
8386 #ifdef CONFIG_X86_64
8387 regs->r8 = kvm_r8_read(vcpu);
8388 regs->r9 = kvm_r9_read(vcpu);
8389 regs->r10 = kvm_r10_read(vcpu);
8390 regs->r11 = kvm_r11_read(vcpu);
8391 regs->r12 = kvm_r12_read(vcpu);
8392 regs->r13 = kvm_r13_read(vcpu);
8393 regs->r14 = kvm_r14_read(vcpu);
8394 regs->r15 = kvm_r15_read(vcpu);
8397 regs->rip = kvm_rip_read(vcpu);
8398 regs->rflags = kvm_get_rflags(vcpu);
8401 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8404 __get_regs(vcpu, regs);
8409 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8411 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8412 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8414 kvm_rax_write(vcpu, regs->rax);
8415 kvm_rbx_write(vcpu, regs->rbx);
8416 kvm_rcx_write(vcpu, regs->rcx);
8417 kvm_rdx_write(vcpu, regs->rdx);
8418 kvm_rsi_write(vcpu, regs->rsi);
8419 kvm_rdi_write(vcpu, regs->rdi);
8420 kvm_rsp_write(vcpu, regs->rsp);
8421 kvm_rbp_write(vcpu, regs->rbp);
8422 #ifdef CONFIG_X86_64
8423 kvm_r8_write(vcpu, regs->r8);
8424 kvm_r9_write(vcpu, regs->r9);
8425 kvm_r10_write(vcpu, regs->r10);
8426 kvm_r11_write(vcpu, regs->r11);
8427 kvm_r12_write(vcpu, regs->r12);
8428 kvm_r13_write(vcpu, regs->r13);
8429 kvm_r14_write(vcpu, regs->r14);
8430 kvm_r15_write(vcpu, regs->r15);
8433 kvm_rip_write(vcpu, regs->rip);
8434 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8436 vcpu->arch.exception.pending = false;
8438 kvm_make_request(KVM_REQ_EVENT, vcpu);
8441 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8444 __set_regs(vcpu, regs);
8449 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8451 struct kvm_segment cs;
8453 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8457 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8459 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8463 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8464 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8465 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8466 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8467 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8468 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8470 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8471 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8473 kvm_x86_ops->get_idt(vcpu, &dt);
8474 sregs->idt.limit = dt.size;
8475 sregs->idt.base = dt.address;
8476 kvm_x86_ops->get_gdt(vcpu, &dt);
8477 sregs->gdt.limit = dt.size;
8478 sregs->gdt.base = dt.address;
8480 sregs->cr0 = kvm_read_cr0(vcpu);
8481 sregs->cr2 = vcpu->arch.cr2;
8482 sregs->cr3 = kvm_read_cr3(vcpu);
8483 sregs->cr4 = kvm_read_cr4(vcpu);
8484 sregs->cr8 = kvm_get_cr8(vcpu);
8485 sregs->efer = vcpu->arch.efer;
8486 sregs->apic_base = kvm_get_apic_base(vcpu);
8488 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8490 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8491 set_bit(vcpu->arch.interrupt.nr,
8492 (unsigned long *)sregs->interrupt_bitmap);
8495 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8496 struct kvm_sregs *sregs)
8499 __get_sregs(vcpu, sregs);
8504 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8505 struct kvm_mp_state *mp_state)
8509 kvm_apic_accept_events(vcpu);
8510 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8511 vcpu->arch.pv.pv_unhalted)
8512 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8514 mp_state->mp_state = vcpu->arch.mp_state;
8520 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8521 struct kvm_mp_state *mp_state)
8527 if (!lapic_in_kernel(vcpu) &&
8528 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8531 /* INITs are latched while in SMM */
8532 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8533 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8534 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8537 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8538 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8539 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8541 vcpu->arch.mp_state = mp_state->mp_state;
8542 kvm_make_request(KVM_REQ_EVENT, vcpu);
8550 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8551 int reason, bool has_error_code, u32 error_code)
8553 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8556 init_emulate_ctxt(vcpu);
8558 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8559 has_error_code, error_code);
8562 return EMULATE_FAIL;
8564 kvm_rip_write(vcpu, ctxt->eip);
8565 kvm_set_rflags(vcpu, ctxt->eflags);
8566 kvm_make_request(KVM_REQ_EVENT, vcpu);
8567 return EMULATE_DONE;
8569 EXPORT_SYMBOL_GPL(kvm_task_switch);
8571 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8573 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8574 (sregs->cr4 & X86_CR4_OSXSAVE))
8577 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8579 * When EFER.LME and CR0.PG are set, the processor is in
8580 * 64-bit mode (though maybe in a 32-bit code segment).
8581 * CR4.PAE and EFER.LMA must be set.
8583 if (!(sregs->cr4 & X86_CR4_PAE)
8584 || !(sregs->efer & EFER_LMA))
8588 * Not in 64-bit mode: EFER.LMA is clear and the code
8589 * segment cannot be 64-bit.
8591 if (sregs->efer & EFER_LMA || sregs->cs.l)
8598 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8600 struct msr_data apic_base_msr;
8601 int mmu_reset_needed = 0;
8602 int cpuid_update_needed = 0;
8603 int pending_vec, max_bits, idx;
8607 if (kvm_valid_sregs(vcpu, sregs))
8610 apic_base_msr.data = sregs->apic_base;
8611 apic_base_msr.host_initiated = true;
8612 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8615 dt.size = sregs->idt.limit;
8616 dt.address = sregs->idt.base;
8617 kvm_x86_ops->set_idt(vcpu, &dt);
8618 dt.size = sregs->gdt.limit;
8619 dt.address = sregs->gdt.base;
8620 kvm_x86_ops->set_gdt(vcpu, &dt);
8622 vcpu->arch.cr2 = sregs->cr2;
8623 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8624 vcpu->arch.cr3 = sregs->cr3;
8625 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8627 kvm_set_cr8(vcpu, sregs->cr8);
8629 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8630 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8632 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8633 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8634 vcpu->arch.cr0 = sregs->cr0;
8636 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8637 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8638 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8639 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8640 if (cpuid_update_needed)
8641 kvm_update_cpuid(vcpu);
8643 idx = srcu_read_lock(&vcpu->kvm->srcu);
8644 if (is_pae_paging(vcpu)) {
8645 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8646 mmu_reset_needed = 1;
8648 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8650 if (mmu_reset_needed)
8651 kvm_mmu_reset_context(vcpu);
8653 max_bits = KVM_NR_INTERRUPTS;
8654 pending_vec = find_first_bit(
8655 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8656 if (pending_vec < max_bits) {
8657 kvm_queue_interrupt(vcpu, pending_vec, false);
8658 pr_debug("Set back pending irq %d\n", pending_vec);
8661 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8662 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8663 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8664 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8665 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8666 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8668 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8669 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8671 update_cr8_intercept(vcpu);
8673 /* Older userspace won't unhalt the vcpu on reset. */
8674 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8675 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8677 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8679 kvm_make_request(KVM_REQ_EVENT, vcpu);
8686 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8687 struct kvm_sregs *sregs)
8692 ret = __set_sregs(vcpu, sregs);
8697 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8698 struct kvm_guest_debug *dbg)
8700 unsigned long rflags;
8705 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8707 if (vcpu->arch.exception.pending)
8709 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8710 kvm_queue_exception(vcpu, DB_VECTOR);
8712 kvm_queue_exception(vcpu, BP_VECTOR);
8716 * Read rflags as long as potentially injected trace flags are still
8719 rflags = kvm_get_rflags(vcpu);
8721 vcpu->guest_debug = dbg->control;
8722 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8723 vcpu->guest_debug = 0;
8725 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8726 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8727 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8728 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8730 for (i = 0; i < KVM_NR_DB_REGS; i++)
8731 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8733 kvm_update_dr7(vcpu);
8735 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8736 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8737 get_segment_base(vcpu, VCPU_SREG_CS);
8740 * Trigger an rflags update that will inject or remove the trace
8743 kvm_set_rflags(vcpu, rflags);
8745 kvm_x86_ops->update_bp_intercept(vcpu);
8755 * Translate a guest virtual address to a guest physical address.
8757 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8758 struct kvm_translation *tr)
8760 unsigned long vaddr = tr->linear_address;
8766 idx = srcu_read_lock(&vcpu->kvm->srcu);
8767 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8768 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8769 tr->physical_address = gpa;
8770 tr->valid = gpa != UNMAPPED_GVA;
8778 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8780 struct fxregs_state *fxsave;
8784 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8785 memcpy(fpu->fpr, fxsave->st_space, 128);
8786 fpu->fcw = fxsave->cwd;
8787 fpu->fsw = fxsave->swd;
8788 fpu->ftwx = fxsave->twd;
8789 fpu->last_opcode = fxsave->fop;
8790 fpu->last_ip = fxsave->rip;
8791 fpu->last_dp = fxsave->rdp;
8792 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8798 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8800 struct fxregs_state *fxsave;
8804 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8806 memcpy(fxsave->st_space, fpu->fpr, 128);
8807 fxsave->cwd = fpu->fcw;
8808 fxsave->swd = fpu->fsw;
8809 fxsave->twd = fpu->ftwx;
8810 fxsave->fop = fpu->last_opcode;
8811 fxsave->rip = fpu->last_ip;
8812 fxsave->rdp = fpu->last_dp;
8813 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8819 static void store_regs(struct kvm_vcpu *vcpu)
8821 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8823 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8824 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8826 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8827 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8829 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8830 kvm_vcpu_ioctl_x86_get_vcpu_events(
8831 vcpu, &vcpu->run->s.regs.events);
8834 static int sync_regs(struct kvm_vcpu *vcpu)
8836 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8839 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8840 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8841 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8843 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8844 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8846 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8848 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8849 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8850 vcpu, &vcpu->run->s.regs.events))
8852 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8858 static void fx_init(struct kvm_vcpu *vcpu)
8860 fpstate_init(&vcpu->arch.guest_fpu->state);
8861 if (boot_cpu_has(X86_FEATURE_XSAVES))
8862 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
8863 host_xcr0 | XSTATE_COMPACTION_ENABLED;
8866 * Ensure guest xcr0 is valid for loading
8868 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8870 vcpu->arch.cr0 |= X86_CR0_ET;
8873 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8875 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8877 kvmclock_reset(vcpu);
8879 kvm_x86_ops->vcpu_free(vcpu);
8880 free_cpumask_var(wbinvd_dirty_mask);
8883 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8886 struct kvm_vcpu *vcpu;
8888 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8889 printk_once(KERN_WARNING
8890 "kvm: SMP vm created on host with unstable TSC; "
8891 "guest TSC will not be reliable\n");
8893 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8898 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8900 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
8901 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8902 kvm_vcpu_mtrr_init(vcpu);
8904 kvm_vcpu_reset(vcpu, false);
8905 kvm_init_mmu(vcpu, false);
8910 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8912 struct msr_data msr;
8913 struct kvm *kvm = vcpu->kvm;
8915 kvm_hv_vcpu_postcreate(vcpu);
8917 if (mutex_lock_killable(&vcpu->mutex))
8921 msr.index = MSR_IA32_TSC;
8922 msr.host_initiated = true;
8923 kvm_write_tsc(vcpu, &msr);
8926 /* poll control enabled by default */
8927 vcpu->arch.msr_kvm_poll_control = 1;
8929 mutex_unlock(&vcpu->mutex);
8931 if (!kvmclock_periodic_sync)
8934 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8935 KVMCLOCK_SYNC_PERIOD);
8938 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8940 vcpu->arch.apf.msr_val = 0;
8943 kvm_mmu_unload(vcpu);
8946 kvm_x86_ops->vcpu_free(vcpu);
8949 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8951 kvm_lapic_reset(vcpu, init_event);
8953 vcpu->arch.hflags = 0;
8955 vcpu->arch.smi_pending = 0;
8956 vcpu->arch.smi_count = 0;
8957 atomic_set(&vcpu->arch.nmi_queued, 0);
8958 vcpu->arch.nmi_pending = 0;
8959 vcpu->arch.nmi_injected = false;
8960 kvm_clear_interrupt_queue(vcpu);
8961 kvm_clear_exception_queue(vcpu);
8962 vcpu->arch.exception.pending = false;
8964 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8965 kvm_update_dr0123(vcpu);
8966 vcpu->arch.dr6 = DR6_INIT;
8967 kvm_update_dr6(vcpu);
8968 vcpu->arch.dr7 = DR7_FIXED_1;
8969 kvm_update_dr7(vcpu);
8973 kvm_make_request(KVM_REQ_EVENT, vcpu);
8974 vcpu->arch.apf.msr_val = 0;
8975 vcpu->arch.st.msr_val = 0;
8977 kvmclock_reset(vcpu);
8979 kvm_clear_async_pf_completion_queue(vcpu);
8980 kvm_async_pf_hash_reset(vcpu);
8981 vcpu->arch.apf.halted = false;
8983 if (kvm_mpx_supported()) {
8984 void *mpx_state_buffer;
8987 * To avoid have the INIT path from kvm_apic_has_events() that be
8988 * called with loaded FPU and does not let userspace fix the state.
8991 kvm_put_guest_fpu(vcpu);
8992 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8994 if (mpx_state_buffer)
8995 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8996 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8998 if (mpx_state_buffer)
8999 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9001 kvm_load_guest_fpu(vcpu);
9005 kvm_pmu_reset(vcpu);
9006 vcpu->arch.smbase = 0x30000;
9008 vcpu->arch.msr_misc_features_enables = 0;
9010 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9013 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9014 vcpu->arch.regs_avail = ~0;
9015 vcpu->arch.regs_dirty = ~0;
9017 vcpu->arch.ia32_xss = 0;
9019 kvm_x86_ops->vcpu_reset(vcpu, init_event);
9022 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9024 struct kvm_segment cs;
9026 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9027 cs.selector = vector << 8;
9028 cs.base = vector << 12;
9029 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9030 kvm_rip_write(vcpu, 0);
9033 int kvm_arch_hardware_enable(void)
9036 struct kvm_vcpu *vcpu;
9041 bool stable, backwards_tsc = false;
9043 kvm_shared_msr_cpu_online();
9044 ret = kvm_x86_ops->hardware_enable();
9048 local_tsc = rdtsc();
9049 stable = !kvm_check_tsc_unstable();
9050 list_for_each_entry(kvm, &vm_list, vm_list) {
9051 kvm_for_each_vcpu(i, vcpu, kvm) {
9052 if (!stable && vcpu->cpu == smp_processor_id())
9053 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9054 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9055 backwards_tsc = true;
9056 if (vcpu->arch.last_host_tsc > max_tsc)
9057 max_tsc = vcpu->arch.last_host_tsc;
9063 * Sometimes, even reliable TSCs go backwards. This happens on
9064 * platforms that reset TSC during suspend or hibernate actions, but
9065 * maintain synchronization. We must compensate. Fortunately, we can
9066 * detect that condition here, which happens early in CPU bringup,
9067 * before any KVM threads can be running. Unfortunately, we can't
9068 * bring the TSCs fully up to date with real time, as we aren't yet far
9069 * enough into CPU bringup that we know how much real time has actually
9070 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
9071 * variables that haven't been updated yet.
9073 * So we simply find the maximum observed TSC above, then record the
9074 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9075 * the adjustment will be applied. Note that we accumulate
9076 * adjustments, in case multiple suspend cycles happen before some VCPU
9077 * gets a chance to run again. In the event that no KVM threads get a
9078 * chance to run, we will miss the entire elapsed period, as we'll have
9079 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9080 * loose cycle time. This isn't too big a deal, since the loss will be
9081 * uniform across all VCPUs (not to mention the scenario is extremely
9082 * unlikely). It is possible that a second hibernate recovery happens
9083 * much faster than a first, causing the observed TSC here to be
9084 * smaller; this would require additional padding adjustment, which is
9085 * why we set last_host_tsc to the local tsc observed here.
9087 * N.B. - this code below runs only on platforms with reliable TSC,
9088 * as that is the only way backwards_tsc is set above. Also note
9089 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9090 * have the same delta_cyc adjustment applied if backwards_tsc
9091 * is detected. Note further, this adjustment is only done once,
9092 * as we reset last_host_tsc on all VCPUs to stop this from being
9093 * called multiple times (one for each physical CPU bringup).
9095 * Platforms with unreliable TSCs don't have to deal with this, they
9096 * will be compensated by the logic in vcpu_load, which sets the TSC to
9097 * catchup mode. This will catchup all VCPUs to real time, but cannot
9098 * guarantee that they stay in perfect synchronization.
9100 if (backwards_tsc) {
9101 u64 delta_cyc = max_tsc - local_tsc;
9102 list_for_each_entry(kvm, &vm_list, vm_list) {
9103 kvm->arch.backwards_tsc_observed = true;
9104 kvm_for_each_vcpu(i, vcpu, kvm) {
9105 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9106 vcpu->arch.last_host_tsc = local_tsc;
9107 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9111 * We have to disable TSC offset matching.. if you were
9112 * booting a VM while issuing an S4 host suspend....
9113 * you may have some problem. Solving this issue is
9114 * left as an exercise to the reader.
9116 kvm->arch.last_tsc_nsec = 0;
9117 kvm->arch.last_tsc_write = 0;
9124 void kvm_arch_hardware_disable(void)
9126 kvm_x86_ops->hardware_disable();
9127 drop_user_return_notifiers();
9130 int kvm_arch_hardware_setup(void)
9134 r = kvm_x86_ops->hardware_setup();
9138 if (kvm_has_tsc_control) {
9140 * Make sure the user can only configure tsc_khz values that
9141 * fit into a signed integer.
9142 * A min value is not calculated because it will always
9143 * be 1 on all machines.
9145 u64 max = min(0x7fffffffULL,
9146 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9147 kvm_max_guest_tsc_khz = max;
9149 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9152 kvm_init_msr_list();
9156 void kvm_arch_hardware_unsetup(void)
9158 kvm_x86_ops->hardware_unsetup();
9161 int kvm_arch_check_processor_compat(void)
9163 return kvm_x86_ops->check_processor_compatibility();
9166 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9168 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9170 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9172 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9174 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9177 struct static_key kvm_no_apic_vcpu __read_mostly;
9178 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9180 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9185 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9186 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9187 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9189 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9191 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9196 vcpu->arch.pio_data = page_address(page);
9198 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9200 r = kvm_mmu_create(vcpu);
9202 goto fail_free_pio_data;
9204 if (irqchip_in_kernel(vcpu->kvm)) {
9205 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9206 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9208 goto fail_mmu_destroy;
9210 static_key_slow_inc(&kvm_no_apic_vcpu);
9212 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9213 GFP_KERNEL_ACCOUNT);
9214 if (!vcpu->arch.mce_banks) {
9216 goto fail_free_lapic;
9218 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9220 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9221 GFP_KERNEL_ACCOUNT)) {
9223 goto fail_free_mce_banks;
9228 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9230 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9232 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9234 kvm_async_pf_hash_reset(vcpu);
9237 vcpu->arch.pending_external_vector = -1;
9238 vcpu->arch.preempted_in_kernel = false;
9240 kvm_hv_vcpu_init(vcpu);
9244 fail_free_mce_banks:
9245 kfree(vcpu->arch.mce_banks);
9247 kvm_free_lapic(vcpu);
9249 kvm_mmu_destroy(vcpu);
9251 free_page((unsigned long)vcpu->arch.pio_data);
9256 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9260 kvm_hv_vcpu_uninit(vcpu);
9261 kvm_pmu_destroy(vcpu);
9262 kfree(vcpu->arch.mce_banks);
9263 kvm_free_lapic(vcpu);
9264 idx = srcu_read_lock(&vcpu->kvm->srcu);
9265 kvm_mmu_destroy(vcpu);
9266 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9267 free_page((unsigned long)vcpu->arch.pio_data);
9268 if (!lapic_in_kernel(vcpu))
9269 static_key_slow_dec(&kvm_no_apic_vcpu);
9272 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9274 vcpu->arch.l1tf_flush_l1d = true;
9275 kvm_x86_ops->sched_in(vcpu, cpu);
9278 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9283 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9284 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9285 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9286 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9288 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9289 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9290 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9291 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9292 &kvm->arch.irq_sources_bitmap);
9294 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9295 mutex_init(&kvm->arch.apic_map_lock);
9296 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9298 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
9299 pvclock_update_vm_gtod_copy(kvm);
9301 kvm->arch.guest_can_read_msr_platform_info = true;
9303 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9304 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9306 kvm_hv_init_vm(kvm);
9307 kvm_page_track_init(kvm);
9308 kvm_mmu_init_vm(kvm);
9310 if (kvm_x86_ops->vm_init)
9311 return kvm_x86_ops->vm_init(kvm);
9316 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9319 kvm_mmu_unload(vcpu);
9323 static void kvm_free_vcpus(struct kvm *kvm)
9326 struct kvm_vcpu *vcpu;
9329 * Unpin any mmu pages first.
9331 kvm_for_each_vcpu(i, vcpu, kvm) {
9332 kvm_clear_async_pf_completion_queue(vcpu);
9333 kvm_unload_vcpu_mmu(vcpu);
9335 kvm_for_each_vcpu(i, vcpu, kvm)
9336 kvm_arch_vcpu_free(vcpu);
9338 mutex_lock(&kvm->lock);
9339 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9340 kvm->vcpus[i] = NULL;
9342 atomic_set(&kvm->online_vcpus, 0);
9343 mutex_unlock(&kvm->lock);
9346 void kvm_arch_sync_events(struct kvm *kvm)
9348 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9349 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9353 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9357 struct kvm_memslots *slots = kvm_memslots(kvm);
9358 struct kvm_memory_slot *slot, old;
9360 /* Called with kvm->slots_lock held. */
9361 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9364 slot = id_to_memslot(slots, id);
9370 * MAP_SHARED to prevent internal slot pages from being moved
9373 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9374 MAP_SHARED | MAP_ANONYMOUS, 0);
9375 if (IS_ERR((void *)hva))
9376 return PTR_ERR((void *)hva);
9385 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9386 struct kvm_userspace_memory_region m;
9388 m.slot = id | (i << 16);
9390 m.guest_phys_addr = gpa;
9391 m.userspace_addr = hva;
9392 m.memory_size = size;
9393 r = __kvm_set_memory_region(kvm, &m);
9399 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9403 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9405 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9409 mutex_lock(&kvm->slots_lock);
9410 r = __x86_set_memory_region(kvm, id, gpa, size);
9411 mutex_unlock(&kvm->slots_lock);
9415 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9417 void kvm_arch_destroy_vm(struct kvm *kvm)
9419 if (current->mm == kvm->mm) {
9421 * Free memory regions allocated on behalf of userspace,
9422 * unless the the memory map has changed due to process exit
9425 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9426 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9427 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9429 if (kvm_x86_ops->vm_destroy)
9430 kvm_x86_ops->vm_destroy(kvm);
9431 kvm_pic_destroy(kvm);
9432 kvm_ioapic_destroy(kvm);
9433 kvm_free_vcpus(kvm);
9434 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9435 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9436 kvm_mmu_uninit_vm(kvm);
9437 kvm_page_track_cleanup(kvm);
9438 kvm_hv_destroy_vm(kvm);
9441 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9442 struct kvm_memory_slot *dont)
9446 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9447 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9448 kvfree(free->arch.rmap[i]);
9449 free->arch.rmap[i] = NULL;
9454 if (!dont || free->arch.lpage_info[i - 1] !=
9455 dont->arch.lpage_info[i - 1]) {
9456 kvfree(free->arch.lpage_info[i - 1]);
9457 free->arch.lpage_info[i - 1] = NULL;
9461 kvm_page_track_free_memslot(free, dont);
9464 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9465 unsigned long npages)
9469 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9470 struct kvm_lpage_info *linfo;
9475 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9476 slot->base_gfn, level) + 1;
9478 slot->arch.rmap[i] =
9479 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9480 GFP_KERNEL_ACCOUNT);
9481 if (!slot->arch.rmap[i])
9486 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9490 slot->arch.lpage_info[i - 1] = linfo;
9492 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9493 linfo[0].disallow_lpage = 1;
9494 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9495 linfo[lpages - 1].disallow_lpage = 1;
9496 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9498 * If the gfn and userspace address are not aligned wrt each
9499 * other, or if explicitly asked to, disable large page
9500 * support for this slot
9502 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9503 !kvm_largepages_enabled()) {
9506 for (j = 0; j < lpages; ++j)
9507 linfo[j].disallow_lpage = 1;
9511 if (kvm_page_track_create_memslot(slot, npages))
9517 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9518 kvfree(slot->arch.rmap[i]);
9519 slot->arch.rmap[i] = NULL;
9523 kvfree(slot->arch.lpage_info[i - 1]);
9524 slot->arch.lpage_info[i - 1] = NULL;
9529 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9532 * memslots->generation has been incremented.
9533 * mmio generation may have reached its maximum value.
9535 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9538 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9539 struct kvm_memory_slot *memslot,
9540 const struct kvm_userspace_memory_region *mem,
9541 enum kvm_mr_change change)
9546 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9547 struct kvm_memory_slot *new)
9549 /* Still write protect RO slot */
9550 if (new->flags & KVM_MEM_READONLY) {
9551 kvm_mmu_slot_remove_write_access(kvm, new);
9556 * Call kvm_x86_ops dirty logging hooks when they are valid.
9558 * kvm_x86_ops->slot_disable_log_dirty is called when:
9560 * - KVM_MR_CREATE with dirty logging is disabled
9561 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9563 * The reason is, in case of PML, we need to set D-bit for any slots
9564 * with dirty logging disabled in order to eliminate unnecessary GPA
9565 * logging in PML buffer (and potential PML buffer full VMEXT). This
9566 * guarantees leaving PML enabled during guest's lifetime won't have
9567 * any additional overhead from PML when guest is running with dirty
9568 * logging disabled for memory slots.
9570 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9571 * to dirty logging mode.
9573 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9575 * In case of write protect:
9577 * Write protect all pages for dirty logging.
9579 * All the sptes including the large sptes which point to this
9580 * slot are set to readonly. We can not create any new large
9581 * spte on this slot until the end of the logging.
9583 * See the comments in fast_page_fault().
9585 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9586 if (kvm_x86_ops->slot_enable_log_dirty)
9587 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9589 kvm_mmu_slot_remove_write_access(kvm, new);
9591 if (kvm_x86_ops->slot_disable_log_dirty)
9592 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9596 void kvm_arch_commit_memory_region(struct kvm *kvm,
9597 const struct kvm_userspace_memory_region *mem,
9598 const struct kvm_memory_slot *old,
9599 const struct kvm_memory_slot *new,
9600 enum kvm_mr_change change)
9602 if (!kvm->arch.n_requested_mmu_pages)
9603 kvm_mmu_change_mmu_pages(kvm,
9604 kvm_mmu_calculate_default_mmu_pages(kvm));
9607 * Dirty logging tracks sptes in 4k granularity, meaning that large
9608 * sptes have to be split. If live migration is successful, the guest
9609 * in the source machine will be destroyed and large sptes will be
9610 * created in the destination. However, if the guest continues to run
9611 * in the source machine (for example if live migration fails), small
9612 * sptes will remain around and cause bad performance.
9614 * Scan sptes if dirty logging has been stopped, dropping those
9615 * which can be collapsed into a single large-page spte. Later
9616 * page faults will create the large-page sptes.
9618 if ((change != KVM_MR_DELETE) &&
9619 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9620 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9621 kvm_mmu_zap_collapsible_sptes(kvm, new);
9624 * Set up write protection and/or dirty logging for the new slot.
9626 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9627 * been zapped so no dirty logging staff is needed for old slot. For
9628 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9629 * new and it's also covered when dealing with the new slot.
9631 * FIXME: const-ify all uses of struct kvm_memory_slot.
9633 if (change != KVM_MR_DELETE)
9634 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9637 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9639 kvm_mmu_zap_all(kvm);
9642 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9643 struct kvm_memory_slot *slot)
9645 kvm_page_track_flush_slot(kvm, slot);
9648 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9650 return (is_guest_mode(vcpu) &&
9651 kvm_x86_ops->guest_apic_has_interrupt &&
9652 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9655 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9657 if (!list_empty_careful(&vcpu->async_pf.done))
9660 if (kvm_apic_has_events(vcpu))
9663 if (vcpu->arch.pv.pv_unhalted)
9666 if (vcpu->arch.exception.pending)
9669 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9670 (vcpu->arch.nmi_pending &&
9671 kvm_x86_ops->nmi_allowed(vcpu)))
9674 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9675 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9678 if (kvm_arch_interrupt_allowed(vcpu) &&
9679 (kvm_cpu_has_interrupt(vcpu) ||
9680 kvm_guest_apic_has_interrupt(vcpu)))
9683 if (kvm_hv_has_stimer_pending(vcpu))
9689 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9691 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9694 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9696 return vcpu->arch.preempted_in_kernel;
9699 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9701 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9704 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9706 return kvm_x86_ops->interrupt_allowed(vcpu);
9709 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9711 if (is_64_bit_mode(vcpu))
9712 return kvm_rip_read(vcpu);
9713 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9714 kvm_rip_read(vcpu));
9716 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9718 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9720 return kvm_get_linear_rip(vcpu) == linear_rip;
9722 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9724 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9726 unsigned long rflags;
9728 rflags = kvm_x86_ops->get_rflags(vcpu);
9729 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9730 rflags &= ~X86_EFLAGS_TF;
9733 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9735 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9737 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9738 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9739 rflags |= X86_EFLAGS_TF;
9740 kvm_x86_ops->set_rflags(vcpu, rflags);
9743 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9745 __kvm_set_rflags(vcpu, rflags);
9746 kvm_make_request(KVM_REQ_EVENT, vcpu);
9748 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9750 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9754 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9758 r = kvm_mmu_reload(vcpu);
9762 if (!vcpu->arch.mmu->direct_map &&
9763 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9766 vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9769 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9771 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9774 static inline u32 kvm_async_pf_next_probe(u32 key)
9776 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9779 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9781 u32 key = kvm_async_pf_hash_fn(gfn);
9783 while (vcpu->arch.apf.gfns[key] != ~0)
9784 key = kvm_async_pf_next_probe(key);
9786 vcpu->arch.apf.gfns[key] = gfn;
9789 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9792 u32 key = kvm_async_pf_hash_fn(gfn);
9794 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9795 (vcpu->arch.apf.gfns[key] != gfn &&
9796 vcpu->arch.apf.gfns[key] != ~0); i++)
9797 key = kvm_async_pf_next_probe(key);
9802 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9804 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9807 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9811 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9813 vcpu->arch.apf.gfns[i] = ~0;
9815 j = kvm_async_pf_next_probe(j);
9816 if (vcpu->arch.apf.gfns[j] == ~0)
9818 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9820 * k lies cyclically in ]i,j]
9822 * |....j i.k.| or |.k..j i...|
9824 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9825 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9830 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9833 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9837 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9840 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9844 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
9846 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
9849 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9850 (vcpu->arch.apf.send_user_only &&
9851 kvm_x86_ops->get_cpl(vcpu) == 0))
9857 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
9859 if (unlikely(!lapic_in_kernel(vcpu) ||
9860 kvm_event_needs_reinjection(vcpu) ||
9861 vcpu->arch.exception.pending))
9864 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
9868 * If interrupts are off we cannot even use an artificial
9871 return kvm_x86_ops->interrupt_allowed(vcpu);
9874 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9875 struct kvm_async_pf *work)
9877 struct x86_exception fault;
9879 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9880 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9882 if (kvm_can_deliver_async_pf(vcpu) &&
9883 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9884 fault.vector = PF_VECTOR;
9885 fault.error_code_valid = true;
9886 fault.error_code = 0;
9887 fault.nested_page_fault = false;
9888 fault.address = work->arch.token;
9889 fault.async_page_fault = true;
9890 kvm_inject_page_fault(vcpu, &fault);
9893 * It is not possible to deliver a paravirtualized asynchronous
9894 * page fault, but putting the guest in an artificial halt state
9895 * can be beneficial nevertheless: if an interrupt arrives, we
9896 * can deliver it timely and perhaps the guest will schedule
9897 * another process. When the instruction that triggered a page
9898 * fault is retried, hopefully the page will be ready in the host.
9900 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9904 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9905 struct kvm_async_pf *work)
9907 struct x86_exception fault;
9910 if (work->wakeup_all)
9911 work->arch.token = ~0; /* broadcast wakeup */
9913 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9914 trace_kvm_async_pf_ready(work->arch.token, work->gva);
9916 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9917 !apf_get_user(vcpu, &val)) {
9918 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9919 vcpu->arch.exception.pending &&
9920 vcpu->arch.exception.nr == PF_VECTOR &&
9921 !apf_put_user(vcpu, 0)) {
9922 vcpu->arch.exception.injected = false;
9923 vcpu->arch.exception.pending = false;
9924 vcpu->arch.exception.nr = 0;
9925 vcpu->arch.exception.has_error_code = false;
9926 vcpu->arch.exception.error_code = 0;
9927 vcpu->arch.exception.has_payload = false;
9928 vcpu->arch.exception.payload = 0;
9929 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9930 fault.vector = PF_VECTOR;
9931 fault.error_code_valid = true;
9932 fault.error_code = 0;
9933 fault.nested_page_fault = false;
9934 fault.address = work->arch.token;
9935 fault.async_page_fault = true;
9936 kvm_inject_page_fault(vcpu, &fault);
9939 vcpu->arch.apf.halted = false;
9940 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9943 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9945 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9948 return kvm_can_do_async_pf(vcpu);
9951 void kvm_arch_start_assignment(struct kvm *kvm)
9953 atomic_inc(&kvm->arch.assigned_device_count);
9955 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9957 void kvm_arch_end_assignment(struct kvm *kvm)
9959 atomic_dec(&kvm->arch.assigned_device_count);
9961 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9963 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9965 return atomic_read(&kvm->arch.assigned_device_count);
9967 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9969 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9971 atomic_inc(&kvm->arch.noncoherent_dma_count);
9973 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9975 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9977 atomic_dec(&kvm->arch.noncoherent_dma_count);
9979 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9981 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9983 return atomic_read(&kvm->arch.noncoherent_dma_count);
9985 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9987 bool kvm_arch_has_irq_bypass(void)
9989 return kvm_x86_ops->update_pi_irte != NULL;
9992 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9993 struct irq_bypass_producer *prod)
9995 struct kvm_kernel_irqfd *irqfd =
9996 container_of(cons, struct kvm_kernel_irqfd, consumer);
9998 irqfd->producer = prod;
10000 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10001 prod->irq, irqfd->gsi, 1);
10004 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10005 struct irq_bypass_producer *prod)
10008 struct kvm_kernel_irqfd *irqfd =
10009 container_of(cons, struct kvm_kernel_irqfd, consumer);
10011 WARN_ON(irqfd->producer != prod);
10012 irqfd->producer = NULL;
10015 * When producer of consumer is unregistered, we change back to
10016 * remapped mode, so we can re-use the current implementation
10017 * when the irq is masked/disabled or the consumer side (KVM
10018 * int this case doesn't want to receive the interrupts.
10020 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10022 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10023 " fails: %d\n", irqfd->consumer.token, ret);
10026 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10027 uint32_t guest_irq, bool set)
10029 if (!kvm_x86_ops->update_pi_irte)
10032 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10035 bool kvm_vector_hashing_enabled(void)
10037 return vector_hashing;
10039 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10041 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10043 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10045 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10048 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10049 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10050 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10051 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10052 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10053 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10054 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10055 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10056 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10057 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10058 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10059 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10060 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10061 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10062 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
10063 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10064 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10065 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10066 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);