1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
5 * derived from drivers/kvm/kvm_main.c
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright (C) 2008 Qumranet, Inc.
9 * Copyright IBM Corporation, 2008
10 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
13 * Avi Kivity <avi@qumranet.com>
14 * Yaniv Kamay <yaniv@qumranet.com>
15 * Amit Shah <amit.shah@qumranet.com>
16 * Ben-Ami Yassour <benami@il.ibm.com>
19 #include <linux/kvm_host.h>
24 #include "kvm_cache_regs.h"
30 #include <linux/clocksource.h>
31 #include <linux/interrupt.h>
32 #include <linux/kvm.h>
34 #include <linux/vmalloc.h>
35 #include <linux/export.h>
36 #include <linux/moduleparam.h>
37 #include <linux/mman.h>
38 #include <linux/highmem.h>
39 #include <linux/iommu.h>
40 #include <linux/intel-iommu.h>
41 #include <linux/cpufreq.h>
42 #include <linux/user-return-notifier.h>
43 #include <linux/srcu.h>
44 #include <linux/slab.h>
45 #include <linux/perf_event.h>
46 #include <linux/uaccess.h>
47 #include <linux/hash.h>
48 #include <linux/pci.h>
49 #include <linux/timekeeper_internal.h>
50 #include <linux/pvclock_gtod.h>
51 #include <linux/kvm_irqfd.h>
52 #include <linux/irqbypass.h>
53 #include <linux/sched/stat.h>
54 #include <linux/sched/isolation.h>
55 #include <linux/mem_encrypt.h>
57 #include <trace/events/kvm.h>
59 #include <asm/debugreg.h>
63 #include <linux/kernel_stat.h>
64 #include <asm/fpu/internal.h> /* Ugh! */
65 #include <asm/pvclock.h>
66 #include <asm/div64.h>
67 #include <asm/irq_remapping.h>
68 #include <asm/mshyperv.h>
69 #include <asm/hypervisor.h>
70 #include <asm/intel_pt.h>
71 #include <clocksource/hyperv_timer.h>
73 #define CREATE_TRACE_POINTS
76 #define MAX_IO_MSRS 256
77 #define KVM_MAX_MCE_BANKS 32
78 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
79 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
81 #define emul_to_vcpu(ctxt) \
82 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
85 * - enable syscall per default because its emulated by KVM
86 * - enable LME and LMA per default on 64 bit KVM
90 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
92 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
95 #define VM_STAT(x, ...) offsetof(struct kvm, stat.x), KVM_STAT_VM, ## __VA_ARGS__
96 #define VCPU_STAT(x, ...) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU, ## __VA_ARGS__
98 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
99 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
101 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
102 static void process_nmi(struct kvm_vcpu *vcpu);
103 static void enter_smm(struct kvm_vcpu *vcpu);
104 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
105 static void store_regs(struct kvm_vcpu *vcpu);
106 static int sync_regs(struct kvm_vcpu *vcpu);
108 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
109 EXPORT_SYMBOL_GPL(kvm_x86_ops);
111 static bool __read_mostly ignore_msrs = 0;
112 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
114 static bool __read_mostly report_ignored_msrs = true;
115 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
117 unsigned int min_timer_period_us = 200;
118 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
120 static bool __read_mostly kvmclock_periodic_sync = true;
121 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
123 bool __read_mostly kvm_has_tsc_control;
124 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
125 u32 __read_mostly kvm_max_guest_tsc_khz;
126 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
127 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
128 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
129 u64 __read_mostly kvm_max_tsc_scaling_ratio;
130 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
131 u64 __read_mostly kvm_default_tsc_scaling_ratio;
132 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
134 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
135 static u32 __read_mostly tsc_tolerance_ppm = 250;
136 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
139 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
140 * adaptive tuning starting from default advancment of 1000ns. '0' disables
141 * advancement entirely. Any other value is used as-is and disables adaptive
142 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
144 static int __read_mostly lapic_timer_advance_ns = -1;
145 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
147 static bool __read_mostly vector_hashing = true;
148 module_param(vector_hashing, bool, S_IRUGO);
150 bool __read_mostly enable_vmware_backdoor = false;
151 module_param(enable_vmware_backdoor, bool, S_IRUGO);
152 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
154 static bool __read_mostly force_emulation_prefix = false;
155 module_param(force_emulation_prefix, bool, S_IRUGO);
157 int __read_mostly pi_inject_timer = -1;
158 module_param(pi_inject_timer, bint, S_IRUGO | S_IWUSR);
160 #define KVM_NR_SHARED_MSRS 16
162 struct kvm_shared_msrs_global {
164 u32 msrs[KVM_NR_SHARED_MSRS];
167 struct kvm_shared_msrs {
168 struct user_return_notifier urn;
170 struct kvm_shared_msr_values {
173 } values[KVM_NR_SHARED_MSRS];
176 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
177 static struct kvm_shared_msrs __percpu *shared_msrs;
179 struct kvm_stats_debugfs_item debugfs_entries[] = {
180 { "pf_fixed", VCPU_STAT(pf_fixed) },
181 { "pf_guest", VCPU_STAT(pf_guest) },
182 { "tlb_flush", VCPU_STAT(tlb_flush) },
183 { "invlpg", VCPU_STAT(invlpg) },
184 { "exits", VCPU_STAT(exits) },
185 { "io_exits", VCPU_STAT(io_exits) },
186 { "mmio_exits", VCPU_STAT(mmio_exits) },
187 { "signal_exits", VCPU_STAT(signal_exits) },
188 { "irq_window", VCPU_STAT(irq_window_exits) },
189 { "nmi_window", VCPU_STAT(nmi_window_exits) },
190 { "halt_exits", VCPU_STAT(halt_exits) },
191 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
192 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
193 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
194 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
195 { "hypercalls", VCPU_STAT(hypercalls) },
196 { "request_irq", VCPU_STAT(request_irq_exits) },
197 { "irq_exits", VCPU_STAT(irq_exits) },
198 { "host_state_reload", VCPU_STAT(host_state_reload) },
199 { "fpu_reload", VCPU_STAT(fpu_reload) },
200 { "insn_emulation", VCPU_STAT(insn_emulation) },
201 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
202 { "irq_injections", VCPU_STAT(irq_injections) },
203 { "nmi_injections", VCPU_STAT(nmi_injections) },
204 { "req_event", VCPU_STAT(req_event) },
205 { "l1d_flush", VCPU_STAT(l1d_flush) },
206 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
207 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
208 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
209 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
210 { "mmu_flooded", VM_STAT(mmu_flooded) },
211 { "mmu_recycled", VM_STAT(mmu_recycled) },
212 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
213 { "mmu_unsync", VM_STAT(mmu_unsync) },
214 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
215 { "largepages", VM_STAT(lpages, .mode = 0444) },
216 { "max_mmu_page_hash_collisions",
217 VM_STAT(max_mmu_page_hash_collisions) },
221 u64 __read_mostly host_xcr0;
223 struct kmem_cache *x86_fpu_cache;
224 EXPORT_SYMBOL_GPL(x86_fpu_cache);
226 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
228 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
231 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
232 vcpu->arch.apf.gfns[i] = ~0;
235 static void kvm_on_user_return(struct user_return_notifier *urn)
238 struct kvm_shared_msrs *locals
239 = container_of(urn, struct kvm_shared_msrs, urn);
240 struct kvm_shared_msr_values *values;
244 * Disabling irqs at this point since the following code could be
245 * interrupted and executed through kvm_arch_hardware_disable()
247 local_irq_save(flags);
248 if (locals->registered) {
249 locals->registered = false;
250 user_return_notifier_unregister(urn);
252 local_irq_restore(flags);
253 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
254 values = &locals->values[slot];
255 if (values->host != values->curr) {
256 wrmsrl(shared_msrs_global.msrs[slot], values->host);
257 values->curr = values->host;
262 static void shared_msr_update(unsigned slot, u32 msr)
265 unsigned int cpu = smp_processor_id();
266 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
268 /* only read, and nobody should modify it at this time,
269 * so don't need lock */
270 if (slot >= shared_msrs_global.nr) {
271 printk(KERN_ERR "kvm: invalid MSR slot!");
274 rdmsrl_safe(msr, &value);
275 smsr->values[slot].host = value;
276 smsr->values[slot].curr = value;
279 void kvm_define_shared_msr(unsigned slot, u32 msr)
281 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
282 shared_msrs_global.msrs[slot] = msr;
283 if (slot >= shared_msrs_global.nr)
284 shared_msrs_global.nr = slot + 1;
286 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
288 static void kvm_shared_msr_cpu_online(void)
292 for (i = 0; i < shared_msrs_global.nr; ++i)
293 shared_msr_update(i, shared_msrs_global.msrs[i]);
296 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
298 unsigned int cpu = smp_processor_id();
299 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
302 if (((value ^ smsr->values[slot].curr) & mask) == 0)
304 smsr->values[slot].curr = value;
305 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
309 if (!smsr->registered) {
310 smsr->urn.on_user_return = kvm_on_user_return;
311 user_return_notifier_register(&smsr->urn);
312 smsr->registered = true;
316 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
318 static void drop_user_return_notifiers(void)
320 unsigned int cpu = smp_processor_id();
321 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
323 if (smsr->registered)
324 kvm_on_user_return(&smsr->urn);
327 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
329 return vcpu->arch.apic_base;
331 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
333 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
335 return kvm_apic_mode(kvm_get_apic_base(vcpu));
337 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
339 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
341 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
342 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
343 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
344 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
346 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
348 if (!msr_info->host_initiated) {
349 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
351 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
355 kvm_lapic_set_base(vcpu, msr_info->data);
358 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
360 asmlinkage __visible void kvm_spurious_fault(void)
362 /* Fault while not rebooting. We want the trace. */
363 BUG_ON(!kvm_rebooting);
365 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
367 #define EXCPT_BENIGN 0
368 #define EXCPT_CONTRIBUTORY 1
371 static int exception_class(int vector)
381 return EXCPT_CONTRIBUTORY;
388 #define EXCPT_FAULT 0
390 #define EXCPT_ABORT 2
391 #define EXCPT_INTERRUPT 3
393 static int exception_type(int vector)
397 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
398 return EXCPT_INTERRUPT;
402 /* #DB is trap, as instruction watchpoints are handled elsewhere */
403 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
406 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
409 /* Reserved exceptions will result in fault */
413 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
415 unsigned nr = vcpu->arch.exception.nr;
416 bool has_payload = vcpu->arch.exception.has_payload;
417 unsigned long payload = vcpu->arch.exception.payload;
425 * "Certain debug exceptions may clear bit 0-3. The
426 * remaining contents of the DR6 register are never
427 * cleared by the processor".
429 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
431 * DR6.RTM is set by all #DB exceptions that don't clear it.
433 vcpu->arch.dr6 |= DR6_RTM;
434 vcpu->arch.dr6 |= payload;
436 * Bit 16 should be set in the payload whenever the #DB
437 * exception should clear DR6.RTM. This makes the payload
438 * compatible with the pending debug exceptions under VMX.
439 * Though not currently documented in the SDM, this also
440 * makes the payload compatible with the exit qualification
441 * for #DB exceptions under VMX.
443 vcpu->arch.dr6 ^= payload & DR6_RTM;
446 vcpu->arch.cr2 = payload;
450 vcpu->arch.exception.has_payload = false;
451 vcpu->arch.exception.payload = 0;
453 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
455 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
456 unsigned nr, bool has_error, u32 error_code,
457 bool has_payload, unsigned long payload, bool reinject)
462 kvm_make_request(KVM_REQ_EVENT, vcpu);
464 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
466 if (has_error && !is_protmode(vcpu))
470 * On vmentry, vcpu->arch.exception.pending is only
471 * true if an event injection was blocked by
472 * nested_run_pending. In that case, however,
473 * vcpu_enter_guest requests an immediate exit,
474 * and the guest shouldn't proceed far enough to
477 WARN_ON_ONCE(vcpu->arch.exception.pending);
478 vcpu->arch.exception.injected = true;
479 if (WARN_ON_ONCE(has_payload)) {
481 * A reinjected event has already
482 * delivered its payload.
488 vcpu->arch.exception.pending = true;
489 vcpu->arch.exception.injected = false;
491 vcpu->arch.exception.has_error_code = has_error;
492 vcpu->arch.exception.nr = nr;
493 vcpu->arch.exception.error_code = error_code;
494 vcpu->arch.exception.has_payload = has_payload;
495 vcpu->arch.exception.payload = payload;
497 * In guest mode, payload delivery should be deferred,
498 * so that the L1 hypervisor can intercept #PF before
499 * CR2 is modified (or intercept #DB before DR6 is
500 * modified under nVMX). However, for ABI
501 * compatibility with KVM_GET_VCPU_EVENTS and
502 * KVM_SET_VCPU_EVENTS, we can't delay payload
503 * delivery unless userspace has enabled this
504 * functionality via the per-VM capability,
505 * KVM_CAP_EXCEPTION_PAYLOAD.
507 if (!vcpu->kvm->arch.exception_payload_enabled ||
508 !is_guest_mode(vcpu))
509 kvm_deliver_exception_payload(vcpu);
513 /* to check exception */
514 prev_nr = vcpu->arch.exception.nr;
515 if (prev_nr == DF_VECTOR) {
516 /* triple fault -> shutdown */
517 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
520 class1 = exception_class(prev_nr);
521 class2 = exception_class(nr);
522 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
523 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
525 * Generate double fault per SDM Table 5-5. Set
526 * exception.pending = true so that the double fault
527 * can trigger a nested vmexit.
529 vcpu->arch.exception.pending = true;
530 vcpu->arch.exception.injected = false;
531 vcpu->arch.exception.has_error_code = true;
532 vcpu->arch.exception.nr = DF_VECTOR;
533 vcpu->arch.exception.error_code = 0;
534 vcpu->arch.exception.has_payload = false;
535 vcpu->arch.exception.payload = 0;
537 /* replace previous exception with a new one in a hope
538 that instruction re-execution will regenerate lost
543 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
545 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
547 EXPORT_SYMBOL_GPL(kvm_queue_exception);
549 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
551 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
553 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
555 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
556 unsigned long payload)
558 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
561 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
562 u32 error_code, unsigned long payload)
564 kvm_multiple_exception(vcpu, nr, true, error_code,
565 true, payload, false);
568 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
571 kvm_inject_gp(vcpu, 0);
573 return kvm_skip_emulated_instruction(vcpu);
577 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
579 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
581 ++vcpu->stat.pf_guest;
582 vcpu->arch.exception.nested_apf =
583 is_guest_mode(vcpu) && fault->async_page_fault;
584 if (vcpu->arch.exception.nested_apf) {
585 vcpu->arch.apf.nested_apf_token = fault->address;
586 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
588 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
592 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
594 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
596 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
597 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
599 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
601 return fault->nested_page_fault;
604 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
606 atomic_inc(&vcpu->arch.nmi_queued);
607 kvm_make_request(KVM_REQ_NMI, vcpu);
609 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
611 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
613 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
615 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
617 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
619 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
621 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
624 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
625 * a #GP and return false.
627 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
629 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
631 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
634 EXPORT_SYMBOL_GPL(kvm_require_cpl);
636 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
638 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
641 kvm_queue_exception(vcpu, UD_VECTOR);
644 EXPORT_SYMBOL_GPL(kvm_require_dr);
647 * This function will be used to read from the physical memory of the currently
648 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
649 * can read from guest physical or from the guest's guest physical memory.
651 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
652 gfn_t ngfn, void *data, int offset, int len,
655 struct x86_exception exception;
659 ngpa = gfn_to_gpa(ngfn);
660 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
661 if (real_gfn == UNMAPPED_GVA)
664 real_gfn = gpa_to_gfn(real_gfn);
666 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
668 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
670 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
671 void *data, int offset, int len, u32 access)
673 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
674 data, offset, len, access);
677 static inline u64 pdptr_rsvd_bits(struct kvm_vcpu *vcpu)
679 return rsvd_bits(cpuid_maxphyaddr(vcpu), 63) | rsvd_bits(5, 8) |
684 * Load the pae pdptrs. Return 1 if they are all valid, 0 otherwise.
686 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
688 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
689 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
692 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
694 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
695 offset * sizeof(u64), sizeof(pdpte),
696 PFERR_USER_MASK|PFERR_WRITE_MASK);
701 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
702 if ((pdpte[i] & PT_PRESENT_MASK) &&
703 (pdpte[i] & pdptr_rsvd_bits(vcpu))) {
710 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
711 __set_bit(VCPU_EXREG_PDPTR,
712 (unsigned long *)&vcpu->arch.regs_avail);
713 __set_bit(VCPU_EXREG_PDPTR,
714 (unsigned long *)&vcpu->arch.regs_dirty);
719 EXPORT_SYMBOL_GPL(load_pdptrs);
721 bool pdptrs_changed(struct kvm_vcpu *vcpu)
723 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
729 if (!is_pae_paging(vcpu))
732 if (!test_bit(VCPU_EXREG_PDPTR,
733 (unsigned long *)&vcpu->arch.regs_avail))
736 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
737 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
738 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
739 PFERR_USER_MASK | PFERR_WRITE_MASK);
742 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
747 EXPORT_SYMBOL_GPL(pdptrs_changed);
749 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
751 unsigned long old_cr0 = kvm_read_cr0(vcpu);
752 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
757 if (cr0 & 0xffffffff00000000UL)
761 cr0 &= ~CR0_RESERVED_BITS;
763 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
766 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
769 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
771 if ((vcpu->arch.efer & EFER_LME)) {
776 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
781 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
786 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
789 kvm_x86_ops->set_cr0(vcpu, cr0);
791 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
792 kvm_clear_async_pf_completion_queue(vcpu);
793 kvm_async_pf_hash_reset(vcpu);
796 if ((cr0 ^ old_cr0) & update_bits)
797 kvm_mmu_reset_context(vcpu);
799 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
800 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
801 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
802 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
806 EXPORT_SYMBOL_GPL(kvm_set_cr0);
808 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
810 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
812 EXPORT_SYMBOL_GPL(kvm_lmsw);
814 void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
816 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
817 !vcpu->guest_xcr0_loaded) {
818 /* kvm_set_xcr() also depends on this */
819 if (vcpu->arch.xcr0 != host_xcr0)
820 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
821 vcpu->guest_xcr0_loaded = 1;
824 EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
826 void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
828 if (vcpu->guest_xcr0_loaded) {
829 if (vcpu->arch.xcr0 != host_xcr0)
830 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
831 vcpu->guest_xcr0_loaded = 0;
834 EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
836 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
839 u64 old_xcr0 = vcpu->arch.xcr0;
842 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
843 if (index != XCR_XFEATURE_ENABLED_MASK)
845 if (!(xcr0 & XFEATURE_MASK_FP))
847 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
851 * Do not allow the guest to set bits that we do not support
852 * saving. However, xcr0 bit 0 is always set, even if the
853 * emulated CPU does not support XSAVE (see fx_init).
855 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
856 if (xcr0 & ~valid_bits)
859 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
860 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
863 if (xcr0 & XFEATURE_MASK_AVX512) {
864 if (!(xcr0 & XFEATURE_MASK_YMM))
866 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
869 vcpu->arch.xcr0 = xcr0;
871 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
872 kvm_update_cpuid(vcpu);
876 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
878 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
879 __kvm_set_xcr(vcpu, index, xcr)) {
880 kvm_inject_gp(vcpu, 0);
885 EXPORT_SYMBOL_GPL(kvm_set_xcr);
887 static int kvm_valid_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
889 if (cr4 & CR4_RESERVED_BITS)
892 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
895 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
898 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
901 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
904 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
907 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
910 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
916 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
918 unsigned long old_cr4 = kvm_read_cr4(vcpu);
919 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
920 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
922 if (kvm_valid_cr4(vcpu, cr4))
925 if (is_long_mode(vcpu)) {
926 if (!(cr4 & X86_CR4_PAE))
928 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
929 && ((cr4 ^ old_cr4) & pdptr_bits)
930 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
934 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
935 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
938 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
939 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
943 if (kvm_x86_ops->set_cr4(vcpu, cr4))
946 if (((cr4 ^ old_cr4) & pdptr_bits) ||
947 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
948 kvm_mmu_reset_context(vcpu);
950 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
951 kvm_update_cpuid(vcpu);
955 EXPORT_SYMBOL_GPL(kvm_set_cr4);
957 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
959 bool skip_tlb_flush = false;
961 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
964 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
965 cr3 &= ~X86_CR3_PCID_NOFLUSH;
969 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
970 if (!skip_tlb_flush) {
971 kvm_mmu_sync_roots(vcpu);
972 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
977 if (is_long_mode(vcpu) &&
978 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
980 else if (is_pae_paging(vcpu) &&
981 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
984 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
985 vcpu->arch.cr3 = cr3;
986 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
990 EXPORT_SYMBOL_GPL(kvm_set_cr3);
992 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
994 if (cr8 & CR8_RESERVED_BITS)
996 if (lapic_in_kernel(vcpu))
997 kvm_lapic_set_tpr(vcpu, cr8);
999 vcpu->arch.cr8 = cr8;
1002 EXPORT_SYMBOL_GPL(kvm_set_cr8);
1004 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
1006 if (lapic_in_kernel(vcpu))
1007 return kvm_lapic_get_cr8(vcpu);
1009 return vcpu->arch.cr8;
1011 EXPORT_SYMBOL_GPL(kvm_get_cr8);
1013 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1017 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1018 for (i = 0; i < KVM_NR_DB_REGS; i++)
1019 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1020 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1024 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1026 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1027 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1030 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1034 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1035 dr7 = vcpu->arch.guest_debug_dr7;
1037 dr7 = vcpu->arch.dr7;
1038 kvm_x86_ops->set_dr7(vcpu, dr7);
1039 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1040 if (dr7 & DR7_BP_EN_MASK)
1041 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1044 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1046 u64 fixed = DR6_FIXED_1;
1048 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1053 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1057 vcpu->arch.db[dr] = val;
1058 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1059 vcpu->arch.eff_db[dr] = val;
1064 if (val & 0xffffffff00000000ULL)
1065 return -1; /* #GP */
1066 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1067 kvm_update_dr6(vcpu);
1072 if (val & 0xffffffff00000000ULL)
1073 return -1; /* #GP */
1074 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1075 kvm_update_dr7(vcpu);
1082 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1084 if (__kvm_set_dr(vcpu, dr, val)) {
1085 kvm_inject_gp(vcpu, 0);
1090 EXPORT_SYMBOL_GPL(kvm_set_dr);
1092 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1096 *val = vcpu->arch.db[dr];
1101 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1102 *val = vcpu->arch.dr6;
1104 *val = kvm_x86_ops->get_dr6(vcpu);
1109 *val = vcpu->arch.dr7;
1114 EXPORT_SYMBOL_GPL(kvm_get_dr);
1116 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1118 u32 ecx = kvm_rcx_read(vcpu);
1122 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1125 kvm_rax_write(vcpu, (u32)data);
1126 kvm_rdx_write(vcpu, data >> 32);
1129 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1132 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1133 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1135 * This list is modified at module load time to reflect the
1136 * capabilities of the host cpu. This capabilities test skips MSRs that are
1137 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1138 * may depend on host virtualization features rather than host cpu features.
1141 static u32 msrs_to_save[] = {
1142 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1144 #ifdef CONFIG_X86_64
1145 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1147 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1148 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1150 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1151 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1152 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1153 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1154 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1155 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1156 MSR_IA32_UMWAIT_CONTROL,
1158 MSR_ARCH_PERFMON_FIXED_CTR0, MSR_ARCH_PERFMON_FIXED_CTR1,
1159 MSR_ARCH_PERFMON_FIXED_CTR0 + 2, MSR_ARCH_PERFMON_FIXED_CTR0 + 3,
1160 MSR_CORE_PERF_FIXED_CTR_CTRL, MSR_CORE_PERF_GLOBAL_STATUS,
1161 MSR_CORE_PERF_GLOBAL_CTRL, MSR_CORE_PERF_GLOBAL_OVF_CTRL,
1162 MSR_ARCH_PERFMON_PERFCTR0, MSR_ARCH_PERFMON_PERFCTR1,
1163 MSR_ARCH_PERFMON_PERFCTR0 + 2, MSR_ARCH_PERFMON_PERFCTR0 + 3,
1164 MSR_ARCH_PERFMON_PERFCTR0 + 4, MSR_ARCH_PERFMON_PERFCTR0 + 5,
1165 MSR_ARCH_PERFMON_PERFCTR0 + 6, MSR_ARCH_PERFMON_PERFCTR0 + 7,
1166 MSR_ARCH_PERFMON_PERFCTR0 + 8, MSR_ARCH_PERFMON_PERFCTR0 + 9,
1167 MSR_ARCH_PERFMON_PERFCTR0 + 10, MSR_ARCH_PERFMON_PERFCTR0 + 11,
1168 MSR_ARCH_PERFMON_PERFCTR0 + 12, MSR_ARCH_PERFMON_PERFCTR0 + 13,
1169 MSR_ARCH_PERFMON_PERFCTR0 + 14, MSR_ARCH_PERFMON_PERFCTR0 + 15,
1170 MSR_ARCH_PERFMON_PERFCTR0 + 16, MSR_ARCH_PERFMON_PERFCTR0 + 17,
1171 MSR_ARCH_PERFMON_EVENTSEL0, MSR_ARCH_PERFMON_EVENTSEL1,
1172 MSR_ARCH_PERFMON_EVENTSEL0 + 2, MSR_ARCH_PERFMON_EVENTSEL0 + 3,
1173 MSR_ARCH_PERFMON_EVENTSEL0 + 4, MSR_ARCH_PERFMON_EVENTSEL0 + 5,
1174 MSR_ARCH_PERFMON_EVENTSEL0 + 6, MSR_ARCH_PERFMON_EVENTSEL0 + 7,
1175 MSR_ARCH_PERFMON_EVENTSEL0 + 8, MSR_ARCH_PERFMON_EVENTSEL0 + 9,
1176 MSR_ARCH_PERFMON_EVENTSEL0 + 10, MSR_ARCH_PERFMON_EVENTSEL0 + 11,
1177 MSR_ARCH_PERFMON_EVENTSEL0 + 12, MSR_ARCH_PERFMON_EVENTSEL0 + 13,
1178 MSR_ARCH_PERFMON_EVENTSEL0 + 14, MSR_ARCH_PERFMON_EVENTSEL0 + 15,
1179 MSR_ARCH_PERFMON_EVENTSEL0 + 16, MSR_ARCH_PERFMON_EVENTSEL0 + 17,
1182 static unsigned num_msrs_to_save;
1184 static u32 emulated_msrs[] = {
1185 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1186 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1187 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1188 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1189 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1190 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1191 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1193 HV_X64_MSR_VP_INDEX,
1194 HV_X64_MSR_VP_RUNTIME,
1195 HV_X64_MSR_SCONTROL,
1196 HV_X64_MSR_STIMER0_CONFIG,
1197 HV_X64_MSR_VP_ASSIST_PAGE,
1198 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1199 HV_X64_MSR_TSC_EMULATION_STATUS,
1201 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1204 MSR_IA32_TSC_ADJUST,
1205 MSR_IA32_TSCDEADLINE,
1206 MSR_IA32_ARCH_CAPABILITIES,
1207 MSR_IA32_MISC_ENABLE,
1208 MSR_IA32_MCG_STATUS,
1210 MSR_IA32_MCG_EXT_CTL,
1214 MSR_MISC_FEATURES_ENABLES,
1215 MSR_AMD64_VIRT_SPEC_CTRL,
1219 * The following list leaves out MSRs whose values are determined
1220 * by arch/x86/kvm/vmx/nested.c based on CPUID or other MSRs.
1221 * We always support the "true" VMX control MSRs, even if the host
1222 * processor does not, so I am putting these registers here rather
1223 * than in msrs_to_save.
1226 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1227 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1228 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1229 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1231 MSR_IA32_VMX_CR0_FIXED0,
1232 MSR_IA32_VMX_CR4_FIXED0,
1233 MSR_IA32_VMX_VMCS_ENUM,
1234 MSR_IA32_VMX_PROCBASED_CTLS2,
1235 MSR_IA32_VMX_EPT_VPID_CAP,
1236 MSR_IA32_VMX_VMFUNC,
1239 MSR_KVM_POLL_CONTROL,
1242 static unsigned num_emulated_msrs;
1245 * List of msr numbers which are used to expose MSR-based features that
1246 * can be used by a hypervisor to validate requested CPU features.
1248 static u32 msr_based_features[] = {
1250 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1251 MSR_IA32_VMX_PINBASED_CTLS,
1252 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1253 MSR_IA32_VMX_PROCBASED_CTLS,
1254 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1255 MSR_IA32_VMX_EXIT_CTLS,
1256 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1257 MSR_IA32_VMX_ENTRY_CTLS,
1259 MSR_IA32_VMX_CR0_FIXED0,
1260 MSR_IA32_VMX_CR0_FIXED1,
1261 MSR_IA32_VMX_CR4_FIXED0,
1262 MSR_IA32_VMX_CR4_FIXED1,
1263 MSR_IA32_VMX_VMCS_ENUM,
1264 MSR_IA32_VMX_PROCBASED_CTLS2,
1265 MSR_IA32_VMX_EPT_VPID_CAP,
1266 MSR_IA32_VMX_VMFUNC,
1270 MSR_IA32_ARCH_CAPABILITIES,
1273 static unsigned int num_msr_based_features;
1275 static u64 kvm_get_arch_capabilities(void)
1279 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1280 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1283 * If we're doing cache flushes (either "always" or "cond")
1284 * we will do one whenever the guest does a vmlaunch/vmresume.
1285 * If an outer hypervisor is doing the cache flush for us
1286 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1287 * capability to the guest too, and if EPT is disabled we're not
1288 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1289 * require a nested hypervisor to do a flush of its own.
1291 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1292 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1294 if (!boot_cpu_has_bug(X86_BUG_CPU_MELTDOWN))
1295 data |= ARCH_CAP_RDCL_NO;
1296 if (!boot_cpu_has_bug(X86_BUG_SPEC_STORE_BYPASS))
1297 data |= ARCH_CAP_SSB_NO;
1298 if (!boot_cpu_has_bug(X86_BUG_MDS))
1299 data |= ARCH_CAP_MDS_NO;
1302 * On TAA affected systems, export MDS_NO=0 when:
1303 * - TSX is enabled on the host, i.e. X86_FEATURE_RTM=1.
1304 * - Updated microcode is present. This is detected by
1305 * the presence of ARCH_CAP_TSX_CTRL_MSR and ensures
1306 * that VERW clears CPU buffers.
1308 * When MDS_NO=0 is exported, guests deploy clear CPU buffer
1309 * mitigation and don't complain:
1311 * "Vulnerable: Clear CPU buffers attempted, no microcode"
1313 * If TSX is disabled on the system, guests are also mitigated against
1314 * TAA and clear CPU buffer mitigation is not required for guests.
1316 if (boot_cpu_has_bug(X86_BUG_TAA) && boot_cpu_has(X86_FEATURE_RTM) &&
1317 (data & ARCH_CAP_TSX_CTRL_MSR))
1318 data &= ~ARCH_CAP_MDS_NO;
1323 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1325 switch (msr->index) {
1326 case MSR_IA32_ARCH_CAPABILITIES:
1327 msr->data = kvm_get_arch_capabilities();
1329 case MSR_IA32_UCODE_REV:
1330 rdmsrl_safe(msr->index, &msr->data);
1333 if (kvm_x86_ops->get_msr_feature(msr))
1339 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1341 struct kvm_msr_entry msr;
1345 r = kvm_get_msr_feature(&msr);
1354 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1356 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1359 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1362 if (efer & (EFER_LME | EFER_LMA) &&
1363 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1366 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1372 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1374 if (efer & efer_reserved_bits)
1377 return __kvm_valid_efer(vcpu, efer);
1379 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1381 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1383 u64 old_efer = vcpu->arch.efer;
1384 u64 efer = msr_info->data;
1386 if (efer & efer_reserved_bits)
1389 if (!msr_info->host_initiated) {
1390 if (!__kvm_valid_efer(vcpu, efer))
1393 if (is_paging(vcpu) &&
1394 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1399 efer |= vcpu->arch.efer & EFER_LMA;
1401 kvm_x86_ops->set_efer(vcpu, efer);
1403 /* Update reserved bits */
1404 if ((efer ^ old_efer) & EFER_NX)
1405 kvm_mmu_reset_context(vcpu);
1410 void kvm_enable_efer_bits(u64 mask)
1412 efer_reserved_bits &= ~mask;
1414 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1417 * Write @data into the MSR specified by @index. Select MSR specific fault
1418 * checks are bypassed if @host_initiated is %true.
1419 * Returns 0 on success, non-0 otherwise.
1420 * Assumes vcpu_load() was already called.
1422 static int __kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data,
1423 bool host_initiated)
1425 struct msr_data msr;
1430 case MSR_KERNEL_GS_BASE:
1433 if (is_noncanonical_address(data, vcpu))
1436 case MSR_IA32_SYSENTER_EIP:
1437 case MSR_IA32_SYSENTER_ESP:
1439 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1440 * non-canonical address is written on Intel but not on
1441 * AMD (which ignores the top 32-bits, because it does
1442 * not implement 64-bit SYSENTER).
1444 * 64-bit code should hence be able to write a non-canonical
1445 * value on AMD. Making the address canonical ensures that
1446 * vmentry does not fail on Intel after writing a non-canonical
1447 * value, and that something deterministic happens if the guest
1448 * invokes 64-bit SYSENTER.
1450 data = get_canonical(data, vcpu_virt_addr_bits(vcpu));
1455 msr.host_initiated = host_initiated;
1457 return kvm_x86_ops->set_msr(vcpu, &msr);
1461 * Read the MSR specified by @index into @data. Select MSR specific fault
1462 * checks are bypassed if @host_initiated is %true.
1463 * Returns 0 on success, non-0 otherwise.
1464 * Assumes vcpu_load() was already called.
1466 static int __kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data,
1467 bool host_initiated)
1469 struct msr_data msr;
1473 msr.host_initiated = host_initiated;
1475 ret = kvm_x86_ops->get_msr(vcpu, &msr);
1481 int kvm_get_msr(struct kvm_vcpu *vcpu, u32 index, u64 *data)
1483 return __kvm_get_msr(vcpu, index, data, false);
1485 EXPORT_SYMBOL_GPL(kvm_get_msr);
1487 int kvm_set_msr(struct kvm_vcpu *vcpu, u32 index, u64 data)
1489 return __kvm_set_msr(vcpu, index, data, false);
1491 EXPORT_SYMBOL_GPL(kvm_set_msr);
1493 int kvm_emulate_rdmsr(struct kvm_vcpu *vcpu)
1495 u32 ecx = kvm_rcx_read(vcpu);
1498 if (kvm_get_msr(vcpu, ecx, &data)) {
1499 trace_kvm_msr_read_ex(ecx);
1500 kvm_inject_gp(vcpu, 0);
1504 trace_kvm_msr_read(ecx, data);
1506 kvm_rax_write(vcpu, data & -1u);
1507 kvm_rdx_write(vcpu, (data >> 32) & -1u);
1508 return kvm_skip_emulated_instruction(vcpu);
1510 EXPORT_SYMBOL_GPL(kvm_emulate_rdmsr);
1512 int kvm_emulate_wrmsr(struct kvm_vcpu *vcpu)
1514 u32 ecx = kvm_rcx_read(vcpu);
1515 u64 data = kvm_read_edx_eax(vcpu);
1517 if (kvm_set_msr(vcpu, ecx, data)) {
1518 trace_kvm_msr_write_ex(ecx, data);
1519 kvm_inject_gp(vcpu, 0);
1523 trace_kvm_msr_write(ecx, data);
1524 return kvm_skip_emulated_instruction(vcpu);
1526 EXPORT_SYMBOL_GPL(kvm_emulate_wrmsr);
1529 * Adapt set_msr() to msr_io()'s calling convention
1531 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1533 return __kvm_get_msr(vcpu, index, data, true);
1536 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1538 return __kvm_set_msr(vcpu, index, *data, true);
1541 #ifdef CONFIG_X86_64
1542 struct pvclock_gtod_data {
1545 struct { /* extract of a clocksource struct */
1558 static struct pvclock_gtod_data pvclock_gtod_data;
1560 static void update_pvclock_gtod(struct timekeeper *tk)
1562 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1565 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1567 write_seqcount_begin(&vdata->seq);
1569 /* copy pvclock gtod data */
1570 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1571 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1572 vdata->clock.mask = tk->tkr_mono.mask;
1573 vdata->clock.mult = tk->tkr_mono.mult;
1574 vdata->clock.shift = tk->tkr_mono.shift;
1576 vdata->boot_ns = boot_ns;
1577 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1579 vdata->wall_time_sec = tk->xtime_sec;
1581 write_seqcount_end(&vdata->seq);
1585 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1587 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1588 kvm_vcpu_kick(vcpu);
1591 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1595 struct pvclock_wall_clock wc;
1596 struct timespec64 boot;
1601 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1606 ++version; /* first time write, random junk */
1610 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1614 * The guest calculates current wall clock time by adding
1615 * system time (updated by kvm_guest_time_update below) to the
1616 * wall clock specified here. guest system time equals host
1617 * system time for us, thus we must fill in host boot time here.
1619 getboottime64(&boot);
1621 if (kvm->arch.kvmclock_offset) {
1622 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1623 boot = timespec64_sub(boot, ts);
1625 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1626 wc.nsec = boot.tv_nsec;
1627 wc.version = version;
1629 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1632 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1635 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1637 do_shl32_div32(dividend, divisor);
1641 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1642 s8 *pshift, u32 *pmultiplier)
1650 scaled64 = scaled_hz;
1651 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1656 tps32 = (uint32_t)tps64;
1657 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1658 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1666 *pmultiplier = div_frac(scaled64, tps32);
1669 #ifdef CONFIG_X86_64
1670 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1673 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1674 static unsigned long max_tsc_khz;
1676 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1678 u64 v = (u64)khz * (1000000 + ppm);
1683 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1687 /* Guest TSC same frequency as host TSC? */
1689 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1693 /* TSC scaling supported? */
1694 if (!kvm_has_tsc_control) {
1695 if (user_tsc_khz > tsc_khz) {
1696 vcpu->arch.tsc_catchup = 1;
1697 vcpu->arch.tsc_always_catchup = 1;
1700 pr_warn_ratelimited("user requested TSC rate below hardware speed\n");
1705 /* TSC scaling required - calculate ratio */
1706 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1707 user_tsc_khz, tsc_khz);
1709 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1710 pr_warn_ratelimited("Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1715 vcpu->arch.tsc_scaling_ratio = ratio;
1719 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1721 u32 thresh_lo, thresh_hi;
1722 int use_scaling = 0;
1724 /* tsc_khz can be zero if TSC calibration fails */
1725 if (user_tsc_khz == 0) {
1726 /* set tsc_scaling_ratio to a safe value */
1727 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1731 /* Compute a scale to convert nanoseconds in TSC cycles */
1732 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1733 &vcpu->arch.virtual_tsc_shift,
1734 &vcpu->arch.virtual_tsc_mult);
1735 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1738 * Compute the variation in TSC rate which is acceptable
1739 * within the range of tolerance and decide if the
1740 * rate being applied is within that bounds of the hardware
1741 * rate. If so, no scaling or compensation need be done.
1743 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1744 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1745 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1746 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1749 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1752 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1754 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1755 vcpu->arch.virtual_tsc_mult,
1756 vcpu->arch.virtual_tsc_shift);
1757 tsc += vcpu->arch.this_tsc_write;
1761 static inline int gtod_is_based_on_tsc(int mode)
1763 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1766 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1768 #ifdef CONFIG_X86_64
1770 struct kvm_arch *ka = &vcpu->kvm->arch;
1771 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1773 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1774 atomic_read(&vcpu->kvm->online_vcpus));
1777 * Once the masterclock is enabled, always perform request in
1778 * order to update it.
1780 * In order to enable masterclock, the host clocksource must be TSC
1781 * and the vcpus need to have matched TSCs. When that happens,
1782 * perform request to enable masterclock.
1784 if (ka->use_master_clock ||
1785 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1786 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1788 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1789 atomic_read(&vcpu->kvm->online_vcpus),
1790 ka->use_master_clock, gtod->clock.vclock_mode);
1794 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1796 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1797 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1801 * Multiply tsc by a fixed point number represented by ratio.
1803 * The most significant 64-N bits (mult) of ratio represent the
1804 * integral part of the fixed point number; the remaining N bits
1805 * (frac) represent the fractional part, ie. ratio represents a fixed
1806 * point number (mult + frac * 2^(-N)).
1808 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1810 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1812 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1815 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1818 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1820 if (ratio != kvm_default_tsc_scaling_ratio)
1821 _tsc = __scale_tsc(ratio, tsc);
1825 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1827 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1831 tsc = kvm_scale_tsc(vcpu, rdtsc());
1833 return target_tsc - tsc;
1836 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1838 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1840 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1842 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1844 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1846 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1849 static inline bool kvm_check_tsc_unstable(void)
1851 #ifdef CONFIG_X86_64
1853 * TSC is marked unstable when we're running on Hyper-V,
1854 * 'TSC page' clocksource is good.
1856 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1859 return check_tsc_unstable();
1862 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1864 struct kvm *kvm = vcpu->kvm;
1865 u64 offset, ns, elapsed;
1866 unsigned long flags;
1868 bool already_matched;
1869 u64 data = msr->data;
1870 bool synchronizing = false;
1872 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1873 offset = kvm_compute_tsc_offset(vcpu, data);
1874 ns = ktime_get_boottime_ns();
1875 elapsed = ns - kvm->arch.last_tsc_nsec;
1877 if (vcpu->arch.virtual_tsc_khz) {
1878 if (data == 0 && msr->host_initiated) {
1880 * detection of vcpu initialization -- need to sync
1881 * with other vCPUs. This particularly helps to keep
1882 * kvm_clock stable after CPU hotplug
1884 synchronizing = true;
1886 u64 tsc_exp = kvm->arch.last_tsc_write +
1887 nsec_to_cycles(vcpu, elapsed);
1888 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1890 * Special case: TSC write with a small delta (1 second)
1891 * of virtual cycle time against real time is
1892 * interpreted as an attempt to synchronize the CPU.
1894 synchronizing = data < tsc_exp + tsc_hz &&
1895 data + tsc_hz > tsc_exp;
1900 * For a reliable TSC, we can match TSC offsets, and for an unstable
1901 * TSC, we add elapsed time in this computation. We could let the
1902 * compensation code attempt to catch up if we fall behind, but
1903 * it's better to try to match offsets from the beginning.
1905 if (synchronizing &&
1906 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1907 if (!kvm_check_tsc_unstable()) {
1908 offset = kvm->arch.cur_tsc_offset;
1910 u64 delta = nsec_to_cycles(vcpu, elapsed);
1912 offset = kvm_compute_tsc_offset(vcpu, data);
1915 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1918 * We split periods of matched TSC writes into generations.
1919 * For each generation, we track the original measured
1920 * nanosecond time, offset, and write, so if TSCs are in
1921 * sync, we can match exact offset, and if not, we can match
1922 * exact software computation in compute_guest_tsc()
1924 * These values are tracked in kvm->arch.cur_xxx variables.
1926 kvm->arch.cur_tsc_generation++;
1927 kvm->arch.cur_tsc_nsec = ns;
1928 kvm->arch.cur_tsc_write = data;
1929 kvm->arch.cur_tsc_offset = offset;
1934 * We also track th most recent recorded KHZ, write and time to
1935 * allow the matching interval to be extended at each write.
1937 kvm->arch.last_tsc_nsec = ns;
1938 kvm->arch.last_tsc_write = data;
1939 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1941 vcpu->arch.last_guest_tsc = data;
1943 /* Keep track of which generation this VCPU has synchronized to */
1944 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1945 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1946 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1948 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1949 update_ia32_tsc_adjust_msr(vcpu, offset);
1951 kvm_vcpu_write_tsc_offset(vcpu, offset);
1952 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1954 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1956 kvm->arch.nr_vcpus_matched_tsc = 0;
1957 } else if (!already_matched) {
1958 kvm->arch.nr_vcpus_matched_tsc++;
1961 kvm_track_tsc_matching(vcpu);
1962 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1965 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1967 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1970 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1971 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1974 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1976 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1977 WARN_ON(adjustment < 0);
1978 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1979 adjust_tsc_offset_guest(vcpu, adjustment);
1982 #ifdef CONFIG_X86_64
1984 static u64 read_tsc(void)
1986 u64 ret = (u64)rdtsc_ordered();
1987 u64 last = pvclock_gtod_data.clock.cycle_last;
1989 if (likely(ret >= last))
1993 * GCC likes to generate cmov here, but this branch is extremely
1994 * predictable (it's just a function of time and the likely is
1995 * very likely) and there's a data dependence, so force GCC
1996 * to generate a branch instead. I don't barrier() because
1997 * we don't actually need a barrier, and if this function
1998 * ever gets inlined it will generate worse code.
2004 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
2007 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2010 switch (gtod->clock.vclock_mode) {
2011 case VCLOCK_HVCLOCK:
2012 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
2014 if (tsc_pg_val != U64_MAX) {
2015 /* TSC page valid */
2016 *mode = VCLOCK_HVCLOCK;
2017 v = (tsc_pg_val - gtod->clock.cycle_last) &
2020 /* TSC page invalid */
2021 *mode = VCLOCK_NONE;
2026 *tsc_timestamp = read_tsc();
2027 v = (*tsc_timestamp - gtod->clock.cycle_last) &
2031 *mode = VCLOCK_NONE;
2034 if (*mode == VCLOCK_NONE)
2035 *tsc_timestamp = v = 0;
2037 return v * gtod->clock.mult;
2040 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
2042 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2048 seq = read_seqcount_begin(>od->seq);
2049 ns = gtod->nsec_base;
2050 ns += vgettsc(tsc_timestamp, &mode);
2051 ns >>= gtod->clock.shift;
2052 ns += gtod->boot_ns;
2053 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2059 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
2061 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
2067 seq = read_seqcount_begin(>od->seq);
2068 ts->tv_sec = gtod->wall_time_sec;
2069 ns = gtod->nsec_base;
2070 ns += vgettsc(tsc_timestamp, &mode);
2071 ns >>= gtod->clock.shift;
2072 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
2074 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
2080 /* returns true if host is using TSC based clocksource */
2081 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
2083 /* checked again under seqlock below */
2084 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2087 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
2091 /* returns true if host is using TSC based clocksource */
2092 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
2095 /* checked again under seqlock below */
2096 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
2099 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
2105 * Assuming a stable TSC across physical CPUS, and a stable TSC
2106 * across virtual CPUs, the following condition is possible.
2107 * Each numbered line represents an event visible to both
2108 * CPUs at the next numbered event.
2110 * "timespecX" represents host monotonic time. "tscX" represents
2113 * VCPU0 on CPU0 | VCPU1 on CPU1
2115 * 1. read timespec0,tsc0
2116 * 2. | timespec1 = timespec0 + N
2118 * 3. transition to guest | transition to guest
2119 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
2120 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
2121 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
2123 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
2126 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
2128 * - 0 < N - M => M < N
2130 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
2131 * always the case (the difference between two distinct xtime instances
2132 * might be smaller then the difference between corresponding TSC reads,
2133 * when updating guest vcpus pvclock areas).
2135 * To avoid that problem, do not allow visibility of distinct
2136 * system_timestamp/tsc_timestamp values simultaneously: use a master
2137 * copy of host monotonic time values. Update that master copy
2140 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2144 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2146 #ifdef CONFIG_X86_64
2147 struct kvm_arch *ka = &kvm->arch;
2149 bool host_tsc_clocksource, vcpus_matched;
2151 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2152 atomic_read(&kvm->online_vcpus));
2155 * If the host uses TSC clock, then passthrough TSC as stable
2158 host_tsc_clocksource = kvm_get_time_and_clockread(
2159 &ka->master_kernel_ns,
2160 &ka->master_cycle_now);
2162 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2163 && !ka->backwards_tsc_observed
2164 && !ka->boot_vcpu_runs_old_kvmclock;
2166 if (ka->use_master_clock)
2167 atomic_set(&kvm_guest_has_master_clock, 1);
2169 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2170 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2175 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2177 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2180 static void kvm_gen_update_masterclock(struct kvm *kvm)
2182 #ifdef CONFIG_X86_64
2184 struct kvm_vcpu *vcpu;
2185 struct kvm_arch *ka = &kvm->arch;
2187 spin_lock(&ka->pvclock_gtod_sync_lock);
2188 kvm_make_mclock_inprogress_request(kvm);
2189 /* no guest entries from this point */
2190 pvclock_update_vm_gtod_copy(kvm);
2192 kvm_for_each_vcpu(i, vcpu, kvm)
2193 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2195 /* guest entries allowed */
2196 kvm_for_each_vcpu(i, vcpu, kvm)
2197 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2199 spin_unlock(&ka->pvclock_gtod_sync_lock);
2203 u64 get_kvmclock_ns(struct kvm *kvm)
2205 struct kvm_arch *ka = &kvm->arch;
2206 struct pvclock_vcpu_time_info hv_clock;
2209 spin_lock(&ka->pvclock_gtod_sync_lock);
2210 if (!ka->use_master_clock) {
2211 spin_unlock(&ka->pvclock_gtod_sync_lock);
2212 return ktime_get_boottime_ns() + ka->kvmclock_offset;
2215 hv_clock.tsc_timestamp = ka->master_cycle_now;
2216 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2217 spin_unlock(&ka->pvclock_gtod_sync_lock);
2219 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2222 if (__this_cpu_read(cpu_tsc_khz)) {
2223 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2224 &hv_clock.tsc_shift,
2225 &hv_clock.tsc_to_system_mul);
2226 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2228 ret = ktime_get_boottime_ns() + ka->kvmclock_offset;
2235 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2237 struct kvm_vcpu_arch *vcpu = &v->arch;
2238 struct pvclock_vcpu_time_info guest_hv_clock;
2240 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2241 &guest_hv_clock, sizeof(guest_hv_clock))))
2244 /* This VCPU is paused, but it's legal for a guest to read another
2245 * VCPU's kvmclock, so we really have to follow the specification where
2246 * it says that version is odd if data is being modified, and even after
2249 * Version field updates must be kept separate. This is because
2250 * kvm_write_guest_cached might use a "rep movs" instruction, and
2251 * writes within a string instruction are weakly ordered. So there
2252 * are three writes overall.
2254 * As a small optimization, only write the version field in the first
2255 * and third write. The vcpu->pv_time cache is still valid, because the
2256 * version field is the first in the struct.
2258 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2260 if (guest_hv_clock.version & 1)
2261 ++guest_hv_clock.version; /* first time write, random junk */
2263 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2264 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2266 sizeof(vcpu->hv_clock.version));
2270 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2271 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2273 if (vcpu->pvclock_set_guest_stopped_request) {
2274 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2275 vcpu->pvclock_set_guest_stopped_request = false;
2278 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2280 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2282 sizeof(vcpu->hv_clock));
2286 vcpu->hv_clock.version++;
2287 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2289 sizeof(vcpu->hv_clock.version));
2292 static int kvm_guest_time_update(struct kvm_vcpu *v)
2294 unsigned long flags, tgt_tsc_khz;
2295 struct kvm_vcpu_arch *vcpu = &v->arch;
2296 struct kvm_arch *ka = &v->kvm->arch;
2298 u64 tsc_timestamp, host_tsc;
2300 bool use_master_clock;
2306 * If the host uses TSC clock, then passthrough TSC as stable
2309 spin_lock(&ka->pvclock_gtod_sync_lock);
2310 use_master_clock = ka->use_master_clock;
2311 if (use_master_clock) {
2312 host_tsc = ka->master_cycle_now;
2313 kernel_ns = ka->master_kernel_ns;
2315 spin_unlock(&ka->pvclock_gtod_sync_lock);
2317 /* Keep irq disabled to prevent changes to the clock */
2318 local_irq_save(flags);
2319 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2320 if (unlikely(tgt_tsc_khz == 0)) {
2321 local_irq_restore(flags);
2322 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2325 if (!use_master_clock) {
2327 kernel_ns = ktime_get_boottime_ns();
2330 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2333 * We may have to catch up the TSC to match elapsed wall clock
2334 * time for two reasons, even if kvmclock is used.
2335 * 1) CPU could have been running below the maximum TSC rate
2336 * 2) Broken TSC compensation resets the base at each VCPU
2337 * entry to avoid unknown leaps of TSC even when running
2338 * again on the same CPU. This may cause apparent elapsed
2339 * time to disappear, and the guest to stand still or run
2342 if (vcpu->tsc_catchup) {
2343 u64 tsc = compute_guest_tsc(v, kernel_ns);
2344 if (tsc > tsc_timestamp) {
2345 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2346 tsc_timestamp = tsc;
2350 local_irq_restore(flags);
2352 /* With all the info we got, fill in the values */
2354 if (kvm_has_tsc_control)
2355 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2357 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2358 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2359 &vcpu->hv_clock.tsc_shift,
2360 &vcpu->hv_clock.tsc_to_system_mul);
2361 vcpu->hw_tsc_khz = tgt_tsc_khz;
2364 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2365 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2366 vcpu->last_guest_tsc = tsc_timestamp;
2368 /* If the host uses TSC clocksource, then it is stable */
2370 if (use_master_clock)
2371 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2373 vcpu->hv_clock.flags = pvclock_flags;
2375 if (vcpu->pv_time_enabled)
2376 kvm_setup_pvclock_page(v);
2377 if (v == kvm_get_vcpu(v->kvm, 0))
2378 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2383 * kvmclock updates which are isolated to a given vcpu, such as
2384 * vcpu->cpu migration, should not allow system_timestamp from
2385 * the rest of the vcpus to remain static. Otherwise ntp frequency
2386 * correction applies to one vcpu's system_timestamp but not
2389 * So in those cases, request a kvmclock update for all vcpus.
2390 * We need to rate-limit these requests though, as they can
2391 * considerably slow guests that have a large number of vcpus.
2392 * The time for a remote vcpu to update its kvmclock is bound
2393 * by the delay we use to rate-limit the updates.
2396 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2398 static void kvmclock_update_fn(struct work_struct *work)
2401 struct delayed_work *dwork = to_delayed_work(work);
2402 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2403 kvmclock_update_work);
2404 struct kvm *kvm = container_of(ka, struct kvm, arch);
2405 struct kvm_vcpu *vcpu;
2407 kvm_for_each_vcpu(i, vcpu, kvm) {
2408 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2409 kvm_vcpu_kick(vcpu);
2413 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2415 struct kvm *kvm = v->kvm;
2417 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2418 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2419 KVMCLOCK_UPDATE_DELAY);
2422 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2424 static void kvmclock_sync_fn(struct work_struct *work)
2426 struct delayed_work *dwork = to_delayed_work(work);
2427 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2428 kvmclock_sync_work);
2429 struct kvm *kvm = container_of(ka, struct kvm, arch);
2431 if (!kvmclock_periodic_sync)
2434 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2435 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2436 KVMCLOCK_SYNC_PERIOD);
2440 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2442 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2444 /* McStatusWrEn enabled? */
2445 if (guest_cpuid_is_amd(vcpu))
2446 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2451 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2453 u64 mcg_cap = vcpu->arch.mcg_cap;
2454 unsigned bank_num = mcg_cap & 0xff;
2455 u32 msr = msr_info->index;
2456 u64 data = msr_info->data;
2459 case MSR_IA32_MCG_STATUS:
2460 vcpu->arch.mcg_status = data;
2462 case MSR_IA32_MCG_CTL:
2463 if (!(mcg_cap & MCG_CTL_P) &&
2464 (data || !msr_info->host_initiated))
2466 if (data != 0 && data != ~(u64)0)
2468 vcpu->arch.mcg_ctl = data;
2471 if (msr >= MSR_IA32_MC0_CTL &&
2472 msr < MSR_IA32_MCx_CTL(bank_num)) {
2473 u32 offset = msr - MSR_IA32_MC0_CTL;
2474 /* only 0 or all 1s can be written to IA32_MCi_CTL
2475 * some Linux kernels though clear bit 10 in bank 4 to
2476 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2477 * this to avoid an uncatched #GP in the guest
2479 if ((offset & 0x3) == 0 &&
2480 data != 0 && (data | (1 << 10)) != ~(u64)0)
2484 if (!msr_info->host_initiated &&
2485 (offset & 0x3) == 1 && data != 0) {
2486 if (!can_set_mci_status(vcpu))
2490 vcpu->arch.mce_banks[offset] = data;
2498 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2500 struct kvm *kvm = vcpu->kvm;
2501 int lm = is_long_mode(vcpu);
2502 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2503 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2504 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2505 : kvm->arch.xen_hvm_config.blob_size_32;
2506 u32 page_num = data & ~PAGE_MASK;
2507 u64 page_addr = data & PAGE_MASK;
2512 if (page_num >= blob_size)
2515 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2520 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2529 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2531 gpa_t gpa = data & ~0x3f;
2533 /* Bits 3:5 are reserved, Should be zero */
2537 vcpu->arch.apf.msr_val = data;
2539 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2540 kvm_clear_async_pf_completion_queue(vcpu);
2541 kvm_async_pf_hash_reset(vcpu);
2545 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2549 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2550 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2551 kvm_async_pf_wakeup_all(vcpu);
2555 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2557 vcpu->arch.pv_time_enabled = false;
2558 vcpu->arch.time = 0;
2561 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2563 ++vcpu->stat.tlb_flush;
2564 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2567 static void record_steal_time(struct kvm_vcpu *vcpu)
2569 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2572 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2573 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2577 * Doing a TLB flush here, on the guest's behalf, can avoid
2580 trace_kvm_pv_tlb_flush(vcpu->vcpu_id,
2581 vcpu->arch.st.steal.preempted & KVM_VCPU_FLUSH_TLB);
2582 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2583 kvm_vcpu_flush_tlb(vcpu, false);
2585 if (vcpu->arch.st.steal.version & 1)
2586 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2588 vcpu->arch.st.steal.version += 1;
2590 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2591 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2595 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2596 vcpu->arch.st.last_steal;
2597 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2599 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2600 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2604 vcpu->arch.st.steal.version += 1;
2606 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2607 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2610 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2613 u32 msr = msr_info->index;
2614 u64 data = msr_info->data;
2617 case MSR_AMD64_NB_CFG:
2618 case MSR_IA32_UCODE_WRITE:
2619 case MSR_VM_HSAVE_PA:
2620 case MSR_AMD64_PATCH_LOADER:
2621 case MSR_AMD64_BU_CFG2:
2622 case MSR_AMD64_DC_CFG:
2623 case MSR_F15H_EX_CFG:
2626 case MSR_IA32_UCODE_REV:
2627 if (msr_info->host_initiated)
2628 vcpu->arch.microcode_version = data;
2630 case MSR_IA32_ARCH_CAPABILITIES:
2631 if (!msr_info->host_initiated)
2633 vcpu->arch.arch_capabilities = data;
2636 return set_efer(vcpu, msr_info);
2638 data &= ~(u64)0x40; /* ignore flush filter disable */
2639 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2640 data &= ~(u64)0x8; /* ignore TLB cache disable */
2642 /* Handle McStatusWrEn */
2643 if (data == BIT_ULL(18)) {
2644 vcpu->arch.msr_hwcr = data;
2645 } else if (data != 0) {
2646 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2651 case MSR_FAM10H_MMIO_CONF_BASE:
2653 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2658 case MSR_IA32_DEBUGCTLMSR:
2660 /* We support the non-activated case already */
2662 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2663 /* Values other than LBR and BTF are vendor-specific,
2664 thus reserved and should throw a #GP */
2667 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2670 case 0x200 ... 0x2ff:
2671 return kvm_mtrr_set_msr(vcpu, msr, data);
2672 case MSR_IA32_APICBASE:
2673 return kvm_set_apic_base(vcpu, msr_info);
2674 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2675 return kvm_x2apic_msr_write(vcpu, msr, data);
2676 case MSR_IA32_TSCDEADLINE:
2677 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2679 case MSR_IA32_TSC_ADJUST:
2680 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2681 if (!msr_info->host_initiated) {
2682 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2683 adjust_tsc_offset_guest(vcpu, adj);
2685 vcpu->arch.ia32_tsc_adjust_msr = data;
2688 case MSR_IA32_MISC_ENABLE:
2689 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2690 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2691 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2693 vcpu->arch.ia32_misc_enable_msr = data;
2694 kvm_update_cpuid(vcpu);
2696 vcpu->arch.ia32_misc_enable_msr = data;
2699 case MSR_IA32_SMBASE:
2700 if (!msr_info->host_initiated)
2702 vcpu->arch.smbase = data;
2704 case MSR_IA32_POWER_CTL:
2705 vcpu->arch.msr_ia32_power_ctl = data;
2708 kvm_write_tsc(vcpu, msr_info);
2711 if (!msr_info->host_initiated)
2713 vcpu->arch.smi_count = data;
2715 case MSR_KVM_WALL_CLOCK_NEW:
2716 case MSR_KVM_WALL_CLOCK:
2717 vcpu->kvm->arch.wall_clock = data;
2718 kvm_write_wall_clock(vcpu->kvm, data);
2720 case MSR_KVM_SYSTEM_TIME_NEW:
2721 case MSR_KVM_SYSTEM_TIME: {
2722 struct kvm_arch *ka = &vcpu->kvm->arch;
2724 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2725 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2727 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2728 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2730 ka->boot_vcpu_runs_old_kvmclock = tmp;
2733 vcpu->arch.time = data;
2734 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2736 /* we verify if the enable bit is set... */
2737 vcpu->arch.pv_time_enabled = false;
2741 if (!kvm_gfn_to_hva_cache_init(vcpu->kvm,
2742 &vcpu->arch.pv_time, data & ~1ULL,
2743 sizeof(struct pvclock_vcpu_time_info)))
2744 vcpu->arch.pv_time_enabled = true;
2748 case MSR_KVM_ASYNC_PF_EN:
2749 if (kvm_pv_enable_async_pf(vcpu, data))
2752 case MSR_KVM_STEAL_TIME:
2754 if (unlikely(!sched_info_on()))
2757 if (data & KVM_STEAL_RESERVED_MASK)
2760 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2761 data & KVM_STEAL_VALID_BITS,
2762 sizeof(struct kvm_steal_time)))
2765 vcpu->arch.st.msr_val = data;
2767 if (!(data & KVM_MSR_ENABLED))
2770 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2773 case MSR_KVM_PV_EOI_EN:
2774 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2778 case MSR_KVM_POLL_CONTROL:
2779 /* only enable bit supported */
2780 if (data & (-1ULL << 1))
2783 vcpu->arch.msr_kvm_poll_control = data;
2786 case MSR_IA32_MCG_CTL:
2787 case MSR_IA32_MCG_STATUS:
2788 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2789 return set_msr_mce(vcpu, msr_info);
2791 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2792 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2793 pr = true; /* fall through */
2794 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2795 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2796 if (kvm_pmu_is_valid_msr(vcpu, msr))
2797 return kvm_pmu_set_msr(vcpu, msr_info);
2799 if (pr || data != 0)
2800 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2801 "0x%x data 0x%llx\n", msr, data);
2803 case MSR_K7_CLK_CTL:
2805 * Ignore all writes to this no longer documented MSR.
2806 * Writes are only relevant for old K7 processors,
2807 * all pre-dating SVM, but a recommended workaround from
2808 * AMD for these chips. It is possible to specify the
2809 * affected processor models on the command line, hence
2810 * the need to ignore the workaround.
2813 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2814 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2815 case HV_X64_MSR_CRASH_CTL:
2816 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2817 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2818 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2819 case HV_X64_MSR_TSC_EMULATION_STATUS:
2820 return kvm_hv_set_msr_common(vcpu, msr, data,
2821 msr_info->host_initiated);
2822 case MSR_IA32_BBL_CR_CTL3:
2823 /* Drop writes to this legacy MSR -- see rdmsr
2824 * counterpart for further detail.
2826 if (report_ignored_msrs)
2827 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2830 case MSR_AMD64_OSVW_ID_LENGTH:
2831 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2833 vcpu->arch.osvw.length = data;
2835 case MSR_AMD64_OSVW_STATUS:
2836 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2838 vcpu->arch.osvw.status = data;
2840 case MSR_PLATFORM_INFO:
2841 if (!msr_info->host_initiated ||
2842 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2843 cpuid_fault_enabled(vcpu)))
2845 vcpu->arch.msr_platform_info = data;
2847 case MSR_MISC_FEATURES_ENABLES:
2848 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2849 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2850 !supports_cpuid_fault(vcpu)))
2852 vcpu->arch.msr_misc_features_enables = data;
2855 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2856 return xen_hvm_config(vcpu, data);
2857 if (kvm_pmu_is_valid_msr(vcpu, msr))
2858 return kvm_pmu_set_msr(vcpu, msr_info);
2860 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2864 if (report_ignored_msrs)
2866 "ignored wrmsr: 0x%x data 0x%llx\n",
2873 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2875 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2878 u64 mcg_cap = vcpu->arch.mcg_cap;
2879 unsigned bank_num = mcg_cap & 0xff;
2882 case MSR_IA32_P5_MC_ADDR:
2883 case MSR_IA32_P5_MC_TYPE:
2886 case MSR_IA32_MCG_CAP:
2887 data = vcpu->arch.mcg_cap;
2889 case MSR_IA32_MCG_CTL:
2890 if (!(mcg_cap & MCG_CTL_P) && !host)
2892 data = vcpu->arch.mcg_ctl;
2894 case MSR_IA32_MCG_STATUS:
2895 data = vcpu->arch.mcg_status;
2898 if (msr >= MSR_IA32_MC0_CTL &&
2899 msr < MSR_IA32_MCx_CTL(bank_num)) {
2900 u32 offset = msr - MSR_IA32_MC0_CTL;
2901 data = vcpu->arch.mce_banks[offset];
2910 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2912 switch (msr_info->index) {
2913 case MSR_IA32_PLATFORM_ID:
2914 case MSR_IA32_EBL_CR_POWERON:
2915 case MSR_IA32_DEBUGCTLMSR:
2916 case MSR_IA32_LASTBRANCHFROMIP:
2917 case MSR_IA32_LASTBRANCHTOIP:
2918 case MSR_IA32_LASTINTFROMIP:
2919 case MSR_IA32_LASTINTTOIP:
2921 case MSR_K8_TSEG_ADDR:
2922 case MSR_K8_TSEG_MASK:
2923 case MSR_VM_HSAVE_PA:
2924 case MSR_K8_INT_PENDING_MSG:
2925 case MSR_AMD64_NB_CFG:
2926 case MSR_FAM10H_MMIO_CONF_BASE:
2927 case MSR_AMD64_BU_CFG2:
2928 case MSR_IA32_PERF_CTL:
2929 case MSR_AMD64_DC_CFG:
2930 case MSR_F15H_EX_CFG:
2933 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2934 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2935 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2936 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2937 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2938 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2939 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2942 case MSR_IA32_UCODE_REV:
2943 msr_info->data = vcpu->arch.microcode_version;
2945 case MSR_IA32_ARCH_CAPABILITIES:
2946 if (!msr_info->host_initiated &&
2947 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2949 msr_info->data = vcpu->arch.arch_capabilities;
2951 case MSR_IA32_POWER_CTL:
2952 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2955 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2958 case 0x200 ... 0x2ff:
2959 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2960 case 0xcd: /* fsb frequency */
2964 * MSR_EBC_FREQUENCY_ID
2965 * Conservative value valid for even the basic CPU models.
2966 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2967 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2968 * and 266MHz for model 3, or 4. Set Core Clock
2969 * Frequency to System Bus Frequency Ratio to 1 (bits
2970 * 31:24) even though these are only valid for CPU
2971 * models > 2, however guests may end up dividing or
2972 * multiplying by zero otherwise.
2974 case MSR_EBC_FREQUENCY_ID:
2975 msr_info->data = 1 << 24;
2977 case MSR_IA32_APICBASE:
2978 msr_info->data = kvm_get_apic_base(vcpu);
2980 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2981 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2983 case MSR_IA32_TSCDEADLINE:
2984 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2986 case MSR_IA32_TSC_ADJUST:
2987 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2989 case MSR_IA32_MISC_ENABLE:
2990 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2992 case MSR_IA32_SMBASE:
2993 if (!msr_info->host_initiated)
2995 msr_info->data = vcpu->arch.smbase;
2998 msr_info->data = vcpu->arch.smi_count;
3000 case MSR_IA32_PERF_STATUS:
3001 /* TSC increment by tick */
3002 msr_info->data = 1000ULL;
3003 /* CPU multiplier */
3004 msr_info->data |= (((uint64_t)4ULL) << 40);
3007 msr_info->data = vcpu->arch.efer;
3009 case MSR_KVM_WALL_CLOCK:
3010 case MSR_KVM_WALL_CLOCK_NEW:
3011 msr_info->data = vcpu->kvm->arch.wall_clock;
3013 case MSR_KVM_SYSTEM_TIME:
3014 case MSR_KVM_SYSTEM_TIME_NEW:
3015 msr_info->data = vcpu->arch.time;
3017 case MSR_KVM_ASYNC_PF_EN:
3018 msr_info->data = vcpu->arch.apf.msr_val;
3020 case MSR_KVM_STEAL_TIME:
3021 msr_info->data = vcpu->arch.st.msr_val;
3023 case MSR_KVM_PV_EOI_EN:
3024 msr_info->data = vcpu->arch.pv_eoi.msr_val;
3026 case MSR_KVM_POLL_CONTROL:
3027 msr_info->data = vcpu->arch.msr_kvm_poll_control;
3029 case MSR_IA32_P5_MC_ADDR:
3030 case MSR_IA32_P5_MC_TYPE:
3031 case MSR_IA32_MCG_CAP:
3032 case MSR_IA32_MCG_CTL:
3033 case MSR_IA32_MCG_STATUS:
3034 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
3035 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
3036 msr_info->host_initiated);
3037 case MSR_K7_CLK_CTL:
3039 * Provide expected ramp-up count for K7. All other
3040 * are set to zero, indicating minimum divisors for
3043 * This prevents guest kernels on AMD host with CPU
3044 * type 6, model 8 and higher from exploding due to
3045 * the rdmsr failing.
3047 msr_info->data = 0x20000000;
3049 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
3050 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
3051 case HV_X64_MSR_CRASH_CTL:
3052 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
3053 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
3054 case HV_X64_MSR_TSC_EMULATION_CONTROL:
3055 case HV_X64_MSR_TSC_EMULATION_STATUS:
3056 return kvm_hv_get_msr_common(vcpu,
3057 msr_info->index, &msr_info->data,
3058 msr_info->host_initiated);
3060 case MSR_IA32_BBL_CR_CTL3:
3061 /* This legacy MSR exists but isn't fully documented in current
3062 * silicon. It is however accessed by winxp in very narrow
3063 * scenarios where it sets bit #19, itself documented as
3064 * a "reserved" bit. Best effort attempt to source coherent
3065 * read data here should the balance of the register be
3066 * interpreted by the guest:
3068 * L2 cache control register 3: 64GB range, 256KB size,
3069 * enabled, latency 0x1, configured
3071 msr_info->data = 0xbe702111;
3073 case MSR_AMD64_OSVW_ID_LENGTH:
3074 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3076 msr_info->data = vcpu->arch.osvw.length;
3078 case MSR_AMD64_OSVW_STATUS:
3079 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
3081 msr_info->data = vcpu->arch.osvw.status;
3083 case MSR_PLATFORM_INFO:
3084 if (!msr_info->host_initiated &&
3085 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
3087 msr_info->data = vcpu->arch.msr_platform_info;
3089 case MSR_MISC_FEATURES_ENABLES:
3090 msr_info->data = vcpu->arch.msr_misc_features_enables;
3093 msr_info->data = vcpu->arch.msr_hwcr;
3096 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
3097 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
3099 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
3103 if (report_ignored_msrs)
3104 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
3112 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
3115 * Read or write a bunch of msrs. All parameters are kernel addresses.
3117 * @return number of msrs set successfully.
3119 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
3120 struct kvm_msr_entry *entries,
3121 int (*do_msr)(struct kvm_vcpu *vcpu,
3122 unsigned index, u64 *data))
3126 for (i = 0; i < msrs->nmsrs; ++i)
3127 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3134 * Read or write a bunch of msrs. Parameters are user addresses.
3136 * @return number of msrs set successfully.
3138 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3139 int (*do_msr)(struct kvm_vcpu *vcpu,
3140 unsigned index, u64 *data),
3143 struct kvm_msrs msrs;
3144 struct kvm_msr_entry *entries;
3149 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3153 if (msrs.nmsrs >= MAX_IO_MSRS)
3156 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3157 entries = memdup_user(user_msrs->entries, size);
3158 if (IS_ERR(entries)) {
3159 r = PTR_ERR(entries);
3163 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3168 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3179 static inline bool kvm_can_mwait_in_guest(void)
3181 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3182 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3183 boot_cpu_has(X86_FEATURE_ARAT);
3186 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3191 case KVM_CAP_IRQCHIP:
3193 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3194 case KVM_CAP_SET_TSS_ADDR:
3195 case KVM_CAP_EXT_CPUID:
3196 case KVM_CAP_EXT_EMUL_CPUID:
3197 case KVM_CAP_CLOCKSOURCE:
3199 case KVM_CAP_NOP_IO_DELAY:
3200 case KVM_CAP_MP_STATE:
3201 case KVM_CAP_SYNC_MMU:
3202 case KVM_CAP_USER_NMI:
3203 case KVM_CAP_REINJECT_CONTROL:
3204 case KVM_CAP_IRQ_INJECT_STATUS:
3205 case KVM_CAP_IOEVENTFD:
3206 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3208 case KVM_CAP_PIT_STATE2:
3209 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3210 case KVM_CAP_XEN_HVM:
3211 case KVM_CAP_VCPU_EVENTS:
3212 case KVM_CAP_HYPERV:
3213 case KVM_CAP_HYPERV_VAPIC:
3214 case KVM_CAP_HYPERV_SPIN:
3215 case KVM_CAP_HYPERV_SYNIC:
3216 case KVM_CAP_HYPERV_SYNIC2:
3217 case KVM_CAP_HYPERV_VP_INDEX:
3218 case KVM_CAP_HYPERV_EVENTFD:
3219 case KVM_CAP_HYPERV_TLBFLUSH:
3220 case KVM_CAP_HYPERV_SEND_IPI:
3221 case KVM_CAP_HYPERV_CPUID:
3222 case KVM_CAP_PCI_SEGMENT:
3223 case KVM_CAP_DEBUGREGS:
3224 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3226 case KVM_CAP_ASYNC_PF:
3227 case KVM_CAP_GET_TSC_KHZ:
3228 case KVM_CAP_KVMCLOCK_CTRL:
3229 case KVM_CAP_READONLY_MEM:
3230 case KVM_CAP_HYPERV_TIME:
3231 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3232 case KVM_CAP_TSC_DEADLINE_TIMER:
3233 case KVM_CAP_DISABLE_QUIRKS:
3234 case KVM_CAP_SET_BOOT_CPU_ID:
3235 case KVM_CAP_SPLIT_IRQCHIP:
3236 case KVM_CAP_IMMEDIATE_EXIT:
3237 case KVM_CAP_PMU_EVENT_FILTER:
3238 case KVM_CAP_GET_MSR_FEATURES:
3239 case KVM_CAP_MSR_PLATFORM_INFO:
3240 case KVM_CAP_EXCEPTION_PAYLOAD:
3243 case KVM_CAP_SYNC_REGS:
3244 r = KVM_SYNC_X86_VALID_FIELDS;
3246 case KVM_CAP_ADJUST_CLOCK:
3247 r = KVM_CLOCK_TSC_STABLE;
3249 case KVM_CAP_X86_DISABLE_EXITS:
3250 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3251 KVM_X86_DISABLE_EXITS_CSTATE;
3252 if(kvm_can_mwait_in_guest())
3253 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3255 case KVM_CAP_X86_SMM:
3256 /* SMBASE is usually relocated above 1M on modern chipsets,
3257 * and SMM handlers might indeed rely on 4G segment limits,
3258 * so do not report SMM to be available if real mode is
3259 * emulated via vm86 mode. Still, do not go to great lengths
3260 * to avoid userspace's usage of the feature, because it is a
3261 * fringe case that is not enabled except via specific settings
3262 * of the module parameters.
3264 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3267 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3269 case KVM_CAP_NR_VCPUS:
3270 r = KVM_SOFT_MAX_VCPUS;
3272 case KVM_CAP_MAX_VCPUS:
3275 case KVM_CAP_MAX_VCPU_ID:
3276 r = KVM_MAX_VCPU_ID;
3278 case KVM_CAP_PV_MMU: /* obsolete */
3282 r = KVM_MAX_MCE_BANKS;
3285 r = boot_cpu_has(X86_FEATURE_XSAVE);
3287 case KVM_CAP_TSC_CONTROL:
3288 r = kvm_has_tsc_control;
3290 case KVM_CAP_X2APIC_API:
3291 r = KVM_X2APIC_API_VALID_FLAGS;
3293 case KVM_CAP_NESTED_STATE:
3294 r = kvm_x86_ops->get_nested_state ?
3295 kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3297 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
3298 r = kvm_x86_ops->enable_direct_tlbflush != NULL;
3300 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3301 r = kvm_x86_ops->nested_enable_evmcs != NULL;
3310 long kvm_arch_dev_ioctl(struct file *filp,
3311 unsigned int ioctl, unsigned long arg)
3313 void __user *argp = (void __user *)arg;
3317 case KVM_GET_MSR_INDEX_LIST: {
3318 struct kvm_msr_list __user *user_msr_list = argp;
3319 struct kvm_msr_list msr_list;
3323 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3326 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3327 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3330 if (n < msr_list.nmsrs)
3333 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3334 num_msrs_to_save * sizeof(u32)))
3336 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3338 num_emulated_msrs * sizeof(u32)))
3343 case KVM_GET_SUPPORTED_CPUID:
3344 case KVM_GET_EMULATED_CPUID: {
3345 struct kvm_cpuid2 __user *cpuid_arg = argp;
3346 struct kvm_cpuid2 cpuid;
3349 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3352 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3358 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3363 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3365 if (copy_to_user(argp, &kvm_mce_cap_supported,
3366 sizeof(kvm_mce_cap_supported)))
3370 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3371 struct kvm_msr_list __user *user_msr_list = argp;
3372 struct kvm_msr_list msr_list;
3376 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3379 msr_list.nmsrs = num_msr_based_features;
3380 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3383 if (n < msr_list.nmsrs)
3386 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3387 num_msr_based_features * sizeof(u32)))
3393 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3403 static void wbinvd_ipi(void *garbage)
3408 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3410 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3413 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3415 /* Address WBINVD may be executed by guest */
3416 if (need_emulate_wbinvd(vcpu)) {
3417 if (kvm_x86_ops->has_wbinvd_exit())
3418 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3419 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3420 smp_call_function_single(vcpu->cpu,
3421 wbinvd_ipi, NULL, 1);
3424 kvm_x86_ops->vcpu_load(vcpu, cpu);
3426 fpregs_assert_state_consistent();
3427 if (test_thread_flag(TIF_NEED_FPU_LOAD))
3428 switch_fpu_return();
3430 /* Apply any externally detected TSC adjustments (due to suspend) */
3431 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3432 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3433 vcpu->arch.tsc_offset_adjustment = 0;
3434 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3437 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3438 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3439 rdtsc() - vcpu->arch.last_host_tsc;
3441 mark_tsc_unstable("KVM discovered backwards TSC");
3443 if (kvm_check_tsc_unstable()) {
3444 u64 offset = kvm_compute_tsc_offset(vcpu,
3445 vcpu->arch.last_guest_tsc);
3446 kvm_vcpu_write_tsc_offset(vcpu, offset);
3447 vcpu->arch.tsc_catchup = 1;
3450 if (kvm_lapic_hv_timer_in_use(vcpu))
3451 kvm_lapic_restart_hv_timer(vcpu);
3454 * On a host with synchronized TSC, there is no need to update
3455 * kvmclock on vcpu->cpu migration
3457 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3458 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3459 if (vcpu->cpu != cpu)
3460 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3464 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3467 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3469 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3472 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3474 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3475 &vcpu->arch.st.steal.preempted,
3476 offsetof(struct kvm_steal_time, preempted),
3477 sizeof(vcpu->arch.st.steal.preempted));
3480 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3484 if (vcpu->preempted)
3485 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3488 * Disable page faults because we're in atomic context here.
3489 * kvm_write_guest_offset_cached() would call might_fault()
3490 * that relies on pagefault_disable() to tell if there's a
3491 * bug. NOTE: the write to guest memory may not go through if
3492 * during postcopy live migration or if there's heavy guest
3495 pagefault_disable();
3497 * kvm_memslots() will be called by
3498 * kvm_write_guest_offset_cached() so take the srcu lock.
3500 idx = srcu_read_lock(&vcpu->kvm->srcu);
3501 kvm_steal_time_set_preempted(vcpu);
3502 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3504 kvm_x86_ops->vcpu_put(vcpu);
3505 vcpu->arch.last_host_tsc = rdtsc();
3507 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3508 * on every vmexit, but if not, we might have a stale dr6 from the
3509 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3514 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3515 struct kvm_lapic_state *s)
3517 if (vcpu->arch.apicv_active)
3518 kvm_x86_ops->sync_pir_to_irr(vcpu);
3520 return kvm_apic_get_state(vcpu, s);
3523 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3524 struct kvm_lapic_state *s)
3528 r = kvm_apic_set_state(vcpu, s);
3531 update_cr8_intercept(vcpu);
3536 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3538 return (!lapic_in_kernel(vcpu) ||
3539 kvm_apic_accept_pic_intr(vcpu));
3543 * if userspace requested an interrupt window, check that the
3544 * interrupt window is open.
3546 * No need to exit to userspace if we already have an interrupt queued.
3548 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3550 return kvm_arch_interrupt_allowed(vcpu) &&
3551 !kvm_cpu_has_interrupt(vcpu) &&
3552 !kvm_event_needs_reinjection(vcpu) &&
3553 kvm_cpu_accept_dm_intr(vcpu);
3556 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3557 struct kvm_interrupt *irq)
3559 if (irq->irq >= KVM_NR_INTERRUPTS)
3562 if (!irqchip_in_kernel(vcpu->kvm)) {
3563 kvm_queue_interrupt(vcpu, irq->irq, false);
3564 kvm_make_request(KVM_REQ_EVENT, vcpu);
3569 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3570 * fail for in-kernel 8259.
3572 if (pic_in_kernel(vcpu->kvm))
3575 if (vcpu->arch.pending_external_vector != -1)
3578 vcpu->arch.pending_external_vector = irq->irq;
3579 kvm_make_request(KVM_REQ_EVENT, vcpu);
3583 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3585 kvm_inject_nmi(vcpu);
3590 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3592 kvm_make_request(KVM_REQ_SMI, vcpu);
3597 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3598 struct kvm_tpr_access_ctl *tac)
3602 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3606 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3610 unsigned bank_num = mcg_cap & 0xff, bank;
3613 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3615 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3618 vcpu->arch.mcg_cap = mcg_cap;
3619 /* Init IA32_MCG_CTL to all 1s */
3620 if (mcg_cap & MCG_CTL_P)
3621 vcpu->arch.mcg_ctl = ~(u64)0;
3622 /* Init IA32_MCi_CTL to all 1s */
3623 for (bank = 0; bank < bank_num; bank++)
3624 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3626 kvm_x86_ops->setup_mce(vcpu);
3631 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3632 struct kvm_x86_mce *mce)
3634 u64 mcg_cap = vcpu->arch.mcg_cap;
3635 unsigned bank_num = mcg_cap & 0xff;
3636 u64 *banks = vcpu->arch.mce_banks;
3638 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3641 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3642 * reporting is disabled
3644 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3645 vcpu->arch.mcg_ctl != ~(u64)0)
3647 banks += 4 * mce->bank;
3649 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3650 * reporting is disabled for the bank
3652 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3654 if (mce->status & MCI_STATUS_UC) {
3655 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3656 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3657 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3660 if (banks[1] & MCI_STATUS_VAL)
3661 mce->status |= MCI_STATUS_OVER;
3662 banks[2] = mce->addr;
3663 banks[3] = mce->misc;
3664 vcpu->arch.mcg_status = mce->mcg_status;
3665 banks[1] = mce->status;
3666 kvm_queue_exception(vcpu, MC_VECTOR);
3667 } else if (!(banks[1] & MCI_STATUS_VAL)
3668 || !(banks[1] & MCI_STATUS_UC)) {
3669 if (banks[1] & MCI_STATUS_VAL)
3670 mce->status |= MCI_STATUS_OVER;
3671 banks[2] = mce->addr;
3672 banks[3] = mce->misc;
3673 banks[1] = mce->status;
3675 banks[1] |= MCI_STATUS_OVER;
3679 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3680 struct kvm_vcpu_events *events)
3685 * The API doesn't provide the instruction length for software
3686 * exceptions, so don't report them. As long as the guest RIP
3687 * isn't advanced, we should expect to encounter the exception
3690 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3691 events->exception.injected = 0;
3692 events->exception.pending = 0;
3694 events->exception.injected = vcpu->arch.exception.injected;
3695 events->exception.pending = vcpu->arch.exception.pending;
3697 * For ABI compatibility, deliberately conflate
3698 * pending and injected exceptions when
3699 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3701 if (!vcpu->kvm->arch.exception_payload_enabled)
3702 events->exception.injected |=
3703 vcpu->arch.exception.pending;
3705 events->exception.nr = vcpu->arch.exception.nr;
3706 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3707 events->exception.error_code = vcpu->arch.exception.error_code;
3708 events->exception_has_payload = vcpu->arch.exception.has_payload;
3709 events->exception_payload = vcpu->arch.exception.payload;
3711 events->interrupt.injected =
3712 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3713 events->interrupt.nr = vcpu->arch.interrupt.nr;
3714 events->interrupt.soft = 0;
3715 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3717 events->nmi.injected = vcpu->arch.nmi_injected;
3718 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3719 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3720 events->nmi.pad = 0;
3722 events->sipi_vector = 0; /* never valid when reporting to user space */
3724 events->smi.smm = is_smm(vcpu);
3725 events->smi.pending = vcpu->arch.smi_pending;
3726 events->smi.smm_inside_nmi =
3727 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3728 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3730 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3731 | KVM_VCPUEVENT_VALID_SHADOW
3732 | KVM_VCPUEVENT_VALID_SMM);
3733 if (vcpu->kvm->arch.exception_payload_enabled)
3734 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3736 memset(&events->reserved, 0, sizeof(events->reserved));
3739 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3741 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3742 struct kvm_vcpu_events *events)
3744 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3745 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3746 | KVM_VCPUEVENT_VALID_SHADOW
3747 | KVM_VCPUEVENT_VALID_SMM
3748 | KVM_VCPUEVENT_VALID_PAYLOAD))
3751 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3752 if (!vcpu->kvm->arch.exception_payload_enabled)
3754 if (events->exception.pending)
3755 events->exception.injected = 0;
3757 events->exception_has_payload = 0;
3759 events->exception.pending = 0;
3760 events->exception_has_payload = 0;
3763 if ((events->exception.injected || events->exception.pending) &&
3764 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3767 /* INITs are latched while in SMM */
3768 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3769 (events->smi.smm || events->smi.pending) &&
3770 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3774 vcpu->arch.exception.injected = events->exception.injected;
3775 vcpu->arch.exception.pending = events->exception.pending;
3776 vcpu->arch.exception.nr = events->exception.nr;
3777 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3778 vcpu->arch.exception.error_code = events->exception.error_code;
3779 vcpu->arch.exception.has_payload = events->exception_has_payload;
3780 vcpu->arch.exception.payload = events->exception_payload;
3782 vcpu->arch.interrupt.injected = events->interrupt.injected;
3783 vcpu->arch.interrupt.nr = events->interrupt.nr;
3784 vcpu->arch.interrupt.soft = events->interrupt.soft;
3785 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3786 kvm_x86_ops->set_interrupt_shadow(vcpu,
3787 events->interrupt.shadow);
3789 vcpu->arch.nmi_injected = events->nmi.injected;
3790 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3791 vcpu->arch.nmi_pending = events->nmi.pending;
3792 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3794 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3795 lapic_in_kernel(vcpu))
3796 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3798 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3799 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3800 if (events->smi.smm)
3801 vcpu->arch.hflags |= HF_SMM_MASK;
3803 vcpu->arch.hflags &= ~HF_SMM_MASK;
3804 kvm_smm_changed(vcpu);
3807 vcpu->arch.smi_pending = events->smi.pending;
3809 if (events->smi.smm) {
3810 if (events->smi.smm_inside_nmi)
3811 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3813 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3814 if (lapic_in_kernel(vcpu)) {
3815 if (events->smi.latched_init)
3816 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3818 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3823 kvm_make_request(KVM_REQ_EVENT, vcpu);
3828 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3829 struct kvm_debugregs *dbgregs)
3833 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3834 kvm_get_dr(vcpu, 6, &val);
3836 dbgregs->dr7 = vcpu->arch.dr7;
3838 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3841 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3842 struct kvm_debugregs *dbgregs)
3847 if (dbgregs->dr6 & ~0xffffffffull)
3849 if (dbgregs->dr7 & ~0xffffffffull)
3852 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3853 kvm_update_dr0123(vcpu);
3854 vcpu->arch.dr6 = dbgregs->dr6;
3855 kvm_update_dr6(vcpu);
3856 vcpu->arch.dr7 = dbgregs->dr7;
3857 kvm_update_dr7(vcpu);
3862 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3864 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3866 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3867 u64 xstate_bv = xsave->header.xfeatures;
3871 * Copy legacy XSAVE area, to avoid complications with CPUID
3872 * leaves 0 and 1 in the loop below.
3874 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3877 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3878 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3881 * Copy each region from the possibly compacted offset to the
3882 * non-compacted offset.
3884 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3886 u64 xfeature_mask = valid & -valid;
3887 int xfeature_nr = fls64(xfeature_mask) - 1;
3888 void *src = get_xsave_addr(xsave, xfeature_nr);
3891 u32 size, offset, ecx, edx;
3892 cpuid_count(XSTATE_CPUID, xfeature_nr,
3893 &size, &offset, &ecx, &edx);
3894 if (xfeature_nr == XFEATURE_PKRU)
3895 memcpy(dest + offset, &vcpu->arch.pkru,
3896 sizeof(vcpu->arch.pkru));
3898 memcpy(dest + offset, src, size);
3902 valid -= xfeature_mask;
3906 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3908 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3909 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3913 * Copy legacy XSAVE area, to avoid complications with CPUID
3914 * leaves 0 and 1 in the loop below.
3916 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3918 /* Set XSTATE_BV and possibly XCOMP_BV. */
3919 xsave->header.xfeatures = xstate_bv;
3920 if (boot_cpu_has(X86_FEATURE_XSAVES))
3921 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3924 * Copy each region from the non-compacted offset to the
3925 * possibly compacted offset.
3927 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3929 u64 xfeature_mask = valid & -valid;
3930 int xfeature_nr = fls64(xfeature_mask) - 1;
3931 void *dest = get_xsave_addr(xsave, xfeature_nr);
3934 u32 size, offset, ecx, edx;
3935 cpuid_count(XSTATE_CPUID, xfeature_nr,
3936 &size, &offset, &ecx, &edx);
3937 if (xfeature_nr == XFEATURE_PKRU)
3938 memcpy(&vcpu->arch.pkru, src + offset,
3939 sizeof(vcpu->arch.pkru));
3941 memcpy(dest, src + offset, size);
3944 valid -= xfeature_mask;
3948 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3949 struct kvm_xsave *guest_xsave)
3951 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3952 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3953 fill_xsave((u8 *) guest_xsave->region, vcpu);
3955 memcpy(guest_xsave->region,
3956 &vcpu->arch.guest_fpu->state.fxsave,
3957 sizeof(struct fxregs_state));
3958 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3959 XFEATURE_MASK_FPSSE;
3963 #define XSAVE_MXCSR_OFFSET 24
3965 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3966 struct kvm_xsave *guest_xsave)
3969 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3970 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3972 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3974 * Here we allow setting states that are not present in
3975 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3976 * with old userspace.
3978 if (xstate_bv & ~kvm_supported_xcr0() ||
3979 mxcsr & ~mxcsr_feature_mask)
3981 load_xsave(vcpu, (u8 *)guest_xsave->region);
3983 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3984 mxcsr & ~mxcsr_feature_mask)
3986 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3987 guest_xsave->region, sizeof(struct fxregs_state));
3992 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3993 struct kvm_xcrs *guest_xcrs)
3995 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3996 guest_xcrs->nr_xcrs = 0;
4000 guest_xcrs->nr_xcrs = 1;
4001 guest_xcrs->flags = 0;
4002 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
4003 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
4006 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
4007 struct kvm_xcrs *guest_xcrs)
4011 if (!boot_cpu_has(X86_FEATURE_XSAVE))
4014 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
4017 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
4018 /* Only support XCR0 currently */
4019 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
4020 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
4021 guest_xcrs->xcrs[i].value);
4030 * kvm_set_guest_paused() indicates to the guest kernel that it has been
4031 * stopped by the hypervisor. This function will be called from the host only.
4032 * EINVAL is returned when the host attempts to set the flag for a guest that
4033 * does not support pv clocks.
4035 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
4037 if (!vcpu->arch.pv_time_enabled)
4039 vcpu->arch.pvclock_set_guest_stopped_request = true;
4040 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
4044 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
4045 struct kvm_enable_cap *cap)
4048 uint16_t vmcs_version;
4049 void __user *user_ptr;
4055 case KVM_CAP_HYPERV_SYNIC2:
4060 case KVM_CAP_HYPERV_SYNIC:
4061 if (!irqchip_in_kernel(vcpu->kvm))
4063 return kvm_hv_activate_synic(vcpu, cap->cap ==
4064 KVM_CAP_HYPERV_SYNIC2);
4065 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
4066 if (!kvm_x86_ops->nested_enable_evmcs)
4068 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
4070 user_ptr = (void __user *)(uintptr_t)cap->args[0];
4071 if (copy_to_user(user_ptr, &vmcs_version,
4072 sizeof(vmcs_version)))
4076 case KVM_CAP_HYPERV_DIRECT_TLBFLUSH:
4077 if (!kvm_x86_ops->enable_direct_tlbflush)
4080 return kvm_x86_ops->enable_direct_tlbflush(vcpu);
4087 long kvm_arch_vcpu_ioctl(struct file *filp,
4088 unsigned int ioctl, unsigned long arg)
4090 struct kvm_vcpu *vcpu = filp->private_data;
4091 void __user *argp = (void __user *)arg;
4094 struct kvm_lapic_state *lapic;
4095 struct kvm_xsave *xsave;
4096 struct kvm_xcrs *xcrs;
4104 case KVM_GET_LAPIC: {
4106 if (!lapic_in_kernel(vcpu))
4108 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
4109 GFP_KERNEL_ACCOUNT);
4114 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
4118 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
4123 case KVM_SET_LAPIC: {
4125 if (!lapic_in_kernel(vcpu))
4127 u.lapic = memdup_user(argp, sizeof(*u.lapic));
4128 if (IS_ERR(u.lapic)) {
4129 r = PTR_ERR(u.lapic);
4133 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
4136 case KVM_INTERRUPT: {
4137 struct kvm_interrupt irq;
4140 if (copy_from_user(&irq, argp, sizeof(irq)))
4142 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4146 r = kvm_vcpu_ioctl_nmi(vcpu);
4150 r = kvm_vcpu_ioctl_smi(vcpu);
4153 case KVM_SET_CPUID: {
4154 struct kvm_cpuid __user *cpuid_arg = argp;
4155 struct kvm_cpuid cpuid;
4158 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4160 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4163 case KVM_SET_CPUID2: {
4164 struct kvm_cpuid2 __user *cpuid_arg = argp;
4165 struct kvm_cpuid2 cpuid;
4168 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4170 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4171 cpuid_arg->entries);
4174 case KVM_GET_CPUID2: {
4175 struct kvm_cpuid2 __user *cpuid_arg = argp;
4176 struct kvm_cpuid2 cpuid;
4179 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4181 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4182 cpuid_arg->entries);
4186 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4191 case KVM_GET_MSRS: {
4192 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4193 r = msr_io(vcpu, argp, do_get_msr, 1);
4194 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4197 case KVM_SET_MSRS: {
4198 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4199 r = msr_io(vcpu, argp, do_set_msr, 0);
4200 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4203 case KVM_TPR_ACCESS_REPORTING: {
4204 struct kvm_tpr_access_ctl tac;
4207 if (copy_from_user(&tac, argp, sizeof(tac)))
4209 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4213 if (copy_to_user(argp, &tac, sizeof(tac)))
4218 case KVM_SET_VAPIC_ADDR: {
4219 struct kvm_vapic_addr va;
4223 if (!lapic_in_kernel(vcpu))
4226 if (copy_from_user(&va, argp, sizeof(va)))
4228 idx = srcu_read_lock(&vcpu->kvm->srcu);
4229 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4230 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4233 case KVM_X86_SETUP_MCE: {
4237 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4239 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4242 case KVM_X86_SET_MCE: {
4243 struct kvm_x86_mce mce;
4246 if (copy_from_user(&mce, argp, sizeof(mce)))
4248 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4251 case KVM_GET_VCPU_EVENTS: {
4252 struct kvm_vcpu_events events;
4254 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4257 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4262 case KVM_SET_VCPU_EVENTS: {
4263 struct kvm_vcpu_events events;
4266 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4269 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4272 case KVM_GET_DEBUGREGS: {
4273 struct kvm_debugregs dbgregs;
4275 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4278 if (copy_to_user(argp, &dbgregs,
4279 sizeof(struct kvm_debugregs)))
4284 case KVM_SET_DEBUGREGS: {
4285 struct kvm_debugregs dbgregs;
4288 if (copy_from_user(&dbgregs, argp,
4289 sizeof(struct kvm_debugregs)))
4292 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4295 case KVM_GET_XSAVE: {
4296 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4301 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4304 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4309 case KVM_SET_XSAVE: {
4310 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4311 if (IS_ERR(u.xsave)) {
4312 r = PTR_ERR(u.xsave);
4316 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4319 case KVM_GET_XCRS: {
4320 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4325 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4328 if (copy_to_user(argp, u.xcrs,
4329 sizeof(struct kvm_xcrs)))
4334 case KVM_SET_XCRS: {
4335 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4336 if (IS_ERR(u.xcrs)) {
4337 r = PTR_ERR(u.xcrs);
4341 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4344 case KVM_SET_TSC_KHZ: {
4348 user_tsc_khz = (u32)arg;
4350 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4353 if (user_tsc_khz == 0)
4354 user_tsc_khz = tsc_khz;
4356 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4361 case KVM_GET_TSC_KHZ: {
4362 r = vcpu->arch.virtual_tsc_khz;
4365 case KVM_KVMCLOCK_CTRL: {
4366 r = kvm_set_guest_paused(vcpu);
4369 case KVM_ENABLE_CAP: {
4370 struct kvm_enable_cap cap;
4373 if (copy_from_user(&cap, argp, sizeof(cap)))
4375 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4378 case KVM_GET_NESTED_STATE: {
4379 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4383 if (!kvm_x86_ops->get_nested_state)
4386 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4388 if (get_user(user_data_size, &user_kvm_nested_state->size))
4391 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4396 if (r > user_data_size) {
4397 if (put_user(r, &user_kvm_nested_state->size))
4407 case KVM_SET_NESTED_STATE: {
4408 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4409 struct kvm_nested_state kvm_state;
4412 if (!kvm_x86_ops->set_nested_state)
4416 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4420 if (kvm_state.size < sizeof(kvm_state))
4423 if (kvm_state.flags &
4424 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4425 | KVM_STATE_NESTED_EVMCS))
4428 /* nested_run_pending implies guest_mode. */
4429 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4430 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4433 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4436 case KVM_GET_SUPPORTED_HV_CPUID: {
4437 struct kvm_cpuid2 __user *cpuid_arg = argp;
4438 struct kvm_cpuid2 cpuid;
4441 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4444 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4445 cpuid_arg->entries);
4450 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4465 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4467 return VM_FAULT_SIGBUS;
4470 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4474 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4476 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4480 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4483 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4486 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4487 unsigned long kvm_nr_mmu_pages)
4489 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4492 mutex_lock(&kvm->slots_lock);
4494 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4495 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4497 mutex_unlock(&kvm->slots_lock);
4501 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4503 return kvm->arch.n_max_mmu_pages;
4506 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4508 struct kvm_pic *pic = kvm->arch.vpic;
4512 switch (chip->chip_id) {
4513 case KVM_IRQCHIP_PIC_MASTER:
4514 memcpy(&chip->chip.pic, &pic->pics[0],
4515 sizeof(struct kvm_pic_state));
4517 case KVM_IRQCHIP_PIC_SLAVE:
4518 memcpy(&chip->chip.pic, &pic->pics[1],
4519 sizeof(struct kvm_pic_state));
4521 case KVM_IRQCHIP_IOAPIC:
4522 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4531 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4533 struct kvm_pic *pic = kvm->arch.vpic;
4537 switch (chip->chip_id) {
4538 case KVM_IRQCHIP_PIC_MASTER:
4539 spin_lock(&pic->lock);
4540 memcpy(&pic->pics[0], &chip->chip.pic,
4541 sizeof(struct kvm_pic_state));
4542 spin_unlock(&pic->lock);
4544 case KVM_IRQCHIP_PIC_SLAVE:
4545 spin_lock(&pic->lock);
4546 memcpy(&pic->pics[1], &chip->chip.pic,
4547 sizeof(struct kvm_pic_state));
4548 spin_unlock(&pic->lock);
4550 case KVM_IRQCHIP_IOAPIC:
4551 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4557 kvm_pic_update_irq(pic);
4561 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4563 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4565 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4567 mutex_lock(&kps->lock);
4568 memcpy(ps, &kps->channels, sizeof(*ps));
4569 mutex_unlock(&kps->lock);
4573 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4576 struct kvm_pit *pit = kvm->arch.vpit;
4578 mutex_lock(&pit->pit_state.lock);
4579 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4580 for (i = 0; i < 3; i++)
4581 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4582 mutex_unlock(&pit->pit_state.lock);
4586 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4588 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4589 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4590 sizeof(ps->channels));
4591 ps->flags = kvm->arch.vpit->pit_state.flags;
4592 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4593 memset(&ps->reserved, 0, sizeof(ps->reserved));
4597 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4601 u32 prev_legacy, cur_legacy;
4602 struct kvm_pit *pit = kvm->arch.vpit;
4604 mutex_lock(&pit->pit_state.lock);
4605 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4606 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4607 if (!prev_legacy && cur_legacy)
4609 memcpy(&pit->pit_state.channels, &ps->channels,
4610 sizeof(pit->pit_state.channels));
4611 pit->pit_state.flags = ps->flags;
4612 for (i = 0; i < 3; i++)
4613 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4615 mutex_unlock(&pit->pit_state.lock);
4619 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4620 struct kvm_reinject_control *control)
4622 struct kvm_pit *pit = kvm->arch.vpit;
4627 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4628 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4629 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4631 mutex_lock(&pit->pit_state.lock);
4632 kvm_pit_set_reinject(pit, control->pit_reinject);
4633 mutex_unlock(&pit->pit_state.lock);
4639 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4640 * @kvm: kvm instance
4641 * @log: slot id and address to which we copy the log
4643 * Steps 1-4 below provide general overview of dirty page logging. See
4644 * kvm_get_dirty_log_protect() function description for additional details.
4646 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4647 * always flush the TLB (step 4) even if previous step failed and the dirty
4648 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4649 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4650 * writes will be marked dirty for next log read.
4652 * 1. Take a snapshot of the bit and clear it if needed.
4653 * 2. Write protect the corresponding page.
4654 * 3. Copy the snapshot to the userspace.
4655 * 4. Flush TLB's if needed.
4657 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4662 mutex_lock(&kvm->slots_lock);
4665 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4667 if (kvm_x86_ops->flush_log_dirty)
4668 kvm_x86_ops->flush_log_dirty(kvm);
4670 r = kvm_get_dirty_log_protect(kvm, log, &flush);
4673 * All the TLBs can be flushed out of mmu lock, see the comments in
4674 * kvm_mmu_slot_remove_write_access().
4676 lockdep_assert_held(&kvm->slots_lock);
4678 kvm_flush_remote_tlbs(kvm);
4680 mutex_unlock(&kvm->slots_lock);
4684 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4689 mutex_lock(&kvm->slots_lock);
4692 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4694 if (kvm_x86_ops->flush_log_dirty)
4695 kvm_x86_ops->flush_log_dirty(kvm);
4697 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4700 * All the TLBs can be flushed out of mmu lock, see the comments in
4701 * kvm_mmu_slot_remove_write_access().
4703 lockdep_assert_held(&kvm->slots_lock);
4705 kvm_flush_remote_tlbs(kvm);
4707 mutex_unlock(&kvm->slots_lock);
4711 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4714 if (!irqchip_in_kernel(kvm))
4717 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4718 irq_event->irq, irq_event->level,
4723 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4724 struct kvm_enable_cap *cap)
4732 case KVM_CAP_DISABLE_QUIRKS:
4733 kvm->arch.disabled_quirks = cap->args[0];
4736 case KVM_CAP_SPLIT_IRQCHIP: {
4737 mutex_lock(&kvm->lock);
4739 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4740 goto split_irqchip_unlock;
4742 if (irqchip_in_kernel(kvm))
4743 goto split_irqchip_unlock;
4744 if (kvm->created_vcpus)
4745 goto split_irqchip_unlock;
4746 r = kvm_setup_empty_irq_routing(kvm);
4748 goto split_irqchip_unlock;
4749 /* Pairs with irqchip_in_kernel. */
4751 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4752 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4754 split_irqchip_unlock:
4755 mutex_unlock(&kvm->lock);
4758 case KVM_CAP_X2APIC_API:
4760 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4763 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4764 kvm->arch.x2apic_format = true;
4765 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4766 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4770 case KVM_CAP_X86_DISABLE_EXITS:
4772 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4775 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4776 kvm_can_mwait_in_guest())
4777 kvm->arch.mwait_in_guest = true;
4778 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4779 kvm->arch.hlt_in_guest = true;
4780 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4781 kvm->arch.pause_in_guest = true;
4782 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4783 kvm->arch.cstate_in_guest = true;
4786 case KVM_CAP_MSR_PLATFORM_INFO:
4787 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4790 case KVM_CAP_EXCEPTION_PAYLOAD:
4791 kvm->arch.exception_payload_enabled = cap->args[0];
4801 long kvm_arch_vm_ioctl(struct file *filp,
4802 unsigned int ioctl, unsigned long arg)
4804 struct kvm *kvm = filp->private_data;
4805 void __user *argp = (void __user *)arg;
4808 * This union makes it completely explicit to gcc-3.x
4809 * that these two variables' stack usage should be
4810 * combined, not added together.
4813 struct kvm_pit_state ps;
4814 struct kvm_pit_state2 ps2;
4815 struct kvm_pit_config pit_config;
4819 case KVM_SET_TSS_ADDR:
4820 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4822 case KVM_SET_IDENTITY_MAP_ADDR: {
4825 mutex_lock(&kvm->lock);
4827 if (kvm->created_vcpus)
4828 goto set_identity_unlock;
4830 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4831 goto set_identity_unlock;
4832 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4833 set_identity_unlock:
4834 mutex_unlock(&kvm->lock);
4837 case KVM_SET_NR_MMU_PAGES:
4838 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4840 case KVM_GET_NR_MMU_PAGES:
4841 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4843 case KVM_CREATE_IRQCHIP: {
4844 mutex_lock(&kvm->lock);
4847 if (irqchip_in_kernel(kvm))
4848 goto create_irqchip_unlock;
4851 if (kvm->created_vcpus)
4852 goto create_irqchip_unlock;
4854 r = kvm_pic_init(kvm);
4856 goto create_irqchip_unlock;
4858 r = kvm_ioapic_init(kvm);
4860 kvm_pic_destroy(kvm);
4861 goto create_irqchip_unlock;
4864 r = kvm_setup_default_irq_routing(kvm);
4866 kvm_ioapic_destroy(kvm);
4867 kvm_pic_destroy(kvm);
4868 goto create_irqchip_unlock;
4870 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4872 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4873 create_irqchip_unlock:
4874 mutex_unlock(&kvm->lock);
4877 case KVM_CREATE_PIT:
4878 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4880 case KVM_CREATE_PIT2:
4882 if (copy_from_user(&u.pit_config, argp,
4883 sizeof(struct kvm_pit_config)))
4886 mutex_lock(&kvm->lock);
4889 goto create_pit_unlock;
4891 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4895 mutex_unlock(&kvm->lock);
4897 case KVM_GET_IRQCHIP: {
4898 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4899 struct kvm_irqchip *chip;
4901 chip = memdup_user(argp, sizeof(*chip));
4908 if (!irqchip_kernel(kvm))
4909 goto get_irqchip_out;
4910 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4912 goto get_irqchip_out;
4914 if (copy_to_user(argp, chip, sizeof(*chip)))
4915 goto get_irqchip_out;
4921 case KVM_SET_IRQCHIP: {
4922 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4923 struct kvm_irqchip *chip;
4925 chip = memdup_user(argp, sizeof(*chip));
4932 if (!irqchip_kernel(kvm))
4933 goto set_irqchip_out;
4934 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4936 goto set_irqchip_out;
4944 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4947 if (!kvm->arch.vpit)
4949 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4953 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4960 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4963 if (!kvm->arch.vpit)
4965 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4968 case KVM_GET_PIT2: {
4970 if (!kvm->arch.vpit)
4972 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4976 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4981 case KVM_SET_PIT2: {
4983 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4986 if (!kvm->arch.vpit)
4988 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4991 case KVM_REINJECT_CONTROL: {
4992 struct kvm_reinject_control control;
4994 if (copy_from_user(&control, argp, sizeof(control)))
4996 r = kvm_vm_ioctl_reinject(kvm, &control);
4999 case KVM_SET_BOOT_CPU_ID:
5001 mutex_lock(&kvm->lock);
5002 if (kvm->created_vcpus)
5005 kvm->arch.bsp_vcpu_id = arg;
5006 mutex_unlock(&kvm->lock);
5008 case KVM_XEN_HVM_CONFIG: {
5009 struct kvm_xen_hvm_config xhc;
5011 if (copy_from_user(&xhc, argp, sizeof(xhc)))
5016 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
5020 case KVM_SET_CLOCK: {
5021 struct kvm_clock_data user_ns;
5025 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
5034 * TODO: userspace has to take care of races with VCPU_RUN, so
5035 * kvm_gen_update_masterclock() can be cut down to locked
5036 * pvclock_update_vm_gtod_copy().
5038 kvm_gen_update_masterclock(kvm);
5039 now_ns = get_kvmclock_ns(kvm);
5040 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
5041 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
5044 case KVM_GET_CLOCK: {
5045 struct kvm_clock_data user_ns;
5048 now_ns = get_kvmclock_ns(kvm);
5049 user_ns.clock = now_ns;
5050 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
5051 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
5054 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
5059 case KVM_MEMORY_ENCRYPT_OP: {
5061 if (kvm_x86_ops->mem_enc_op)
5062 r = kvm_x86_ops->mem_enc_op(kvm, argp);
5065 case KVM_MEMORY_ENCRYPT_REG_REGION: {
5066 struct kvm_enc_region region;
5069 if (copy_from_user(®ion, argp, sizeof(region)))
5073 if (kvm_x86_ops->mem_enc_reg_region)
5074 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
5077 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
5078 struct kvm_enc_region region;
5081 if (copy_from_user(®ion, argp, sizeof(region)))
5085 if (kvm_x86_ops->mem_enc_unreg_region)
5086 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
5089 case KVM_HYPERV_EVENTFD: {
5090 struct kvm_hyperv_eventfd hvevfd;
5093 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
5095 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
5098 case KVM_SET_PMU_EVENT_FILTER:
5099 r = kvm_vm_ioctl_set_pmu_event_filter(kvm, argp);
5108 static void kvm_init_msr_list(void)
5110 struct x86_pmu_capability x86_pmu;
5114 BUILD_BUG_ON_MSG(INTEL_PMC_MAX_FIXED != 4,
5115 "Please update the fixed PMCs in msrs_to_save[]");
5117 perf_get_x86_pmu_capability(&x86_pmu);
5119 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
5120 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
5124 * Even MSRs that are valid in the host may not be exposed
5125 * to the guests in some cases.
5127 switch (msrs_to_save[i]) {
5128 case MSR_IA32_BNDCFGS:
5129 if (!kvm_mpx_supported())
5133 if (!kvm_x86_ops->rdtscp_supported())
5136 case MSR_IA32_RTIT_CTL:
5137 case MSR_IA32_RTIT_STATUS:
5138 if (!kvm_x86_ops->pt_supported())
5141 case MSR_IA32_RTIT_CR3_MATCH:
5142 if (!kvm_x86_ops->pt_supported() ||
5143 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5146 case MSR_IA32_RTIT_OUTPUT_BASE:
5147 case MSR_IA32_RTIT_OUTPUT_MASK:
5148 if (!kvm_x86_ops->pt_supported() ||
5149 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5150 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5153 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5154 if (!kvm_x86_ops->pt_supported() ||
5155 msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
5156 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5159 case MSR_ARCH_PERFMON_PERFCTR0 ... MSR_ARCH_PERFMON_PERFCTR0 + 17:
5160 if (msrs_to_save[i] - MSR_ARCH_PERFMON_PERFCTR0 >=
5161 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5164 case MSR_ARCH_PERFMON_EVENTSEL0 ... MSR_ARCH_PERFMON_EVENTSEL0 + 17:
5165 if (msrs_to_save[i] - MSR_ARCH_PERFMON_EVENTSEL0 >=
5166 min(INTEL_PMC_MAX_GENERIC, x86_pmu.num_counters_gp))
5174 msrs_to_save[j] = msrs_to_save[i];
5177 num_msrs_to_save = j;
5179 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
5180 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
5184 emulated_msrs[j] = emulated_msrs[i];
5187 num_emulated_msrs = j;
5189 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
5190 struct kvm_msr_entry msr;
5192 msr.index = msr_based_features[i];
5193 if (kvm_get_msr_feature(&msr))
5197 msr_based_features[j] = msr_based_features[i];
5200 num_msr_based_features = j;
5203 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5211 if (!(lapic_in_kernel(vcpu) &&
5212 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5213 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5224 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5231 if (!(lapic_in_kernel(vcpu) &&
5232 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5234 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5236 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5246 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5247 struct kvm_segment *var, int seg)
5249 kvm_x86_ops->set_segment(vcpu, var, seg);
5252 void kvm_get_segment(struct kvm_vcpu *vcpu,
5253 struct kvm_segment *var, int seg)
5255 kvm_x86_ops->get_segment(vcpu, var, seg);
5258 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5259 struct x86_exception *exception)
5263 BUG_ON(!mmu_is_nested(vcpu));
5265 /* NPT walks are always user-walks */
5266 access |= PFERR_USER_MASK;
5267 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5272 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5273 struct x86_exception *exception)
5275 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5276 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5279 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5280 struct x86_exception *exception)
5282 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5283 access |= PFERR_FETCH_MASK;
5284 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5287 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5288 struct x86_exception *exception)
5290 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5291 access |= PFERR_WRITE_MASK;
5292 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5295 /* uses this to access any guest's mapped memory without checking CPL */
5296 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5297 struct x86_exception *exception)
5299 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5302 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5303 struct kvm_vcpu *vcpu, u32 access,
5304 struct x86_exception *exception)
5307 int r = X86EMUL_CONTINUE;
5310 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5312 unsigned offset = addr & (PAGE_SIZE-1);
5313 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5316 if (gpa == UNMAPPED_GVA)
5317 return X86EMUL_PROPAGATE_FAULT;
5318 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5321 r = X86EMUL_IO_NEEDED;
5333 /* used for instruction fetching */
5334 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5335 gva_t addr, void *val, unsigned int bytes,
5336 struct x86_exception *exception)
5338 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5339 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5343 /* Inline kvm_read_guest_virt_helper for speed. */
5344 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5346 if (unlikely(gpa == UNMAPPED_GVA))
5347 return X86EMUL_PROPAGATE_FAULT;
5349 offset = addr & (PAGE_SIZE-1);
5350 if (WARN_ON(offset + bytes > PAGE_SIZE))
5351 bytes = (unsigned)PAGE_SIZE - offset;
5352 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5354 if (unlikely(ret < 0))
5355 return X86EMUL_IO_NEEDED;
5357 return X86EMUL_CONTINUE;
5360 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5361 gva_t addr, void *val, unsigned int bytes,
5362 struct x86_exception *exception)
5364 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5367 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5368 * is returned, but our callers are not ready for that and they blindly
5369 * call kvm_inject_page_fault. Ensure that they at least do not leak
5370 * uninitialized kernel stack memory into cr2 and error code.
5372 memset(exception, 0, sizeof(*exception));
5373 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5376 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5378 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5379 gva_t addr, void *val, unsigned int bytes,
5380 struct x86_exception *exception, bool system)
5382 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5385 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5386 access |= PFERR_USER_MASK;
5388 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5391 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5392 unsigned long addr, void *val, unsigned int bytes)
5394 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5395 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5397 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5400 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5401 struct kvm_vcpu *vcpu, u32 access,
5402 struct x86_exception *exception)
5405 int r = X86EMUL_CONTINUE;
5408 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5411 unsigned offset = addr & (PAGE_SIZE-1);
5412 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5415 if (gpa == UNMAPPED_GVA)
5416 return X86EMUL_PROPAGATE_FAULT;
5417 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5419 r = X86EMUL_IO_NEEDED;
5431 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5432 unsigned int bytes, struct x86_exception *exception,
5435 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5436 u32 access = PFERR_WRITE_MASK;
5438 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5439 access |= PFERR_USER_MASK;
5441 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5445 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5446 unsigned int bytes, struct x86_exception *exception)
5448 /* kvm_write_guest_virt_system can pull in tons of pages. */
5449 vcpu->arch.l1tf_flush_l1d = true;
5452 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5453 * is returned, but our callers are not ready for that and they blindly
5454 * call kvm_inject_page_fault. Ensure that they at least do not leak
5455 * uninitialized kernel stack memory into cr2 and error code.
5457 memset(exception, 0, sizeof(*exception));
5458 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5459 PFERR_WRITE_MASK, exception);
5461 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5463 int handle_ud(struct kvm_vcpu *vcpu)
5465 int emul_type = EMULTYPE_TRAP_UD;
5466 char sig[5]; /* ud2; .ascii "kvm" */
5467 struct x86_exception e;
5469 if (force_emulation_prefix &&
5470 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5471 sig, sizeof(sig), &e) == 0 &&
5472 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5473 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5474 emul_type = EMULTYPE_TRAP_UD_FORCED;
5477 return kvm_emulate_instruction(vcpu, emul_type);
5479 EXPORT_SYMBOL_GPL(handle_ud);
5481 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5482 gpa_t gpa, bool write)
5484 /* For APIC access vmexit */
5485 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5488 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5489 trace_vcpu_match_mmio(gva, gpa, write, true);
5496 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5497 gpa_t *gpa, struct x86_exception *exception,
5500 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5501 | (write ? PFERR_WRITE_MASK : 0);
5504 * currently PKRU is only applied to ept enabled guest so
5505 * there is no pkey in EPT page table for L1 guest or EPT
5506 * shadow page table for L2 guest.
5508 if (vcpu_match_mmio_gva(vcpu, gva)
5509 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5510 vcpu->arch.mmio_access, 0, access)) {
5511 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5512 (gva & (PAGE_SIZE - 1));
5513 trace_vcpu_match_mmio(gva, *gpa, write, false);
5517 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5519 if (*gpa == UNMAPPED_GVA)
5522 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5525 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5526 const void *val, int bytes)
5530 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5533 kvm_page_track_write(vcpu, gpa, val, bytes);
5537 struct read_write_emulator_ops {
5538 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5540 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5541 void *val, int bytes);
5542 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5543 int bytes, void *val);
5544 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5545 void *val, int bytes);
5549 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5551 if (vcpu->mmio_read_completed) {
5552 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5553 vcpu->mmio_fragments[0].gpa, val);
5554 vcpu->mmio_read_completed = 0;
5561 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5562 void *val, int bytes)
5564 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5567 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5568 void *val, int bytes)
5570 return emulator_write_phys(vcpu, gpa, val, bytes);
5573 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5575 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5576 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5579 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5580 void *val, int bytes)
5582 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5583 return X86EMUL_IO_NEEDED;
5586 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5587 void *val, int bytes)
5589 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5591 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5592 return X86EMUL_CONTINUE;
5595 static const struct read_write_emulator_ops read_emultor = {
5596 .read_write_prepare = read_prepare,
5597 .read_write_emulate = read_emulate,
5598 .read_write_mmio = vcpu_mmio_read,
5599 .read_write_exit_mmio = read_exit_mmio,
5602 static const struct read_write_emulator_ops write_emultor = {
5603 .read_write_emulate = write_emulate,
5604 .read_write_mmio = write_mmio,
5605 .read_write_exit_mmio = write_exit_mmio,
5609 static int emulator_read_write_onepage(unsigned long addr, void *val,
5611 struct x86_exception *exception,
5612 struct kvm_vcpu *vcpu,
5613 const struct read_write_emulator_ops *ops)
5617 bool write = ops->write;
5618 struct kvm_mmio_fragment *frag;
5619 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5622 * If the exit was due to a NPF we may already have a GPA.
5623 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5624 * Note, this cannot be used on string operations since string
5625 * operation using rep will only have the initial GPA from the NPF
5628 if (vcpu->arch.gpa_available &&
5629 emulator_can_use_gpa(ctxt) &&
5630 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5631 gpa = vcpu->arch.gpa_val;
5632 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5634 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5636 return X86EMUL_PROPAGATE_FAULT;
5639 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5640 return X86EMUL_CONTINUE;
5643 * Is this MMIO handled locally?
5645 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5646 if (handled == bytes)
5647 return X86EMUL_CONTINUE;
5653 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5654 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5658 return X86EMUL_CONTINUE;
5661 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5663 void *val, unsigned int bytes,
5664 struct x86_exception *exception,
5665 const struct read_write_emulator_ops *ops)
5667 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5671 if (ops->read_write_prepare &&
5672 ops->read_write_prepare(vcpu, val, bytes))
5673 return X86EMUL_CONTINUE;
5675 vcpu->mmio_nr_fragments = 0;
5677 /* Crossing a page boundary? */
5678 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5681 now = -addr & ~PAGE_MASK;
5682 rc = emulator_read_write_onepage(addr, val, now, exception,
5685 if (rc != X86EMUL_CONTINUE)
5688 if (ctxt->mode != X86EMUL_MODE_PROT64)
5694 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5696 if (rc != X86EMUL_CONTINUE)
5699 if (!vcpu->mmio_nr_fragments)
5702 gpa = vcpu->mmio_fragments[0].gpa;
5704 vcpu->mmio_needed = 1;
5705 vcpu->mmio_cur_fragment = 0;
5707 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5708 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5709 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5710 vcpu->run->mmio.phys_addr = gpa;
5712 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5715 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5719 struct x86_exception *exception)
5721 return emulator_read_write(ctxt, addr, val, bytes,
5722 exception, &read_emultor);
5725 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5729 struct x86_exception *exception)
5731 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5732 exception, &write_emultor);
5735 #define CMPXCHG_TYPE(t, ptr, old, new) \
5736 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5738 #ifdef CONFIG_X86_64
5739 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5741 # define CMPXCHG64(ptr, old, new) \
5742 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5745 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5750 struct x86_exception *exception)
5752 struct kvm_host_map map;
5753 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5758 /* guests cmpxchg8b have to be emulated atomically */
5759 if (bytes > 8 || (bytes & (bytes - 1)))
5762 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5764 if (gpa == UNMAPPED_GVA ||
5765 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5768 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5771 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5774 kaddr = map.hva + offset_in_page(gpa);
5778 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5781 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5784 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5787 exchanged = CMPXCHG64(kaddr, old, new);
5793 kvm_vcpu_unmap(vcpu, &map, true);
5796 return X86EMUL_CMPXCHG_FAILED;
5798 kvm_page_track_write(vcpu, gpa, new, bytes);
5800 return X86EMUL_CONTINUE;
5803 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5805 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5808 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5812 for (i = 0; i < vcpu->arch.pio.count; i++) {
5813 if (vcpu->arch.pio.in)
5814 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5815 vcpu->arch.pio.size, pd);
5817 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5818 vcpu->arch.pio.port, vcpu->arch.pio.size,
5822 pd += vcpu->arch.pio.size;
5827 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5828 unsigned short port, void *val,
5829 unsigned int count, bool in)
5831 vcpu->arch.pio.port = port;
5832 vcpu->arch.pio.in = in;
5833 vcpu->arch.pio.count = count;
5834 vcpu->arch.pio.size = size;
5836 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5837 vcpu->arch.pio.count = 0;
5841 vcpu->run->exit_reason = KVM_EXIT_IO;
5842 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5843 vcpu->run->io.size = size;
5844 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5845 vcpu->run->io.count = count;
5846 vcpu->run->io.port = port;
5851 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5852 int size, unsigned short port, void *val,
5855 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5858 if (vcpu->arch.pio.count)
5861 memset(vcpu->arch.pio_data, 0, size * count);
5863 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5866 memcpy(val, vcpu->arch.pio_data, size * count);
5867 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5868 vcpu->arch.pio.count = 0;
5875 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5876 int size, unsigned short port,
5877 const void *val, unsigned int count)
5879 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5881 memcpy(vcpu->arch.pio_data, val, size * count);
5882 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5883 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5886 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5888 return kvm_x86_ops->get_segment_base(vcpu, seg);
5891 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5893 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5896 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5898 if (!need_emulate_wbinvd(vcpu))
5899 return X86EMUL_CONTINUE;
5901 if (kvm_x86_ops->has_wbinvd_exit()) {
5902 int cpu = get_cpu();
5904 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5905 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5906 wbinvd_ipi, NULL, 1);
5908 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5911 return X86EMUL_CONTINUE;
5914 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5916 kvm_emulate_wbinvd_noskip(vcpu);
5917 return kvm_skip_emulated_instruction(vcpu);
5919 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5923 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5925 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5928 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5929 unsigned long *dest)
5931 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5934 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5935 unsigned long value)
5938 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5941 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5943 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5946 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5948 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5949 unsigned long value;
5953 value = kvm_read_cr0(vcpu);
5956 value = vcpu->arch.cr2;
5959 value = kvm_read_cr3(vcpu);
5962 value = kvm_read_cr4(vcpu);
5965 value = kvm_get_cr8(vcpu);
5968 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5975 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5977 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5982 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5985 vcpu->arch.cr2 = val;
5988 res = kvm_set_cr3(vcpu, val);
5991 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5994 res = kvm_set_cr8(vcpu, val);
5997 kvm_err("%s: unexpected cr %u\n", __func__, cr);
6004 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
6006 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
6009 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6011 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
6014 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6016 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
6019 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6021 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
6024 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
6026 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
6029 static unsigned long emulator_get_cached_segment_base(
6030 struct x86_emulate_ctxt *ctxt, int seg)
6032 return get_segment_base(emul_to_vcpu(ctxt), seg);
6035 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
6036 struct desc_struct *desc, u32 *base3,
6039 struct kvm_segment var;
6041 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
6042 *selector = var.selector;
6045 memset(desc, 0, sizeof(*desc));
6053 set_desc_limit(desc, var.limit);
6054 set_desc_base(desc, (unsigned long)var.base);
6055 #ifdef CONFIG_X86_64
6057 *base3 = var.base >> 32;
6059 desc->type = var.type;
6061 desc->dpl = var.dpl;
6062 desc->p = var.present;
6063 desc->avl = var.avl;
6071 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
6072 struct desc_struct *desc, u32 base3,
6075 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6076 struct kvm_segment var;
6078 var.selector = selector;
6079 var.base = get_desc_base(desc);
6080 #ifdef CONFIG_X86_64
6081 var.base |= ((u64)base3) << 32;
6083 var.limit = get_desc_limit(desc);
6085 var.limit = (var.limit << 12) | 0xfff;
6086 var.type = desc->type;
6087 var.dpl = desc->dpl;
6092 var.avl = desc->avl;
6093 var.present = desc->p;
6094 var.unusable = !var.present;
6097 kvm_set_segment(vcpu, &var, seg);
6101 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
6102 u32 msr_index, u64 *pdata)
6104 return kvm_get_msr(emul_to_vcpu(ctxt), msr_index, pdata);
6107 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
6108 u32 msr_index, u64 data)
6110 return kvm_set_msr(emul_to_vcpu(ctxt), msr_index, data);
6113 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
6115 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6117 return vcpu->arch.smbase;
6120 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
6122 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6124 vcpu->arch.smbase = smbase;
6127 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
6130 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
6133 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
6134 u32 pmc, u64 *pdata)
6136 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
6139 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6141 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6144 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6145 struct x86_instruction_info *info,
6146 enum x86_intercept_stage stage)
6148 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6151 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6152 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6154 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6157 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6159 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6162 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6164 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6167 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6169 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6172 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6174 return emul_to_vcpu(ctxt)->arch.hflags;
6177 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6179 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6182 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6183 const char *smstate)
6185 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6188 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6190 kvm_smm_changed(emul_to_vcpu(ctxt));
6193 static int emulator_set_xcr(struct x86_emulate_ctxt *ctxt, u32 index, u64 xcr)
6195 return __kvm_set_xcr(emul_to_vcpu(ctxt), index, xcr);
6198 static const struct x86_emulate_ops emulate_ops = {
6199 .read_gpr = emulator_read_gpr,
6200 .write_gpr = emulator_write_gpr,
6201 .read_std = emulator_read_std,
6202 .write_std = emulator_write_std,
6203 .read_phys = kvm_read_guest_phys_system,
6204 .fetch = kvm_fetch_guest_virt,
6205 .read_emulated = emulator_read_emulated,
6206 .write_emulated = emulator_write_emulated,
6207 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6208 .invlpg = emulator_invlpg,
6209 .pio_in_emulated = emulator_pio_in_emulated,
6210 .pio_out_emulated = emulator_pio_out_emulated,
6211 .get_segment = emulator_get_segment,
6212 .set_segment = emulator_set_segment,
6213 .get_cached_segment_base = emulator_get_cached_segment_base,
6214 .get_gdt = emulator_get_gdt,
6215 .get_idt = emulator_get_idt,
6216 .set_gdt = emulator_set_gdt,
6217 .set_idt = emulator_set_idt,
6218 .get_cr = emulator_get_cr,
6219 .set_cr = emulator_set_cr,
6220 .cpl = emulator_get_cpl,
6221 .get_dr = emulator_get_dr,
6222 .set_dr = emulator_set_dr,
6223 .get_smbase = emulator_get_smbase,
6224 .set_smbase = emulator_set_smbase,
6225 .set_msr = emulator_set_msr,
6226 .get_msr = emulator_get_msr,
6227 .check_pmc = emulator_check_pmc,
6228 .read_pmc = emulator_read_pmc,
6229 .halt = emulator_halt,
6230 .wbinvd = emulator_wbinvd,
6231 .fix_hypercall = emulator_fix_hypercall,
6232 .intercept = emulator_intercept,
6233 .get_cpuid = emulator_get_cpuid,
6234 .set_nmi_mask = emulator_set_nmi_mask,
6235 .get_hflags = emulator_get_hflags,
6236 .set_hflags = emulator_set_hflags,
6237 .pre_leave_smm = emulator_pre_leave_smm,
6238 .post_leave_smm = emulator_post_leave_smm,
6239 .set_xcr = emulator_set_xcr,
6242 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6244 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6246 * an sti; sti; sequence only disable interrupts for the first
6247 * instruction. So, if the last instruction, be it emulated or
6248 * not, left the system with the INT_STI flag enabled, it
6249 * means that the last instruction is an sti. We should not
6250 * leave the flag on in this case. The same goes for mov ss
6252 if (int_shadow & mask)
6254 if (unlikely(int_shadow || mask)) {
6255 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6257 kvm_make_request(KVM_REQ_EVENT, vcpu);
6261 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6263 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6264 if (ctxt->exception.vector == PF_VECTOR)
6265 return kvm_propagate_fault(vcpu, &ctxt->exception);
6267 if (ctxt->exception.error_code_valid)
6268 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6269 ctxt->exception.error_code);
6271 kvm_queue_exception(vcpu, ctxt->exception.vector);
6275 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6277 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6280 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6282 ctxt->eflags = kvm_get_rflags(vcpu);
6283 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6285 ctxt->eip = kvm_rip_read(vcpu);
6286 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6287 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6288 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6289 cs_db ? X86EMUL_MODE_PROT32 :
6290 X86EMUL_MODE_PROT16;
6291 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6292 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6293 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6295 init_decode_cache(ctxt);
6296 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6299 void kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6301 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6304 init_emulate_ctxt(vcpu);
6308 ctxt->_eip = ctxt->eip + inc_eip;
6309 ret = emulate_int_real(ctxt, irq);
6311 if (ret != X86EMUL_CONTINUE) {
6312 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6314 ctxt->eip = ctxt->_eip;
6315 kvm_rip_write(vcpu, ctxt->eip);
6316 kvm_set_rflags(vcpu, ctxt->eflags);
6319 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6321 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6323 ++vcpu->stat.insn_emulation_fail;
6324 trace_kvm_emulate_insn_failed(vcpu);
6326 if (emulation_type & EMULTYPE_VMWARE_GP) {
6327 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6331 if (emulation_type & EMULTYPE_SKIP) {
6332 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6333 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6334 vcpu->run->internal.ndata = 0;
6338 kvm_queue_exception(vcpu, UD_VECTOR);
6340 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6341 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6342 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6343 vcpu->run->internal.ndata = 0;
6350 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6351 bool write_fault_to_shadow_pgtable,
6357 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6360 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6363 if (!vcpu->arch.mmu->direct_map) {
6365 * Write permission should be allowed since only
6366 * write access need to be emulated.
6368 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6371 * If the mapping is invalid in guest, let cpu retry
6372 * it to generate fault.
6374 if (gpa == UNMAPPED_GVA)
6379 * Do not retry the unhandleable instruction if it faults on the
6380 * readonly host memory, otherwise it will goto a infinite loop:
6381 * retry instruction -> write #PF -> emulation fail -> retry
6382 * instruction -> ...
6384 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6387 * If the instruction failed on the error pfn, it can not be fixed,
6388 * report the error to userspace.
6390 if (is_error_noslot_pfn(pfn))
6393 kvm_release_pfn_clean(pfn);
6395 /* The instructions are well-emulated on direct mmu. */
6396 if (vcpu->arch.mmu->direct_map) {
6397 unsigned int indirect_shadow_pages;
6399 spin_lock(&vcpu->kvm->mmu_lock);
6400 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6401 spin_unlock(&vcpu->kvm->mmu_lock);
6403 if (indirect_shadow_pages)
6404 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6410 * if emulation was due to access to shadowed page table
6411 * and it failed try to unshadow page and re-enter the
6412 * guest to let CPU execute the instruction.
6414 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6417 * If the access faults on its page table, it can not
6418 * be fixed by unprotecting shadow page and it should
6419 * be reported to userspace.
6421 return !write_fault_to_shadow_pgtable;
6424 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6425 unsigned long cr2, int emulation_type)
6427 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6428 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6430 last_retry_eip = vcpu->arch.last_retry_eip;
6431 last_retry_addr = vcpu->arch.last_retry_addr;
6434 * If the emulation is caused by #PF and it is non-page_table
6435 * writing instruction, it means the VM-EXIT is caused by shadow
6436 * page protected, we can zap the shadow page and retry this
6437 * instruction directly.
6439 * Note: if the guest uses a non-page-table modifying instruction
6440 * on the PDE that points to the instruction, then we will unmap
6441 * the instruction and go to an infinite loop. So, we cache the
6442 * last retried eip and the last fault address, if we meet the eip
6443 * and the address again, we can break out of the potential infinite
6446 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6448 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6451 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6454 if (x86_page_table_writing_insn(ctxt))
6457 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6460 vcpu->arch.last_retry_eip = ctxt->eip;
6461 vcpu->arch.last_retry_addr = cr2;
6463 if (!vcpu->arch.mmu->direct_map)
6464 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6466 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6471 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6472 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6474 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6476 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6477 /* This is a good place to trace that we are exiting SMM. */
6478 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6480 /* Process a latched INIT or SMI, if any. */
6481 kvm_make_request(KVM_REQ_EVENT, vcpu);
6484 kvm_mmu_reset_context(vcpu);
6487 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6496 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6497 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6502 static int kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu)
6504 struct kvm_run *kvm_run = vcpu->run;
6506 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6507 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6508 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6509 kvm_run->debug.arch.exception = DB_VECTOR;
6510 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6513 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6517 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6519 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6522 r = kvm_x86_ops->skip_emulated_instruction(vcpu);
6527 * rflags is the old, "raw" value of the flags. The new value has
6528 * not been saved yet.
6530 * This is correct even for TF set by the guest, because "the
6531 * processor will not generate this exception after the instruction
6532 * that sets the TF flag".
6534 if (unlikely(rflags & X86_EFLAGS_TF))
6535 r = kvm_vcpu_do_singlestep(vcpu);
6538 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6540 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6542 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6543 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6544 struct kvm_run *kvm_run = vcpu->run;
6545 unsigned long eip = kvm_get_linear_rip(vcpu);
6546 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6547 vcpu->arch.guest_debug_dr7,
6551 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6552 kvm_run->debug.arch.pc = eip;
6553 kvm_run->debug.arch.exception = DB_VECTOR;
6554 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6560 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6561 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6562 unsigned long eip = kvm_get_linear_rip(vcpu);
6563 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6568 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6569 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6570 kvm_queue_exception(vcpu, DB_VECTOR);
6579 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6581 switch (ctxt->opcode_len) {
6588 case 0xe6: /* OUT */
6592 case 0x6c: /* INS */
6594 case 0x6e: /* OUTS */
6601 case 0x33: /* RDPMC */
6610 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6617 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6618 bool writeback = true;
6619 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6621 vcpu->arch.l1tf_flush_l1d = true;
6624 * Clear write_fault_to_shadow_pgtable here to ensure it is
6627 vcpu->arch.write_fault_to_shadow_pgtable = false;
6628 kvm_clear_exception_queue(vcpu);
6630 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6631 init_emulate_ctxt(vcpu);
6634 * We will reenter on the same instruction since
6635 * we do not set complete_userspace_io. This does not
6636 * handle watchpoints yet, those would be handled in
6639 if (!(emulation_type & EMULTYPE_SKIP) &&
6640 kvm_vcpu_check_breakpoint(vcpu, &r))
6643 ctxt->interruptibility = 0;
6644 ctxt->have_exception = false;
6645 ctxt->exception.vector = -1;
6646 ctxt->perm_ok = false;
6648 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6650 r = x86_decode_insn(ctxt, insn, insn_len);
6652 trace_kvm_emulate_insn_start(vcpu);
6653 ++vcpu->stat.insn_emulation;
6654 if (r != EMULATION_OK) {
6655 if ((emulation_type & EMULTYPE_TRAP_UD) ||
6656 (emulation_type & EMULTYPE_TRAP_UD_FORCED)) {
6657 kvm_queue_exception(vcpu, UD_VECTOR);
6660 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6663 if (ctxt->have_exception) {
6665 * #UD should result in just EMULATION_FAILED, and trap-like
6666 * exception should not be encountered during decode.
6668 WARN_ON_ONCE(ctxt->exception.vector == UD_VECTOR ||
6669 exception_type(ctxt->exception.vector) == EXCPT_TRAP);
6670 inject_emulated_exception(vcpu);
6673 return handle_emulation_failure(vcpu, emulation_type);
6677 if ((emulation_type & EMULTYPE_VMWARE_GP) &&
6678 !is_vmware_backdoor_opcode(ctxt)) {
6679 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6684 * Note, EMULTYPE_SKIP is intended for use *only* by vendor callbacks
6685 * for kvm_skip_emulated_instruction(). The caller is responsible for
6686 * updating interruptibility state and injecting single-step #DBs.
6688 if (emulation_type & EMULTYPE_SKIP) {
6689 kvm_rip_write(vcpu, ctxt->_eip);
6690 if (ctxt->eflags & X86_EFLAGS_RF)
6691 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6695 if (retry_instruction(ctxt, cr2, emulation_type))
6698 /* this is needed for vmware backdoor interface to work since it
6699 changes registers values during IO operation */
6700 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6701 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6702 emulator_invalidate_register_cache(ctxt);
6706 /* Save the faulting GPA (cr2) in the address field */
6707 ctxt->exception.address = cr2;
6709 r = x86_emulate_insn(ctxt);
6711 if (r == EMULATION_INTERCEPTED)
6714 if (r == EMULATION_FAILED) {
6715 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6719 return handle_emulation_failure(vcpu, emulation_type);
6722 if (ctxt->have_exception) {
6724 if (inject_emulated_exception(vcpu))
6726 } else if (vcpu->arch.pio.count) {
6727 if (!vcpu->arch.pio.in) {
6728 /* FIXME: return into emulator if single-stepping. */
6729 vcpu->arch.pio.count = 0;
6732 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6735 } else if (vcpu->mmio_needed) {
6736 ++vcpu->stat.mmio_exits;
6738 if (!vcpu->mmio_is_write)
6741 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6742 } else if (r == EMULATION_RESTART)
6748 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6749 toggle_interruptibility(vcpu, ctxt->interruptibility);
6750 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6751 if (!ctxt->have_exception ||
6752 exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
6753 kvm_rip_write(vcpu, ctxt->eip);
6755 r = kvm_vcpu_do_singlestep(vcpu);
6756 __kvm_set_rflags(vcpu, ctxt->eflags);
6760 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6761 * do nothing, and it will be requested again as soon as
6762 * the shadow expires. But we still need to check here,
6763 * because POPF has no interrupt shadow.
6765 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6766 kvm_make_request(KVM_REQ_EVENT, vcpu);
6768 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6773 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6775 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6777 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6779 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6780 void *insn, int insn_len)
6782 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6784 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6786 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6788 vcpu->arch.pio.count = 0;
6792 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6794 vcpu->arch.pio.count = 0;
6796 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6799 return kvm_skip_emulated_instruction(vcpu);
6802 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6803 unsigned short port)
6805 unsigned long val = kvm_rax_read(vcpu);
6806 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6807 size, port, &val, 1);
6812 * Workaround userspace that relies on old KVM behavior of %rip being
6813 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6816 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6817 vcpu->arch.complete_userspace_io =
6818 complete_fast_pio_out_port_0x7e;
6819 kvm_skip_emulated_instruction(vcpu);
6821 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6822 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6827 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6831 /* We should only ever be called with arch.pio.count equal to 1 */
6832 BUG_ON(vcpu->arch.pio.count != 1);
6834 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6835 vcpu->arch.pio.count = 0;
6839 /* For size less than 4 we merge, else we zero extend */
6840 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6843 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6844 * the copy and tracing
6846 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6847 vcpu->arch.pio.port, &val, 1);
6848 kvm_rax_write(vcpu, val);
6850 return kvm_skip_emulated_instruction(vcpu);
6853 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6854 unsigned short port)
6859 /* For size less than 4 we merge, else we zero extend */
6860 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6862 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6865 kvm_rax_write(vcpu, val);
6869 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6870 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6875 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6880 ret = kvm_fast_pio_in(vcpu, size, port);
6882 ret = kvm_fast_pio_out(vcpu, size, port);
6883 return ret && kvm_skip_emulated_instruction(vcpu);
6885 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6887 static int kvmclock_cpu_down_prep(unsigned int cpu)
6889 __this_cpu_write(cpu_tsc_khz, 0);
6893 static void tsc_khz_changed(void *data)
6895 struct cpufreq_freqs *freq = data;
6896 unsigned long khz = 0;
6900 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6901 khz = cpufreq_quick_get(raw_smp_processor_id());
6904 __this_cpu_write(cpu_tsc_khz, khz);
6907 #ifdef CONFIG_X86_64
6908 static void kvm_hyperv_tsc_notifier(void)
6911 struct kvm_vcpu *vcpu;
6914 mutex_lock(&kvm_lock);
6915 list_for_each_entry(kvm, &vm_list, vm_list)
6916 kvm_make_mclock_inprogress_request(kvm);
6918 hyperv_stop_tsc_emulation();
6920 /* TSC frequency always matches when on Hyper-V */
6921 for_each_present_cpu(cpu)
6922 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6923 kvm_max_guest_tsc_khz = tsc_khz;
6925 list_for_each_entry(kvm, &vm_list, vm_list) {
6926 struct kvm_arch *ka = &kvm->arch;
6928 spin_lock(&ka->pvclock_gtod_sync_lock);
6930 pvclock_update_vm_gtod_copy(kvm);
6932 kvm_for_each_vcpu(cpu, vcpu, kvm)
6933 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6935 kvm_for_each_vcpu(cpu, vcpu, kvm)
6936 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6938 spin_unlock(&ka->pvclock_gtod_sync_lock);
6940 mutex_unlock(&kvm_lock);
6944 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6947 struct kvm_vcpu *vcpu;
6948 int i, send_ipi = 0;
6951 * We allow guests to temporarily run on slowing clocks,
6952 * provided we notify them after, or to run on accelerating
6953 * clocks, provided we notify them before. Thus time never
6956 * However, we have a problem. We can't atomically update
6957 * the frequency of a given CPU from this function; it is
6958 * merely a notifier, which can be called from any CPU.
6959 * Changing the TSC frequency at arbitrary points in time
6960 * requires a recomputation of local variables related to
6961 * the TSC for each VCPU. We must flag these local variables
6962 * to be updated and be sure the update takes place with the
6963 * new frequency before any guests proceed.
6965 * Unfortunately, the combination of hotplug CPU and frequency
6966 * change creates an intractable locking scenario; the order
6967 * of when these callouts happen is undefined with respect to
6968 * CPU hotplug, and they can race with each other. As such,
6969 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6970 * undefined; you can actually have a CPU frequency change take
6971 * place in between the computation of X and the setting of the
6972 * variable. To protect against this problem, all updates of
6973 * the per_cpu tsc_khz variable are done in an interrupt
6974 * protected IPI, and all callers wishing to update the value
6975 * must wait for a synchronous IPI to complete (which is trivial
6976 * if the caller is on the CPU already). This establishes the
6977 * necessary total order on variable updates.
6979 * Note that because a guest time update may take place
6980 * anytime after the setting of the VCPU's request bit, the
6981 * correct TSC value must be set before the request. However,
6982 * to ensure the update actually makes it to any guest which
6983 * starts running in hardware virtualization between the set
6984 * and the acquisition of the spinlock, we must also ping the
6985 * CPU after setting the request bit.
6989 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6991 mutex_lock(&kvm_lock);
6992 list_for_each_entry(kvm, &vm_list, vm_list) {
6993 kvm_for_each_vcpu(i, vcpu, kvm) {
6994 if (vcpu->cpu != cpu)
6996 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6997 if (vcpu->cpu != raw_smp_processor_id())
7001 mutex_unlock(&kvm_lock);
7003 if (freq->old < freq->new && send_ipi) {
7005 * We upscale the frequency. Must make the guest
7006 * doesn't see old kvmclock values while running with
7007 * the new frequency, otherwise we risk the guest sees
7008 * time go backwards.
7010 * In case we update the frequency for another cpu
7011 * (which might be in guest context) send an interrupt
7012 * to kick the cpu out of guest context. Next time
7013 * guest context is entered kvmclock will be updated,
7014 * so the guest will not see stale values.
7016 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
7020 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
7023 struct cpufreq_freqs *freq = data;
7026 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
7028 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
7031 for_each_cpu(cpu, freq->policy->cpus)
7032 __kvmclock_cpufreq_notifier(freq, cpu);
7037 static struct notifier_block kvmclock_cpufreq_notifier_block = {
7038 .notifier_call = kvmclock_cpufreq_notifier
7041 static int kvmclock_cpu_online(unsigned int cpu)
7043 tsc_khz_changed(NULL);
7047 static void kvm_timer_init(void)
7049 max_tsc_khz = tsc_khz;
7051 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
7052 #ifdef CONFIG_CPU_FREQ
7053 struct cpufreq_policy policy;
7056 memset(&policy, 0, sizeof(policy));
7058 cpufreq_get_policy(&policy, cpu);
7059 if (policy.cpuinfo.max_freq)
7060 max_tsc_khz = policy.cpuinfo.max_freq;
7063 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
7064 CPUFREQ_TRANSITION_NOTIFIER);
7067 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
7068 kvmclock_cpu_online, kvmclock_cpu_down_prep);
7071 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
7072 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
7074 int kvm_is_in_guest(void)
7076 return __this_cpu_read(current_vcpu) != NULL;
7079 static int kvm_is_user_mode(void)
7083 if (__this_cpu_read(current_vcpu))
7084 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
7086 return user_mode != 0;
7089 static unsigned long kvm_get_guest_ip(void)
7091 unsigned long ip = 0;
7093 if (__this_cpu_read(current_vcpu))
7094 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
7099 static void kvm_handle_intel_pt_intr(void)
7101 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
7103 kvm_make_request(KVM_REQ_PMI, vcpu);
7104 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
7105 (unsigned long *)&vcpu->arch.pmu.global_status);
7108 static struct perf_guest_info_callbacks kvm_guest_cbs = {
7109 .is_in_guest = kvm_is_in_guest,
7110 .is_user_mode = kvm_is_user_mode,
7111 .get_guest_ip = kvm_get_guest_ip,
7112 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
7115 #ifdef CONFIG_X86_64
7116 static void pvclock_gtod_update_fn(struct work_struct *work)
7120 struct kvm_vcpu *vcpu;
7123 mutex_lock(&kvm_lock);
7124 list_for_each_entry(kvm, &vm_list, vm_list)
7125 kvm_for_each_vcpu(i, vcpu, kvm)
7126 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
7127 atomic_set(&kvm_guest_has_master_clock, 0);
7128 mutex_unlock(&kvm_lock);
7131 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
7134 * Notification about pvclock gtod data update.
7136 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
7139 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
7140 struct timekeeper *tk = priv;
7142 update_pvclock_gtod(tk);
7144 /* disable master clock if host does not trust, or does not
7145 * use, TSC based clocksource.
7147 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
7148 atomic_read(&kvm_guest_has_master_clock) != 0)
7149 queue_work(system_long_wq, &pvclock_gtod_work);
7154 static struct notifier_block pvclock_gtod_notifier = {
7155 .notifier_call = pvclock_gtod_notify,
7159 int kvm_arch_init(void *opaque)
7162 struct kvm_x86_ops *ops = opaque;
7165 printk(KERN_ERR "kvm: already loaded the other module\n");
7170 if (!ops->cpu_has_kvm_support()) {
7171 printk(KERN_ERR "kvm: no hardware support\n");
7175 if (ops->disabled_by_bios()) {
7176 printk(KERN_ERR "kvm: disabled by bios\n");
7182 * KVM explicitly assumes that the guest has an FPU and
7183 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7184 * vCPU's FPU state as a fxregs_state struct.
7186 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7187 printk(KERN_ERR "kvm: inadequate fpu\n");
7193 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7194 __alignof__(struct fpu), SLAB_ACCOUNT,
7196 if (!x86_fpu_cache) {
7197 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7201 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7203 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7204 goto out_free_x86_fpu_cache;
7207 r = kvm_mmu_module_init();
7209 goto out_free_percpu;
7213 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7214 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7215 PT_PRESENT_MASK, 0, sme_me_mask);
7218 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7220 if (boot_cpu_has(X86_FEATURE_XSAVE))
7221 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7224 if (pi_inject_timer == -1)
7225 pi_inject_timer = housekeeping_enabled(HK_FLAG_TIMER);
7226 #ifdef CONFIG_X86_64
7227 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7229 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7230 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7236 free_percpu(shared_msrs);
7237 out_free_x86_fpu_cache:
7238 kmem_cache_destroy(x86_fpu_cache);
7243 void kvm_arch_exit(void)
7245 #ifdef CONFIG_X86_64
7246 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7247 clear_hv_tscchange_cb();
7250 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7252 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7253 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7254 CPUFREQ_TRANSITION_NOTIFIER);
7255 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7256 #ifdef CONFIG_X86_64
7257 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7260 kvm_mmu_module_exit();
7261 free_percpu(shared_msrs);
7262 kmem_cache_destroy(x86_fpu_cache);
7265 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7267 ++vcpu->stat.halt_exits;
7268 if (lapic_in_kernel(vcpu)) {
7269 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7272 vcpu->run->exit_reason = KVM_EXIT_HLT;
7276 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7278 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7280 int ret = kvm_skip_emulated_instruction(vcpu);
7282 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7283 * KVM_EXIT_DEBUG here.
7285 return kvm_vcpu_halt(vcpu) && ret;
7287 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7289 #ifdef CONFIG_X86_64
7290 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7291 unsigned long clock_type)
7293 struct kvm_clock_pairing clock_pairing;
7294 struct timespec64 ts;
7298 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7299 return -KVM_EOPNOTSUPP;
7301 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7302 return -KVM_EOPNOTSUPP;
7304 clock_pairing.sec = ts.tv_sec;
7305 clock_pairing.nsec = ts.tv_nsec;
7306 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7307 clock_pairing.flags = 0;
7308 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7311 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7312 sizeof(struct kvm_clock_pairing)))
7320 * kvm_pv_kick_cpu_op: Kick a vcpu.
7322 * @apicid - apicid of vcpu to be kicked.
7324 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7326 struct kvm_lapic_irq lapic_irq;
7328 lapic_irq.shorthand = 0;
7329 lapic_irq.dest_mode = 0;
7330 lapic_irq.level = 0;
7331 lapic_irq.dest_id = apicid;
7332 lapic_irq.msi_redir_hint = false;
7334 lapic_irq.delivery_mode = APIC_DM_REMRD;
7335 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7338 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7340 if (!lapic_in_kernel(vcpu)) {
7341 WARN_ON_ONCE(vcpu->arch.apicv_active);
7344 if (!vcpu->arch.apicv_active)
7347 vcpu->arch.apicv_active = false;
7348 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7351 static void kvm_sched_yield(struct kvm *kvm, unsigned long dest_id)
7353 struct kvm_vcpu *target = NULL;
7354 struct kvm_apic_map *map;
7357 map = rcu_dereference(kvm->arch.apic_map);
7359 if (likely(map) && dest_id <= map->max_apic_id && map->phys_map[dest_id])
7360 target = map->phys_map[dest_id]->vcpu;
7364 if (target && READ_ONCE(target->ready))
7365 kvm_vcpu_yield_to(target);
7368 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7370 unsigned long nr, a0, a1, a2, a3, ret;
7373 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7374 return kvm_hv_hypercall(vcpu);
7376 nr = kvm_rax_read(vcpu);
7377 a0 = kvm_rbx_read(vcpu);
7378 a1 = kvm_rcx_read(vcpu);
7379 a2 = kvm_rdx_read(vcpu);
7380 a3 = kvm_rsi_read(vcpu);
7382 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7384 op_64_bit = is_64_bit_mode(vcpu);
7393 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7399 case KVM_HC_VAPIC_POLL_IRQ:
7402 case KVM_HC_KICK_CPU:
7403 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7404 kvm_sched_yield(vcpu->kvm, a1);
7407 #ifdef CONFIG_X86_64
7408 case KVM_HC_CLOCK_PAIRING:
7409 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7412 case KVM_HC_SEND_IPI:
7413 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7415 case KVM_HC_SCHED_YIELD:
7416 kvm_sched_yield(vcpu->kvm, a0);
7426 kvm_rax_write(vcpu, ret);
7428 ++vcpu->stat.hypercalls;
7429 return kvm_skip_emulated_instruction(vcpu);
7431 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7433 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7435 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7436 char instruction[3];
7437 unsigned long rip = kvm_rip_read(vcpu);
7439 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7441 return emulator_write_emulated(ctxt, rip, instruction, 3,
7445 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7447 return vcpu->run->request_interrupt_window &&
7448 likely(!pic_in_kernel(vcpu->kvm));
7451 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7453 struct kvm_run *kvm_run = vcpu->run;
7455 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7456 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7457 kvm_run->cr8 = kvm_get_cr8(vcpu);
7458 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7459 kvm_run->ready_for_interrupt_injection =
7460 pic_in_kernel(vcpu->kvm) ||
7461 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7464 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7468 if (!kvm_x86_ops->update_cr8_intercept)
7471 if (!lapic_in_kernel(vcpu))
7474 if (vcpu->arch.apicv_active)
7477 if (!vcpu->arch.apic->vapic_addr)
7478 max_irr = kvm_lapic_find_highest_irr(vcpu);
7485 tpr = kvm_lapic_get_cr8(vcpu);
7487 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7490 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7494 /* try to reinject previous events if any */
7496 if (vcpu->arch.exception.injected)
7497 kvm_x86_ops->queue_exception(vcpu);
7499 * Do not inject an NMI or interrupt if there is a pending
7500 * exception. Exceptions and interrupts are recognized at
7501 * instruction boundaries, i.e. the start of an instruction.
7502 * Trap-like exceptions, e.g. #DB, have higher priority than
7503 * NMIs and interrupts, i.e. traps are recognized before an
7504 * NMI/interrupt that's pending on the same instruction.
7505 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7506 * priority, but are only generated (pended) during instruction
7507 * execution, i.e. a pending fault-like exception means the
7508 * fault occurred on the *previous* instruction and must be
7509 * serviced prior to recognizing any new events in order to
7510 * fully complete the previous instruction.
7512 else if (!vcpu->arch.exception.pending) {
7513 if (vcpu->arch.nmi_injected)
7514 kvm_x86_ops->set_nmi(vcpu);
7515 else if (vcpu->arch.interrupt.injected)
7516 kvm_x86_ops->set_irq(vcpu);
7520 * Call check_nested_events() even if we reinjected a previous event
7521 * in order for caller to determine if it should require immediate-exit
7522 * from L2 to L1 due to pending L1 events which require exit
7525 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7526 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7531 /* try to inject new event if pending */
7532 if (vcpu->arch.exception.pending) {
7533 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7534 vcpu->arch.exception.has_error_code,
7535 vcpu->arch.exception.error_code);
7537 WARN_ON_ONCE(vcpu->arch.exception.injected);
7538 vcpu->arch.exception.pending = false;
7539 vcpu->arch.exception.injected = true;
7541 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7542 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7545 if (vcpu->arch.exception.nr == DB_VECTOR) {
7547 * This code assumes that nSVM doesn't use
7548 * check_nested_events(). If it does, the
7549 * DR6/DR7 changes should happen before L1
7550 * gets a #VMEXIT for an intercepted #DB in
7551 * L2. (Under VMX, on the other hand, the
7552 * DR6/DR7 changes should not happen in the
7553 * event of a VM-exit to L1 for an intercepted
7556 kvm_deliver_exception_payload(vcpu);
7557 if (vcpu->arch.dr7 & DR7_GD) {
7558 vcpu->arch.dr7 &= ~DR7_GD;
7559 kvm_update_dr7(vcpu);
7563 kvm_x86_ops->queue_exception(vcpu);
7566 /* Don't consider new event if we re-injected an event */
7567 if (kvm_event_needs_reinjection(vcpu))
7570 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7571 kvm_x86_ops->smi_allowed(vcpu)) {
7572 vcpu->arch.smi_pending = false;
7573 ++vcpu->arch.smi_count;
7575 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7576 --vcpu->arch.nmi_pending;
7577 vcpu->arch.nmi_injected = true;
7578 kvm_x86_ops->set_nmi(vcpu);
7579 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7581 * Because interrupts can be injected asynchronously, we are
7582 * calling check_nested_events again here to avoid a race condition.
7583 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7584 * proposal and current concerns. Perhaps we should be setting
7585 * KVM_REQ_EVENT only on certain events and not unconditionally?
7587 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7588 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7592 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7593 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7595 kvm_x86_ops->set_irq(vcpu);
7602 static void process_nmi(struct kvm_vcpu *vcpu)
7607 * x86 is limited to one NMI running, and one NMI pending after it.
7608 * If an NMI is already in progress, limit further NMIs to just one.
7609 * Otherwise, allow two (and we'll inject the first one immediately).
7611 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7614 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7615 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7616 kvm_make_request(KVM_REQ_EVENT, vcpu);
7619 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7622 flags |= seg->g << 23;
7623 flags |= seg->db << 22;
7624 flags |= seg->l << 21;
7625 flags |= seg->avl << 20;
7626 flags |= seg->present << 15;
7627 flags |= seg->dpl << 13;
7628 flags |= seg->s << 12;
7629 flags |= seg->type << 8;
7633 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7635 struct kvm_segment seg;
7638 kvm_get_segment(vcpu, &seg, n);
7639 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7642 offset = 0x7f84 + n * 12;
7644 offset = 0x7f2c + (n - 3) * 12;
7646 put_smstate(u32, buf, offset + 8, seg.base);
7647 put_smstate(u32, buf, offset + 4, seg.limit);
7648 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7651 #ifdef CONFIG_X86_64
7652 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7654 struct kvm_segment seg;
7658 kvm_get_segment(vcpu, &seg, n);
7659 offset = 0x7e00 + n * 16;
7661 flags = enter_smm_get_segment_flags(&seg) >> 8;
7662 put_smstate(u16, buf, offset, seg.selector);
7663 put_smstate(u16, buf, offset + 2, flags);
7664 put_smstate(u32, buf, offset + 4, seg.limit);
7665 put_smstate(u64, buf, offset + 8, seg.base);
7669 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7672 struct kvm_segment seg;
7676 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7677 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7678 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7679 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7681 for (i = 0; i < 8; i++)
7682 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7684 kvm_get_dr(vcpu, 6, &val);
7685 put_smstate(u32, buf, 0x7fcc, (u32)val);
7686 kvm_get_dr(vcpu, 7, &val);
7687 put_smstate(u32, buf, 0x7fc8, (u32)val);
7689 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7690 put_smstate(u32, buf, 0x7fc4, seg.selector);
7691 put_smstate(u32, buf, 0x7f64, seg.base);
7692 put_smstate(u32, buf, 0x7f60, seg.limit);
7693 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7695 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7696 put_smstate(u32, buf, 0x7fc0, seg.selector);
7697 put_smstate(u32, buf, 0x7f80, seg.base);
7698 put_smstate(u32, buf, 0x7f7c, seg.limit);
7699 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7701 kvm_x86_ops->get_gdt(vcpu, &dt);
7702 put_smstate(u32, buf, 0x7f74, dt.address);
7703 put_smstate(u32, buf, 0x7f70, dt.size);
7705 kvm_x86_ops->get_idt(vcpu, &dt);
7706 put_smstate(u32, buf, 0x7f58, dt.address);
7707 put_smstate(u32, buf, 0x7f54, dt.size);
7709 for (i = 0; i < 6; i++)
7710 enter_smm_save_seg_32(vcpu, buf, i);
7712 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7715 put_smstate(u32, buf, 0x7efc, 0x00020000);
7716 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7719 #ifdef CONFIG_X86_64
7720 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7723 struct kvm_segment seg;
7727 for (i = 0; i < 16; i++)
7728 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7730 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7731 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7733 kvm_get_dr(vcpu, 6, &val);
7734 put_smstate(u64, buf, 0x7f68, val);
7735 kvm_get_dr(vcpu, 7, &val);
7736 put_smstate(u64, buf, 0x7f60, val);
7738 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7739 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7740 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7742 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7745 put_smstate(u32, buf, 0x7efc, 0x00020064);
7747 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7749 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7750 put_smstate(u16, buf, 0x7e90, seg.selector);
7751 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7752 put_smstate(u32, buf, 0x7e94, seg.limit);
7753 put_smstate(u64, buf, 0x7e98, seg.base);
7755 kvm_x86_ops->get_idt(vcpu, &dt);
7756 put_smstate(u32, buf, 0x7e84, dt.size);
7757 put_smstate(u64, buf, 0x7e88, dt.address);
7759 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7760 put_smstate(u16, buf, 0x7e70, seg.selector);
7761 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7762 put_smstate(u32, buf, 0x7e74, seg.limit);
7763 put_smstate(u64, buf, 0x7e78, seg.base);
7765 kvm_x86_ops->get_gdt(vcpu, &dt);
7766 put_smstate(u32, buf, 0x7e64, dt.size);
7767 put_smstate(u64, buf, 0x7e68, dt.address);
7769 for (i = 0; i < 6; i++)
7770 enter_smm_save_seg_64(vcpu, buf, i);
7774 static void enter_smm(struct kvm_vcpu *vcpu)
7776 struct kvm_segment cs, ds;
7781 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7782 memset(buf, 0, 512);
7783 #ifdef CONFIG_X86_64
7784 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7785 enter_smm_save_state_64(vcpu, buf);
7788 enter_smm_save_state_32(vcpu, buf);
7791 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7792 * vCPU state (e.g. leave guest mode) after we've saved the state into
7793 * the SMM state-save area.
7795 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7797 vcpu->arch.hflags |= HF_SMM_MASK;
7798 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7800 if (kvm_x86_ops->get_nmi_mask(vcpu))
7801 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7803 kvm_x86_ops->set_nmi_mask(vcpu, true);
7805 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7806 kvm_rip_write(vcpu, 0x8000);
7808 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7809 kvm_x86_ops->set_cr0(vcpu, cr0);
7810 vcpu->arch.cr0 = cr0;
7812 kvm_x86_ops->set_cr4(vcpu, 0);
7814 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7815 dt.address = dt.size = 0;
7816 kvm_x86_ops->set_idt(vcpu, &dt);
7818 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7820 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7821 cs.base = vcpu->arch.smbase;
7826 cs.limit = ds.limit = 0xffffffff;
7827 cs.type = ds.type = 0x3;
7828 cs.dpl = ds.dpl = 0;
7833 cs.avl = ds.avl = 0;
7834 cs.present = ds.present = 1;
7835 cs.unusable = ds.unusable = 0;
7836 cs.padding = ds.padding = 0;
7838 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7839 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7840 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7841 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7842 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7843 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7845 #ifdef CONFIG_X86_64
7846 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7847 kvm_x86_ops->set_efer(vcpu, 0);
7850 kvm_update_cpuid(vcpu);
7851 kvm_mmu_reset_context(vcpu);
7854 static void process_smi(struct kvm_vcpu *vcpu)
7856 vcpu->arch.smi_pending = true;
7857 kvm_make_request(KVM_REQ_EVENT, vcpu);
7860 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7862 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7865 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7867 if (!kvm_apic_present(vcpu))
7870 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7872 if (irqchip_split(vcpu->kvm))
7873 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7875 if (vcpu->arch.apicv_active)
7876 kvm_x86_ops->sync_pir_to_irr(vcpu);
7877 if (ioapic_in_kernel(vcpu->kvm))
7878 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7881 if (is_guest_mode(vcpu))
7882 vcpu->arch.load_eoi_exitmap_pending = true;
7884 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7887 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7889 u64 eoi_exit_bitmap[4];
7891 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7894 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7895 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7896 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7899 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7900 unsigned long start, unsigned long end,
7903 unsigned long apic_address;
7906 * The physical address of apic access page is stored in the VMCS.
7907 * Update it when it becomes invalid.
7909 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7910 if (start <= apic_address && apic_address < end)
7911 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7916 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7918 struct page *page = NULL;
7920 if (!lapic_in_kernel(vcpu))
7923 if (!kvm_x86_ops->set_apic_access_page_addr)
7926 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7927 if (is_error_page(page))
7929 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7932 * Do not pin apic access page in memory, the MMU notifier
7933 * will call us again if it is migrated or swapped out.
7937 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7939 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7941 smp_send_reschedule(vcpu->cpu);
7943 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7946 * Returns 1 to let vcpu_run() continue the guest execution loop without
7947 * exiting to the userspace. Otherwise, the value will be returned to the
7950 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7954 dm_request_for_irq_injection(vcpu) &&
7955 kvm_cpu_accept_dm_intr(vcpu);
7957 bool req_immediate_exit = false;
7959 if (kvm_request_pending(vcpu)) {
7960 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu)) {
7961 if (unlikely(!kvm_x86_ops->get_vmcs12_pages(vcpu))) {
7966 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7967 kvm_mmu_unload(vcpu);
7968 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7969 __kvm_migrate_timers(vcpu);
7970 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7971 kvm_gen_update_masterclock(vcpu->kvm);
7972 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7973 kvm_gen_kvmclock_update(vcpu);
7974 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7975 r = kvm_guest_time_update(vcpu);
7979 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7980 kvm_mmu_sync_roots(vcpu);
7981 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7982 kvm_mmu_load_cr3(vcpu);
7983 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7984 kvm_vcpu_flush_tlb(vcpu, true);
7985 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7986 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7990 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7991 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7992 vcpu->mmio_needed = 0;
7996 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7997 /* Page is swapped out. Do synthetic halt */
7998 vcpu->arch.apf.halted = true;
8002 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
8003 record_steal_time(vcpu);
8004 if (kvm_check_request(KVM_REQ_SMI, vcpu))
8006 if (kvm_check_request(KVM_REQ_NMI, vcpu))
8008 if (kvm_check_request(KVM_REQ_PMU, vcpu))
8009 kvm_pmu_handle_event(vcpu);
8010 if (kvm_check_request(KVM_REQ_PMI, vcpu))
8011 kvm_pmu_deliver_pmi(vcpu);
8012 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
8013 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
8014 if (test_bit(vcpu->arch.pending_ioapic_eoi,
8015 vcpu->arch.ioapic_handled_vectors)) {
8016 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
8017 vcpu->run->eoi.vector =
8018 vcpu->arch.pending_ioapic_eoi;
8023 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
8024 vcpu_scan_ioapic(vcpu);
8025 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
8026 vcpu_load_eoi_exitmap(vcpu);
8027 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
8028 kvm_vcpu_reload_apic_access_page(vcpu);
8029 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
8030 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8031 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
8035 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
8036 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
8037 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
8041 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
8042 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
8043 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
8049 * KVM_REQ_HV_STIMER has to be processed after
8050 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
8051 * depend on the guest clock being up-to-date
8053 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
8054 kvm_hv_process_stimers(vcpu);
8057 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
8058 ++vcpu->stat.req_event;
8059 kvm_apic_accept_events(vcpu);
8060 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
8065 if (inject_pending_event(vcpu, req_int_win) != 0)
8066 req_immediate_exit = true;
8068 /* Enable SMI/NMI/IRQ window open exits if needed.
8070 * SMIs have three cases:
8071 * 1) They can be nested, and then there is nothing to
8072 * do here because RSM will cause a vmexit anyway.
8073 * 2) There is an ISA-specific reason why SMI cannot be
8074 * injected, and the moment when this changes can be
8076 * 3) Or the SMI can be pending because
8077 * inject_pending_event has completed the injection
8078 * of an IRQ or NMI from the previous vmexit, and
8079 * then we request an immediate exit to inject the
8082 if (vcpu->arch.smi_pending && !is_smm(vcpu))
8083 if (!kvm_x86_ops->enable_smi_window(vcpu))
8084 req_immediate_exit = true;
8085 if (vcpu->arch.nmi_pending)
8086 kvm_x86_ops->enable_nmi_window(vcpu);
8087 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
8088 kvm_x86_ops->enable_irq_window(vcpu);
8089 WARN_ON(vcpu->arch.exception.pending);
8092 if (kvm_lapic_enabled(vcpu)) {
8093 update_cr8_intercept(vcpu);
8094 kvm_lapic_sync_to_vapic(vcpu);
8098 r = kvm_mmu_reload(vcpu);
8100 goto cancel_injection;
8105 kvm_x86_ops->prepare_guest_switch(vcpu);
8108 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
8109 * IPI are then delayed after guest entry, which ensures that they
8110 * result in virtual interrupt delivery.
8112 local_irq_disable();
8113 vcpu->mode = IN_GUEST_MODE;
8115 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8118 * 1) We should set ->mode before checking ->requests. Please see
8119 * the comment in kvm_vcpu_exiting_guest_mode().
8121 * 2) For APICv, we should set ->mode before checking PID.ON. This
8122 * pairs with the memory barrier implicit in pi_test_and_set_on
8123 * (see vmx_deliver_posted_interrupt).
8125 * 3) This also orders the write to mode from any reads to the page
8126 * tables done while the VCPU is running. Please see the comment
8127 * in kvm_flush_remote_tlbs.
8129 smp_mb__after_srcu_read_unlock();
8132 * This handles the case where a posted interrupt was
8133 * notified with kvm_vcpu_kick.
8135 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
8136 kvm_x86_ops->sync_pir_to_irr(vcpu);
8138 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
8139 || need_resched() || signal_pending(current)) {
8140 vcpu->mode = OUTSIDE_GUEST_MODE;
8144 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8146 goto cancel_injection;
8149 if (req_immediate_exit) {
8150 kvm_make_request(KVM_REQ_EVENT, vcpu);
8151 kvm_x86_ops->request_immediate_exit(vcpu);
8154 trace_kvm_entry(vcpu->vcpu_id);
8155 guest_enter_irqoff();
8157 /* The preempt notifier should have taken care of the FPU already. */
8158 WARN_ON_ONCE(test_thread_flag(TIF_NEED_FPU_LOAD));
8160 if (unlikely(vcpu->arch.switch_db_regs)) {
8162 set_debugreg(vcpu->arch.eff_db[0], 0);
8163 set_debugreg(vcpu->arch.eff_db[1], 1);
8164 set_debugreg(vcpu->arch.eff_db[2], 2);
8165 set_debugreg(vcpu->arch.eff_db[3], 3);
8166 set_debugreg(vcpu->arch.dr6, 6);
8167 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8170 kvm_x86_ops->run(vcpu);
8173 * Do this here before restoring debug registers on the host. And
8174 * since we do this before handling the vmexit, a DR access vmexit
8175 * can (a) read the correct value of the debug registers, (b) set
8176 * KVM_DEBUGREG_WONT_EXIT again.
8178 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
8179 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
8180 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
8181 kvm_update_dr0123(vcpu);
8182 kvm_update_dr6(vcpu);
8183 kvm_update_dr7(vcpu);
8184 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
8188 * If the guest has used debug registers, at least dr7
8189 * will be disabled while returning to the host.
8190 * If we don't have active breakpoints in the host, we don't
8191 * care about the messed up debug address registers. But if
8192 * we have some of them active, restore the old state.
8194 if (hw_breakpoint_active())
8195 hw_breakpoint_restore();
8197 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
8199 vcpu->mode = OUTSIDE_GUEST_MODE;
8202 kvm_x86_ops->handle_exit_irqoff(vcpu);
8205 * Consume any pending interrupts, including the possible source of
8206 * VM-Exit on SVM and any ticks that occur between VM-Exit and now.
8207 * An instruction is required after local_irq_enable() to fully unblock
8208 * interrupts on processors that implement an interrupt shadow, the
8209 * stat.exits increment will do nicely.
8211 kvm_before_interrupt(vcpu);
8214 local_irq_disable();
8215 kvm_after_interrupt(vcpu);
8217 guest_exit_irqoff();
8218 if (lapic_in_kernel(vcpu)) {
8219 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8220 if (delta != S64_MIN) {
8221 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8222 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8229 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8232 * Profile KVM exit RIPs:
8234 if (unlikely(prof_on == KVM_PROFILING)) {
8235 unsigned long rip = kvm_rip_read(vcpu);
8236 profile_hit(KVM_PROFILING, (void *)rip);
8239 if (unlikely(vcpu->arch.tsc_always_catchup))
8240 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8242 if (vcpu->arch.apic_attention)
8243 kvm_lapic_sync_from_vapic(vcpu);
8245 vcpu->arch.gpa_available = false;
8246 r = kvm_x86_ops->handle_exit(vcpu);
8250 kvm_x86_ops->cancel_injection(vcpu);
8251 if (unlikely(vcpu->arch.apic_attention))
8252 kvm_lapic_sync_from_vapic(vcpu);
8257 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8259 if (!kvm_arch_vcpu_runnable(vcpu) &&
8260 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8261 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8262 kvm_vcpu_block(vcpu);
8263 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8265 if (kvm_x86_ops->post_block)
8266 kvm_x86_ops->post_block(vcpu);
8268 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8272 kvm_apic_accept_events(vcpu);
8273 switch(vcpu->arch.mp_state) {
8274 case KVM_MP_STATE_HALTED:
8275 vcpu->arch.pv.pv_unhalted = false;
8276 vcpu->arch.mp_state =
8277 KVM_MP_STATE_RUNNABLE;
8279 case KVM_MP_STATE_RUNNABLE:
8280 vcpu->arch.apf.halted = false;
8282 case KVM_MP_STATE_INIT_RECEIVED:
8291 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8293 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8294 kvm_x86_ops->check_nested_events(vcpu, false);
8296 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8297 !vcpu->arch.apf.halted);
8300 static int vcpu_run(struct kvm_vcpu *vcpu)
8303 struct kvm *kvm = vcpu->kvm;
8305 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8306 vcpu->arch.l1tf_flush_l1d = true;
8309 if (kvm_vcpu_running(vcpu)) {
8310 r = vcpu_enter_guest(vcpu);
8312 r = vcpu_block(kvm, vcpu);
8318 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8319 if (kvm_cpu_has_pending_timer(vcpu))
8320 kvm_inject_pending_timer_irqs(vcpu);
8322 if (dm_request_for_irq_injection(vcpu) &&
8323 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8325 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8326 ++vcpu->stat.request_irq_exits;
8330 kvm_check_async_pf_completion(vcpu);
8332 if (signal_pending(current)) {
8334 vcpu->run->exit_reason = KVM_EXIT_INTR;
8335 ++vcpu->stat.signal_exits;
8338 if (need_resched()) {
8339 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8341 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8345 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8350 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8354 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8355 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8356 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8360 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8362 BUG_ON(!vcpu->arch.pio.count);
8364 return complete_emulated_io(vcpu);
8368 * Implements the following, as a state machine:
8372 * for each mmio piece in the fragment
8380 * for each mmio piece in the fragment
8385 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8387 struct kvm_run *run = vcpu->run;
8388 struct kvm_mmio_fragment *frag;
8391 BUG_ON(!vcpu->mmio_needed);
8393 /* Complete previous fragment */
8394 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8395 len = min(8u, frag->len);
8396 if (!vcpu->mmio_is_write)
8397 memcpy(frag->data, run->mmio.data, len);
8399 if (frag->len <= 8) {
8400 /* Switch to the next fragment. */
8402 vcpu->mmio_cur_fragment++;
8404 /* Go forward to the next mmio piece. */
8410 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8411 vcpu->mmio_needed = 0;
8413 /* FIXME: return into emulator if single-stepping. */
8414 if (vcpu->mmio_is_write)
8416 vcpu->mmio_read_completed = 1;
8417 return complete_emulated_io(vcpu);
8420 run->exit_reason = KVM_EXIT_MMIO;
8421 run->mmio.phys_addr = frag->gpa;
8422 if (vcpu->mmio_is_write)
8423 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8424 run->mmio.len = min(8u, frag->len);
8425 run->mmio.is_write = vcpu->mmio_is_write;
8426 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8430 /* Swap (qemu) user FPU context for the guest FPU context. */
8431 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8435 copy_fpregs_to_fpstate(vcpu->arch.user_fpu);
8436 /* PKRU is separately restored in kvm_x86_ops->run. */
8437 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8438 ~XFEATURE_MASK_PKRU);
8440 fpregs_mark_activate();
8446 /* When vcpu_run ends, restore user space FPU context. */
8447 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8451 copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8452 copy_kernel_to_fpregs(&vcpu->arch.user_fpu->state);
8454 fpregs_mark_activate();
8457 ++vcpu->stat.fpu_reload;
8461 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8466 kvm_sigset_activate(vcpu);
8467 kvm_load_guest_fpu(vcpu);
8469 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8470 if (kvm_run->immediate_exit) {
8474 kvm_vcpu_block(vcpu);
8475 kvm_apic_accept_events(vcpu);
8476 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8478 if (signal_pending(current)) {
8480 vcpu->run->exit_reason = KVM_EXIT_INTR;
8481 ++vcpu->stat.signal_exits;
8486 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8491 if (vcpu->run->kvm_dirty_regs) {
8492 r = sync_regs(vcpu);
8497 /* re-sync apic's tpr */
8498 if (!lapic_in_kernel(vcpu)) {
8499 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8505 if (unlikely(vcpu->arch.complete_userspace_io)) {
8506 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8507 vcpu->arch.complete_userspace_io = NULL;
8512 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8514 if (kvm_run->immediate_exit)
8520 kvm_put_guest_fpu(vcpu);
8521 if (vcpu->run->kvm_valid_regs)
8523 post_kvm_run_save(vcpu);
8524 kvm_sigset_deactivate(vcpu);
8530 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8532 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8534 * We are here if userspace calls get_regs() in the middle of
8535 * instruction emulation. Registers state needs to be copied
8536 * back from emulation context to vcpu. Userspace shouldn't do
8537 * that usually, but some bad designed PV devices (vmware
8538 * backdoor interface) need this to work
8540 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8541 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8543 regs->rax = kvm_rax_read(vcpu);
8544 regs->rbx = kvm_rbx_read(vcpu);
8545 regs->rcx = kvm_rcx_read(vcpu);
8546 regs->rdx = kvm_rdx_read(vcpu);
8547 regs->rsi = kvm_rsi_read(vcpu);
8548 regs->rdi = kvm_rdi_read(vcpu);
8549 regs->rsp = kvm_rsp_read(vcpu);
8550 regs->rbp = kvm_rbp_read(vcpu);
8551 #ifdef CONFIG_X86_64
8552 regs->r8 = kvm_r8_read(vcpu);
8553 regs->r9 = kvm_r9_read(vcpu);
8554 regs->r10 = kvm_r10_read(vcpu);
8555 regs->r11 = kvm_r11_read(vcpu);
8556 regs->r12 = kvm_r12_read(vcpu);
8557 regs->r13 = kvm_r13_read(vcpu);
8558 regs->r14 = kvm_r14_read(vcpu);
8559 regs->r15 = kvm_r15_read(vcpu);
8562 regs->rip = kvm_rip_read(vcpu);
8563 regs->rflags = kvm_get_rflags(vcpu);
8566 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8569 __get_regs(vcpu, regs);
8574 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8576 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8577 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8579 kvm_rax_write(vcpu, regs->rax);
8580 kvm_rbx_write(vcpu, regs->rbx);
8581 kvm_rcx_write(vcpu, regs->rcx);
8582 kvm_rdx_write(vcpu, regs->rdx);
8583 kvm_rsi_write(vcpu, regs->rsi);
8584 kvm_rdi_write(vcpu, regs->rdi);
8585 kvm_rsp_write(vcpu, regs->rsp);
8586 kvm_rbp_write(vcpu, regs->rbp);
8587 #ifdef CONFIG_X86_64
8588 kvm_r8_write(vcpu, regs->r8);
8589 kvm_r9_write(vcpu, regs->r9);
8590 kvm_r10_write(vcpu, regs->r10);
8591 kvm_r11_write(vcpu, regs->r11);
8592 kvm_r12_write(vcpu, regs->r12);
8593 kvm_r13_write(vcpu, regs->r13);
8594 kvm_r14_write(vcpu, regs->r14);
8595 kvm_r15_write(vcpu, regs->r15);
8598 kvm_rip_write(vcpu, regs->rip);
8599 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8601 vcpu->arch.exception.pending = false;
8603 kvm_make_request(KVM_REQ_EVENT, vcpu);
8606 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8609 __set_regs(vcpu, regs);
8614 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8616 struct kvm_segment cs;
8618 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8622 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8624 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8628 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8629 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8630 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8631 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8632 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8633 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8635 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8636 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8638 kvm_x86_ops->get_idt(vcpu, &dt);
8639 sregs->idt.limit = dt.size;
8640 sregs->idt.base = dt.address;
8641 kvm_x86_ops->get_gdt(vcpu, &dt);
8642 sregs->gdt.limit = dt.size;
8643 sregs->gdt.base = dt.address;
8645 sregs->cr0 = kvm_read_cr0(vcpu);
8646 sregs->cr2 = vcpu->arch.cr2;
8647 sregs->cr3 = kvm_read_cr3(vcpu);
8648 sregs->cr4 = kvm_read_cr4(vcpu);
8649 sregs->cr8 = kvm_get_cr8(vcpu);
8650 sregs->efer = vcpu->arch.efer;
8651 sregs->apic_base = kvm_get_apic_base(vcpu);
8653 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8655 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8656 set_bit(vcpu->arch.interrupt.nr,
8657 (unsigned long *)sregs->interrupt_bitmap);
8660 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8661 struct kvm_sregs *sregs)
8664 __get_sregs(vcpu, sregs);
8669 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8670 struct kvm_mp_state *mp_state)
8674 kvm_apic_accept_events(vcpu);
8675 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8676 vcpu->arch.pv.pv_unhalted)
8677 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8679 mp_state->mp_state = vcpu->arch.mp_state;
8685 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8686 struct kvm_mp_state *mp_state)
8692 if (!lapic_in_kernel(vcpu) &&
8693 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8696 /* INITs are latched while in SMM */
8697 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8698 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8699 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8702 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8703 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8704 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8706 vcpu->arch.mp_state = mp_state->mp_state;
8707 kvm_make_request(KVM_REQ_EVENT, vcpu);
8715 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8716 int reason, bool has_error_code, u32 error_code)
8718 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8721 init_emulate_ctxt(vcpu);
8723 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8724 has_error_code, error_code);
8726 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8727 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
8728 vcpu->run->internal.ndata = 0;
8732 kvm_rip_write(vcpu, ctxt->eip);
8733 kvm_set_rflags(vcpu, ctxt->eflags);
8734 kvm_make_request(KVM_REQ_EVENT, vcpu);
8737 EXPORT_SYMBOL_GPL(kvm_task_switch);
8739 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8741 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8743 * When EFER.LME and CR0.PG are set, the processor is in
8744 * 64-bit mode (though maybe in a 32-bit code segment).
8745 * CR4.PAE and EFER.LMA must be set.
8747 if (!(sregs->cr4 & X86_CR4_PAE)
8748 || !(sregs->efer & EFER_LMA))
8752 * Not in 64-bit mode: EFER.LMA is clear and the code
8753 * segment cannot be 64-bit.
8755 if (sregs->efer & EFER_LMA || sregs->cs.l)
8759 return kvm_valid_cr4(vcpu, sregs->cr4);
8762 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8764 struct msr_data apic_base_msr;
8765 int mmu_reset_needed = 0;
8766 int cpuid_update_needed = 0;
8767 int pending_vec, max_bits, idx;
8771 if (kvm_valid_sregs(vcpu, sregs))
8774 apic_base_msr.data = sregs->apic_base;
8775 apic_base_msr.host_initiated = true;
8776 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8779 dt.size = sregs->idt.limit;
8780 dt.address = sregs->idt.base;
8781 kvm_x86_ops->set_idt(vcpu, &dt);
8782 dt.size = sregs->gdt.limit;
8783 dt.address = sregs->gdt.base;
8784 kvm_x86_ops->set_gdt(vcpu, &dt);
8786 vcpu->arch.cr2 = sregs->cr2;
8787 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8788 vcpu->arch.cr3 = sregs->cr3;
8789 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8791 kvm_set_cr8(vcpu, sregs->cr8);
8793 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8794 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8796 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8797 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8798 vcpu->arch.cr0 = sregs->cr0;
8800 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8801 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8802 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8803 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8804 if (cpuid_update_needed)
8805 kvm_update_cpuid(vcpu);
8807 idx = srcu_read_lock(&vcpu->kvm->srcu);
8808 if (is_pae_paging(vcpu)) {
8809 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8810 mmu_reset_needed = 1;
8812 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8814 if (mmu_reset_needed)
8815 kvm_mmu_reset_context(vcpu);
8817 max_bits = KVM_NR_INTERRUPTS;
8818 pending_vec = find_first_bit(
8819 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8820 if (pending_vec < max_bits) {
8821 kvm_queue_interrupt(vcpu, pending_vec, false);
8822 pr_debug("Set back pending irq %d\n", pending_vec);
8825 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8826 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8827 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8828 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8829 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8830 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8832 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8833 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8835 update_cr8_intercept(vcpu);
8837 /* Older userspace won't unhalt the vcpu on reset. */
8838 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8839 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8841 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8843 kvm_make_request(KVM_REQ_EVENT, vcpu);
8850 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8851 struct kvm_sregs *sregs)
8856 ret = __set_sregs(vcpu, sregs);
8861 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8862 struct kvm_guest_debug *dbg)
8864 unsigned long rflags;
8869 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8871 if (vcpu->arch.exception.pending)
8873 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8874 kvm_queue_exception(vcpu, DB_VECTOR);
8876 kvm_queue_exception(vcpu, BP_VECTOR);
8880 * Read rflags as long as potentially injected trace flags are still
8883 rflags = kvm_get_rflags(vcpu);
8885 vcpu->guest_debug = dbg->control;
8886 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8887 vcpu->guest_debug = 0;
8889 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8890 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8891 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8892 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8894 for (i = 0; i < KVM_NR_DB_REGS; i++)
8895 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8897 kvm_update_dr7(vcpu);
8899 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8900 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8901 get_segment_base(vcpu, VCPU_SREG_CS);
8904 * Trigger an rflags update that will inject or remove the trace
8907 kvm_set_rflags(vcpu, rflags);
8909 kvm_x86_ops->update_bp_intercept(vcpu);
8919 * Translate a guest virtual address to a guest physical address.
8921 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8922 struct kvm_translation *tr)
8924 unsigned long vaddr = tr->linear_address;
8930 idx = srcu_read_lock(&vcpu->kvm->srcu);
8931 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8932 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8933 tr->physical_address = gpa;
8934 tr->valid = gpa != UNMAPPED_GVA;
8942 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8944 struct fxregs_state *fxsave;
8948 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8949 memcpy(fpu->fpr, fxsave->st_space, 128);
8950 fpu->fcw = fxsave->cwd;
8951 fpu->fsw = fxsave->swd;
8952 fpu->ftwx = fxsave->twd;
8953 fpu->last_opcode = fxsave->fop;
8954 fpu->last_ip = fxsave->rip;
8955 fpu->last_dp = fxsave->rdp;
8956 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8962 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8964 struct fxregs_state *fxsave;
8968 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8970 memcpy(fxsave->st_space, fpu->fpr, 128);
8971 fxsave->cwd = fpu->fcw;
8972 fxsave->swd = fpu->fsw;
8973 fxsave->twd = fpu->ftwx;
8974 fxsave->fop = fpu->last_opcode;
8975 fxsave->rip = fpu->last_ip;
8976 fxsave->rdp = fpu->last_dp;
8977 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8983 static void store_regs(struct kvm_vcpu *vcpu)
8985 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8987 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8988 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8990 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8991 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8993 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8994 kvm_vcpu_ioctl_x86_get_vcpu_events(
8995 vcpu, &vcpu->run->s.regs.events);
8998 static int sync_regs(struct kvm_vcpu *vcpu)
9000 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
9003 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
9004 __set_regs(vcpu, &vcpu->run->s.regs.regs);
9005 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
9007 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
9008 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
9010 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
9012 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
9013 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
9014 vcpu, &vcpu->run->s.regs.events))
9016 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
9022 static void fx_init(struct kvm_vcpu *vcpu)
9024 fpstate_init(&vcpu->arch.guest_fpu->state);
9025 if (boot_cpu_has(X86_FEATURE_XSAVES))
9026 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
9027 host_xcr0 | XSTATE_COMPACTION_ENABLED;
9030 * Ensure guest xcr0 is valid for loading
9032 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9034 vcpu->arch.cr0 |= X86_CR0_ET;
9037 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
9039 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
9041 kvmclock_reset(vcpu);
9043 kvm_x86_ops->vcpu_free(vcpu);
9044 free_cpumask_var(wbinvd_dirty_mask);
9047 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
9050 struct kvm_vcpu *vcpu;
9052 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
9053 printk_once(KERN_WARNING
9054 "kvm: SMP vm created on host with unstable TSC; "
9055 "guest TSC will not be reliable\n");
9057 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
9062 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
9064 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
9065 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
9066 kvm_vcpu_mtrr_init(vcpu);
9068 kvm_vcpu_reset(vcpu, false);
9069 kvm_init_mmu(vcpu, false);
9074 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
9076 struct msr_data msr;
9077 struct kvm *kvm = vcpu->kvm;
9079 kvm_hv_vcpu_postcreate(vcpu);
9081 if (mutex_lock_killable(&vcpu->mutex))
9085 msr.index = MSR_IA32_TSC;
9086 msr.host_initiated = true;
9087 kvm_write_tsc(vcpu, &msr);
9090 /* poll control enabled by default */
9091 vcpu->arch.msr_kvm_poll_control = 1;
9093 mutex_unlock(&vcpu->mutex);
9095 if (!kvmclock_periodic_sync)
9098 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
9099 KVMCLOCK_SYNC_PERIOD);
9102 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
9104 vcpu->arch.apf.msr_val = 0;
9107 kvm_mmu_unload(vcpu);
9110 kvm_x86_ops->vcpu_free(vcpu);
9113 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
9115 kvm_lapic_reset(vcpu, init_event);
9117 vcpu->arch.hflags = 0;
9119 vcpu->arch.smi_pending = 0;
9120 vcpu->arch.smi_count = 0;
9121 atomic_set(&vcpu->arch.nmi_queued, 0);
9122 vcpu->arch.nmi_pending = 0;
9123 vcpu->arch.nmi_injected = false;
9124 kvm_clear_interrupt_queue(vcpu);
9125 kvm_clear_exception_queue(vcpu);
9126 vcpu->arch.exception.pending = false;
9128 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
9129 kvm_update_dr0123(vcpu);
9130 vcpu->arch.dr6 = DR6_INIT;
9131 kvm_update_dr6(vcpu);
9132 vcpu->arch.dr7 = DR7_FIXED_1;
9133 kvm_update_dr7(vcpu);
9137 kvm_make_request(KVM_REQ_EVENT, vcpu);
9138 vcpu->arch.apf.msr_val = 0;
9139 vcpu->arch.st.msr_val = 0;
9141 kvmclock_reset(vcpu);
9143 kvm_clear_async_pf_completion_queue(vcpu);
9144 kvm_async_pf_hash_reset(vcpu);
9145 vcpu->arch.apf.halted = false;
9147 if (kvm_mpx_supported()) {
9148 void *mpx_state_buffer;
9151 * To avoid have the INIT path from kvm_apic_has_events() that be
9152 * called with loaded FPU and does not let userspace fix the state.
9155 kvm_put_guest_fpu(vcpu);
9156 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9158 if (mpx_state_buffer)
9159 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
9160 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
9162 if (mpx_state_buffer)
9163 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
9165 kvm_load_guest_fpu(vcpu);
9169 kvm_pmu_reset(vcpu);
9170 vcpu->arch.smbase = 0x30000;
9172 vcpu->arch.msr_misc_features_enables = 0;
9174 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
9177 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
9178 vcpu->arch.regs_avail = ~0;
9179 vcpu->arch.regs_dirty = ~0;
9181 vcpu->arch.ia32_xss = 0;
9183 kvm_x86_ops->vcpu_reset(vcpu, init_event);
9186 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
9188 struct kvm_segment cs;
9190 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
9191 cs.selector = vector << 8;
9192 cs.base = vector << 12;
9193 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
9194 kvm_rip_write(vcpu, 0);
9197 int kvm_arch_hardware_enable(void)
9200 struct kvm_vcpu *vcpu;
9205 bool stable, backwards_tsc = false;
9207 kvm_shared_msr_cpu_online();
9208 ret = kvm_x86_ops->hardware_enable();
9212 local_tsc = rdtsc();
9213 stable = !kvm_check_tsc_unstable();
9214 list_for_each_entry(kvm, &vm_list, vm_list) {
9215 kvm_for_each_vcpu(i, vcpu, kvm) {
9216 if (!stable && vcpu->cpu == smp_processor_id())
9217 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9218 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9219 backwards_tsc = true;
9220 if (vcpu->arch.last_host_tsc > max_tsc)
9221 max_tsc = vcpu->arch.last_host_tsc;
9227 * Sometimes, even reliable TSCs go backwards. This happens on
9228 * platforms that reset TSC during suspend or hibernate actions, but
9229 * maintain synchronization. We must compensate. Fortunately, we can
9230 * detect that condition here, which happens early in CPU bringup,
9231 * before any KVM threads can be running. Unfortunately, we can't
9232 * bring the TSCs fully up to date with real time, as we aren't yet far
9233 * enough into CPU bringup that we know how much real time has actually
9234 * elapsed; our helper function, ktime_get_boottime_ns() will be using boot
9235 * variables that haven't been updated yet.
9237 * So we simply find the maximum observed TSC above, then record the
9238 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9239 * the adjustment will be applied. Note that we accumulate
9240 * adjustments, in case multiple suspend cycles happen before some VCPU
9241 * gets a chance to run again. In the event that no KVM threads get a
9242 * chance to run, we will miss the entire elapsed period, as we'll have
9243 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9244 * loose cycle time. This isn't too big a deal, since the loss will be
9245 * uniform across all VCPUs (not to mention the scenario is extremely
9246 * unlikely). It is possible that a second hibernate recovery happens
9247 * much faster than a first, causing the observed TSC here to be
9248 * smaller; this would require additional padding adjustment, which is
9249 * why we set last_host_tsc to the local tsc observed here.
9251 * N.B. - this code below runs only on platforms with reliable TSC,
9252 * as that is the only way backwards_tsc is set above. Also note
9253 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9254 * have the same delta_cyc adjustment applied if backwards_tsc
9255 * is detected. Note further, this adjustment is only done once,
9256 * as we reset last_host_tsc on all VCPUs to stop this from being
9257 * called multiple times (one for each physical CPU bringup).
9259 * Platforms with unreliable TSCs don't have to deal with this, they
9260 * will be compensated by the logic in vcpu_load, which sets the TSC to
9261 * catchup mode. This will catchup all VCPUs to real time, but cannot
9262 * guarantee that they stay in perfect synchronization.
9264 if (backwards_tsc) {
9265 u64 delta_cyc = max_tsc - local_tsc;
9266 list_for_each_entry(kvm, &vm_list, vm_list) {
9267 kvm->arch.backwards_tsc_observed = true;
9268 kvm_for_each_vcpu(i, vcpu, kvm) {
9269 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9270 vcpu->arch.last_host_tsc = local_tsc;
9271 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9275 * We have to disable TSC offset matching.. if you were
9276 * booting a VM while issuing an S4 host suspend....
9277 * you may have some problem. Solving this issue is
9278 * left as an exercise to the reader.
9280 kvm->arch.last_tsc_nsec = 0;
9281 kvm->arch.last_tsc_write = 0;
9288 void kvm_arch_hardware_disable(void)
9290 kvm_x86_ops->hardware_disable();
9291 drop_user_return_notifiers();
9294 int kvm_arch_hardware_setup(void)
9298 r = kvm_x86_ops->hardware_setup();
9302 if (kvm_has_tsc_control) {
9304 * Make sure the user can only configure tsc_khz values that
9305 * fit into a signed integer.
9306 * A min value is not calculated because it will always
9307 * be 1 on all machines.
9309 u64 max = min(0x7fffffffULL,
9310 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9311 kvm_max_guest_tsc_khz = max;
9313 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9316 kvm_init_msr_list();
9320 void kvm_arch_hardware_unsetup(void)
9322 kvm_x86_ops->hardware_unsetup();
9325 int kvm_arch_check_processor_compat(void)
9327 return kvm_x86_ops->check_processor_compatibility();
9330 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9332 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9334 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9336 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9338 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9341 struct static_key kvm_no_apic_vcpu __read_mostly;
9342 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9344 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9349 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9350 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9351 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9353 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9355 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9360 vcpu->arch.pio_data = page_address(page);
9362 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9364 r = kvm_mmu_create(vcpu);
9366 goto fail_free_pio_data;
9368 if (irqchip_in_kernel(vcpu->kvm)) {
9369 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9370 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9372 goto fail_mmu_destroy;
9374 static_key_slow_inc(&kvm_no_apic_vcpu);
9376 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9377 GFP_KERNEL_ACCOUNT);
9378 if (!vcpu->arch.mce_banks) {
9380 goto fail_free_lapic;
9382 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9384 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9385 GFP_KERNEL_ACCOUNT)) {
9387 goto fail_free_mce_banks;
9392 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9394 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9396 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9398 kvm_async_pf_hash_reset(vcpu);
9401 vcpu->arch.pending_external_vector = -1;
9402 vcpu->arch.preempted_in_kernel = false;
9404 kvm_hv_vcpu_init(vcpu);
9408 fail_free_mce_banks:
9409 kfree(vcpu->arch.mce_banks);
9411 kvm_free_lapic(vcpu);
9413 kvm_mmu_destroy(vcpu);
9415 free_page((unsigned long)vcpu->arch.pio_data);
9420 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9424 kvm_hv_vcpu_uninit(vcpu);
9425 kvm_pmu_destroy(vcpu);
9426 kfree(vcpu->arch.mce_banks);
9427 kvm_free_lapic(vcpu);
9428 idx = srcu_read_lock(&vcpu->kvm->srcu);
9429 kvm_mmu_destroy(vcpu);
9430 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9431 free_page((unsigned long)vcpu->arch.pio_data);
9432 if (!lapic_in_kernel(vcpu))
9433 static_key_slow_dec(&kvm_no_apic_vcpu);
9436 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9438 vcpu->arch.l1tf_flush_l1d = true;
9439 kvm_x86_ops->sched_in(vcpu, cpu);
9442 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9447 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9448 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9449 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9450 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9451 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9453 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9454 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9455 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9456 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9457 &kvm->arch.irq_sources_bitmap);
9459 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9460 mutex_init(&kvm->arch.apic_map_lock);
9461 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9463 kvm->arch.kvmclock_offset = -ktime_get_boottime_ns();
9464 pvclock_update_vm_gtod_copy(kvm);
9466 kvm->arch.guest_can_read_msr_platform_info = true;
9468 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9469 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9471 kvm_hv_init_vm(kvm);
9472 kvm_page_track_init(kvm);
9473 kvm_mmu_init_vm(kvm);
9475 return kvm_x86_ops->vm_init(kvm);
9478 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9481 kvm_mmu_unload(vcpu);
9485 static void kvm_free_vcpus(struct kvm *kvm)
9488 struct kvm_vcpu *vcpu;
9491 * Unpin any mmu pages first.
9493 kvm_for_each_vcpu(i, vcpu, kvm) {
9494 kvm_clear_async_pf_completion_queue(vcpu);
9495 kvm_unload_vcpu_mmu(vcpu);
9497 kvm_for_each_vcpu(i, vcpu, kvm)
9498 kvm_arch_vcpu_free(vcpu);
9500 mutex_lock(&kvm->lock);
9501 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9502 kvm->vcpus[i] = NULL;
9504 atomic_set(&kvm->online_vcpus, 0);
9505 mutex_unlock(&kvm->lock);
9508 void kvm_arch_sync_events(struct kvm *kvm)
9510 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9511 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9515 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9519 struct kvm_memslots *slots = kvm_memslots(kvm);
9520 struct kvm_memory_slot *slot, old;
9522 /* Called with kvm->slots_lock held. */
9523 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9526 slot = id_to_memslot(slots, id);
9532 * MAP_SHARED to prevent internal slot pages from being moved
9535 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9536 MAP_SHARED | MAP_ANONYMOUS, 0);
9537 if (IS_ERR((void *)hva))
9538 return PTR_ERR((void *)hva);
9547 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9548 struct kvm_userspace_memory_region m;
9550 m.slot = id | (i << 16);
9552 m.guest_phys_addr = gpa;
9553 m.userspace_addr = hva;
9554 m.memory_size = size;
9555 r = __kvm_set_memory_region(kvm, &m);
9561 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9565 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9567 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9571 mutex_lock(&kvm->slots_lock);
9572 r = __x86_set_memory_region(kvm, id, gpa, size);
9573 mutex_unlock(&kvm->slots_lock);
9577 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9579 void kvm_arch_destroy_vm(struct kvm *kvm)
9581 if (current->mm == kvm->mm) {
9583 * Free memory regions allocated on behalf of userspace,
9584 * unless the the memory map has changed due to process exit
9587 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9588 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9589 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9591 if (kvm_x86_ops->vm_destroy)
9592 kvm_x86_ops->vm_destroy(kvm);
9593 kvm_pic_destroy(kvm);
9594 kvm_ioapic_destroy(kvm);
9595 kvm_free_vcpus(kvm);
9596 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9597 kfree(srcu_dereference_check(kvm->arch.pmu_event_filter, &kvm->srcu, 1));
9598 kvm_mmu_uninit_vm(kvm);
9599 kvm_page_track_cleanup(kvm);
9600 kvm_hv_destroy_vm(kvm);
9603 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9604 struct kvm_memory_slot *dont)
9608 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9609 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9610 kvfree(free->arch.rmap[i]);
9611 free->arch.rmap[i] = NULL;
9616 if (!dont || free->arch.lpage_info[i - 1] !=
9617 dont->arch.lpage_info[i - 1]) {
9618 kvfree(free->arch.lpage_info[i - 1]);
9619 free->arch.lpage_info[i - 1] = NULL;
9623 kvm_page_track_free_memslot(free, dont);
9626 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9627 unsigned long npages)
9631 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9632 struct kvm_lpage_info *linfo;
9637 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9638 slot->base_gfn, level) + 1;
9640 slot->arch.rmap[i] =
9641 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9642 GFP_KERNEL_ACCOUNT);
9643 if (!slot->arch.rmap[i])
9648 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9652 slot->arch.lpage_info[i - 1] = linfo;
9654 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9655 linfo[0].disallow_lpage = 1;
9656 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9657 linfo[lpages - 1].disallow_lpage = 1;
9658 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9660 * If the gfn and userspace address are not aligned wrt each
9661 * other, or if explicitly asked to, disable large page
9662 * support for this slot
9664 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9665 !kvm_largepages_enabled()) {
9668 for (j = 0; j < lpages; ++j)
9669 linfo[j].disallow_lpage = 1;
9673 if (kvm_page_track_create_memslot(slot, npages))
9679 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9680 kvfree(slot->arch.rmap[i]);
9681 slot->arch.rmap[i] = NULL;
9685 kvfree(slot->arch.lpage_info[i - 1]);
9686 slot->arch.lpage_info[i - 1] = NULL;
9691 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9694 * memslots->generation has been incremented.
9695 * mmio generation may have reached its maximum value.
9697 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9700 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9701 struct kvm_memory_slot *memslot,
9702 const struct kvm_userspace_memory_region *mem,
9703 enum kvm_mr_change change)
9708 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9709 struct kvm_memory_slot *new)
9711 /* Still write protect RO slot */
9712 if (new->flags & KVM_MEM_READONLY) {
9713 kvm_mmu_slot_remove_write_access(kvm, new);
9718 * Call kvm_x86_ops dirty logging hooks when they are valid.
9720 * kvm_x86_ops->slot_disable_log_dirty is called when:
9722 * - KVM_MR_CREATE with dirty logging is disabled
9723 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9725 * The reason is, in case of PML, we need to set D-bit for any slots
9726 * with dirty logging disabled in order to eliminate unnecessary GPA
9727 * logging in PML buffer (and potential PML buffer full VMEXT). This
9728 * guarantees leaving PML enabled during guest's lifetime won't have
9729 * any additional overhead from PML when guest is running with dirty
9730 * logging disabled for memory slots.
9732 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9733 * to dirty logging mode.
9735 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9737 * In case of write protect:
9739 * Write protect all pages for dirty logging.
9741 * All the sptes including the large sptes which point to this
9742 * slot are set to readonly. We can not create any new large
9743 * spte on this slot until the end of the logging.
9745 * See the comments in fast_page_fault().
9747 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9748 if (kvm_x86_ops->slot_enable_log_dirty)
9749 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9751 kvm_mmu_slot_remove_write_access(kvm, new);
9753 if (kvm_x86_ops->slot_disable_log_dirty)
9754 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9758 void kvm_arch_commit_memory_region(struct kvm *kvm,
9759 const struct kvm_userspace_memory_region *mem,
9760 const struct kvm_memory_slot *old,
9761 const struct kvm_memory_slot *new,
9762 enum kvm_mr_change change)
9764 if (!kvm->arch.n_requested_mmu_pages)
9765 kvm_mmu_change_mmu_pages(kvm,
9766 kvm_mmu_calculate_default_mmu_pages(kvm));
9769 * Dirty logging tracks sptes in 4k granularity, meaning that large
9770 * sptes have to be split. If live migration is successful, the guest
9771 * in the source machine will be destroyed and large sptes will be
9772 * created in the destination. However, if the guest continues to run
9773 * in the source machine (for example if live migration fails), small
9774 * sptes will remain around and cause bad performance.
9776 * Scan sptes if dirty logging has been stopped, dropping those
9777 * which can be collapsed into a single large-page spte. Later
9778 * page faults will create the large-page sptes.
9780 * There is no need to do this in any of the following cases:
9781 * CREATE: No dirty mappings will already exist.
9782 * MOVE/DELETE: The old mappings will already have been cleaned up by
9783 * kvm_arch_flush_shadow_memslot()
9785 if (change == KVM_MR_FLAGS_ONLY &&
9786 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9787 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9788 kvm_mmu_zap_collapsible_sptes(kvm, new);
9791 * Set up write protection and/or dirty logging for the new slot.
9793 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9794 * been zapped so no dirty logging staff is needed for old slot. For
9795 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9796 * new and it's also covered when dealing with the new slot.
9798 * FIXME: const-ify all uses of struct kvm_memory_slot.
9800 if (change != KVM_MR_DELETE)
9801 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9804 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9806 kvm_mmu_zap_all(kvm);
9809 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9810 struct kvm_memory_slot *slot)
9812 kvm_page_track_flush_slot(kvm, slot);
9815 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9817 return (is_guest_mode(vcpu) &&
9818 kvm_x86_ops->guest_apic_has_interrupt &&
9819 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9822 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9824 if (!list_empty_careful(&vcpu->async_pf.done))
9827 if (kvm_apic_has_events(vcpu))
9830 if (vcpu->arch.pv.pv_unhalted)
9833 if (vcpu->arch.exception.pending)
9836 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9837 (vcpu->arch.nmi_pending &&
9838 kvm_x86_ops->nmi_allowed(vcpu)))
9841 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9842 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9845 if (kvm_arch_interrupt_allowed(vcpu) &&
9846 (kvm_cpu_has_interrupt(vcpu) ||
9847 kvm_guest_apic_has_interrupt(vcpu)))
9850 if (kvm_hv_has_stimer_pending(vcpu))
9856 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9858 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9861 bool kvm_arch_dy_runnable(struct kvm_vcpu *vcpu)
9863 if (READ_ONCE(vcpu->arch.pv.pv_unhalted))
9866 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9867 kvm_test_request(KVM_REQ_SMI, vcpu) ||
9868 kvm_test_request(KVM_REQ_EVENT, vcpu))
9871 if (vcpu->arch.apicv_active && kvm_x86_ops->dy_apicv_has_pending_interrupt(vcpu))
9877 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9879 return vcpu->arch.preempted_in_kernel;
9882 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9884 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9887 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9889 return kvm_x86_ops->interrupt_allowed(vcpu);
9892 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9894 if (is_64_bit_mode(vcpu))
9895 return kvm_rip_read(vcpu);
9896 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9897 kvm_rip_read(vcpu));
9899 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9901 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9903 return kvm_get_linear_rip(vcpu) == linear_rip;
9905 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9907 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9909 unsigned long rflags;
9911 rflags = kvm_x86_ops->get_rflags(vcpu);
9912 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9913 rflags &= ~X86_EFLAGS_TF;
9916 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9918 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9920 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9921 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9922 rflags |= X86_EFLAGS_TF;
9923 kvm_x86_ops->set_rflags(vcpu, rflags);
9926 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9928 __kvm_set_rflags(vcpu, rflags);
9929 kvm_make_request(KVM_REQ_EVENT, vcpu);
9931 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9933 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9937 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9941 r = kvm_mmu_reload(vcpu);
9945 if (!vcpu->arch.mmu->direct_map &&
9946 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9949 vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9952 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9954 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9957 static inline u32 kvm_async_pf_next_probe(u32 key)
9959 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9962 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9964 u32 key = kvm_async_pf_hash_fn(gfn);
9966 while (vcpu->arch.apf.gfns[key] != ~0)
9967 key = kvm_async_pf_next_probe(key);
9969 vcpu->arch.apf.gfns[key] = gfn;
9972 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9975 u32 key = kvm_async_pf_hash_fn(gfn);
9977 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9978 (vcpu->arch.apf.gfns[key] != gfn &&
9979 vcpu->arch.apf.gfns[key] != ~0); i++)
9980 key = kvm_async_pf_next_probe(key);
9985 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9987 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9990 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9994 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9996 vcpu->arch.apf.gfns[i] = ~0;
9998 j = kvm_async_pf_next_probe(j);
9999 if (vcpu->arch.apf.gfns[j] == ~0)
10001 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
10003 * k lies cyclically in ]i,j]
10005 * |....j i.k.| or |.k..j i...|
10007 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
10008 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
10013 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
10016 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
10020 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
10023 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
10027 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
10029 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
10032 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
10033 (vcpu->arch.apf.send_user_only &&
10034 kvm_x86_ops->get_cpl(vcpu) == 0))
10040 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
10042 if (unlikely(!lapic_in_kernel(vcpu) ||
10043 kvm_event_needs_reinjection(vcpu) ||
10044 vcpu->arch.exception.pending))
10047 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
10051 * If interrupts are off we cannot even use an artificial
10054 return kvm_x86_ops->interrupt_allowed(vcpu);
10057 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
10058 struct kvm_async_pf *work)
10060 struct x86_exception fault;
10062 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
10063 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
10065 if (kvm_can_deliver_async_pf(vcpu) &&
10066 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
10067 fault.vector = PF_VECTOR;
10068 fault.error_code_valid = true;
10069 fault.error_code = 0;
10070 fault.nested_page_fault = false;
10071 fault.address = work->arch.token;
10072 fault.async_page_fault = true;
10073 kvm_inject_page_fault(vcpu, &fault);
10076 * It is not possible to deliver a paravirtualized asynchronous
10077 * page fault, but putting the guest in an artificial halt state
10078 * can be beneficial nevertheless: if an interrupt arrives, we
10079 * can deliver it timely and perhaps the guest will schedule
10080 * another process. When the instruction that triggered a page
10081 * fault is retried, hopefully the page will be ready in the host.
10083 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
10087 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
10088 struct kvm_async_pf *work)
10090 struct x86_exception fault;
10093 if (work->wakeup_all)
10094 work->arch.token = ~0; /* broadcast wakeup */
10096 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
10097 trace_kvm_async_pf_ready(work->arch.token, work->gva);
10099 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
10100 !apf_get_user(vcpu, &val)) {
10101 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
10102 vcpu->arch.exception.pending &&
10103 vcpu->arch.exception.nr == PF_VECTOR &&
10104 !apf_put_user(vcpu, 0)) {
10105 vcpu->arch.exception.injected = false;
10106 vcpu->arch.exception.pending = false;
10107 vcpu->arch.exception.nr = 0;
10108 vcpu->arch.exception.has_error_code = false;
10109 vcpu->arch.exception.error_code = 0;
10110 vcpu->arch.exception.has_payload = false;
10111 vcpu->arch.exception.payload = 0;
10112 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
10113 fault.vector = PF_VECTOR;
10114 fault.error_code_valid = true;
10115 fault.error_code = 0;
10116 fault.nested_page_fault = false;
10117 fault.address = work->arch.token;
10118 fault.async_page_fault = true;
10119 kvm_inject_page_fault(vcpu, &fault);
10122 vcpu->arch.apf.halted = false;
10123 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10126 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
10128 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
10131 return kvm_can_do_async_pf(vcpu);
10134 void kvm_arch_start_assignment(struct kvm *kvm)
10136 atomic_inc(&kvm->arch.assigned_device_count);
10138 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
10140 void kvm_arch_end_assignment(struct kvm *kvm)
10142 atomic_dec(&kvm->arch.assigned_device_count);
10144 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
10146 bool kvm_arch_has_assigned_device(struct kvm *kvm)
10148 return atomic_read(&kvm->arch.assigned_device_count);
10150 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
10152 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
10154 atomic_inc(&kvm->arch.noncoherent_dma_count);
10156 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
10158 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
10160 atomic_dec(&kvm->arch.noncoherent_dma_count);
10162 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
10164 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
10166 return atomic_read(&kvm->arch.noncoherent_dma_count);
10168 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
10170 bool kvm_arch_has_irq_bypass(void)
10175 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
10176 struct irq_bypass_producer *prod)
10178 struct kvm_kernel_irqfd *irqfd =
10179 container_of(cons, struct kvm_kernel_irqfd, consumer);
10181 irqfd->producer = prod;
10183 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
10184 prod->irq, irqfd->gsi, 1);
10187 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
10188 struct irq_bypass_producer *prod)
10191 struct kvm_kernel_irqfd *irqfd =
10192 container_of(cons, struct kvm_kernel_irqfd, consumer);
10194 WARN_ON(irqfd->producer != prod);
10195 irqfd->producer = NULL;
10198 * When producer of consumer is unregistered, we change back to
10199 * remapped mode, so we can re-use the current implementation
10200 * when the irq is masked/disabled or the consumer side (KVM
10201 * int this case doesn't want to receive the interrupts.
10203 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
10205 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
10206 " fails: %d\n", irqfd->consumer.token, ret);
10209 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
10210 uint32_t guest_irq, bool set)
10212 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
10215 bool kvm_vector_hashing_enabled(void)
10217 return vector_hashing;
10219 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
10221 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
10223 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
10225 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10228 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10229 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10230 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10231 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10232 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10233 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10234 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10235 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10236 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10237 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10238 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmenter_failed);
10239 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10240 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10241 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10242 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10243 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window_update);
10244 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10245 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10246 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10247 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);