2 * Kernel-based Virtual Machine driver for Linux
4 * derived from drivers/kvm/kvm_main.c
6 * Copyright (C) 2006 Qumranet, Inc.
7 * Copyright (C) 2008 Qumranet, Inc.
8 * Copyright IBM Corporation, 2008
9 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
12 * Avi Kivity <avi@qumranet.com>
13 * Yaniv Kamay <yaniv@qumranet.com>
14 * Amit Shah <amit.shah@qumranet.com>
15 * Ben-Ami Yassour <benami@il.ibm.com>
17 * This work is licensed under the terms of the GNU GPL, version 2. See
18 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
27 #include "kvm_cache_regs.h"
33 #include <linux/clocksource.h>
34 #include <linux/interrupt.h>
35 #include <linux/kvm.h>
37 #include <linux/vmalloc.h>
38 #include <linux/export.h>
39 #include <linux/moduleparam.h>
40 #include <linux/mman.h>
41 #include <linux/highmem.h>
42 #include <linux/iommu.h>
43 #include <linux/intel-iommu.h>
44 #include <linux/cpufreq.h>
45 #include <linux/user-return-notifier.h>
46 #include <linux/srcu.h>
47 #include <linux/slab.h>
48 #include <linux/perf_event.h>
49 #include <linux/uaccess.h>
50 #include <linux/hash.h>
51 #include <linux/pci.h>
52 #include <linux/timekeeper_internal.h>
53 #include <linux/pvclock_gtod.h>
54 #include <linux/kvm_irqfd.h>
55 #include <linux/irqbypass.h>
56 #include <linux/sched/stat.h>
57 #include <linux/mem_encrypt.h>
59 #include <trace/events/kvm.h>
61 #include <asm/debugreg.h>
65 #include <linux/kernel_stat.h>
66 #include <asm/fpu/internal.h> /* Ugh! */
67 #include <asm/pvclock.h>
68 #include <asm/div64.h>
69 #include <asm/irq_remapping.h>
70 #include <asm/mshyperv.h>
71 #include <asm/hypervisor.h>
72 #include <asm/intel_pt.h>
74 #define CREATE_TRACE_POINTS
77 #define MAX_IO_MSRS 256
78 #define KVM_MAX_MCE_BANKS 32
79 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
80 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
82 #define emul_to_vcpu(ctxt) \
83 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
86 * - enable syscall per default because its emulated by KVM
87 * - enable LME and LMA per default on 64 bit KVM
91 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
93 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
96 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
97 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
99 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
100 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
102 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
103 static void process_nmi(struct kvm_vcpu *vcpu);
104 static void enter_smm(struct kvm_vcpu *vcpu);
105 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
106 static void store_regs(struct kvm_vcpu *vcpu);
107 static int sync_regs(struct kvm_vcpu *vcpu);
109 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
110 EXPORT_SYMBOL_GPL(kvm_x86_ops);
112 static bool __read_mostly ignore_msrs = 0;
113 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
115 static bool __read_mostly report_ignored_msrs = true;
116 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
118 unsigned int min_timer_period_us = 200;
119 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
121 static bool __read_mostly kvmclock_periodic_sync = true;
122 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
124 bool __read_mostly kvm_has_tsc_control;
125 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
126 u32 __read_mostly kvm_max_guest_tsc_khz;
127 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
128 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
129 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
130 u64 __read_mostly kvm_max_tsc_scaling_ratio;
131 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
132 u64 __read_mostly kvm_default_tsc_scaling_ratio;
133 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
135 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
136 static u32 __read_mostly tsc_tolerance_ppm = 250;
137 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
140 * lapic timer advance (tscdeadline mode only) in nanoseconds. '-1' enables
141 * adaptive tuning starting from default advancment of 1000ns. '0' disables
142 * advancement entirely. Any other value is used as-is and disables adaptive
143 * tuning, i.e. allows priveleged userspace to set an exact advancement time.
145 static int __read_mostly lapic_timer_advance_ns = -1;
146 module_param(lapic_timer_advance_ns, int, S_IRUGO | S_IWUSR);
148 static bool __read_mostly vector_hashing = true;
149 module_param(vector_hashing, bool, S_IRUGO);
151 bool __read_mostly enable_vmware_backdoor = false;
152 module_param(enable_vmware_backdoor, bool, S_IRUGO);
153 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
155 static bool __read_mostly force_emulation_prefix = false;
156 module_param(force_emulation_prefix, bool, S_IRUGO);
158 #define KVM_NR_SHARED_MSRS 16
160 struct kvm_shared_msrs_global {
162 u32 msrs[KVM_NR_SHARED_MSRS];
165 struct kvm_shared_msrs {
166 struct user_return_notifier urn;
168 struct kvm_shared_msr_values {
171 } values[KVM_NR_SHARED_MSRS];
174 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
175 static struct kvm_shared_msrs __percpu *shared_msrs;
177 struct kvm_stats_debugfs_item debugfs_entries[] = {
178 { "pf_fixed", VCPU_STAT(pf_fixed) },
179 { "pf_guest", VCPU_STAT(pf_guest) },
180 { "tlb_flush", VCPU_STAT(tlb_flush) },
181 { "invlpg", VCPU_STAT(invlpg) },
182 { "exits", VCPU_STAT(exits) },
183 { "io_exits", VCPU_STAT(io_exits) },
184 { "mmio_exits", VCPU_STAT(mmio_exits) },
185 { "signal_exits", VCPU_STAT(signal_exits) },
186 { "irq_window", VCPU_STAT(irq_window_exits) },
187 { "nmi_window", VCPU_STAT(nmi_window_exits) },
188 { "halt_exits", VCPU_STAT(halt_exits) },
189 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
190 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
191 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
192 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
193 { "hypercalls", VCPU_STAT(hypercalls) },
194 { "request_irq", VCPU_STAT(request_irq_exits) },
195 { "irq_exits", VCPU_STAT(irq_exits) },
196 { "host_state_reload", VCPU_STAT(host_state_reload) },
197 { "fpu_reload", VCPU_STAT(fpu_reload) },
198 { "insn_emulation", VCPU_STAT(insn_emulation) },
199 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
200 { "irq_injections", VCPU_STAT(irq_injections) },
201 { "nmi_injections", VCPU_STAT(nmi_injections) },
202 { "req_event", VCPU_STAT(req_event) },
203 { "l1d_flush", VCPU_STAT(l1d_flush) },
204 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
205 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
206 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
207 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
208 { "mmu_flooded", VM_STAT(mmu_flooded) },
209 { "mmu_recycled", VM_STAT(mmu_recycled) },
210 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
211 { "mmu_unsync", VM_STAT(mmu_unsync) },
212 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
213 { "largepages", VM_STAT(lpages) },
214 { "max_mmu_page_hash_collisions",
215 VM_STAT(max_mmu_page_hash_collisions) },
219 u64 __read_mostly host_xcr0;
221 struct kmem_cache *x86_fpu_cache;
222 EXPORT_SYMBOL_GPL(x86_fpu_cache);
224 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
226 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
229 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
230 vcpu->arch.apf.gfns[i] = ~0;
233 static void kvm_on_user_return(struct user_return_notifier *urn)
236 struct kvm_shared_msrs *locals
237 = container_of(urn, struct kvm_shared_msrs, urn);
238 struct kvm_shared_msr_values *values;
242 * Disabling irqs at this point since the following code could be
243 * interrupted and executed through kvm_arch_hardware_disable()
245 local_irq_save(flags);
246 if (locals->registered) {
247 locals->registered = false;
248 user_return_notifier_unregister(urn);
250 local_irq_restore(flags);
251 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
252 values = &locals->values[slot];
253 if (values->host != values->curr) {
254 wrmsrl(shared_msrs_global.msrs[slot], values->host);
255 values->curr = values->host;
260 static void shared_msr_update(unsigned slot, u32 msr)
263 unsigned int cpu = smp_processor_id();
264 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
266 /* only read, and nobody should modify it at this time,
267 * so don't need lock */
268 if (slot >= shared_msrs_global.nr) {
269 printk(KERN_ERR "kvm: invalid MSR slot!");
272 rdmsrl_safe(msr, &value);
273 smsr->values[slot].host = value;
274 smsr->values[slot].curr = value;
277 void kvm_define_shared_msr(unsigned slot, u32 msr)
279 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
280 shared_msrs_global.msrs[slot] = msr;
281 if (slot >= shared_msrs_global.nr)
282 shared_msrs_global.nr = slot + 1;
284 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
286 static void kvm_shared_msr_cpu_online(void)
290 for (i = 0; i < shared_msrs_global.nr; ++i)
291 shared_msr_update(i, shared_msrs_global.msrs[i]);
294 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
296 unsigned int cpu = smp_processor_id();
297 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
300 if (((value ^ smsr->values[slot].curr) & mask) == 0)
302 smsr->values[slot].curr = value;
303 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
307 if (!smsr->registered) {
308 smsr->urn.on_user_return = kvm_on_user_return;
309 user_return_notifier_register(&smsr->urn);
310 smsr->registered = true;
314 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
316 static void drop_user_return_notifiers(void)
318 unsigned int cpu = smp_processor_id();
319 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
321 if (smsr->registered)
322 kvm_on_user_return(&smsr->urn);
325 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
327 return vcpu->arch.apic_base;
329 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
331 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
333 return kvm_apic_mode(kvm_get_apic_base(vcpu));
335 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
337 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
339 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
340 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
341 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
342 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
344 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
346 if (!msr_info->host_initiated) {
347 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
349 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
353 kvm_lapic_set_base(vcpu, msr_info->data);
356 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
358 asmlinkage __visible void kvm_spurious_fault(void)
360 /* Fault while not rebooting. We want the trace. */
363 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
365 #define EXCPT_BENIGN 0
366 #define EXCPT_CONTRIBUTORY 1
369 static int exception_class(int vector)
379 return EXCPT_CONTRIBUTORY;
386 #define EXCPT_FAULT 0
388 #define EXCPT_ABORT 2
389 #define EXCPT_INTERRUPT 3
391 static int exception_type(int vector)
395 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
396 return EXCPT_INTERRUPT;
400 /* #DB is trap, as instruction watchpoints are handled elsewhere */
401 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
404 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
407 /* Reserved exceptions will result in fault */
411 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
413 unsigned nr = vcpu->arch.exception.nr;
414 bool has_payload = vcpu->arch.exception.has_payload;
415 unsigned long payload = vcpu->arch.exception.payload;
423 * "Certain debug exceptions may clear bit 0-3. The
424 * remaining contents of the DR6 register are never
425 * cleared by the processor".
427 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
429 * DR6.RTM is set by all #DB exceptions that don't clear it.
431 vcpu->arch.dr6 |= DR6_RTM;
432 vcpu->arch.dr6 |= payload;
434 * Bit 16 should be set in the payload whenever the #DB
435 * exception should clear DR6.RTM. This makes the payload
436 * compatible with the pending debug exceptions under VMX.
437 * Though not currently documented in the SDM, this also
438 * makes the payload compatible with the exit qualification
439 * for #DB exceptions under VMX.
441 vcpu->arch.dr6 ^= payload & DR6_RTM;
444 vcpu->arch.cr2 = payload;
448 vcpu->arch.exception.has_payload = false;
449 vcpu->arch.exception.payload = 0;
451 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
453 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
454 unsigned nr, bool has_error, u32 error_code,
455 bool has_payload, unsigned long payload, bool reinject)
460 kvm_make_request(KVM_REQ_EVENT, vcpu);
462 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
464 if (has_error && !is_protmode(vcpu))
468 * On vmentry, vcpu->arch.exception.pending is only
469 * true if an event injection was blocked by
470 * nested_run_pending. In that case, however,
471 * vcpu_enter_guest requests an immediate exit,
472 * and the guest shouldn't proceed far enough to
475 WARN_ON_ONCE(vcpu->arch.exception.pending);
476 vcpu->arch.exception.injected = true;
477 if (WARN_ON_ONCE(has_payload)) {
479 * A reinjected event has already
480 * delivered its payload.
486 vcpu->arch.exception.pending = true;
487 vcpu->arch.exception.injected = false;
489 vcpu->arch.exception.has_error_code = has_error;
490 vcpu->arch.exception.nr = nr;
491 vcpu->arch.exception.error_code = error_code;
492 vcpu->arch.exception.has_payload = has_payload;
493 vcpu->arch.exception.payload = payload;
495 * In guest mode, payload delivery should be deferred,
496 * so that the L1 hypervisor can intercept #PF before
497 * CR2 is modified (or intercept #DB before DR6 is
498 * modified under nVMX). However, for ABI
499 * compatibility with KVM_GET_VCPU_EVENTS and
500 * KVM_SET_VCPU_EVENTS, we can't delay payload
501 * delivery unless userspace has enabled this
502 * functionality via the per-VM capability,
503 * KVM_CAP_EXCEPTION_PAYLOAD.
505 if (!vcpu->kvm->arch.exception_payload_enabled ||
506 !is_guest_mode(vcpu))
507 kvm_deliver_exception_payload(vcpu);
511 /* to check exception */
512 prev_nr = vcpu->arch.exception.nr;
513 if (prev_nr == DF_VECTOR) {
514 /* triple fault -> shutdown */
515 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
518 class1 = exception_class(prev_nr);
519 class2 = exception_class(nr);
520 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
521 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
523 * Generate double fault per SDM Table 5-5. Set
524 * exception.pending = true so that the double fault
525 * can trigger a nested vmexit.
527 vcpu->arch.exception.pending = true;
528 vcpu->arch.exception.injected = false;
529 vcpu->arch.exception.has_error_code = true;
530 vcpu->arch.exception.nr = DF_VECTOR;
531 vcpu->arch.exception.error_code = 0;
532 vcpu->arch.exception.has_payload = false;
533 vcpu->arch.exception.payload = 0;
535 /* replace previous exception with a new one in a hope
536 that instruction re-execution will regenerate lost
541 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
543 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
545 EXPORT_SYMBOL_GPL(kvm_queue_exception);
547 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
549 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
551 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
553 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
554 unsigned long payload)
556 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
559 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
560 u32 error_code, unsigned long payload)
562 kvm_multiple_exception(vcpu, nr, true, error_code,
563 true, payload, false);
566 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
569 kvm_inject_gp(vcpu, 0);
571 return kvm_skip_emulated_instruction(vcpu);
575 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
577 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
579 ++vcpu->stat.pf_guest;
580 vcpu->arch.exception.nested_apf =
581 is_guest_mode(vcpu) && fault->async_page_fault;
582 if (vcpu->arch.exception.nested_apf) {
583 vcpu->arch.apf.nested_apf_token = fault->address;
584 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
586 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
590 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
592 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
594 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
595 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
597 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
599 return fault->nested_page_fault;
602 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
604 atomic_inc(&vcpu->arch.nmi_queued);
605 kvm_make_request(KVM_REQ_NMI, vcpu);
607 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
609 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
611 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
613 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
615 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
617 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
619 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
622 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
623 * a #GP and return false.
625 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
627 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
629 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
632 EXPORT_SYMBOL_GPL(kvm_require_cpl);
634 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
636 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
639 kvm_queue_exception(vcpu, UD_VECTOR);
642 EXPORT_SYMBOL_GPL(kvm_require_dr);
645 * This function will be used to read from the physical memory of the currently
646 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
647 * can read from guest physical or from the guest's guest physical memory.
649 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
650 gfn_t ngfn, void *data, int offset, int len,
653 struct x86_exception exception;
657 ngpa = gfn_to_gpa(ngfn);
658 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
659 if (real_gfn == UNMAPPED_GVA)
662 real_gfn = gpa_to_gfn(real_gfn);
664 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
666 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
668 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
669 void *data, int offset, int len, u32 access)
671 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
672 data, offset, len, access);
676 * Load the pae pdptrs. Return true is they are all valid.
678 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
680 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
681 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
684 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
686 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
687 offset * sizeof(u64), sizeof(pdpte),
688 PFERR_USER_MASK|PFERR_WRITE_MASK);
693 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
694 if ((pdpte[i] & PT_PRESENT_MASK) &&
696 vcpu->arch.mmu->guest_rsvd_check.rsvd_bits_mask[0][2])) {
703 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
704 __set_bit(VCPU_EXREG_PDPTR,
705 (unsigned long *)&vcpu->arch.regs_avail);
706 __set_bit(VCPU_EXREG_PDPTR,
707 (unsigned long *)&vcpu->arch.regs_dirty);
712 EXPORT_SYMBOL_GPL(load_pdptrs);
714 bool pdptrs_changed(struct kvm_vcpu *vcpu)
716 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
722 if (is_long_mode(vcpu) || !is_pae(vcpu) || !is_paging(vcpu))
725 if (!test_bit(VCPU_EXREG_PDPTR,
726 (unsigned long *)&vcpu->arch.regs_avail))
729 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
730 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
731 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
732 PFERR_USER_MASK | PFERR_WRITE_MASK);
735 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
740 EXPORT_SYMBOL_GPL(pdptrs_changed);
742 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
744 unsigned long old_cr0 = kvm_read_cr0(vcpu);
745 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
750 if (cr0 & 0xffffffff00000000UL)
754 cr0 &= ~CR0_RESERVED_BITS;
756 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
759 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
762 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
764 if ((vcpu->arch.efer & EFER_LME)) {
769 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
774 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
779 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
782 kvm_x86_ops->set_cr0(vcpu, cr0);
784 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
785 kvm_clear_async_pf_completion_queue(vcpu);
786 kvm_async_pf_hash_reset(vcpu);
789 if ((cr0 ^ old_cr0) & update_bits)
790 kvm_mmu_reset_context(vcpu);
792 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
793 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
794 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
795 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
799 EXPORT_SYMBOL_GPL(kvm_set_cr0);
801 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
803 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
805 EXPORT_SYMBOL_GPL(kvm_lmsw);
807 void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
809 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
810 !vcpu->guest_xcr0_loaded) {
811 /* kvm_set_xcr() also depends on this */
812 if (vcpu->arch.xcr0 != host_xcr0)
813 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
814 vcpu->guest_xcr0_loaded = 1;
817 EXPORT_SYMBOL_GPL(kvm_load_guest_xcr0);
819 void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
821 if (vcpu->guest_xcr0_loaded) {
822 if (vcpu->arch.xcr0 != host_xcr0)
823 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
824 vcpu->guest_xcr0_loaded = 0;
827 EXPORT_SYMBOL_GPL(kvm_put_guest_xcr0);
829 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
832 u64 old_xcr0 = vcpu->arch.xcr0;
835 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
836 if (index != XCR_XFEATURE_ENABLED_MASK)
838 if (!(xcr0 & XFEATURE_MASK_FP))
840 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
844 * Do not allow the guest to set bits that we do not support
845 * saving. However, xcr0 bit 0 is always set, even if the
846 * emulated CPU does not support XSAVE (see fx_init).
848 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
849 if (xcr0 & ~valid_bits)
852 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
853 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
856 if (xcr0 & XFEATURE_MASK_AVX512) {
857 if (!(xcr0 & XFEATURE_MASK_YMM))
859 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
862 vcpu->arch.xcr0 = xcr0;
864 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
865 kvm_update_cpuid(vcpu);
869 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
871 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
872 __kvm_set_xcr(vcpu, index, xcr)) {
873 kvm_inject_gp(vcpu, 0);
878 EXPORT_SYMBOL_GPL(kvm_set_xcr);
880 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
882 unsigned long old_cr4 = kvm_read_cr4(vcpu);
883 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
884 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
886 if (cr4 & CR4_RESERVED_BITS)
889 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
892 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
895 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
898 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
901 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
904 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
907 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
910 if (is_long_mode(vcpu)) {
911 if (!(cr4 & X86_CR4_PAE))
913 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
914 && ((cr4 ^ old_cr4) & pdptr_bits)
915 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
919 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
920 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
923 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
924 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
928 if (kvm_x86_ops->set_cr4(vcpu, cr4))
931 if (((cr4 ^ old_cr4) & pdptr_bits) ||
932 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
933 kvm_mmu_reset_context(vcpu);
935 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
936 kvm_update_cpuid(vcpu);
940 EXPORT_SYMBOL_GPL(kvm_set_cr4);
942 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
944 bool skip_tlb_flush = false;
946 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
949 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
950 cr3 &= ~X86_CR3_PCID_NOFLUSH;
954 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
955 if (!skip_tlb_flush) {
956 kvm_mmu_sync_roots(vcpu);
957 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
962 if (is_long_mode(vcpu) &&
963 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
965 else if (is_pae(vcpu) && is_paging(vcpu) &&
966 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
969 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
970 vcpu->arch.cr3 = cr3;
971 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
975 EXPORT_SYMBOL_GPL(kvm_set_cr3);
977 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
979 if (cr8 & CR8_RESERVED_BITS)
981 if (lapic_in_kernel(vcpu))
982 kvm_lapic_set_tpr(vcpu, cr8);
984 vcpu->arch.cr8 = cr8;
987 EXPORT_SYMBOL_GPL(kvm_set_cr8);
989 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
991 if (lapic_in_kernel(vcpu))
992 return kvm_lapic_get_cr8(vcpu);
994 return vcpu->arch.cr8;
996 EXPORT_SYMBOL_GPL(kvm_get_cr8);
998 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
1002 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
1003 for (i = 0; i < KVM_NR_DB_REGS; i++)
1004 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
1005 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1009 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1011 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1012 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1015 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1019 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1020 dr7 = vcpu->arch.guest_debug_dr7;
1022 dr7 = vcpu->arch.dr7;
1023 kvm_x86_ops->set_dr7(vcpu, dr7);
1024 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1025 if (dr7 & DR7_BP_EN_MASK)
1026 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1029 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1031 u64 fixed = DR6_FIXED_1;
1033 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1038 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1042 vcpu->arch.db[dr] = val;
1043 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1044 vcpu->arch.eff_db[dr] = val;
1049 if (val & 0xffffffff00000000ULL)
1050 return -1; /* #GP */
1051 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1052 kvm_update_dr6(vcpu);
1057 if (val & 0xffffffff00000000ULL)
1058 return -1; /* #GP */
1059 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1060 kvm_update_dr7(vcpu);
1067 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1069 if (__kvm_set_dr(vcpu, dr, val)) {
1070 kvm_inject_gp(vcpu, 0);
1075 EXPORT_SYMBOL_GPL(kvm_set_dr);
1077 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1081 *val = vcpu->arch.db[dr];
1086 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1087 *val = vcpu->arch.dr6;
1089 *val = kvm_x86_ops->get_dr6(vcpu);
1094 *val = vcpu->arch.dr7;
1099 EXPORT_SYMBOL_GPL(kvm_get_dr);
1101 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1103 u32 ecx = kvm_rcx_read(vcpu);
1107 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1110 kvm_rax_write(vcpu, (u32)data);
1111 kvm_rdx_write(vcpu, data >> 32);
1114 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1117 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1118 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1120 * This list is modified at module load time to reflect the
1121 * capabilities of the host cpu. This capabilities test skips MSRs that are
1122 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1123 * may depend on host virtualization features rather than host cpu features.
1126 static u32 msrs_to_save[] = {
1127 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1129 #ifdef CONFIG_X86_64
1130 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1132 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1133 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1135 MSR_IA32_RTIT_CTL, MSR_IA32_RTIT_STATUS, MSR_IA32_RTIT_CR3_MATCH,
1136 MSR_IA32_RTIT_OUTPUT_BASE, MSR_IA32_RTIT_OUTPUT_MASK,
1137 MSR_IA32_RTIT_ADDR0_A, MSR_IA32_RTIT_ADDR0_B,
1138 MSR_IA32_RTIT_ADDR1_A, MSR_IA32_RTIT_ADDR1_B,
1139 MSR_IA32_RTIT_ADDR2_A, MSR_IA32_RTIT_ADDR2_B,
1140 MSR_IA32_RTIT_ADDR3_A, MSR_IA32_RTIT_ADDR3_B,
1143 static unsigned num_msrs_to_save;
1145 static u32 emulated_msrs[] = {
1146 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1147 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1148 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1149 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1150 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1151 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1152 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1154 HV_X64_MSR_VP_INDEX,
1155 HV_X64_MSR_VP_RUNTIME,
1156 HV_X64_MSR_SCONTROL,
1157 HV_X64_MSR_STIMER0_CONFIG,
1158 HV_X64_MSR_VP_ASSIST_PAGE,
1159 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1160 HV_X64_MSR_TSC_EMULATION_STATUS,
1162 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1165 MSR_IA32_TSC_ADJUST,
1166 MSR_IA32_TSCDEADLINE,
1167 MSR_IA32_ARCH_CAPABILITIES,
1168 MSR_IA32_MISC_ENABLE,
1169 MSR_IA32_MCG_STATUS,
1171 MSR_IA32_MCG_EXT_CTL,
1175 MSR_MISC_FEATURES_ENABLES,
1176 MSR_AMD64_VIRT_SPEC_CTRL,
1180 MSR_KVM_POLL_CONTROL,
1183 static unsigned num_emulated_msrs;
1186 * List of msr numbers which are used to expose MSR-based features that
1187 * can be used by a hypervisor to validate requested CPU features.
1189 static u32 msr_based_features[] = {
1191 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1192 MSR_IA32_VMX_PINBASED_CTLS,
1193 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1194 MSR_IA32_VMX_PROCBASED_CTLS,
1195 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1196 MSR_IA32_VMX_EXIT_CTLS,
1197 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1198 MSR_IA32_VMX_ENTRY_CTLS,
1200 MSR_IA32_VMX_CR0_FIXED0,
1201 MSR_IA32_VMX_CR0_FIXED1,
1202 MSR_IA32_VMX_CR4_FIXED0,
1203 MSR_IA32_VMX_CR4_FIXED1,
1204 MSR_IA32_VMX_VMCS_ENUM,
1205 MSR_IA32_VMX_PROCBASED_CTLS2,
1206 MSR_IA32_VMX_EPT_VPID_CAP,
1207 MSR_IA32_VMX_VMFUNC,
1211 MSR_IA32_ARCH_CAPABILITIES,
1214 static unsigned int num_msr_based_features;
1216 static u64 kvm_get_arch_capabilities(void)
1220 if (boot_cpu_has(X86_FEATURE_ARCH_CAPABILITIES))
1221 rdmsrl(MSR_IA32_ARCH_CAPABILITIES, data);
1224 * If we're doing cache flushes (either "always" or "cond")
1225 * we will do one whenever the guest does a vmlaunch/vmresume.
1226 * If an outer hypervisor is doing the cache flush for us
1227 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1228 * capability to the guest too, and if EPT is disabled we're not
1229 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1230 * require a nested hypervisor to do a flush of its own.
1232 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1233 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1238 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1240 switch (msr->index) {
1241 case MSR_IA32_ARCH_CAPABILITIES:
1242 msr->data = kvm_get_arch_capabilities();
1244 case MSR_IA32_UCODE_REV:
1245 rdmsrl_safe(msr->index, &msr->data);
1248 if (kvm_x86_ops->get_msr_feature(msr))
1254 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1256 struct kvm_msr_entry msr;
1260 r = kvm_get_msr_feature(&msr);
1269 static bool __kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1271 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1274 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1277 if (efer & (EFER_LME | EFER_LMA) &&
1278 !guest_cpuid_has(vcpu, X86_FEATURE_LM))
1281 if (efer & EFER_NX && !guest_cpuid_has(vcpu, X86_FEATURE_NX))
1287 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1289 if (efer & efer_reserved_bits)
1292 return __kvm_valid_efer(vcpu, efer);
1294 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1296 static int set_efer(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
1298 u64 old_efer = vcpu->arch.efer;
1299 u64 efer = msr_info->data;
1301 if (efer & efer_reserved_bits)
1304 if (!msr_info->host_initiated) {
1305 if (!__kvm_valid_efer(vcpu, efer))
1308 if (is_paging(vcpu) &&
1309 (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1314 efer |= vcpu->arch.efer & EFER_LMA;
1316 kvm_x86_ops->set_efer(vcpu, efer);
1318 /* Update reserved bits */
1319 if ((efer ^ old_efer) & EFER_NX)
1320 kvm_mmu_reset_context(vcpu);
1325 void kvm_enable_efer_bits(u64 mask)
1327 efer_reserved_bits &= ~mask;
1329 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1332 * Writes msr value into into the appropriate "register".
1333 * Returns 0 on success, non-0 otherwise.
1334 * Assumes vcpu_load() was already called.
1336 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1338 switch (msr->index) {
1341 case MSR_KERNEL_GS_BASE:
1344 if (is_noncanonical_address(msr->data, vcpu))
1347 case MSR_IA32_SYSENTER_EIP:
1348 case MSR_IA32_SYSENTER_ESP:
1350 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1351 * non-canonical address is written on Intel but not on
1352 * AMD (which ignores the top 32-bits, because it does
1353 * not implement 64-bit SYSENTER).
1355 * 64-bit code should hence be able to write a non-canonical
1356 * value on AMD. Making the address canonical ensures that
1357 * vmentry does not fail on Intel after writing a non-canonical
1358 * value, and that something deterministic happens if the guest
1359 * invokes 64-bit SYSENTER.
1361 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1363 return kvm_x86_ops->set_msr(vcpu, msr);
1365 EXPORT_SYMBOL_GPL(kvm_set_msr);
1368 * Adapt set_msr() to msr_io()'s calling convention
1370 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1372 struct msr_data msr;
1376 msr.host_initiated = true;
1377 r = kvm_get_msr(vcpu, &msr);
1385 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1387 struct msr_data msr;
1391 msr.host_initiated = true;
1392 return kvm_set_msr(vcpu, &msr);
1395 #ifdef CONFIG_X86_64
1396 struct pvclock_gtod_data {
1399 struct { /* extract of a clocksource struct */
1412 static struct pvclock_gtod_data pvclock_gtod_data;
1414 static void update_pvclock_gtod(struct timekeeper *tk)
1416 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1419 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1421 write_seqcount_begin(&vdata->seq);
1423 /* copy pvclock gtod data */
1424 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1425 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1426 vdata->clock.mask = tk->tkr_mono.mask;
1427 vdata->clock.mult = tk->tkr_mono.mult;
1428 vdata->clock.shift = tk->tkr_mono.shift;
1430 vdata->boot_ns = boot_ns;
1431 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1433 vdata->wall_time_sec = tk->xtime_sec;
1435 write_seqcount_end(&vdata->seq);
1439 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1442 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1443 * vcpu_enter_guest. This function is only called from
1444 * the physical CPU that is running vcpu.
1446 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1449 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1453 struct pvclock_wall_clock wc;
1454 struct timespec64 boot;
1459 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1464 ++version; /* first time write, random junk */
1468 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1472 * The guest calculates current wall clock time by adding
1473 * system time (updated by kvm_guest_time_update below) to the
1474 * wall clock specified here. guest system time equals host
1475 * system time for us, thus we must fill in host boot time here.
1477 getboottime64(&boot);
1479 if (kvm->arch.kvmclock_offset) {
1480 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1481 boot = timespec64_sub(boot, ts);
1483 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1484 wc.nsec = boot.tv_nsec;
1485 wc.version = version;
1487 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1490 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1493 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1495 do_shl32_div32(dividend, divisor);
1499 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1500 s8 *pshift, u32 *pmultiplier)
1508 scaled64 = scaled_hz;
1509 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1514 tps32 = (uint32_t)tps64;
1515 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1516 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1524 *pmultiplier = div_frac(scaled64, tps32);
1526 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1527 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1530 #ifdef CONFIG_X86_64
1531 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1534 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1535 static unsigned long max_tsc_khz;
1537 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1539 u64 v = (u64)khz * (1000000 + ppm);
1544 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1548 /* Guest TSC same frequency as host TSC? */
1550 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1554 /* TSC scaling supported? */
1555 if (!kvm_has_tsc_control) {
1556 if (user_tsc_khz > tsc_khz) {
1557 vcpu->arch.tsc_catchup = 1;
1558 vcpu->arch.tsc_always_catchup = 1;
1561 WARN(1, "user requested TSC rate below hardware speed\n");
1566 /* TSC scaling required - calculate ratio */
1567 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1568 user_tsc_khz, tsc_khz);
1570 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1571 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1576 vcpu->arch.tsc_scaling_ratio = ratio;
1580 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1582 u32 thresh_lo, thresh_hi;
1583 int use_scaling = 0;
1585 /* tsc_khz can be zero if TSC calibration fails */
1586 if (user_tsc_khz == 0) {
1587 /* set tsc_scaling_ratio to a safe value */
1588 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1592 /* Compute a scale to convert nanoseconds in TSC cycles */
1593 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1594 &vcpu->arch.virtual_tsc_shift,
1595 &vcpu->arch.virtual_tsc_mult);
1596 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1599 * Compute the variation in TSC rate which is acceptable
1600 * within the range of tolerance and decide if the
1601 * rate being applied is within that bounds of the hardware
1602 * rate. If so, no scaling or compensation need be done.
1604 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1605 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1606 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1607 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1610 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1613 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1615 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1616 vcpu->arch.virtual_tsc_mult,
1617 vcpu->arch.virtual_tsc_shift);
1618 tsc += vcpu->arch.this_tsc_write;
1622 static inline int gtod_is_based_on_tsc(int mode)
1624 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1627 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1629 #ifdef CONFIG_X86_64
1631 struct kvm_arch *ka = &vcpu->kvm->arch;
1632 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1634 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1635 atomic_read(&vcpu->kvm->online_vcpus));
1638 * Once the masterclock is enabled, always perform request in
1639 * order to update it.
1641 * In order to enable masterclock, the host clocksource must be TSC
1642 * and the vcpus need to have matched TSCs. When that happens,
1643 * perform request to enable masterclock.
1645 if (ka->use_master_clock ||
1646 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1647 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1649 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1650 atomic_read(&vcpu->kvm->online_vcpus),
1651 ka->use_master_clock, gtod->clock.vclock_mode);
1655 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1657 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1658 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1662 * Multiply tsc by a fixed point number represented by ratio.
1664 * The most significant 64-N bits (mult) of ratio represent the
1665 * integral part of the fixed point number; the remaining N bits
1666 * (frac) represent the fractional part, ie. ratio represents a fixed
1667 * point number (mult + frac * 2^(-N)).
1669 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1671 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1673 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1676 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1679 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1681 if (ratio != kvm_default_tsc_scaling_ratio)
1682 _tsc = __scale_tsc(ratio, tsc);
1686 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1688 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1692 tsc = kvm_scale_tsc(vcpu, rdtsc());
1694 return target_tsc - tsc;
1697 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1699 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1701 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1703 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1705 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1707 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1710 static inline bool kvm_check_tsc_unstable(void)
1712 #ifdef CONFIG_X86_64
1714 * TSC is marked unstable when we're running on Hyper-V,
1715 * 'TSC page' clocksource is good.
1717 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1720 return check_tsc_unstable();
1723 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1725 struct kvm *kvm = vcpu->kvm;
1726 u64 offset, ns, elapsed;
1727 unsigned long flags;
1729 bool already_matched;
1730 u64 data = msr->data;
1731 bool synchronizing = false;
1733 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1734 offset = kvm_compute_tsc_offset(vcpu, data);
1735 ns = ktime_get_boot_ns();
1736 elapsed = ns - kvm->arch.last_tsc_nsec;
1738 if (vcpu->arch.virtual_tsc_khz) {
1739 if (data == 0 && msr->host_initiated) {
1741 * detection of vcpu initialization -- need to sync
1742 * with other vCPUs. This particularly helps to keep
1743 * kvm_clock stable after CPU hotplug
1745 synchronizing = true;
1747 u64 tsc_exp = kvm->arch.last_tsc_write +
1748 nsec_to_cycles(vcpu, elapsed);
1749 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1751 * Special case: TSC write with a small delta (1 second)
1752 * of virtual cycle time against real time is
1753 * interpreted as an attempt to synchronize the CPU.
1755 synchronizing = data < tsc_exp + tsc_hz &&
1756 data + tsc_hz > tsc_exp;
1761 * For a reliable TSC, we can match TSC offsets, and for an unstable
1762 * TSC, we add elapsed time in this computation. We could let the
1763 * compensation code attempt to catch up if we fall behind, but
1764 * it's better to try to match offsets from the beginning.
1766 if (synchronizing &&
1767 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1768 if (!kvm_check_tsc_unstable()) {
1769 offset = kvm->arch.cur_tsc_offset;
1770 pr_debug("kvm: matched tsc offset for %llu\n", data);
1772 u64 delta = nsec_to_cycles(vcpu, elapsed);
1774 offset = kvm_compute_tsc_offset(vcpu, data);
1775 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1778 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1781 * We split periods of matched TSC writes into generations.
1782 * For each generation, we track the original measured
1783 * nanosecond time, offset, and write, so if TSCs are in
1784 * sync, we can match exact offset, and if not, we can match
1785 * exact software computation in compute_guest_tsc()
1787 * These values are tracked in kvm->arch.cur_xxx variables.
1789 kvm->arch.cur_tsc_generation++;
1790 kvm->arch.cur_tsc_nsec = ns;
1791 kvm->arch.cur_tsc_write = data;
1792 kvm->arch.cur_tsc_offset = offset;
1794 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1795 kvm->arch.cur_tsc_generation, data);
1799 * We also track th most recent recorded KHZ, write and time to
1800 * allow the matching interval to be extended at each write.
1802 kvm->arch.last_tsc_nsec = ns;
1803 kvm->arch.last_tsc_write = data;
1804 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1806 vcpu->arch.last_guest_tsc = data;
1808 /* Keep track of which generation this VCPU has synchronized to */
1809 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1810 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1811 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1813 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1814 update_ia32_tsc_adjust_msr(vcpu, offset);
1816 kvm_vcpu_write_tsc_offset(vcpu, offset);
1817 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1819 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1821 kvm->arch.nr_vcpus_matched_tsc = 0;
1822 } else if (!already_matched) {
1823 kvm->arch.nr_vcpus_matched_tsc++;
1826 kvm_track_tsc_matching(vcpu);
1827 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1830 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1832 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1835 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1836 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1839 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1841 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1842 WARN_ON(adjustment < 0);
1843 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1844 adjust_tsc_offset_guest(vcpu, adjustment);
1847 #ifdef CONFIG_X86_64
1849 static u64 read_tsc(void)
1851 u64 ret = (u64)rdtsc_ordered();
1852 u64 last = pvclock_gtod_data.clock.cycle_last;
1854 if (likely(ret >= last))
1858 * GCC likes to generate cmov here, but this branch is extremely
1859 * predictable (it's just a function of time and the likely is
1860 * very likely) and there's a data dependence, so force GCC
1861 * to generate a branch instead. I don't barrier() because
1862 * we don't actually need a barrier, and if this function
1863 * ever gets inlined it will generate worse code.
1869 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1872 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1875 switch (gtod->clock.vclock_mode) {
1876 case VCLOCK_HVCLOCK:
1877 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1879 if (tsc_pg_val != U64_MAX) {
1880 /* TSC page valid */
1881 *mode = VCLOCK_HVCLOCK;
1882 v = (tsc_pg_val - gtod->clock.cycle_last) &
1885 /* TSC page invalid */
1886 *mode = VCLOCK_NONE;
1891 *tsc_timestamp = read_tsc();
1892 v = (*tsc_timestamp - gtod->clock.cycle_last) &
1896 *mode = VCLOCK_NONE;
1899 if (*mode == VCLOCK_NONE)
1900 *tsc_timestamp = v = 0;
1902 return v * gtod->clock.mult;
1905 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1907 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1913 seq = read_seqcount_begin(>od->seq);
1914 ns = gtod->nsec_base;
1915 ns += vgettsc(tsc_timestamp, &mode);
1916 ns >>= gtod->clock.shift;
1917 ns += gtod->boot_ns;
1918 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1924 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1926 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1932 seq = read_seqcount_begin(>od->seq);
1933 ts->tv_sec = gtod->wall_time_sec;
1934 ns = gtod->nsec_base;
1935 ns += vgettsc(tsc_timestamp, &mode);
1936 ns >>= gtod->clock.shift;
1937 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1939 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1945 /* returns true if host is using TSC based clocksource */
1946 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1948 /* checked again under seqlock below */
1949 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1952 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1956 /* returns true if host is using TSC based clocksource */
1957 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1960 /* checked again under seqlock below */
1961 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1964 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1970 * Assuming a stable TSC across physical CPUS, and a stable TSC
1971 * across virtual CPUs, the following condition is possible.
1972 * Each numbered line represents an event visible to both
1973 * CPUs at the next numbered event.
1975 * "timespecX" represents host monotonic time. "tscX" represents
1978 * VCPU0 on CPU0 | VCPU1 on CPU1
1980 * 1. read timespec0,tsc0
1981 * 2. | timespec1 = timespec0 + N
1983 * 3. transition to guest | transition to guest
1984 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1985 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1986 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1988 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1991 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1993 * - 0 < N - M => M < N
1995 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1996 * always the case (the difference between two distinct xtime instances
1997 * might be smaller then the difference between corresponding TSC reads,
1998 * when updating guest vcpus pvclock areas).
2000 * To avoid that problem, do not allow visibility of distinct
2001 * system_timestamp/tsc_timestamp values simultaneously: use a master
2002 * copy of host monotonic time values. Update that master copy
2005 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
2009 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
2011 #ifdef CONFIG_X86_64
2012 struct kvm_arch *ka = &kvm->arch;
2014 bool host_tsc_clocksource, vcpus_matched;
2016 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
2017 atomic_read(&kvm->online_vcpus));
2020 * If the host uses TSC clock, then passthrough TSC as stable
2023 host_tsc_clocksource = kvm_get_time_and_clockread(
2024 &ka->master_kernel_ns,
2025 &ka->master_cycle_now);
2027 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
2028 && !ka->backwards_tsc_observed
2029 && !ka->boot_vcpu_runs_old_kvmclock;
2031 if (ka->use_master_clock)
2032 atomic_set(&kvm_guest_has_master_clock, 1);
2034 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
2035 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2040 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2042 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2045 static void kvm_gen_update_masterclock(struct kvm *kvm)
2047 #ifdef CONFIG_X86_64
2049 struct kvm_vcpu *vcpu;
2050 struct kvm_arch *ka = &kvm->arch;
2052 spin_lock(&ka->pvclock_gtod_sync_lock);
2053 kvm_make_mclock_inprogress_request(kvm);
2054 /* no guest entries from this point */
2055 pvclock_update_vm_gtod_copy(kvm);
2057 kvm_for_each_vcpu(i, vcpu, kvm)
2058 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2060 /* guest entries allowed */
2061 kvm_for_each_vcpu(i, vcpu, kvm)
2062 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2064 spin_unlock(&ka->pvclock_gtod_sync_lock);
2068 u64 get_kvmclock_ns(struct kvm *kvm)
2070 struct kvm_arch *ka = &kvm->arch;
2071 struct pvclock_vcpu_time_info hv_clock;
2074 spin_lock(&ka->pvclock_gtod_sync_lock);
2075 if (!ka->use_master_clock) {
2076 spin_unlock(&ka->pvclock_gtod_sync_lock);
2077 return ktime_get_boot_ns() + ka->kvmclock_offset;
2080 hv_clock.tsc_timestamp = ka->master_cycle_now;
2081 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2082 spin_unlock(&ka->pvclock_gtod_sync_lock);
2084 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2087 if (__this_cpu_read(cpu_tsc_khz)) {
2088 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2089 &hv_clock.tsc_shift,
2090 &hv_clock.tsc_to_system_mul);
2091 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2093 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
2100 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2102 struct kvm_vcpu_arch *vcpu = &v->arch;
2103 struct pvclock_vcpu_time_info guest_hv_clock;
2105 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2106 &guest_hv_clock, sizeof(guest_hv_clock))))
2109 /* This VCPU is paused, but it's legal for a guest to read another
2110 * VCPU's kvmclock, so we really have to follow the specification where
2111 * it says that version is odd if data is being modified, and even after
2114 * Version field updates must be kept separate. This is because
2115 * kvm_write_guest_cached might use a "rep movs" instruction, and
2116 * writes within a string instruction are weakly ordered. So there
2117 * are three writes overall.
2119 * As a small optimization, only write the version field in the first
2120 * and third write. The vcpu->pv_time cache is still valid, because the
2121 * version field is the first in the struct.
2123 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2125 if (guest_hv_clock.version & 1)
2126 ++guest_hv_clock.version; /* first time write, random junk */
2128 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2129 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2131 sizeof(vcpu->hv_clock.version));
2135 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2136 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2138 if (vcpu->pvclock_set_guest_stopped_request) {
2139 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2140 vcpu->pvclock_set_guest_stopped_request = false;
2143 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2145 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2147 sizeof(vcpu->hv_clock));
2151 vcpu->hv_clock.version++;
2152 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2154 sizeof(vcpu->hv_clock.version));
2157 static int kvm_guest_time_update(struct kvm_vcpu *v)
2159 unsigned long flags, tgt_tsc_khz;
2160 struct kvm_vcpu_arch *vcpu = &v->arch;
2161 struct kvm_arch *ka = &v->kvm->arch;
2163 u64 tsc_timestamp, host_tsc;
2165 bool use_master_clock;
2171 * If the host uses TSC clock, then passthrough TSC as stable
2174 spin_lock(&ka->pvclock_gtod_sync_lock);
2175 use_master_clock = ka->use_master_clock;
2176 if (use_master_clock) {
2177 host_tsc = ka->master_cycle_now;
2178 kernel_ns = ka->master_kernel_ns;
2180 spin_unlock(&ka->pvclock_gtod_sync_lock);
2182 /* Keep irq disabled to prevent changes to the clock */
2183 local_irq_save(flags);
2184 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2185 if (unlikely(tgt_tsc_khz == 0)) {
2186 local_irq_restore(flags);
2187 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2190 if (!use_master_clock) {
2192 kernel_ns = ktime_get_boot_ns();
2195 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2198 * We may have to catch up the TSC to match elapsed wall clock
2199 * time for two reasons, even if kvmclock is used.
2200 * 1) CPU could have been running below the maximum TSC rate
2201 * 2) Broken TSC compensation resets the base at each VCPU
2202 * entry to avoid unknown leaps of TSC even when running
2203 * again on the same CPU. This may cause apparent elapsed
2204 * time to disappear, and the guest to stand still or run
2207 if (vcpu->tsc_catchup) {
2208 u64 tsc = compute_guest_tsc(v, kernel_ns);
2209 if (tsc > tsc_timestamp) {
2210 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2211 tsc_timestamp = tsc;
2215 local_irq_restore(flags);
2217 /* With all the info we got, fill in the values */
2219 if (kvm_has_tsc_control)
2220 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2222 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2223 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2224 &vcpu->hv_clock.tsc_shift,
2225 &vcpu->hv_clock.tsc_to_system_mul);
2226 vcpu->hw_tsc_khz = tgt_tsc_khz;
2229 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2230 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2231 vcpu->last_guest_tsc = tsc_timestamp;
2233 /* If the host uses TSC clocksource, then it is stable */
2235 if (use_master_clock)
2236 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2238 vcpu->hv_clock.flags = pvclock_flags;
2240 if (vcpu->pv_time_enabled)
2241 kvm_setup_pvclock_page(v);
2242 if (v == kvm_get_vcpu(v->kvm, 0))
2243 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2248 * kvmclock updates which are isolated to a given vcpu, such as
2249 * vcpu->cpu migration, should not allow system_timestamp from
2250 * the rest of the vcpus to remain static. Otherwise ntp frequency
2251 * correction applies to one vcpu's system_timestamp but not
2254 * So in those cases, request a kvmclock update for all vcpus.
2255 * We need to rate-limit these requests though, as they can
2256 * considerably slow guests that have a large number of vcpus.
2257 * The time for a remote vcpu to update its kvmclock is bound
2258 * by the delay we use to rate-limit the updates.
2261 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2263 static void kvmclock_update_fn(struct work_struct *work)
2266 struct delayed_work *dwork = to_delayed_work(work);
2267 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2268 kvmclock_update_work);
2269 struct kvm *kvm = container_of(ka, struct kvm, arch);
2270 struct kvm_vcpu *vcpu;
2272 kvm_for_each_vcpu(i, vcpu, kvm) {
2273 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2274 kvm_vcpu_kick(vcpu);
2278 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2280 struct kvm *kvm = v->kvm;
2282 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2283 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2284 KVMCLOCK_UPDATE_DELAY);
2287 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2289 static void kvmclock_sync_fn(struct work_struct *work)
2291 struct delayed_work *dwork = to_delayed_work(work);
2292 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2293 kvmclock_sync_work);
2294 struct kvm *kvm = container_of(ka, struct kvm, arch);
2296 if (!kvmclock_periodic_sync)
2299 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2300 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2301 KVMCLOCK_SYNC_PERIOD);
2305 * On AMD, HWCR[McStatusWrEn] controls whether setting MCi_STATUS results in #GP.
2307 static bool can_set_mci_status(struct kvm_vcpu *vcpu)
2309 /* McStatusWrEn enabled? */
2310 if (guest_cpuid_is_amd(vcpu))
2311 return !!(vcpu->arch.msr_hwcr & BIT_ULL(18));
2316 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2318 u64 mcg_cap = vcpu->arch.mcg_cap;
2319 unsigned bank_num = mcg_cap & 0xff;
2320 u32 msr = msr_info->index;
2321 u64 data = msr_info->data;
2324 case MSR_IA32_MCG_STATUS:
2325 vcpu->arch.mcg_status = data;
2327 case MSR_IA32_MCG_CTL:
2328 if (!(mcg_cap & MCG_CTL_P) &&
2329 (data || !msr_info->host_initiated))
2331 if (data != 0 && data != ~(u64)0)
2333 vcpu->arch.mcg_ctl = data;
2336 if (msr >= MSR_IA32_MC0_CTL &&
2337 msr < MSR_IA32_MCx_CTL(bank_num)) {
2338 u32 offset = msr - MSR_IA32_MC0_CTL;
2339 /* only 0 or all 1s can be written to IA32_MCi_CTL
2340 * some Linux kernels though clear bit 10 in bank 4 to
2341 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2342 * this to avoid an uncatched #GP in the guest
2344 if ((offset & 0x3) == 0 &&
2345 data != 0 && (data | (1 << 10)) != ~(u64)0)
2349 if (!msr_info->host_initiated &&
2350 (offset & 0x3) == 1 && data != 0) {
2351 if (!can_set_mci_status(vcpu))
2355 vcpu->arch.mce_banks[offset] = data;
2363 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2365 struct kvm *kvm = vcpu->kvm;
2366 int lm = is_long_mode(vcpu);
2367 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2368 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2369 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2370 : kvm->arch.xen_hvm_config.blob_size_32;
2371 u32 page_num = data & ~PAGE_MASK;
2372 u64 page_addr = data & PAGE_MASK;
2377 if (page_num >= blob_size)
2380 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2385 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2394 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2396 gpa_t gpa = data & ~0x3f;
2398 /* Bits 3:5 are reserved, Should be zero */
2402 vcpu->arch.apf.msr_val = data;
2404 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2405 kvm_clear_async_pf_completion_queue(vcpu);
2406 kvm_async_pf_hash_reset(vcpu);
2410 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2414 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2415 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2416 kvm_async_pf_wakeup_all(vcpu);
2420 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2422 vcpu->arch.pv_time_enabled = false;
2425 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2427 ++vcpu->stat.tlb_flush;
2428 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2431 static void record_steal_time(struct kvm_vcpu *vcpu)
2433 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2436 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2437 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2441 * Doing a TLB flush here, on the guest's behalf, can avoid
2444 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2445 kvm_vcpu_flush_tlb(vcpu, false);
2447 if (vcpu->arch.st.steal.version & 1)
2448 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2450 vcpu->arch.st.steal.version += 1;
2452 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2453 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2457 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2458 vcpu->arch.st.last_steal;
2459 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2461 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2462 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2466 vcpu->arch.st.steal.version += 1;
2468 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2469 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2472 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2475 u32 msr = msr_info->index;
2476 u64 data = msr_info->data;
2479 case MSR_AMD64_NB_CFG:
2480 case MSR_IA32_UCODE_WRITE:
2481 case MSR_VM_HSAVE_PA:
2482 case MSR_AMD64_PATCH_LOADER:
2483 case MSR_AMD64_BU_CFG2:
2484 case MSR_AMD64_DC_CFG:
2485 case MSR_F15H_EX_CFG:
2488 case MSR_IA32_UCODE_REV:
2489 if (msr_info->host_initiated)
2490 vcpu->arch.microcode_version = data;
2492 case MSR_IA32_ARCH_CAPABILITIES:
2493 if (!msr_info->host_initiated)
2495 vcpu->arch.arch_capabilities = data;
2498 return set_efer(vcpu, msr_info);
2500 data &= ~(u64)0x40; /* ignore flush filter disable */
2501 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2502 data &= ~(u64)0x8; /* ignore TLB cache disable */
2504 /* Handle McStatusWrEn */
2505 if (data == BIT_ULL(18)) {
2506 vcpu->arch.msr_hwcr = data;
2507 } else if (data != 0) {
2508 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2513 case MSR_FAM10H_MMIO_CONF_BASE:
2515 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2520 case MSR_IA32_DEBUGCTLMSR:
2522 /* We support the non-activated case already */
2524 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2525 /* Values other than LBR and BTF are vendor-specific,
2526 thus reserved and should throw a #GP */
2529 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2532 case 0x200 ... 0x2ff:
2533 return kvm_mtrr_set_msr(vcpu, msr, data);
2534 case MSR_IA32_APICBASE:
2535 return kvm_set_apic_base(vcpu, msr_info);
2536 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2537 return kvm_x2apic_msr_write(vcpu, msr, data);
2538 case MSR_IA32_TSCDEADLINE:
2539 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2541 case MSR_IA32_TSC_ADJUST:
2542 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2543 if (!msr_info->host_initiated) {
2544 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2545 adjust_tsc_offset_guest(vcpu, adj);
2547 vcpu->arch.ia32_tsc_adjust_msr = data;
2550 case MSR_IA32_MISC_ENABLE:
2551 if (!kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_MISC_ENABLE_NO_MWAIT) &&
2552 ((vcpu->arch.ia32_misc_enable_msr ^ data) & MSR_IA32_MISC_ENABLE_MWAIT)) {
2553 if (!guest_cpuid_has(vcpu, X86_FEATURE_XMM3))
2555 vcpu->arch.ia32_misc_enable_msr = data;
2556 kvm_update_cpuid(vcpu);
2558 vcpu->arch.ia32_misc_enable_msr = data;
2561 case MSR_IA32_SMBASE:
2562 if (!msr_info->host_initiated)
2564 vcpu->arch.smbase = data;
2566 case MSR_IA32_POWER_CTL:
2567 vcpu->arch.msr_ia32_power_ctl = data;
2570 kvm_write_tsc(vcpu, msr_info);
2573 if (!msr_info->host_initiated)
2575 vcpu->arch.smi_count = data;
2577 case MSR_KVM_WALL_CLOCK_NEW:
2578 case MSR_KVM_WALL_CLOCK:
2579 vcpu->kvm->arch.wall_clock = data;
2580 kvm_write_wall_clock(vcpu->kvm, data);
2582 case MSR_KVM_SYSTEM_TIME_NEW:
2583 case MSR_KVM_SYSTEM_TIME: {
2584 struct kvm_arch *ka = &vcpu->kvm->arch;
2586 kvmclock_reset(vcpu);
2588 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2589 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2591 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2592 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2594 ka->boot_vcpu_runs_old_kvmclock = tmp;
2597 vcpu->arch.time = data;
2598 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2600 /* we verify if the enable bit is set... */
2604 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2605 &vcpu->arch.pv_time, data & ~1ULL,
2606 sizeof(struct pvclock_vcpu_time_info)))
2607 vcpu->arch.pv_time_enabled = false;
2609 vcpu->arch.pv_time_enabled = true;
2613 case MSR_KVM_ASYNC_PF_EN:
2614 if (kvm_pv_enable_async_pf(vcpu, data))
2617 case MSR_KVM_STEAL_TIME:
2619 if (unlikely(!sched_info_on()))
2622 if (data & KVM_STEAL_RESERVED_MASK)
2625 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2626 data & KVM_STEAL_VALID_BITS,
2627 sizeof(struct kvm_steal_time)))
2630 vcpu->arch.st.msr_val = data;
2632 if (!(data & KVM_MSR_ENABLED))
2635 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2638 case MSR_KVM_PV_EOI_EN:
2639 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2643 case MSR_KVM_POLL_CONTROL:
2644 /* only enable bit supported */
2645 if (data & (-1ULL << 1))
2648 vcpu->arch.msr_kvm_poll_control = data;
2651 case MSR_IA32_MCG_CTL:
2652 case MSR_IA32_MCG_STATUS:
2653 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2654 return set_msr_mce(vcpu, msr_info);
2656 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2657 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2658 pr = true; /* fall through */
2659 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2660 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2661 if (kvm_pmu_is_valid_msr(vcpu, msr))
2662 return kvm_pmu_set_msr(vcpu, msr_info);
2664 if (pr || data != 0)
2665 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2666 "0x%x data 0x%llx\n", msr, data);
2668 case MSR_K7_CLK_CTL:
2670 * Ignore all writes to this no longer documented MSR.
2671 * Writes are only relevant for old K7 processors,
2672 * all pre-dating SVM, but a recommended workaround from
2673 * AMD for these chips. It is possible to specify the
2674 * affected processor models on the command line, hence
2675 * the need to ignore the workaround.
2678 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2679 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2680 case HV_X64_MSR_CRASH_CTL:
2681 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2682 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2683 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2684 case HV_X64_MSR_TSC_EMULATION_STATUS:
2685 return kvm_hv_set_msr_common(vcpu, msr, data,
2686 msr_info->host_initiated);
2687 case MSR_IA32_BBL_CR_CTL3:
2688 /* Drop writes to this legacy MSR -- see rdmsr
2689 * counterpart for further detail.
2691 if (report_ignored_msrs)
2692 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2695 case MSR_AMD64_OSVW_ID_LENGTH:
2696 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2698 vcpu->arch.osvw.length = data;
2700 case MSR_AMD64_OSVW_STATUS:
2701 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2703 vcpu->arch.osvw.status = data;
2705 case MSR_PLATFORM_INFO:
2706 if (!msr_info->host_initiated ||
2707 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2708 cpuid_fault_enabled(vcpu)))
2710 vcpu->arch.msr_platform_info = data;
2712 case MSR_MISC_FEATURES_ENABLES:
2713 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2714 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2715 !supports_cpuid_fault(vcpu)))
2717 vcpu->arch.msr_misc_features_enables = data;
2720 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2721 return xen_hvm_config(vcpu, data);
2722 if (kvm_pmu_is_valid_msr(vcpu, msr))
2723 return kvm_pmu_set_msr(vcpu, msr_info);
2725 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2729 if (report_ignored_msrs)
2731 "ignored wrmsr: 0x%x data 0x%llx\n",
2738 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2742 * Reads an msr value (of 'msr_index') into 'pdata'.
2743 * Returns 0 on success, non-0 otherwise.
2744 * Assumes vcpu_load() was already called.
2746 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2748 return kvm_x86_ops->get_msr(vcpu, msr);
2750 EXPORT_SYMBOL_GPL(kvm_get_msr);
2752 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2755 u64 mcg_cap = vcpu->arch.mcg_cap;
2756 unsigned bank_num = mcg_cap & 0xff;
2759 case MSR_IA32_P5_MC_ADDR:
2760 case MSR_IA32_P5_MC_TYPE:
2763 case MSR_IA32_MCG_CAP:
2764 data = vcpu->arch.mcg_cap;
2766 case MSR_IA32_MCG_CTL:
2767 if (!(mcg_cap & MCG_CTL_P) && !host)
2769 data = vcpu->arch.mcg_ctl;
2771 case MSR_IA32_MCG_STATUS:
2772 data = vcpu->arch.mcg_status;
2775 if (msr >= MSR_IA32_MC0_CTL &&
2776 msr < MSR_IA32_MCx_CTL(bank_num)) {
2777 u32 offset = msr - MSR_IA32_MC0_CTL;
2778 data = vcpu->arch.mce_banks[offset];
2787 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2789 switch (msr_info->index) {
2790 case MSR_IA32_PLATFORM_ID:
2791 case MSR_IA32_EBL_CR_POWERON:
2792 case MSR_IA32_DEBUGCTLMSR:
2793 case MSR_IA32_LASTBRANCHFROMIP:
2794 case MSR_IA32_LASTBRANCHTOIP:
2795 case MSR_IA32_LASTINTFROMIP:
2796 case MSR_IA32_LASTINTTOIP:
2798 case MSR_K8_TSEG_ADDR:
2799 case MSR_K8_TSEG_MASK:
2800 case MSR_VM_HSAVE_PA:
2801 case MSR_K8_INT_PENDING_MSG:
2802 case MSR_AMD64_NB_CFG:
2803 case MSR_FAM10H_MMIO_CONF_BASE:
2804 case MSR_AMD64_BU_CFG2:
2805 case MSR_IA32_PERF_CTL:
2806 case MSR_AMD64_DC_CFG:
2807 case MSR_F15H_EX_CFG:
2810 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2811 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2812 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2813 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2814 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2815 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2816 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2819 case MSR_IA32_UCODE_REV:
2820 msr_info->data = vcpu->arch.microcode_version;
2822 case MSR_IA32_ARCH_CAPABILITIES:
2823 if (!msr_info->host_initiated &&
2824 !guest_cpuid_has(vcpu, X86_FEATURE_ARCH_CAPABILITIES))
2826 msr_info->data = vcpu->arch.arch_capabilities;
2828 case MSR_IA32_POWER_CTL:
2829 msr_info->data = vcpu->arch.msr_ia32_power_ctl;
2832 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2835 case 0x200 ... 0x2ff:
2836 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2837 case 0xcd: /* fsb frequency */
2841 * MSR_EBC_FREQUENCY_ID
2842 * Conservative value valid for even the basic CPU models.
2843 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2844 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2845 * and 266MHz for model 3, or 4. Set Core Clock
2846 * Frequency to System Bus Frequency Ratio to 1 (bits
2847 * 31:24) even though these are only valid for CPU
2848 * models > 2, however guests may end up dividing or
2849 * multiplying by zero otherwise.
2851 case MSR_EBC_FREQUENCY_ID:
2852 msr_info->data = 1 << 24;
2854 case MSR_IA32_APICBASE:
2855 msr_info->data = kvm_get_apic_base(vcpu);
2857 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2858 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2860 case MSR_IA32_TSCDEADLINE:
2861 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2863 case MSR_IA32_TSC_ADJUST:
2864 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2866 case MSR_IA32_MISC_ENABLE:
2867 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2869 case MSR_IA32_SMBASE:
2870 if (!msr_info->host_initiated)
2872 msr_info->data = vcpu->arch.smbase;
2875 msr_info->data = vcpu->arch.smi_count;
2877 case MSR_IA32_PERF_STATUS:
2878 /* TSC increment by tick */
2879 msr_info->data = 1000ULL;
2880 /* CPU multiplier */
2881 msr_info->data |= (((uint64_t)4ULL) << 40);
2884 msr_info->data = vcpu->arch.efer;
2886 case MSR_KVM_WALL_CLOCK:
2887 case MSR_KVM_WALL_CLOCK_NEW:
2888 msr_info->data = vcpu->kvm->arch.wall_clock;
2890 case MSR_KVM_SYSTEM_TIME:
2891 case MSR_KVM_SYSTEM_TIME_NEW:
2892 msr_info->data = vcpu->arch.time;
2894 case MSR_KVM_ASYNC_PF_EN:
2895 msr_info->data = vcpu->arch.apf.msr_val;
2897 case MSR_KVM_STEAL_TIME:
2898 msr_info->data = vcpu->arch.st.msr_val;
2900 case MSR_KVM_PV_EOI_EN:
2901 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2903 case MSR_KVM_POLL_CONTROL:
2904 msr_info->data = vcpu->arch.msr_kvm_poll_control;
2906 case MSR_IA32_P5_MC_ADDR:
2907 case MSR_IA32_P5_MC_TYPE:
2908 case MSR_IA32_MCG_CAP:
2909 case MSR_IA32_MCG_CTL:
2910 case MSR_IA32_MCG_STATUS:
2911 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2912 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2913 msr_info->host_initiated);
2914 case MSR_K7_CLK_CTL:
2916 * Provide expected ramp-up count for K7. All other
2917 * are set to zero, indicating minimum divisors for
2920 * This prevents guest kernels on AMD host with CPU
2921 * type 6, model 8 and higher from exploding due to
2922 * the rdmsr failing.
2924 msr_info->data = 0x20000000;
2926 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2927 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2928 case HV_X64_MSR_CRASH_CTL:
2929 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2930 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2931 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2932 case HV_X64_MSR_TSC_EMULATION_STATUS:
2933 return kvm_hv_get_msr_common(vcpu,
2934 msr_info->index, &msr_info->data,
2935 msr_info->host_initiated);
2937 case MSR_IA32_BBL_CR_CTL3:
2938 /* This legacy MSR exists but isn't fully documented in current
2939 * silicon. It is however accessed by winxp in very narrow
2940 * scenarios where it sets bit #19, itself documented as
2941 * a "reserved" bit. Best effort attempt to source coherent
2942 * read data here should the balance of the register be
2943 * interpreted by the guest:
2945 * L2 cache control register 3: 64GB range, 256KB size,
2946 * enabled, latency 0x1, configured
2948 msr_info->data = 0xbe702111;
2950 case MSR_AMD64_OSVW_ID_LENGTH:
2951 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2953 msr_info->data = vcpu->arch.osvw.length;
2955 case MSR_AMD64_OSVW_STATUS:
2956 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2958 msr_info->data = vcpu->arch.osvw.status;
2960 case MSR_PLATFORM_INFO:
2961 if (!msr_info->host_initiated &&
2962 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2964 msr_info->data = vcpu->arch.msr_platform_info;
2966 case MSR_MISC_FEATURES_ENABLES:
2967 msr_info->data = vcpu->arch.msr_misc_features_enables;
2970 msr_info->data = vcpu->arch.msr_hwcr;
2973 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2974 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2976 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2980 if (report_ignored_msrs)
2981 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2989 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2992 * Read or write a bunch of msrs. All parameters are kernel addresses.
2994 * @return number of msrs set successfully.
2996 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2997 struct kvm_msr_entry *entries,
2998 int (*do_msr)(struct kvm_vcpu *vcpu,
2999 unsigned index, u64 *data))
3003 for (i = 0; i < msrs->nmsrs; ++i)
3004 if (do_msr(vcpu, entries[i].index, &entries[i].data))
3011 * Read or write a bunch of msrs. Parameters are user addresses.
3013 * @return number of msrs set successfully.
3015 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
3016 int (*do_msr)(struct kvm_vcpu *vcpu,
3017 unsigned index, u64 *data),
3020 struct kvm_msrs msrs;
3021 struct kvm_msr_entry *entries;
3026 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
3030 if (msrs.nmsrs >= MAX_IO_MSRS)
3033 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
3034 entries = memdup_user(user_msrs->entries, size);
3035 if (IS_ERR(entries)) {
3036 r = PTR_ERR(entries);
3040 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
3045 if (writeback && copy_to_user(user_msrs->entries, entries, size))
3056 static inline bool kvm_can_mwait_in_guest(void)
3058 return boot_cpu_has(X86_FEATURE_MWAIT) &&
3059 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
3060 boot_cpu_has(X86_FEATURE_ARAT);
3063 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
3068 case KVM_CAP_IRQCHIP:
3070 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
3071 case KVM_CAP_SET_TSS_ADDR:
3072 case KVM_CAP_EXT_CPUID:
3073 case KVM_CAP_EXT_EMUL_CPUID:
3074 case KVM_CAP_CLOCKSOURCE:
3076 case KVM_CAP_NOP_IO_DELAY:
3077 case KVM_CAP_MP_STATE:
3078 case KVM_CAP_SYNC_MMU:
3079 case KVM_CAP_USER_NMI:
3080 case KVM_CAP_REINJECT_CONTROL:
3081 case KVM_CAP_IRQ_INJECT_STATUS:
3082 case KVM_CAP_IOEVENTFD:
3083 case KVM_CAP_IOEVENTFD_NO_LENGTH:
3085 case KVM_CAP_PIT_STATE2:
3086 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
3087 case KVM_CAP_XEN_HVM:
3088 case KVM_CAP_VCPU_EVENTS:
3089 case KVM_CAP_HYPERV:
3090 case KVM_CAP_HYPERV_VAPIC:
3091 case KVM_CAP_HYPERV_SPIN:
3092 case KVM_CAP_HYPERV_SYNIC:
3093 case KVM_CAP_HYPERV_SYNIC2:
3094 case KVM_CAP_HYPERV_VP_INDEX:
3095 case KVM_CAP_HYPERV_EVENTFD:
3096 case KVM_CAP_HYPERV_TLBFLUSH:
3097 case KVM_CAP_HYPERV_SEND_IPI:
3098 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3099 case KVM_CAP_HYPERV_CPUID:
3100 case KVM_CAP_PCI_SEGMENT:
3101 case KVM_CAP_DEBUGREGS:
3102 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3104 case KVM_CAP_ASYNC_PF:
3105 case KVM_CAP_GET_TSC_KHZ:
3106 case KVM_CAP_KVMCLOCK_CTRL:
3107 case KVM_CAP_READONLY_MEM:
3108 case KVM_CAP_HYPERV_TIME:
3109 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3110 case KVM_CAP_TSC_DEADLINE_TIMER:
3111 case KVM_CAP_DISABLE_QUIRKS:
3112 case KVM_CAP_SET_BOOT_CPU_ID:
3113 case KVM_CAP_SPLIT_IRQCHIP:
3114 case KVM_CAP_IMMEDIATE_EXIT:
3115 case KVM_CAP_GET_MSR_FEATURES:
3116 case KVM_CAP_MSR_PLATFORM_INFO:
3117 case KVM_CAP_EXCEPTION_PAYLOAD:
3120 case KVM_CAP_SYNC_REGS:
3121 r = KVM_SYNC_X86_VALID_FIELDS;
3123 case KVM_CAP_ADJUST_CLOCK:
3124 r = KVM_CLOCK_TSC_STABLE;
3126 case KVM_CAP_X86_DISABLE_EXITS:
3127 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE |
3128 KVM_X86_DISABLE_EXITS_CSTATE;
3129 if(kvm_can_mwait_in_guest())
3130 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3132 case KVM_CAP_X86_SMM:
3133 /* SMBASE is usually relocated above 1M on modern chipsets,
3134 * and SMM handlers might indeed rely on 4G segment limits,
3135 * so do not report SMM to be available if real mode is
3136 * emulated via vm86 mode. Still, do not go to great lengths
3137 * to avoid userspace's usage of the feature, because it is a
3138 * fringe case that is not enabled except via specific settings
3139 * of the module parameters.
3141 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3144 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3146 case KVM_CAP_NR_VCPUS:
3147 r = KVM_SOFT_MAX_VCPUS;
3149 case KVM_CAP_MAX_VCPUS:
3152 case KVM_CAP_MAX_VCPU_ID:
3153 r = KVM_MAX_VCPU_ID;
3155 case KVM_CAP_PV_MMU: /* obsolete */
3159 r = KVM_MAX_MCE_BANKS;
3162 r = boot_cpu_has(X86_FEATURE_XSAVE);
3164 case KVM_CAP_TSC_CONTROL:
3165 r = kvm_has_tsc_control;
3167 case KVM_CAP_X2APIC_API:
3168 r = KVM_X2APIC_API_VALID_FLAGS;
3170 case KVM_CAP_NESTED_STATE:
3171 r = kvm_x86_ops->get_nested_state ?
3172 kvm_x86_ops->get_nested_state(NULL, NULL, 0) : 0;
3181 long kvm_arch_dev_ioctl(struct file *filp,
3182 unsigned int ioctl, unsigned long arg)
3184 void __user *argp = (void __user *)arg;
3188 case KVM_GET_MSR_INDEX_LIST: {
3189 struct kvm_msr_list __user *user_msr_list = argp;
3190 struct kvm_msr_list msr_list;
3194 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3197 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3198 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3201 if (n < msr_list.nmsrs)
3204 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3205 num_msrs_to_save * sizeof(u32)))
3207 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3209 num_emulated_msrs * sizeof(u32)))
3214 case KVM_GET_SUPPORTED_CPUID:
3215 case KVM_GET_EMULATED_CPUID: {
3216 struct kvm_cpuid2 __user *cpuid_arg = argp;
3217 struct kvm_cpuid2 cpuid;
3220 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3223 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3229 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3234 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3236 if (copy_to_user(argp, &kvm_mce_cap_supported,
3237 sizeof(kvm_mce_cap_supported)))
3241 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3242 struct kvm_msr_list __user *user_msr_list = argp;
3243 struct kvm_msr_list msr_list;
3247 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3250 msr_list.nmsrs = num_msr_based_features;
3251 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3254 if (n < msr_list.nmsrs)
3257 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3258 num_msr_based_features * sizeof(u32)))
3264 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3274 static void wbinvd_ipi(void *garbage)
3279 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3281 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3284 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3286 /* Address WBINVD may be executed by guest */
3287 if (need_emulate_wbinvd(vcpu)) {
3288 if (kvm_x86_ops->has_wbinvd_exit())
3289 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3290 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3291 smp_call_function_single(vcpu->cpu,
3292 wbinvd_ipi, NULL, 1);
3295 kvm_x86_ops->vcpu_load(vcpu, cpu);
3297 /* Apply any externally detected TSC adjustments (due to suspend) */
3298 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3299 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3300 vcpu->arch.tsc_offset_adjustment = 0;
3301 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3304 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3305 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3306 rdtsc() - vcpu->arch.last_host_tsc;
3308 mark_tsc_unstable("KVM discovered backwards TSC");
3310 if (kvm_check_tsc_unstable()) {
3311 u64 offset = kvm_compute_tsc_offset(vcpu,
3312 vcpu->arch.last_guest_tsc);
3313 kvm_vcpu_write_tsc_offset(vcpu, offset);
3314 vcpu->arch.tsc_catchup = 1;
3317 if (kvm_lapic_hv_timer_in_use(vcpu))
3318 kvm_lapic_restart_hv_timer(vcpu);
3321 * On a host with synchronized TSC, there is no need to update
3322 * kvmclock on vcpu->cpu migration
3324 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3325 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3326 if (vcpu->cpu != cpu)
3327 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3331 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3334 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3336 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3339 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3341 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3342 &vcpu->arch.st.steal.preempted,
3343 offsetof(struct kvm_steal_time, preempted),
3344 sizeof(vcpu->arch.st.steal.preempted));
3347 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3351 if (vcpu->preempted)
3352 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3355 * Disable page faults because we're in atomic context here.
3356 * kvm_write_guest_offset_cached() would call might_fault()
3357 * that relies on pagefault_disable() to tell if there's a
3358 * bug. NOTE: the write to guest memory may not go through if
3359 * during postcopy live migration or if there's heavy guest
3362 pagefault_disable();
3364 * kvm_memslots() will be called by
3365 * kvm_write_guest_offset_cached() so take the srcu lock.
3367 idx = srcu_read_lock(&vcpu->kvm->srcu);
3368 kvm_steal_time_set_preempted(vcpu);
3369 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3371 kvm_x86_ops->vcpu_put(vcpu);
3372 vcpu->arch.last_host_tsc = rdtsc();
3374 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3375 * on every vmexit, but if not, we might have a stale dr6 from the
3376 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3381 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3382 struct kvm_lapic_state *s)
3384 if (vcpu->arch.apicv_active)
3385 kvm_x86_ops->sync_pir_to_irr(vcpu);
3387 return kvm_apic_get_state(vcpu, s);
3390 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3391 struct kvm_lapic_state *s)
3395 r = kvm_apic_set_state(vcpu, s);
3398 update_cr8_intercept(vcpu);
3403 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3405 return (!lapic_in_kernel(vcpu) ||
3406 kvm_apic_accept_pic_intr(vcpu));
3410 * if userspace requested an interrupt window, check that the
3411 * interrupt window is open.
3413 * No need to exit to userspace if we already have an interrupt queued.
3415 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3417 return kvm_arch_interrupt_allowed(vcpu) &&
3418 !kvm_cpu_has_interrupt(vcpu) &&
3419 !kvm_event_needs_reinjection(vcpu) &&
3420 kvm_cpu_accept_dm_intr(vcpu);
3423 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3424 struct kvm_interrupt *irq)
3426 if (irq->irq >= KVM_NR_INTERRUPTS)
3429 if (!irqchip_in_kernel(vcpu->kvm)) {
3430 kvm_queue_interrupt(vcpu, irq->irq, false);
3431 kvm_make_request(KVM_REQ_EVENT, vcpu);
3436 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3437 * fail for in-kernel 8259.
3439 if (pic_in_kernel(vcpu->kvm))
3442 if (vcpu->arch.pending_external_vector != -1)
3445 vcpu->arch.pending_external_vector = irq->irq;
3446 kvm_make_request(KVM_REQ_EVENT, vcpu);
3450 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3452 kvm_inject_nmi(vcpu);
3457 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3459 kvm_make_request(KVM_REQ_SMI, vcpu);
3464 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3465 struct kvm_tpr_access_ctl *tac)
3469 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3473 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3477 unsigned bank_num = mcg_cap & 0xff, bank;
3480 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3482 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3485 vcpu->arch.mcg_cap = mcg_cap;
3486 /* Init IA32_MCG_CTL to all 1s */
3487 if (mcg_cap & MCG_CTL_P)
3488 vcpu->arch.mcg_ctl = ~(u64)0;
3489 /* Init IA32_MCi_CTL to all 1s */
3490 for (bank = 0; bank < bank_num; bank++)
3491 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3493 if (kvm_x86_ops->setup_mce)
3494 kvm_x86_ops->setup_mce(vcpu);
3499 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3500 struct kvm_x86_mce *mce)
3502 u64 mcg_cap = vcpu->arch.mcg_cap;
3503 unsigned bank_num = mcg_cap & 0xff;
3504 u64 *banks = vcpu->arch.mce_banks;
3506 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3509 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3510 * reporting is disabled
3512 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3513 vcpu->arch.mcg_ctl != ~(u64)0)
3515 banks += 4 * mce->bank;
3517 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3518 * reporting is disabled for the bank
3520 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3522 if (mce->status & MCI_STATUS_UC) {
3523 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3524 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3525 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3528 if (banks[1] & MCI_STATUS_VAL)
3529 mce->status |= MCI_STATUS_OVER;
3530 banks[2] = mce->addr;
3531 banks[3] = mce->misc;
3532 vcpu->arch.mcg_status = mce->mcg_status;
3533 banks[1] = mce->status;
3534 kvm_queue_exception(vcpu, MC_VECTOR);
3535 } else if (!(banks[1] & MCI_STATUS_VAL)
3536 || !(banks[1] & MCI_STATUS_UC)) {
3537 if (banks[1] & MCI_STATUS_VAL)
3538 mce->status |= MCI_STATUS_OVER;
3539 banks[2] = mce->addr;
3540 banks[3] = mce->misc;
3541 banks[1] = mce->status;
3543 banks[1] |= MCI_STATUS_OVER;
3547 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3548 struct kvm_vcpu_events *events)
3553 * The API doesn't provide the instruction length for software
3554 * exceptions, so don't report them. As long as the guest RIP
3555 * isn't advanced, we should expect to encounter the exception
3558 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3559 events->exception.injected = 0;
3560 events->exception.pending = 0;
3562 events->exception.injected = vcpu->arch.exception.injected;
3563 events->exception.pending = vcpu->arch.exception.pending;
3565 * For ABI compatibility, deliberately conflate
3566 * pending and injected exceptions when
3567 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3569 if (!vcpu->kvm->arch.exception_payload_enabled)
3570 events->exception.injected |=
3571 vcpu->arch.exception.pending;
3573 events->exception.nr = vcpu->arch.exception.nr;
3574 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3575 events->exception.error_code = vcpu->arch.exception.error_code;
3576 events->exception_has_payload = vcpu->arch.exception.has_payload;
3577 events->exception_payload = vcpu->arch.exception.payload;
3579 events->interrupt.injected =
3580 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3581 events->interrupt.nr = vcpu->arch.interrupt.nr;
3582 events->interrupt.soft = 0;
3583 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3585 events->nmi.injected = vcpu->arch.nmi_injected;
3586 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3587 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3588 events->nmi.pad = 0;
3590 events->sipi_vector = 0; /* never valid when reporting to user space */
3592 events->smi.smm = is_smm(vcpu);
3593 events->smi.pending = vcpu->arch.smi_pending;
3594 events->smi.smm_inside_nmi =
3595 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3596 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3598 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3599 | KVM_VCPUEVENT_VALID_SHADOW
3600 | KVM_VCPUEVENT_VALID_SMM);
3601 if (vcpu->kvm->arch.exception_payload_enabled)
3602 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3604 memset(&events->reserved, 0, sizeof(events->reserved));
3607 static void kvm_smm_changed(struct kvm_vcpu *vcpu);
3609 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3610 struct kvm_vcpu_events *events)
3612 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3613 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3614 | KVM_VCPUEVENT_VALID_SHADOW
3615 | KVM_VCPUEVENT_VALID_SMM
3616 | KVM_VCPUEVENT_VALID_PAYLOAD))
3619 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3620 if (!vcpu->kvm->arch.exception_payload_enabled)
3622 if (events->exception.pending)
3623 events->exception.injected = 0;
3625 events->exception_has_payload = 0;
3627 events->exception.pending = 0;
3628 events->exception_has_payload = 0;
3631 if ((events->exception.injected || events->exception.pending) &&
3632 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3635 /* INITs are latched while in SMM */
3636 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3637 (events->smi.smm || events->smi.pending) &&
3638 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3642 vcpu->arch.exception.injected = events->exception.injected;
3643 vcpu->arch.exception.pending = events->exception.pending;
3644 vcpu->arch.exception.nr = events->exception.nr;
3645 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3646 vcpu->arch.exception.error_code = events->exception.error_code;
3647 vcpu->arch.exception.has_payload = events->exception_has_payload;
3648 vcpu->arch.exception.payload = events->exception_payload;
3650 vcpu->arch.interrupt.injected = events->interrupt.injected;
3651 vcpu->arch.interrupt.nr = events->interrupt.nr;
3652 vcpu->arch.interrupt.soft = events->interrupt.soft;
3653 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3654 kvm_x86_ops->set_interrupt_shadow(vcpu,
3655 events->interrupt.shadow);
3657 vcpu->arch.nmi_injected = events->nmi.injected;
3658 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3659 vcpu->arch.nmi_pending = events->nmi.pending;
3660 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3662 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3663 lapic_in_kernel(vcpu))
3664 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3666 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3667 if (!!(vcpu->arch.hflags & HF_SMM_MASK) != events->smi.smm) {
3668 if (events->smi.smm)
3669 vcpu->arch.hflags |= HF_SMM_MASK;
3671 vcpu->arch.hflags &= ~HF_SMM_MASK;
3672 kvm_smm_changed(vcpu);
3675 vcpu->arch.smi_pending = events->smi.pending;
3677 if (events->smi.smm) {
3678 if (events->smi.smm_inside_nmi)
3679 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3681 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3682 if (lapic_in_kernel(vcpu)) {
3683 if (events->smi.latched_init)
3684 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3686 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3691 kvm_make_request(KVM_REQ_EVENT, vcpu);
3696 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3697 struct kvm_debugregs *dbgregs)
3701 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3702 kvm_get_dr(vcpu, 6, &val);
3704 dbgregs->dr7 = vcpu->arch.dr7;
3706 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3709 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3710 struct kvm_debugregs *dbgregs)
3715 if (dbgregs->dr6 & ~0xffffffffull)
3717 if (dbgregs->dr7 & ~0xffffffffull)
3720 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3721 kvm_update_dr0123(vcpu);
3722 vcpu->arch.dr6 = dbgregs->dr6;
3723 kvm_update_dr6(vcpu);
3724 vcpu->arch.dr7 = dbgregs->dr7;
3725 kvm_update_dr7(vcpu);
3730 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3732 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3734 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3735 u64 xstate_bv = xsave->header.xfeatures;
3739 * Copy legacy XSAVE area, to avoid complications with CPUID
3740 * leaves 0 and 1 in the loop below.
3742 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3745 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3746 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3749 * Copy each region from the possibly compacted offset to the
3750 * non-compacted offset.
3752 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3754 u64 xfeature_mask = valid & -valid;
3755 int xfeature_nr = fls64(xfeature_mask) - 1;
3756 void *src = get_xsave_addr(xsave, xfeature_nr);
3759 u32 size, offset, ecx, edx;
3760 cpuid_count(XSTATE_CPUID, xfeature_nr,
3761 &size, &offset, &ecx, &edx);
3762 if (xfeature_nr == XFEATURE_PKRU)
3763 memcpy(dest + offset, &vcpu->arch.pkru,
3764 sizeof(vcpu->arch.pkru));
3766 memcpy(dest + offset, src, size);
3770 valid -= xfeature_mask;
3774 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3776 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3777 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3781 * Copy legacy XSAVE area, to avoid complications with CPUID
3782 * leaves 0 and 1 in the loop below.
3784 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3786 /* Set XSTATE_BV and possibly XCOMP_BV. */
3787 xsave->header.xfeatures = xstate_bv;
3788 if (boot_cpu_has(X86_FEATURE_XSAVES))
3789 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3792 * Copy each region from the non-compacted offset to the
3793 * possibly compacted offset.
3795 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3797 u64 xfeature_mask = valid & -valid;
3798 int xfeature_nr = fls64(xfeature_mask) - 1;
3799 void *dest = get_xsave_addr(xsave, xfeature_nr);
3802 u32 size, offset, ecx, edx;
3803 cpuid_count(XSTATE_CPUID, xfeature_nr,
3804 &size, &offset, &ecx, &edx);
3805 if (xfeature_nr == XFEATURE_PKRU)
3806 memcpy(&vcpu->arch.pkru, src + offset,
3807 sizeof(vcpu->arch.pkru));
3809 memcpy(dest, src + offset, size);
3812 valid -= xfeature_mask;
3816 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3817 struct kvm_xsave *guest_xsave)
3819 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3820 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3821 fill_xsave((u8 *) guest_xsave->region, vcpu);
3823 memcpy(guest_xsave->region,
3824 &vcpu->arch.guest_fpu->state.fxsave,
3825 sizeof(struct fxregs_state));
3826 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3827 XFEATURE_MASK_FPSSE;
3831 #define XSAVE_MXCSR_OFFSET 24
3833 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3834 struct kvm_xsave *guest_xsave)
3837 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3838 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3840 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3842 * Here we allow setting states that are not present in
3843 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3844 * with old userspace.
3846 if (xstate_bv & ~kvm_supported_xcr0() ||
3847 mxcsr & ~mxcsr_feature_mask)
3849 load_xsave(vcpu, (u8 *)guest_xsave->region);
3851 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3852 mxcsr & ~mxcsr_feature_mask)
3854 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3855 guest_xsave->region, sizeof(struct fxregs_state));
3860 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3861 struct kvm_xcrs *guest_xcrs)
3863 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3864 guest_xcrs->nr_xcrs = 0;
3868 guest_xcrs->nr_xcrs = 1;
3869 guest_xcrs->flags = 0;
3870 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3871 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3874 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3875 struct kvm_xcrs *guest_xcrs)
3879 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3882 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3885 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3886 /* Only support XCR0 currently */
3887 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3888 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3889 guest_xcrs->xcrs[i].value);
3898 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3899 * stopped by the hypervisor. This function will be called from the host only.
3900 * EINVAL is returned when the host attempts to set the flag for a guest that
3901 * does not support pv clocks.
3903 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3905 if (!vcpu->arch.pv_time_enabled)
3907 vcpu->arch.pvclock_set_guest_stopped_request = true;
3908 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3912 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3913 struct kvm_enable_cap *cap)
3916 uint16_t vmcs_version;
3917 void __user *user_ptr;
3923 case KVM_CAP_HYPERV_SYNIC2:
3928 case KVM_CAP_HYPERV_SYNIC:
3929 if (!irqchip_in_kernel(vcpu->kvm))
3931 return kvm_hv_activate_synic(vcpu, cap->cap ==
3932 KVM_CAP_HYPERV_SYNIC2);
3933 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3934 if (!kvm_x86_ops->nested_enable_evmcs)
3936 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
3938 user_ptr = (void __user *)(uintptr_t)cap->args[0];
3939 if (copy_to_user(user_ptr, &vmcs_version,
3940 sizeof(vmcs_version)))
3950 long kvm_arch_vcpu_ioctl(struct file *filp,
3951 unsigned int ioctl, unsigned long arg)
3953 struct kvm_vcpu *vcpu = filp->private_data;
3954 void __user *argp = (void __user *)arg;
3957 struct kvm_lapic_state *lapic;
3958 struct kvm_xsave *xsave;
3959 struct kvm_xcrs *xcrs;
3967 case KVM_GET_LAPIC: {
3969 if (!lapic_in_kernel(vcpu))
3971 u.lapic = kzalloc(sizeof(struct kvm_lapic_state),
3972 GFP_KERNEL_ACCOUNT);
3977 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3981 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3986 case KVM_SET_LAPIC: {
3988 if (!lapic_in_kernel(vcpu))
3990 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3991 if (IS_ERR(u.lapic)) {
3992 r = PTR_ERR(u.lapic);
3996 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3999 case KVM_INTERRUPT: {
4000 struct kvm_interrupt irq;
4003 if (copy_from_user(&irq, argp, sizeof(irq)))
4005 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
4009 r = kvm_vcpu_ioctl_nmi(vcpu);
4013 r = kvm_vcpu_ioctl_smi(vcpu);
4016 case KVM_SET_CPUID: {
4017 struct kvm_cpuid __user *cpuid_arg = argp;
4018 struct kvm_cpuid cpuid;
4021 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4023 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
4026 case KVM_SET_CPUID2: {
4027 struct kvm_cpuid2 __user *cpuid_arg = argp;
4028 struct kvm_cpuid2 cpuid;
4031 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4033 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
4034 cpuid_arg->entries);
4037 case KVM_GET_CPUID2: {
4038 struct kvm_cpuid2 __user *cpuid_arg = argp;
4039 struct kvm_cpuid2 cpuid;
4042 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4044 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
4045 cpuid_arg->entries);
4049 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4054 case KVM_GET_MSRS: {
4055 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4056 r = msr_io(vcpu, argp, do_get_msr, 1);
4057 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4060 case KVM_SET_MSRS: {
4061 int idx = srcu_read_lock(&vcpu->kvm->srcu);
4062 r = msr_io(vcpu, argp, do_set_msr, 0);
4063 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4066 case KVM_TPR_ACCESS_REPORTING: {
4067 struct kvm_tpr_access_ctl tac;
4070 if (copy_from_user(&tac, argp, sizeof(tac)))
4072 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
4076 if (copy_to_user(argp, &tac, sizeof(tac)))
4081 case KVM_SET_VAPIC_ADDR: {
4082 struct kvm_vapic_addr va;
4086 if (!lapic_in_kernel(vcpu))
4089 if (copy_from_user(&va, argp, sizeof(va)))
4091 idx = srcu_read_lock(&vcpu->kvm->srcu);
4092 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
4093 srcu_read_unlock(&vcpu->kvm->srcu, idx);
4096 case KVM_X86_SETUP_MCE: {
4100 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4102 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4105 case KVM_X86_SET_MCE: {
4106 struct kvm_x86_mce mce;
4109 if (copy_from_user(&mce, argp, sizeof(mce)))
4111 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4114 case KVM_GET_VCPU_EVENTS: {
4115 struct kvm_vcpu_events events;
4117 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4120 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4125 case KVM_SET_VCPU_EVENTS: {
4126 struct kvm_vcpu_events events;
4129 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4132 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4135 case KVM_GET_DEBUGREGS: {
4136 struct kvm_debugregs dbgregs;
4138 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4141 if (copy_to_user(argp, &dbgregs,
4142 sizeof(struct kvm_debugregs)))
4147 case KVM_SET_DEBUGREGS: {
4148 struct kvm_debugregs dbgregs;
4151 if (copy_from_user(&dbgregs, argp,
4152 sizeof(struct kvm_debugregs)))
4155 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4158 case KVM_GET_XSAVE: {
4159 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL_ACCOUNT);
4164 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4167 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4172 case KVM_SET_XSAVE: {
4173 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4174 if (IS_ERR(u.xsave)) {
4175 r = PTR_ERR(u.xsave);
4179 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4182 case KVM_GET_XCRS: {
4183 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL_ACCOUNT);
4188 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4191 if (copy_to_user(argp, u.xcrs,
4192 sizeof(struct kvm_xcrs)))
4197 case KVM_SET_XCRS: {
4198 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4199 if (IS_ERR(u.xcrs)) {
4200 r = PTR_ERR(u.xcrs);
4204 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4207 case KVM_SET_TSC_KHZ: {
4211 user_tsc_khz = (u32)arg;
4213 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4216 if (user_tsc_khz == 0)
4217 user_tsc_khz = tsc_khz;
4219 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4224 case KVM_GET_TSC_KHZ: {
4225 r = vcpu->arch.virtual_tsc_khz;
4228 case KVM_KVMCLOCK_CTRL: {
4229 r = kvm_set_guest_paused(vcpu);
4232 case KVM_ENABLE_CAP: {
4233 struct kvm_enable_cap cap;
4236 if (copy_from_user(&cap, argp, sizeof(cap)))
4238 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4241 case KVM_GET_NESTED_STATE: {
4242 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4246 if (!kvm_x86_ops->get_nested_state)
4249 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4251 if (get_user(user_data_size, &user_kvm_nested_state->size))
4254 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4259 if (r > user_data_size) {
4260 if (put_user(r, &user_kvm_nested_state->size))
4270 case KVM_SET_NESTED_STATE: {
4271 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4272 struct kvm_nested_state kvm_state;
4275 if (!kvm_x86_ops->set_nested_state)
4279 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4283 if (kvm_state.size < sizeof(kvm_state))
4286 if (kvm_state.flags &
4287 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4288 | KVM_STATE_NESTED_EVMCS))
4291 /* nested_run_pending implies guest_mode. */
4292 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4293 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4296 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4299 case KVM_GET_SUPPORTED_HV_CPUID: {
4300 struct kvm_cpuid2 __user *cpuid_arg = argp;
4301 struct kvm_cpuid2 cpuid;
4304 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4307 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4308 cpuid_arg->entries);
4313 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4328 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4330 return VM_FAULT_SIGBUS;
4333 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4337 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4339 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4343 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4346 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4349 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4350 unsigned long kvm_nr_mmu_pages)
4352 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4355 mutex_lock(&kvm->slots_lock);
4357 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4358 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4360 mutex_unlock(&kvm->slots_lock);
4364 static unsigned long kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4366 return kvm->arch.n_max_mmu_pages;
4369 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4371 struct kvm_pic *pic = kvm->arch.vpic;
4375 switch (chip->chip_id) {
4376 case KVM_IRQCHIP_PIC_MASTER:
4377 memcpy(&chip->chip.pic, &pic->pics[0],
4378 sizeof(struct kvm_pic_state));
4380 case KVM_IRQCHIP_PIC_SLAVE:
4381 memcpy(&chip->chip.pic, &pic->pics[1],
4382 sizeof(struct kvm_pic_state));
4384 case KVM_IRQCHIP_IOAPIC:
4385 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4394 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4396 struct kvm_pic *pic = kvm->arch.vpic;
4400 switch (chip->chip_id) {
4401 case KVM_IRQCHIP_PIC_MASTER:
4402 spin_lock(&pic->lock);
4403 memcpy(&pic->pics[0], &chip->chip.pic,
4404 sizeof(struct kvm_pic_state));
4405 spin_unlock(&pic->lock);
4407 case KVM_IRQCHIP_PIC_SLAVE:
4408 spin_lock(&pic->lock);
4409 memcpy(&pic->pics[1], &chip->chip.pic,
4410 sizeof(struct kvm_pic_state));
4411 spin_unlock(&pic->lock);
4413 case KVM_IRQCHIP_IOAPIC:
4414 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4420 kvm_pic_update_irq(pic);
4424 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4426 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4428 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4430 mutex_lock(&kps->lock);
4431 memcpy(ps, &kps->channels, sizeof(*ps));
4432 mutex_unlock(&kps->lock);
4436 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4439 struct kvm_pit *pit = kvm->arch.vpit;
4441 mutex_lock(&pit->pit_state.lock);
4442 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4443 for (i = 0; i < 3; i++)
4444 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4445 mutex_unlock(&pit->pit_state.lock);
4449 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4451 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4452 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4453 sizeof(ps->channels));
4454 ps->flags = kvm->arch.vpit->pit_state.flags;
4455 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4456 memset(&ps->reserved, 0, sizeof(ps->reserved));
4460 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4464 u32 prev_legacy, cur_legacy;
4465 struct kvm_pit *pit = kvm->arch.vpit;
4467 mutex_lock(&pit->pit_state.lock);
4468 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4469 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4470 if (!prev_legacy && cur_legacy)
4472 memcpy(&pit->pit_state.channels, &ps->channels,
4473 sizeof(pit->pit_state.channels));
4474 pit->pit_state.flags = ps->flags;
4475 for (i = 0; i < 3; i++)
4476 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4478 mutex_unlock(&pit->pit_state.lock);
4482 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4483 struct kvm_reinject_control *control)
4485 struct kvm_pit *pit = kvm->arch.vpit;
4490 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4491 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4492 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4494 mutex_lock(&pit->pit_state.lock);
4495 kvm_pit_set_reinject(pit, control->pit_reinject);
4496 mutex_unlock(&pit->pit_state.lock);
4502 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4503 * @kvm: kvm instance
4504 * @log: slot id and address to which we copy the log
4506 * Steps 1-4 below provide general overview of dirty page logging. See
4507 * kvm_get_dirty_log_protect() function description for additional details.
4509 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4510 * always flush the TLB (step 4) even if previous step failed and the dirty
4511 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4512 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4513 * writes will be marked dirty for next log read.
4515 * 1. Take a snapshot of the bit and clear it if needed.
4516 * 2. Write protect the corresponding page.
4517 * 3. Copy the snapshot to the userspace.
4518 * 4. Flush TLB's if needed.
4520 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4525 mutex_lock(&kvm->slots_lock);
4528 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4530 if (kvm_x86_ops->flush_log_dirty)
4531 kvm_x86_ops->flush_log_dirty(kvm);
4533 r = kvm_get_dirty_log_protect(kvm, log, &flush);
4536 * All the TLBs can be flushed out of mmu lock, see the comments in
4537 * kvm_mmu_slot_remove_write_access().
4539 lockdep_assert_held(&kvm->slots_lock);
4541 kvm_flush_remote_tlbs(kvm);
4543 mutex_unlock(&kvm->slots_lock);
4547 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4552 mutex_lock(&kvm->slots_lock);
4555 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4557 if (kvm_x86_ops->flush_log_dirty)
4558 kvm_x86_ops->flush_log_dirty(kvm);
4560 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4563 * All the TLBs can be flushed out of mmu lock, see the comments in
4564 * kvm_mmu_slot_remove_write_access().
4566 lockdep_assert_held(&kvm->slots_lock);
4568 kvm_flush_remote_tlbs(kvm);
4570 mutex_unlock(&kvm->slots_lock);
4574 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4577 if (!irqchip_in_kernel(kvm))
4580 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4581 irq_event->irq, irq_event->level,
4586 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4587 struct kvm_enable_cap *cap)
4595 case KVM_CAP_DISABLE_QUIRKS:
4596 kvm->arch.disabled_quirks = cap->args[0];
4599 case KVM_CAP_SPLIT_IRQCHIP: {
4600 mutex_lock(&kvm->lock);
4602 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4603 goto split_irqchip_unlock;
4605 if (irqchip_in_kernel(kvm))
4606 goto split_irqchip_unlock;
4607 if (kvm->created_vcpus)
4608 goto split_irqchip_unlock;
4609 r = kvm_setup_empty_irq_routing(kvm);
4611 goto split_irqchip_unlock;
4612 /* Pairs with irqchip_in_kernel. */
4614 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4615 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4617 split_irqchip_unlock:
4618 mutex_unlock(&kvm->lock);
4621 case KVM_CAP_X2APIC_API:
4623 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4626 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4627 kvm->arch.x2apic_format = true;
4628 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4629 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4633 case KVM_CAP_X86_DISABLE_EXITS:
4635 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4638 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4639 kvm_can_mwait_in_guest())
4640 kvm->arch.mwait_in_guest = true;
4641 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4642 kvm->arch.hlt_in_guest = true;
4643 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4644 kvm->arch.pause_in_guest = true;
4645 if (cap->args[0] & KVM_X86_DISABLE_EXITS_CSTATE)
4646 kvm->arch.cstate_in_guest = true;
4649 case KVM_CAP_MSR_PLATFORM_INFO:
4650 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4653 case KVM_CAP_EXCEPTION_PAYLOAD:
4654 kvm->arch.exception_payload_enabled = cap->args[0];
4664 long kvm_arch_vm_ioctl(struct file *filp,
4665 unsigned int ioctl, unsigned long arg)
4667 struct kvm *kvm = filp->private_data;
4668 void __user *argp = (void __user *)arg;
4671 * This union makes it completely explicit to gcc-3.x
4672 * that these two variables' stack usage should be
4673 * combined, not added together.
4676 struct kvm_pit_state ps;
4677 struct kvm_pit_state2 ps2;
4678 struct kvm_pit_config pit_config;
4682 case KVM_SET_TSS_ADDR:
4683 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4685 case KVM_SET_IDENTITY_MAP_ADDR: {
4688 mutex_lock(&kvm->lock);
4690 if (kvm->created_vcpus)
4691 goto set_identity_unlock;
4693 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4694 goto set_identity_unlock;
4695 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4696 set_identity_unlock:
4697 mutex_unlock(&kvm->lock);
4700 case KVM_SET_NR_MMU_PAGES:
4701 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4703 case KVM_GET_NR_MMU_PAGES:
4704 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4706 case KVM_CREATE_IRQCHIP: {
4707 mutex_lock(&kvm->lock);
4710 if (irqchip_in_kernel(kvm))
4711 goto create_irqchip_unlock;
4714 if (kvm->created_vcpus)
4715 goto create_irqchip_unlock;
4717 r = kvm_pic_init(kvm);
4719 goto create_irqchip_unlock;
4721 r = kvm_ioapic_init(kvm);
4723 kvm_pic_destroy(kvm);
4724 goto create_irqchip_unlock;
4727 r = kvm_setup_default_irq_routing(kvm);
4729 kvm_ioapic_destroy(kvm);
4730 kvm_pic_destroy(kvm);
4731 goto create_irqchip_unlock;
4733 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4735 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4736 create_irqchip_unlock:
4737 mutex_unlock(&kvm->lock);
4740 case KVM_CREATE_PIT:
4741 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4743 case KVM_CREATE_PIT2:
4745 if (copy_from_user(&u.pit_config, argp,
4746 sizeof(struct kvm_pit_config)))
4749 mutex_lock(&kvm->lock);
4752 goto create_pit_unlock;
4754 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4758 mutex_unlock(&kvm->lock);
4760 case KVM_GET_IRQCHIP: {
4761 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4762 struct kvm_irqchip *chip;
4764 chip = memdup_user(argp, sizeof(*chip));
4771 if (!irqchip_kernel(kvm))
4772 goto get_irqchip_out;
4773 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4775 goto get_irqchip_out;
4777 if (copy_to_user(argp, chip, sizeof(*chip)))
4778 goto get_irqchip_out;
4784 case KVM_SET_IRQCHIP: {
4785 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4786 struct kvm_irqchip *chip;
4788 chip = memdup_user(argp, sizeof(*chip));
4795 if (!irqchip_kernel(kvm))
4796 goto set_irqchip_out;
4797 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4799 goto set_irqchip_out;
4807 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4810 if (!kvm->arch.vpit)
4812 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4816 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4823 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4826 if (!kvm->arch.vpit)
4828 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4831 case KVM_GET_PIT2: {
4833 if (!kvm->arch.vpit)
4835 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4839 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4844 case KVM_SET_PIT2: {
4846 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4849 if (!kvm->arch.vpit)
4851 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4854 case KVM_REINJECT_CONTROL: {
4855 struct kvm_reinject_control control;
4857 if (copy_from_user(&control, argp, sizeof(control)))
4859 r = kvm_vm_ioctl_reinject(kvm, &control);
4862 case KVM_SET_BOOT_CPU_ID:
4864 mutex_lock(&kvm->lock);
4865 if (kvm->created_vcpus)
4868 kvm->arch.bsp_vcpu_id = arg;
4869 mutex_unlock(&kvm->lock);
4871 case KVM_XEN_HVM_CONFIG: {
4872 struct kvm_xen_hvm_config xhc;
4874 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4879 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4883 case KVM_SET_CLOCK: {
4884 struct kvm_clock_data user_ns;
4888 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4897 * TODO: userspace has to take care of races with VCPU_RUN, so
4898 * kvm_gen_update_masterclock() can be cut down to locked
4899 * pvclock_update_vm_gtod_copy().
4901 kvm_gen_update_masterclock(kvm);
4902 now_ns = get_kvmclock_ns(kvm);
4903 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4904 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4907 case KVM_GET_CLOCK: {
4908 struct kvm_clock_data user_ns;
4911 now_ns = get_kvmclock_ns(kvm);
4912 user_ns.clock = now_ns;
4913 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4914 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4917 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4922 case KVM_MEMORY_ENCRYPT_OP: {
4924 if (kvm_x86_ops->mem_enc_op)
4925 r = kvm_x86_ops->mem_enc_op(kvm, argp);
4928 case KVM_MEMORY_ENCRYPT_REG_REGION: {
4929 struct kvm_enc_region region;
4932 if (copy_from_user(®ion, argp, sizeof(region)))
4936 if (kvm_x86_ops->mem_enc_reg_region)
4937 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
4940 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4941 struct kvm_enc_region region;
4944 if (copy_from_user(®ion, argp, sizeof(region)))
4948 if (kvm_x86_ops->mem_enc_unreg_region)
4949 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
4952 case KVM_HYPERV_EVENTFD: {
4953 struct kvm_hyperv_eventfd hvevfd;
4956 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4958 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4968 static void kvm_init_msr_list(void)
4973 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4974 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4978 * Even MSRs that are valid in the host may not be exposed
4979 * to the guests in some cases.
4981 switch (msrs_to_save[i]) {
4982 case MSR_IA32_BNDCFGS:
4983 if (!kvm_mpx_supported())
4987 if (!kvm_x86_ops->rdtscp_supported())
4990 case MSR_IA32_RTIT_CTL:
4991 case MSR_IA32_RTIT_STATUS:
4992 if (!kvm_x86_ops->pt_supported())
4995 case MSR_IA32_RTIT_CR3_MATCH:
4996 if (!kvm_x86_ops->pt_supported() ||
4997 !intel_pt_validate_hw_cap(PT_CAP_cr3_filtering))
5000 case MSR_IA32_RTIT_OUTPUT_BASE:
5001 case MSR_IA32_RTIT_OUTPUT_MASK:
5002 if (!kvm_x86_ops->pt_supported() ||
5003 (!intel_pt_validate_hw_cap(PT_CAP_topa_output) &&
5004 !intel_pt_validate_hw_cap(PT_CAP_single_range_output)))
5007 case MSR_IA32_RTIT_ADDR0_A ... MSR_IA32_RTIT_ADDR3_B: {
5008 if (!kvm_x86_ops->pt_supported() ||
5009 msrs_to_save[i] - MSR_IA32_RTIT_ADDR0_A >=
5010 intel_pt_validate_hw_cap(PT_CAP_num_address_ranges) * 2)
5019 msrs_to_save[j] = msrs_to_save[i];
5022 num_msrs_to_save = j;
5024 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
5025 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
5029 emulated_msrs[j] = emulated_msrs[i];
5032 num_emulated_msrs = j;
5034 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
5035 struct kvm_msr_entry msr;
5037 msr.index = msr_based_features[i];
5038 if (kvm_get_msr_feature(&msr))
5042 msr_based_features[j] = msr_based_features[i];
5045 num_msr_based_features = j;
5048 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
5056 if (!(lapic_in_kernel(vcpu) &&
5057 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
5058 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
5069 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
5076 if (!(lapic_in_kernel(vcpu) &&
5077 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
5079 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
5081 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
5091 static void kvm_set_segment(struct kvm_vcpu *vcpu,
5092 struct kvm_segment *var, int seg)
5094 kvm_x86_ops->set_segment(vcpu, var, seg);
5097 void kvm_get_segment(struct kvm_vcpu *vcpu,
5098 struct kvm_segment *var, int seg)
5100 kvm_x86_ops->get_segment(vcpu, var, seg);
5103 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
5104 struct x86_exception *exception)
5108 BUG_ON(!mmu_is_nested(vcpu));
5110 /* NPT walks are always user-walks */
5111 access |= PFERR_USER_MASK;
5112 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
5117 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
5118 struct x86_exception *exception)
5120 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5121 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5124 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
5125 struct x86_exception *exception)
5127 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5128 access |= PFERR_FETCH_MASK;
5129 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5132 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5133 struct x86_exception *exception)
5135 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5136 access |= PFERR_WRITE_MASK;
5137 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5140 /* uses this to access any guest's mapped memory without checking CPL */
5141 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5142 struct x86_exception *exception)
5144 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5147 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5148 struct kvm_vcpu *vcpu, u32 access,
5149 struct x86_exception *exception)
5152 int r = X86EMUL_CONTINUE;
5155 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5157 unsigned offset = addr & (PAGE_SIZE-1);
5158 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5161 if (gpa == UNMAPPED_GVA)
5162 return X86EMUL_PROPAGATE_FAULT;
5163 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5166 r = X86EMUL_IO_NEEDED;
5178 /* used for instruction fetching */
5179 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5180 gva_t addr, void *val, unsigned int bytes,
5181 struct x86_exception *exception)
5183 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5184 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5188 /* Inline kvm_read_guest_virt_helper for speed. */
5189 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5191 if (unlikely(gpa == UNMAPPED_GVA))
5192 return X86EMUL_PROPAGATE_FAULT;
5194 offset = addr & (PAGE_SIZE-1);
5195 if (WARN_ON(offset + bytes > PAGE_SIZE))
5196 bytes = (unsigned)PAGE_SIZE - offset;
5197 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5199 if (unlikely(ret < 0))
5200 return X86EMUL_IO_NEEDED;
5202 return X86EMUL_CONTINUE;
5205 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5206 gva_t addr, void *val, unsigned int bytes,
5207 struct x86_exception *exception)
5209 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5212 * FIXME: this should call handle_emulation_failure if X86EMUL_IO_NEEDED
5213 * is returned, but our callers are not ready for that and they blindly
5214 * call kvm_inject_page_fault. Ensure that they at least do not leak
5215 * uninitialized kernel stack memory into cr2 and error code.
5217 memset(exception, 0, sizeof(*exception));
5218 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5221 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5223 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5224 gva_t addr, void *val, unsigned int bytes,
5225 struct x86_exception *exception, bool system)
5227 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5230 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5231 access |= PFERR_USER_MASK;
5233 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5236 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5237 unsigned long addr, void *val, unsigned int bytes)
5239 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5240 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5242 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5245 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5246 struct kvm_vcpu *vcpu, u32 access,
5247 struct x86_exception *exception)
5250 int r = X86EMUL_CONTINUE;
5253 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5256 unsigned offset = addr & (PAGE_SIZE-1);
5257 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5260 if (gpa == UNMAPPED_GVA)
5261 return X86EMUL_PROPAGATE_FAULT;
5262 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5264 r = X86EMUL_IO_NEEDED;
5276 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5277 unsigned int bytes, struct x86_exception *exception,
5280 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5281 u32 access = PFERR_WRITE_MASK;
5283 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5284 access |= PFERR_USER_MASK;
5286 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5290 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5291 unsigned int bytes, struct x86_exception *exception)
5293 /* kvm_write_guest_virt_system can pull in tons of pages. */
5294 vcpu->arch.l1tf_flush_l1d = true;
5296 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5297 PFERR_WRITE_MASK, exception);
5299 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5301 int handle_ud(struct kvm_vcpu *vcpu)
5303 int emul_type = EMULTYPE_TRAP_UD;
5304 enum emulation_result er;
5305 char sig[5]; /* ud2; .ascii "kvm" */
5306 struct x86_exception e;
5308 if (force_emulation_prefix &&
5309 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5310 sig, sizeof(sig), &e) == 0 &&
5311 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5312 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5316 er = kvm_emulate_instruction(vcpu, emul_type);
5317 if (er == EMULATE_USER_EXIT)
5319 if (er != EMULATE_DONE)
5320 kvm_queue_exception(vcpu, UD_VECTOR);
5323 EXPORT_SYMBOL_GPL(handle_ud);
5325 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5326 gpa_t gpa, bool write)
5328 /* For APIC access vmexit */
5329 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5332 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5333 trace_vcpu_match_mmio(gva, gpa, write, true);
5340 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5341 gpa_t *gpa, struct x86_exception *exception,
5344 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5345 | (write ? PFERR_WRITE_MASK : 0);
5348 * currently PKRU is only applied to ept enabled guest so
5349 * there is no pkey in EPT page table for L1 guest or EPT
5350 * shadow page table for L2 guest.
5352 if (vcpu_match_mmio_gva(vcpu, gva)
5353 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5354 vcpu->arch.access, 0, access)) {
5355 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5356 (gva & (PAGE_SIZE - 1));
5357 trace_vcpu_match_mmio(gva, *gpa, write, false);
5361 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5363 if (*gpa == UNMAPPED_GVA)
5366 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5369 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5370 const void *val, int bytes)
5374 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5377 kvm_page_track_write(vcpu, gpa, val, bytes);
5381 struct read_write_emulator_ops {
5382 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5384 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5385 void *val, int bytes);
5386 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5387 int bytes, void *val);
5388 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5389 void *val, int bytes);
5393 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5395 if (vcpu->mmio_read_completed) {
5396 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5397 vcpu->mmio_fragments[0].gpa, val);
5398 vcpu->mmio_read_completed = 0;
5405 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5406 void *val, int bytes)
5408 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5411 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5412 void *val, int bytes)
5414 return emulator_write_phys(vcpu, gpa, val, bytes);
5417 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5419 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5420 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5423 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5424 void *val, int bytes)
5426 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5427 return X86EMUL_IO_NEEDED;
5430 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5431 void *val, int bytes)
5433 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5435 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5436 return X86EMUL_CONTINUE;
5439 static const struct read_write_emulator_ops read_emultor = {
5440 .read_write_prepare = read_prepare,
5441 .read_write_emulate = read_emulate,
5442 .read_write_mmio = vcpu_mmio_read,
5443 .read_write_exit_mmio = read_exit_mmio,
5446 static const struct read_write_emulator_ops write_emultor = {
5447 .read_write_emulate = write_emulate,
5448 .read_write_mmio = write_mmio,
5449 .read_write_exit_mmio = write_exit_mmio,
5453 static int emulator_read_write_onepage(unsigned long addr, void *val,
5455 struct x86_exception *exception,
5456 struct kvm_vcpu *vcpu,
5457 const struct read_write_emulator_ops *ops)
5461 bool write = ops->write;
5462 struct kvm_mmio_fragment *frag;
5463 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5466 * If the exit was due to a NPF we may already have a GPA.
5467 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5468 * Note, this cannot be used on string operations since string
5469 * operation using rep will only have the initial GPA from the NPF
5472 if (vcpu->arch.gpa_available &&
5473 emulator_can_use_gpa(ctxt) &&
5474 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5475 gpa = vcpu->arch.gpa_val;
5476 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5478 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5480 return X86EMUL_PROPAGATE_FAULT;
5483 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5484 return X86EMUL_CONTINUE;
5487 * Is this MMIO handled locally?
5489 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5490 if (handled == bytes)
5491 return X86EMUL_CONTINUE;
5497 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5498 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5502 return X86EMUL_CONTINUE;
5505 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5507 void *val, unsigned int bytes,
5508 struct x86_exception *exception,
5509 const struct read_write_emulator_ops *ops)
5511 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5515 if (ops->read_write_prepare &&
5516 ops->read_write_prepare(vcpu, val, bytes))
5517 return X86EMUL_CONTINUE;
5519 vcpu->mmio_nr_fragments = 0;
5521 /* Crossing a page boundary? */
5522 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5525 now = -addr & ~PAGE_MASK;
5526 rc = emulator_read_write_onepage(addr, val, now, exception,
5529 if (rc != X86EMUL_CONTINUE)
5532 if (ctxt->mode != X86EMUL_MODE_PROT64)
5538 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5540 if (rc != X86EMUL_CONTINUE)
5543 if (!vcpu->mmio_nr_fragments)
5546 gpa = vcpu->mmio_fragments[0].gpa;
5548 vcpu->mmio_needed = 1;
5549 vcpu->mmio_cur_fragment = 0;
5551 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5552 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5553 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5554 vcpu->run->mmio.phys_addr = gpa;
5556 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5559 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5563 struct x86_exception *exception)
5565 return emulator_read_write(ctxt, addr, val, bytes,
5566 exception, &read_emultor);
5569 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5573 struct x86_exception *exception)
5575 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5576 exception, &write_emultor);
5579 #define CMPXCHG_TYPE(t, ptr, old, new) \
5580 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5582 #ifdef CONFIG_X86_64
5583 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5585 # define CMPXCHG64(ptr, old, new) \
5586 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5589 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5594 struct x86_exception *exception)
5596 struct kvm_host_map map;
5597 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5602 /* guests cmpxchg8b have to be emulated atomically */
5603 if (bytes > 8 || (bytes & (bytes - 1)))
5606 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5608 if (gpa == UNMAPPED_GVA ||
5609 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5612 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5615 if (kvm_vcpu_map(vcpu, gpa_to_gfn(gpa), &map))
5618 kaddr = map.hva + offset_in_page(gpa);
5622 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5625 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5628 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5631 exchanged = CMPXCHG64(kaddr, old, new);
5637 kvm_vcpu_unmap(vcpu, &map, true);
5640 return X86EMUL_CMPXCHG_FAILED;
5642 kvm_page_track_write(vcpu, gpa, new, bytes);
5644 return X86EMUL_CONTINUE;
5647 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5649 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5652 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5656 for (i = 0; i < vcpu->arch.pio.count; i++) {
5657 if (vcpu->arch.pio.in)
5658 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5659 vcpu->arch.pio.size, pd);
5661 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5662 vcpu->arch.pio.port, vcpu->arch.pio.size,
5666 pd += vcpu->arch.pio.size;
5671 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5672 unsigned short port, void *val,
5673 unsigned int count, bool in)
5675 vcpu->arch.pio.port = port;
5676 vcpu->arch.pio.in = in;
5677 vcpu->arch.pio.count = count;
5678 vcpu->arch.pio.size = size;
5680 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5681 vcpu->arch.pio.count = 0;
5685 vcpu->run->exit_reason = KVM_EXIT_IO;
5686 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5687 vcpu->run->io.size = size;
5688 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5689 vcpu->run->io.count = count;
5690 vcpu->run->io.port = port;
5695 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5696 int size, unsigned short port, void *val,
5699 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5702 if (vcpu->arch.pio.count)
5705 memset(vcpu->arch.pio_data, 0, size * count);
5707 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5710 memcpy(val, vcpu->arch.pio_data, size * count);
5711 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5712 vcpu->arch.pio.count = 0;
5719 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5720 int size, unsigned short port,
5721 const void *val, unsigned int count)
5723 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5725 memcpy(vcpu->arch.pio_data, val, size * count);
5726 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5727 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5730 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5732 return kvm_x86_ops->get_segment_base(vcpu, seg);
5735 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5737 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5740 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5742 if (!need_emulate_wbinvd(vcpu))
5743 return X86EMUL_CONTINUE;
5745 if (kvm_x86_ops->has_wbinvd_exit()) {
5746 int cpu = get_cpu();
5748 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5749 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5750 wbinvd_ipi, NULL, 1);
5752 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5755 return X86EMUL_CONTINUE;
5758 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5760 kvm_emulate_wbinvd_noskip(vcpu);
5761 return kvm_skip_emulated_instruction(vcpu);
5763 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5767 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5769 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5772 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5773 unsigned long *dest)
5775 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5778 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5779 unsigned long value)
5782 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5785 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5787 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5790 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5792 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5793 unsigned long value;
5797 value = kvm_read_cr0(vcpu);
5800 value = vcpu->arch.cr2;
5803 value = kvm_read_cr3(vcpu);
5806 value = kvm_read_cr4(vcpu);
5809 value = kvm_get_cr8(vcpu);
5812 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5819 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5821 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5826 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5829 vcpu->arch.cr2 = val;
5832 res = kvm_set_cr3(vcpu, val);
5835 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5838 res = kvm_set_cr8(vcpu, val);
5841 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5848 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5850 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5853 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5855 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5858 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5860 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5863 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5865 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5868 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5870 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5873 static unsigned long emulator_get_cached_segment_base(
5874 struct x86_emulate_ctxt *ctxt, int seg)
5876 return get_segment_base(emul_to_vcpu(ctxt), seg);
5879 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5880 struct desc_struct *desc, u32 *base3,
5883 struct kvm_segment var;
5885 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5886 *selector = var.selector;
5889 memset(desc, 0, sizeof(*desc));
5897 set_desc_limit(desc, var.limit);
5898 set_desc_base(desc, (unsigned long)var.base);
5899 #ifdef CONFIG_X86_64
5901 *base3 = var.base >> 32;
5903 desc->type = var.type;
5905 desc->dpl = var.dpl;
5906 desc->p = var.present;
5907 desc->avl = var.avl;
5915 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5916 struct desc_struct *desc, u32 base3,
5919 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5920 struct kvm_segment var;
5922 var.selector = selector;
5923 var.base = get_desc_base(desc);
5924 #ifdef CONFIG_X86_64
5925 var.base |= ((u64)base3) << 32;
5927 var.limit = get_desc_limit(desc);
5929 var.limit = (var.limit << 12) | 0xfff;
5930 var.type = desc->type;
5931 var.dpl = desc->dpl;
5936 var.avl = desc->avl;
5937 var.present = desc->p;
5938 var.unusable = !var.present;
5941 kvm_set_segment(vcpu, &var, seg);
5945 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5946 u32 msr_index, u64 *pdata)
5948 struct msr_data msr;
5951 msr.index = msr_index;
5952 msr.host_initiated = false;
5953 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5961 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5962 u32 msr_index, u64 data)
5964 struct msr_data msr;
5967 msr.index = msr_index;
5968 msr.host_initiated = false;
5969 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5972 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5974 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5976 return vcpu->arch.smbase;
5979 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5981 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5983 vcpu->arch.smbase = smbase;
5986 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5989 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5992 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5993 u32 pmc, u64 *pdata)
5995 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5998 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
6000 emul_to_vcpu(ctxt)->arch.halt_request = 1;
6003 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
6004 struct x86_instruction_info *info,
6005 enum x86_intercept_stage stage)
6007 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
6010 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
6011 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
6013 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
6016 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
6018 return kvm_register_read(emul_to_vcpu(ctxt), reg);
6021 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
6023 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
6026 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
6028 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
6031 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
6033 return emul_to_vcpu(ctxt)->arch.hflags;
6036 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
6038 emul_to_vcpu(ctxt)->arch.hflags = emul_flags;
6041 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt,
6042 const char *smstate)
6044 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smstate);
6047 static void emulator_post_leave_smm(struct x86_emulate_ctxt *ctxt)
6049 kvm_smm_changed(emul_to_vcpu(ctxt));
6052 static const struct x86_emulate_ops emulate_ops = {
6053 .read_gpr = emulator_read_gpr,
6054 .write_gpr = emulator_write_gpr,
6055 .read_std = emulator_read_std,
6056 .write_std = emulator_write_std,
6057 .read_phys = kvm_read_guest_phys_system,
6058 .fetch = kvm_fetch_guest_virt,
6059 .read_emulated = emulator_read_emulated,
6060 .write_emulated = emulator_write_emulated,
6061 .cmpxchg_emulated = emulator_cmpxchg_emulated,
6062 .invlpg = emulator_invlpg,
6063 .pio_in_emulated = emulator_pio_in_emulated,
6064 .pio_out_emulated = emulator_pio_out_emulated,
6065 .get_segment = emulator_get_segment,
6066 .set_segment = emulator_set_segment,
6067 .get_cached_segment_base = emulator_get_cached_segment_base,
6068 .get_gdt = emulator_get_gdt,
6069 .get_idt = emulator_get_idt,
6070 .set_gdt = emulator_set_gdt,
6071 .set_idt = emulator_set_idt,
6072 .get_cr = emulator_get_cr,
6073 .set_cr = emulator_set_cr,
6074 .cpl = emulator_get_cpl,
6075 .get_dr = emulator_get_dr,
6076 .set_dr = emulator_set_dr,
6077 .get_smbase = emulator_get_smbase,
6078 .set_smbase = emulator_set_smbase,
6079 .set_msr = emulator_set_msr,
6080 .get_msr = emulator_get_msr,
6081 .check_pmc = emulator_check_pmc,
6082 .read_pmc = emulator_read_pmc,
6083 .halt = emulator_halt,
6084 .wbinvd = emulator_wbinvd,
6085 .fix_hypercall = emulator_fix_hypercall,
6086 .intercept = emulator_intercept,
6087 .get_cpuid = emulator_get_cpuid,
6088 .set_nmi_mask = emulator_set_nmi_mask,
6089 .get_hflags = emulator_get_hflags,
6090 .set_hflags = emulator_set_hflags,
6091 .pre_leave_smm = emulator_pre_leave_smm,
6092 .post_leave_smm = emulator_post_leave_smm,
6095 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
6097 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
6099 * an sti; sti; sequence only disable interrupts for the first
6100 * instruction. So, if the last instruction, be it emulated or
6101 * not, left the system with the INT_STI flag enabled, it
6102 * means that the last instruction is an sti. We should not
6103 * leave the flag on in this case. The same goes for mov ss
6105 if (int_shadow & mask)
6107 if (unlikely(int_shadow || mask)) {
6108 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
6110 kvm_make_request(KVM_REQ_EVENT, vcpu);
6114 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
6116 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6117 if (ctxt->exception.vector == PF_VECTOR)
6118 return kvm_propagate_fault(vcpu, &ctxt->exception);
6120 if (ctxt->exception.error_code_valid)
6121 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
6122 ctxt->exception.error_code);
6124 kvm_queue_exception(vcpu, ctxt->exception.vector);
6128 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
6130 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6133 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
6135 ctxt->eflags = kvm_get_rflags(vcpu);
6136 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
6138 ctxt->eip = kvm_rip_read(vcpu);
6139 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6140 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6141 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6142 cs_db ? X86EMUL_MODE_PROT32 :
6143 X86EMUL_MODE_PROT16;
6144 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6145 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6146 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6148 init_decode_cache(ctxt);
6149 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6152 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6154 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6157 init_emulate_ctxt(vcpu);
6161 ctxt->_eip = ctxt->eip + inc_eip;
6162 ret = emulate_int_real(ctxt, irq);
6164 if (ret != X86EMUL_CONTINUE)
6165 return EMULATE_FAIL;
6167 ctxt->eip = ctxt->_eip;
6168 kvm_rip_write(vcpu, ctxt->eip);
6169 kvm_set_rflags(vcpu, ctxt->eflags);
6171 return EMULATE_DONE;
6173 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6175 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6177 int r = EMULATE_DONE;
6179 ++vcpu->stat.insn_emulation_fail;
6180 trace_kvm_emulate_insn_failed(vcpu);
6182 if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6183 return EMULATE_FAIL;
6185 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6186 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6187 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6188 vcpu->run->internal.ndata = 0;
6189 r = EMULATE_USER_EXIT;
6192 kvm_queue_exception(vcpu, UD_VECTOR);
6197 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6198 bool write_fault_to_shadow_pgtable,
6204 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6207 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6210 if (!vcpu->arch.mmu->direct_map) {
6212 * Write permission should be allowed since only
6213 * write access need to be emulated.
6215 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6218 * If the mapping is invalid in guest, let cpu retry
6219 * it to generate fault.
6221 if (gpa == UNMAPPED_GVA)
6226 * Do not retry the unhandleable instruction if it faults on the
6227 * readonly host memory, otherwise it will goto a infinite loop:
6228 * retry instruction -> write #PF -> emulation fail -> retry
6229 * instruction -> ...
6231 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6234 * If the instruction failed on the error pfn, it can not be fixed,
6235 * report the error to userspace.
6237 if (is_error_noslot_pfn(pfn))
6240 kvm_release_pfn_clean(pfn);
6242 /* The instructions are well-emulated on direct mmu. */
6243 if (vcpu->arch.mmu->direct_map) {
6244 unsigned int indirect_shadow_pages;
6246 spin_lock(&vcpu->kvm->mmu_lock);
6247 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6248 spin_unlock(&vcpu->kvm->mmu_lock);
6250 if (indirect_shadow_pages)
6251 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6257 * if emulation was due to access to shadowed page table
6258 * and it failed try to unshadow page and re-enter the
6259 * guest to let CPU execute the instruction.
6261 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6264 * If the access faults on its page table, it can not
6265 * be fixed by unprotecting shadow page and it should
6266 * be reported to userspace.
6268 return !write_fault_to_shadow_pgtable;
6271 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6272 unsigned long cr2, int emulation_type)
6274 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6275 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6277 last_retry_eip = vcpu->arch.last_retry_eip;
6278 last_retry_addr = vcpu->arch.last_retry_addr;
6281 * If the emulation is caused by #PF and it is non-page_table
6282 * writing instruction, it means the VM-EXIT is caused by shadow
6283 * page protected, we can zap the shadow page and retry this
6284 * instruction directly.
6286 * Note: if the guest uses a non-page-table modifying instruction
6287 * on the PDE that points to the instruction, then we will unmap
6288 * the instruction and go to an infinite loop. So, we cache the
6289 * last retried eip and the last fault address, if we meet the eip
6290 * and the address again, we can break out of the potential infinite
6293 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6295 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6298 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6301 if (x86_page_table_writing_insn(ctxt))
6304 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6307 vcpu->arch.last_retry_eip = ctxt->eip;
6308 vcpu->arch.last_retry_addr = cr2;
6310 if (!vcpu->arch.mmu->direct_map)
6311 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6313 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6318 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6319 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6321 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6323 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6324 /* This is a good place to trace that we are exiting SMM. */
6325 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6327 /* Process a latched INIT or SMI, if any. */
6328 kvm_make_request(KVM_REQ_EVENT, vcpu);
6331 kvm_mmu_reset_context(vcpu);
6334 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6343 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6344 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6349 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6351 struct kvm_run *kvm_run = vcpu->run;
6353 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6354 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6355 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6356 kvm_run->debug.arch.exception = DB_VECTOR;
6357 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6358 *r = EMULATE_USER_EXIT;
6360 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6364 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6366 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6367 int r = EMULATE_DONE;
6369 kvm_x86_ops->skip_emulated_instruction(vcpu);
6372 * rflags is the old, "raw" value of the flags. The new value has
6373 * not been saved yet.
6375 * This is correct even for TF set by the guest, because "the
6376 * processor will not generate this exception after the instruction
6377 * that sets the TF flag".
6379 if (unlikely(rflags & X86_EFLAGS_TF))
6380 kvm_vcpu_do_singlestep(vcpu, &r);
6381 return r == EMULATE_DONE;
6383 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6385 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6387 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6388 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6389 struct kvm_run *kvm_run = vcpu->run;
6390 unsigned long eip = kvm_get_linear_rip(vcpu);
6391 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6392 vcpu->arch.guest_debug_dr7,
6396 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6397 kvm_run->debug.arch.pc = eip;
6398 kvm_run->debug.arch.exception = DB_VECTOR;
6399 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6400 *r = EMULATE_USER_EXIT;
6405 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6406 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6407 unsigned long eip = kvm_get_linear_rip(vcpu);
6408 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6413 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
6414 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6415 kvm_queue_exception(vcpu, DB_VECTOR);
6424 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6426 switch (ctxt->opcode_len) {
6433 case 0xe6: /* OUT */
6437 case 0x6c: /* INS */
6439 case 0x6e: /* OUTS */
6446 case 0x33: /* RDPMC */
6455 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6462 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6463 bool writeback = true;
6464 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6466 vcpu->arch.l1tf_flush_l1d = true;
6469 * Clear write_fault_to_shadow_pgtable here to ensure it is
6472 vcpu->arch.write_fault_to_shadow_pgtable = false;
6473 kvm_clear_exception_queue(vcpu);
6475 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6476 init_emulate_ctxt(vcpu);
6479 * We will reenter on the same instruction since
6480 * we do not set complete_userspace_io. This does not
6481 * handle watchpoints yet, those would be handled in
6484 if (!(emulation_type & EMULTYPE_SKIP) &&
6485 kvm_vcpu_check_breakpoint(vcpu, &r))
6488 ctxt->interruptibility = 0;
6489 ctxt->have_exception = false;
6490 ctxt->exception.vector = -1;
6491 ctxt->perm_ok = false;
6493 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6495 r = x86_decode_insn(ctxt, insn, insn_len);
6497 trace_kvm_emulate_insn_start(vcpu);
6498 ++vcpu->stat.insn_emulation;
6499 if (r != EMULATION_OK) {
6500 if (emulation_type & EMULTYPE_TRAP_UD)
6501 return EMULATE_FAIL;
6502 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6504 return EMULATE_DONE;
6505 if (ctxt->have_exception && inject_emulated_exception(vcpu))
6506 return EMULATE_DONE;
6507 if (emulation_type & EMULTYPE_SKIP)
6508 return EMULATE_FAIL;
6509 return handle_emulation_failure(vcpu, emulation_type);
6513 if ((emulation_type & EMULTYPE_VMWARE) &&
6514 !is_vmware_backdoor_opcode(ctxt))
6515 return EMULATE_FAIL;
6517 if (emulation_type & EMULTYPE_SKIP) {
6518 kvm_rip_write(vcpu, ctxt->_eip);
6519 if (ctxt->eflags & X86_EFLAGS_RF)
6520 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6521 return EMULATE_DONE;
6524 if (retry_instruction(ctxt, cr2, emulation_type))
6525 return EMULATE_DONE;
6527 /* this is needed for vmware backdoor interface to work since it
6528 changes registers values during IO operation */
6529 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6530 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6531 emulator_invalidate_register_cache(ctxt);
6535 /* Save the faulting GPA (cr2) in the address field */
6536 ctxt->exception.address = cr2;
6538 r = x86_emulate_insn(ctxt);
6540 if (r == EMULATION_INTERCEPTED)
6541 return EMULATE_DONE;
6543 if (r == EMULATION_FAILED) {
6544 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6546 return EMULATE_DONE;
6548 return handle_emulation_failure(vcpu, emulation_type);
6551 if (ctxt->have_exception) {
6553 if (inject_emulated_exception(vcpu))
6555 } else if (vcpu->arch.pio.count) {
6556 if (!vcpu->arch.pio.in) {
6557 /* FIXME: return into emulator if single-stepping. */
6558 vcpu->arch.pio.count = 0;
6561 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6563 r = EMULATE_USER_EXIT;
6564 } else if (vcpu->mmio_needed) {
6565 if (!vcpu->mmio_is_write)
6567 r = EMULATE_USER_EXIT;
6568 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6569 } else if (r == EMULATION_RESTART)
6575 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6576 toggle_interruptibility(vcpu, ctxt->interruptibility);
6577 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6578 kvm_rip_write(vcpu, ctxt->eip);
6579 if (r == EMULATE_DONE && ctxt->tf)
6580 kvm_vcpu_do_singlestep(vcpu, &r);
6581 if (!ctxt->have_exception ||
6582 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6583 __kvm_set_rflags(vcpu, ctxt->eflags);
6586 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6587 * do nothing, and it will be requested again as soon as
6588 * the shadow expires. But we still need to check here,
6589 * because POPF has no interrupt shadow.
6591 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6592 kvm_make_request(KVM_REQ_EVENT, vcpu);
6594 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6599 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6601 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6603 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6605 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6606 void *insn, int insn_len)
6608 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6610 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6612 static int complete_fast_pio_out_port_0x7e(struct kvm_vcpu *vcpu)
6614 vcpu->arch.pio.count = 0;
6618 static int complete_fast_pio_out(struct kvm_vcpu *vcpu)
6620 vcpu->arch.pio.count = 0;
6622 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip)))
6625 return kvm_skip_emulated_instruction(vcpu);
6628 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6629 unsigned short port)
6631 unsigned long val = kvm_rax_read(vcpu);
6632 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6633 size, port, &val, 1);
6638 * Workaround userspace that relies on old KVM behavior of %rip being
6639 * incremented prior to exiting to userspace to handle "OUT 0x7e".
6642 kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_OUT_7E_INC_RIP)) {
6643 vcpu->arch.complete_userspace_io =
6644 complete_fast_pio_out_port_0x7e;
6645 kvm_skip_emulated_instruction(vcpu);
6647 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6648 vcpu->arch.complete_userspace_io = complete_fast_pio_out;
6653 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6657 /* We should only ever be called with arch.pio.count equal to 1 */
6658 BUG_ON(vcpu->arch.pio.count != 1);
6660 if (unlikely(!kvm_is_linear_rip(vcpu, vcpu->arch.pio.linear_rip))) {
6661 vcpu->arch.pio.count = 0;
6665 /* For size less than 4 we merge, else we zero extend */
6666 val = (vcpu->arch.pio.size < 4) ? kvm_rax_read(vcpu) : 0;
6669 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6670 * the copy and tracing
6672 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6673 vcpu->arch.pio.port, &val, 1);
6674 kvm_rax_write(vcpu, val);
6676 return kvm_skip_emulated_instruction(vcpu);
6679 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6680 unsigned short port)
6685 /* For size less than 4 we merge, else we zero extend */
6686 val = (size < 4) ? kvm_rax_read(vcpu) : 0;
6688 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6691 kvm_rax_write(vcpu, val);
6695 vcpu->arch.pio.linear_rip = kvm_get_linear_rip(vcpu);
6696 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6701 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6706 ret = kvm_fast_pio_in(vcpu, size, port);
6708 ret = kvm_fast_pio_out(vcpu, size, port);
6709 return ret && kvm_skip_emulated_instruction(vcpu);
6711 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6713 static int kvmclock_cpu_down_prep(unsigned int cpu)
6715 __this_cpu_write(cpu_tsc_khz, 0);
6719 static void tsc_khz_changed(void *data)
6721 struct cpufreq_freqs *freq = data;
6722 unsigned long khz = 0;
6726 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6727 khz = cpufreq_quick_get(raw_smp_processor_id());
6730 __this_cpu_write(cpu_tsc_khz, khz);
6733 #ifdef CONFIG_X86_64
6734 static void kvm_hyperv_tsc_notifier(void)
6737 struct kvm_vcpu *vcpu;
6740 mutex_lock(&kvm_lock);
6741 list_for_each_entry(kvm, &vm_list, vm_list)
6742 kvm_make_mclock_inprogress_request(kvm);
6744 hyperv_stop_tsc_emulation();
6746 /* TSC frequency always matches when on Hyper-V */
6747 for_each_present_cpu(cpu)
6748 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6749 kvm_max_guest_tsc_khz = tsc_khz;
6751 list_for_each_entry(kvm, &vm_list, vm_list) {
6752 struct kvm_arch *ka = &kvm->arch;
6754 spin_lock(&ka->pvclock_gtod_sync_lock);
6756 pvclock_update_vm_gtod_copy(kvm);
6758 kvm_for_each_vcpu(cpu, vcpu, kvm)
6759 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6761 kvm_for_each_vcpu(cpu, vcpu, kvm)
6762 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6764 spin_unlock(&ka->pvclock_gtod_sync_lock);
6766 mutex_unlock(&kvm_lock);
6770 static void __kvmclock_cpufreq_notifier(struct cpufreq_freqs *freq, int cpu)
6773 struct kvm_vcpu *vcpu;
6774 int i, send_ipi = 0;
6777 * We allow guests to temporarily run on slowing clocks,
6778 * provided we notify them after, or to run on accelerating
6779 * clocks, provided we notify them before. Thus time never
6782 * However, we have a problem. We can't atomically update
6783 * the frequency of a given CPU from this function; it is
6784 * merely a notifier, which can be called from any CPU.
6785 * Changing the TSC frequency at arbitrary points in time
6786 * requires a recomputation of local variables related to
6787 * the TSC for each VCPU. We must flag these local variables
6788 * to be updated and be sure the update takes place with the
6789 * new frequency before any guests proceed.
6791 * Unfortunately, the combination of hotplug CPU and frequency
6792 * change creates an intractable locking scenario; the order
6793 * of when these callouts happen is undefined with respect to
6794 * CPU hotplug, and they can race with each other. As such,
6795 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6796 * undefined; you can actually have a CPU frequency change take
6797 * place in between the computation of X and the setting of the
6798 * variable. To protect against this problem, all updates of
6799 * the per_cpu tsc_khz variable are done in an interrupt
6800 * protected IPI, and all callers wishing to update the value
6801 * must wait for a synchronous IPI to complete (which is trivial
6802 * if the caller is on the CPU already). This establishes the
6803 * necessary total order on variable updates.
6805 * Note that because a guest time update may take place
6806 * anytime after the setting of the VCPU's request bit, the
6807 * correct TSC value must be set before the request. However,
6808 * to ensure the update actually makes it to any guest which
6809 * starts running in hardware virtualization between the set
6810 * and the acquisition of the spinlock, we must also ping the
6811 * CPU after setting the request bit.
6815 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6817 mutex_lock(&kvm_lock);
6818 list_for_each_entry(kvm, &vm_list, vm_list) {
6819 kvm_for_each_vcpu(i, vcpu, kvm) {
6820 if (vcpu->cpu != cpu)
6822 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6823 if (vcpu->cpu != raw_smp_processor_id())
6827 mutex_unlock(&kvm_lock);
6829 if (freq->old < freq->new && send_ipi) {
6831 * We upscale the frequency. Must make the guest
6832 * doesn't see old kvmclock values while running with
6833 * the new frequency, otherwise we risk the guest sees
6834 * time go backwards.
6836 * In case we update the frequency for another cpu
6837 * (which might be in guest context) send an interrupt
6838 * to kick the cpu out of guest context. Next time
6839 * guest context is entered kvmclock will be updated,
6840 * so the guest will not see stale values.
6842 smp_call_function_single(cpu, tsc_khz_changed, freq, 1);
6846 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6849 struct cpufreq_freqs *freq = data;
6852 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6854 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6857 for_each_cpu(cpu, freq->policy->cpus)
6858 __kvmclock_cpufreq_notifier(freq, cpu);
6863 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6864 .notifier_call = kvmclock_cpufreq_notifier
6867 static int kvmclock_cpu_online(unsigned int cpu)
6869 tsc_khz_changed(NULL);
6873 static void kvm_timer_init(void)
6875 max_tsc_khz = tsc_khz;
6877 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6878 #ifdef CONFIG_CPU_FREQ
6879 struct cpufreq_policy policy;
6882 memset(&policy, 0, sizeof(policy));
6884 cpufreq_get_policy(&policy, cpu);
6885 if (policy.cpuinfo.max_freq)
6886 max_tsc_khz = policy.cpuinfo.max_freq;
6889 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6890 CPUFREQ_TRANSITION_NOTIFIER);
6892 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6894 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6895 kvmclock_cpu_online, kvmclock_cpu_down_prep);
6898 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6899 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6901 int kvm_is_in_guest(void)
6903 return __this_cpu_read(current_vcpu) != NULL;
6906 static int kvm_is_user_mode(void)
6910 if (__this_cpu_read(current_vcpu))
6911 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6913 return user_mode != 0;
6916 static unsigned long kvm_get_guest_ip(void)
6918 unsigned long ip = 0;
6920 if (__this_cpu_read(current_vcpu))
6921 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6926 static void kvm_handle_intel_pt_intr(void)
6928 struct kvm_vcpu *vcpu = __this_cpu_read(current_vcpu);
6930 kvm_make_request(KVM_REQ_PMI, vcpu);
6931 __set_bit(MSR_CORE_PERF_GLOBAL_OVF_CTRL_TRACE_TOPA_PMI_BIT,
6932 (unsigned long *)&vcpu->arch.pmu.global_status);
6935 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6936 .is_in_guest = kvm_is_in_guest,
6937 .is_user_mode = kvm_is_user_mode,
6938 .get_guest_ip = kvm_get_guest_ip,
6939 .handle_intel_pt_intr = kvm_handle_intel_pt_intr,
6942 #ifdef CONFIG_X86_64
6943 static void pvclock_gtod_update_fn(struct work_struct *work)
6947 struct kvm_vcpu *vcpu;
6950 mutex_lock(&kvm_lock);
6951 list_for_each_entry(kvm, &vm_list, vm_list)
6952 kvm_for_each_vcpu(i, vcpu, kvm)
6953 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6954 atomic_set(&kvm_guest_has_master_clock, 0);
6955 mutex_unlock(&kvm_lock);
6958 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6961 * Notification about pvclock gtod data update.
6963 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6966 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6967 struct timekeeper *tk = priv;
6969 update_pvclock_gtod(tk);
6971 /* disable master clock if host does not trust, or does not
6972 * use, TSC based clocksource.
6974 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6975 atomic_read(&kvm_guest_has_master_clock) != 0)
6976 queue_work(system_long_wq, &pvclock_gtod_work);
6981 static struct notifier_block pvclock_gtod_notifier = {
6982 .notifier_call = pvclock_gtod_notify,
6986 int kvm_arch_init(void *opaque)
6989 struct kvm_x86_ops *ops = opaque;
6992 printk(KERN_ERR "kvm: already loaded the other module\n");
6997 if (!ops->cpu_has_kvm_support()) {
6998 printk(KERN_ERR "kvm: no hardware support\n");
7002 if (ops->disabled_by_bios()) {
7003 printk(KERN_ERR "kvm: disabled by bios\n");
7009 * KVM explicitly assumes that the guest has an FPU and
7010 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
7011 * vCPU's FPU state as a fxregs_state struct.
7013 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
7014 printk(KERN_ERR "kvm: inadequate fpu\n");
7020 x86_fpu_cache = kmem_cache_create("x86_fpu", sizeof(struct fpu),
7021 __alignof__(struct fpu), SLAB_ACCOUNT,
7023 if (!x86_fpu_cache) {
7024 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
7028 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
7030 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
7031 goto out_free_x86_fpu_cache;
7034 r = kvm_mmu_module_init();
7036 goto out_free_percpu;
7040 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
7041 PT_DIRTY_MASK, PT64_NX_MASK, 0,
7042 PT_PRESENT_MASK, 0, sme_me_mask);
7045 perf_register_guest_info_callbacks(&kvm_guest_cbs);
7047 if (boot_cpu_has(X86_FEATURE_XSAVE))
7048 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
7051 #ifdef CONFIG_X86_64
7052 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
7054 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7055 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
7061 free_percpu(shared_msrs);
7062 out_free_x86_fpu_cache:
7063 kmem_cache_destroy(x86_fpu_cache);
7068 void kvm_arch_exit(void)
7070 #ifdef CONFIG_X86_64
7071 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
7072 clear_hv_tscchange_cb();
7075 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
7077 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
7078 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
7079 CPUFREQ_TRANSITION_NOTIFIER);
7080 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
7081 #ifdef CONFIG_X86_64
7082 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
7085 kvm_mmu_module_exit();
7086 free_percpu(shared_msrs);
7087 kmem_cache_destroy(x86_fpu_cache);
7090 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
7092 ++vcpu->stat.halt_exits;
7093 if (lapic_in_kernel(vcpu)) {
7094 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
7097 vcpu->run->exit_reason = KVM_EXIT_HLT;
7101 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
7103 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
7105 int ret = kvm_skip_emulated_instruction(vcpu);
7107 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
7108 * KVM_EXIT_DEBUG here.
7110 return kvm_vcpu_halt(vcpu) && ret;
7112 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
7114 #ifdef CONFIG_X86_64
7115 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
7116 unsigned long clock_type)
7118 struct kvm_clock_pairing clock_pairing;
7119 struct timespec64 ts;
7123 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
7124 return -KVM_EOPNOTSUPP;
7126 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
7127 return -KVM_EOPNOTSUPP;
7129 clock_pairing.sec = ts.tv_sec;
7130 clock_pairing.nsec = ts.tv_nsec;
7131 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
7132 clock_pairing.flags = 0;
7133 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
7136 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
7137 sizeof(struct kvm_clock_pairing)))
7145 * kvm_pv_kick_cpu_op: Kick a vcpu.
7147 * @apicid - apicid of vcpu to be kicked.
7149 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7151 struct kvm_lapic_irq lapic_irq;
7153 lapic_irq.shorthand = 0;
7154 lapic_irq.dest_mode = 0;
7155 lapic_irq.level = 0;
7156 lapic_irq.dest_id = apicid;
7157 lapic_irq.msi_redir_hint = false;
7159 lapic_irq.delivery_mode = APIC_DM_REMRD;
7160 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7163 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7165 if (!lapic_in_kernel(vcpu)) {
7166 WARN_ON_ONCE(vcpu->arch.apicv_active);
7169 if (!vcpu->arch.apicv_active)
7172 vcpu->arch.apicv_active = false;
7173 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7176 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7178 unsigned long nr, a0, a1, a2, a3, ret;
7181 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7182 return kvm_hv_hypercall(vcpu);
7184 nr = kvm_rax_read(vcpu);
7185 a0 = kvm_rbx_read(vcpu);
7186 a1 = kvm_rcx_read(vcpu);
7187 a2 = kvm_rdx_read(vcpu);
7188 a3 = kvm_rsi_read(vcpu);
7190 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7192 op_64_bit = is_64_bit_mode(vcpu);
7201 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7207 case KVM_HC_VAPIC_POLL_IRQ:
7210 case KVM_HC_KICK_CPU:
7211 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7214 #ifdef CONFIG_X86_64
7215 case KVM_HC_CLOCK_PAIRING:
7216 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7219 case KVM_HC_SEND_IPI:
7220 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7229 kvm_rax_write(vcpu, ret);
7231 ++vcpu->stat.hypercalls;
7232 return kvm_skip_emulated_instruction(vcpu);
7234 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7236 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7238 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7239 char instruction[3];
7240 unsigned long rip = kvm_rip_read(vcpu);
7242 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7244 return emulator_write_emulated(ctxt, rip, instruction, 3,
7248 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7250 return vcpu->run->request_interrupt_window &&
7251 likely(!pic_in_kernel(vcpu->kvm));
7254 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7256 struct kvm_run *kvm_run = vcpu->run;
7258 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7259 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7260 kvm_run->cr8 = kvm_get_cr8(vcpu);
7261 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7262 kvm_run->ready_for_interrupt_injection =
7263 pic_in_kernel(vcpu->kvm) ||
7264 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7267 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7271 if (!kvm_x86_ops->update_cr8_intercept)
7274 if (!lapic_in_kernel(vcpu))
7277 if (vcpu->arch.apicv_active)
7280 if (!vcpu->arch.apic->vapic_addr)
7281 max_irr = kvm_lapic_find_highest_irr(vcpu);
7288 tpr = kvm_lapic_get_cr8(vcpu);
7290 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7293 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7297 /* try to reinject previous events if any */
7299 if (vcpu->arch.exception.injected)
7300 kvm_x86_ops->queue_exception(vcpu);
7302 * Do not inject an NMI or interrupt if there is a pending
7303 * exception. Exceptions and interrupts are recognized at
7304 * instruction boundaries, i.e. the start of an instruction.
7305 * Trap-like exceptions, e.g. #DB, have higher priority than
7306 * NMIs and interrupts, i.e. traps are recognized before an
7307 * NMI/interrupt that's pending on the same instruction.
7308 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7309 * priority, but are only generated (pended) during instruction
7310 * execution, i.e. a pending fault-like exception means the
7311 * fault occurred on the *previous* instruction and must be
7312 * serviced prior to recognizing any new events in order to
7313 * fully complete the previous instruction.
7315 else if (!vcpu->arch.exception.pending) {
7316 if (vcpu->arch.nmi_injected)
7317 kvm_x86_ops->set_nmi(vcpu);
7318 else if (vcpu->arch.interrupt.injected)
7319 kvm_x86_ops->set_irq(vcpu);
7323 * Call check_nested_events() even if we reinjected a previous event
7324 * in order for caller to determine if it should require immediate-exit
7325 * from L2 to L1 due to pending L1 events which require exit
7328 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7329 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7334 /* try to inject new event if pending */
7335 if (vcpu->arch.exception.pending) {
7336 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7337 vcpu->arch.exception.has_error_code,
7338 vcpu->arch.exception.error_code);
7340 WARN_ON_ONCE(vcpu->arch.exception.injected);
7341 vcpu->arch.exception.pending = false;
7342 vcpu->arch.exception.injected = true;
7344 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7345 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7348 if (vcpu->arch.exception.nr == DB_VECTOR) {
7350 * This code assumes that nSVM doesn't use
7351 * check_nested_events(). If it does, the
7352 * DR6/DR7 changes should happen before L1
7353 * gets a #VMEXIT for an intercepted #DB in
7354 * L2. (Under VMX, on the other hand, the
7355 * DR6/DR7 changes should not happen in the
7356 * event of a VM-exit to L1 for an intercepted
7359 kvm_deliver_exception_payload(vcpu);
7360 if (vcpu->arch.dr7 & DR7_GD) {
7361 vcpu->arch.dr7 &= ~DR7_GD;
7362 kvm_update_dr7(vcpu);
7366 kvm_x86_ops->queue_exception(vcpu);
7369 /* Don't consider new event if we re-injected an event */
7370 if (kvm_event_needs_reinjection(vcpu))
7373 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7374 kvm_x86_ops->smi_allowed(vcpu)) {
7375 vcpu->arch.smi_pending = false;
7376 ++vcpu->arch.smi_count;
7378 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7379 --vcpu->arch.nmi_pending;
7380 vcpu->arch.nmi_injected = true;
7381 kvm_x86_ops->set_nmi(vcpu);
7382 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7384 * Because interrupts can be injected asynchronously, we are
7385 * calling check_nested_events again here to avoid a race condition.
7386 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7387 * proposal and current concerns. Perhaps we should be setting
7388 * KVM_REQ_EVENT only on certain events and not unconditionally?
7390 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7391 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7395 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7396 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7398 kvm_x86_ops->set_irq(vcpu);
7405 static void process_nmi(struct kvm_vcpu *vcpu)
7410 * x86 is limited to one NMI running, and one NMI pending after it.
7411 * If an NMI is already in progress, limit further NMIs to just one.
7412 * Otherwise, allow two (and we'll inject the first one immediately).
7414 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7417 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7418 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7419 kvm_make_request(KVM_REQ_EVENT, vcpu);
7422 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7425 flags |= seg->g << 23;
7426 flags |= seg->db << 22;
7427 flags |= seg->l << 21;
7428 flags |= seg->avl << 20;
7429 flags |= seg->present << 15;
7430 flags |= seg->dpl << 13;
7431 flags |= seg->s << 12;
7432 flags |= seg->type << 8;
7436 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7438 struct kvm_segment seg;
7441 kvm_get_segment(vcpu, &seg, n);
7442 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7445 offset = 0x7f84 + n * 12;
7447 offset = 0x7f2c + (n - 3) * 12;
7449 put_smstate(u32, buf, offset + 8, seg.base);
7450 put_smstate(u32, buf, offset + 4, seg.limit);
7451 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7454 #ifdef CONFIG_X86_64
7455 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7457 struct kvm_segment seg;
7461 kvm_get_segment(vcpu, &seg, n);
7462 offset = 0x7e00 + n * 16;
7464 flags = enter_smm_get_segment_flags(&seg) >> 8;
7465 put_smstate(u16, buf, offset, seg.selector);
7466 put_smstate(u16, buf, offset + 2, flags);
7467 put_smstate(u32, buf, offset + 4, seg.limit);
7468 put_smstate(u64, buf, offset + 8, seg.base);
7472 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7475 struct kvm_segment seg;
7479 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7480 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7481 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7482 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7484 for (i = 0; i < 8; i++)
7485 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7487 kvm_get_dr(vcpu, 6, &val);
7488 put_smstate(u32, buf, 0x7fcc, (u32)val);
7489 kvm_get_dr(vcpu, 7, &val);
7490 put_smstate(u32, buf, 0x7fc8, (u32)val);
7492 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7493 put_smstate(u32, buf, 0x7fc4, seg.selector);
7494 put_smstate(u32, buf, 0x7f64, seg.base);
7495 put_smstate(u32, buf, 0x7f60, seg.limit);
7496 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7498 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7499 put_smstate(u32, buf, 0x7fc0, seg.selector);
7500 put_smstate(u32, buf, 0x7f80, seg.base);
7501 put_smstate(u32, buf, 0x7f7c, seg.limit);
7502 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7504 kvm_x86_ops->get_gdt(vcpu, &dt);
7505 put_smstate(u32, buf, 0x7f74, dt.address);
7506 put_smstate(u32, buf, 0x7f70, dt.size);
7508 kvm_x86_ops->get_idt(vcpu, &dt);
7509 put_smstate(u32, buf, 0x7f58, dt.address);
7510 put_smstate(u32, buf, 0x7f54, dt.size);
7512 for (i = 0; i < 6; i++)
7513 enter_smm_save_seg_32(vcpu, buf, i);
7515 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7518 put_smstate(u32, buf, 0x7efc, 0x00020000);
7519 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7522 #ifdef CONFIG_X86_64
7523 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7526 struct kvm_segment seg;
7530 for (i = 0; i < 16; i++)
7531 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7533 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7534 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7536 kvm_get_dr(vcpu, 6, &val);
7537 put_smstate(u64, buf, 0x7f68, val);
7538 kvm_get_dr(vcpu, 7, &val);
7539 put_smstate(u64, buf, 0x7f60, val);
7541 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7542 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7543 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7545 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7548 put_smstate(u32, buf, 0x7efc, 0x00020064);
7550 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7552 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7553 put_smstate(u16, buf, 0x7e90, seg.selector);
7554 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7555 put_smstate(u32, buf, 0x7e94, seg.limit);
7556 put_smstate(u64, buf, 0x7e98, seg.base);
7558 kvm_x86_ops->get_idt(vcpu, &dt);
7559 put_smstate(u32, buf, 0x7e84, dt.size);
7560 put_smstate(u64, buf, 0x7e88, dt.address);
7562 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7563 put_smstate(u16, buf, 0x7e70, seg.selector);
7564 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7565 put_smstate(u32, buf, 0x7e74, seg.limit);
7566 put_smstate(u64, buf, 0x7e78, seg.base);
7568 kvm_x86_ops->get_gdt(vcpu, &dt);
7569 put_smstate(u32, buf, 0x7e64, dt.size);
7570 put_smstate(u64, buf, 0x7e68, dt.address);
7572 for (i = 0; i < 6; i++)
7573 enter_smm_save_seg_64(vcpu, buf, i);
7577 static void enter_smm(struct kvm_vcpu *vcpu)
7579 struct kvm_segment cs, ds;
7584 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7585 memset(buf, 0, 512);
7586 #ifdef CONFIG_X86_64
7587 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7588 enter_smm_save_state_64(vcpu, buf);
7591 enter_smm_save_state_32(vcpu, buf);
7594 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7595 * vCPU state (e.g. leave guest mode) after we've saved the state into
7596 * the SMM state-save area.
7598 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7600 vcpu->arch.hflags |= HF_SMM_MASK;
7601 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7603 if (kvm_x86_ops->get_nmi_mask(vcpu))
7604 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7606 kvm_x86_ops->set_nmi_mask(vcpu, true);
7608 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7609 kvm_rip_write(vcpu, 0x8000);
7611 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7612 kvm_x86_ops->set_cr0(vcpu, cr0);
7613 vcpu->arch.cr0 = cr0;
7615 kvm_x86_ops->set_cr4(vcpu, 0);
7617 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7618 dt.address = dt.size = 0;
7619 kvm_x86_ops->set_idt(vcpu, &dt);
7621 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7623 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7624 cs.base = vcpu->arch.smbase;
7629 cs.limit = ds.limit = 0xffffffff;
7630 cs.type = ds.type = 0x3;
7631 cs.dpl = ds.dpl = 0;
7636 cs.avl = ds.avl = 0;
7637 cs.present = ds.present = 1;
7638 cs.unusable = ds.unusable = 0;
7639 cs.padding = ds.padding = 0;
7641 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7642 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7643 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7644 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7645 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7646 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7648 #ifdef CONFIG_X86_64
7649 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7650 kvm_x86_ops->set_efer(vcpu, 0);
7653 kvm_update_cpuid(vcpu);
7654 kvm_mmu_reset_context(vcpu);
7657 static void process_smi(struct kvm_vcpu *vcpu)
7659 vcpu->arch.smi_pending = true;
7660 kvm_make_request(KVM_REQ_EVENT, vcpu);
7663 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7665 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7668 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7670 if (!kvm_apic_present(vcpu))
7673 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7675 if (irqchip_split(vcpu->kvm))
7676 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7678 if (vcpu->arch.apicv_active)
7679 kvm_x86_ops->sync_pir_to_irr(vcpu);
7680 if (ioapic_in_kernel(vcpu->kvm))
7681 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7684 if (is_guest_mode(vcpu))
7685 vcpu->arch.load_eoi_exitmap_pending = true;
7687 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7690 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7692 u64 eoi_exit_bitmap[4];
7694 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7697 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7698 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7699 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7702 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7703 unsigned long start, unsigned long end,
7706 unsigned long apic_address;
7709 * The physical address of apic access page is stored in the VMCS.
7710 * Update it when it becomes invalid.
7712 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7713 if (start <= apic_address && apic_address < end)
7714 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7719 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7721 struct page *page = NULL;
7723 if (!lapic_in_kernel(vcpu))
7726 if (!kvm_x86_ops->set_apic_access_page_addr)
7729 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7730 if (is_error_page(page))
7732 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7735 * Do not pin apic access page in memory, the MMU notifier
7736 * will call us again if it is migrated or swapped out.
7740 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7742 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7744 smp_send_reschedule(vcpu->cpu);
7746 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7749 * Returns 1 to let vcpu_run() continue the guest execution loop without
7750 * exiting to the userspace. Otherwise, the value will be returned to the
7753 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7757 dm_request_for_irq_injection(vcpu) &&
7758 kvm_cpu_accept_dm_intr(vcpu);
7760 bool req_immediate_exit = false;
7762 if (kvm_request_pending(vcpu)) {
7763 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7764 kvm_x86_ops->get_vmcs12_pages(vcpu);
7765 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7766 kvm_mmu_unload(vcpu);
7767 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7768 __kvm_migrate_timers(vcpu);
7769 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7770 kvm_gen_update_masterclock(vcpu->kvm);
7771 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7772 kvm_gen_kvmclock_update(vcpu);
7773 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7774 r = kvm_guest_time_update(vcpu);
7778 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7779 kvm_mmu_sync_roots(vcpu);
7780 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7781 kvm_mmu_load_cr3(vcpu);
7782 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7783 kvm_vcpu_flush_tlb(vcpu, true);
7784 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7785 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7789 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7790 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7791 vcpu->mmio_needed = 0;
7795 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7796 /* Page is swapped out. Do synthetic halt */
7797 vcpu->arch.apf.halted = true;
7801 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7802 record_steal_time(vcpu);
7803 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7805 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7807 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7808 kvm_pmu_handle_event(vcpu);
7809 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7810 kvm_pmu_deliver_pmi(vcpu);
7811 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7812 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7813 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7814 vcpu->arch.ioapic_handled_vectors)) {
7815 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7816 vcpu->run->eoi.vector =
7817 vcpu->arch.pending_ioapic_eoi;
7822 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7823 vcpu_scan_ioapic(vcpu);
7824 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7825 vcpu_load_eoi_exitmap(vcpu);
7826 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7827 kvm_vcpu_reload_apic_access_page(vcpu);
7828 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7829 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7830 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7834 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7835 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7836 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7840 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7841 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7842 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7848 * KVM_REQ_HV_STIMER has to be processed after
7849 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7850 * depend on the guest clock being up-to-date
7852 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7853 kvm_hv_process_stimers(vcpu);
7856 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7857 ++vcpu->stat.req_event;
7858 kvm_apic_accept_events(vcpu);
7859 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7864 if (inject_pending_event(vcpu, req_int_win) != 0)
7865 req_immediate_exit = true;
7867 /* Enable SMI/NMI/IRQ window open exits if needed.
7869 * SMIs have three cases:
7870 * 1) They can be nested, and then there is nothing to
7871 * do here because RSM will cause a vmexit anyway.
7872 * 2) There is an ISA-specific reason why SMI cannot be
7873 * injected, and the moment when this changes can be
7875 * 3) Or the SMI can be pending because
7876 * inject_pending_event has completed the injection
7877 * of an IRQ or NMI from the previous vmexit, and
7878 * then we request an immediate exit to inject the
7881 if (vcpu->arch.smi_pending && !is_smm(vcpu))
7882 if (!kvm_x86_ops->enable_smi_window(vcpu))
7883 req_immediate_exit = true;
7884 if (vcpu->arch.nmi_pending)
7885 kvm_x86_ops->enable_nmi_window(vcpu);
7886 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7887 kvm_x86_ops->enable_irq_window(vcpu);
7888 WARN_ON(vcpu->arch.exception.pending);
7891 if (kvm_lapic_enabled(vcpu)) {
7892 update_cr8_intercept(vcpu);
7893 kvm_lapic_sync_to_vapic(vcpu);
7897 r = kvm_mmu_reload(vcpu);
7899 goto cancel_injection;
7904 kvm_x86_ops->prepare_guest_switch(vcpu);
7907 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7908 * IPI are then delayed after guest entry, which ensures that they
7909 * result in virtual interrupt delivery.
7911 local_irq_disable();
7912 vcpu->mode = IN_GUEST_MODE;
7914 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7917 * 1) We should set ->mode before checking ->requests. Please see
7918 * the comment in kvm_vcpu_exiting_guest_mode().
7920 * 2) For APICv, we should set ->mode before checking PID.ON. This
7921 * pairs with the memory barrier implicit in pi_test_and_set_on
7922 * (see vmx_deliver_posted_interrupt).
7924 * 3) This also orders the write to mode from any reads to the page
7925 * tables done while the VCPU is running. Please see the comment
7926 * in kvm_flush_remote_tlbs.
7928 smp_mb__after_srcu_read_unlock();
7931 * This handles the case where a posted interrupt was
7932 * notified with kvm_vcpu_kick.
7934 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7935 kvm_x86_ops->sync_pir_to_irr(vcpu);
7937 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7938 || need_resched() || signal_pending(current)) {
7939 vcpu->mode = OUTSIDE_GUEST_MODE;
7943 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7945 goto cancel_injection;
7948 if (req_immediate_exit) {
7949 kvm_make_request(KVM_REQ_EVENT, vcpu);
7950 kvm_x86_ops->request_immediate_exit(vcpu);
7953 trace_kvm_entry(vcpu->vcpu_id);
7954 guest_enter_irqoff();
7956 fpregs_assert_state_consistent();
7957 if (test_thread_flag(TIF_NEED_FPU_LOAD))
7958 switch_fpu_return();
7960 if (unlikely(vcpu->arch.switch_db_regs)) {
7962 set_debugreg(vcpu->arch.eff_db[0], 0);
7963 set_debugreg(vcpu->arch.eff_db[1], 1);
7964 set_debugreg(vcpu->arch.eff_db[2], 2);
7965 set_debugreg(vcpu->arch.eff_db[3], 3);
7966 set_debugreg(vcpu->arch.dr6, 6);
7967 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7970 kvm_x86_ops->run(vcpu);
7973 * Do this here before restoring debug registers on the host. And
7974 * since we do this before handling the vmexit, a DR access vmexit
7975 * can (a) read the correct value of the debug registers, (b) set
7976 * KVM_DEBUGREG_WONT_EXIT again.
7978 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7979 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7980 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7981 kvm_update_dr0123(vcpu);
7982 kvm_update_dr6(vcpu);
7983 kvm_update_dr7(vcpu);
7984 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7988 * If the guest has used debug registers, at least dr7
7989 * will be disabled while returning to the host.
7990 * If we don't have active breakpoints in the host, we don't
7991 * care about the messed up debug address registers. But if
7992 * we have some of them active, restore the old state.
7994 if (hw_breakpoint_active())
7995 hw_breakpoint_restore();
7997 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7999 vcpu->mode = OUTSIDE_GUEST_MODE;
8002 kvm_x86_ops->handle_external_intr(vcpu);
8006 guest_exit_irqoff();
8007 if (lapic_in_kernel(vcpu)) {
8008 s64 delta = vcpu->arch.apic->lapic_timer.advance_expire_delta;
8009 if (delta != S64_MIN) {
8010 trace_kvm_wait_lapic_expire(vcpu->vcpu_id, delta);
8011 vcpu->arch.apic->lapic_timer.advance_expire_delta = S64_MIN;
8018 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8021 * Profile KVM exit RIPs:
8023 if (unlikely(prof_on == KVM_PROFILING)) {
8024 unsigned long rip = kvm_rip_read(vcpu);
8025 profile_hit(KVM_PROFILING, (void *)rip);
8028 if (unlikely(vcpu->arch.tsc_always_catchup))
8029 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8031 if (vcpu->arch.apic_attention)
8032 kvm_lapic_sync_from_vapic(vcpu);
8034 vcpu->arch.gpa_available = false;
8035 r = kvm_x86_ops->handle_exit(vcpu);
8039 kvm_x86_ops->cancel_injection(vcpu);
8040 if (unlikely(vcpu->arch.apic_attention))
8041 kvm_lapic_sync_from_vapic(vcpu);
8046 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
8048 if (!kvm_arch_vcpu_runnable(vcpu) &&
8049 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
8050 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8051 kvm_vcpu_block(vcpu);
8052 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8054 if (kvm_x86_ops->post_block)
8055 kvm_x86_ops->post_block(vcpu);
8057 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
8061 kvm_apic_accept_events(vcpu);
8062 switch(vcpu->arch.mp_state) {
8063 case KVM_MP_STATE_HALTED:
8064 vcpu->arch.pv.pv_unhalted = false;
8065 vcpu->arch.mp_state =
8066 KVM_MP_STATE_RUNNABLE;
8068 case KVM_MP_STATE_RUNNABLE:
8069 vcpu->arch.apf.halted = false;
8071 case KVM_MP_STATE_INIT_RECEIVED:
8080 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
8082 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
8083 kvm_x86_ops->check_nested_events(vcpu, false);
8085 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
8086 !vcpu->arch.apf.halted);
8089 static int vcpu_run(struct kvm_vcpu *vcpu)
8092 struct kvm *kvm = vcpu->kvm;
8094 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8095 vcpu->arch.l1tf_flush_l1d = true;
8098 if (kvm_vcpu_running(vcpu)) {
8099 r = vcpu_enter_guest(vcpu);
8101 r = vcpu_block(kvm, vcpu);
8107 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
8108 if (kvm_cpu_has_pending_timer(vcpu))
8109 kvm_inject_pending_timer_irqs(vcpu);
8111 if (dm_request_for_irq_injection(vcpu) &&
8112 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
8114 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
8115 ++vcpu->stat.request_irq_exits;
8119 kvm_check_async_pf_completion(vcpu);
8121 if (signal_pending(current)) {
8123 vcpu->run->exit_reason = KVM_EXIT_INTR;
8124 ++vcpu->stat.signal_exits;
8127 if (need_resched()) {
8128 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8130 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
8134 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
8139 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
8142 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
8143 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
8144 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
8145 if (r != EMULATE_DONE)
8150 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
8152 BUG_ON(!vcpu->arch.pio.count);
8154 return complete_emulated_io(vcpu);
8158 * Implements the following, as a state machine:
8162 * for each mmio piece in the fragment
8170 * for each mmio piece in the fragment
8175 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8177 struct kvm_run *run = vcpu->run;
8178 struct kvm_mmio_fragment *frag;
8181 BUG_ON(!vcpu->mmio_needed);
8183 /* Complete previous fragment */
8184 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8185 len = min(8u, frag->len);
8186 if (!vcpu->mmio_is_write)
8187 memcpy(frag->data, run->mmio.data, len);
8189 if (frag->len <= 8) {
8190 /* Switch to the next fragment. */
8192 vcpu->mmio_cur_fragment++;
8194 /* Go forward to the next mmio piece. */
8200 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8201 vcpu->mmio_needed = 0;
8203 /* FIXME: return into emulator if single-stepping. */
8204 if (vcpu->mmio_is_write)
8206 vcpu->mmio_read_completed = 1;
8207 return complete_emulated_io(vcpu);
8210 run->exit_reason = KVM_EXIT_MMIO;
8211 run->mmio.phys_addr = frag->gpa;
8212 if (vcpu->mmio_is_write)
8213 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8214 run->mmio.len = min(8u, frag->len);
8215 run->mmio.is_write = vcpu->mmio_is_write;
8216 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8220 /* Swap (qemu) user FPU context for the guest FPU context. */
8221 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8225 copy_fpregs_to_fpstate(¤t->thread.fpu);
8226 /* PKRU is separately restored in kvm_x86_ops->run. */
8227 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8228 ~XFEATURE_MASK_PKRU);
8230 fpregs_mark_activate();
8236 /* When vcpu_run ends, restore user space FPU context. */
8237 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8241 copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8242 copy_kernel_to_fpregs(¤t->thread.fpu.state);
8244 fpregs_mark_activate();
8247 ++vcpu->stat.fpu_reload;
8251 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8256 kvm_sigset_activate(vcpu);
8257 kvm_load_guest_fpu(vcpu);
8259 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8260 if (kvm_run->immediate_exit) {
8264 kvm_vcpu_block(vcpu);
8265 kvm_apic_accept_events(vcpu);
8266 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8268 if (signal_pending(current)) {
8270 vcpu->run->exit_reason = KVM_EXIT_INTR;
8271 ++vcpu->stat.signal_exits;
8276 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8281 if (vcpu->run->kvm_dirty_regs) {
8282 r = sync_regs(vcpu);
8287 /* re-sync apic's tpr */
8288 if (!lapic_in_kernel(vcpu)) {
8289 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8295 if (unlikely(vcpu->arch.complete_userspace_io)) {
8296 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8297 vcpu->arch.complete_userspace_io = NULL;
8302 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8304 if (kvm_run->immediate_exit)
8310 kvm_put_guest_fpu(vcpu);
8311 if (vcpu->run->kvm_valid_regs)
8313 post_kvm_run_save(vcpu);
8314 kvm_sigset_deactivate(vcpu);
8320 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8322 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8324 * We are here if userspace calls get_regs() in the middle of
8325 * instruction emulation. Registers state needs to be copied
8326 * back from emulation context to vcpu. Userspace shouldn't do
8327 * that usually, but some bad designed PV devices (vmware
8328 * backdoor interface) need this to work
8330 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8331 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8333 regs->rax = kvm_rax_read(vcpu);
8334 regs->rbx = kvm_rbx_read(vcpu);
8335 regs->rcx = kvm_rcx_read(vcpu);
8336 regs->rdx = kvm_rdx_read(vcpu);
8337 regs->rsi = kvm_rsi_read(vcpu);
8338 regs->rdi = kvm_rdi_read(vcpu);
8339 regs->rsp = kvm_rsp_read(vcpu);
8340 regs->rbp = kvm_rbp_read(vcpu);
8341 #ifdef CONFIG_X86_64
8342 regs->r8 = kvm_r8_read(vcpu);
8343 regs->r9 = kvm_r9_read(vcpu);
8344 regs->r10 = kvm_r10_read(vcpu);
8345 regs->r11 = kvm_r11_read(vcpu);
8346 regs->r12 = kvm_r12_read(vcpu);
8347 regs->r13 = kvm_r13_read(vcpu);
8348 regs->r14 = kvm_r14_read(vcpu);
8349 regs->r15 = kvm_r15_read(vcpu);
8352 regs->rip = kvm_rip_read(vcpu);
8353 regs->rflags = kvm_get_rflags(vcpu);
8356 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8359 __get_regs(vcpu, regs);
8364 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8366 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8367 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8369 kvm_rax_write(vcpu, regs->rax);
8370 kvm_rbx_write(vcpu, regs->rbx);
8371 kvm_rcx_write(vcpu, regs->rcx);
8372 kvm_rdx_write(vcpu, regs->rdx);
8373 kvm_rsi_write(vcpu, regs->rsi);
8374 kvm_rdi_write(vcpu, regs->rdi);
8375 kvm_rsp_write(vcpu, regs->rsp);
8376 kvm_rbp_write(vcpu, regs->rbp);
8377 #ifdef CONFIG_X86_64
8378 kvm_r8_write(vcpu, regs->r8);
8379 kvm_r9_write(vcpu, regs->r9);
8380 kvm_r10_write(vcpu, regs->r10);
8381 kvm_r11_write(vcpu, regs->r11);
8382 kvm_r12_write(vcpu, regs->r12);
8383 kvm_r13_write(vcpu, regs->r13);
8384 kvm_r14_write(vcpu, regs->r14);
8385 kvm_r15_write(vcpu, regs->r15);
8388 kvm_rip_write(vcpu, regs->rip);
8389 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8391 vcpu->arch.exception.pending = false;
8393 kvm_make_request(KVM_REQ_EVENT, vcpu);
8396 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8399 __set_regs(vcpu, regs);
8404 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8406 struct kvm_segment cs;
8408 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8412 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8414 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8418 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8419 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8420 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8421 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8422 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8423 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8425 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8426 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8428 kvm_x86_ops->get_idt(vcpu, &dt);
8429 sregs->idt.limit = dt.size;
8430 sregs->idt.base = dt.address;
8431 kvm_x86_ops->get_gdt(vcpu, &dt);
8432 sregs->gdt.limit = dt.size;
8433 sregs->gdt.base = dt.address;
8435 sregs->cr0 = kvm_read_cr0(vcpu);
8436 sregs->cr2 = vcpu->arch.cr2;
8437 sregs->cr3 = kvm_read_cr3(vcpu);
8438 sregs->cr4 = kvm_read_cr4(vcpu);
8439 sregs->cr8 = kvm_get_cr8(vcpu);
8440 sregs->efer = vcpu->arch.efer;
8441 sregs->apic_base = kvm_get_apic_base(vcpu);
8443 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8445 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8446 set_bit(vcpu->arch.interrupt.nr,
8447 (unsigned long *)sregs->interrupt_bitmap);
8450 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8451 struct kvm_sregs *sregs)
8454 __get_sregs(vcpu, sregs);
8459 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8460 struct kvm_mp_state *mp_state)
8464 kvm_apic_accept_events(vcpu);
8465 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8466 vcpu->arch.pv.pv_unhalted)
8467 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8469 mp_state->mp_state = vcpu->arch.mp_state;
8475 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8476 struct kvm_mp_state *mp_state)
8482 if (!lapic_in_kernel(vcpu) &&
8483 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8486 /* INITs are latched while in SMM */
8487 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8488 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8489 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8492 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8493 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8494 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8496 vcpu->arch.mp_state = mp_state->mp_state;
8497 kvm_make_request(KVM_REQ_EVENT, vcpu);
8505 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8506 int reason, bool has_error_code, u32 error_code)
8508 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8511 init_emulate_ctxt(vcpu);
8513 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8514 has_error_code, error_code);
8517 return EMULATE_FAIL;
8519 kvm_rip_write(vcpu, ctxt->eip);
8520 kvm_set_rflags(vcpu, ctxt->eflags);
8521 kvm_make_request(KVM_REQ_EVENT, vcpu);
8522 return EMULATE_DONE;
8524 EXPORT_SYMBOL_GPL(kvm_task_switch);
8526 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8528 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8529 (sregs->cr4 & X86_CR4_OSXSAVE))
8532 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8534 * When EFER.LME and CR0.PG are set, the processor is in
8535 * 64-bit mode (though maybe in a 32-bit code segment).
8536 * CR4.PAE and EFER.LMA must be set.
8538 if (!(sregs->cr4 & X86_CR4_PAE)
8539 || !(sregs->efer & EFER_LMA))
8543 * Not in 64-bit mode: EFER.LMA is clear and the code
8544 * segment cannot be 64-bit.
8546 if (sregs->efer & EFER_LMA || sregs->cs.l)
8553 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8555 struct msr_data apic_base_msr;
8556 int mmu_reset_needed = 0;
8557 int cpuid_update_needed = 0;
8558 int pending_vec, max_bits, idx;
8562 if (kvm_valid_sregs(vcpu, sregs))
8565 apic_base_msr.data = sregs->apic_base;
8566 apic_base_msr.host_initiated = true;
8567 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8570 dt.size = sregs->idt.limit;
8571 dt.address = sregs->idt.base;
8572 kvm_x86_ops->set_idt(vcpu, &dt);
8573 dt.size = sregs->gdt.limit;
8574 dt.address = sregs->gdt.base;
8575 kvm_x86_ops->set_gdt(vcpu, &dt);
8577 vcpu->arch.cr2 = sregs->cr2;
8578 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8579 vcpu->arch.cr3 = sregs->cr3;
8580 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8582 kvm_set_cr8(vcpu, sregs->cr8);
8584 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8585 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8587 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8588 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8589 vcpu->arch.cr0 = sregs->cr0;
8591 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8592 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8593 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8594 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8595 if (cpuid_update_needed)
8596 kvm_update_cpuid(vcpu);
8598 idx = srcu_read_lock(&vcpu->kvm->srcu);
8599 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu)) {
8600 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8601 mmu_reset_needed = 1;
8603 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8605 if (mmu_reset_needed)
8606 kvm_mmu_reset_context(vcpu);
8608 max_bits = KVM_NR_INTERRUPTS;
8609 pending_vec = find_first_bit(
8610 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8611 if (pending_vec < max_bits) {
8612 kvm_queue_interrupt(vcpu, pending_vec, false);
8613 pr_debug("Set back pending irq %d\n", pending_vec);
8616 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8617 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8618 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8619 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8620 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8621 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8623 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8624 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8626 update_cr8_intercept(vcpu);
8628 /* Older userspace won't unhalt the vcpu on reset. */
8629 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8630 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8632 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8634 kvm_make_request(KVM_REQ_EVENT, vcpu);
8641 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8642 struct kvm_sregs *sregs)
8647 ret = __set_sregs(vcpu, sregs);
8652 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8653 struct kvm_guest_debug *dbg)
8655 unsigned long rflags;
8660 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8662 if (vcpu->arch.exception.pending)
8664 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8665 kvm_queue_exception(vcpu, DB_VECTOR);
8667 kvm_queue_exception(vcpu, BP_VECTOR);
8671 * Read rflags as long as potentially injected trace flags are still
8674 rflags = kvm_get_rflags(vcpu);
8676 vcpu->guest_debug = dbg->control;
8677 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8678 vcpu->guest_debug = 0;
8680 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8681 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8682 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8683 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8685 for (i = 0; i < KVM_NR_DB_REGS; i++)
8686 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8688 kvm_update_dr7(vcpu);
8690 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8691 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8692 get_segment_base(vcpu, VCPU_SREG_CS);
8695 * Trigger an rflags update that will inject or remove the trace
8698 kvm_set_rflags(vcpu, rflags);
8700 kvm_x86_ops->update_bp_intercept(vcpu);
8710 * Translate a guest virtual address to a guest physical address.
8712 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8713 struct kvm_translation *tr)
8715 unsigned long vaddr = tr->linear_address;
8721 idx = srcu_read_lock(&vcpu->kvm->srcu);
8722 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8723 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8724 tr->physical_address = gpa;
8725 tr->valid = gpa != UNMAPPED_GVA;
8733 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8735 struct fxregs_state *fxsave;
8739 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8740 memcpy(fpu->fpr, fxsave->st_space, 128);
8741 fpu->fcw = fxsave->cwd;
8742 fpu->fsw = fxsave->swd;
8743 fpu->ftwx = fxsave->twd;
8744 fpu->last_opcode = fxsave->fop;
8745 fpu->last_ip = fxsave->rip;
8746 fpu->last_dp = fxsave->rdp;
8747 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8753 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8755 struct fxregs_state *fxsave;
8759 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8761 memcpy(fxsave->st_space, fpu->fpr, 128);
8762 fxsave->cwd = fpu->fcw;
8763 fxsave->swd = fpu->fsw;
8764 fxsave->twd = fpu->ftwx;
8765 fxsave->fop = fpu->last_opcode;
8766 fxsave->rip = fpu->last_ip;
8767 fxsave->rdp = fpu->last_dp;
8768 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8774 static void store_regs(struct kvm_vcpu *vcpu)
8776 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8778 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8779 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8781 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8782 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8784 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8785 kvm_vcpu_ioctl_x86_get_vcpu_events(
8786 vcpu, &vcpu->run->s.regs.events);
8789 static int sync_regs(struct kvm_vcpu *vcpu)
8791 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8794 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8795 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8796 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8798 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8799 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8801 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8803 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8804 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8805 vcpu, &vcpu->run->s.regs.events))
8807 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8813 static void fx_init(struct kvm_vcpu *vcpu)
8815 fpstate_init(&vcpu->arch.guest_fpu->state);
8816 if (boot_cpu_has(X86_FEATURE_XSAVES))
8817 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
8818 host_xcr0 | XSTATE_COMPACTION_ENABLED;
8821 * Ensure guest xcr0 is valid for loading
8823 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8825 vcpu->arch.cr0 |= X86_CR0_ET;
8828 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8830 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8832 kvmclock_reset(vcpu);
8834 kvm_x86_ops->vcpu_free(vcpu);
8835 free_cpumask_var(wbinvd_dirty_mask);
8838 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8841 struct kvm_vcpu *vcpu;
8843 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8844 printk_once(KERN_WARNING
8845 "kvm: SMP vm created on host with unstable TSC; "
8846 "guest TSC will not be reliable\n");
8848 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8853 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8855 vcpu->arch.arch_capabilities = kvm_get_arch_capabilities();
8856 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8857 kvm_vcpu_mtrr_init(vcpu);
8859 kvm_vcpu_reset(vcpu, false);
8860 kvm_init_mmu(vcpu, false);
8865 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8867 struct msr_data msr;
8868 struct kvm *kvm = vcpu->kvm;
8870 kvm_hv_vcpu_postcreate(vcpu);
8872 if (mutex_lock_killable(&vcpu->mutex))
8876 msr.index = MSR_IA32_TSC;
8877 msr.host_initiated = true;
8878 kvm_write_tsc(vcpu, &msr);
8881 /* poll control enabled by default */
8882 vcpu->arch.msr_kvm_poll_control = 1;
8884 mutex_unlock(&vcpu->mutex);
8886 if (!kvmclock_periodic_sync)
8889 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8890 KVMCLOCK_SYNC_PERIOD);
8893 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8895 vcpu->arch.apf.msr_val = 0;
8898 kvm_mmu_unload(vcpu);
8901 kvm_x86_ops->vcpu_free(vcpu);
8904 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8906 kvm_lapic_reset(vcpu, init_event);
8908 vcpu->arch.hflags = 0;
8910 vcpu->arch.smi_pending = 0;
8911 vcpu->arch.smi_count = 0;
8912 atomic_set(&vcpu->arch.nmi_queued, 0);
8913 vcpu->arch.nmi_pending = 0;
8914 vcpu->arch.nmi_injected = false;
8915 kvm_clear_interrupt_queue(vcpu);
8916 kvm_clear_exception_queue(vcpu);
8917 vcpu->arch.exception.pending = false;
8919 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8920 kvm_update_dr0123(vcpu);
8921 vcpu->arch.dr6 = DR6_INIT;
8922 kvm_update_dr6(vcpu);
8923 vcpu->arch.dr7 = DR7_FIXED_1;
8924 kvm_update_dr7(vcpu);
8928 kvm_make_request(KVM_REQ_EVENT, vcpu);
8929 vcpu->arch.apf.msr_val = 0;
8930 vcpu->arch.st.msr_val = 0;
8932 kvmclock_reset(vcpu);
8934 kvm_clear_async_pf_completion_queue(vcpu);
8935 kvm_async_pf_hash_reset(vcpu);
8936 vcpu->arch.apf.halted = false;
8938 if (kvm_mpx_supported()) {
8939 void *mpx_state_buffer;
8942 * To avoid have the INIT path from kvm_apic_has_events() that be
8943 * called with loaded FPU and does not let userspace fix the state.
8946 kvm_put_guest_fpu(vcpu);
8947 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8949 if (mpx_state_buffer)
8950 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8951 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8953 if (mpx_state_buffer)
8954 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8956 kvm_load_guest_fpu(vcpu);
8960 kvm_pmu_reset(vcpu);
8961 vcpu->arch.smbase = 0x30000;
8963 vcpu->arch.msr_misc_features_enables = 0;
8965 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8968 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8969 vcpu->arch.regs_avail = ~0;
8970 vcpu->arch.regs_dirty = ~0;
8972 vcpu->arch.ia32_xss = 0;
8974 kvm_x86_ops->vcpu_reset(vcpu, init_event);
8977 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8979 struct kvm_segment cs;
8981 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8982 cs.selector = vector << 8;
8983 cs.base = vector << 12;
8984 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8985 kvm_rip_write(vcpu, 0);
8988 int kvm_arch_hardware_enable(void)
8991 struct kvm_vcpu *vcpu;
8996 bool stable, backwards_tsc = false;
8998 kvm_shared_msr_cpu_online();
8999 ret = kvm_x86_ops->hardware_enable();
9003 local_tsc = rdtsc();
9004 stable = !kvm_check_tsc_unstable();
9005 list_for_each_entry(kvm, &vm_list, vm_list) {
9006 kvm_for_each_vcpu(i, vcpu, kvm) {
9007 if (!stable && vcpu->cpu == smp_processor_id())
9008 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
9009 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
9010 backwards_tsc = true;
9011 if (vcpu->arch.last_host_tsc > max_tsc)
9012 max_tsc = vcpu->arch.last_host_tsc;
9018 * Sometimes, even reliable TSCs go backwards. This happens on
9019 * platforms that reset TSC during suspend or hibernate actions, but
9020 * maintain synchronization. We must compensate. Fortunately, we can
9021 * detect that condition here, which happens early in CPU bringup,
9022 * before any KVM threads can be running. Unfortunately, we can't
9023 * bring the TSCs fully up to date with real time, as we aren't yet far
9024 * enough into CPU bringup that we know how much real time has actually
9025 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
9026 * variables that haven't been updated yet.
9028 * So we simply find the maximum observed TSC above, then record the
9029 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
9030 * the adjustment will be applied. Note that we accumulate
9031 * adjustments, in case multiple suspend cycles happen before some VCPU
9032 * gets a chance to run again. In the event that no KVM threads get a
9033 * chance to run, we will miss the entire elapsed period, as we'll have
9034 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
9035 * loose cycle time. This isn't too big a deal, since the loss will be
9036 * uniform across all VCPUs (not to mention the scenario is extremely
9037 * unlikely). It is possible that a second hibernate recovery happens
9038 * much faster than a first, causing the observed TSC here to be
9039 * smaller; this would require additional padding adjustment, which is
9040 * why we set last_host_tsc to the local tsc observed here.
9042 * N.B. - this code below runs only on platforms with reliable TSC,
9043 * as that is the only way backwards_tsc is set above. Also note
9044 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
9045 * have the same delta_cyc adjustment applied if backwards_tsc
9046 * is detected. Note further, this adjustment is only done once,
9047 * as we reset last_host_tsc on all VCPUs to stop this from being
9048 * called multiple times (one for each physical CPU bringup).
9050 * Platforms with unreliable TSCs don't have to deal with this, they
9051 * will be compensated by the logic in vcpu_load, which sets the TSC to
9052 * catchup mode. This will catchup all VCPUs to real time, but cannot
9053 * guarantee that they stay in perfect synchronization.
9055 if (backwards_tsc) {
9056 u64 delta_cyc = max_tsc - local_tsc;
9057 list_for_each_entry(kvm, &vm_list, vm_list) {
9058 kvm->arch.backwards_tsc_observed = true;
9059 kvm_for_each_vcpu(i, vcpu, kvm) {
9060 vcpu->arch.tsc_offset_adjustment += delta_cyc;
9061 vcpu->arch.last_host_tsc = local_tsc;
9062 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
9066 * We have to disable TSC offset matching.. if you were
9067 * booting a VM while issuing an S4 host suspend....
9068 * you may have some problem. Solving this issue is
9069 * left as an exercise to the reader.
9071 kvm->arch.last_tsc_nsec = 0;
9072 kvm->arch.last_tsc_write = 0;
9079 void kvm_arch_hardware_disable(void)
9081 kvm_x86_ops->hardware_disable();
9082 drop_user_return_notifiers();
9085 int kvm_arch_hardware_setup(void)
9089 r = kvm_x86_ops->hardware_setup();
9093 if (kvm_has_tsc_control) {
9095 * Make sure the user can only configure tsc_khz values that
9096 * fit into a signed integer.
9097 * A min value is not calculated because it will always
9098 * be 1 on all machines.
9100 u64 max = min(0x7fffffffULL,
9101 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
9102 kvm_max_guest_tsc_khz = max;
9104 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
9107 kvm_init_msr_list();
9111 void kvm_arch_hardware_unsetup(void)
9113 kvm_x86_ops->hardware_unsetup();
9116 int kvm_arch_check_processor_compat(void)
9118 return kvm_x86_ops->check_processor_compatibility();
9121 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
9123 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
9125 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
9127 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
9129 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
9132 struct static_key kvm_no_apic_vcpu __read_mostly;
9133 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
9135 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
9140 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
9141 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
9142 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9144 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
9146 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
9151 vcpu->arch.pio_data = page_address(page);
9153 kvm_set_tsc_khz(vcpu, max_tsc_khz);
9155 r = kvm_mmu_create(vcpu);
9157 goto fail_free_pio_data;
9159 if (irqchip_in_kernel(vcpu->kvm)) {
9160 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
9161 r = kvm_create_lapic(vcpu, lapic_timer_advance_ns);
9163 goto fail_mmu_destroy;
9165 static_key_slow_inc(&kvm_no_apic_vcpu);
9167 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
9168 GFP_KERNEL_ACCOUNT);
9169 if (!vcpu->arch.mce_banks) {
9171 goto fail_free_lapic;
9173 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9175 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask,
9176 GFP_KERNEL_ACCOUNT)) {
9178 goto fail_free_mce_banks;
9183 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9185 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9187 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9189 kvm_async_pf_hash_reset(vcpu);
9192 vcpu->arch.pending_external_vector = -1;
9193 vcpu->arch.preempted_in_kernel = false;
9195 kvm_hv_vcpu_init(vcpu);
9199 fail_free_mce_banks:
9200 kfree(vcpu->arch.mce_banks);
9202 kvm_free_lapic(vcpu);
9204 kvm_mmu_destroy(vcpu);
9206 free_page((unsigned long)vcpu->arch.pio_data);
9211 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9215 kvm_hv_vcpu_uninit(vcpu);
9216 kvm_pmu_destroy(vcpu);
9217 kfree(vcpu->arch.mce_banks);
9218 kvm_free_lapic(vcpu);
9219 idx = srcu_read_lock(&vcpu->kvm->srcu);
9220 kvm_mmu_destroy(vcpu);
9221 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9222 free_page((unsigned long)vcpu->arch.pio_data);
9223 if (!lapic_in_kernel(vcpu))
9224 static_key_slow_dec(&kvm_no_apic_vcpu);
9227 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9229 vcpu->arch.l1tf_flush_l1d = true;
9230 kvm_x86_ops->sched_in(vcpu, cpu);
9233 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9238 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9239 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9240 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9241 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9243 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9244 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9245 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9246 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9247 &kvm->arch.irq_sources_bitmap);
9249 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9250 mutex_init(&kvm->arch.apic_map_lock);
9251 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9253 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
9254 pvclock_update_vm_gtod_copy(kvm);
9256 kvm->arch.guest_can_read_msr_platform_info = true;
9258 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9259 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9261 kvm_hv_init_vm(kvm);
9262 kvm_page_track_init(kvm);
9263 kvm_mmu_init_vm(kvm);
9265 if (kvm_x86_ops->vm_init)
9266 return kvm_x86_ops->vm_init(kvm);
9271 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9274 kvm_mmu_unload(vcpu);
9278 static void kvm_free_vcpus(struct kvm *kvm)
9281 struct kvm_vcpu *vcpu;
9284 * Unpin any mmu pages first.
9286 kvm_for_each_vcpu(i, vcpu, kvm) {
9287 kvm_clear_async_pf_completion_queue(vcpu);
9288 kvm_unload_vcpu_mmu(vcpu);
9290 kvm_for_each_vcpu(i, vcpu, kvm)
9291 kvm_arch_vcpu_free(vcpu);
9293 mutex_lock(&kvm->lock);
9294 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9295 kvm->vcpus[i] = NULL;
9297 atomic_set(&kvm->online_vcpus, 0);
9298 mutex_unlock(&kvm->lock);
9301 void kvm_arch_sync_events(struct kvm *kvm)
9303 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9304 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9308 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9312 struct kvm_memslots *slots = kvm_memslots(kvm);
9313 struct kvm_memory_slot *slot, old;
9315 /* Called with kvm->slots_lock held. */
9316 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9319 slot = id_to_memslot(slots, id);
9325 * MAP_SHARED to prevent internal slot pages from being moved
9328 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9329 MAP_SHARED | MAP_ANONYMOUS, 0);
9330 if (IS_ERR((void *)hva))
9331 return PTR_ERR((void *)hva);
9340 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9341 struct kvm_userspace_memory_region m;
9343 m.slot = id | (i << 16);
9345 m.guest_phys_addr = gpa;
9346 m.userspace_addr = hva;
9347 m.memory_size = size;
9348 r = __kvm_set_memory_region(kvm, &m);
9354 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9358 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9360 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9364 mutex_lock(&kvm->slots_lock);
9365 r = __x86_set_memory_region(kvm, id, gpa, size);
9366 mutex_unlock(&kvm->slots_lock);
9370 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9372 void kvm_arch_destroy_vm(struct kvm *kvm)
9374 if (current->mm == kvm->mm) {
9376 * Free memory regions allocated on behalf of userspace,
9377 * unless the the memory map has changed due to process exit
9380 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9381 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9382 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9384 if (kvm_x86_ops->vm_destroy)
9385 kvm_x86_ops->vm_destroy(kvm);
9386 kvm_pic_destroy(kvm);
9387 kvm_ioapic_destroy(kvm);
9388 kvm_free_vcpus(kvm);
9389 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9390 kvm_mmu_uninit_vm(kvm);
9391 kvm_page_track_cleanup(kvm);
9392 kvm_hv_destroy_vm(kvm);
9395 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9396 struct kvm_memory_slot *dont)
9400 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9401 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9402 kvfree(free->arch.rmap[i]);
9403 free->arch.rmap[i] = NULL;
9408 if (!dont || free->arch.lpage_info[i - 1] !=
9409 dont->arch.lpage_info[i - 1]) {
9410 kvfree(free->arch.lpage_info[i - 1]);
9411 free->arch.lpage_info[i - 1] = NULL;
9415 kvm_page_track_free_memslot(free, dont);
9418 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9419 unsigned long npages)
9423 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9424 struct kvm_lpage_info *linfo;
9429 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9430 slot->base_gfn, level) + 1;
9432 slot->arch.rmap[i] =
9433 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9434 GFP_KERNEL_ACCOUNT);
9435 if (!slot->arch.rmap[i])
9440 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL_ACCOUNT);
9444 slot->arch.lpage_info[i - 1] = linfo;
9446 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9447 linfo[0].disallow_lpage = 1;
9448 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9449 linfo[lpages - 1].disallow_lpage = 1;
9450 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9452 * If the gfn and userspace address are not aligned wrt each
9453 * other, or if explicitly asked to, disable large page
9454 * support for this slot
9456 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9457 !kvm_largepages_enabled()) {
9460 for (j = 0; j < lpages; ++j)
9461 linfo[j].disallow_lpage = 1;
9465 if (kvm_page_track_create_memslot(slot, npages))
9471 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9472 kvfree(slot->arch.rmap[i]);
9473 slot->arch.rmap[i] = NULL;
9477 kvfree(slot->arch.lpage_info[i - 1]);
9478 slot->arch.lpage_info[i - 1] = NULL;
9483 void kvm_arch_memslots_updated(struct kvm *kvm, u64 gen)
9486 * memslots->generation has been incremented.
9487 * mmio generation may have reached its maximum value.
9489 kvm_mmu_invalidate_mmio_sptes(kvm, gen);
9492 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9493 struct kvm_memory_slot *memslot,
9494 const struct kvm_userspace_memory_region *mem,
9495 enum kvm_mr_change change)
9500 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9501 struct kvm_memory_slot *new)
9503 /* Still write protect RO slot */
9504 if (new->flags & KVM_MEM_READONLY) {
9505 kvm_mmu_slot_remove_write_access(kvm, new);
9510 * Call kvm_x86_ops dirty logging hooks when they are valid.
9512 * kvm_x86_ops->slot_disable_log_dirty is called when:
9514 * - KVM_MR_CREATE with dirty logging is disabled
9515 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9517 * The reason is, in case of PML, we need to set D-bit for any slots
9518 * with dirty logging disabled in order to eliminate unnecessary GPA
9519 * logging in PML buffer (and potential PML buffer full VMEXT). This
9520 * guarantees leaving PML enabled during guest's lifetime won't have
9521 * any additional overhead from PML when guest is running with dirty
9522 * logging disabled for memory slots.
9524 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9525 * to dirty logging mode.
9527 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9529 * In case of write protect:
9531 * Write protect all pages for dirty logging.
9533 * All the sptes including the large sptes which point to this
9534 * slot are set to readonly. We can not create any new large
9535 * spte on this slot until the end of the logging.
9537 * See the comments in fast_page_fault().
9539 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9540 if (kvm_x86_ops->slot_enable_log_dirty)
9541 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9543 kvm_mmu_slot_remove_write_access(kvm, new);
9545 if (kvm_x86_ops->slot_disable_log_dirty)
9546 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9550 void kvm_arch_commit_memory_region(struct kvm *kvm,
9551 const struct kvm_userspace_memory_region *mem,
9552 const struct kvm_memory_slot *old,
9553 const struct kvm_memory_slot *new,
9554 enum kvm_mr_change change)
9556 if (!kvm->arch.n_requested_mmu_pages)
9557 kvm_mmu_change_mmu_pages(kvm,
9558 kvm_mmu_calculate_default_mmu_pages(kvm));
9561 * Dirty logging tracks sptes in 4k granularity, meaning that large
9562 * sptes have to be split. If live migration is successful, the guest
9563 * in the source machine will be destroyed and large sptes will be
9564 * created in the destination. However, if the guest continues to run
9565 * in the source machine (for example if live migration fails), small
9566 * sptes will remain around and cause bad performance.
9568 * Scan sptes if dirty logging has been stopped, dropping those
9569 * which can be collapsed into a single large-page spte. Later
9570 * page faults will create the large-page sptes.
9572 if ((change != KVM_MR_DELETE) &&
9573 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9574 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9575 kvm_mmu_zap_collapsible_sptes(kvm, new);
9578 * Set up write protection and/or dirty logging for the new slot.
9580 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9581 * been zapped so no dirty logging staff is needed for old slot. For
9582 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9583 * new and it's also covered when dealing with the new slot.
9585 * FIXME: const-ify all uses of struct kvm_memory_slot.
9587 if (change != KVM_MR_DELETE)
9588 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9591 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9593 kvm_mmu_zap_all(kvm);
9596 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9597 struct kvm_memory_slot *slot)
9599 kvm_page_track_flush_slot(kvm, slot);
9602 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9604 return (is_guest_mode(vcpu) &&
9605 kvm_x86_ops->guest_apic_has_interrupt &&
9606 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9609 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9611 if (!list_empty_careful(&vcpu->async_pf.done))
9614 if (kvm_apic_has_events(vcpu))
9617 if (vcpu->arch.pv.pv_unhalted)
9620 if (vcpu->arch.exception.pending)
9623 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9624 (vcpu->arch.nmi_pending &&
9625 kvm_x86_ops->nmi_allowed(vcpu)))
9628 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9629 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9632 if (kvm_arch_interrupt_allowed(vcpu) &&
9633 (kvm_cpu_has_interrupt(vcpu) ||
9634 kvm_guest_apic_has_interrupt(vcpu)))
9637 if (kvm_hv_has_stimer_pending(vcpu))
9643 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9645 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9648 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9650 return vcpu->arch.preempted_in_kernel;
9653 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9655 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9658 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9660 return kvm_x86_ops->interrupt_allowed(vcpu);
9663 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9665 if (is_64_bit_mode(vcpu))
9666 return kvm_rip_read(vcpu);
9667 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9668 kvm_rip_read(vcpu));
9670 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9672 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9674 return kvm_get_linear_rip(vcpu) == linear_rip;
9676 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9678 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9680 unsigned long rflags;
9682 rflags = kvm_x86_ops->get_rflags(vcpu);
9683 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9684 rflags &= ~X86_EFLAGS_TF;
9687 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9689 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9691 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9692 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9693 rflags |= X86_EFLAGS_TF;
9694 kvm_x86_ops->set_rflags(vcpu, rflags);
9697 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9699 __kvm_set_rflags(vcpu, rflags);
9700 kvm_make_request(KVM_REQ_EVENT, vcpu);
9702 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9704 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9708 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9712 r = kvm_mmu_reload(vcpu);
9716 if (!vcpu->arch.mmu->direct_map &&
9717 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9720 vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9723 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9725 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9728 static inline u32 kvm_async_pf_next_probe(u32 key)
9730 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9733 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9735 u32 key = kvm_async_pf_hash_fn(gfn);
9737 while (vcpu->arch.apf.gfns[key] != ~0)
9738 key = kvm_async_pf_next_probe(key);
9740 vcpu->arch.apf.gfns[key] = gfn;
9743 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9746 u32 key = kvm_async_pf_hash_fn(gfn);
9748 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9749 (vcpu->arch.apf.gfns[key] != gfn &&
9750 vcpu->arch.apf.gfns[key] != ~0); i++)
9751 key = kvm_async_pf_next_probe(key);
9756 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9758 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9761 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9765 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9767 vcpu->arch.apf.gfns[i] = ~0;
9769 j = kvm_async_pf_next_probe(j);
9770 if (vcpu->arch.apf.gfns[j] == ~0)
9772 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9774 * k lies cyclically in ]i,j]
9776 * |....j i.k.| or |.k..j i...|
9778 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9779 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9784 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9787 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9791 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9794 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9798 static bool kvm_can_deliver_async_pf(struct kvm_vcpu *vcpu)
9800 if (!vcpu->arch.apf.delivery_as_pf_vmexit && is_guest_mode(vcpu))
9803 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9804 (vcpu->arch.apf.send_user_only &&
9805 kvm_x86_ops->get_cpl(vcpu) == 0))
9811 bool kvm_can_do_async_pf(struct kvm_vcpu *vcpu)
9813 if (unlikely(!lapic_in_kernel(vcpu) ||
9814 kvm_event_needs_reinjection(vcpu) ||
9815 vcpu->arch.exception.pending))
9818 if (kvm_hlt_in_guest(vcpu->kvm) && !kvm_can_deliver_async_pf(vcpu))
9822 * If interrupts are off we cannot even use an artificial
9825 return kvm_x86_ops->interrupt_allowed(vcpu);
9828 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9829 struct kvm_async_pf *work)
9831 struct x86_exception fault;
9833 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9834 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9836 if (kvm_can_deliver_async_pf(vcpu) &&
9837 !apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9838 fault.vector = PF_VECTOR;
9839 fault.error_code_valid = true;
9840 fault.error_code = 0;
9841 fault.nested_page_fault = false;
9842 fault.address = work->arch.token;
9843 fault.async_page_fault = true;
9844 kvm_inject_page_fault(vcpu, &fault);
9847 * It is not possible to deliver a paravirtualized asynchronous
9848 * page fault, but putting the guest in an artificial halt state
9849 * can be beneficial nevertheless: if an interrupt arrives, we
9850 * can deliver it timely and perhaps the guest will schedule
9851 * another process. When the instruction that triggered a page
9852 * fault is retried, hopefully the page will be ready in the host.
9854 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9858 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9859 struct kvm_async_pf *work)
9861 struct x86_exception fault;
9864 if (work->wakeup_all)
9865 work->arch.token = ~0; /* broadcast wakeup */
9867 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9868 trace_kvm_async_pf_ready(work->arch.token, work->gva);
9870 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9871 !apf_get_user(vcpu, &val)) {
9872 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9873 vcpu->arch.exception.pending &&
9874 vcpu->arch.exception.nr == PF_VECTOR &&
9875 !apf_put_user(vcpu, 0)) {
9876 vcpu->arch.exception.injected = false;
9877 vcpu->arch.exception.pending = false;
9878 vcpu->arch.exception.nr = 0;
9879 vcpu->arch.exception.has_error_code = false;
9880 vcpu->arch.exception.error_code = 0;
9881 vcpu->arch.exception.has_payload = false;
9882 vcpu->arch.exception.payload = 0;
9883 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9884 fault.vector = PF_VECTOR;
9885 fault.error_code_valid = true;
9886 fault.error_code = 0;
9887 fault.nested_page_fault = false;
9888 fault.address = work->arch.token;
9889 fault.async_page_fault = true;
9890 kvm_inject_page_fault(vcpu, &fault);
9893 vcpu->arch.apf.halted = false;
9894 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9897 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9899 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9902 return kvm_can_do_async_pf(vcpu);
9905 void kvm_arch_start_assignment(struct kvm *kvm)
9907 atomic_inc(&kvm->arch.assigned_device_count);
9909 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9911 void kvm_arch_end_assignment(struct kvm *kvm)
9913 atomic_dec(&kvm->arch.assigned_device_count);
9915 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9917 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9919 return atomic_read(&kvm->arch.assigned_device_count);
9921 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9923 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9925 atomic_inc(&kvm->arch.noncoherent_dma_count);
9927 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9929 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9931 atomic_dec(&kvm->arch.noncoherent_dma_count);
9933 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9935 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9937 return atomic_read(&kvm->arch.noncoherent_dma_count);
9939 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9941 bool kvm_arch_has_irq_bypass(void)
9943 return kvm_x86_ops->update_pi_irte != NULL;
9946 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9947 struct irq_bypass_producer *prod)
9949 struct kvm_kernel_irqfd *irqfd =
9950 container_of(cons, struct kvm_kernel_irqfd, consumer);
9952 irqfd->producer = prod;
9954 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9955 prod->irq, irqfd->gsi, 1);
9958 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9959 struct irq_bypass_producer *prod)
9962 struct kvm_kernel_irqfd *irqfd =
9963 container_of(cons, struct kvm_kernel_irqfd, consumer);
9965 WARN_ON(irqfd->producer != prod);
9966 irqfd->producer = NULL;
9969 * When producer of consumer is unregistered, we change back to
9970 * remapped mode, so we can re-use the current implementation
9971 * when the irq is masked/disabled or the consumer side (KVM
9972 * int this case doesn't want to receive the interrupts.
9974 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9976 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9977 " fails: %d\n", irqfd->consumer.token, ret);
9980 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9981 uint32_t guest_irq, bool set)
9983 if (!kvm_x86_ops->update_pi_irte)
9986 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9989 bool kvm_vector_hashing_enabled(void)
9991 return vector_hashing;
9993 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9995 bool kvm_arch_no_poll(struct kvm_vcpu *vcpu)
9997 return (vcpu->arch.msr_kvm_poll_control & 1) == 0;
9999 EXPORT_SYMBOL_GPL(kvm_arch_no_poll);
10002 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
10003 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
10004 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
10005 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
10006 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
10007 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
10008 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
10009 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
10010 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
10011 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
10012 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
10013 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
10014 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
10015 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
10016 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
10017 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
10018 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
10019 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
10020 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);