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>
73 #define CREATE_TRACE_POINTS
76 #define MAX_IO_MSRS 256
77 #define KVM_MAX_MCE_BANKS 32
78 u64 __read_mostly kvm_mce_cap_supported = MCG_CTL_P | MCG_SER_P;
79 EXPORT_SYMBOL_GPL(kvm_mce_cap_supported);
81 #define emul_to_vcpu(ctxt) \
82 container_of(ctxt, struct kvm_vcpu, arch.emulate_ctxt)
85 * - enable syscall per default because its emulated by KVM
86 * - enable LME and LMA per default on 64 bit KVM
90 u64 __read_mostly efer_reserved_bits = ~((u64)(EFER_SCE | EFER_LME | EFER_LMA));
92 static u64 __read_mostly efer_reserved_bits = ~((u64)EFER_SCE);
95 #define VM_STAT(x) offsetof(struct kvm, stat.x), KVM_STAT_VM
96 #define VCPU_STAT(x) offsetof(struct kvm_vcpu, stat.x), KVM_STAT_VCPU
98 #define KVM_X2APIC_API_VALID_FLAGS (KVM_X2APIC_API_USE_32BIT_IDS | \
99 KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
101 static void update_cr8_intercept(struct kvm_vcpu *vcpu);
102 static void process_nmi(struct kvm_vcpu *vcpu);
103 static void enter_smm(struct kvm_vcpu *vcpu);
104 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags);
105 static void store_regs(struct kvm_vcpu *vcpu);
106 static int sync_regs(struct kvm_vcpu *vcpu);
108 struct kvm_x86_ops *kvm_x86_ops __read_mostly;
109 EXPORT_SYMBOL_GPL(kvm_x86_ops);
111 static bool __read_mostly ignore_msrs = 0;
112 module_param(ignore_msrs, bool, S_IRUGO | S_IWUSR);
114 static bool __read_mostly report_ignored_msrs = true;
115 module_param(report_ignored_msrs, bool, S_IRUGO | S_IWUSR);
117 unsigned int min_timer_period_us = 200;
118 module_param(min_timer_period_us, uint, S_IRUGO | S_IWUSR);
120 static bool __read_mostly kvmclock_periodic_sync = true;
121 module_param(kvmclock_periodic_sync, bool, S_IRUGO);
123 bool __read_mostly kvm_has_tsc_control;
124 EXPORT_SYMBOL_GPL(kvm_has_tsc_control);
125 u32 __read_mostly kvm_max_guest_tsc_khz;
126 EXPORT_SYMBOL_GPL(kvm_max_guest_tsc_khz);
127 u8 __read_mostly kvm_tsc_scaling_ratio_frac_bits;
128 EXPORT_SYMBOL_GPL(kvm_tsc_scaling_ratio_frac_bits);
129 u64 __read_mostly kvm_max_tsc_scaling_ratio;
130 EXPORT_SYMBOL_GPL(kvm_max_tsc_scaling_ratio);
131 u64 __read_mostly kvm_default_tsc_scaling_ratio;
132 EXPORT_SYMBOL_GPL(kvm_default_tsc_scaling_ratio);
134 /* tsc tolerance in parts per million - default to 1/2 of the NTP threshold */
135 static u32 __read_mostly tsc_tolerance_ppm = 250;
136 module_param(tsc_tolerance_ppm, uint, S_IRUGO | S_IWUSR);
138 /* lapic timer advance (tscdeadline mode only) in nanoseconds */
139 unsigned int __read_mostly lapic_timer_advance_ns = 1000;
140 module_param(lapic_timer_advance_ns, uint, S_IRUGO | S_IWUSR);
141 EXPORT_SYMBOL_GPL(lapic_timer_advance_ns);
143 static bool __read_mostly vector_hashing = true;
144 module_param(vector_hashing, bool, S_IRUGO);
146 bool __read_mostly enable_vmware_backdoor = false;
147 module_param(enable_vmware_backdoor, bool, S_IRUGO);
148 EXPORT_SYMBOL_GPL(enable_vmware_backdoor);
150 static bool __read_mostly force_emulation_prefix = false;
151 module_param(force_emulation_prefix, bool, S_IRUGO);
153 #define KVM_NR_SHARED_MSRS 16
155 struct kvm_shared_msrs_global {
157 u32 msrs[KVM_NR_SHARED_MSRS];
160 struct kvm_shared_msrs {
161 struct user_return_notifier urn;
163 struct kvm_shared_msr_values {
166 } values[KVM_NR_SHARED_MSRS];
169 static struct kvm_shared_msrs_global __read_mostly shared_msrs_global;
170 static struct kvm_shared_msrs __percpu *shared_msrs;
172 struct kvm_stats_debugfs_item debugfs_entries[] = {
173 { "pf_fixed", VCPU_STAT(pf_fixed) },
174 { "pf_guest", VCPU_STAT(pf_guest) },
175 { "tlb_flush", VCPU_STAT(tlb_flush) },
176 { "invlpg", VCPU_STAT(invlpg) },
177 { "exits", VCPU_STAT(exits) },
178 { "io_exits", VCPU_STAT(io_exits) },
179 { "mmio_exits", VCPU_STAT(mmio_exits) },
180 { "signal_exits", VCPU_STAT(signal_exits) },
181 { "irq_window", VCPU_STAT(irq_window_exits) },
182 { "nmi_window", VCPU_STAT(nmi_window_exits) },
183 { "halt_exits", VCPU_STAT(halt_exits) },
184 { "halt_successful_poll", VCPU_STAT(halt_successful_poll) },
185 { "halt_attempted_poll", VCPU_STAT(halt_attempted_poll) },
186 { "halt_poll_invalid", VCPU_STAT(halt_poll_invalid) },
187 { "halt_wakeup", VCPU_STAT(halt_wakeup) },
188 { "hypercalls", VCPU_STAT(hypercalls) },
189 { "request_irq", VCPU_STAT(request_irq_exits) },
190 { "irq_exits", VCPU_STAT(irq_exits) },
191 { "host_state_reload", VCPU_STAT(host_state_reload) },
192 { "fpu_reload", VCPU_STAT(fpu_reload) },
193 { "insn_emulation", VCPU_STAT(insn_emulation) },
194 { "insn_emulation_fail", VCPU_STAT(insn_emulation_fail) },
195 { "irq_injections", VCPU_STAT(irq_injections) },
196 { "nmi_injections", VCPU_STAT(nmi_injections) },
197 { "req_event", VCPU_STAT(req_event) },
198 { "l1d_flush", VCPU_STAT(l1d_flush) },
199 { "mmu_shadow_zapped", VM_STAT(mmu_shadow_zapped) },
200 { "mmu_pte_write", VM_STAT(mmu_pte_write) },
201 { "mmu_pte_updated", VM_STAT(mmu_pte_updated) },
202 { "mmu_pde_zapped", VM_STAT(mmu_pde_zapped) },
203 { "mmu_flooded", VM_STAT(mmu_flooded) },
204 { "mmu_recycled", VM_STAT(mmu_recycled) },
205 { "mmu_cache_miss", VM_STAT(mmu_cache_miss) },
206 { "mmu_unsync", VM_STAT(mmu_unsync) },
207 { "remote_tlb_flush", VM_STAT(remote_tlb_flush) },
208 { "largepages", VM_STAT(lpages) },
209 { "max_mmu_page_hash_collisions",
210 VM_STAT(max_mmu_page_hash_collisions) },
214 u64 __read_mostly host_xcr0;
216 struct kmem_cache *x86_fpu_cache;
217 EXPORT_SYMBOL_GPL(x86_fpu_cache);
219 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt);
221 static inline void kvm_async_pf_hash_reset(struct kvm_vcpu *vcpu)
224 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU); i++)
225 vcpu->arch.apf.gfns[i] = ~0;
228 static void kvm_on_user_return(struct user_return_notifier *urn)
231 struct kvm_shared_msrs *locals
232 = container_of(urn, struct kvm_shared_msrs, urn);
233 struct kvm_shared_msr_values *values;
237 * Disabling irqs at this point since the following code could be
238 * interrupted and executed through kvm_arch_hardware_disable()
240 local_irq_save(flags);
241 if (locals->registered) {
242 locals->registered = false;
243 user_return_notifier_unregister(urn);
245 local_irq_restore(flags);
246 for (slot = 0; slot < shared_msrs_global.nr; ++slot) {
247 values = &locals->values[slot];
248 if (values->host != values->curr) {
249 wrmsrl(shared_msrs_global.msrs[slot], values->host);
250 values->curr = values->host;
255 static void shared_msr_update(unsigned slot, u32 msr)
258 unsigned int cpu = smp_processor_id();
259 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
261 /* only read, and nobody should modify it at this time,
262 * so don't need lock */
263 if (slot >= shared_msrs_global.nr) {
264 printk(KERN_ERR "kvm: invalid MSR slot!");
267 rdmsrl_safe(msr, &value);
268 smsr->values[slot].host = value;
269 smsr->values[slot].curr = value;
272 void kvm_define_shared_msr(unsigned slot, u32 msr)
274 BUG_ON(slot >= KVM_NR_SHARED_MSRS);
275 shared_msrs_global.msrs[slot] = msr;
276 if (slot >= shared_msrs_global.nr)
277 shared_msrs_global.nr = slot + 1;
279 EXPORT_SYMBOL_GPL(kvm_define_shared_msr);
281 static void kvm_shared_msr_cpu_online(void)
285 for (i = 0; i < shared_msrs_global.nr; ++i)
286 shared_msr_update(i, shared_msrs_global.msrs[i]);
289 int kvm_set_shared_msr(unsigned slot, u64 value, u64 mask)
291 unsigned int cpu = smp_processor_id();
292 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
295 if (((value ^ smsr->values[slot].curr) & mask) == 0)
297 smsr->values[slot].curr = value;
298 err = wrmsrl_safe(shared_msrs_global.msrs[slot], value);
302 if (!smsr->registered) {
303 smsr->urn.on_user_return = kvm_on_user_return;
304 user_return_notifier_register(&smsr->urn);
305 smsr->registered = true;
309 EXPORT_SYMBOL_GPL(kvm_set_shared_msr);
311 static void drop_user_return_notifiers(void)
313 unsigned int cpu = smp_processor_id();
314 struct kvm_shared_msrs *smsr = per_cpu_ptr(shared_msrs, cpu);
316 if (smsr->registered)
317 kvm_on_user_return(&smsr->urn);
320 u64 kvm_get_apic_base(struct kvm_vcpu *vcpu)
322 return vcpu->arch.apic_base;
324 EXPORT_SYMBOL_GPL(kvm_get_apic_base);
326 enum lapic_mode kvm_get_apic_mode(struct kvm_vcpu *vcpu)
328 return kvm_apic_mode(kvm_get_apic_base(vcpu));
330 EXPORT_SYMBOL_GPL(kvm_get_apic_mode);
332 int kvm_set_apic_base(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
334 enum lapic_mode old_mode = kvm_get_apic_mode(vcpu);
335 enum lapic_mode new_mode = kvm_apic_mode(msr_info->data);
336 u64 reserved_bits = ((~0ULL) << cpuid_maxphyaddr(vcpu)) | 0x2ff |
337 (guest_cpuid_has(vcpu, X86_FEATURE_X2APIC) ? 0 : X2APIC_ENABLE);
339 if ((msr_info->data & reserved_bits) != 0 || new_mode == LAPIC_MODE_INVALID)
341 if (!msr_info->host_initiated) {
342 if (old_mode == LAPIC_MODE_X2APIC && new_mode == LAPIC_MODE_XAPIC)
344 if (old_mode == LAPIC_MODE_DISABLED && new_mode == LAPIC_MODE_X2APIC)
348 kvm_lapic_set_base(vcpu, msr_info->data);
351 EXPORT_SYMBOL_GPL(kvm_set_apic_base);
353 asmlinkage __visible void kvm_spurious_fault(void)
355 /* Fault while not rebooting. We want the trace. */
358 EXPORT_SYMBOL_GPL(kvm_spurious_fault);
360 #define EXCPT_BENIGN 0
361 #define EXCPT_CONTRIBUTORY 1
364 static int exception_class(int vector)
374 return EXCPT_CONTRIBUTORY;
381 #define EXCPT_FAULT 0
383 #define EXCPT_ABORT 2
384 #define EXCPT_INTERRUPT 3
386 static int exception_type(int vector)
390 if (WARN_ON(vector > 31 || vector == NMI_VECTOR))
391 return EXCPT_INTERRUPT;
395 /* #DB is trap, as instruction watchpoints are handled elsewhere */
396 if (mask & ((1 << DB_VECTOR) | (1 << BP_VECTOR) | (1 << OF_VECTOR)))
399 if (mask & ((1 << DF_VECTOR) | (1 << MC_VECTOR)))
402 /* Reserved exceptions will result in fault */
406 void kvm_deliver_exception_payload(struct kvm_vcpu *vcpu)
408 unsigned nr = vcpu->arch.exception.nr;
409 bool has_payload = vcpu->arch.exception.has_payload;
410 unsigned long payload = vcpu->arch.exception.payload;
418 * "Certain debug exceptions may clear bit 0-3. The
419 * remaining contents of the DR6 register are never
420 * cleared by the processor".
422 vcpu->arch.dr6 &= ~DR_TRAP_BITS;
424 * DR6.RTM is set by all #DB exceptions that don't clear it.
426 vcpu->arch.dr6 |= DR6_RTM;
427 vcpu->arch.dr6 |= payload;
429 * Bit 16 should be set in the payload whenever the #DB
430 * exception should clear DR6.RTM. This makes the payload
431 * compatible with the pending debug exceptions under VMX.
432 * Though not currently documented in the SDM, this also
433 * makes the payload compatible with the exit qualification
434 * for #DB exceptions under VMX.
436 vcpu->arch.dr6 ^= payload & DR6_RTM;
439 vcpu->arch.cr2 = payload;
443 vcpu->arch.exception.has_payload = false;
444 vcpu->arch.exception.payload = 0;
446 EXPORT_SYMBOL_GPL(kvm_deliver_exception_payload);
448 static void kvm_multiple_exception(struct kvm_vcpu *vcpu,
449 unsigned nr, bool has_error, u32 error_code,
450 bool has_payload, unsigned long payload, bool reinject)
455 kvm_make_request(KVM_REQ_EVENT, vcpu);
457 if (!vcpu->arch.exception.pending && !vcpu->arch.exception.injected) {
459 if (has_error && !is_protmode(vcpu))
463 * On vmentry, vcpu->arch.exception.pending is only
464 * true if an event injection was blocked by
465 * nested_run_pending. In that case, however,
466 * vcpu_enter_guest requests an immediate exit,
467 * and the guest shouldn't proceed far enough to
470 WARN_ON_ONCE(vcpu->arch.exception.pending);
471 vcpu->arch.exception.injected = true;
472 if (WARN_ON_ONCE(has_payload)) {
474 * A reinjected event has already
475 * delivered its payload.
481 vcpu->arch.exception.pending = true;
482 vcpu->arch.exception.injected = false;
484 vcpu->arch.exception.has_error_code = has_error;
485 vcpu->arch.exception.nr = nr;
486 vcpu->arch.exception.error_code = error_code;
487 vcpu->arch.exception.has_payload = has_payload;
488 vcpu->arch.exception.payload = payload;
490 * In guest mode, payload delivery should be deferred,
491 * so that the L1 hypervisor can intercept #PF before
492 * CR2 is modified (or intercept #DB before DR6 is
493 * modified under nVMX). However, for ABI
494 * compatibility with KVM_GET_VCPU_EVENTS and
495 * KVM_SET_VCPU_EVENTS, we can't delay payload
496 * delivery unless userspace has enabled this
497 * functionality via the per-VM capability,
498 * KVM_CAP_EXCEPTION_PAYLOAD.
500 if (!vcpu->kvm->arch.exception_payload_enabled ||
501 !is_guest_mode(vcpu))
502 kvm_deliver_exception_payload(vcpu);
506 /* to check exception */
507 prev_nr = vcpu->arch.exception.nr;
508 if (prev_nr == DF_VECTOR) {
509 /* triple fault -> shutdown */
510 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
513 class1 = exception_class(prev_nr);
514 class2 = exception_class(nr);
515 if ((class1 == EXCPT_CONTRIBUTORY && class2 == EXCPT_CONTRIBUTORY)
516 || (class1 == EXCPT_PF && class2 != EXCPT_BENIGN)) {
518 * Generate double fault per SDM Table 5-5. Set
519 * exception.pending = true so that the double fault
520 * can trigger a nested vmexit.
522 vcpu->arch.exception.pending = true;
523 vcpu->arch.exception.injected = false;
524 vcpu->arch.exception.has_error_code = true;
525 vcpu->arch.exception.nr = DF_VECTOR;
526 vcpu->arch.exception.error_code = 0;
527 vcpu->arch.exception.has_payload = false;
528 vcpu->arch.exception.payload = 0;
530 /* replace previous exception with a new one in a hope
531 that instruction re-execution will regenerate lost
536 void kvm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr)
538 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, false);
540 EXPORT_SYMBOL_GPL(kvm_queue_exception);
542 void kvm_requeue_exception(struct kvm_vcpu *vcpu, unsigned nr)
544 kvm_multiple_exception(vcpu, nr, false, 0, false, 0, true);
546 EXPORT_SYMBOL_GPL(kvm_requeue_exception);
548 static void kvm_queue_exception_p(struct kvm_vcpu *vcpu, unsigned nr,
549 unsigned long payload)
551 kvm_multiple_exception(vcpu, nr, false, 0, true, payload, false);
554 static void kvm_queue_exception_e_p(struct kvm_vcpu *vcpu, unsigned nr,
555 u32 error_code, unsigned long payload)
557 kvm_multiple_exception(vcpu, nr, true, error_code,
558 true, payload, false);
561 int kvm_complete_insn_gp(struct kvm_vcpu *vcpu, int err)
564 kvm_inject_gp(vcpu, 0);
566 return kvm_skip_emulated_instruction(vcpu);
570 EXPORT_SYMBOL_GPL(kvm_complete_insn_gp);
572 void kvm_inject_page_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
574 ++vcpu->stat.pf_guest;
575 vcpu->arch.exception.nested_apf =
576 is_guest_mode(vcpu) && fault->async_page_fault;
577 if (vcpu->arch.exception.nested_apf) {
578 vcpu->arch.apf.nested_apf_token = fault->address;
579 kvm_queue_exception_e(vcpu, PF_VECTOR, fault->error_code);
581 kvm_queue_exception_e_p(vcpu, PF_VECTOR, fault->error_code,
585 EXPORT_SYMBOL_GPL(kvm_inject_page_fault);
587 static bool kvm_propagate_fault(struct kvm_vcpu *vcpu, struct x86_exception *fault)
589 if (mmu_is_nested(vcpu) && !fault->nested_page_fault)
590 vcpu->arch.nested_mmu.inject_page_fault(vcpu, fault);
592 vcpu->arch.mmu->inject_page_fault(vcpu, fault);
594 return fault->nested_page_fault;
597 void kvm_inject_nmi(struct kvm_vcpu *vcpu)
599 atomic_inc(&vcpu->arch.nmi_queued);
600 kvm_make_request(KVM_REQ_NMI, vcpu);
602 EXPORT_SYMBOL_GPL(kvm_inject_nmi);
604 void kvm_queue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
606 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, false);
608 EXPORT_SYMBOL_GPL(kvm_queue_exception_e);
610 void kvm_requeue_exception_e(struct kvm_vcpu *vcpu, unsigned nr, u32 error_code)
612 kvm_multiple_exception(vcpu, nr, true, error_code, false, 0, true);
614 EXPORT_SYMBOL_GPL(kvm_requeue_exception_e);
617 * Checks if cpl <= required_cpl; if true, return true. Otherwise queue
618 * a #GP and return false.
620 bool kvm_require_cpl(struct kvm_vcpu *vcpu, int required_cpl)
622 if (kvm_x86_ops->get_cpl(vcpu) <= required_cpl)
624 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
627 EXPORT_SYMBOL_GPL(kvm_require_cpl);
629 bool kvm_require_dr(struct kvm_vcpu *vcpu, int dr)
631 if ((dr != 4 && dr != 5) || !kvm_read_cr4_bits(vcpu, X86_CR4_DE))
634 kvm_queue_exception(vcpu, UD_VECTOR);
637 EXPORT_SYMBOL_GPL(kvm_require_dr);
640 * This function will be used to read from the physical memory of the currently
641 * running guest. The difference to kvm_vcpu_read_guest_page is that this function
642 * can read from guest physical or from the guest's guest physical memory.
644 int kvm_read_guest_page_mmu(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu,
645 gfn_t ngfn, void *data, int offset, int len,
648 struct x86_exception exception;
652 ngpa = gfn_to_gpa(ngfn);
653 real_gfn = mmu->translate_gpa(vcpu, ngpa, access, &exception);
654 if (real_gfn == UNMAPPED_GVA)
657 real_gfn = gpa_to_gfn(real_gfn);
659 return kvm_vcpu_read_guest_page(vcpu, real_gfn, data, offset, len);
661 EXPORT_SYMBOL_GPL(kvm_read_guest_page_mmu);
663 static int kvm_read_nested_guest_page(struct kvm_vcpu *vcpu, gfn_t gfn,
664 void *data, int offset, int len, u32 access)
666 return kvm_read_guest_page_mmu(vcpu, vcpu->arch.walk_mmu, gfn,
667 data, offset, len, access);
671 * Load the pae pdptrs. Return true is they are all valid.
673 int load_pdptrs(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu, unsigned long cr3)
675 gfn_t pdpt_gfn = cr3 >> PAGE_SHIFT;
676 unsigned offset = ((cr3 & (PAGE_SIZE-1)) >> 5) << 2;
679 u64 pdpte[ARRAY_SIZE(mmu->pdptrs)];
681 ret = kvm_read_guest_page_mmu(vcpu, mmu, pdpt_gfn, pdpte,
682 offset * sizeof(u64), sizeof(pdpte),
683 PFERR_USER_MASK|PFERR_WRITE_MASK);
688 for (i = 0; i < ARRAY_SIZE(pdpte); ++i) {
689 if ((pdpte[i] & PT_PRESENT_MASK) &&
691 vcpu->arch.mmu->guest_rsvd_check.rsvd_bits_mask[0][2])) {
698 memcpy(mmu->pdptrs, pdpte, sizeof(mmu->pdptrs));
699 __set_bit(VCPU_EXREG_PDPTR,
700 (unsigned long *)&vcpu->arch.regs_avail);
701 __set_bit(VCPU_EXREG_PDPTR,
702 (unsigned long *)&vcpu->arch.regs_dirty);
707 EXPORT_SYMBOL_GPL(load_pdptrs);
709 bool pdptrs_changed(struct kvm_vcpu *vcpu)
711 u64 pdpte[ARRAY_SIZE(vcpu->arch.walk_mmu->pdptrs)];
717 if (is_long_mode(vcpu) || !is_pae(vcpu) || !is_paging(vcpu))
720 if (!test_bit(VCPU_EXREG_PDPTR,
721 (unsigned long *)&vcpu->arch.regs_avail))
724 gfn = (kvm_read_cr3(vcpu) & 0xffffffe0ul) >> PAGE_SHIFT;
725 offset = (kvm_read_cr3(vcpu) & 0xffffffe0ul) & (PAGE_SIZE - 1);
726 r = kvm_read_nested_guest_page(vcpu, gfn, pdpte, offset, sizeof(pdpte),
727 PFERR_USER_MASK | PFERR_WRITE_MASK);
730 changed = memcmp(pdpte, vcpu->arch.walk_mmu->pdptrs, sizeof(pdpte)) != 0;
735 EXPORT_SYMBOL_GPL(pdptrs_changed);
737 int kvm_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
739 unsigned long old_cr0 = kvm_read_cr0(vcpu);
740 unsigned long update_bits = X86_CR0_PG | X86_CR0_WP;
745 if (cr0 & 0xffffffff00000000UL)
749 cr0 &= ~CR0_RESERVED_BITS;
751 if ((cr0 & X86_CR0_NW) && !(cr0 & X86_CR0_CD))
754 if ((cr0 & X86_CR0_PG) && !(cr0 & X86_CR0_PE))
757 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG)) {
759 if ((vcpu->arch.efer & EFER_LME)) {
764 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
769 if (is_pae(vcpu) && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
774 if (!(cr0 & X86_CR0_PG) && kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE))
777 kvm_x86_ops->set_cr0(vcpu, cr0);
779 if ((cr0 ^ old_cr0) & X86_CR0_PG) {
780 kvm_clear_async_pf_completion_queue(vcpu);
781 kvm_async_pf_hash_reset(vcpu);
784 if ((cr0 ^ old_cr0) & update_bits)
785 kvm_mmu_reset_context(vcpu);
787 if (((cr0 ^ old_cr0) & X86_CR0_CD) &&
788 kvm_arch_has_noncoherent_dma(vcpu->kvm) &&
789 !kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
790 kvm_zap_gfn_range(vcpu->kvm, 0, ~0ULL);
794 EXPORT_SYMBOL_GPL(kvm_set_cr0);
796 void kvm_lmsw(struct kvm_vcpu *vcpu, unsigned long msw)
798 (void)kvm_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~0x0eul) | (msw & 0x0f));
800 EXPORT_SYMBOL_GPL(kvm_lmsw);
802 static void kvm_load_guest_xcr0(struct kvm_vcpu *vcpu)
804 if (kvm_read_cr4_bits(vcpu, X86_CR4_OSXSAVE) &&
805 !vcpu->guest_xcr0_loaded) {
806 /* kvm_set_xcr() also depends on this */
807 if (vcpu->arch.xcr0 != host_xcr0)
808 xsetbv(XCR_XFEATURE_ENABLED_MASK, vcpu->arch.xcr0);
809 vcpu->guest_xcr0_loaded = 1;
813 static void kvm_put_guest_xcr0(struct kvm_vcpu *vcpu)
815 if (vcpu->guest_xcr0_loaded) {
816 if (vcpu->arch.xcr0 != host_xcr0)
817 xsetbv(XCR_XFEATURE_ENABLED_MASK, host_xcr0);
818 vcpu->guest_xcr0_loaded = 0;
822 static int __kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
825 u64 old_xcr0 = vcpu->arch.xcr0;
828 /* Only support XCR_XFEATURE_ENABLED_MASK(xcr0) now */
829 if (index != XCR_XFEATURE_ENABLED_MASK)
831 if (!(xcr0 & XFEATURE_MASK_FP))
833 if ((xcr0 & XFEATURE_MASK_YMM) && !(xcr0 & XFEATURE_MASK_SSE))
837 * Do not allow the guest to set bits that we do not support
838 * saving. However, xcr0 bit 0 is always set, even if the
839 * emulated CPU does not support XSAVE (see fx_init).
841 valid_bits = vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FP;
842 if (xcr0 & ~valid_bits)
845 if ((!(xcr0 & XFEATURE_MASK_BNDREGS)) !=
846 (!(xcr0 & XFEATURE_MASK_BNDCSR)))
849 if (xcr0 & XFEATURE_MASK_AVX512) {
850 if (!(xcr0 & XFEATURE_MASK_YMM))
852 if ((xcr0 & XFEATURE_MASK_AVX512) != XFEATURE_MASK_AVX512)
855 vcpu->arch.xcr0 = xcr0;
857 if ((xcr0 ^ old_xcr0) & XFEATURE_MASK_EXTEND)
858 kvm_update_cpuid(vcpu);
862 int kvm_set_xcr(struct kvm_vcpu *vcpu, u32 index, u64 xcr)
864 if (kvm_x86_ops->get_cpl(vcpu) != 0 ||
865 __kvm_set_xcr(vcpu, index, xcr)) {
866 kvm_inject_gp(vcpu, 0);
871 EXPORT_SYMBOL_GPL(kvm_set_xcr);
873 int kvm_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
875 unsigned long old_cr4 = kvm_read_cr4(vcpu);
876 unsigned long pdptr_bits = X86_CR4_PGE | X86_CR4_PSE | X86_CR4_PAE |
877 X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE;
879 if (cr4 & CR4_RESERVED_BITS)
882 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) && (cr4 & X86_CR4_OSXSAVE))
885 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMEP) && (cr4 & X86_CR4_SMEP))
888 if (!guest_cpuid_has(vcpu, X86_FEATURE_SMAP) && (cr4 & X86_CR4_SMAP))
891 if (!guest_cpuid_has(vcpu, X86_FEATURE_FSGSBASE) && (cr4 & X86_CR4_FSGSBASE))
894 if (!guest_cpuid_has(vcpu, X86_FEATURE_PKU) && (cr4 & X86_CR4_PKE))
897 if (!guest_cpuid_has(vcpu, X86_FEATURE_LA57) && (cr4 & X86_CR4_LA57))
900 if (!guest_cpuid_has(vcpu, X86_FEATURE_UMIP) && (cr4 & X86_CR4_UMIP))
903 if (is_long_mode(vcpu)) {
904 if (!(cr4 & X86_CR4_PAE))
906 } else if (is_paging(vcpu) && (cr4 & X86_CR4_PAE)
907 && ((cr4 ^ old_cr4) & pdptr_bits)
908 && !load_pdptrs(vcpu, vcpu->arch.walk_mmu,
912 if ((cr4 & X86_CR4_PCIDE) && !(old_cr4 & X86_CR4_PCIDE)) {
913 if (!guest_cpuid_has(vcpu, X86_FEATURE_PCID))
916 /* PCID can not be enabled when cr3[11:0]!=000H or EFER.LMA=0 */
917 if ((kvm_read_cr3(vcpu) & X86_CR3_PCID_MASK) || !is_long_mode(vcpu))
921 if (kvm_x86_ops->set_cr4(vcpu, cr4))
924 if (((cr4 ^ old_cr4) & pdptr_bits) ||
925 (!(cr4 & X86_CR4_PCIDE) && (old_cr4 & X86_CR4_PCIDE)))
926 kvm_mmu_reset_context(vcpu);
928 if ((cr4 ^ old_cr4) & (X86_CR4_OSXSAVE | X86_CR4_PKE))
929 kvm_update_cpuid(vcpu);
933 EXPORT_SYMBOL_GPL(kvm_set_cr4);
935 int kvm_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
937 bool skip_tlb_flush = false;
939 bool pcid_enabled = kvm_read_cr4_bits(vcpu, X86_CR4_PCIDE);
942 skip_tlb_flush = cr3 & X86_CR3_PCID_NOFLUSH;
943 cr3 &= ~X86_CR3_PCID_NOFLUSH;
947 if (cr3 == kvm_read_cr3(vcpu) && !pdptrs_changed(vcpu)) {
948 if (!skip_tlb_flush) {
949 kvm_mmu_sync_roots(vcpu);
950 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
955 if (is_long_mode(vcpu) &&
956 (cr3 & rsvd_bits(cpuid_maxphyaddr(vcpu), 63)))
958 else if (is_pae(vcpu) && is_paging(vcpu) &&
959 !load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3))
962 kvm_mmu_new_cr3(vcpu, cr3, skip_tlb_flush);
963 vcpu->arch.cr3 = cr3;
964 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
968 EXPORT_SYMBOL_GPL(kvm_set_cr3);
970 int kvm_set_cr8(struct kvm_vcpu *vcpu, unsigned long cr8)
972 if (cr8 & CR8_RESERVED_BITS)
974 if (lapic_in_kernel(vcpu))
975 kvm_lapic_set_tpr(vcpu, cr8);
977 vcpu->arch.cr8 = cr8;
980 EXPORT_SYMBOL_GPL(kvm_set_cr8);
982 unsigned long kvm_get_cr8(struct kvm_vcpu *vcpu)
984 if (lapic_in_kernel(vcpu))
985 return kvm_lapic_get_cr8(vcpu);
987 return vcpu->arch.cr8;
989 EXPORT_SYMBOL_GPL(kvm_get_cr8);
991 static void kvm_update_dr0123(struct kvm_vcpu *vcpu)
995 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
996 for (i = 0; i < KVM_NR_DB_REGS; i++)
997 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
998 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_RELOAD;
1002 static void kvm_update_dr6(struct kvm_vcpu *vcpu)
1004 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1005 kvm_x86_ops->set_dr6(vcpu, vcpu->arch.dr6);
1008 static void kvm_update_dr7(struct kvm_vcpu *vcpu)
1012 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1013 dr7 = vcpu->arch.guest_debug_dr7;
1015 dr7 = vcpu->arch.dr7;
1016 kvm_x86_ops->set_dr7(vcpu, dr7);
1017 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_BP_ENABLED;
1018 if (dr7 & DR7_BP_EN_MASK)
1019 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_BP_ENABLED;
1022 static u64 kvm_dr6_fixed(struct kvm_vcpu *vcpu)
1024 u64 fixed = DR6_FIXED_1;
1026 if (!guest_cpuid_has(vcpu, X86_FEATURE_RTM))
1031 static int __kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1035 vcpu->arch.db[dr] = val;
1036 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
1037 vcpu->arch.eff_db[dr] = val;
1042 if (val & 0xffffffff00000000ULL)
1043 return -1; /* #GP */
1044 vcpu->arch.dr6 = (val & DR6_VOLATILE) | kvm_dr6_fixed(vcpu);
1045 kvm_update_dr6(vcpu);
1050 if (val & 0xffffffff00000000ULL)
1051 return -1; /* #GP */
1052 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
1053 kvm_update_dr7(vcpu);
1060 int kvm_set_dr(struct kvm_vcpu *vcpu, int dr, unsigned long val)
1062 if (__kvm_set_dr(vcpu, dr, val)) {
1063 kvm_inject_gp(vcpu, 0);
1068 EXPORT_SYMBOL_GPL(kvm_set_dr);
1070 int kvm_get_dr(struct kvm_vcpu *vcpu, int dr, unsigned long *val)
1074 *val = vcpu->arch.db[dr];
1079 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1080 *val = vcpu->arch.dr6;
1082 *val = kvm_x86_ops->get_dr6(vcpu);
1087 *val = vcpu->arch.dr7;
1092 EXPORT_SYMBOL_GPL(kvm_get_dr);
1094 bool kvm_rdpmc(struct kvm_vcpu *vcpu)
1096 u32 ecx = kvm_register_read(vcpu, VCPU_REGS_RCX);
1100 err = kvm_pmu_rdpmc(vcpu, ecx, &data);
1103 kvm_register_write(vcpu, VCPU_REGS_RAX, (u32)data);
1104 kvm_register_write(vcpu, VCPU_REGS_RDX, data >> 32);
1107 EXPORT_SYMBOL_GPL(kvm_rdpmc);
1110 * List of msr numbers which we expose to userspace through KVM_GET_MSRS
1111 * and KVM_SET_MSRS, and KVM_GET_MSR_INDEX_LIST.
1113 * This list is modified at module load time to reflect the
1114 * capabilities of the host cpu. This capabilities test skips MSRs that are
1115 * kvm-specific. Those are put in emulated_msrs; filtering of emulated_msrs
1116 * may depend on host virtualization features rather than host cpu features.
1119 static u32 msrs_to_save[] = {
1120 MSR_IA32_SYSENTER_CS, MSR_IA32_SYSENTER_ESP, MSR_IA32_SYSENTER_EIP,
1122 #ifdef CONFIG_X86_64
1123 MSR_CSTAR, MSR_KERNEL_GS_BASE, MSR_SYSCALL_MASK, MSR_LSTAR,
1125 MSR_IA32_TSC, MSR_IA32_CR_PAT, MSR_VM_HSAVE_PA,
1126 MSR_IA32_FEATURE_CONTROL, MSR_IA32_BNDCFGS, MSR_TSC_AUX,
1127 MSR_IA32_SPEC_CTRL, MSR_IA32_ARCH_CAPABILITIES
1130 static unsigned num_msrs_to_save;
1132 static u32 emulated_msrs[] = {
1133 MSR_KVM_SYSTEM_TIME, MSR_KVM_WALL_CLOCK,
1134 MSR_KVM_SYSTEM_TIME_NEW, MSR_KVM_WALL_CLOCK_NEW,
1135 HV_X64_MSR_GUEST_OS_ID, HV_X64_MSR_HYPERCALL,
1136 HV_X64_MSR_TIME_REF_COUNT, HV_X64_MSR_REFERENCE_TSC,
1137 HV_X64_MSR_TSC_FREQUENCY, HV_X64_MSR_APIC_FREQUENCY,
1138 HV_X64_MSR_CRASH_P0, HV_X64_MSR_CRASH_P1, HV_X64_MSR_CRASH_P2,
1139 HV_X64_MSR_CRASH_P3, HV_X64_MSR_CRASH_P4, HV_X64_MSR_CRASH_CTL,
1141 HV_X64_MSR_VP_INDEX,
1142 HV_X64_MSR_VP_RUNTIME,
1143 HV_X64_MSR_SCONTROL,
1144 HV_X64_MSR_STIMER0_CONFIG,
1145 HV_X64_MSR_VP_ASSIST_PAGE,
1146 HV_X64_MSR_REENLIGHTENMENT_CONTROL, HV_X64_MSR_TSC_EMULATION_CONTROL,
1147 HV_X64_MSR_TSC_EMULATION_STATUS,
1149 MSR_KVM_ASYNC_PF_EN, MSR_KVM_STEAL_TIME,
1152 MSR_IA32_TSC_ADJUST,
1153 MSR_IA32_TSCDEADLINE,
1154 MSR_IA32_MISC_ENABLE,
1155 MSR_IA32_MCG_STATUS,
1157 MSR_IA32_MCG_EXT_CTL,
1161 MSR_MISC_FEATURES_ENABLES,
1162 MSR_AMD64_VIRT_SPEC_CTRL,
1165 static unsigned num_emulated_msrs;
1168 * List of msr numbers which are used to expose MSR-based features that
1169 * can be used by a hypervisor to validate requested CPU features.
1171 static u32 msr_based_features[] = {
1173 MSR_IA32_VMX_TRUE_PINBASED_CTLS,
1174 MSR_IA32_VMX_PINBASED_CTLS,
1175 MSR_IA32_VMX_TRUE_PROCBASED_CTLS,
1176 MSR_IA32_VMX_PROCBASED_CTLS,
1177 MSR_IA32_VMX_TRUE_EXIT_CTLS,
1178 MSR_IA32_VMX_EXIT_CTLS,
1179 MSR_IA32_VMX_TRUE_ENTRY_CTLS,
1180 MSR_IA32_VMX_ENTRY_CTLS,
1182 MSR_IA32_VMX_CR0_FIXED0,
1183 MSR_IA32_VMX_CR0_FIXED1,
1184 MSR_IA32_VMX_CR4_FIXED0,
1185 MSR_IA32_VMX_CR4_FIXED1,
1186 MSR_IA32_VMX_VMCS_ENUM,
1187 MSR_IA32_VMX_PROCBASED_CTLS2,
1188 MSR_IA32_VMX_EPT_VPID_CAP,
1189 MSR_IA32_VMX_VMFUNC,
1193 MSR_IA32_ARCH_CAPABILITIES,
1196 static unsigned int num_msr_based_features;
1198 u64 kvm_get_arch_capabilities(void)
1202 rdmsrl_safe(MSR_IA32_ARCH_CAPABILITIES, &data);
1205 * If we're doing cache flushes (either "always" or "cond")
1206 * we will do one whenever the guest does a vmlaunch/vmresume.
1207 * If an outer hypervisor is doing the cache flush for us
1208 * (VMENTER_L1D_FLUSH_NESTED_VM), we can safely pass that
1209 * capability to the guest too, and if EPT is disabled we're not
1210 * vulnerable. Overall, only VMENTER_L1D_FLUSH_NEVER will
1211 * require a nested hypervisor to do a flush of its own.
1213 if (l1tf_vmx_mitigation != VMENTER_L1D_FLUSH_NEVER)
1214 data |= ARCH_CAP_SKIP_VMENTRY_L1DFLUSH;
1218 EXPORT_SYMBOL_GPL(kvm_get_arch_capabilities);
1220 static int kvm_get_msr_feature(struct kvm_msr_entry *msr)
1222 switch (msr->index) {
1223 case MSR_IA32_ARCH_CAPABILITIES:
1224 msr->data = kvm_get_arch_capabilities();
1226 case MSR_IA32_UCODE_REV:
1227 rdmsrl_safe(msr->index, &msr->data);
1230 if (kvm_x86_ops->get_msr_feature(msr))
1236 static int do_get_msr_feature(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1238 struct kvm_msr_entry msr;
1242 r = kvm_get_msr_feature(&msr);
1251 bool kvm_valid_efer(struct kvm_vcpu *vcpu, u64 efer)
1253 if (efer & efer_reserved_bits)
1256 if (efer & EFER_FFXSR && !guest_cpuid_has(vcpu, X86_FEATURE_FXSR_OPT))
1259 if (efer & EFER_SVME && !guest_cpuid_has(vcpu, X86_FEATURE_SVM))
1264 EXPORT_SYMBOL_GPL(kvm_valid_efer);
1266 static int set_efer(struct kvm_vcpu *vcpu, u64 efer)
1268 u64 old_efer = vcpu->arch.efer;
1270 if (!kvm_valid_efer(vcpu, efer))
1274 && (vcpu->arch.efer & EFER_LME) != (efer & EFER_LME))
1278 efer |= vcpu->arch.efer & EFER_LMA;
1280 kvm_x86_ops->set_efer(vcpu, efer);
1282 /* Update reserved bits */
1283 if ((efer ^ old_efer) & EFER_NX)
1284 kvm_mmu_reset_context(vcpu);
1289 void kvm_enable_efer_bits(u64 mask)
1291 efer_reserved_bits &= ~mask;
1293 EXPORT_SYMBOL_GPL(kvm_enable_efer_bits);
1296 * Writes msr value into into the appropriate "register".
1297 * Returns 0 on success, non-0 otherwise.
1298 * Assumes vcpu_load() was already called.
1300 int kvm_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
1302 switch (msr->index) {
1305 case MSR_KERNEL_GS_BASE:
1308 if (is_noncanonical_address(msr->data, vcpu))
1311 case MSR_IA32_SYSENTER_EIP:
1312 case MSR_IA32_SYSENTER_ESP:
1314 * IA32_SYSENTER_ESP and IA32_SYSENTER_EIP cause #GP if
1315 * non-canonical address is written on Intel but not on
1316 * AMD (which ignores the top 32-bits, because it does
1317 * not implement 64-bit SYSENTER).
1319 * 64-bit code should hence be able to write a non-canonical
1320 * value on AMD. Making the address canonical ensures that
1321 * vmentry does not fail on Intel after writing a non-canonical
1322 * value, and that something deterministic happens if the guest
1323 * invokes 64-bit SYSENTER.
1325 msr->data = get_canonical(msr->data, vcpu_virt_addr_bits(vcpu));
1327 return kvm_x86_ops->set_msr(vcpu, msr);
1329 EXPORT_SYMBOL_GPL(kvm_set_msr);
1332 * Adapt set_msr() to msr_io()'s calling convention
1334 static int do_get_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1336 struct msr_data msr;
1340 msr.host_initiated = true;
1341 r = kvm_get_msr(vcpu, &msr);
1349 static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
1351 struct msr_data msr;
1355 msr.host_initiated = true;
1356 return kvm_set_msr(vcpu, &msr);
1359 #ifdef CONFIG_X86_64
1360 struct pvclock_gtod_data {
1363 struct { /* extract of a clocksource struct */
1376 static struct pvclock_gtod_data pvclock_gtod_data;
1378 static void update_pvclock_gtod(struct timekeeper *tk)
1380 struct pvclock_gtod_data *vdata = &pvclock_gtod_data;
1383 boot_ns = ktime_to_ns(ktime_add(tk->tkr_mono.base, tk->offs_boot));
1385 write_seqcount_begin(&vdata->seq);
1387 /* copy pvclock gtod data */
1388 vdata->clock.vclock_mode = tk->tkr_mono.clock->archdata.vclock_mode;
1389 vdata->clock.cycle_last = tk->tkr_mono.cycle_last;
1390 vdata->clock.mask = tk->tkr_mono.mask;
1391 vdata->clock.mult = tk->tkr_mono.mult;
1392 vdata->clock.shift = tk->tkr_mono.shift;
1394 vdata->boot_ns = boot_ns;
1395 vdata->nsec_base = tk->tkr_mono.xtime_nsec;
1397 vdata->wall_time_sec = tk->xtime_sec;
1399 write_seqcount_end(&vdata->seq);
1403 void kvm_set_pending_timer(struct kvm_vcpu *vcpu)
1406 * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
1407 * vcpu_enter_guest. This function is only called from
1408 * the physical CPU that is running vcpu.
1410 kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
1413 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
1417 struct pvclock_wall_clock wc;
1418 struct timespec64 boot;
1423 r = kvm_read_guest(kvm, wall_clock, &version, sizeof(version));
1428 ++version; /* first time write, random junk */
1432 if (kvm_write_guest(kvm, wall_clock, &version, sizeof(version)))
1436 * The guest calculates current wall clock time by adding
1437 * system time (updated by kvm_guest_time_update below) to the
1438 * wall clock specified here. guest system time equals host
1439 * system time for us, thus we must fill in host boot time here.
1441 getboottime64(&boot);
1443 if (kvm->arch.kvmclock_offset) {
1444 struct timespec64 ts = ns_to_timespec64(kvm->arch.kvmclock_offset);
1445 boot = timespec64_sub(boot, ts);
1447 wc.sec = (u32)boot.tv_sec; /* overflow in 2106 guest time */
1448 wc.nsec = boot.tv_nsec;
1449 wc.version = version;
1451 kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
1454 kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
1457 static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
1459 do_shl32_div32(dividend, divisor);
1463 static void kvm_get_time_scale(uint64_t scaled_hz, uint64_t base_hz,
1464 s8 *pshift, u32 *pmultiplier)
1472 scaled64 = scaled_hz;
1473 while (tps64 > scaled64*2 || tps64 & 0xffffffff00000000ULL) {
1478 tps32 = (uint32_t)tps64;
1479 while (tps32 <= scaled64 || scaled64 & 0xffffffff00000000ULL) {
1480 if (scaled64 & 0xffffffff00000000ULL || tps32 & 0x80000000)
1488 *pmultiplier = div_frac(scaled64, tps32);
1490 pr_debug("%s: base_hz %llu => %llu, shift %d, mul %u\n",
1491 __func__, base_hz, scaled_hz, shift, *pmultiplier);
1494 #ifdef CONFIG_X86_64
1495 static atomic_t kvm_guest_has_master_clock = ATOMIC_INIT(0);
1498 static DEFINE_PER_CPU(unsigned long, cpu_tsc_khz);
1499 static unsigned long max_tsc_khz;
1501 static u32 adjust_tsc_khz(u32 khz, s32 ppm)
1503 u64 v = (u64)khz * (1000000 + ppm);
1508 static int set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1512 /* Guest TSC same frequency as host TSC? */
1514 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1518 /* TSC scaling supported? */
1519 if (!kvm_has_tsc_control) {
1520 if (user_tsc_khz > tsc_khz) {
1521 vcpu->arch.tsc_catchup = 1;
1522 vcpu->arch.tsc_always_catchup = 1;
1525 WARN(1, "user requested TSC rate below hardware speed\n");
1530 /* TSC scaling required - calculate ratio */
1531 ratio = mul_u64_u32_div(1ULL << kvm_tsc_scaling_ratio_frac_bits,
1532 user_tsc_khz, tsc_khz);
1534 if (ratio == 0 || ratio >= kvm_max_tsc_scaling_ratio) {
1535 WARN_ONCE(1, "Invalid TSC scaling ratio - virtual-tsc-khz=%u\n",
1540 vcpu->arch.tsc_scaling_ratio = ratio;
1544 static int kvm_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1546 u32 thresh_lo, thresh_hi;
1547 int use_scaling = 0;
1549 /* tsc_khz can be zero if TSC calibration fails */
1550 if (user_tsc_khz == 0) {
1551 /* set tsc_scaling_ratio to a safe value */
1552 vcpu->arch.tsc_scaling_ratio = kvm_default_tsc_scaling_ratio;
1556 /* Compute a scale to convert nanoseconds in TSC cycles */
1557 kvm_get_time_scale(user_tsc_khz * 1000LL, NSEC_PER_SEC,
1558 &vcpu->arch.virtual_tsc_shift,
1559 &vcpu->arch.virtual_tsc_mult);
1560 vcpu->arch.virtual_tsc_khz = user_tsc_khz;
1563 * Compute the variation in TSC rate which is acceptable
1564 * within the range of tolerance and decide if the
1565 * rate being applied is within that bounds of the hardware
1566 * rate. If so, no scaling or compensation need be done.
1568 thresh_lo = adjust_tsc_khz(tsc_khz, -tsc_tolerance_ppm);
1569 thresh_hi = adjust_tsc_khz(tsc_khz, tsc_tolerance_ppm);
1570 if (user_tsc_khz < thresh_lo || user_tsc_khz > thresh_hi) {
1571 pr_debug("kvm: requested TSC rate %u falls outside tolerance [%u,%u]\n", user_tsc_khz, thresh_lo, thresh_hi);
1574 return set_tsc_khz(vcpu, user_tsc_khz, use_scaling);
1577 static u64 compute_guest_tsc(struct kvm_vcpu *vcpu, s64 kernel_ns)
1579 u64 tsc = pvclock_scale_delta(kernel_ns-vcpu->arch.this_tsc_nsec,
1580 vcpu->arch.virtual_tsc_mult,
1581 vcpu->arch.virtual_tsc_shift);
1582 tsc += vcpu->arch.this_tsc_write;
1586 static inline int gtod_is_based_on_tsc(int mode)
1588 return mode == VCLOCK_TSC || mode == VCLOCK_HVCLOCK;
1591 static void kvm_track_tsc_matching(struct kvm_vcpu *vcpu)
1593 #ifdef CONFIG_X86_64
1595 struct kvm_arch *ka = &vcpu->kvm->arch;
1596 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1598 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1599 atomic_read(&vcpu->kvm->online_vcpus));
1602 * Once the masterclock is enabled, always perform request in
1603 * order to update it.
1605 * In order to enable masterclock, the host clocksource must be TSC
1606 * and the vcpus need to have matched TSCs. When that happens,
1607 * perform request to enable masterclock.
1609 if (ka->use_master_clock ||
1610 (gtod_is_based_on_tsc(gtod->clock.vclock_mode) && vcpus_matched))
1611 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
1613 trace_kvm_track_tsc(vcpu->vcpu_id, ka->nr_vcpus_matched_tsc,
1614 atomic_read(&vcpu->kvm->online_vcpus),
1615 ka->use_master_clock, gtod->clock.vclock_mode);
1619 static void update_ia32_tsc_adjust_msr(struct kvm_vcpu *vcpu, s64 offset)
1621 u64 curr_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1622 vcpu->arch.ia32_tsc_adjust_msr += offset - curr_offset;
1626 * Multiply tsc by a fixed point number represented by ratio.
1628 * The most significant 64-N bits (mult) of ratio represent the
1629 * integral part of the fixed point number; the remaining N bits
1630 * (frac) represent the fractional part, ie. ratio represents a fixed
1631 * point number (mult + frac * 2^(-N)).
1633 * N equals to kvm_tsc_scaling_ratio_frac_bits.
1635 static inline u64 __scale_tsc(u64 ratio, u64 tsc)
1637 return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
1640 u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
1643 u64 ratio = vcpu->arch.tsc_scaling_ratio;
1645 if (ratio != kvm_default_tsc_scaling_ratio)
1646 _tsc = __scale_tsc(ratio, tsc);
1650 EXPORT_SYMBOL_GPL(kvm_scale_tsc);
1652 static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1656 tsc = kvm_scale_tsc(vcpu, rdtsc());
1658 return target_tsc - tsc;
1661 u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
1663 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1665 return tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
1667 EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
1669 static void kvm_vcpu_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1671 vcpu->arch.tsc_offset = kvm_x86_ops->write_l1_tsc_offset(vcpu, offset);
1674 static inline bool kvm_check_tsc_unstable(void)
1676 #ifdef CONFIG_X86_64
1678 * TSC is marked unstable when we're running on Hyper-V,
1679 * 'TSC page' clocksource is good.
1681 if (pvclock_gtod_data.clock.vclock_mode == VCLOCK_HVCLOCK)
1684 return check_tsc_unstable();
1687 void kvm_write_tsc(struct kvm_vcpu *vcpu, struct msr_data *msr)
1689 struct kvm *kvm = vcpu->kvm;
1690 u64 offset, ns, elapsed;
1691 unsigned long flags;
1693 bool already_matched;
1694 u64 data = msr->data;
1695 bool synchronizing = false;
1697 raw_spin_lock_irqsave(&kvm->arch.tsc_write_lock, flags);
1698 offset = kvm_compute_tsc_offset(vcpu, data);
1699 ns = ktime_get_boot_ns();
1700 elapsed = ns - kvm->arch.last_tsc_nsec;
1702 if (vcpu->arch.virtual_tsc_khz) {
1703 if (data == 0 && msr->host_initiated) {
1705 * detection of vcpu initialization -- need to sync
1706 * with other vCPUs. This particularly helps to keep
1707 * kvm_clock stable after CPU hotplug
1709 synchronizing = true;
1711 u64 tsc_exp = kvm->arch.last_tsc_write +
1712 nsec_to_cycles(vcpu, elapsed);
1713 u64 tsc_hz = vcpu->arch.virtual_tsc_khz * 1000LL;
1715 * Special case: TSC write with a small delta (1 second)
1716 * of virtual cycle time against real time is
1717 * interpreted as an attempt to synchronize the CPU.
1719 synchronizing = data < tsc_exp + tsc_hz &&
1720 data + tsc_hz > tsc_exp;
1725 * For a reliable TSC, we can match TSC offsets, and for an unstable
1726 * TSC, we add elapsed time in this computation. We could let the
1727 * compensation code attempt to catch up if we fall behind, but
1728 * it's better to try to match offsets from the beginning.
1730 if (synchronizing &&
1731 vcpu->arch.virtual_tsc_khz == kvm->arch.last_tsc_khz) {
1732 if (!kvm_check_tsc_unstable()) {
1733 offset = kvm->arch.cur_tsc_offset;
1734 pr_debug("kvm: matched tsc offset for %llu\n", data);
1736 u64 delta = nsec_to_cycles(vcpu, elapsed);
1738 offset = kvm_compute_tsc_offset(vcpu, data);
1739 pr_debug("kvm: adjusted tsc offset by %llu\n", delta);
1742 already_matched = (vcpu->arch.this_tsc_generation == kvm->arch.cur_tsc_generation);
1745 * We split periods of matched TSC writes into generations.
1746 * For each generation, we track the original measured
1747 * nanosecond time, offset, and write, so if TSCs are in
1748 * sync, we can match exact offset, and if not, we can match
1749 * exact software computation in compute_guest_tsc()
1751 * These values are tracked in kvm->arch.cur_xxx variables.
1753 kvm->arch.cur_tsc_generation++;
1754 kvm->arch.cur_tsc_nsec = ns;
1755 kvm->arch.cur_tsc_write = data;
1756 kvm->arch.cur_tsc_offset = offset;
1758 pr_debug("kvm: new tsc generation %llu, clock %llu\n",
1759 kvm->arch.cur_tsc_generation, data);
1763 * We also track th most recent recorded KHZ, write and time to
1764 * allow the matching interval to be extended at each write.
1766 kvm->arch.last_tsc_nsec = ns;
1767 kvm->arch.last_tsc_write = data;
1768 kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
1770 vcpu->arch.last_guest_tsc = data;
1772 /* Keep track of which generation this VCPU has synchronized to */
1773 vcpu->arch.this_tsc_generation = kvm->arch.cur_tsc_generation;
1774 vcpu->arch.this_tsc_nsec = kvm->arch.cur_tsc_nsec;
1775 vcpu->arch.this_tsc_write = kvm->arch.cur_tsc_write;
1777 if (!msr->host_initiated && guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST))
1778 update_ia32_tsc_adjust_msr(vcpu, offset);
1780 kvm_vcpu_write_tsc_offset(vcpu, offset);
1781 raw_spin_unlock_irqrestore(&kvm->arch.tsc_write_lock, flags);
1783 spin_lock(&kvm->arch.pvclock_gtod_sync_lock);
1785 kvm->arch.nr_vcpus_matched_tsc = 0;
1786 } else if (!already_matched) {
1787 kvm->arch.nr_vcpus_matched_tsc++;
1790 kvm_track_tsc_matching(vcpu);
1791 spin_unlock(&kvm->arch.pvclock_gtod_sync_lock);
1794 EXPORT_SYMBOL_GPL(kvm_write_tsc);
1796 static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
1799 u64 tsc_offset = kvm_x86_ops->read_l1_tsc_offset(vcpu);
1800 kvm_vcpu_write_tsc_offset(vcpu, tsc_offset + adjustment);
1803 static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
1805 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
1806 WARN_ON(adjustment < 0);
1807 adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
1808 adjust_tsc_offset_guest(vcpu, adjustment);
1811 #ifdef CONFIG_X86_64
1813 static u64 read_tsc(void)
1815 u64 ret = (u64)rdtsc_ordered();
1816 u64 last = pvclock_gtod_data.clock.cycle_last;
1818 if (likely(ret >= last))
1822 * GCC likes to generate cmov here, but this branch is extremely
1823 * predictable (it's just a function of time and the likely is
1824 * very likely) and there's a data dependence, so force GCC
1825 * to generate a branch instead. I don't barrier() because
1826 * we don't actually need a barrier, and if this function
1827 * ever gets inlined it will generate worse code.
1833 static inline u64 vgettsc(u64 *tsc_timestamp, int *mode)
1836 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1839 switch (gtod->clock.vclock_mode) {
1840 case VCLOCK_HVCLOCK:
1841 tsc_pg_val = hv_read_tsc_page_tsc(hv_get_tsc_page(),
1843 if (tsc_pg_val != U64_MAX) {
1844 /* TSC page valid */
1845 *mode = VCLOCK_HVCLOCK;
1846 v = (tsc_pg_val - gtod->clock.cycle_last) &
1849 /* TSC page invalid */
1850 *mode = VCLOCK_NONE;
1855 *tsc_timestamp = read_tsc();
1856 v = (*tsc_timestamp - gtod->clock.cycle_last) &
1860 *mode = VCLOCK_NONE;
1863 if (*mode == VCLOCK_NONE)
1864 *tsc_timestamp = v = 0;
1866 return v * gtod->clock.mult;
1869 static int do_monotonic_boot(s64 *t, u64 *tsc_timestamp)
1871 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1877 seq = read_seqcount_begin(>od->seq);
1878 ns = gtod->nsec_base;
1879 ns += vgettsc(tsc_timestamp, &mode);
1880 ns >>= gtod->clock.shift;
1881 ns += gtod->boot_ns;
1882 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1888 static int do_realtime(struct timespec64 *ts, u64 *tsc_timestamp)
1890 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
1896 seq = read_seqcount_begin(>od->seq);
1897 ts->tv_sec = gtod->wall_time_sec;
1898 ns = gtod->nsec_base;
1899 ns += vgettsc(tsc_timestamp, &mode);
1900 ns >>= gtod->clock.shift;
1901 } while (unlikely(read_seqcount_retry(>od->seq, seq)));
1903 ts->tv_sec += __iter_div_u64_rem(ns, NSEC_PER_SEC, &ns);
1909 /* returns true if host is using TSC based clocksource */
1910 static bool kvm_get_time_and_clockread(s64 *kernel_ns, u64 *tsc_timestamp)
1912 /* checked again under seqlock below */
1913 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1916 return gtod_is_based_on_tsc(do_monotonic_boot(kernel_ns,
1920 /* returns true if host is using TSC based clocksource */
1921 static bool kvm_get_walltime_and_clockread(struct timespec64 *ts,
1924 /* checked again under seqlock below */
1925 if (!gtod_is_based_on_tsc(pvclock_gtod_data.clock.vclock_mode))
1928 return gtod_is_based_on_tsc(do_realtime(ts, tsc_timestamp));
1934 * Assuming a stable TSC across physical CPUS, and a stable TSC
1935 * across virtual CPUs, the following condition is possible.
1936 * Each numbered line represents an event visible to both
1937 * CPUs at the next numbered event.
1939 * "timespecX" represents host monotonic time. "tscX" represents
1942 * VCPU0 on CPU0 | VCPU1 on CPU1
1944 * 1. read timespec0,tsc0
1945 * 2. | timespec1 = timespec0 + N
1947 * 3. transition to guest | transition to guest
1948 * 4. ret0 = timespec0 + (rdtsc - tsc0) |
1949 * 5. | ret1 = timespec1 + (rdtsc - tsc1)
1950 * | ret1 = timespec0 + N + (rdtsc - (tsc0 + M))
1952 * Since ret0 update is visible to VCPU1 at time 5, to obey monotonicity:
1955 * - timespec0 + (rdtsc - tsc0) < timespec0 + N + (rdtsc - (tsc0 + M))
1957 * - 0 < N - M => M < N
1959 * That is, when timespec0 != timespec1, M < N. Unfortunately that is not
1960 * always the case (the difference between two distinct xtime instances
1961 * might be smaller then the difference between corresponding TSC reads,
1962 * when updating guest vcpus pvclock areas).
1964 * To avoid that problem, do not allow visibility of distinct
1965 * system_timestamp/tsc_timestamp values simultaneously: use a master
1966 * copy of host monotonic time values. Update that master copy
1969 * Rely on synchronization of host TSCs and guest TSCs for monotonicity.
1973 static void pvclock_update_vm_gtod_copy(struct kvm *kvm)
1975 #ifdef CONFIG_X86_64
1976 struct kvm_arch *ka = &kvm->arch;
1978 bool host_tsc_clocksource, vcpus_matched;
1980 vcpus_matched = (ka->nr_vcpus_matched_tsc + 1 ==
1981 atomic_read(&kvm->online_vcpus));
1984 * If the host uses TSC clock, then passthrough TSC as stable
1987 host_tsc_clocksource = kvm_get_time_and_clockread(
1988 &ka->master_kernel_ns,
1989 &ka->master_cycle_now);
1991 ka->use_master_clock = host_tsc_clocksource && vcpus_matched
1992 && !ka->backwards_tsc_observed
1993 && !ka->boot_vcpu_runs_old_kvmclock;
1995 if (ka->use_master_clock)
1996 atomic_set(&kvm_guest_has_master_clock, 1);
1998 vclock_mode = pvclock_gtod_data.clock.vclock_mode;
1999 trace_kvm_update_master_clock(ka->use_master_clock, vclock_mode,
2004 void kvm_make_mclock_inprogress_request(struct kvm *kvm)
2006 kvm_make_all_cpus_request(kvm, KVM_REQ_MCLOCK_INPROGRESS);
2009 static void kvm_gen_update_masterclock(struct kvm *kvm)
2011 #ifdef CONFIG_X86_64
2013 struct kvm_vcpu *vcpu;
2014 struct kvm_arch *ka = &kvm->arch;
2016 spin_lock(&ka->pvclock_gtod_sync_lock);
2017 kvm_make_mclock_inprogress_request(kvm);
2018 /* no guest entries from this point */
2019 pvclock_update_vm_gtod_copy(kvm);
2021 kvm_for_each_vcpu(i, vcpu, kvm)
2022 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2024 /* guest entries allowed */
2025 kvm_for_each_vcpu(i, vcpu, kvm)
2026 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
2028 spin_unlock(&ka->pvclock_gtod_sync_lock);
2032 u64 get_kvmclock_ns(struct kvm *kvm)
2034 struct kvm_arch *ka = &kvm->arch;
2035 struct pvclock_vcpu_time_info hv_clock;
2038 spin_lock(&ka->pvclock_gtod_sync_lock);
2039 if (!ka->use_master_clock) {
2040 spin_unlock(&ka->pvclock_gtod_sync_lock);
2041 return ktime_get_boot_ns() + ka->kvmclock_offset;
2044 hv_clock.tsc_timestamp = ka->master_cycle_now;
2045 hv_clock.system_time = ka->master_kernel_ns + ka->kvmclock_offset;
2046 spin_unlock(&ka->pvclock_gtod_sync_lock);
2048 /* both __this_cpu_read() and rdtsc() should be on the same cpu */
2051 if (__this_cpu_read(cpu_tsc_khz)) {
2052 kvm_get_time_scale(NSEC_PER_SEC, __this_cpu_read(cpu_tsc_khz) * 1000LL,
2053 &hv_clock.tsc_shift,
2054 &hv_clock.tsc_to_system_mul);
2055 ret = __pvclock_read_cycles(&hv_clock, rdtsc());
2057 ret = ktime_get_boot_ns() + ka->kvmclock_offset;
2064 static void kvm_setup_pvclock_page(struct kvm_vcpu *v)
2066 struct kvm_vcpu_arch *vcpu = &v->arch;
2067 struct pvclock_vcpu_time_info guest_hv_clock;
2069 if (unlikely(kvm_read_guest_cached(v->kvm, &vcpu->pv_time,
2070 &guest_hv_clock, sizeof(guest_hv_clock))))
2073 /* This VCPU is paused, but it's legal for a guest to read another
2074 * VCPU's kvmclock, so we really have to follow the specification where
2075 * it says that version is odd if data is being modified, and even after
2078 * Version field updates must be kept separate. This is because
2079 * kvm_write_guest_cached might use a "rep movs" instruction, and
2080 * writes within a string instruction are weakly ordered. So there
2081 * are three writes overall.
2083 * As a small optimization, only write the version field in the first
2084 * and third write. The vcpu->pv_time cache is still valid, because the
2085 * version field is the first in the struct.
2087 BUILD_BUG_ON(offsetof(struct pvclock_vcpu_time_info, version) != 0);
2089 if (guest_hv_clock.version & 1)
2090 ++guest_hv_clock.version; /* first time write, random junk */
2092 vcpu->hv_clock.version = guest_hv_clock.version + 1;
2093 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2095 sizeof(vcpu->hv_clock.version));
2099 /* retain PVCLOCK_GUEST_STOPPED if set in guest copy */
2100 vcpu->hv_clock.flags |= (guest_hv_clock.flags & PVCLOCK_GUEST_STOPPED);
2102 if (vcpu->pvclock_set_guest_stopped_request) {
2103 vcpu->hv_clock.flags |= PVCLOCK_GUEST_STOPPED;
2104 vcpu->pvclock_set_guest_stopped_request = false;
2107 trace_kvm_pvclock_update(v->vcpu_id, &vcpu->hv_clock);
2109 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2111 sizeof(vcpu->hv_clock));
2115 vcpu->hv_clock.version++;
2116 kvm_write_guest_cached(v->kvm, &vcpu->pv_time,
2118 sizeof(vcpu->hv_clock.version));
2121 static int kvm_guest_time_update(struct kvm_vcpu *v)
2123 unsigned long flags, tgt_tsc_khz;
2124 struct kvm_vcpu_arch *vcpu = &v->arch;
2125 struct kvm_arch *ka = &v->kvm->arch;
2127 u64 tsc_timestamp, host_tsc;
2129 bool use_master_clock;
2135 * If the host uses TSC clock, then passthrough TSC as stable
2138 spin_lock(&ka->pvclock_gtod_sync_lock);
2139 use_master_clock = ka->use_master_clock;
2140 if (use_master_clock) {
2141 host_tsc = ka->master_cycle_now;
2142 kernel_ns = ka->master_kernel_ns;
2144 spin_unlock(&ka->pvclock_gtod_sync_lock);
2146 /* Keep irq disabled to prevent changes to the clock */
2147 local_irq_save(flags);
2148 tgt_tsc_khz = __this_cpu_read(cpu_tsc_khz);
2149 if (unlikely(tgt_tsc_khz == 0)) {
2150 local_irq_restore(flags);
2151 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2154 if (!use_master_clock) {
2156 kernel_ns = ktime_get_boot_ns();
2159 tsc_timestamp = kvm_read_l1_tsc(v, host_tsc);
2162 * We may have to catch up the TSC to match elapsed wall clock
2163 * time for two reasons, even if kvmclock is used.
2164 * 1) CPU could have been running below the maximum TSC rate
2165 * 2) Broken TSC compensation resets the base at each VCPU
2166 * entry to avoid unknown leaps of TSC even when running
2167 * again on the same CPU. This may cause apparent elapsed
2168 * time to disappear, and the guest to stand still or run
2171 if (vcpu->tsc_catchup) {
2172 u64 tsc = compute_guest_tsc(v, kernel_ns);
2173 if (tsc > tsc_timestamp) {
2174 adjust_tsc_offset_guest(v, tsc - tsc_timestamp);
2175 tsc_timestamp = tsc;
2179 local_irq_restore(flags);
2181 /* With all the info we got, fill in the values */
2183 if (kvm_has_tsc_control)
2184 tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
2186 if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
2187 kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
2188 &vcpu->hv_clock.tsc_shift,
2189 &vcpu->hv_clock.tsc_to_system_mul);
2190 vcpu->hw_tsc_khz = tgt_tsc_khz;
2193 vcpu->hv_clock.tsc_timestamp = tsc_timestamp;
2194 vcpu->hv_clock.system_time = kernel_ns + v->kvm->arch.kvmclock_offset;
2195 vcpu->last_guest_tsc = tsc_timestamp;
2197 /* If the host uses TSC clocksource, then it is stable */
2199 if (use_master_clock)
2200 pvclock_flags |= PVCLOCK_TSC_STABLE_BIT;
2202 vcpu->hv_clock.flags = pvclock_flags;
2204 if (vcpu->pv_time_enabled)
2205 kvm_setup_pvclock_page(v);
2206 if (v == kvm_get_vcpu(v->kvm, 0))
2207 kvm_hv_setup_tsc_page(v->kvm, &vcpu->hv_clock);
2212 * kvmclock updates which are isolated to a given vcpu, such as
2213 * vcpu->cpu migration, should not allow system_timestamp from
2214 * the rest of the vcpus to remain static. Otherwise ntp frequency
2215 * correction applies to one vcpu's system_timestamp but not
2218 * So in those cases, request a kvmclock update for all vcpus.
2219 * We need to rate-limit these requests though, as they can
2220 * considerably slow guests that have a large number of vcpus.
2221 * The time for a remote vcpu to update its kvmclock is bound
2222 * by the delay we use to rate-limit the updates.
2225 #define KVMCLOCK_UPDATE_DELAY msecs_to_jiffies(100)
2227 static void kvmclock_update_fn(struct work_struct *work)
2230 struct delayed_work *dwork = to_delayed_work(work);
2231 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2232 kvmclock_update_work);
2233 struct kvm *kvm = container_of(ka, struct kvm, arch);
2234 struct kvm_vcpu *vcpu;
2236 kvm_for_each_vcpu(i, vcpu, kvm) {
2237 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
2238 kvm_vcpu_kick(vcpu);
2242 static void kvm_gen_kvmclock_update(struct kvm_vcpu *v)
2244 struct kvm *kvm = v->kvm;
2246 kvm_make_request(KVM_REQ_CLOCK_UPDATE, v);
2247 schedule_delayed_work(&kvm->arch.kvmclock_update_work,
2248 KVMCLOCK_UPDATE_DELAY);
2251 #define KVMCLOCK_SYNC_PERIOD (300 * HZ)
2253 static void kvmclock_sync_fn(struct work_struct *work)
2255 struct delayed_work *dwork = to_delayed_work(work);
2256 struct kvm_arch *ka = container_of(dwork, struct kvm_arch,
2257 kvmclock_sync_work);
2258 struct kvm *kvm = container_of(ka, struct kvm, arch);
2260 if (!kvmclock_periodic_sync)
2263 schedule_delayed_work(&kvm->arch.kvmclock_update_work, 0);
2264 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
2265 KVMCLOCK_SYNC_PERIOD);
2268 static int set_msr_mce(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2270 u64 mcg_cap = vcpu->arch.mcg_cap;
2271 unsigned bank_num = mcg_cap & 0xff;
2272 u32 msr = msr_info->index;
2273 u64 data = msr_info->data;
2276 case MSR_IA32_MCG_STATUS:
2277 vcpu->arch.mcg_status = data;
2279 case MSR_IA32_MCG_CTL:
2280 if (!(mcg_cap & MCG_CTL_P) &&
2281 (data || !msr_info->host_initiated))
2283 if (data != 0 && data != ~(u64)0)
2285 vcpu->arch.mcg_ctl = data;
2288 if (msr >= MSR_IA32_MC0_CTL &&
2289 msr < MSR_IA32_MCx_CTL(bank_num)) {
2290 u32 offset = msr - MSR_IA32_MC0_CTL;
2291 /* only 0 or all 1s can be written to IA32_MCi_CTL
2292 * some Linux kernels though clear bit 10 in bank 4 to
2293 * workaround a BIOS/GART TBL issue on AMD K8s, ignore
2294 * this to avoid an uncatched #GP in the guest
2296 if ((offset & 0x3) == 0 &&
2297 data != 0 && (data | (1 << 10)) != ~(u64)0)
2299 if (!msr_info->host_initiated &&
2300 (offset & 0x3) == 1 && data != 0)
2302 vcpu->arch.mce_banks[offset] = data;
2310 static int xen_hvm_config(struct kvm_vcpu *vcpu, u64 data)
2312 struct kvm *kvm = vcpu->kvm;
2313 int lm = is_long_mode(vcpu);
2314 u8 *blob_addr = lm ? (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_64
2315 : (u8 *)(long)kvm->arch.xen_hvm_config.blob_addr_32;
2316 u8 blob_size = lm ? kvm->arch.xen_hvm_config.blob_size_64
2317 : kvm->arch.xen_hvm_config.blob_size_32;
2318 u32 page_num = data & ~PAGE_MASK;
2319 u64 page_addr = data & PAGE_MASK;
2324 if (page_num >= blob_size)
2327 page = memdup_user(blob_addr + (page_num * PAGE_SIZE), PAGE_SIZE);
2332 if (kvm_vcpu_write_guest(vcpu, page_addr, page, PAGE_SIZE))
2341 static int kvm_pv_enable_async_pf(struct kvm_vcpu *vcpu, u64 data)
2343 gpa_t gpa = data & ~0x3f;
2345 /* Bits 3:5 are reserved, Should be zero */
2349 vcpu->arch.apf.msr_val = data;
2351 if (!(data & KVM_ASYNC_PF_ENABLED)) {
2352 kvm_clear_async_pf_completion_queue(vcpu);
2353 kvm_async_pf_hash_reset(vcpu);
2357 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.apf.data, gpa,
2361 vcpu->arch.apf.send_user_only = !(data & KVM_ASYNC_PF_SEND_ALWAYS);
2362 vcpu->arch.apf.delivery_as_pf_vmexit = data & KVM_ASYNC_PF_DELIVERY_AS_PF_VMEXIT;
2363 kvm_async_pf_wakeup_all(vcpu);
2367 static void kvmclock_reset(struct kvm_vcpu *vcpu)
2369 vcpu->arch.pv_time_enabled = false;
2372 static void kvm_vcpu_flush_tlb(struct kvm_vcpu *vcpu, bool invalidate_gpa)
2374 ++vcpu->stat.tlb_flush;
2375 kvm_x86_ops->tlb_flush(vcpu, invalidate_gpa);
2378 static void record_steal_time(struct kvm_vcpu *vcpu)
2380 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
2383 if (unlikely(kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2384 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time))))
2388 * Doing a TLB flush here, on the guest's behalf, can avoid
2391 if (xchg(&vcpu->arch.st.steal.preempted, 0) & KVM_VCPU_FLUSH_TLB)
2392 kvm_vcpu_flush_tlb(vcpu, false);
2394 if (vcpu->arch.st.steal.version & 1)
2395 vcpu->arch.st.steal.version += 1; /* first time write, random junk */
2397 vcpu->arch.st.steal.version += 1;
2399 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2400 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2404 vcpu->arch.st.steal.steal += current->sched_info.run_delay -
2405 vcpu->arch.st.last_steal;
2406 vcpu->arch.st.last_steal = current->sched_info.run_delay;
2408 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2409 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2413 vcpu->arch.st.steal.version += 1;
2415 kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.st.stime,
2416 &vcpu->arch.st.steal, sizeof(struct kvm_steal_time));
2419 int kvm_set_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2422 u32 msr = msr_info->index;
2423 u64 data = msr_info->data;
2426 case MSR_AMD64_NB_CFG:
2427 case MSR_IA32_UCODE_WRITE:
2428 case MSR_VM_HSAVE_PA:
2429 case MSR_AMD64_PATCH_LOADER:
2430 case MSR_AMD64_BU_CFG2:
2431 case MSR_AMD64_DC_CFG:
2434 case MSR_IA32_UCODE_REV:
2435 if (msr_info->host_initiated)
2436 vcpu->arch.microcode_version = data;
2439 return set_efer(vcpu, data);
2441 data &= ~(u64)0x40; /* ignore flush filter disable */
2442 data &= ~(u64)0x100; /* ignore ignne emulation enable */
2443 data &= ~(u64)0x8; /* ignore TLB cache disable */
2444 data &= ~(u64)0x40000; /* ignore Mc status write enable */
2446 vcpu_unimpl(vcpu, "unimplemented HWCR wrmsr: 0x%llx\n",
2451 case MSR_FAM10H_MMIO_CONF_BASE:
2453 vcpu_unimpl(vcpu, "unimplemented MMIO_CONF_BASE wrmsr: "
2458 case MSR_IA32_DEBUGCTLMSR:
2460 /* We support the non-activated case already */
2462 } else if (data & ~(DEBUGCTLMSR_LBR | DEBUGCTLMSR_BTF)) {
2463 /* Values other than LBR and BTF are vendor-specific,
2464 thus reserved and should throw a #GP */
2467 vcpu_unimpl(vcpu, "%s: MSR_IA32_DEBUGCTLMSR 0x%llx, nop\n",
2470 case 0x200 ... 0x2ff:
2471 return kvm_mtrr_set_msr(vcpu, msr, data);
2472 case MSR_IA32_APICBASE:
2473 return kvm_set_apic_base(vcpu, msr_info);
2474 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2475 return kvm_x2apic_msr_write(vcpu, msr, data);
2476 case MSR_IA32_TSCDEADLINE:
2477 kvm_set_lapic_tscdeadline_msr(vcpu, data);
2479 case MSR_IA32_TSC_ADJUST:
2480 if (guest_cpuid_has(vcpu, X86_FEATURE_TSC_ADJUST)) {
2481 if (!msr_info->host_initiated) {
2482 s64 adj = data - vcpu->arch.ia32_tsc_adjust_msr;
2483 adjust_tsc_offset_guest(vcpu, adj);
2485 vcpu->arch.ia32_tsc_adjust_msr = data;
2488 case MSR_IA32_MISC_ENABLE:
2489 vcpu->arch.ia32_misc_enable_msr = data;
2491 case MSR_IA32_SMBASE:
2492 if (!msr_info->host_initiated)
2494 vcpu->arch.smbase = data;
2497 kvm_write_tsc(vcpu, msr_info);
2500 if (!msr_info->host_initiated)
2502 vcpu->arch.smi_count = data;
2504 case MSR_KVM_WALL_CLOCK_NEW:
2505 case MSR_KVM_WALL_CLOCK:
2506 vcpu->kvm->arch.wall_clock = data;
2507 kvm_write_wall_clock(vcpu->kvm, data);
2509 case MSR_KVM_SYSTEM_TIME_NEW:
2510 case MSR_KVM_SYSTEM_TIME: {
2511 struct kvm_arch *ka = &vcpu->kvm->arch;
2513 kvmclock_reset(vcpu);
2515 if (vcpu->vcpu_id == 0 && !msr_info->host_initiated) {
2516 bool tmp = (msr == MSR_KVM_SYSTEM_TIME);
2518 if (ka->boot_vcpu_runs_old_kvmclock != tmp)
2519 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
2521 ka->boot_vcpu_runs_old_kvmclock = tmp;
2524 vcpu->arch.time = data;
2525 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
2527 /* we verify if the enable bit is set... */
2531 if (kvm_gfn_to_hva_cache_init(vcpu->kvm,
2532 &vcpu->arch.pv_time, data & ~1ULL,
2533 sizeof(struct pvclock_vcpu_time_info)))
2534 vcpu->arch.pv_time_enabled = false;
2536 vcpu->arch.pv_time_enabled = true;
2540 case MSR_KVM_ASYNC_PF_EN:
2541 if (kvm_pv_enable_async_pf(vcpu, data))
2544 case MSR_KVM_STEAL_TIME:
2546 if (unlikely(!sched_info_on()))
2549 if (data & KVM_STEAL_RESERVED_MASK)
2552 if (kvm_gfn_to_hva_cache_init(vcpu->kvm, &vcpu->arch.st.stime,
2553 data & KVM_STEAL_VALID_BITS,
2554 sizeof(struct kvm_steal_time)))
2557 vcpu->arch.st.msr_val = data;
2559 if (!(data & KVM_MSR_ENABLED))
2562 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
2565 case MSR_KVM_PV_EOI_EN:
2566 if (kvm_lapic_enable_pv_eoi(vcpu, data, sizeof(u8)))
2570 case MSR_IA32_MCG_CTL:
2571 case MSR_IA32_MCG_STATUS:
2572 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2573 return set_msr_mce(vcpu, msr_info);
2575 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2576 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2577 pr = true; /* fall through */
2578 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2579 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2580 if (kvm_pmu_is_valid_msr(vcpu, msr))
2581 return kvm_pmu_set_msr(vcpu, msr_info);
2583 if (pr || data != 0)
2584 vcpu_unimpl(vcpu, "disabled perfctr wrmsr: "
2585 "0x%x data 0x%llx\n", msr, data);
2587 case MSR_K7_CLK_CTL:
2589 * Ignore all writes to this no longer documented MSR.
2590 * Writes are only relevant for old K7 processors,
2591 * all pre-dating SVM, but a recommended workaround from
2592 * AMD for these chips. It is possible to specify the
2593 * affected processor models on the command line, hence
2594 * the need to ignore the workaround.
2597 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2598 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2599 case HV_X64_MSR_CRASH_CTL:
2600 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2601 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2602 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2603 case HV_X64_MSR_TSC_EMULATION_STATUS:
2604 return kvm_hv_set_msr_common(vcpu, msr, data,
2605 msr_info->host_initiated);
2606 case MSR_IA32_BBL_CR_CTL3:
2607 /* Drop writes to this legacy MSR -- see rdmsr
2608 * counterpart for further detail.
2610 if (report_ignored_msrs)
2611 vcpu_unimpl(vcpu, "ignored wrmsr: 0x%x data 0x%llx\n",
2614 case MSR_AMD64_OSVW_ID_LENGTH:
2615 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2617 vcpu->arch.osvw.length = data;
2619 case MSR_AMD64_OSVW_STATUS:
2620 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2622 vcpu->arch.osvw.status = data;
2624 case MSR_PLATFORM_INFO:
2625 if (!msr_info->host_initiated ||
2626 (!(data & MSR_PLATFORM_INFO_CPUID_FAULT) &&
2627 cpuid_fault_enabled(vcpu)))
2629 vcpu->arch.msr_platform_info = data;
2631 case MSR_MISC_FEATURES_ENABLES:
2632 if (data & ~MSR_MISC_FEATURES_ENABLES_CPUID_FAULT ||
2633 (data & MSR_MISC_FEATURES_ENABLES_CPUID_FAULT &&
2634 !supports_cpuid_fault(vcpu)))
2636 vcpu->arch.msr_misc_features_enables = data;
2639 if (msr && (msr == vcpu->kvm->arch.xen_hvm_config.msr))
2640 return xen_hvm_config(vcpu, data);
2641 if (kvm_pmu_is_valid_msr(vcpu, msr))
2642 return kvm_pmu_set_msr(vcpu, msr_info);
2644 vcpu_debug_ratelimited(vcpu, "unhandled wrmsr: 0x%x data 0x%llx\n",
2648 if (report_ignored_msrs)
2650 "ignored wrmsr: 0x%x data 0x%llx\n",
2657 EXPORT_SYMBOL_GPL(kvm_set_msr_common);
2661 * Reads an msr value (of 'msr_index') into 'pdata'.
2662 * Returns 0 on success, non-0 otherwise.
2663 * Assumes vcpu_load() was already called.
2665 int kvm_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr)
2667 return kvm_x86_ops->get_msr(vcpu, msr);
2669 EXPORT_SYMBOL_GPL(kvm_get_msr);
2671 static int get_msr_mce(struct kvm_vcpu *vcpu, u32 msr, u64 *pdata, bool host)
2674 u64 mcg_cap = vcpu->arch.mcg_cap;
2675 unsigned bank_num = mcg_cap & 0xff;
2678 case MSR_IA32_P5_MC_ADDR:
2679 case MSR_IA32_P5_MC_TYPE:
2682 case MSR_IA32_MCG_CAP:
2683 data = vcpu->arch.mcg_cap;
2685 case MSR_IA32_MCG_CTL:
2686 if (!(mcg_cap & MCG_CTL_P) && !host)
2688 data = vcpu->arch.mcg_ctl;
2690 case MSR_IA32_MCG_STATUS:
2691 data = vcpu->arch.mcg_status;
2694 if (msr >= MSR_IA32_MC0_CTL &&
2695 msr < MSR_IA32_MCx_CTL(bank_num)) {
2696 u32 offset = msr - MSR_IA32_MC0_CTL;
2697 data = vcpu->arch.mce_banks[offset];
2706 int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2708 switch (msr_info->index) {
2709 case MSR_IA32_PLATFORM_ID:
2710 case MSR_IA32_EBL_CR_POWERON:
2711 case MSR_IA32_DEBUGCTLMSR:
2712 case MSR_IA32_LASTBRANCHFROMIP:
2713 case MSR_IA32_LASTBRANCHTOIP:
2714 case MSR_IA32_LASTINTFROMIP:
2715 case MSR_IA32_LASTINTTOIP:
2717 case MSR_K8_TSEG_ADDR:
2718 case MSR_K8_TSEG_MASK:
2720 case MSR_VM_HSAVE_PA:
2721 case MSR_K8_INT_PENDING_MSG:
2722 case MSR_AMD64_NB_CFG:
2723 case MSR_FAM10H_MMIO_CONF_BASE:
2724 case MSR_AMD64_BU_CFG2:
2725 case MSR_IA32_PERF_CTL:
2726 case MSR_AMD64_DC_CFG:
2729 case MSR_F15H_PERF_CTL0 ... MSR_F15H_PERF_CTR5:
2730 case MSR_K7_EVNTSEL0 ... MSR_K7_EVNTSEL3:
2731 case MSR_K7_PERFCTR0 ... MSR_K7_PERFCTR3:
2732 case MSR_P6_PERFCTR0 ... MSR_P6_PERFCTR1:
2733 case MSR_P6_EVNTSEL0 ... MSR_P6_EVNTSEL1:
2734 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2735 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2738 case MSR_IA32_UCODE_REV:
2739 msr_info->data = vcpu->arch.microcode_version;
2742 msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + vcpu->arch.tsc_offset;
2745 case 0x200 ... 0x2ff:
2746 return kvm_mtrr_get_msr(vcpu, msr_info->index, &msr_info->data);
2747 case 0xcd: /* fsb frequency */
2751 * MSR_EBC_FREQUENCY_ID
2752 * Conservative value valid for even the basic CPU models.
2753 * Models 0,1: 000 in bits 23:21 indicating a bus speed of
2754 * 100MHz, model 2 000 in bits 18:16 indicating 100MHz,
2755 * and 266MHz for model 3, or 4. Set Core Clock
2756 * Frequency to System Bus Frequency Ratio to 1 (bits
2757 * 31:24) even though these are only valid for CPU
2758 * models > 2, however guests may end up dividing or
2759 * multiplying by zero otherwise.
2761 case MSR_EBC_FREQUENCY_ID:
2762 msr_info->data = 1 << 24;
2764 case MSR_IA32_APICBASE:
2765 msr_info->data = kvm_get_apic_base(vcpu);
2767 case APIC_BASE_MSR ... APIC_BASE_MSR + 0x3ff:
2768 return kvm_x2apic_msr_read(vcpu, msr_info->index, &msr_info->data);
2770 case MSR_IA32_TSCDEADLINE:
2771 msr_info->data = kvm_get_lapic_tscdeadline_msr(vcpu);
2773 case MSR_IA32_TSC_ADJUST:
2774 msr_info->data = (u64)vcpu->arch.ia32_tsc_adjust_msr;
2776 case MSR_IA32_MISC_ENABLE:
2777 msr_info->data = vcpu->arch.ia32_misc_enable_msr;
2779 case MSR_IA32_SMBASE:
2780 if (!msr_info->host_initiated)
2782 msr_info->data = vcpu->arch.smbase;
2785 msr_info->data = vcpu->arch.smi_count;
2787 case MSR_IA32_PERF_STATUS:
2788 /* TSC increment by tick */
2789 msr_info->data = 1000ULL;
2790 /* CPU multiplier */
2791 msr_info->data |= (((uint64_t)4ULL) << 40);
2794 msr_info->data = vcpu->arch.efer;
2796 case MSR_KVM_WALL_CLOCK:
2797 case MSR_KVM_WALL_CLOCK_NEW:
2798 msr_info->data = vcpu->kvm->arch.wall_clock;
2800 case MSR_KVM_SYSTEM_TIME:
2801 case MSR_KVM_SYSTEM_TIME_NEW:
2802 msr_info->data = vcpu->arch.time;
2804 case MSR_KVM_ASYNC_PF_EN:
2805 msr_info->data = vcpu->arch.apf.msr_val;
2807 case MSR_KVM_STEAL_TIME:
2808 msr_info->data = vcpu->arch.st.msr_val;
2810 case MSR_KVM_PV_EOI_EN:
2811 msr_info->data = vcpu->arch.pv_eoi.msr_val;
2813 case MSR_IA32_P5_MC_ADDR:
2814 case MSR_IA32_P5_MC_TYPE:
2815 case MSR_IA32_MCG_CAP:
2816 case MSR_IA32_MCG_CTL:
2817 case MSR_IA32_MCG_STATUS:
2818 case MSR_IA32_MC0_CTL ... MSR_IA32_MCx_CTL(KVM_MAX_MCE_BANKS) - 1:
2819 return get_msr_mce(vcpu, msr_info->index, &msr_info->data,
2820 msr_info->host_initiated);
2821 case MSR_K7_CLK_CTL:
2823 * Provide expected ramp-up count for K7. All other
2824 * are set to zero, indicating minimum divisors for
2827 * This prevents guest kernels on AMD host with CPU
2828 * type 6, model 8 and higher from exploding due to
2829 * the rdmsr failing.
2831 msr_info->data = 0x20000000;
2833 case HV_X64_MSR_GUEST_OS_ID ... HV_X64_MSR_SINT15:
2834 case HV_X64_MSR_CRASH_P0 ... HV_X64_MSR_CRASH_P4:
2835 case HV_X64_MSR_CRASH_CTL:
2836 case HV_X64_MSR_STIMER0_CONFIG ... HV_X64_MSR_STIMER3_COUNT:
2837 case HV_X64_MSR_REENLIGHTENMENT_CONTROL:
2838 case HV_X64_MSR_TSC_EMULATION_CONTROL:
2839 case HV_X64_MSR_TSC_EMULATION_STATUS:
2840 return kvm_hv_get_msr_common(vcpu,
2841 msr_info->index, &msr_info->data,
2842 msr_info->host_initiated);
2844 case MSR_IA32_BBL_CR_CTL3:
2845 /* This legacy MSR exists but isn't fully documented in current
2846 * silicon. It is however accessed by winxp in very narrow
2847 * scenarios where it sets bit #19, itself documented as
2848 * a "reserved" bit. Best effort attempt to source coherent
2849 * read data here should the balance of the register be
2850 * interpreted by the guest:
2852 * L2 cache control register 3: 64GB range, 256KB size,
2853 * enabled, latency 0x1, configured
2855 msr_info->data = 0xbe702111;
2857 case MSR_AMD64_OSVW_ID_LENGTH:
2858 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2860 msr_info->data = vcpu->arch.osvw.length;
2862 case MSR_AMD64_OSVW_STATUS:
2863 if (!guest_cpuid_has(vcpu, X86_FEATURE_OSVW))
2865 msr_info->data = vcpu->arch.osvw.status;
2867 case MSR_PLATFORM_INFO:
2868 if (!msr_info->host_initiated &&
2869 !vcpu->kvm->arch.guest_can_read_msr_platform_info)
2871 msr_info->data = vcpu->arch.msr_platform_info;
2873 case MSR_MISC_FEATURES_ENABLES:
2874 msr_info->data = vcpu->arch.msr_misc_features_enables;
2877 if (kvm_pmu_is_valid_msr(vcpu, msr_info->index))
2878 return kvm_pmu_get_msr(vcpu, msr_info->index, &msr_info->data);
2880 vcpu_debug_ratelimited(vcpu, "unhandled rdmsr: 0x%x\n",
2884 if (report_ignored_msrs)
2885 vcpu_unimpl(vcpu, "ignored rdmsr: 0x%x\n",
2893 EXPORT_SYMBOL_GPL(kvm_get_msr_common);
2896 * Read or write a bunch of msrs. All parameters are kernel addresses.
2898 * @return number of msrs set successfully.
2900 static int __msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs *msrs,
2901 struct kvm_msr_entry *entries,
2902 int (*do_msr)(struct kvm_vcpu *vcpu,
2903 unsigned index, u64 *data))
2907 for (i = 0; i < msrs->nmsrs; ++i)
2908 if (do_msr(vcpu, entries[i].index, &entries[i].data))
2915 * Read or write a bunch of msrs. Parameters are user addresses.
2917 * @return number of msrs set successfully.
2919 static int msr_io(struct kvm_vcpu *vcpu, struct kvm_msrs __user *user_msrs,
2920 int (*do_msr)(struct kvm_vcpu *vcpu,
2921 unsigned index, u64 *data),
2924 struct kvm_msrs msrs;
2925 struct kvm_msr_entry *entries;
2930 if (copy_from_user(&msrs, user_msrs, sizeof(msrs)))
2934 if (msrs.nmsrs >= MAX_IO_MSRS)
2937 size = sizeof(struct kvm_msr_entry) * msrs.nmsrs;
2938 entries = memdup_user(user_msrs->entries, size);
2939 if (IS_ERR(entries)) {
2940 r = PTR_ERR(entries);
2944 r = n = __msr_io(vcpu, &msrs, entries, do_msr);
2949 if (writeback && copy_to_user(user_msrs->entries, entries, size))
2960 static inline bool kvm_can_mwait_in_guest(void)
2962 return boot_cpu_has(X86_FEATURE_MWAIT) &&
2963 !boot_cpu_has_bug(X86_BUG_MONITOR) &&
2964 boot_cpu_has(X86_FEATURE_ARAT);
2967 int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
2972 case KVM_CAP_IRQCHIP:
2974 case KVM_CAP_MMU_SHADOW_CACHE_CONTROL:
2975 case KVM_CAP_SET_TSS_ADDR:
2976 case KVM_CAP_EXT_CPUID:
2977 case KVM_CAP_EXT_EMUL_CPUID:
2978 case KVM_CAP_CLOCKSOURCE:
2980 case KVM_CAP_NOP_IO_DELAY:
2981 case KVM_CAP_MP_STATE:
2982 case KVM_CAP_SYNC_MMU:
2983 case KVM_CAP_USER_NMI:
2984 case KVM_CAP_REINJECT_CONTROL:
2985 case KVM_CAP_IRQ_INJECT_STATUS:
2986 case KVM_CAP_IOEVENTFD:
2987 case KVM_CAP_IOEVENTFD_NO_LENGTH:
2989 case KVM_CAP_PIT_STATE2:
2990 case KVM_CAP_SET_IDENTITY_MAP_ADDR:
2991 case KVM_CAP_XEN_HVM:
2992 case KVM_CAP_VCPU_EVENTS:
2993 case KVM_CAP_HYPERV:
2994 case KVM_CAP_HYPERV_VAPIC:
2995 case KVM_CAP_HYPERV_SPIN:
2996 case KVM_CAP_HYPERV_SYNIC:
2997 case KVM_CAP_HYPERV_SYNIC2:
2998 case KVM_CAP_HYPERV_VP_INDEX:
2999 case KVM_CAP_HYPERV_EVENTFD:
3000 case KVM_CAP_HYPERV_TLBFLUSH:
3001 case KVM_CAP_HYPERV_SEND_IPI:
3002 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3003 case KVM_CAP_HYPERV_CPUID:
3004 case KVM_CAP_PCI_SEGMENT:
3005 case KVM_CAP_DEBUGREGS:
3006 case KVM_CAP_X86_ROBUST_SINGLESTEP:
3008 case KVM_CAP_ASYNC_PF:
3009 case KVM_CAP_GET_TSC_KHZ:
3010 case KVM_CAP_KVMCLOCK_CTRL:
3011 case KVM_CAP_READONLY_MEM:
3012 case KVM_CAP_HYPERV_TIME:
3013 case KVM_CAP_IOAPIC_POLARITY_IGNORED:
3014 case KVM_CAP_TSC_DEADLINE_TIMER:
3015 case KVM_CAP_DISABLE_QUIRKS:
3016 case KVM_CAP_SET_BOOT_CPU_ID:
3017 case KVM_CAP_SPLIT_IRQCHIP:
3018 case KVM_CAP_IMMEDIATE_EXIT:
3019 case KVM_CAP_GET_MSR_FEATURES:
3020 case KVM_CAP_MSR_PLATFORM_INFO:
3021 case KVM_CAP_EXCEPTION_PAYLOAD:
3024 case KVM_CAP_SYNC_REGS:
3025 r = KVM_SYNC_X86_VALID_FIELDS;
3027 case KVM_CAP_ADJUST_CLOCK:
3028 r = KVM_CLOCK_TSC_STABLE;
3030 case KVM_CAP_X86_DISABLE_EXITS:
3031 r |= KVM_X86_DISABLE_EXITS_HLT | KVM_X86_DISABLE_EXITS_PAUSE;
3032 if(kvm_can_mwait_in_guest())
3033 r |= KVM_X86_DISABLE_EXITS_MWAIT;
3035 case KVM_CAP_X86_SMM:
3036 /* SMBASE is usually relocated above 1M on modern chipsets,
3037 * and SMM handlers might indeed rely on 4G segment limits,
3038 * so do not report SMM to be available if real mode is
3039 * emulated via vm86 mode. Still, do not go to great lengths
3040 * to avoid userspace's usage of the feature, because it is a
3041 * fringe case that is not enabled except via specific settings
3042 * of the module parameters.
3044 r = kvm_x86_ops->has_emulated_msr(MSR_IA32_SMBASE);
3047 r = !kvm_x86_ops->cpu_has_accelerated_tpr();
3049 case KVM_CAP_NR_VCPUS:
3050 r = KVM_SOFT_MAX_VCPUS;
3052 case KVM_CAP_MAX_VCPUS:
3055 case KVM_CAP_NR_MEMSLOTS:
3056 r = KVM_USER_MEM_SLOTS;
3058 case KVM_CAP_PV_MMU: /* obsolete */
3062 r = KVM_MAX_MCE_BANKS;
3065 r = boot_cpu_has(X86_FEATURE_XSAVE);
3067 case KVM_CAP_TSC_CONTROL:
3068 r = kvm_has_tsc_control;
3070 case KVM_CAP_X2APIC_API:
3071 r = KVM_X2APIC_API_VALID_FLAGS;
3073 case KVM_CAP_NESTED_STATE:
3074 r = kvm_x86_ops->get_nested_state ?
3075 kvm_x86_ops->get_nested_state(NULL, 0, 0) : 0;
3084 long kvm_arch_dev_ioctl(struct file *filp,
3085 unsigned int ioctl, unsigned long arg)
3087 void __user *argp = (void __user *)arg;
3091 case KVM_GET_MSR_INDEX_LIST: {
3092 struct kvm_msr_list __user *user_msr_list = argp;
3093 struct kvm_msr_list msr_list;
3097 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3100 msr_list.nmsrs = num_msrs_to_save + num_emulated_msrs;
3101 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3104 if (n < msr_list.nmsrs)
3107 if (copy_to_user(user_msr_list->indices, &msrs_to_save,
3108 num_msrs_to_save * sizeof(u32)))
3110 if (copy_to_user(user_msr_list->indices + num_msrs_to_save,
3112 num_emulated_msrs * sizeof(u32)))
3117 case KVM_GET_SUPPORTED_CPUID:
3118 case KVM_GET_EMULATED_CPUID: {
3119 struct kvm_cpuid2 __user *cpuid_arg = argp;
3120 struct kvm_cpuid2 cpuid;
3123 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3126 r = kvm_dev_ioctl_get_cpuid(&cpuid, cpuid_arg->entries,
3132 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3137 case KVM_X86_GET_MCE_CAP_SUPPORTED: {
3139 if (copy_to_user(argp, &kvm_mce_cap_supported,
3140 sizeof(kvm_mce_cap_supported)))
3144 case KVM_GET_MSR_FEATURE_INDEX_LIST: {
3145 struct kvm_msr_list __user *user_msr_list = argp;
3146 struct kvm_msr_list msr_list;
3150 if (copy_from_user(&msr_list, user_msr_list, sizeof(msr_list)))
3153 msr_list.nmsrs = num_msr_based_features;
3154 if (copy_to_user(user_msr_list, &msr_list, sizeof(msr_list)))
3157 if (n < msr_list.nmsrs)
3160 if (copy_to_user(user_msr_list->indices, &msr_based_features,
3161 num_msr_based_features * sizeof(u32)))
3167 r = msr_io(NULL, argp, do_get_msr_feature, 1);
3177 static void wbinvd_ipi(void *garbage)
3182 static bool need_emulate_wbinvd(struct kvm_vcpu *vcpu)
3184 return kvm_arch_has_noncoherent_dma(vcpu->kvm);
3187 void kvm_arch_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
3189 /* Address WBINVD may be executed by guest */
3190 if (need_emulate_wbinvd(vcpu)) {
3191 if (kvm_x86_ops->has_wbinvd_exit())
3192 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
3193 else if (vcpu->cpu != -1 && vcpu->cpu != cpu)
3194 smp_call_function_single(vcpu->cpu,
3195 wbinvd_ipi, NULL, 1);
3198 kvm_x86_ops->vcpu_load(vcpu, cpu);
3200 /* Apply any externally detected TSC adjustments (due to suspend) */
3201 if (unlikely(vcpu->arch.tsc_offset_adjustment)) {
3202 adjust_tsc_offset_host(vcpu, vcpu->arch.tsc_offset_adjustment);
3203 vcpu->arch.tsc_offset_adjustment = 0;
3204 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3207 if (unlikely(vcpu->cpu != cpu) || kvm_check_tsc_unstable()) {
3208 s64 tsc_delta = !vcpu->arch.last_host_tsc ? 0 :
3209 rdtsc() - vcpu->arch.last_host_tsc;
3211 mark_tsc_unstable("KVM discovered backwards TSC");
3213 if (kvm_check_tsc_unstable()) {
3214 u64 offset = kvm_compute_tsc_offset(vcpu,
3215 vcpu->arch.last_guest_tsc);
3216 kvm_vcpu_write_tsc_offset(vcpu, offset);
3217 vcpu->arch.tsc_catchup = 1;
3220 if (kvm_lapic_hv_timer_in_use(vcpu))
3221 kvm_lapic_restart_hv_timer(vcpu);
3224 * On a host with synchronized TSC, there is no need to update
3225 * kvmclock on vcpu->cpu migration
3227 if (!vcpu->kvm->arch.use_master_clock || vcpu->cpu == -1)
3228 kvm_make_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu);
3229 if (vcpu->cpu != cpu)
3230 kvm_make_request(KVM_REQ_MIGRATE_TIMER, vcpu);
3234 kvm_make_request(KVM_REQ_STEAL_UPDATE, vcpu);
3237 static void kvm_steal_time_set_preempted(struct kvm_vcpu *vcpu)
3239 if (!(vcpu->arch.st.msr_val & KVM_MSR_ENABLED))
3242 vcpu->arch.st.steal.preempted = KVM_VCPU_PREEMPTED;
3244 kvm_write_guest_offset_cached(vcpu->kvm, &vcpu->arch.st.stime,
3245 &vcpu->arch.st.steal.preempted,
3246 offsetof(struct kvm_steal_time, preempted),
3247 sizeof(vcpu->arch.st.steal.preempted));
3250 void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
3254 if (vcpu->preempted)
3255 vcpu->arch.preempted_in_kernel = !kvm_x86_ops->get_cpl(vcpu);
3258 * Disable page faults because we're in atomic context here.
3259 * kvm_write_guest_offset_cached() would call might_fault()
3260 * that relies on pagefault_disable() to tell if there's a
3261 * bug. NOTE: the write to guest memory may not go through if
3262 * during postcopy live migration or if there's heavy guest
3265 pagefault_disable();
3267 * kvm_memslots() will be called by
3268 * kvm_write_guest_offset_cached() so take the srcu lock.
3270 idx = srcu_read_lock(&vcpu->kvm->srcu);
3271 kvm_steal_time_set_preempted(vcpu);
3272 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3274 kvm_x86_ops->vcpu_put(vcpu);
3275 vcpu->arch.last_host_tsc = rdtsc();
3277 * If userspace has set any breakpoints or watchpoints, dr6 is restored
3278 * on every vmexit, but if not, we might have a stale dr6 from the
3279 * guest. do_debug expects dr6 to be cleared after it runs, do the same.
3284 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,
3285 struct kvm_lapic_state *s)
3287 if (vcpu->arch.apicv_active)
3288 kvm_x86_ops->sync_pir_to_irr(vcpu);
3290 return kvm_apic_get_state(vcpu, s);
3293 static int kvm_vcpu_ioctl_set_lapic(struct kvm_vcpu *vcpu,
3294 struct kvm_lapic_state *s)
3298 r = kvm_apic_set_state(vcpu, s);
3301 update_cr8_intercept(vcpu);
3306 static int kvm_cpu_accept_dm_intr(struct kvm_vcpu *vcpu)
3308 return (!lapic_in_kernel(vcpu) ||
3309 kvm_apic_accept_pic_intr(vcpu));
3313 * if userspace requested an interrupt window, check that the
3314 * interrupt window is open.
3316 * No need to exit to userspace if we already have an interrupt queued.
3318 static int kvm_vcpu_ready_for_interrupt_injection(struct kvm_vcpu *vcpu)
3320 return kvm_arch_interrupt_allowed(vcpu) &&
3321 !kvm_cpu_has_interrupt(vcpu) &&
3322 !kvm_event_needs_reinjection(vcpu) &&
3323 kvm_cpu_accept_dm_intr(vcpu);
3326 static int kvm_vcpu_ioctl_interrupt(struct kvm_vcpu *vcpu,
3327 struct kvm_interrupt *irq)
3329 if (irq->irq >= KVM_NR_INTERRUPTS)
3332 if (!irqchip_in_kernel(vcpu->kvm)) {
3333 kvm_queue_interrupt(vcpu, irq->irq, false);
3334 kvm_make_request(KVM_REQ_EVENT, vcpu);
3339 * With in-kernel LAPIC, we only use this to inject EXTINT, so
3340 * fail for in-kernel 8259.
3342 if (pic_in_kernel(vcpu->kvm))
3345 if (vcpu->arch.pending_external_vector != -1)
3348 vcpu->arch.pending_external_vector = irq->irq;
3349 kvm_make_request(KVM_REQ_EVENT, vcpu);
3353 static int kvm_vcpu_ioctl_nmi(struct kvm_vcpu *vcpu)
3355 kvm_inject_nmi(vcpu);
3360 static int kvm_vcpu_ioctl_smi(struct kvm_vcpu *vcpu)
3362 kvm_make_request(KVM_REQ_SMI, vcpu);
3367 static int vcpu_ioctl_tpr_access_reporting(struct kvm_vcpu *vcpu,
3368 struct kvm_tpr_access_ctl *tac)
3372 vcpu->arch.tpr_access_reporting = !!tac->enabled;
3376 static int kvm_vcpu_ioctl_x86_setup_mce(struct kvm_vcpu *vcpu,
3380 unsigned bank_num = mcg_cap & 0xff, bank;
3383 if (!bank_num || bank_num >= KVM_MAX_MCE_BANKS)
3385 if (mcg_cap & ~(kvm_mce_cap_supported | 0xff | 0xff0000))
3388 vcpu->arch.mcg_cap = mcg_cap;
3389 /* Init IA32_MCG_CTL to all 1s */
3390 if (mcg_cap & MCG_CTL_P)
3391 vcpu->arch.mcg_ctl = ~(u64)0;
3392 /* Init IA32_MCi_CTL to all 1s */
3393 for (bank = 0; bank < bank_num; bank++)
3394 vcpu->arch.mce_banks[bank*4] = ~(u64)0;
3396 if (kvm_x86_ops->setup_mce)
3397 kvm_x86_ops->setup_mce(vcpu);
3402 static int kvm_vcpu_ioctl_x86_set_mce(struct kvm_vcpu *vcpu,
3403 struct kvm_x86_mce *mce)
3405 u64 mcg_cap = vcpu->arch.mcg_cap;
3406 unsigned bank_num = mcg_cap & 0xff;
3407 u64 *banks = vcpu->arch.mce_banks;
3409 if (mce->bank >= bank_num || !(mce->status & MCI_STATUS_VAL))
3412 * if IA32_MCG_CTL is not all 1s, the uncorrected error
3413 * reporting is disabled
3415 if ((mce->status & MCI_STATUS_UC) && (mcg_cap & MCG_CTL_P) &&
3416 vcpu->arch.mcg_ctl != ~(u64)0)
3418 banks += 4 * mce->bank;
3420 * if IA32_MCi_CTL is not all 1s, the uncorrected error
3421 * reporting is disabled for the bank
3423 if ((mce->status & MCI_STATUS_UC) && banks[0] != ~(u64)0)
3425 if (mce->status & MCI_STATUS_UC) {
3426 if ((vcpu->arch.mcg_status & MCG_STATUS_MCIP) ||
3427 !kvm_read_cr4_bits(vcpu, X86_CR4_MCE)) {
3428 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3431 if (banks[1] & MCI_STATUS_VAL)
3432 mce->status |= MCI_STATUS_OVER;
3433 banks[2] = mce->addr;
3434 banks[3] = mce->misc;
3435 vcpu->arch.mcg_status = mce->mcg_status;
3436 banks[1] = mce->status;
3437 kvm_queue_exception(vcpu, MC_VECTOR);
3438 } else if (!(banks[1] & MCI_STATUS_VAL)
3439 || !(banks[1] & MCI_STATUS_UC)) {
3440 if (banks[1] & MCI_STATUS_VAL)
3441 mce->status |= MCI_STATUS_OVER;
3442 banks[2] = mce->addr;
3443 banks[3] = mce->misc;
3444 banks[1] = mce->status;
3446 banks[1] |= MCI_STATUS_OVER;
3450 static void kvm_vcpu_ioctl_x86_get_vcpu_events(struct kvm_vcpu *vcpu,
3451 struct kvm_vcpu_events *events)
3456 * The API doesn't provide the instruction length for software
3457 * exceptions, so don't report them. As long as the guest RIP
3458 * isn't advanced, we should expect to encounter the exception
3461 if (kvm_exception_is_soft(vcpu->arch.exception.nr)) {
3462 events->exception.injected = 0;
3463 events->exception.pending = 0;
3465 events->exception.injected = vcpu->arch.exception.injected;
3466 events->exception.pending = vcpu->arch.exception.pending;
3468 * For ABI compatibility, deliberately conflate
3469 * pending and injected exceptions when
3470 * KVM_CAP_EXCEPTION_PAYLOAD isn't enabled.
3472 if (!vcpu->kvm->arch.exception_payload_enabled)
3473 events->exception.injected |=
3474 vcpu->arch.exception.pending;
3476 events->exception.nr = vcpu->arch.exception.nr;
3477 events->exception.has_error_code = vcpu->arch.exception.has_error_code;
3478 events->exception.error_code = vcpu->arch.exception.error_code;
3479 events->exception_has_payload = vcpu->arch.exception.has_payload;
3480 events->exception_payload = vcpu->arch.exception.payload;
3482 events->interrupt.injected =
3483 vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft;
3484 events->interrupt.nr = vcpu->arch.interrupt.nr;
3485 events->interrupt.soft = 0;
3486 events->interrupt.shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
3488 events->nmi.injected = vcpu->arch.nmi_injected;
3489 events->nmi.pending = vcpu->arch.nmi_pending != 0;
3490 events->nmi.masked = kvm_x86_ops->get_nmi_mask(vcpu);
3491 events->nmi.pad = 0;
3493 events->sipi_vector = 0; /* never valid when reporting to user space */
3495 events->smi.smm = is_smm(vcpu);
3496 events->smi.pending = vcpu->arch.smi_pending;
3497 events->smi.smm_inside_nmi =
3498 !!(vcpu->arch.hflags & HF_SMM_INSIDE_NMI_MASK);
3499 events->smi.latched_init = kvm_lapic_latched_init(vcpu);
3501 events->flags = (KVM_VCPUEVENT_VALID_NMI_PENDING
3502 | KVM_VCPUEVENT_VALID_SHADOW
3503 | KVM_VCPUEVENT_VALID_SMM);
3504 if (vcpu->kvm->arch.exception_payload_enabled)
3505 events->flags |= KVM_VCPUEVENT_VALID_PAYLOAD;
3507 memset(&events->reserved, 0, sizeof(events->reserved));
3510 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags);
3512 static int kvm_vcpu_ioctl_x86_set_vcpu_events(struct kvm_vcpu *vcpu,
3513 struct kvm_vcpu_events *events)
3515 if (events->flags & ~(KVM_VCPUEVENT_VALID_NMI_PENDING
3516 | KVM_VCPUEVENT_VALID_SIPI_VECTOR
3517 | KVM_VCPUEVENT_VALID_SHADOW
3518 | KVM_VCPUEVENT_VALID_SMM
3519 | KVM_VCPUEVENT_VALID_PAYLOAD))
3522 if (events->flags & KVM_VCPUEVENT_VALID_PAYLOAD) {
3523 if (!vcpu->kvm->arch.exception_payload_enabled)
3525 if (events->exception.pending)
3526 events->exception.injected = 0;
3528 events->exception_has_payload = 0;
3530 events->exception.pending = 0;
3531 events->exception_has_payload = 0;
3534 if ((events->exception.injected || events->exception.pending) &&
3535 (events->exception.nr > 31 || events->exception.nr == NMI_VECTOR))
3538 /* INITs are latched while in SMM */
3539 if (events->flags & KVM_VCPUEVENT_VALID_SMM &&
3540 (events->smi.smm || events->smi.pending) &&
3541 vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED)
3545 vcpu->arch.exception.injected = events->exception.injected;
3546 vcpu->arch.exception.pending = events->exception.pending;
3547 vcpu->arch.exception.nr = events->exception.nr;
3548 vcpu->arch.exception.has_error_code = events->exception.has_error_code;
3549 vcpu->arch.exception.error_code = events->exception.error_code;
3550 vcpu->arch.exception.has_payload = events->exception_has_payload;
3551 vcpu->arch.exception.payload = events->exception_payload;
3553 vcpu->arch.interrupt.injected = events->interrupt.injected;
3554 vcpu->arch.interrupt.nr = events->interrupt.nr;
3555 vcpu->arch.interrupt.soft = events->interrupt.soft;
3556 if (events->flags & KVM_VCPUEVENT_VALID_SHADOW)
3557 kvm_x86_ops->set_interrupt_shadow(vcpu,
3558 events->interrupt.shadow);
3560 vcpu->arch.nmi_injected = events->nmi.injected;
3561 if (events->flags & KVM_VCPUEVENT_VALID_NMI_PENDING)
3562 vcpu->arch.nmi_pending = events->nmi.pending;
3563 kvm_x86_ops->set_nmi_mask(vcpu, events->nmi.masked);
3565 if (events->flags & KVM_VCPUEVENT_VALID_SIPI_VECTOR &&
3566 lapic_in_kernel(vcpu))
3567 vcpu->arch.apic->sipi_vector = events->sipi_vector;
3569 if (events->flags & KVM_VCPUEVENT_VALID_SMM) {
3570 u32 hflags = vcpu->arch.hflags;
3571 if (events->smi.smm)
3572 hflags |= HF_SMM_MASK;
3574 hflags &= ~HF_SMM_MASK;
3575 kvm_set_hflags(vcpu, hflags);
3577 vcpu->arch.smi_pending = events->smi.pending;
3579 if (events->smi.smm) {
3580 if (events->smi.smm_inside_nmi)
3581 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
3583 vcpu->arch.hflags &= ~HF_SMM_INSIDE_NMI_MASK;
3584 if (lapic_in_kernel(vcpu)) {
3585 if (events->smi.latched_init)
3586 set_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3588 clear_bit(KVM_APIC_INIT, &vcpu->arch.apic->pending_events);
3593 kvm_make_request(KVM_REQ_EVENT, vcpu);
3598 static void kvm_vcpu_ioctl_x86_get_debugregs(struct kvm_vcpu *vcpu,
3599 struct kvm_debugregs *dbgregs)
3603 memcpy(dbgregs->db, vcpu->arch.db, sizeof(vcpu->arch.db));
3604 kvm_get_dr(vcpu, 6, &val);
3606 dbgregs->dr7 = vcpu->arch.dr7;
3608 memset(&dbgregs->reserved, 0, sizeof(dbgregs->reserved));
3611 static int kvm_vcpu_ioctl_x86_set_debugregs(struct kvm_vcpu *vcpu,
3612 struct kvm_debugregs *dbgregs)
3617 if (dbgregs->dr6 & ~0xffffffffull)
3619 if (dbgregs->dr7 & ~0xffffffffull)
3622 memcpy(vcpu->arch.db, dbgregs->db, sizeof(vcpu->arch.db));
3623 kvm_update_dr0123(vcpu);
3624 vcpu->arch.dr6 = dbgregs->dr6;
3625 kvm_update_dr6(vcpu);
3626 vcpu->arch.dr7 = dbgregs->dr7;
3627 kvm_update_dr7(vcpu);
3632 #define XSTATE_COMPACTION_ENABLED (1ULL << 63)
3634 static void fill_xsave(u8 *dest, struct kvm_vcpu *vcpu)
3636 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3637 u64 xstate_bv = xsave->header.xfeatures;
3641 * Copy legacy XSAVE area, to avoid complications with CPUID
3642 * leaves 0 and 1 in the loop below.
3644 memcpy(dest, xsave, XSAVE_HDR_OFFSET);
3647 xstate_bv &= vcpu->arch.guest_supported_xcr0 | XFEATURE_MASK_FPSSE;
3648 *(u64 *)(dest + XSAVE_HDR_OFFSET) = xstate_bv;
3651 * Copy each region from the possibly compacted offset to the
3652 * non-compacted offset.
3654 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3656 u64 feature = valid & -valid;
3657 int index = fls64(feature) - 1;
3658 void *src = get_xsave_addr(xsave, feature);
3661 u32 size, offset, ecx, edx;
3662 cpuid_count(XSTATE_CPUID, index,
3663 &size, &offset, &ecx, &edx);
3664 if (feature == XFEATURE_MASK_PKRU)
3665 memcpy(dest + offset, &vcpu->arch.pkru,
3666 sizeof(vcpu->arch.pkru));
3668 memcpy(dest + offset, src, size);
3676 static void load_xsave(struct kvm_vcpu *vcpu, u8 *src)
3678 struct xregs_state *xsave = &vcpu->arch.guest_fpu->state.xsave;
3679 u64 xstate_bv = *(u64 *)(src + XSAVE_HDR_OFFSET);
3683 * Copy legacy XSAVE area, to avoid complications with CPUID
3684 * leaves 0 and 1 in the loop below.
3686 memcpy(xsave, src, XSAVE_HDR_OFFSET);
3688 /* Set XSTATE_BV and possibly XCOMP_BV. */
3689 xsave->header.xfeatures = xstate_bv;
3690 if (boot_cpu_has(X86_FEATURE_XSAVES))
3691 xsave->header.xcomp_bv = host_xcr0 | XSTATE_COMPACTION_ENABLED;
3694 * Copy each region from the non-compacted offset to the
3695 * possibly compacted offset.
3697 valid = xstate_bv & ~XFEATURE_MASK_FPSSE;
3699 u64 feature = valid & -valid;
3700 int index = fls64(feature) - 1;
3701 void *dest = get_xsave_addr(xsave, feature);
3704 u32 size, offset, ecx, edx;
3705 cpuid_count(XSTATE_CPUID, index,
3706 &size, &offset, &ecx, &edx);
3707 if (feature == XFEATURE_MASK_PKRU)
3708 memcpy(&vcpu->arch.pkru, src + offset,
3709 sizeof(vcpu->arch.pkru));
3711 memcpy(dest, src + offset, size);
3718 static void kvm_vcpu_ioctl_x86_get_xsave(struct kvm_vcpu *vcpu,
3719 struct kvm_xsave *guest_xsave)
3721 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3722 memset(guest_xsave, 0, sizeof(struct kvm_xsave));
3723 fill_xsave((u8 *) guest_xsave->region, vcpu);
3725 memcpy(guest_xsave->region,
3726 &vcpu->arch.guest_fpu->state.fxsave,
3727 sizeof(struct fxregs_state));
3728 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)] =
3729 XFEATURE_MASK_FPSSE;
3733 #define XSAVE_MXCSR_OFFSET 24
3735 static int kvm_vcpu_ioctl_x86_set_xsave(struct kvm_vcpu *vcpu,
3736 struct kvm_xsave *guest_xsave)
3739 *(u64 *)&guest_xsave->region[XSAVE_HDR_OFFSET / sizeof(u32)];
3740 u32 mxcsr = *(u32 *)&guest_xsave->region[XSAVE_MXCSR_OFFSET / sizeof(u32)];
3742 if (boot_cpu_has(X86_FEATURE_XSAVE)) {
3744 * Here we allow setting states that are not present in
3745 * CPUID leaf 0xD, index 0, EDX:EAX. This is for compatibility
3746 * with old userspace.
3748 if (xstate_bv & ~kvm_supported_xcr0() ||
3749 mxcsr & ~mxcsr_feature_mask)
3751 load_xsave(vcpu, (u8 *)guest_xsave->region);
3753 if (xstate_bv & ~XFEATURE_MASK_FPSSE ||
3754 mxcsr & ~mxcsr_feature_mask)
3756 memcpy(&vcpu->arch.guest_fpu->state.fxsave,
3757 guest_xsave->region, sizeof(struct fxregs_state));
3762 static void kvm_vcpu_ioctl_x86_get_xcrs(struct kvm_vcpu *vcpu,
3763 struct kvm_xcrs *guest_xcrs)
3765 if (!boot_cpu_has(X86_FEATURE_XSAVE)) {
3766 guest_xcrs->nr_xcrs = 0;
3770 guest_xcrs->nr_xcrs = 1;
3771 guest_xcrs->flags = 0;
3772 guest_xcrs->xcrs[0].xcr = XCR_XFEATURE_ENABLED_MASK;
3773 guest_xcrs->xcrs[0].value = vcpu->arch.xcr0;
3776 static int kvm_vcpu_ioctl_x86_set_xcrs(struct kvm_vcpu *vcpu,
3777 struct kvm_xcrs *guest_xcrs)
3781 if (!boot_cpu_has(X86_FEATURE_XSAVE))
3784 if (guest_xcrs->nr_xcrs > KVM_MAX_XCRS || guest_xcrs->flags)
3787 for (i = 0; i < guest_xcrs->nr_xcrs; i++)
3788 /* Only support XCR0 currently */
3789 if (guest_xcrs->xcrs[i].xcr == XCR_XFEATURE_ENABLED_MASK) {
3790 r = __kvm_set_xcr(vcpu, XCR_XFEATURE_ENABLED_MASK,
3791 guest_xcrs->xcrs[i].value);
3800 * kvm_set_guest_paused() indicates to the guest kernel that it has been
3801 * stopped by the hypervisor. This function will be called from the host only.
3802 * EINVAL is returned when the host attempts to set the flag for a guest that
3803 * does not support pv clocks.
3805 static int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
3807 if (!vcpu->arch.pv_time_enabled)
3809 vcpu->arch.pvclock_set_guest_stopped_request = true;
3810 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
3814 static int kvm_vcpu_ioctl_enable_cap(struct kvm_vcpu *vcpu,
3815 struct kvm_enable_cap *cap)
3818 uint16_t vmcs_version;
3819 void __user *user_ptr;
3825 case KVM_CAP_HYPERV_SYNIC2:
3828 case KVM_CAP_HYPERV_SYNIC:
3829 if (!irqchip_in_kernel(vcpu->kvm))
3831 return kvm_hv_activate_synic(vcpu, cap->cap ==
3832 KVM_CAP_HYPERV_SYNIC2);
3833 case KVM_CAP_HYPERV_ENLIGHTENED_VMCS:
3834 if (!kvm_x86_ops->nested_enable_evmcs)
3836 r = kvm_x86_ops->nested_enable_evmcs(vcpu, &vmcs_version);
3838 user_ptr = (void __user *)(uintptr_t)cap->args[0];
3839 if (copy_to_user(user_ptr, &vmcs_version,
3840 sizeof(vmcs_version)))
3850 long kvm_arch_vcpu_ioctl(struct file *filp,
3851 unsigned int ioctl, unsigned long arg)
3853 struct kvm_vcpu *vcpu = filp->private_data;
3854 void __user *argp = (void __user *)arg;
3857 struct kvm_lapic_state *lapic;
3858 struct kvm_xsave *xsave;
3859 struct kvm_xcrs *xcrs;
3867 case KVM_GET_LAPIC: {
3869 if (!lapic_in_kernel(vcpu))
3871 u.lapic = kzalloc(sizeof(struct kvm_lapic_state), GFP_KERNEL);
3876 r = kvm_vcpu_ioctl_get_lapic(vcpu, u.lapic);
3880 if (copy_to_user(argp, u.lapic, sizeof(struct kvm_lapic_state)))
3885 case KVM_SET_LAPIC: {
3887 if (!lapic_in_kernel(vcpu))
3889 u.lapic = memdup_user(argp, sizeof(*u.lapic));
3890 if (IS_ERR(u.lapic)) {
3891 r = PTR_ERR(u.lapic);
3895 r = kvm_vcpu_ioctl_set_lapic(vcpu, u.lapic);
3898 case KVM_INTERRUPT: {
3899 struct kvm_interrupt irq;
3902 if (copy_from_user(&irq, argp, sizeof(irq)))
3904 r = kvm_vcpu_ioctl_interrupt(vcpu, &irq);
3908 r = kvm_vcpu_ioctl_nmi(vcpu);
3912 r = kvm_vcpu_ioctl_smi(vcpu);
3915 case KVM_SET_CPUID: {
3916 struct kvm_cpuid __user *cpuid_arg = argp;
3917 struct kvm_cpuid cpuid;
3920 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3922 r = kvm_vcpu_ioctl_set_cpuid(vcpu, &cpuid, cpuid_arg->entries);
3925 case KVM_SET_CPUID2: {
3926 struct kvm_cpuid2 __user *cpuid_arg = argp;
3927 struct kvm_cpuid2 cpuid;
3930 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3932 r = kvm_vcpu_ioctl_set_cpuid2(vcpu, &cpuid,
3933 cpuid_arg->entries);
3936 case KVM_GET_CPUID2: {
3937 struct kvm_cpuid2 __user *cpuid_arg = argp;
3938 struct kvm_cpuid2 cpuid;
3941 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
3943 r = kvm_vcpu_ioctl_get_cpuid2(vcpu, &cpuid,
3944 cpuid_arg->entries);
3948 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
3953 case KVM_GET_MSRS: {
3954 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3955 r = msr_io(vcpu, argp, do_get_msr, 1);
3956 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3959 case KVM_SET_MSRS: {
3960 int idx = srcu_read_lock(&vcpu->kvm->srcu);
3961 r = msr_io(vcpu, argp, do_set_msr, 0);
3962 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3965 case KVM_TPR_ACCESS_REPORTING: {
3966 struct kvm_tpr_access_ctl tac;
3969 if (copy_from_user(&tac, argp, sizeof(tac)))
3971 r = vcpu_ioctl_tpr_access_reporting(vcpu, &tac);
3975 if (copy_to_user(argp, &tac, sizeof(tac)))
3980 case KVM_SET_VAPIC_ADDR: {
3981 struct kvm_vapic_addr va;
3985 if (!lapic_in_kernel(vcpu))
3988 if (copy_from_user(&va, argp, sizeof(va)))
3990 idx = srcu_read_lock(&vcpu->kvm->srcu);
3991 r = kvm_lapic_set_vapic_addr(vcpu, va.vapic_addr);
3992 srcu_read_unlock(&vcpu->kvm->srcu, idx);
3995 case KVM_X86_SETUP_MCE: {
3999 if (copy_from_user(&mcg_cap, argp, sizeof(mcg_cap)))
4001 r = kvm_vcpu_ioctl_x86_setup_mce(vcpu, mcg_cap);
4004 case KVM_X86_SET_MCE: {
4005 struct kvm_x86_mce mce;
4008 if (copy_from_user(&mce, argp, sizeof(mce)))
4010 r = kvm_vcpu_ioctl_x86_set_mce(vcpu, &mce);
4013 case KVM_GET_VCPU_EVENTS: {
4014 struct kvm_vcpu_events events;
4016 kvm_vcpu_ioctl_x86_get_vcpu_events(vcpu, &events);
4019 if (copy_to_user(argp, &events, sizeof(struct kvm_vcpu_events)))
4024 case KVM_SET_VCPU_EVENTS: {
4025 struct kvm_vcpu_events events;
4028 if (copy_from_user(&events, argp, sizeof(struct kvm_vcpu_events)))
4031 r = kvm_vcpu_ioctl_x86_set_vcpu_events(vcpu, &events);
4034 case KVM_GET_DEBUGREGS: {
4035 struct kvm_debugregs dbgregs;
4037 kvm_vcpu_ioctl_x86_get_debugregs(vcpu, &dbgregs);
4040 if (copy_to_user(argp, &dbgregs,
4041 sizeof(struct kvm_debugregs)))
4046 case KVM_SET_DEBUGREGS: {
4047 struct kvm_debugregs dbgregs;
4050 if (copy_from_user(&dbgregs, argp,
4051 sizeof(struct kvm_debugregs)))
4054 r = kvm_vcpu_ioctl_x86_set_debugregs(vcpu, &dbgregs);
4057 case KVM_GET_XSAVE: {
4058 u.xsave = kzalloc(sizeof(struct kvm_xsave), GFP_KERNEL);
4063 kvm_vcpu_ioctl_x86_get_xsave(vcpu, u.xsave);
4066 if (copy_to_user(argp, u.xsave, sizeof(struct kvm_xsave)))
4071 case KVM_SET_XSAVE: {
4072 u.xsave = memdup_user(argp, sizeof(*u.xsave));
4073 if (IS_ERR(u.xsave)) {
4074 r = PTR_ERR(u.xsave);
4078 r = kvm_vcpu_ioctl_x86_set_xsave(vcpu, u.xsave);
4081 case KVM_GET_XCRS: {
4082 u.xcrs = kzalloc(sizeof(struct kvm_xcrs), GFP_KERNEL);
4087 kvm_vcpu_ioctl_x86_get_xcrs(vcpu, u.xcrs);
4090 if (copy_to_user(argp, u.xcrs,
4091 sizeof(struct kvm_xcrs)))
4096 case KVM_SET_XCRS: {
4097 u.xcrs = memdup_user(argp, sizeof(*u.xcrs));
4098 if (IS_ERR(u.xcrs)) {
4099 r = PTR_ERR(u.xcrs);
4103 r = kvm_vcpu_ioctl_x86_set_xcrs(vcpu, u.xcrs);
4106 case KVM_SET_TSC_KHZ: {
4110 user_tsc_khz = (u32)arg;
4112 if (user_tsc_khz >= kvm_max_guest_tsc_khz)
4115 if (user_tsc_khz == 0)
4116 user_tsc_khz = tsc_khz;
4118 if (!kvm_set_tsc_khz(vcpu, user_tsc_khz))
4123 case KVM_GET_TSC_KHZ: {
4124 r = vcpu->arch.virtual_tsc_khz;
4127 case KVM_KVMCLOCK_CTRL: {
4128 r = kvm_set_guest_paused(vcpu);
4131 case KVM_ENABLE_CAP: {
4132 struct kvm_enable_cap cap;
4135 if (copy_from_user(&cap, argp, sizeof(cap)))
4137 r = kvm_vcpu_ioctl_enable_cap(vcpu, &cap);
4140 case KVM_GET_NESTED_STATE: {
4141 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4145 if (!kvm_x86_ops->get_nested_state)
4148 BUILD_BUG_ON(sizeof(user_data_size) != sizeof(user_kvm_nested_state->size));
4150 if (get_user(user_data_size, &user_kvm_nested_state->size))
4153 r = kvm_x86_ops->get_nested_state(vcpu, user_kvm_nested_state,
4158 if (r > user_data_size) {
4159 if (put_user(r, &user_kvm_nested_state->size))
4169 case KVM_SET_NESTED_STATE: {
4170 struct kvm_nested_state __user *user_kvm_nested_state = argp;
4171 struct kvm_nested_state kvm_state;
4174 if (!kvm_x86_ops->set_nested_state)
4178 if (copy_from_user(&kvm_state, user_kvm_nested_state, sizeof(kvm_state)))
4182 if (kvm_state.size < sizeof(kvm_state))
4185 if (kvm_state.flags &
4186 ~(KVM_STATE_NESTED_RUN_PENDING | KVM_STATE_NESTED_GUEST_MODE
4187 | KVM_STATE_NESTED_EVMCS))
4190 /* nested_run_pending implies guest_mode. */
4191 if ((kvm_state.flags & KVM_STATE_NESTED_RUN_PENDING)
4192 && !(kvm_state.flags & KVM_STATE_NESTED_GUEST_MODE))
4195 r = kvm_x86_ops->set_nested_state(vcpu, user_kvm_nested_state, &kvm_state);
4198 case KVM_GET_SUPPORTED_HV_CPUID: {
4199 struct kvm_cpuid2 __user *cpuid_arg = argp;
4200 struct kvm_cpuid2 cpuid;
4203 if (copy_from_user(&cpuid, cpuid_arg, sizeof(cpuid)))
4206 r = kvm_vcpu_ioctl_get_hv_cpuid(vcpu, &cpuid,
4207 cpuid_arg->entries);
4212 if (copy_to_user(cpuid_arg, &cpuid, sizeof(cpuid)))
4227 vm_fault_t kvm_arch_vcpu_fault(struct kvm_vcpu *vcpu, struct vm_fault *vmf)
4229 return VM_FAULT_SIGBUS;
4232 static int kvm_vm_ioctl_set_tss_addr(struct kvm *kvm, unsigned long addr)
4236 if (addr > (unsigned int)(-3 * PAGE_SIZE))
4238 ret = kvm_x86_ops->set_tss_addr(kvm, addr);
4242 static int kvm_vm_ioctl_set_identity_map_addr(struct kvm *kvm,
4245 return kvm_x86_ops->set_identity_map_addr(kvm, ident_addr);
4248 static int kvm_vm_ioctl_set_nr_mmu_pages(struct kvm *kvm,
4249 u32 kvm_nr_mmu_pages)
4251 if (kvm_nr_mmu_pages < KVM_MIN_ALLOC_MMU_PAGES)
4254 mutex_lock(&kvm->slots_lock);
4256 kvm_mmu_change_mmu_pages(kvm, kvm_nr_mmu_pages);
4257 kvm->arch.n_requested_mmu_pages = kvm_nr_mmu_pages;
4259 mutex_unlock(&kvm->slots_lock);
4263 static int kvm_vm_ioctl_get_nr_mmu_pages(struct kvm *kvm)
4265 return kvm->arch.n_max_mmu_pages;
4268 static int kvm_vm_ioctl_get_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4270 struct kvm_pic *pic = kvm->arch.vpic;
4274 switch (chip->chip_id) {
4275 case KVM_IRQCHIP_PIC_MASTER:
4276 memcpy(&chip->chip.pic, &pic->pics[0],
4277 sizeof(struct kvm_pic_state));
4279 case KVM_IRQCHIP_PIC_SLAVE:
4280 memcpy(&chip->chip.pic, &pic->pics[1],
4281 sizeof(struct kvm_pic_state));
4283 case KVM_IRQCHIP_IOAPIC:
4284 kvm_get_ioapic(kvm, &chip->chip.ioapic);
4293 static int kvm_vm_ioctl_set_irqchip(struct kvm *kvm, struct kvm_irqchip *chip)
4295 struct kvm_pic *pic = kvm->arch.vpic;
4299 switch (chip->chip_id) {
4300 case KVM_IRQCHIP_PIC_MASTER:
4301 spin_lock(&pic->lock);
4302 memcpy(&pic->pics[0], &chip->chip.pic,
4303 sizeof(struct kvm_pic_state));
4304 spin_unlock(&pic->lock);
4306 case KVM_IRQCHIP_PIC_SLAVE:
4307 spin_lock(&pic->lock);
4308 memcpy(&pic->pics[1], &chip->chip.pic,
4309 sizeof(struct kvm_pic_state));
4310 spin_unlock(&pic->lock);
4312 case KVM_IRQCHIP_IOAPIC:
4313 kvm_set_ioapic(kvm, &chip->chip.ioapic);
4319 kvm_pic_update_irq(pic);
4323 static int kvm_vm_ioctl_get_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4325 struct kvm_kpit_state *kps = &kvm->arch.vpit->pit_state;
4327 BUILD_BUG_ON(sizeof(*ps) != sizeof(kps->channels));
4329 mutex_lock(&kps->lock);
4330 memcpy(ps, &kps->channels, sizeof(*ps));
4331 mutex_unlock(&kps->lock);
4335 static int kvm_vm_ioctl_set_pit(struct kvm *kvm, struct kvm_pit_state *ps)
4338 struct kvm_pit *pit = kvm->arch.vpit;
4340 mutex_lock(&pit->pit_state.lock);
4341 memcpy(&pit->pit_state.channels, ps, sizeof(*ps));
4342 for (i = 0; i < 3; i++)
4343 kvm_pit_load_count(pit, i, ps->channels[i].count, 0);
4344 mutex_unlock(&pit->pit_state.lock);
4348 static int kvm_vm_ioctl_get_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4350 mutex_lock(&kvm->arch.vpit->pit_state.lock);
4351 memcpy(ps->channels, &kvm->arch.vpit->pit_state.channels,
4352 sizeof(ps->channels));
4353 ps->flags = kvm->arch.vpit->pit_state.flags;
4354 mutex_unlock(&kvm->arch.vpit->pit_state.lock);
4355 memset(&ps->reserved, 0, sizeof(ps->reserved));
4359 static int kvm_vm_ioctl_set_pit2(struct kvm *kvm, struct kvm_pit_state2 *ps)
4363 u32 prev_legacy, cur_legacy;
4364 struct kvm_pit *pit = kvm->arch.vpit;
4366 mutex_lock(&pit->pit_state.lock);
4367 prev_legacy = pit->pit_state.flags & KVM_PIT_FLAGS_HPET_LEGACY;
4368 cur_legacy = ps->flags & KVM_PIT_FLAGS_HPET_LEGACY;
4369 if (!prev_legacy && cur_legacy)
4371 memcpy(&pit->pit_state.channels, &ps->channels,
4372 sizeof(pit->pit_state.channels));
4373 pit->pit_state.flags = ps->flags;
4374 for (i = 0; i < 3; i++)
4375 kvm_pit_load_count(pit, i, pit->pit_state.channels[i].count,
4377 mutex_unlock(&pit->pit_state.lock);
4381 static int kvm_vm_ioctl_reinject(struct kvm *kvm,
4382 struct kvm_reinject_control *control)
4384 struct kvm_pit *pit = kvm->arch.vpit;
4389 /* pit->pit_state.lock was overloaded to prevent userspace from getting
4390 * an inconsistent state after running multiple KVM_REINJECT_CONTROL
4391 * ioctls in parallel. Use a separate lock if that ioctl isn't rare.
4393 mutex_lock(&pit->pit_state.lock);
4394 kvm_pit_set_reinject(pit, control->pit_reinject);
4395 mutex_unlock(&pit->pit_state.lock);
4401 * kvm_vm_ioctl_get_dirty_log - get and clear the log of dirty pages in a slot
4402 * @kvm: kvm instance
4403 * @log: slot id and address to which we copy the log
4405 * Steps 1-4 below provide general overview of dirty page logging. See
4406 * kvm_get_dirty_log_protect() function description for additional details.
4408 * We call kvm_get_dirty_log_protect() to handle steps 1-3, upon return we
4409 * always flush the TLB (step 4) even if previous step failed and the dirty
4410 * bitmap may be corrupt. Regardless of previous outcome the KVM logging API
4411 * does not preclude user space subsequent dirty log read. Flushing TLB ensures
4412 * writes will be marked dirty for next log read.
4414 * 1. Take a snapshot of the bit and clear it if needed.
4415 * 2. Write protect the corresponding page.
4416 * 3. Copy the snapshot to the userspace.
4417 * 4. Flush TLB's if needed.
4419 int kvm_vm_ioctl_get_dirty_log(struct kvm *kvm, struct kvm_dirty_log *log)
4424 mutex_lock(&kvm->slots_lock);
4427 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4429 if (kvm_x86_ops->flush_log_dirty)
4430 kvm_x86_ops->flush_log_dirty(kvm);
4432 r = kvm_get_dirty_log_protect(kvm, log, &flush);
4435 * All the TLBs can be flushed out of mmu lock, see the comments in
4436 * kvm_mmu_slot_remove_write_access().
4438 lockdep_assert_held(&kvm->slots_lock);
4440 kvm_flush_remote_tlbs(kvm);
4442 mutex_unlock(&kvm->slots_lock);
4446 int kvm_vm_ioctl_clear_dirty_log(struct kvm *kvm, struct kvm_clear_dirty_log *log)
4451 mutex_lock(&kvm->slots_lock);
4454 * Flush potentially hardware-cached dirty pages to dirty_bitmap.
4456 if (kvm_x86_ops->flush_log_dirty)
4457 kvm_x86_ops->flush_log_dirty(kvm);
4459 r = kvm_clear_dirty_log_protect(kvm, log, &flush);
4462 * All the TLBs can be flushed out of mmu lock, see the comments in
4463 * kvm_mmu_slot_remove_write_access().
4465 lockdep_assert_held(&kvm->slots_lock);
4467 kvm_flush_remote_tlbs(kvm);
4469 mutex_unlock(&kvm->slots_lock);
4473 int kvm_vm_ioctl_irq_line(struct kvm *kvm, struct kvm_irq_level *irq_event,
4476 if (!irqchip_in_kernel(kvm))
4479 irq_event->status = kvm_set_irq(kvm, KVM_USERSPACE_IRQ_SOURCE_ID,
4480 irq_event->irq, irq_event->level,
4485 int kvm_vm_ioctl_enable_cap(struct kvm *kvm,
4486 struct kvm_enable_cap *cap)
4494 case KVM_CAP_DISABLE_QUIRKS:
4495 kvm->arch.disabled_quirks = cap->args[0];
4498 case KVM_CAP_SPLIT_IRQCHIP: {
4499 mutex_lock(&kvm->lock);
4501 if (cap->args[0] > MAX_NR_RESERVED_IOAPIC_PINS)
4502 goto split_irqchip_unlock;
4504 if (irqchip_in_kernel(kvm))
4505 goto split_irqchip_unlock;
4506 if (kvm->created_vcpus)
4507 goto split_irqchip_unlock;
4508 r = kvm_setup_empty_irq_routing(kvm);
4510 goto split_irqchip_unlock;
4511 /* Pairs with irqchip_in_kernel. */
4513 kvm->arch.irqchip_mode = KVM_IRQCHIP_SPLIT;
4514 kvm->arch.nr_reserved_ioapic_pins = cap->args[0];
4516 split_irqchip_unlock:
4517 mutex_unlock(&kvm->lock);
4520 case KVM_CAP_X2APIC_API:
4522 if (cap->args[0] & ~KVM_X2APIC_API_VALID_FLAGS)
4525 if (cap->args[0] & KVM_X2APIC_API_USE_32BIT_IDS)
4526 kvm->arch.x2apic_format = true;
4527 if (cap->args[0] & KVM_X2APIC_API_DISABLE_BROADCAST_QUIRK)
4528 kvm->arch.x2apic_broadcast_quirk_disabled = true;
4532 case KVM_CAP_X86_DISABLE_EXITS:
4534 if (cap->args[0] & ~KVM_X86_DISABLE_VALID_EXITS)
4537 if ((cap->args[0] & KVM_X86_DISABLE_EXITS_MWAIT) &&
4538 kvm_can_mwait_in_guest())
4539 kvm->arch.mwait_in_guest = true;
4540 if (cap->args[0] & KVM_X86_DISABLE_EXITS_HLT)
4541 kvm->arch.hlt_in_guest = true;
4542 if (cap->args[0] & KVM_X86_DISABLE_EXITS_PAUSE)
4543 kvm->arch.pause_in_guest = true;
4546 case KVM_CAP_MSR_PLATFORM_INFO:
4547 kvm->arch.guest_can_read_msr_platform_info = cap->args[0];
4550 case KVM_CAP_EXCEPTION_PAYLOAD:
4551 kvm->arch.exception_payload_enabled = cap->args[0];
4561 long kvm_arch_vm_ioctl(struct file *filp,
4562 unsigned int ioctl, unsigned long arg)
4564 struct kvm *kvm = filp->private_data;
4565 void __user *argp = (void __user *)arg;
4568 * This union makes it completely explicit to gcc-3.x
4569 * that these two variables' stack usage should be
4570 * combined, not added together.
4573 struct kvm_pit_state ps;
4574 struct kvm_pit_state2 ps2;
4575 struct kvm_pit_config pit_config;
4579 case KVM_SET_TSS_ADDR:
4580 r = kvm_vm_ioctl_set_tss_addr(kvm, arg);
4582 case KVM_SET_IDENTITY_MAP_ADDR: {
4585 mutex_lock(&kvm->lock);
4587 if (kvm->created_vcpus)
4588 goto set_identity_unlock;
4590 if (copy_from_user(&ident_addr, argp, sizeof(ident_addr)))
4591 goto set_identity_unlock;
4592 r = kvm_vm_ioctl_set_identity_map_addr(kvm, ident_addr);
4593 set_identity_unlock:
4594 mutex_unlock(&kvm->lock);
4597 case KVM_SET_NR_MMU_PAGES:
4598 r = kvm_vm_ioctl_set_nr_mmu_pages(kvm, arg);
4600 case KVM_GET_NR_MMU_PAGES:
4601 r = kvm_vm_ioctl_get_nr_mmu_pages(kvm);
4603 case KVM_CREATE_IRQCHIP: {
4604 mutex_lock(&kvm->lock);
4607 if (irqchip_in_kernel(kvm))
4608 goto create_irqchip_unlock;
4611 if (kvm->created_vcpus)
4612 goto create_irqchip_unlock;
4614 r = kvm_pic_init(kvm);
4616 goto create_irqchip_unlock;
4618 r = kvm_ioapic_init(kvm);
4620 kvm_pic_destroy(kvm);
4621 goto create_irqchip_unlock;
4624 r = kvm_setup_default_irq_routing(kvm);
4626 kvm_ioapic_destroy(kvm);
4627 kvm_pic_destroy(kvm);
4628 goto create_irqchip_unlock;
4630 /* Write kvm->irq_routing before enabling irqchip_in_kernel. */
4632 kvm->arch.irqchip_mode = KVM_IRQCHIP_KERNEL;
4633 create_irqchip_unlock:
4634 mutex_unlock(&kvm->lock);
4637 case KVM_CREATE_PIT:
4638 u.pit_config.flags = KVM_PIT_SPEAKER_DUMMY;
4640 case KVM_CREATE_PIT2:
4642 if (copy_from_user(&u.pit_config, argp,
4643 sizeof(struct kvm_pit_config)))
4646 mutex_lock(&kvm->lock);
4649 goto create_pit_unlock;
4651 kvm->arch.vpit = kvm_create_pit(kvm, u.pit_config.flags);
4655 mutex_unlock(&kvm->lock);
4657 case KVM_GET_IRQCHIP: {
4658 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4659 struct kvm_irqchip *chip;
4661 chip = memdup_user(argp, sizeof(*chip));
4668 if (!irqchip_kernel(kvm))
4669 goto get_irqchip_out;
4670 r = kvm_vm_ioctl_get_irqchip(kvm, chip);
4672 goto get_irqchip_out;
4674 if (copy_to_user(argp, chip, sizeof(*chip)))
4675 goto get_irqchip_out;
4681 case KVM_SET_IRQCHIP: {
4682 /* 0: PIC master, 1: PIC slave, 2: IOAPIC */
4683 struct kvm_irqchip *chip;
4685 chip = memdup_user(argp, sizeof(*chip));
4692 if (!irqchip_kernel(kvm))
4693 goto set_irqchip_out;
4694 r = kvm_vm_ioctl_set_irqchip(kvm, chip);
4696 goto set_irqchip_out;
4704 if (copy_from_user(&u.ps, argp, sizeof(struct kvm_pit_state)))
4707 if (!kvm->arch.vpit)
4709 r = kvm_vm_ioctl_get_pit(kvm, &u.ps);
4713 if (copy_to_user(argp, &u.ps, sizeof(struct kvm_pit_state)))
4720 if (copy_from_user(&u.ps, argp, sizeof(u.ps)))
4723 if (!kvm->arch.vpit)
4725 r = kvm_vm_ioctl_set_pit(kvm, &u.ps);
4728 case KVM_GET_PIT2: {
4730 if (!kvm->arch.vpit)
4732 r = kvm_vm_ioctl_get_pit2(kvm, &u.ps2);
4736 if (copy_to_user(argp, &u.ps2, sizeof(u.ps2)))
4741 case KVM_SET_PIT2: {
4743 if (copy_from_user(&u.ps2, argp, sizeof(u.ps2)))
4746 if (!kvm->arch.vpit)
4748 r = kvm_vm_ioctl_set_pit2(kvm, &u.ps2);
4751 case KVM_REINJECT_CONTROL: {
4752 struct kvm_reinject_control control;
4754 if (copy_from_user(&control, argp, sizeof(control)))
4756 r = kvm_vm_ioctl_reinject(kvm, &control);
4759 case KVM_SET_BOOT_CPU_ID:
4761 mutex_lock(&kvm->lock);
4762 if (kvm->created_vcpus)
4765 kvm->arch.bsp_vcpu_id = arg;
4766 mutex_unlock(&kvm->lock);
4768 case KVM_XEN_HVM_CONFIG: {
4769 struct kvm_xen_hvm_config xhc;
4771 if (copy_from_user(&xhc, argp, sizeof(xhc)))
4776 memcpy(&kvm->arch.xen_hvm_config, &xhc, sizeof(xhc));
4780 case KVM_SET_CLOCK: {
4781 struct kvm_clock_data user_ns;
4785 if (copy_from_user(&user_ns, argp, sizeof(user_ns)))
4794 * TODO: userspace has to take care of races with VCPU_RUN, so
4795 * kvm_gen_update_masterclock() can be cut down to locked
4796 * pvclock_update_vm_gtod_copy().
4798 kvm_gen_update_masterclock(kvm);
4799 now_ns = get_kvmclock_ns(kvm);
4800 kvm->arch.kvmclock_offset += user_ns.clock - now_ns;
4801 kvm_make_all_cpus_request(kvm, KVM_REQ_CLOCK_UPDATE);
4804 case KVM_GET_CLOCK: {
4805 struct kvm_clock_data user_ns;
4808 now_ns = get_kvmclock_ns(kvm);
4809 user_ns.clock = now_ns;
4810 user_ns.flags = kvm->arch.use_master_clock ? KVM_CLOCK_TSC_STABLE : 0;
4811 memset(&user_ns.pad, 0, sizeof(user_ns.pad));
4814 if (copy_to_user(argp, &user_ns, sizeof(user_ns)))
4819 case KVM_MEMORY_ENCRYPT_OP: {
4821 if (kvm_x86_ops->mem_enc_op)
4822 r = kvm_x86_ops->mem_enc_op(kvm, argp);
4825 case KVM_MEMORY_ENCRYPT_REG_REGION: {
4826 struct kvm_enc_region region;
4829 if (copy_from_user(®ion, argp, sizeof(region)))
4833 if (kvm_x86_ops->mem_enc_reg_region)
4834 r = kvm_x86_ops->mem_enc_reg_region(kvm, ®ion);
4837 case KVM_MEMORY_ENCRYPT_UNREG_REGION: {
4838 struct kvm_enc_region region;
4841 if (copy_from_user(®ion, argp, sizeof(region)))
4845 if (kvm_x86_ops->mem_enc_unreg_region)
4846 r = kvm_x86_ops->mem_enc_unreg_region(kvm, ®ion);
4849 case KVM_HYPERV_EVENTFD: {
4850 struct kvm_hyperv_eventfd hvevfd;
4853 if (copy_from_user(&hvevfd, argp, sizeof(hvevfd)))
4855 r = kvm_vm_ioctl_hv_eventfd(kvm, &hvevfd);
4865 static void kvm_init_msr_list(void)
4870 for (i = j = 0; i < ARRAY_SIZE(msrs_to_save); i++) {
4871 if (rdmsr_safe(msrs_to_save[i], &dummy[0], &dummy[1]) < 0)
4875 * Even MSRs that are valid in the host may not be exposed
4876 * to the guests in some cases.
4878 switch (msrs_to_save[i]) {
4879 case MSR_IA32_BNDCFGS:
4880 if (!kvm_mpx_supported())
4884 if (!kvm_x86_ops->rdtscp_supported())
4892 msrs_to_save[j] = msrs_to_save[i];
4895 num_msrs_to_save = j;
4897 for (i = j = 0; i < ARRAY_SIZE(emulated_msrs); i++) {
4898 if (!kvm_x86_ops->has_emulated_msr(emulated_msrs[i]))
4902 emulated_msrs[j] = emulated_msrs[i];
4905 num_emulated_msrs = j;
4907 for (i = j = 0; i < ARRAY_SIZE(msr_based_features); i++) {
4908 struct kvm_msr_entry msr;
4910 msr.index = msr_based_features[i];
4911 if (kvm_get_msr_feature(&msr))
4915 msr_based_features[j] = msr_based_features[i];
4918 num_msr_based_features = j;
4921 static int vcpu_mmio_write(struct kvm_vcpu *vcpu, gpa_t addr, int len,
4929 if (!(lapic_in_kernel(vcpu) &&
4930 !kvm_iodevice_write(vcpu, &vcpu->arch.apic->dev, addr, n, v))
4931 && kvm_io_bus_write(vcpu, KVM_MMIO_BUS, addr, n, v))
4942 static int vcpu_mmio_read(struct kvm_vcpu *vcpu, gpa_t addr, int len, void *v)
4949 if (!(lapic_in_kernel(vcpu) &&
4950 !kvm_iodevice_read(vcpu, &vcpu->arch.apic->dev,
4952 && kvm_io_bus_read(vcpu, KVM_MMIO_BUS, addr, n, v))
4954 trace_kvm_mmio(KVM_TRACE_MMIO_READ, n, addr, v);
4964 static void kvm_set_segment(struct kvm_vcpu *vcpu,
4965 struct kvm_segment *var, int seg)
4967 kvm_x86_ops->set_segment(vcpu, var, seg);
4970 void kvm_get_segment(struct kvm_vcpu *vcpu,
4971 struct kvm_segment *var, int seg)
4973 kvm_x86_ops->get_segment(vcpu, var, seg);
4976 gpa_t translate_nested_gpa(struct kvm_vcpu *vcpu, gpa_t gpa, u32 access,
4977 struct x86_exception *exception)
4981 BUG_ON(!mmu_is_nested(vcpu));
4983 /* NPT walks are always user-walks */
4984 access |= PFERR_USER_MASK;
4985 t_gpa = vcpu->arch.mmu->gva_to_gpa(vcpu, gpa, access, exception);
4990 gpa_t kvm_mmu_gva_to_gpa_read(struct kvm_vcpu *vcpu, gva_t gva,
4991 struct x86_exception *exception)
4993 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
4994 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
4997 gpa_t kvm_mmu_gva_to_gpa_fetch(struct kvm_vcpu *vcpu, gva_t gva,
4998 struct x86_exception *exception)
5000 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5001 access |= PFERR_FETCH_MASK;
5002 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5005 gpa_t kvm_mmu_gva_to_gpa_write(struct kvm_vcpu *vcpu, gva_t gva,
5006 struct x86_exception *exception)
5008 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5009 access |= PFERR_WRITE_MASK;
5010 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5013 /* uses this to access any guest's mapped memory without checking CPL */
5014 gpa_t kvm_mmu_gva_to_gpa_system(struct kvm_vcpu *vcpu, gva_t gva,
5015 struct x86_exception *exception)
5017 return vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, 0, exception);
5020 static int kvm_read_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5021 struct kvm_vcpu *vcpu, u32 access,
5022 struct x86_exception *exception)
5025 int r = X86EMUL_CONTINUE;
5028 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access,
5030 unsigned offset = addr & (PAGE_SIZE-1);
5031 unsigned toread = min(bytes, (unsigned)PAGE_SIZE - offset);
5034 if (gpa == UNMAPPED_GVA)
5035 return X86EMUL_PROPAGATE_FAULT;
5036 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, data,
5039 r = X86EMUL_IO_NEEDED;
5051 /* used for instruction fetching */
5052 static int kvm_fetch_guest_virt(struct x86_emulate_ctxt *ctxt,
5053 gva_t addr, void *val, unsigned int bytes,
5054 struct x86_exception *exception)
5056 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5057 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5061 /* Inline kvm_read_guest_virt_helper for speed. */
5062 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr, access|PFERR_FETCH_MASK,
5064 if (unlikely(gpa == UNMAPPED_GVA))
5065 return X86EMUL_PROPAGATE_FAULT;
5067 offset = addr & (PAGE_SIZE-1);
5068 if (WARN_ON(offset + bytes > PAGE_SIZE))
5069 bytes = (unsigned)PAGE_SIZE - offset;
5070 ret = kvm_vcpu_read_guest_page(vcpu, gpa >> PAGE_SHIFT, val,
5072 if (unlikely(ret < 0))
5073 return X86EMUL_IO_NEEDED;
5075 return X86EMUL_CONTINUE;
5078 int kvm_read_guest_virt(struct kvm_vcpu *vcpu,
5079 gva_t addr, void *val, unsigned int bytes,
5080 struct x86_exception *exception)
5082 u32 access = (kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0;
5084 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access,
5087 EXPORT_SYMBOL_GPL(kvm_read_guest_virt);
5089 static int emulator_read_std(struct x86_emulate_ctxt *ctxt,
5090 gva_t addr, void *val, unsigned int bytes,
5091 struct x86_exception *exception, bool system)
5093 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5096 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5097 access |= PFERR_USER_MASK;
5099 return kvm_read_guest_virt_helper(addr, val, bytes, vcpu, access, exception);
5102 static int kvm_read_guest_phys_system(struct x86_emulate_ctxt *ctxt,
5103 unsigned long addr, void *val, unsigned int bytes)
5105 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5106 int r = kvm_vcpu_read_guest(vcpu, addr, val, bytes);
5108 return r < 0 ? X86EMUL_IO_NEEDED : X86EMUL_CONTINUE;
5111 static int kvm_write_guest_virt_helper(gva_t addr, void *val, unsigned int bytes,
5112 struct kvm_vcpu *vcpu, u32 access,
5113 struct x86_exception *exception)
5116 int r = X86EMUL_CONTINUE;
5119 gpa_t gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, addr,
5122 unsigned offset = addr & (PAGE_SIZE-1);
5123 unsigned towrite = min(bytes, (unsigned)PAGE_SIZE - offset);
5126 if (gpa == UNMAPPED_GVA)
5127 return X86EMUL_PROPAGATE_FAULT;
5128 ret = kvm_vcpu_write_guest(vcpu, gpa, data, towrite);
5130 r = X86EMUL_IO_NEEDED;
5142 static int emulator_write_std(struct x86_emulate_ctxt *ctxt, gva_t addr, void *val,
5143 unsigned int bytes, struct x86_exception *exception,
5146 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5147 u32 access = PFERR_WRITE_MASK;
5149 if (!system && kvm_x86_ops->get_cpl(vcpu) == 3)
5150 access |= PFERR_USER_MASK;
5152 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5156 int kvm_write_guest_virt_system(struct kvm_vcpu *vcpu, gva_t addr, void *val,
5157 unsigned int bytes, struct x86_exception *exception)
5159 /* kvm_write_guest_virt_system can pull in tons of pages. */
5160 vcpu->arch.l1tf_flush_l1d = true;
5162 return kvm_write_guest_virt_helper(addr, val, bytes, vcpu,
5163 PFERR_WRITE_MASK, exception);
5165 EXPORT_SYMBOL_GPL(kvm_write_guest_virt_system);
5167 int handle_ud(struct kvm_vcpu *vcpu)
5169 int emul_type = EMULTYPE_TRAP_UD;
5170 enum emulation_result er;
5171 char sig[5]; /* ud2; .ascii "kvm" */
5172 struct x86_exception e;
5174 if (force_emulation_prefix &&
5175 kvm_read_guest_virt(vcpu, kvm_get_linear_rip(vcpu),
5176 sig, sizeof(sig), &e) == 0 &&
5177 memcmp(sig, "\xf\xbkvm", sizeof(sig)) == 0) {
5178 kvm_rip_write(vcpu, kvm_rip_read(vcpu) + sizeof(sig));
5182 er = kvm_emulate_instruction(vcpu, emul_type);
5183 if (er == EMULATE_USER_EXIT)
5185 if (er != EMULATE_DONE)
5186 kvm_queue_exception(vcpu, UD_VECTOR);
5189 EXPORT_SYMBOL_GPL(handle_ud);
5191 static int vcpu_is_mmio_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5192 gpa_t gpa, bool write)
5194 /* For APIC access vmexit */
5195 if ((gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5198 if (vcpu_match_mmio_gpa(vcpu, gpa)) {
5199 trace_vcpu_match_mmio(gva, gpa, write, true);
5206 static int vcpu_mmio_gva_to_gpa(struct kvm_vcpu *vcpu, unsigned long gva,
5207 gpa_t *gpa, struct x86_exception *exception,
5210 u32 access = ((kvm_x86_ops->get_cpl(vcpu) == 3) ? PFERR_USER_MASK : 0)
5211 | (write ? PFERR_WRITE_MASK : 0);
5214 * currently PKRU is only applied to ept enabled guest so
5215 * there is no pkey in EPT page table for L1 guest or EPT
5216 * shadow page table for L2 guest.
5218 if (vcpu_match_mmio_gva(vcpu, gva)
5219 && !permission_fault(vcpu, vcpu->arch.walk_mmu,
5220 vcpu->arch.access, 0, access)) {
5221 *gpa = vcpu->arch.mmio_gfn << PAGE_SHIFT |
5222 (gva & (PAGE_SIZE - 1));
5223 trace_vcpu_match_mmio(gva, *gpa, write, false);
5227 *gpa = vcpu->arch.walk_mmu->gva_to_gpa(vcpu, gva, access, exception);
5229 if (*gpa == UNMAPPED_GVA)
5232 return vcpu_is_mmio_gpa(vcpu, gva, *gpa, write);
5235 int emulator_write_phys(struct kvm_vcpu *vcpu, gpa_t gpa,
5236 const void *val, int bytes)
5240 ret = kvm_vcpu_write_guest(vcpu, gpa, val, bytes);
5243 kvm_page_track_write(vcpu, gpa, val, bytes);
5247 struct read_write_emulator_ops {
5248 int (*read_write_prepare)(struct kvm_vcpu *vcpu, void *val,
5250 int (*read_write_emulate)(struct kvm_vcpu *vcpu, gpa_t gpa,
5251 void *val, int bytes);
5252 int (*read_write_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5253 int bytes, void *val);
5254 int (*read_write_exit_mmio)(struct kvm_vcpu *vcpu, gpa_t gpa,
5255 void *val, int bytes);
5259 static int read_prepare(struct kvm_vcpu *vcpu, void *val, int bytes)
5261 if (vcpu->mmio_read_completed) {
5262 trace_kvm_mmio(KVM_TRACE_MMIO_READ, bytes,
5263 vcpu->mmio_fragments[0].gpa, val);
5264 vcpu->mmio_read_completed = 0;
5271 static int read_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5272 void *val, int bytes)
5274 return !kvm_vcpu_read_guest(vcpu, gpa, val, bytes);
5277 static int write_emulate(struct kvm_vcpu *vcpu, gpa_t gpa,
5278 void *val, int bytes)
5280 return emulator_write_phys(vcpu, gpa, val, bytes);
5283 static int write_mmio(struct kvm_vcpu *vcpu, gpa_t gpa, int bytes, void *val)
5285 trace_kvm_mmio(KVM_TRACE_MMIO_WRITE, bytes, gpa, val);
5286 return vcpu_mmio_write(vcpu, gpa, bytes, val);
5289 static int read_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5290 void *val, int bytes)
5292 trace_kvm_mmio(KVM_TRACE_MMIO_READ_UNSATISFIED, bytes, gpa, NULL);
5293 return X86EMUL_IO_NEEDED;
5296 static int write_exit_mmio(struct kvm_vcpu *vcpu, gpa_t gpa,
5297 void *val, int bytes)
5299 struct kvm_mmio_fragment *frag = &vcpu->mmio_fragments[0];
5301 memcpy(vcpu->run->mmio.data, frag->data, min(8u, frag->len));
5302 return X86EMUL_CONTINUE;
5305 static const struct read_write_emulator_ops read_emultor = {
5306 .read_write_prepare = read_prepare,
5307 .read_write_emulate = read_emulate,
5308 .read_write_mmio = vcpu_mmio_read,
5309 .read_write_exit_mmio = read_exit_mmio,
5312 static const struct read_write_emulator_ops write_emultor = {
5313 .read_write_emulate = write_emulate,
5314 .read_write_mmio = write_mmio,
5315 .read_write_exit_mmio = write_exit_mmio,
5319 static int emulator_read_write_onepage(unsigned long addr, void *val,
5321 struct x86_exception *exception,
5322 struct kvm_vcpu *vcpu,
5323 const struct read_write_emulator_ops *ops)
5327 bool write = ops->write;
5328 struct kvm_mmio_fragment *frag;
5329 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5332 * If the exit was due to a NPF we may already have a GPA.
5333 * If the GPA is present, use it to avoid the GVA to GPA table walk.
5334 * Note, this cannot be used on string operations since string
5335 * operation using rep will only have the initial GPA from the NPF
5338 if (vcpu->arch.gpa_available &&
5339 emulator_can_use_gpa(ctxt) &&
5340 (addr & ~PAGE_MASK) == (vcpu->arch.gpa_val & ~PAGE_MASK)) {
5341 gpa = vcpu->arch.gpa_val;
5342 ret = vcpu_is_mmio_gpa(vcpu, addr, gpa, write);
5344 ret = vcpu_mmio_gva_to_gpa(vcpu, addr, &gpa, exception, write);
5346 return X86EMUL_PROPAGATE_FAULT;
5349 if (!ret && ops->read_write_emulate(vcpu, gpa, val, bytes))
5350 return X86EMUL_CONTINUE;
5353 * Is this MMIO handled locally?
5355 handled = ops->read_write_mmio(vcpu, gpa, bytes, val);
5356 if (handled == bytes)
5357 return X86EMUL_CONTINUE;
5363 WARN_ON(vcpu->mmio_nr_fragments >= KVM_MAX_MMIO_FRAGMENTS);
5364 frag = &vcpu->mmio_fragments[vcpu->mmio_nr_fragments++];
5368 return X86EMUL_CONTINUE;
5371 static int emulator_read_write(struct x86_emulate_ctxt *ctxt,
5373 void *val, unsigned int bytes,
5374 struct x86_exception *exception,
5375 const struct read_write_emulator_ops *ops)
5377 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5381 if (ops->read_write_prepare &&
5382 ops->read_write_prepare(vcpu, val, bytes))
5383 return X86EMUL_CONTINUE;
5385 vcpu->mmio_nr_fragments = 0;
5387 /* Crossing a page boundary? */
5388 if (((addr + bytes - 1) ^ addr) & PAGE_MASK) {
5391 now = -addr & ~PAGE_MASK;
5392 rc = emulator_read_write_onepage(addr, val, now, exception,
5395 if (rc != X86EMUL_CONTINUE)
5398 if (ctxt->mode != X86EMUL_MODE_PROT64)
5404 rc = emulator_read_write_onepage(addr, val, bytes, exception,
5406 if (rc != X86EMUL_CONTINUE)
5409 if (!vcpu->mmio_nr_fragments)
5412 gpa = vcpu->mmio_fragments[0].gpa;
5414 vcpu->mmio_needed = 1;
5415 vcpu->mmio_cur_fragment = 0;
5417 vcpu->run->mmio.len = min(8u, vcpu->mmio_fragments[0].len);
5418 vcpu->run->mmio.is_write = vcpu->mmio_is_write = ops->write;
5419 vcpu->run->exit_reason = KVM_EXIT_MMIO;
5420 vcpu->run->mmio.phys_addr = gpa;
5422 return ops->read_write_exit_mmio(vcpu, gpa, val, bytes);
5425 static int emulator_read_emulated(struct x86_emulate_ctxt *ctxt,
5429 struct x86_exception *exception)
5431 return emulator_read_write(ctxt, addr, val, bytes,
5432 exception, &read_emultor);
5435 static int emulator_write_emulated(struct x86_emulate_ctxt *ctxt,
5439 struct x86_exception *exception)
5441 return emulator_read_write(ctxt, addr, (void *)val, bytes,
5442 exception, &write_emultor);
5445 #define CMPXCHG_TYPE(t, ptr, old, new) \
5446 (cmpxchg((t *)(ptr), *(t *)(old), *(t *)(new)) == *(t *)(old))
5448 #ifdef CONFIG_X86_64
5449 # define CMPXCHG64(ptr, old, new) CMPXCHG_TYPE(u64, ptr, old, new)
5451 # define CMPXCHG64(ptr, old, new) \
5452 (cmpxchg64((u64 *)(ptr), *(u64 *)(old), *(u64 *)(new)) == *(u64 *)(old))
5455 static int emulator_cmpxchg_emulated(struct x86_emulate_ctxt *ctxt,
5460 struct x86_exception *exception)
5462 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5468 /* guests cmpxchg8b have to be emulated atomically */
5469 if (bytes > 8 || (bytes & (bytes - 1)))
5472 gpa = kvm_mmu_gva_to_gpa_write(vcpu, addr, NULL);
5474 if (gpa == UNMAPPED_GVA ||
5475 (gpa & PAGE_MASK) == APIC_DEFAULT_PHYS_BASE)
5478 if (((gpa + bytes - 1) & PAGE_MASK) != (gpa & PAGE_MASK))
5481 page = kvm_vcpu_gfn_to_page(vcpu, gpa >> PAGE_SHIFT);
5482 if (is_error_page(page))
5485 kaddr = kmap_atomic(page);
5486 kaddr += offset_in_page(gpa);
5489 exchanged = CMPXCHG_TYPE(u8, kaddr, old, new);
5492 exchanged = CMPXCHG_TYPE(u16, kaddr, old, new);
5495 exchanged = CMPXCHG_TYPE(u32, kaddr, old, new);
5498 exchanged = CMPXCHG64(kaddr, old, new);
5503 kunmap_atomic(kaddr);
5504 kvm_release_page_dirty(page);
5507 return X86EMUL_CMPXCHG_FAILED;
5509 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
5510 kvm_page_track_write(vcpu, gpa, new, bytes);
5512 return X86EMUL_CONTINUE;
5515 printk_once(KERN_WARNING "kvm: emulating exchange as write\n");
5517 return emulator_write_emulated(ctxt, addr, new, bytes, exception);
5520 static int kernel_pio(struct kvm_vcpu *vcpu, void *pd)
5524 for (i = 0; i < vcpu->arch.pio.count; i++) {
5525 if (vcpu->arch.pio.in)
5526 r = kvm_io_bus_read(vcpu, KVM_PIO_BUS, vcpu->arch.pio.port,
5527 vcpu->arch.pio.size, pd);
5529 r = kvm_io_bus_write(vcpu, KVM_PIO_BUS,
5530 vcpu->arch.pio.port, vcpu->arch.pio.size,
5534 pd += vcpu->arch.pio.size;
5539 static int emulator_pio_in_out(struct kvm_vcpu *vcpu, int size,
5540 unsigned short port, void *val,
5541 unsigned int count, bool in)
5543 vcpu->arch.pio.port = port;
5544 vcpu->arch.pio.in = in;
5545 vcpu->arch.pio.count = count;
5546 vcpu->arch.pio.size = size;
5548 if (!kernel_pio(vcpu, vcpu->arch.pio_data)) {
5549 vcpu->arch.pio.count = 0;
5553 vcpu->run->exit_reason = KVM_EXIT_IO;
5554 vcpu->run->io.direction = in ? KVM_EXIT_IO_IN : KVM_EXIT_IO_OUT;
5555 vcpu->run->io.size = size;
5556 vcpu->run->io.data_offset = KVM_PIO_PAGE_OFFSET * PAGE_SIZE;
5557 vcpu->run->io.count = count;
5558 vcpu->run->io.port = port;
5563 static int emulator_pio_in_emulated(struct x86_emulate_ctxt *ctxt,
5564 int size, unsigned short port, void *val,
5567 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5570 if (vcpu->arch.pio.count)
5573 memset(vcpu->arch.pio_data, 0, size * count);
5575 ret = emulator_pio_in_out(vcpu, size, port, val, count, true);
5578 memcpy(val, vcpu->arch.pio_data, size * count);
5579 trace_kvm_pio(KVM_PIO_IN, port, size, count, vcpu->arch.pio_data);
5580 vcpu->arch.pio.count = 0;
5587 static int emulator_pio_out_emulated(struct x86_emulate_ctxt *ctxt,
5588 int size, unsigned short port,
5589 const void *val, unsigned int count)
5591 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5593 memcpy(vcpu->arch.pio_data, val, size * count);
5594 trace_kvm_pio(KVM_PIO_OUT, port, size, count, vcpu->arch.pio_data);
5595 return emulator_pio_in_out(vcpu, size, port, (void *)val, count, false);
5598 static unsigned long get_segment_base(struct kvm_vcpu *vcpu, int seg)
5600 return kvm_x86_ops->get_segment_base(vcpu, seg);
5603 static void emulator_invlpg(struct x86_emulate_ctxt *ctxt, ulong address)
5605 kvm_mmu_invlpg(emul_to_vcpu(ctxt), address);
5608 static int kvm_emulate_wbinvd_noskip(struct kvm_vcpu *vcpu)
5610 if (!need_emulate_wbinvd(vcpu))
5611 return X86EMUL_CONTINUE;
5613 if (kvm_x86_ops->has_wbinvd_exit()) {
5614 int cpu = get_cpu();
5616 cpumask_set_cpu(cpu, vcpu->arch.wbinvd_dirty_mask);
5617 smp_call_function_many(vcpu->arch.wbinvd_dirty_mask,
5618 wbinvd_ipi, NULL, 1);
5620 cpumask_clear(vcpu->arch.wbinvd_dirty_mask);
5623 return X86EMUL_CONTINUE;
5626 int kvm_emulate_wbinvd(struct kvm_vcpu *vcpu)
5628 kvm_emulate_wbinvd_noskip(vcpu);
5629 return kvm_skip_emulated_instruction(vcpu);
5631 EXPORT_SYMBOL_GPL(kvm_emulate_wbinvd);
5635 static void emulator_wbinvd(struct x86_emulate_ctxt *ctxt)
5637 kvm_emulate_wbinvd_noskip(emul_to_vcpu(ctxt));
5640 static int emulator_get_dr(struct x86_emulate_ctxt *ctxt, int dr,
5641 unsigned long *dest)
5643 return kvm_get_dr(emul_to_vcpu(ctxt), dr, dest);
5646 static int emulator_set_dr(struct x86_emulate_ctxt *ctxt, int dr,
5647 unsigned long value)
5650 return __kvm_set_dr(emul_to_vcpu(ctxt), dr, value);
5653 static u64 mk_cr_64(u64 curr_cr, u32 new_val)
5655 return (curr_cr & ~((1ULL << 32) - 1)) | new_val;
5658 static unsigned long emulator_get_cr(struct x86_emulate_ctxt *ctxt, int cr)
5660 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5661 unsigned long value;
5665 value = kvm_read_cr0(vcpu);
5668 value = vcpu->arch.cr2;
5671 value = kvm_read_cr3(vcpu);
5674 value = kvm_read_cr4(vcpu);
5677 value = kvm_get_cr8(vcpu);
5680 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5687 static int emulator_set_cr(struct x86_emulate_ctxt *ctxt, int cr, ulong val)
5689 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5694 res = kvm_set_cr0(vcpu, mk_cr_64(kvm_read_cr0(vcpu), val));
5697 vcpu->arch.cr2 = val;
5700 res = kvm_set_cr3(vcpu, val);
5703 res = kvm_set_cr4(vcpu, mk_cr_64(kvm_read_cr4(vcpu), val));
5706 res = kvm_set_cr8(vcpu, val);
5709 kvm_err("%s: unexpected cr %u\n", __func__, cr);
5716 static int emulator_get_cpl(struct x86_emulate_ctxt *ctxt)
5718 return kvm_x86_ops->get_cpl(emul_to_vcpu(ctxt));
5721 static void emulator_get_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5723 kvm_x86_ops->get_gdt(emul_to_vcpu(ctxt), dt);
5726 static void emulator_get_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5728 kvm_x86_ops->get_idt(emul_to_vcpu(ctxt), dt);
5731 static void emulator_set_gdt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5733 kvm_x86_ops->set_gdt(emul_to_vcpu(ctxt), dt);
5736 static void emulator_set_idt(struct x86_emulate_ctxt *ctxt, struct desc_ptr *dt)
5738 kvm_x86_ops->set_idt(emul_to_vcpu(ctxt), dt);
5741 static unsigned long emulator_get_cached_segment_base(
5742 struct x86_emulate_ctxt *ctxt, int seg)
5744 return get_segment_base(emul_to_vcpu(ctxt), seg);
5747 static bool emulator_get_segment(struct x86_emulate_ctxt *ctxt, u16 *selector,
5748 struct desc_struct *desc, u32 *base3,
5751 struct kvm_segment var;
5753 kvm_get_segment(emul_to_vcpu(ctxt), &var, seg);
5754 *selector = var.selector;
5757 memset(desc, 0, sizeof(*desc));
5765 set_desc_limit(desc, var.limit);
5766 set_desc_base(desc, (unsigned long)var.base);
5767 #ifdef CONFIG_X86_64
5769 *base3 = var.base >> 32;
5771 desc->type = var.type;
5773 desc->dpl = var.dpl;
5774 desc->p = var.present;
5775 desc->avl = var.avl;
5783 static void emulator_set_segment(struct x86_emulate_ctxt *ctxt, u16 selector,
5784 struct desc_struct *desc, u32 base3,
5787 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5788 struct kvm_segment var;
5790 var.selector = selector;
5791 var.base = get_desc_base(desc);
5792 #ifdef CONFIG_X86_64
5793 var.base |= ((u64)base3) << 32;
5795 var.limit = get_desc_limit(desc);
5797 var.limit = (var.limit << 12) | 0xfff;
5798 var.type = desc->type;
5799 var.dpl = desc->dpl;
5804 var.avl = desc->avl;
5805 var.present = desc->p;
5806 var.unusable = !var.present;
5809 kvm_set_segment(vcpu, &var, seg);
5813 static int emulator_get_msr(struct x86_emulate_ctxt *ctxt,
5814 u32 msr_index, u64 *pdata)
5816 struct msr_data msr;
5819 msr.index = msr_index;
5820 msr.host_initiated = false;
5821 r = kvm_get_msr(emul_to_vcpu(ctxt), &msr);
5829 static int emulator_set_msr(struct x86_emulate_ctxt *ctxt,
5830 u32 msr_index, u64 data)
5832 struct msr_data msr;
5835 msr.index = msr_index;
5836 msr.host_initiated = false;
5837 return kvm_set_msr(emul_to_vcpu(ctxt), &msr);
5840 static u64 emulator_get_smbase(struct x86_emulate_ctxt *ctxt)
5842 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5844 return vcpu->arch.smbase;
5847 static void emulator_set_smbase(struct x86_emulate_ctxt *ctxt, u64 smbase)
5849 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
5851 vcpu->arch.smbase = smbase;
5854 static int emulator_check_pmc(struct x86_emulate_ctxt *ctxt,
5857 return kvm_pmu_is_valid_msr_idx(emul_to_vcpu(ctxt), pmc);
5860 static int emulator_read_pmc(struct x86_emulate_ctxt *ctxt,
5861 u32 pmc, u64 *pdata)
5863 return kvm_pmu_rdpmc(emul_to_vcpu(ctxt), pmc, pdata);
5866 static void emulator_halt(struct x86_emulate_ctxt *ctxt)
5868 emul_to_vcpu(ctxt)->arch.halt_request = 1;
5871 static int emulator_intercept(struct x86_emulate_ctxt *ctxt,
5872 struct x86_instruction_info *info,
5873 enum x86_intercept_stage stage)
5875 return kvm_x86_ops->check_intercept(emul_to_vcpu(ctxt), info, stage);
5878 static bool emulator_get_cpuid(struct x86_emulate_ctxt *ctxt,
5879 u32 *eax, u32 *ebx, u32 *ecx, u32 *edx, bool check_limit)
5881 return kvm_cpuid(emul_to_vcpu(ctxt), eax, ebx, ecx, edx, check_limit);
5884 static ulong emulator_read_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg)
5886 return kvm_register_read(emul_to_vcpu(ctxt), reg);
5889 static void emulator_write_gpr(struct x86_emulate_ctxt *ctxt, unsigned reg, ulong val)
5891 kvm_register_write(emul_to_vcpu(ctxt), reg, val);
5894 static void emulator_set_nmi_mask(struct x86_emulate_ctxt *ctxt, bool masked)
5896 kvm_x86_ops->set_nmi_mask(emul_to_vcpu(ctxt), masked);
5899 static unsigned emulator_get_hflags(struct x86_emulate_ctxt *ctxt)
5901 return emul_to_vcpu(ctxt)->arch.hflags;
5904 static void emulator_set_hflags(struct x86_emulate_ctxt *ctxt, unsigned emul_flags)
5906 kvm_set_hflags(emul_to_vcpu(ctxt), emul_flags);
5909 static int emulator_pre_leave_smm(struct x86_emulate_ctxt *ctxt, u64 smbase)
5911 return kvm_x86_ops->pre_leave_smm(emul_to_vcpu(ctxt), smbase);
5914 static const struct x86_emulate_ops emulate_ops = {
5915 .read_gpr = emulator_read_gpr,
5916 .write_gpr = emulator_write_gpr,
5917 .read_std = emulator_read_std,
5918 .write_std = emulator_write_std,
5919 .read_phys = kvm_read_guest_phys_system,
5920 .fetch = kvm_fetch_guest_virt,
5921 .read_emulated = emulator_read_emulated,
5922 .write_emulated = emulator_write_emulated,
5923 .cmpxchg_emulated = emulator_cmpxchg_emulated,
5924 .invlpg = emulator_invlpg,
5925 .pio_in_emulated = emulator_pio_in_emulated,
5926 .pio_out_emulated = emulator_pio_out_emulated,
5927 .get_segment = emulator_get_segment,
5928 .set_segment = emulator_set_segment,
5929 .get_cached_segment_base = emulator_get_cached_segment_base,
5930 .get_gdt = emulator_get_gdt,
5931 .get_idt = emulator_get_idt,
5932 .set_gdt = emulator_set_gdt,
5933 .set_idt = emulator_set_idt,
5934 .get_cr = emulator_get_cr,
5935 .set_cr = emulator_set_cr,
5936 .cpl = emulator_get_cpl,
5937 .get_dr = emulator_get_dr,
5938 .set_dr = emulator_set_dr,
5939 .get_smbase = emulator_get_smbase,
5940 .set_smbase = emulator_set_smbase,
5941 .set_msr = emulator_set_msr,
5942 .get_msr = emulator_get_msr,
5943 .check_pmc = emulator_check_pmc,
5944 .read_pmc = emulator_read_pmc,
5945 .halt = emulator_halt,
5946 .wbinvd = emulator_wbinvd,
5947 .fix_hypercall = emulator_fix_hypercall,
5948 .intercept = emulator_intercept,
5949 .get_cpuid = emulator_get_cpuid,
5950 .set_nmi_mask = emulator_set_nmi_mask,
5951 .get_hflags = emulator_get_hflags,
5952 .set_hflags = emulator_set_hflags,
5953 .pre_leave_smm = emulator_pre_leave_smm,
5956 static void toggle_interruptibility(struct kvm_vcpu *vcpu, u32 mask)
5958 u32 int_shadow = kvm_x86_ops->get_interrupt_shadow(vcpu);
5960 * an sti; sti; sequence only disable interrupts for the first
5961 * instruction. So, if the last instruction, be it emulated or
5962 * not, left the system with the INT_STI flag enabled, it
5963 * means that the last instruction is an sti. We should not
5964 * leave the flag on in this case. The same goes for mov ss
5966 if (int_shadow & mask)
5968 if (unlikely(int_shadow || mask)) {
5969 kvm_x86_ops->set_interrupt_shadow(vcpu, mask);
5971 kvm_make_request(KVM_REQ_EVENT, vcpu);
5975 static bool inject_emulated_exception(struct kvm_vcpu *vcpu)
5977 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5978 if (ctxt->exception.vector == PF_VECTOR)
5979 return kvm_propagate_fault(vcpu, &ctxt->exception);
5981 if (ctxt->exception.error_code_valid)
5982 kvm_queue_exception_e(vcpu, ctxt->exception.vector,
5983 ctxt->exception.error_code);
5985 kvm_queue_exception(vcpu, ctxt->exception.vector);
5989 static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
5991 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
5994 kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
5996 ctxt->eflags = kvm_get_rflags(vcpu);
5997 ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
5999 ctxt->eip = kvm_rip_read(vcpu);
6000 ctxt->mode = (!is_protmode(vcpu)) ? X86EMUL_MODE_REAL :
6001 (ctxt->eflags & X86_EFLAGS_VM) ? X86EMUL_MODE_VM86 :
6002 (cs_l && is_long_mode(vcpu)) ? X86EMUL_MODE_PROT64 :
6003 cs_db ? X86EMUL_MODE_PROT32 :
6004 X86EMUL_MODE_PROT16;
6005 BUILD_BUG_ON(HF_GUEST_MASK != X86EMUL_GUEST_MASK);
6006 BUILD_BUG_ON(HF_SMM_MASK != X86EMUL_SMM_MASK);
6007 BUILD_BUG_ON(HF_SMM_INSIDE_NMI_MASK != X86EMUL_SMM_INSIDE_NMI_MASK);
6009 init_decode_cache(ctxt);
6010 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6013 int kvm_inject_realmode_interrupt(struct kvm_vcpu *vcpu, int irq, int inc_eip)
6015 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6018 init_emulate_ctxt(vcpu);
6022 ctxt->_eip = ctxt->eip + inc_eip;
6023 ret = emulate_int_real(ctxt, irq);
6025 if (ret != X86EMUL_CONTINUE)
6026 return EMULATE_FAIL;
6028 ctxt->eip = ctxt->_eip;
6029 kvm_rip_write(vcpu, ctxt->eip);
6030 kvm_set_rflags(vcpu, ctxt->eflags);
6032 return EMULATE_DONE;
6034 EXPORT_SYMBOL_GPL(kvm_inject_realmode_interrupt);
6036 static int handle_emulation_failure(struct kvm_vcpu *vcpu, int emulation_type)
6038 int r = EMULATE_DONE;
6040 ++vcpu->stat.insn_emulation_fail;
6041 trace_kvm_emulate_insn_failed(vcpu);
6043 if (emulation_type & EMULTYPE_NO_UD_ON_FAIL)
6044 return EMULATE_FAIL;
6046 if (!is_guest_mode(vcpu) && kvm_x86_ops->get_cpl(vcpu) == 0) {
6047 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6048 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6049 vcpu->run->internal.ndata = 0;
6050 r = EMULATE_USER_EXIT;
6053 kvm_queue_exception(vcpu, UD_VECTOR);
6058 static bool reexecute_instruction(struct kvm_vcpu *vcpu, gva_t cr2,
6059 bool write_fault_to_shadow_pgtable,
6065 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6068 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6071 if (!vcpu->arch.mmu->direct_map) {
6073 * Write permission should be allowed since only
6074 * write access need to be emulated.
6076 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6079 * If the mapping is invalid in guest, let cpu retry
6080 * it to generate fault.
6082 if (gpa == UNMAPPED_GVA)
6087 * Do not retry the unhandleable instruction if it faults on the
6088 * readonly host memory, otherwise it will goto a infinite loop:
6089 * retry instruction -> write #PF -> emulation fail -> retry
6090 * instruction -> ...
6092 pfn = gfn_to_pfn(vcpu->kvm, gpa_to_gfn(gpa));
6095 * If the instruction failed on the error pfn, it can not be fixed,
6096 * report the error to userspace.
6098 if (is_error_noslot_pfn(pfn))
6101 kvm_release_pfn_clean(pfn);
6103 /* The instructions are well-emulated on direct mmu. */
6104 if (vcpu->arch.mmu->direct_map) {
6105 unsigned int indirect_shadow_pages;
6107 spin_lock(&vcpu->kvm->mmu_lock);
6108 indirect_shadow_pages = vcpu->kvm->arch.indirect_shadow_pages;
6109 spin_unlock(&vcpu->kvm->mmu_lock);
6111 if (indirect_shadow_pages)
6112 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6118 * if emulation was due to access to shadowed page table
6119 * and it failed try to unshadow page and re-enter the
6120 * guest to let CPU execute the instruction.
6122 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6125 * If the access faults on its page table, it can not
6126 * be fixed by unprotecting shadow page and it should
6127 * be reported to userspace.
6129 return !write_fault_to_shadow_pgtable;
6132 static bool retry_instruction(struct x86_emulate_ctxt *ctxt,
6133 unsigned long cr2, int emulation_type)
6135 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
6136 unsigned long last_retry_eip, last_retry_addr, gpa = cr2;
6138 last_retry_eip = vcpu->arch.last_retry_eip;
6139 last_retry_addr = vcpu->arch.last_retry_addr;
6142 * If the emulation is caused by #PF and it is non-page_table
6143 * writing instruction, it means the VM-EXIT is caused by shadow
6144 * page protected, we can zap the shadow page and retry this
6145 * instruction directly.
6147 * Note: if the guest uses a non-page-table modifying instruction
6148 * on the PDE that points to the instruction, then we will unmap
6149 * the instruction and go to an infinite loop. So, we cache the
6150 * last retried eip and the last fault address, if we meet the eip
6151 * and the address again, we can break out of the potential infinite
6154 vcpu->arch.last_retry_eip = vcpu->arch.last_retry_addr = 0;
6156 if (!(emulation_type & EMULTYPE_ALLOW_RETRY))
6159 if (WARN_ON_ONCE(is_guest_mode(vcpu)))
6162 if (x86_page_table_writing_insn(ctxt))
6165 if (ctxt->eip == last_retry_eip && last_retry_addr == cr2)
6168 vcpu->arch.last_retry_eip = ctxt->eip;
6169 vcpu->arch.last_retry_addr = cr2;
6171 if (!vcpu->arch.mmu->direct_map)
6172 gpa = kvm_mmu_gva_to_gpa_write(vcpu, cr2, NULL);
6174 kvm_mmu_unprotect_page(vcpu->kvm, gpa_to_gfn(gpa));
6179 static int complete_emulated_mmio(struct kvm_vcpu *vcpu);
6180 static int complete_emulated_pio(struct kvm_vcpu *vcpu);
6182 static void kvm_smm_changed(struct kvm_vcpu *vcpu)
6184 if (!(vcpu->arch.hflags & HF_SMM_MASK)) {
6185 /* This is a good place to trace that we are exiting SMM. */
6186 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, false);
6188 /* Process a latched INIT or SMI, if any. */
6189 kvm_make_request(KVM_REQ_EVENT, vcpu);
6192 kvm_mmu_reset_context(vcpu);
6195 static void kvm_set_hflags(struct kvm_vcpu *vcpu, unsigned emul_flags)
6197 unsigned changed = vcpu->arch.hflags ^ emul_flags;
6199 vcpu->arch.hflags = emul_flags;
6201 if (changed & HF_SMM_MASK)
6202 kvm_smm_changed(vcpu);
6205 static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
6214 for (i = 0; i < 4; i++, enable >>= 2, rwlen >>= 4)
6215 if ((enable & 3) && (rwlen & 15) == type && db[i] == addr)
6220 static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
6222 struct kvm_run *kvm_run = vcpu->run;
6224 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
6225 kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
6226 kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
6227 kvm_run->debug.arch.exception = DB_VECTOR;
6228 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6229 *r = EMULATE_USER_EXIT;
6231 kvm_queue_exception_p(vcpu, DB_VECTOR, DR6_BS);
6235 int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
6237 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6238 int r = EMULATE_DONE;
6240 kvm_x86_ops->skip_emulated_instruction(vcpu);
6243 * rflags is the old, "raw" value of the flags. The new value has
6244 * not been saved yet.
6246 * This is correct even for TF set by the guest, because "the
6247 * processor will not generate this exception after the instruction
6248 * that sets the TF flag".
6250 if (unlikely(rflags & X86_EFLAGS_TF))
6251 kvm_vcpu_do_singlestep(vcpu, &r);
6252 return r == EMULATE_DONE;
6254 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
6256 static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
6258 if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
6259 (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
6260 struct kvm_run *kvm_run = vcpu->run;
6261 unsigned long eip = kvm_get_linear_rip(vcpu);
6262 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6263 vcpu->arch.guest_debug_dr7,
6267 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1 | DR6_RTM;
6268 kvm_run->debug.arch.pc = eip;
6269 kvm_run->debug.arch.exception = DB_VECTOR;
6270 kvm_run->exit_reason = KVM_EXIT_DEBUG;
6271 *r = EMULATE_USER_EXIT;
6276 if (unlikely(vcpu->arch.dr7 & DR7_BP_EN_MASK) &&
6277 !(kvm_get_rflags(vcpu) & X86_EFLAGS_RF)) {
6278 unsigned long eip = kvm_get_linear_rip(vcpu);
6279 u32 dr6 = kvm_vcpu_check_hw_bp(eip, 0,
6284 vcpu->arch.dr6 &= ~15;
6285 vcpu->arch.dr6 |= dr6 | DR6_RTM;
6286 kvm_queue_exception(vcpu, DB_VECTOR);
6295 static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
6297 switch (ctxt->opcode_len) {
6304 case 0xe6: /* OUT */
6308 case 0x6c: /* INS */
6310 case 0x6e: /* OUTS */
6317 case 0x33: /* RDPMC */
6326 int x86_emulate_instruction(struct kvm_vcpu *vcpu,
6333 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
6334 bool writeback = true;
6335 bool write_fault_to_spt = vcpu->arch.write_fault_to_shadow_pgtable;
6337 vcpu->arch.l1tf_flush_l1d = true;
6340 * Clear write_fault_to_shadow_pgtable here to ensure it is
6343 vcpu->arch.write_fault_to_shadow_pgtable = false;
6344 kvm_clear_exception_queue(vcpu);
6346 if (!(emulation_type & EMULTYPE_NO_DECODE)) {
6347 init_emulate_ctxt(vcpu);
6350 * We will reenter on the same instruction since
6351 * we do not set complete_userspace_io. This does not
6352 * handle watchpoints yet, those would be handled in
6355 if (!(emulation_type & EMULTYPE_SKIP) &&
6356 kvm_vcpu_check_breakpoint(vcpu, &r))
6359 ctxt->interruptibility = 0;
6360 ctxt->have_exception = false;
6361 ctxt->exception.vector = -1;
6362 ctxt->perm_ok = false;
6364 ctxt->ud = emulation_type & EMULTYPE_TRAP_UD;
6366 r = x86_decode_insn(ctxt, insn, insn_len);
6368 trace_kvm_emulate_insn_start(vcpu);
6369 ++vcpu->stat.insn_emulation;
6370 if (r != EMULATION_OK) {
6371 if (emulation_type & EMULTYPE_TRAP_UD)
6372 return EMULATE_FAIL;
6373 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6375 return EMULATE_DONE;
6376 if (ctxt->have_exception && inject_emulated_exception(vcpu))
6377 return EMULATE_DONE;
6378 if (emulation_type & EMULTYPE_SKIP)
6379 return EMULATE_FAIL;
6380 return handle_emulation_failure(vcpu, emulation_type);
6384 if ((emulation_type & EMULTYPE_VMWARE) &&
6385 !is_vmware_backdoor_opcode(ctxt))
6386 return EMULATE_FAIL;
6388 if (emulation_type & EMULTYPE_SKIP) {
6389 kvm_rip_write(vcpu, ctxt->_eip);
6390 if (ctxt->eflags & X86_EFLAGS_RF)
6391 kvm_set_rflags(vcpu, ctxt->eflags & ~X86_EFLAGS_RF);
6392 return EMULATE_DONE;
6395 if (retry_instruction(ctxt, cr2, emulation_type))
6396 return EMULATE_DONE;
6398 /* this is needed for vmware backdoor interface to work since it
6399 changes registers values during IO operation */
6400 if (vcpu->arch.emulate_regs_need_sync_from_vcpu) {
6401 vcpu->arch.emulate_regs_need_sync_from_vcpu = false;
6402 emulator_invalidate_register_cache(ctxt);
6406 /* Save the faulting GPA (cr2) in the address field */
6407 ctxt->exception.address = cr2;
6409 r = x86_emulate_insn(ctxt);
6411 if (r == EMULATION_INTERCEPTED)
6412 return EMULATE_DONE;
6414 if (r == EMULATION_FAILED) {
6415 if (reexecute_instruction(vcpu, cr2, write_fault_to_spt,
6417 return EMULATE_DONE;
6419 return handle_emulation_failure(vcpu, emulation_type);
6422 if (ctxt->have_exception) {
6424 if (inject_emulated_exception(vcpu))
6426 } else if (vcpu->arch.pio.count) {
6427 if (!vcpu->arch.pio.in) {
6428 /* FIXME: return into emulator if single-stepping. */
6429 vcpu->arch.pio.count = 0;
6432 vcpu->arch.complete_userspace_io = complete_emulated_pio;
6434 r = EMULATE_USER_EXIT;
6435 } else if (vcpu->mmio_needed) {
6436 if (!vcpu->mmio_is_write)
6438 r = EMULATE_USER_EXIT;
6439 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
6440 } else if (r == EMULATION_RESTART)
6446 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
6447 toggle_interruptibility(vcpu, ctxt->interruptibility);
6448 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
6449 kvm_rip_write(vcpu, ctxt->eip);
6450 if (r == EMULATE_DONE &&
6451 (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
6452 kvm_vcpu_do_singlestep(vcpu, &r);
6453 if (!ctxt->have_exception ||
6454 exception_type(ctxt->exception.vector) == EXCPT_TRAP)
6455 __kvm_set_rflags(vcpu, ctxt->eflags);
6458 * For STI, interrupts are shadowed; so KVM_REQ_EVENT will
6459 * do nothing, and it will be requested again as soon as
6460 * the shadow expires. But we still need to check here,
6461 * because POPF has no interrupt shadow.
6463 if (unlikely((ctxt->eflags & ~rflags) & X86_EFLAGS_IF))
6464 kvm_make_request(KVM_REQ_EVENT, vcpu);
6466 vcpu->arch.emulate_regs_need_sync_to_vcpu = true;
6471 int kvm_emulate_instruction(struct kvm_vcpu *vcpu, int emulation_type)
6473 return x86_emulate_instruction(vcpu, 0, emulation_type, NULL, 0);
6475 EXPORT_SYMBOL_GPL(kvm_emulate_instruction);
6477 int kvm_emulate_instruction_from_buffer(struct kvm_vcpu *vcpu,
6478 void *insn, int insn_len)
6480 return x86_emulate_instruction(vcpu, 0, 0, insn, insn_len);
6482 EXPORT_SYMBOL_GPL(kvm_emulate_instruction_from_buffer);
6484 static int kvm_fast_pio_out(struct kvm_vcpu *vcpu, int size,
6485 unsigned short port)
6487 unsigned long val = kvm_register_read(vcpu, VCPU_REGS_RAX);
6488 int ret = emulator_pio_out_emulated(&vcpu->arch.emulate_ctxt,
6489 size, port, &val, 1);
6490 /* do not return to emulator after return from userspace */
6491 vcpu->arch.pio.count = 0;
6495 static int complete_fast_pio_in(struct kvm_vcpu *vcpu)
6499 /* We should only ever be called with arch.pio.count equal to 1 */
6500 BUG_ON(vcpu->arch.pio.count != 1);
6502 /* For size less than 4 we merge, else we zero extend */
6503 val = (vcpu->arch.pio.size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX)
6507 * Since vcpu->arch.pio.count == 1 let emulator_pio_in_emulated perform
6508 * the copy and tracing
6510 emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, vcpu->arch.pio.size,
6511 vcpu->arch.pio.port, &val, 1);
6512 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6517 static int kvm_fast_pio_in(struct kvm_vcpu *vcpu, int size,
6518 unsigned short port)
6523 /* For size less than 4 we merge, else we zero extend */
6524 val = (size < 4) ? kvm_register_read(vcpu, VCPU_REGS_RAX) : 0;
6526 ret = emulator_pio_in_emulated(&vcpu->arch.emulate_ctxt, size, port,
6529 kvm_register_write(vcpu, VCPU_REGS_RAX, val);
6533 vcpu->arch.complete_userspace_io = complete_fast_pio_in;
6538 int kvm_fast_pio(struct kvm_vcpu *vcpu, int size, unsigned short port, int in)
6540 int ret = kvm_skip_emulated_instruction(vcpu);
6543 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6544 * KVM_EXIT_DEBUG here.
6547 return kvm_fast_pio_in(vcpu, size, port) && ret;
6549 return kvm_fast_pio_out(vcpu, size, port) && ret;
6551 EXPORT_SYMBOL_GPL(kvm_fast_pio);
6553 static int kvmclock_cpu_down_prep(unsigned int cpu)
6555 __this_cpu_write(cpu_tsc_khz, 0);
6559 static void tsc_khz_changed(void *data)
6561 struct cpufreq_freqs *freq = data;
6562 unsigned long khz = 0;
6566 else if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6567 khz = cpufreq_quick_get(raw_smp_processor_id());
6570 __this_cpu_write(cpu_tsc_khz, khz);
6573 #ifdef CONFIG_X86_64
6574 static void kvm_hyperv_tsc_notifier(void)
6577 struct kvm_vcpu *vcpu;
6580 spin_lock(&kvm_lock);
6581 list_for_each_entry(kvm, &vm_list, vm_list)
6582 kvm_make_mclock_inprogress_request(kvm);
6584 hyperv_stop_tsc_emulation();
6586 /* TSC frequency always matches when on Hyper-V */
6587 for_each_present_cpu(cpu)
6588 per_cpu(cpu_tsc_khz, cpu) = tsc_khz;
6589 kvm_max_guest_tsc_khz = tsc_khz;
6591 list_for_each_entry(kvm, &vm_list, vm_list) {
6592 struct kvm_arch *ka = &kvm->arch;
6594 spin_lock(&ka->pvclock_gtod_sync_lock);
6596 pvclock_update_vm_gtod_copy(kvm);
6598 kvm_for_each_vcpu(cpu, vcpu, kvm)
6599 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6601 kvm_for_each_vcpu(cpu, vcpu, kvm)
6602 kvm_clear_request(KVM_REQ_MCLOCK_INPROGRESS, vcpu);
6604 spin_unlock(&ka->pvclock_gtod_sync_lock);
6606 spin_unlock(&kvm_lock);
6610 static int kvmclock_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
6613 struct cpufreq_freqs *freq = data;
6615 struct kvm_vcpu *vcpu;
6616 int i, send_ipi = 0;
6619 * We allow guests to temporarily run on slowing clocks,
6620 * provided we notify them after, or to run on accelerating
6621 * clocks, provided we notify them before. Thus time never
6624 * However, we have a problem. We can't atomically update
6625 * the frequency of a given CPU from this function; it is
6626 * merely a notifier, which can be called from any CPU.
6627 * Changing the TSC frequency at arbitrary points in time
6628 * requires a recomputation of local variables related to
6629 * the TSC for each VCPU. We must flag these local variables
6630 * to be updated and be sure the update takes place with the
6631 * new frequency before any guests proceed.
6633 * Unfortunately, the combination of hotplug CPU and frequency
6634 * change creates an intractable locking scenario; the order
6635 * of when these callouts happen is undefined with respect to
6636 * CPU hotplug, and they can race with each other. As such,
6637 * merely setting per_cpu(cpu_tsc_khz) = X during a hotadd is
6638 * undefined; you can actually have a CPU frequency change take
6639 * place in between the computation of X and the setting of the
6640 * variable. To protect against this problem, all updates of
6641 * the per_cpu tsc_khz variable are done in an interrupt
6642 * protected IPI, and all callers wishing to update the value
6643 * must wait for a synchronous IPI to complete (which is trivial
6644 * if the caller is on the CPU already). This establishes the
6645 * necessary total order on variable updates.
6647 * Note that because a guest time update may take place
6648 * anytime after the setting of the VCPU's request bit, the
6649 * correct TSC value must be set before the request. However,
6650 * to ensure the update actually makes it to any guest which
6651 * starts running in hardware virtualization between the set
6652 * and the acquisition of the spinlock, we must also ping the
6653 * CPU after setting the request bit.
6657 if (val == CPUFREQ_PRECHANGE && freq->old > freq->new)
6659 if (val == CPUFREQ_POSTCHANGE && freq->old < freq->new)
6662 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6664 spin_lock(&kvm_lock);
6665 list_for_each_entry(kvm, &vm_list, vm_list) {
6666 kvm_for_each_vcpu(i, vcpu, kvm) {
6667 if (vcpu->cpu != freq->cpu)
6669 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
6670 if (vcpu->cpu != smp_processor_id())
6674 spin_unlock(&kvm_lock);
6676 if (freq->old < freq->new && send_ipi) {
6678 * We upscale the frequency. Must make the guest
6679 * doesn't see old kvmclock values while running with
6680 * the new frequency, otherwise we risk the guest sees
6681 * time go backwards.
6683 * In case we update the frequency for another cpu
6684 * (which might be in guest context) send an interrupt
6685 * to kick the cpu out of guest context. Next time
6686 * guest context is entered kvmclock will be updated,
6687 * so the guest will not see stale values.
6689 smp_call_function_single(freq->cpu, tsc_khz_changed, freq, 1);
6694 static struct notifier_block kvmclock_cpufreq_notifier_block = {
6695 .notifier_call = kvmclock_cpufreq_notifier
6698 static int kvmclock_cpu_online(unsigned int cpu)
6700 tsc_khz_changed(NULL);
6704 static void kvm_timer_init(void)
6706 max_tsc_khz = tsc_khz;
6708 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC)) {
6709 #ifdef CONFIG_CPU_FREQ
6710 struct cpufreq_policy policy;
6713 memset(&policy, 0, sizeof(policy));
6715 cpufreq_get_policy(&policy, cpu);
6716 if (policy.cpuinfo.max_freq)
6717 max_tsc_khz = policy.cpuinfo.max_freq;
6720 cpufreq_register_notifier(&kvmclock_cpufreq_notifier_block,
6721 CPUFREQ_TRANSITION_NOTIFIER);
6723 pr_debug("kvm: max_tsc_khz = %ld\n", max_tsc_khz);
6725 cpuhp_setup_state(CPUHP_AP_X86_KVM_CLK_ONLINE, "x86/kvm/clk:online",
6726 kvmclock_cpu_online, kvmclock_cpu_down_prep);
6729 DEFINE_PER_CPU(struct kvm_vcpu *, current_vcpu);
6730 EXPORT_PER_CPU_SYMBOL_GPL(current_vcpu);
6732 int kvm_is_in_guest(void)
6734 return __this_cpu_read(current_vcpu) != NULL;
6737 static int kvm_is_user_mode(void)
6741 if (__this_cpu_read(current_vcpu))
6742 user_mode = kvm_x86_ops->get_cpl(__this_cpu_read(current_vcpu));
6744 return user_mode != 0;
6747 static unsigned long kvm_get_guest_ip(void)
6749 unsigned long ip = 0;
6751 if (__this_cpu_read(current_vcpu))
6752 ip = kvm_rip_read(__this_cpu_read(current_vcpu));
6757 static struct perf_guest_info_callbacks kvm_guest_cbs = {
6758 .is_in_guest = kvm_is_in_guest,
6759 .is_user_mode = kvm_is_user_mode,
6760 .get_guest_ip = kvm_get_guest_ip,
6763 static void kvm_set_mmio_spte_mask(void)
6766 int maxphyaddr = boot_cpu_data.x86_phys_bits;
6769 * Set the reserved bits and the present bit of an paging-structure
6770 * entry to generate page fault with PFER.RSV = 1.
6774 * Mask the uppermost physical address bit, which would be reserved as
6775 * long as the supported physical address width is less than 52.
6779 /* Set the present bit. */
6783 * If reserved bit is not supported, clear the present bit to disable
6786 if (IS_ENABLED(CONFIG_X86_64) && maxphyaddr == 52)
6789 kvm_mmu_set_mmio_spte_mask(mask, mask);
6792 #ifdef CONFIG_X86_64
6793 static void pvclock_gtod_update_fn(struct work_struct *work)
6797 struct kvm_vcpu *vcpu;
6800 spin_lock(&kvm_lock);
6801 list_for_each_entry(kvm, &vm_list, vm_list)
6802 kvm_for_each_vcpu(i, vcpu, kvm)
6803 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
6804 atomic_set(&kvm_guest_has_master_clock, 0);
6805 spin_unlock(&kvm_lock);
6808 static DECLARE_WORK(pvclock_gtod_work, pvclock_gtod_update_fn);
6811 * Notification about pvclock gtod data update.
6813 static int pvclock_gtod_notify(struct notifier_block *nb, unsigned long unused,
6816 struct pvclock_gtod_data *gtod = &pvclock_gtod_data;
6817 struct timekeeper *tk = priv;
6819 update_pvclock_gtod(tk);
6821 /* disable master clock if host does not trust, or does not
6822 * use, TSC based clocksource.
6824 if (!gtod_is_based_on_tsc(gtod->clock.vclock_mode) &&
6825 atomic_read(&kvm_guest_has_master_clock) != 0)
6826 queue_work(system_long_wq, &pvclock_gtod_work);
6831 static struct notifier_block pvclock_gtod_notifier = {
6832 .notifier_call = pvclock_gtod_notify,
6836 int kvm_arch_init(void *opaque)
6839 struct kvm_x86_ops *ops = opaque;
6842 printk(KERN_ERR "kvm: already loaded the other module\n");
6847 if (!ops->cpu_has_kvm_support()) {
6848 printk(KERN_ERR "kvm: no hardware support\n");
6852 if (ops->disabled_by_bios()) {
6853 printk(KERN_ERR "kvm: disabled by bios\n");
6859 * KVM explicitly assumes that the guest has an FPU and
6860 * FXSAVE/FXRSTOR. For example, the KVM_GET_FPU explicitly casts the
6861 * vCPU's FPU state as a fxregs_state struct.
6863 if (!boot_cpu_has(X86_FEATURE_FPU) || !boot_cpu_has(X86_FEATURE_FXSR)) {
6864 printk(KERN_ERR "kvm: inadequate fpu\n");
6870 x86_fpu_cache = kmem_cache_create("x86_fpu", fpu_kernel_xstate_size,
6871 __alignof__(struct fpu), SLAB_ACCOUNT,
6873 if (!x86_fpu_cache) {
6874 printk(KERN_ERR "kvm: failed to allocate cache for x86 fpu\n");
6878 shared_msrs = alloc_percpu(struct kvm_shared_msrs);
6880 printk(KERN_ERR "kvm: failed to allocate percpu kvm_shared_msrs\n");
6881 goto out_free_x86_fpu_cache;
6884 r = kvm_mmu_module_init();
6886 goto out_free_percpu;
6888 kvm_set_mmio_spte_mask();
6892 kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
6893 PT_DIRTY_MASK, PT64_NX_MASK, 0,
6894 PT_PRESENT_MASK, 0, sme_me_mask);
6897 perf_register_guest_info_callbacks(&kvm_guest_cbs);
6899 if (boot_cpu_has(X86_FEATURE_XSAVE))
6900 host_xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
6903 #ifdef CONFIG_X86_64
6904 pvclock_gtod_register_notifier(&pvclock_gtod_notifier);
6906 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6907 set_hv_tscchange_cb(kvm_hyperv_tsc_notifier);
6913 free_percpu(shared_msrs);
6914 out_free_x86_fpu_cache:
6915 kmem_cache_destroy(x86_fpu_cache);
6920 void kvm_arch_exit(void)
6922 #ifdef CONFIG_X86_64
6923 if (hypervisor_is_type(X86_HYPER_MS_HYPERV))
6924 clear_hv_tscchange_cb();
6927 perf_unregister_guest_info_callbacks(&kvm_guest_cbs);
6929 if (!boot_cpu_has(X86_FEATURE_CONSTANT_TSC))
6930 cpufreq_unregister_notifier(&kvmclock_cpufreq_notifier_block,
6931 CPUFREQ_TRANSITION_NOTIFIER);
6932 cpuhp_remove_state_nocalls(CPUHP_AP_X86_KVM_CLK_ONLINE);
6933 #ifdef CONFIG_X86_64
6934 pvclock_gtod_unregister_notifier(&pvclock_gtod_notifier);
6937 kvm_mmu_module_exit();
6938 free_percpu(shared_msrs);
6939 kmem_cache_destroy(x86_fpu_cache);
6942 int kvm_vcpu_halt(struct kvm_vcpu *vcpu)
6944 ++vcpu->stat.halt_exits;
6945 if (lapic_in_kernel(vcpu)) {
6946 vcpu->arch.mp_state = KVM_MP_STATE_HALTED;
6949 vcpu->run->exit_reason = KVM_EXIT_HLT;
6953 EXPORT_SYMBOL_GPL(kvm_vcpu_halt);
6955 int kvm_emulate_halt(struct kvm_vcpu *vcpu)
6957 int ret = kvm_skip_emulated_instruction(vcpu);
6959 * TODO: we might be squashing a GUESTDBG_SINGLESTEP-triggered
6960 * KVM_EXIT_DEBUG here.
6962 return kvm_vcpu_halt(vcpu) && ret;
6964 EXPORT_SYMBOL_GPL(kvm_emulate_halt);
6966 #ifdef CONFIG_X86_64
6967 static int kvm_pv_clock_pairing(struct kvm_vcpu *vcpu, gpa_t paddr,
6968 unsigned long clock_type)
6970 struct kvm_clock_pairing clock_pairing;
6971 struct timespec64 ts;
6975 if (clock_type != KVM_CLOCK_PAIRING_WALLCLOCK)
6976 return -KVM_EOPNOTSUPP;
6978 if (kvm_get_walltime_and_clockread(&ts, &cycle) == false)
6979 return -KVM_EOPNOTSUPP;
6981 clock_pairing.sec = ts.tv_sec;
6982 clock_pairing.nsec = ts.tv_nsec;
6983 clock_pairing.tsc = kvm_read_l1_tsc(vcpu, cycle);
6984 clock_pairing.flags = 0;
6985 memset(&clock_pairing.pad, 0, sizeof(clock_pairing.pad));
6988 if (kvm_write_guest(vcpu->kvm, paddr, &clock_pairing,
6989 sizeof(struct kvm_clock_pairing)))
6997 * kvm_pv_kick_cpu_op: Kick a vcpu.
6999 * @apicid - apicid of vcpu to be kicked.
7001 static void kvm_pv_kick_cpu_op(struct kvm *kvm, unsigned long flags, int apicid)
7003 struct kvm_lapic_irq lapic_irq;
7005 lapic_irq.shorthand = 0;
7006 lapic_irq.dest_mode = 0;
7007 lapic_irq.level = 0;
7008 lapic_irq.dest_id = apicid;
7009 lapic_irq.msi_redir_hint = false;
7011 lapic_irq.delivery_mode = APIC_DM_REMRD;
7012 kvm_irq_delivery_to_apic(kvm, NULL, &lapic_irq, NULL);
7015 void kvm_vcpu_deactivate_apicv(struct kvm_vcpu *vcpu)
7017 vcpu->arch.apicv_active = false;
7018 kvm_x86_ops->refresh_apicv_exec_ctrl(vcpu);
7021 int kvm_emulate_hypercall(struct kvm_vcpu *vcpu)
7023 unsigned long nr, a0, a1, a2, a3, ret;
7026 if (kvm_hv_hypercall_enabled(vcpu->kvm))
7027 return kvm_hv_hypercall(vcpu);
7029 nr = kvm_register_read(vcpu, VCPU_REGS_RAX);
7030 a0 = kvm_register_read(vcpu, VCPU_REGS_RBX);
7031 a1 = kvm_register_read(vcpu, VCPU_REGS_RCX);
7032 a2 = kvm_register_read(vcpu, VCPU_REGS_RDX);
7033 a3 = kvm_register_read(vcpu, VCPU_REGS_RSI);
7035 trace_kvm_hypercall(nr, a0, a1, a2, a3);
7037 op_64_bit = is_64_bit_mode(vcpu);
7046 if (kvm_x86_ops->get_cpl(vcpu) != 0) {
7052 case KVM_HC_VAPIC_POLL_IRQ:
7055 case KVM_HC_KICK_CPU:
7056 kvm_pv_kick_cpu_op(vcpu->kvm, a0, a1);
7059 #ifdef CONFIG_X86_64
7060 case KVM_HC_CLOCK_PAIRING:
7061 ret = kvm_pv_clock_pairing(vcpu, a0, a1);
7063 case KVM_HC_SEND_IPI:
7064 ret = kvm_pv_send_ipi(vcpu->kvm, a0, a1, a2, a3, op_64_bit);
7074 kvm_register_write(vcpu, VCPU_REGS_RAX, ret);
7076 ++vcpu->stat.hypercalls;
7077 return kvm_skip_emulated_instruction(vcpu);
7079 EXPORT_SYMBOL_GPL(kvm_emulate_hypercall);
7081 static int emulator_fix_hypercall(struct x86_emulate_ctxt *ctxt)
7083 struct kvm_vcpu *vcpu = emul_to_vcpu(ctxt);
7084 char instruction[3];
7085 unsigned long rip = kvm_rip_read(vcpu);
7087 kvm_x86_ops->patch_hypercall(vcpu, instruction);
7089 return emulator_write_emulated(ctxt, rip, instruction, 3,
7093 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu)
7095 return vcpu->run->request_interrupt_window &&
7096 likely(!pic_in_kernel(vcpu->kvm));
7099 static void post_kvm_run_save(struct kvm_vcpu *vcpu)
7101 struct kvm_run *kvm_run = vcpu->run;
7103 kvm_run->if_flag = (kvm_get_rflags(vcpu) & X86_EFLAGS_IF) != 0;
7104 kvm_run->flags = is_smm(vcpu) ? KVM_RUN_X86_SMM : 0;
7105 kvm_run->cr8 = kvm_get_cr8(vcpu);
7106 kvm_run->apic_base = kvm_get_apic_base(vcpu);
7107 kvm_run->ready_for_interrupt_injection =
7108 pic_in_kernel(vcpu->kvm) ||
7109 kvm_vcpu_ready_for_interrupt_injection(vcpu);
7112 static void update_cr8_intercept(struct kvm_vcpu *vcpu)
7116 if (!kvm_x86_ops->update_cr8_intercept)
7119 if (!lapic_in_kernel(vcpu))
7122 if (vcpu->arch.apicv_active)
7125 if (!vcpu->arch.apic->vapic_addr)
7126 max_irr = kvm_lapic_find_highest_irr(vcpu);
7133 tpr = kvm_lapic_get_cr8(vcpu);
7135 kvm_x86_ops->update_cr8_intercept(vcpu, tpr, max_irr);
7138 static int inject_pending_event(struct kvm_vcpu *vcpu, bool req_int_win)
7142 /* try to reinject previous events if any */
7144 if (vcpu->arch.exception.injected)
7145 kvm_x86_ops->queue_exception(vcpu);
7147 * Do not inject an NMI or interrupt if there is a pending
7148 * exception. Exceptions and interrupts are recognized at
7149 * instruction boundaries, i.e. the start of an instruction.
7150 * Trap-like exceptions, e.g. #DB, have higher priority than
7151 * NMIs and interrupts, i.e. traps are recognized before an
7152 * NMI/interrupt that's pending on the same instruction.
7153 * Fault-like exceptions, e.g. #GP and #PF, are the lowest
7154 * priority, but are only generated (pended) during instruction
7155 * execution, i.e. a pending fault-like exception means the
7156 * fault occurred on the *previous* instruction and must be
7157 * serviced prior to recognizing any new events in order to
7158 * fully complete the previous instruction.
7160 else if (!vcpu->arch.exception.pending) {
7161 if (vcpu->arch.nmi_injected)
7162 kvm_x86_ops->set_nmi(vcpu);
7163 else if (vcpu->arch.interrupt.injected)
7164 kvm_x86_ops->set_irq(vcpu);
7168 * Call check_nested_events() even if we reinjected a previous event
7169 * in order for caller to determine if it should require immediate-exit
7170 * from L2 to L1 due to pending L1 events which require exit
7173 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7174 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7179 /* try to inject new event if pending */
7180 if (vcpu->arch.exception.pending) {
7181 trace_kvm_inj_exception(vcpu->arch.exception.nr,
7182 vcpu->arch.exception.has_error_code,
7183 vcpu->arch.exception.error_code);
7185 WARN_ON_ONCE(vcpu->arch.exception.injected);
7186 vcpu->arch.exception.pending = false;
7187 vcpu->arch.exception.injected = true;
7189 if (exception_type(vcpu->arch.exception.nr) == EXCPT_FAULT)
7190 __kvm_set_rflags(vcpu, kvm_get_rflags(vcpu) |
7193 if (vcpu->arch.exception.nr == DB_VECTOR) {
7195 * This code assumes that nSVM doesn't use
7196 * check_nested_events(). If it does, the
7197 * DR6/DR7 changes should happen before L1
7198 * gets a #VMEXIT for an intercepted #DB in
7199 * L2. (Under VMX, on the other hand, the
7200 * DR6/DR7 changes should not happen in the
7201 * event of a VM-exit to L1 for an intercepted
7204 kvm_deliver_exception_payload(vcpu);
7205 if (vcpu->arch.dr7 & DR7_GD) {
7206 vcpu->arch.dr7 &= ~DR7_GD;
7207 kvm_update_dr7(vcpu);
7211 kvm_x86_ops->queue_exception(vcpu);
7214 /* Don't consider new event if we re-injected an event */
7215 if (kvm_event_needs_reinjection(vcpu))
7218 if (vcpu->arch.smi_pending && !is_smm(vcpu) &&
7219 kvm_x86_ops->smi_allowed(vcpu)) {
7220 vcpu->arch.smi_pending = false;
7221 ++vcpu->arch.smi_count;
7223 } else if (vcpu->arch.nmi_pending && kvm_x86_ops->nmi_allowed(vcpu)) {
7224 --vcpu->arch.nmi_pending;
7225 vcpu->arch.nmi_injected = true;
7226 kvm_x86_ops->set_nmi(vcpu);
7227 } else if (kvm_cpu_has_injectable_intr(vcpu)) {
7229 * Because interrupts can be injected asynchronously, we are
7230 * calling check_nested_events again here to avoid a race condition.
7231 * See https://lkml.org/lkml/2014/7/2/60 for discussion about this
7232 * proposal and current concerns. Perhaps we should be setting
7233 * KVM_REQ_EVENT only on certain events and not unconditionally?
7235 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events) {
7236 r = kvm_x86_ops->check_nested_events(vcpu, req_int_win);
7240 if (kvm_x86_ops->interrupt_allowed(vcpu)) {
7241 kvm_queue_interrupt(vcpu, kvm_cpu_get_interrupt(vcpu),
7243 kvm_x86_ops->set_irq(vcpu);
7250 static void process_nmi(struct kvm_vcpu *vcpu)
7255 * x86 is limited to one NMI running, and one NMI pending after it.
7256 * If an NMI is already in progress, limit further NMIs to just one.
7257 * Otherwise, allow two (and we'll inject the first one immediately).
7259 if (kvm_x86_ops->get_nmi_mask(vcpu) || vcpu->arch.nmi_injected)
7262 vcpu->arch.nmi_pending += atomic_xchg(&vcpu->arch.nmi_queued, 0);
7263 vcpu->arch.nmi_pending = min(vcpu->arch.nmi_pending, limit);
7264 kvm_make_request(KVM_REQ_EVENT, vcpu);
7267 static u32 enter_smm_get_segment_flags(struct kvm_segment *seg)
7270 flags |= seg->g << 23;
7271 flags |= seg->db << 22;
7272 flags |= seg->l << 21;
7273 flags |= seg->avl << 20;
7274 flags |= seg->present << 15;
7275 flags |= seg->dpl << 13;
7276 flags |= seg->s << 12;
7277 flags |= seg->type << 8;
7281 static void enter_smm_save_seg_32(struct kvm_vcpu *vcpu, char *buf, int n)
7283 struct kvm_segment seg;
7286 kvm_get_segment(vcpu, &seg, n);
7287 put_smstate(u32, buf, 0x7fa8 + n * 4, seg.selector);
7290 offset = 0x7f84 + n * 12;
7292 offset = 0x7f2c + (n - 3) * 12;
7294 put_smstate(u32, buf, offset + 8, seg.base);
7295 put_smstate(u32, buf, offset + 4, seg.limit);
7296 put_smstate(u32, buf, offset, enter_smm_get_segment_flags(&seg));
7299 #ifdef CONFIG_X86_64
7300 static void enter_smm_save_seg_64(struct kvm_vcpu *vcpu, char *buf, int n)
7302 struct kvm_segment seg;
7306 kvm_get_segment(vcpu, &seg, n);
7307 offset = 0x7e00 + n * 16;
7309 flags = enter_smm_get_segment_flags(&seg) >> 8;
7310 put_smstate(u16, buf, offset, seg.selector);
7311 put_smstate(u16, buf, offset + 2, flags);
7312 put_smstate(u32, buf, offset + 4, seg.limit);
7313 put_smstate(u64, buf, offset + 8, seg.base);
7317 static void enter_smm_save_state_32(struct kvm_vcpu *vcpu, char *buf)
7320 struct kvm_segment seg;
7324 put_smstate(u32, buf, 0x7ffc, kvm_read_cr0(vcpu));
7325 put_smstate(u32, buf, 0x7ff8, kvm_read_cr3(vcpu));
7326 put_smstate(u32, buf, 0x7ff4, kvm_get_rflags(vcpu));
7327 put_smstate(u32, buf, 0x7ff0, kvm_rip_read(vcpu));
7329 for (i = 0; i < 8; i++)
7330 put_smstate(u32, buf, 0x7fd0 + i * 4, kvm_register_read(vcpu, i));
7332 kvm_get_dr(vcpu, 6, &val);
7333 put_smstate(u32, buf, 0x7fcc, (u32)val);
7334 kvm_get_dr(vcpu, 7, &val);
7335 put_smstate(u32, buf, 0x7fc8, (u32)val);
7337 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7338 put_smstate(u32, buf, 0x7fc4, seg.selector);
7339 put_smstate(u32, buf, 0x7f64, seg.base);
7340 put_smstate(u32, buf, 0x7f60, seg.limit);
7341 put_smstate(u32, buf, 0x7f5c, enter_smm_get_segment_flags(&seg));
7343 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7344 put_smstate(u32, buf, 0x7fc0, seg.selector);
7345 put_smstate(u32, buf, 0x7f80, seg.base);
7346 put_smstate(u32, buf, 0x7f7c, seg.limit);
7347 put_smstate(u32, buf, 0x7f78, enter_smm_get_segment_flags(&seg));
7349 kvm_x86_ops->get_gdt(vcpu, &dt);
7350 put_smstate(u32, buf, 0x7f74, dt.address);
7351 put_smstate(u32, buf, 0x7f70, dt.size);
7353 kvm_x86_ops->get_idt(vcpu, &dt);
7354 put_smstate(u32, buf, 0x7f58, dt.address);
7355 put_smstate(u32, buf, 0x7f54, dt.size);
7357 for (i = 0; i < 6; i++)
7358 enter_smm_save_seg_32(vcpu, buf, i);
7360 put_smstate(u32, buf, 0x7f14, kvm_read_cr4(vcpu));
7363 put_smstate(u32, buf, 0x7efc, 0x00020000);
7364 put_smstate(u32, buf, 0x7ef8, vcpu->arch.smbase);
7367 static void enter_smm_save_state_64(struct kvm_vcpu *vcpu, char *buf)
7369 #ifdef CONFIG_X86_64
7371 struct kvm_segment seg;
7375 for (i = 0; i < 16; i++)
7376 put_smstate(u64, buf, 0x7ff8 - i * 8, kvm_register_read(vcpu, i));
7378 put_smstate(u64, buf, 0x7f78, kvm_rip_read(vcpu));
7379 put_smstate(u32, buf, 0x7f70, kvm_get_rflags(vcpu));
7381 kvm_get_dr(vcpu, 6, &val);
7382 put_smstate(u64, buf, 0x7f68, val);
7383 kvm_get_dr(vcpu, 7, &val);
7384 put_smstate(u64, buf, 0x7f60, val);
7386 put_smstate(u64, buf, 0x7f58, kvm_read_cr0(vcpu));
7387 put_smstate(u64, buf, 0x7f50, kvm_read_cr3(vcpu));
7388 put_smstate(u64, buf, 0x7f48, kvm_read_cr4(vcpu));
7390 put_smstate(u32, buf, 0x7f00, vcpu->arch.smbase);
7393 put_smstate(u32, buf, 0x7efc, 0x00020064);
7395 put_smstate(u64, buf, 0x7ed0, vcpu->arch.efer);
7397 kvm_get_segment(vcpu, &seg, VCPU_SREG_TR);
7398 put_smstate(u16, buf, 0x7e90, seg.selector);
7399 put_smstate(u16, buf, 0x7e92, enter_smm_get_segment_flags(&seg) >> 8);
7400 put_smstate(u32, buf, 0x7e94, seg.limit);
7401 put_smstate(u64, buf, 0x7e98, seg.base);
7403 kvm_x86_ops->get_idt(vcpu, &dt);
7404 put_smstate(u32, buf, 0x7e84, dt.size);
7405 put_smstate(u64, buf, 0x7e88, dt.address);
7407 kvm_get_segment(vcpu, &seg, VCPU_SREG_LDTR);
7408 put_smstate(u16, buf, 0x7e70, seg.selector);
7409 put_smstate(u16, buf, 0x7e72, enter_smm_get_segment_flags(&seg) >> 8);
7410 put_smstate(u32, buf, 0x7e74, seg.limit);
7411 put_smstate(u64, buf, 0x7e78, seg.base);
7413 kvm_x86_ops->get_gdt(vcpu, &dt);
7414 put_smstate(u32, buf, 0x7e64, dt.size);
7415 put_smstate(u64, buf, 0x7e68, dt.address);
7417 for (i = 0; i < 6; i++)
7418 enter_smm_save_seg_64(vcpu, buf, i);
7424 static void enter_smm(struct kvm_vcpu *vcpu)
7426 struct kvm_segment cs, ds;
7431 trace_kvm_enter_smm(vcpu->vcpu_id, vcpu->arch.smbase, true);
7432 memset(buf, 0, 512);
7433 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7434 enter_smm_save_state_64(vcpu, buf);
7436 enter_smm_save_state_32(vcpu, buf);
7439 * Give pre_enter_smm() a chance to make ISA-specific changes to the
7440 * vCPU state (e.g. leave guest mode) after we've saved the state into
7441 * the SMM state-save area.
7443 kvm_x86_ops->pre_enter_smm(vcpu, buf);
7445 vcpu->arch.hflags |= HF_SMM_MASK;
7446 kvm_vcpu_write_guest(vcpu, vcpu->arch.smbase + 0xfe00, buf, sizeof(buf));
7448 if (kvm_x86_ops->get_nmi_mask(vcpu))
7449 vcpu->arch.hflags |= HF_SMM_INSIDE_NMI_MASK;
7451 kvm_x86_ops->set_nmi_mask(vcpu, true);
7453 kvm_set_rflags(vcpu, X86_EFLAGS_FIXED);
7454 kvm_rip_write(vcpu, 0x8000);
7456 cr0 = vcpu->arch.cr0 & ~(X86_CR0_PE | X86_CR0_EM | X86_CR0_TS | X86_CR0_PG);
7457 kvm_x86_ops->set_cr0(vcpu, cr0);
7458 vcpu->arch.cr0 = cr0;
7460 kvm_x86_ops->set_cr4(vcpu, 0);
7462 /* Undocumented: IDT limit is set to zero on entry to SMM. */
7463 dt.address = dt.size = 0;
7464 kvm_x86_ops->set_idt(vcpu, &dt);
7466 __kvm_set_dr(vcpu, 7, DR7_FIXED_1);
7468 cs.selector = (vcpu->arch.smbase >> 4) & 0xffff;
7469 cs.base = vcpu->arch.smbase;
7474 cs.limit = ds.limit = 0xffffffff;
7475 cs.type = ds.type = 0x3;
7476 cs.dpl = ds.dpl = 0;
7481 cs.avl = ds.avl = 0;
7482 cs.present = ds.present = 1;
7483 cs.unusable = ds.unusable = 0;
7484 cs.padding = ds.padding = 0;
7486 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
7487 kvm_set_segment(vcpu, &ds, VCPU_SREG_DS);
7488 kvm_set_segment(vcpu, &ds, VCPU_SREG_ES);
7489 kvm_set_segment(vcpu, &ds, VCPU_SREG_FS);
7490 kvm_set_segment(vcpu, &ds, VCPU_SREG_GS);
7491 kvm_set_segment(vcpu, &ds, VCPU_SREG_SS);
7493 if (guest_cpuid_has(vcpu, X86_FEATURE_LM))
7494 kvm_x86_ops->set_efer(vcpu, 0);
7496 kvm_update_cpuid(vcpu);
7497 kvm_mmu_reset_context(vcpu);
7500 static void process_smi(struct kvm_vcpu *vcpu)
7502 vcpu->arch.smi_pending = true;
7503 kvm_make_request(KVM_REQ_EVENT, vcpu);
7506 void kvm_make_scan_ioapic_request(struct kvm *kvm)
7508 kvm_make_all_cpus_request(kvm, KVM_REQ_SCAN_IOAPIC);
7511 static void vcpu_scan_ioapic(struct kvm_vcpu *vcpu)
7513 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7516 bitmap_zero(vcpu->arch.ioapic_handled_vectors, 256);
7518 if (irqchip_split(vcpu->kvm))
7519 kvm_scan_ioapic_routes(vcpu, vcpu->arch.ioapic_handled_vectors);
7521 if (vcpu->arch.apicv_active)
7522 kvm_x86_ops->sync_pir_to_irr(vcpu);
7523 if (ioapic_in_kernel(vcpu->kvm))
7524 kvm_ioapic_scan_entry(vcpu, vcpu->arch.ioapic_handled_vectors);
7527 if (is_guest_mode(vcpu))
7528 vcpu->arch.load_eoi_exitmap_pending = true;
7530 kvm_make_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu);
7533 static void vcpu_load_eoi_exitmap(struct kvm_vcpu *vcpu)
7535 u64 eoi_exit_bitmap[4];
7537 if (!kvm_apic_hw_enabled(vcpu->arch.apic))
7540 bitmap_or((ulong *)eoi_exit_bitmap, vcpu->arch.ioapic_handled_vectors,
7541 vcpu_to_synic(vcpu)->vec_bitmap, 256);
7542 kvm_x86_ops->load_eoi_exitmap(vcpu, eoi_exit_bitmap);
7545 int kvm_arch_mmu_notifier_invalidate_range(struct kvm *kvm,
7546 unsigned long start, unsigned long end,
7549 unsigned long apic_address;
7552 * The physical address of apic access page is stored in the VMCS.
7553 * Update it when it becomes invalid.
7555 apic_address = gfn_to_hva(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7556 if (start <= apic_address && apic_address < end)
7557 kvm_make_all_cpus_request(kvm, KVM_REQ_APIC_PAGE_RELOAD);
7562 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu)
7564 struct page *page = NULL;
7566 if (!lapic_in_kernel(vcpu))
7569 if (!kvm_x86_ops->set_apic_access_page_addr)
7572 page = gfn_to_page(vcpu->kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
7573 if (is_error_page(page))
7575 kvm_x86_ops->set_apic_access_page_addr(vcpu, page_to_phys(page));
7578 * Do not pin apic access page in memory, the MMU notifier
7579 * will call us again if it is migrated or swapped out.
7583 EXPORT_SYMBOL_GPL(kvm_vcpu_reload_apic_access_page);
7585 void __kvm_request_immediate_exit(struct kvm_vcpu *vcpu)
7587 smp_send_reschedule(vcpu->cpu);
7589 EXPORT_SYMBOL_GPL(__kvm_request_immediate_exit);
7592 * Returns 1 to let vcpu_run() continue the guest execution loop without
7593 * exiting to the userspace. Otherwise, the value will be returned to the
7596 static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
7600 dm_request_for_irq_injection(vcpu) &&
7601 kvm_cpu_accept_dm_intr(vcpu);
7603 bool req_immediate_exit = false;
7605 if (kvm_request_pending(vcpu)) {
7606 if (kvm_check_request(KVM_REQ_GET_VMCS12_PAGES, vcpu))
7607 kvm_x86_ops->get_vmcs12_pages(vcpu);
7608 if (kvm_check_request(KVM_REQ_MMU_RELOAD, vcpu))
7609 kvm_mmu_unload(vcpu);
7610 if (kvm_check_request(KVM_REQ_MIGRATE_TIMER, vcpu))
7611 __kvm_migrate_timers(vcpu);
7612 if (kvm_check_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu))
7613 kvm_gen_update_masterclock(vcpu->kvm);
7614 if (kvm_check_request(KVM_REQ_GLOBAL_CLOCK_UPDATE, vcpu))
7615 kvm_gen_kvmclock_update(vcpu);
7616 if (kvm_check_request(KVM_REQ_CLOCK_UPDATE, vcpu)) {
7617 r = kvm_guest_time_update(vcpu);
7621 if (kvm_check_request(KVM_REQ_MMU_SYNC, vcpu))
7622 kvm_mmu_sync_roots(vcpu);
7623 if (kvm_check_request(KVM_REQ_LOAD_CR3, vcpu))
7624 kvm_mmu_load_cr3(vcpu);
7625 if (kvm_check_request(KVM_REQ_TLB_FLUSH, vcpu))
7626 kvm_vcpu_flush_tlb(vcpu, true);
7627 if (kvm_check_request(KVM_REQ_REPORT_TPR_ACCESS, vcpu)) {
7628 vcpu->run->exit_reason = KVM_EXIT_TPR_ACCESS;
7632 if (kvm_check_request(KVM_REQ_TRIPLE_FAULT, vcpu)) {
7633 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
7634 vcpu->mmio_needed = 0;
7638 if (kvm_check_request(KVM_REQ_APF_HALT, vcpu)) {
7639 /* Page is swapped out. Do synthetic halt */
7640 vcpu->arch.apf.halted = true;
7644 if (kvm_check_request(KVM_REQ_STEAL_UPDATE, vcpu))
7645 record_steal_time(vcpu);
7646 if (kvm_check_request(KVM_REQ_SMI, vcpu))
7648 if (kvm_check_request(KVM_REQ_NMI, vcpu))
7650 if (kvm_check_request(KVM_REQ_PMU, vcpu))
7651 kvm_pmu_handle_event(vcpu);
7652 if (kvm_check_request(KVM_REQ_PMI, vcpu))
7653 kvm_pmu_deliver_pmi(vcpu);
7654 if (kvm_check_request(KVM_REQ_IOAPIC_EOI_EXIT, vcpu)) {
7655 BUG_ON(vcpu->arch.pending_ioapic_eoi > 255);
7656 if (test_bit(vcpu->arch.pending_ioapic_eoi,
7657 vcpu->arch.ioapic_handled_vectors)) {
7658 vcpu->run->exit_reason = KVM_EXIT_IOAPIC_EOI;
7659 vcpu->run->eoi.vector =
7660 vcpu->arch.pending_ioapic_eoi;
7665 if (kvm_check_request(KVM_REQ_SCAN_IOAPIC, vcpu))
7666 vcpu_scan_ioapic(vcpu);
7667 if (kvm_check_request(KVM_REQ_LOAD_EOI_EXITMAP, vcpu))
7668 vcpu_load_eoi_exitmap(vcpu);
7669 if (kvm_check_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu))
7670 kvm_vcpu_reload_apic_access_page(vcpu);
7671 if (kvm_check_request(KVM_REQ_HV_CRASH, vcpu)) {
7672 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7673 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_CRASH;
7677 if (kvm_check_request(KVM_REQ_HV_RESET, vcpu)) {
7678 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
7679 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_RESET;
7683 if (kvm_check_request(KVM_REQ_HV_EXIT, vcpu)) {
7684 vcpu->run->exit_reason = KVM_EXIT_HYPERV;
7685 vcpu->run->hyperv = vcpu->arch.hyperv.exit;
7691 * KVM_REQ_HV_STIMER has to be processed after
7692 * KVM_REQ_CLOCK_UPDATE, because Hyper-V SynIC timers
7693 * depend on the guest clock being up-to-date
7695 if (kvm_check_request(KVM_REQ_HV_STIMER, vcpu))
7696 kvm_hv_process_stimers(vcpu);
7699 if (kvm_check_request(KVM_REQ_EVENT, vcpu) || req_int_win) {
7700 ++vcpu->stat.req_event;
7701 kvm_apic_accept_events(vcpu);
7702 if (vcpu->arch.mp_state == KVM_MP_STATE_INIT_RECEIVED) {
7707 if (inject_pending_event(vcpu, req_int_win) != 0)
7708 req_immediate_exit = true;
7710 /* Enable SMI/NMI/IRQ window open exits if needed.
7712 * SMIs have three cases:
7713 * 1) They can be nested, and then there is nothing to
7714 * do here because RSM will cause a vmexit anyway.
7715 * 2) There is an ISA-specific reason why SMI cannot be
7716 * injected, and the moment when this changes can be
7718 * 3) Or the SMI can be pending because
7719 * inject_pending_event has completed the injection
7720 * of an IRQ or NMI from the previous vmexit, and
7721 * then we request an immediate exit to inject the
7724 if (vcpu->arch.smi_pending && !is_smm(vcpu))
7725 if (!kvm_x86_ops->enable_smi_window(vcpu))
7726 req_immediate_exit = true;
7727 if (vcpu->arch.nmi_pending)
7728 kvm_x86_ops->enable_nmi_window(vcpu);
7729 if (kvm_cpu_has_injectable_intr(vcpu) || req_int_win)
7730 kvm_x86_ops->enable_irq_window(vcpu);
7731 WARN_ON(vcpu->arch.exception.pending);
7734 if (kvm_lapic_enabled(vcpu)) {
7735 update_cr8_intercept(vcpu);
7736 kvm_lapic_sync_to_vapic(vcpu);
7740 r = kvm_mmu_reload(vcpu);
7742 goto cancel_injection;
7747 kvm_x86_ops->prepare_guest_switch(vcpu);
7750 * Disable IRQs before setting IN_GUEST_MODE. Posted interrupt
7751 * IPI are then delayed after guest entry, which ensures that they
7752 * result in virtual interrupt delivery.
7754 local_irq_disable();
7755 vcpu->mode = IN_GUEST_MODE;
7757 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7760 * 1) We should set ->mode before checking ->requests. Please see
7761 * the comment in kvm_vcpu_exiting_guest_mode().
7763 * 2) For APICv, we should set ->mode before checking PIR.ON. This
7764 * pairs with the memory barrier implicit in pi_test_and_set_on
7765 * (see vmx_deliver_posted_interrupt).
7767 * 3) This also orders the write to mode from any reads to the page
7768 * tables done while the VCPU is running. Please see the comment
7769 * in kvm_flush_remote_tlbs.
7771 smp_mb__after_srcu_read_unlock();
7774 * This handles the case where a posted interrupt was
7775 * notified with kvm_vcpu_kick.
7777 if (kvm_lapic_enabled(vcpu) && vcpu->arch.apicv_active)
7778 kvm_x86_ops->sync_pir_to_irr(vcpu);
7780 if (vcpu->mode == EXITING_GUEST_MODE || kvm_request_pending(vcpu)
7781 || need_resched() || signal_pending(current)) {
7782 vcpu->mode = OUTSIDE_GUEST_MODE;
7786 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7788 goto cancel_injection;
7791 kvm_load_guest_xcr0(vcpu);
7793 if (req_immediate_exit) {
7794 kvm_make_request(KVM_REQ_EVENT, vcpu);
7795 kvm_x86_ops->request_immediate_exit(vcpu);
7798 trace_kvm_entry(vcpu->vcpu_id);
7799 if (lapic_timer_advance_ns)
7800 wait_lapic_expire(vcpu);
7801 guest_enter_irqoff();
7803 if (unlikely(vcpu->arch.switch_db_regs)) {
7805 set_debugreg(vcpu->arch.eff_db[0], 0);
7806 set_debugreg(vcpu->arch.eff_db[1], 1);
7807 set_debugreg(vcpu->arch.eff_db[2], 2);
7808 set_debugreg(vcpu->arch.eff_db[3], 3);
7809 set_debugreg(vcpu->arch.dr6, 6);
7810 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7813 kvm_x86_ops->run(vcpu);
7816 * Do this here before restoring debug registers on the host. And
7817 * since we do this before handling the vmexit, a DR access vmexit
7818 * can (a) read the correct value of the debug registers, (b) set
7819 * KVM_DEBUGREG_WONT_EXIT again.
7821 if (unlikely(vcpu->arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)) {
7822 WARN_ON(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP);
7823 kvm_x86_ops->sync_dirty_debug_regs(vcpu);
7824 kvm_update_dr0123(vcpu);
7825 kvm_update_dr6(vcpu);
7826 kvm_update_dr7(vcpu);
7827 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_RELOAD;
7831 * If the guest has used debug registers, at least dr7
7832 * will be disabled while returning to the host.
7833 * If we don't have active breakpoints in the host, we don't
7834 * care about the messed up debug address registers. But if
7835 * we have some of them active, restore the old state.
7837 if (hw_breakpoint_active())
7838 hw_breakpoint_restore();
7840 vcpu->arch.last_guest_tsc = kvm_read_l1_tsc(vcpu, rdtsc());
7842 vcpu->mode = OUTSIDE_GUEST_MODE;
7845 kvm_put_guest_xcr0(vcpu);
7847 kvm_before_interrupt(vcpu);
7848 kvm_x86_ops->handle_external_intr(vcpu);
7849 kvm_after_interrupt(vcpu);
7853 guest_exit_irqoff();
7858 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7861 * Profile KVM exit RIPs:
7863 if (unlikely(prof_on == KVM_PROFILING)) {
7864 unsigned long rip = kvm_rip_read(vcpu);
7865 profile_hit(KVM_PROFILING, (void *)rip);
7868 if (unlikely(vcpu->arch.tsc_always_catchup))
7869 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
7871 if (vcpu->arch.apic_attention)
7872 kvm_lapic_sync_from_vapic(vcpu);
7874 vcpu->arch.gpa_available = false;
7875 r = kvm_x86_ops->handle_exit(vcpu);
7879 kvm_x86_ops->cancel_injection(vcpu);
7880 if (unlikely(vcpu->arch.apic_attention))
7881 kvm_lapic_sync_from_vapic(vcpu);
7886 static inline int vcpu_block(struct kvm *kvm, struct kvm_vcpu *vcpu)
7888 if (!kvm_arch_vcpu_runnable(vcpu) &&
7889 (!kvm_x86_ops->pre_block || kvm_x86_ops->pre_block(vcpu) == 0)) {
7890 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7891 kvm_vcpu_block(vcpu);
7892 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7894 if (kvm_x86_ops->post_block)
7895 kvm_x86_ops->post_block(vcpu);
7897 if (!kvm_check_request(KVM_REQ_UNHALT, vcpu))
7901 kvm_apic_accept_events(vcpu);
7902 switch(vcpu->arch.mp_state) {
7903 case KVM_MP_STATE_HALTED:
7904 vcpu->arch.pv.pv_unhalted = false;
7905 vcpu->arch.mp_state =
7906 KVM_MP_STATE_RUNNABLE;
7907 case KVM_MP_STATE_RUNNABLE:
7908 vcpu->arch.apf.halted = false;
7910 case KVM_MP_STATE_INIT_RECEIVED:
7919 static inline bool kvm_vcpu_running(struct kvm_vcpu *vcpu)
7921 if (is_guest_mode(vcpu) && kvm_x86_ops->check_nested_events)
7922 kvm_x86_ops->check_nested_events(vcpu, false);
7924 return (vcpu->arch.mp_state == KVM_MP_STATE_RUNNABLE &&
7925 !vcpu->arch.apf.halted);
7928 static int vcpu_run(struct kvm_vcpu *vcpu)
7931 struct kvm *kvm = vcpu->kvm;
7933 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7934 vcpu->arch.l1tf_flush_l1d = true;
7937 if (kvm_vcpu_running(vcpu)) {
7938 r = vcpu_enter_guest(vcpu);
7940 r = vcpu_block(kvm, vcpu);
7946 kvm_clear_request(KVM_REQ_PENDING_TIMER, vcpu);
7947 if (kvm_cpu_has_pending_timer(vcpu))
7948 kvm_inject_pending_timer_irqs(vcpu);
7950 if (dm_request_for_irq_injection(vcpu) &&
7951 kvm_vcpu_ready_for_interrupt_injection(vcpu)) {
7953 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
7954 ++vcpu->stat.request_irq_exits;
7958 kvm_check_async_pf_completion(vcpu);
7960 if (signal_pending(current)) {
7962 vcpu->run->exit_reason = KVM_EXIT_INTR;
7963 ++vcpu->stat.signal_exits;
7966 if (need_resched()) {
7967 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7969 vcpu->srcu_idx = srcu_read_lock(&kvm->srcu);
7973 srcu_read_unlock(&kvm->srcu, vcpu->srcu_idx);
7978 static inline int complete_emulated_io(struct kvm_vcpu *vcpu)
7981 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
7982 r = kvm_emulate_instruction(vcpu, EMULTYPE_NO_DECODE);
7983 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
7984 if (r != EMULATE_DONE)
7989 static int complete_emulated_pio(struct kvm_vcpu *vcpu)
7991 BUG_ON(!vcpu->arch.pio.count);
7993 return complete_emulated_io(vcpu);
7997 * Implements the following, as a state machine:
8001 * for each mmio piece in the fragment
8009 * for each mmio piece in the fragment
8014 static int complete_emulated_mmio(struct kvm_vcpu *vcpu)
8016 struct kvm_run *run = vcpu->run;
8017 struct kvm_mmio_fragment *frag;
8020 BUG_ON(!vcpu->mmio_needed);
8022 /* Complete previous fragment */
8023 frag = &vcpu->mmio_fragments[vcpu->mmio_cur_fragment];
8024 len = min(8u, frag->len);
8025 if (!vcpu->mmio_is_write)
8026 memcpy(frag->data, run->mmio.data, len);
8028 if (frag->len <= 8) {
8029 /* Switch to the next fragment. */
8031 vcpu->mmio_cur_fragment++;
8033 /* Go forward to the next mmio piece. */
8039 if (vcpu->mmio_cur_fragment >= vcpu->mmio_nr_fragments) {
8040 vcpu->mmio_needed = 0;
8042 /* FIXME: return into emulator if single-stepping. */
8043 if (vcpu->mmio_is_write)
8045 vcpu->mmio_read_completed = 1;
8046 return complete_emulated_io(vcpu);
8049 run->exit_reason = KVM_EXIT_MMIO;
8050 run->mmio.phys_addr = frag->gpa;
8051 if (vcpu->mmio_is_write)
8052 memcpy(run->mmio.data, frag->data, min(8u, frag->len));
8053 run->mmio.len = min(8u, frag->len);
8054 run->mmio.is_write = vcpu->mmio_is_write;
8055 vcpu->arch.complete_userspace_io = complete_emulated_mmio;
8059 /* Swap (qemu) user FPU context for the guest FPU context. */
8060 static void kvm_load_guest_fpu(struct kvm_vcpu *vcpu)
8063 copy_fpregs_to_fpstate(¤t->thread.fpu);
8064 /* PKRU is separately restored in kvm_x86_ops->run. */
8065 __copy_kernel_to_fpregs(&vcpu->arch.guest_fpu->state,
8066 ~XFEATURE_MASK_PKRU);
8071 /* When vcpu_run ends, restore user space FPU context. */
8072 static void kvm_put_guest_fpu(struct kvm_vcpu *vcpu)
8075 copy_fpregs_to_fpstate(vcpu->arch.guest_fpu);
8076 copy_kernel_to_fpregs(¤t->thread.fpu.state);
8078 ++vcpu->stat.fpu_reload;
8082 int kvm_arch_vcpu_ioctl_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
8087 kvm_sigset_activate(vcpu);
8088 kvm_load_guest_fpu(vcpu);
8090 if (unlikely(vcpu->arch.mp_state == KVM_MP_STATE_UNINITIALIZED)) {
8091 if (kvm_run->immediate_exit) {
8095 kvm_vcpu_block(vcpu);
8096 kvm_apic_accept_events(vcpu);
8097 kvm_clear_request(KVM_REQ_UNHALT, vcpu);
8099 if (signal_pending(current)) {
8101 vcpu->run->exit_reason = KVM_EXIT_INTR;
8102 ++vcpu->stat.signal_exits;
8107 if (vcpu->run->kvm_valid_regs & ~KVM_SYNC_X86_VALID_FIELDS) {
8112 if (vcpu->run->kvm_dirty_regs) {
8113 r = sync_regs(vcpu);
8118 /* re-sync apic's tpr */
8119 if (!lapic_in_kernel(vcpu)) {
8120 if (kvm_set_cr8(vcpu, kvm_run->cr8) != 0) {
8126 if (unlikely(vcpu->arch.complete_userspace_io)) {
8127 int (*cui)(struct kvm_vcpu *) = vcpu->arch.complete_userspace_io;
8128 vcpu->arch.complete_userspace_io = NULL;
8133 WARN_ON(vcpu->arch.pio.count || vcpu->mmio_needed);
8135 if (kvm_run->immediate_exit)
8141 kvm_put_guest_fpu(vcpu);
8142 if (vcpu->run->kvm_valid_regs)
8144 post_kvm_run_save(vcpu);
8145 kvm_sigset_deactivate(vcpu);
8151 static void __get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8153 if (vcpu->arch.emulate_regs_need_sync_to_vcpu) {
8155 * We are here if userspace calls get_regs() in the middle of
8156 * instruction emulation. Registers state needs to be copied
8157 * back from emulation context to vcpu. Userspace shouldn't do
8158 * that usually, but some bad designed PV devices (vmware
8159 * backdoor interface) need this to work
8161 emulator_writeback_register_cache(&vcpu->arch.emulate_ctxt);
8162 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8164 regs->rax = kvm_register_read(vcpu, VCPU_REGS_RAX);
8165 regs->rbx = kvm_register_read(vcpu, VCPU_REGS_RBX);
8166 regs->rcx = kvm_register_read(vcpu, VCPU_REGS_RCX);
8167 regs->rdx = kvm_register_read(vcpu, VCPU_REGS_RDX);
8168 regs->rsi = kvm_register_read(vcpu, VCPU_REGS_RSI);
8169 regs->rdi = kvm_register_read(vcpu, VCPU_REGS_RDI);
8170 regs->rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
8171 regs->rbp = kvm_register_read(vcpu, VCPU_REGS_RBP);
8172 #ifdef CONFIG_X86_64
8173 regs->r8 = kvm_register_read(vcpu, VCPU_REGS_R8);
8174 regs->r9 = kvm_register_read(vcpu, VCPU_REGS_R9);
8175 regs->r10 = kvm_register_read(vcpu, VCPU_REGS_R10);
8176 regs->r11 = kvm_register_read(vcpu, VCPU_REGS_R11);
8177 regs->r12 = kvm_register_read(vcpu, VCPU_REGS_R12);
8178 regs->r13 = kvm_register_read(vcpu, VCPU_REGS_R13);
8179 regs->r14 = kvm_register_read(vcpu, VCPU_REGS_R14);
8180 regs->r15 = kvm_register_read(vcpu, VCPU_REGS_R15);
8183 regs->rip = kvm_rip_read(vcpu);
8184 regs->rflags = kvm_get_rflags(vcpu);
8187 int kvm_arch_vcpu_ioctl_get_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8190 __get_regs(vcpu, regs);
8195 static void __set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8197 vcpu->arch.emulate_regs_need_sync_from_vcpu = true;
8198 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
8200 kvm_register_write(vcpu, VCPU_REGS_RAX, regs->rax);
8201 kvm_register_write(vcpu, VCPU_REGS_RBX, regs->rbx);
8202 kvm_register_write(vcpu, VCPU_REGS_RCX, regs->rcx);
8203 kvm_register_write(vcpu, VCPU_REGS_RDX, regs->rdx);
8204 kvm_register_write(vcpu, VCPU_REGS_RSI, regs->rsi);
8205 kvm_register_write(vcpu, VCPU_REGS_RDI, regs->rdi);
8206 kvm_register_write(vcpu, VCPU_REGS_RSP, regs->rsp);
8207 kvm_register_write(vcpu, VCPU_REGS_RBP, regs->rbp);
8208 #ifdef CONFIG_X86_64
8209 kvm_register_write(vcpu, VCPU_REGS_R8, regs->r8);
8210 kvm_register_write(vcpu, VCPU_REGS_R9, regs->r9);
8211 kvm_register_write(vcpu, VCPU_REGS_R10, regs->r10);
8212 kvm_register_write(vcpu, VCPU_REGS_R11, regs->r11);
8213 kvm_register_write(vcpu, VCPU_REGS_R12, regs->r12);
8214 kvm_register_write(vcpu, VCPU_REGS_R13, regs->r13);
8215 kvm_register_write(vcpu, VCPU_REGS_R14, regs->r14);
8216 kvm_register_write(vcpu, VCPU_REGS_R15, regs->r15);
8219 kvm_rip_write(vcpu, regs->rip);
8220 kvm_set_rflags(vcpu, regs->rflags | X86_EFLAGS_FIXED);
8222 vcpu->arch.exception.pending = false;
8224 kvm_make_request(KVM_REQ_EVENT, vcpu);
8227 int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
8230 __set_regs(vcpu, regs);
8235 void kvm_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
8237 struct kvm_segment cs;
8239 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8243 EXPORT_SYMBOL_GPL(kvm_get_cs_db_l_bits);
8245 static void __get_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8249 kvm_get_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8250 kvm_get_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8251 kvm_get_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8252 kvm_get_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8253 kvm_get_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8254 kvm_get_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8256 kvm_get_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8257 kvm_get_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8259 kvm_x86_ops->get_idt(vcpu, &dt);
8260 sregs->idt.limit = dt.size;
8261 sregs->idt.base = dt.address;
8262 kvm_x86_ops->get_gdt(vcpu, &dt);
8263 sregs->gdt.limit = dt.size;
8264 sregs->gdt.base = dt.address;
8266 sregs->cr0 = kvm_read_cr0(vcpu);
8267 sregs->cr2 = vcpu->arch.cr2;
8268 sregs->cr3 = kvm_read_cr3(vcpu);
8269 sregs->cr4 = kvm_read_cr4(vcpu);
8270 sregs->cr8 = kvm_get_cr8(vcpu);
8271 sregs->efer = vcpu->arch.efer;
8272 sregs->apic_base = kvm_get_apic_base(vcpu);
8274 memset(sregs->interrupt_bitmap, 0, sizeof(sregs->interrupt_bitmap));
8276 if (vcpu->arch.interrupt.injected && !vcpu->arch.interrupt.soft)
8277 set_bit(vcpu->arch.interrupt.nr,
8278 (unsigned long *)sregs->interrupt_bitmap);
8281 int kvm_arch_vcpu_ioctl_get_sregs(struct kvm_vcpu *vcpu,
8282 struct kvm_sregs *sregs)
8285 __get_sregs(vcpu, sregs);
8290 int kvm_arch_vcpu_ioctl_get_mpstate(struct kvm_vcpu *vcpu,
8291 struct kvm_mp_state *mp_state)
8295 kvm_apic_accept_events(vcpu);
8296 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED &&
8297 vcpu->arch.pv.pv_unhalted)
8298 mp_state->mp_state = KVM_MP_STATE_RUNNABLE;
8300 mp_state->mp_state = vcpu->arch.mp_state;
8306 int kvm_arch_vcpu_ioctl_set_mpstate(struct kvm_vcpu *vcpu,
8307 struct kvm_mp_state *mp_state)
8313 if (!lapic_in_kernel(vcpu) &&
8314 mp_state->mp_state != KVM_MP_STATE_RUNNABLE)
8317 /* INITs are latched while in SMM */
8318 if ((is_smm(vcpu) || vcpu->arch.smi_pending) &&
8319 (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED ||
8320 mp_state->mp_state == KVM_MP_STATE_INIT_RECEIVED))
8323 if (mp_state->mp_state == KVM_MP_STATE_SIPI_RECEIVED) {
8324 vcpu->arch.mp_state = KVM_MP_STATE_INIT_RECEIVED;
8325 set_bit(KVM_APIC_SIPI, &vcpu->arch.apic->pending_events);
8327 vcpu->arch.mp_state = mp_state->mp_state;
8328 kvm_make_request(KVM_REQ_EVENT, vcpu);
8336 int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int idt_index,
8337 int reason, bool has_error_code, u32 error_code)
8339 struct x86_emulate_ctxt *ctxt = &vcpu->arch.emulate_ctxt;
8342 init_emulate_ctxt(vcpu);
8344 ret = emulator_task_switch(ctxt, tss_selector, idt_index, reason,
8345 has_error_code, error_code);
8348 return EMULATE_FAIL;
8350 kvm_rip_write(vcpu, ctxt->eip);
8351 kvm_set_rflags(vcpu, ctxt->eflags);
8352 kvm_make_request(KVM_REQ_EVENT, vcpu);
8353 return EMULATE_DONE;
8355 EXPORT_SYMBOL_GPL(kvm_task_switch);
8357 static int kvm_valid_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8359 if (!guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
8360 (sregs->cr4 & X86_CR4_OSXSAVE))
8363 if ((sregs->efer & EFER_LME) && (sregs->cr0 & X86_CR0_PG)) {
8365 * When EFER.LME and CR0.PG are set, the processor is in
8366 * 64-bit mode (though maybe in a 32-bit code segment).
8367 * CR4.PAE and EFER.LMA must be set.
8369 if (!(sregs->cr4 & X86_CR4_PAE)
8370 || !(sregs->efer & EFER_LMA))
8374 * Not in 64-bit mode: EFER.LMA is clear and the code
8375 * segment cannot be 64-bit.
8377 if (sregs->efer & EFER_LMA || sregs->cs.l)
8384 static int __set_sregs(struct kvm_vcpu *vcpu, struct kvm_sregs *sregs)
8386 struct msr_data apic_base_msr;
8387 int mmu_reset_needed = 0;
8388 int cpuid_update_needed = 0;
8389 int pending_vec, max_bits, idx;
8393 if (kvm_valid_sregs(vcpu, sregs))
8396 apic_base_msr.data = sregs->apic_base;
8397 apic_base_msr.host_initiated = true;
8398 if (kvm_set_apic_base(vcpu, &apic_base_msr))
8401 dt.size = sregs->idt.limit;
8402 dt.address = sregs->idt.base;
8403 kvm_x86_ops->set_idt(vcpu, &dt);
8404 dt.size = sregs->gdt.limit;
8405 dt.address = sregs->gdt.base;
8406 kvm_x86_ops->set_gdt(vcpu, &dt);
8408 vcpu->arch.cr2 = sregs->cr2;
8409 mmu_reset_needed |= kvm_read_cr3(vcpu) != sregs->cr3;
8410 vcpu->arch.cr3 = sregs->cr3;
8411 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
8413 kvm_set_cr8(vcpu, sregs->cr8);
8415 mmu_reset_needed |= vcpu->arch.efer != sregs->efer;
8416 kvm_x86_ops->set_efer(vcpu, sregs->efer);
8418 mmu_reset_needed |= kvm_read_cr0(vcpu) != sregs->cr0;
8419 kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
8420 vcpu->arch.cr0 = sregs->cr0;
8422 mmu_reset_needed |= kvm_read_cr4(vcpu) != sregs->cr4;
8423 cpuid_update_needed |= ((kvm_read_cr4(vcpu) ^ sregs->cr4) &
8424 (X86_CR4_OSXSAVE | X86_CR4_PKE));
8425 kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
8426 if (cpuid_update_needed)
8427 kvm_update_cpuid(vcpu);
8429 idx = srcu_read_lock(&vcpu->kvm->srcu);
8430 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu)) {
8431 load_pdptrs(vcpu, vcpu->arch.walk_mmu, kvm_read_cr3(vcpu));
8432 mmu_reset_needed = 1;
8434 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8436 if (mmu_reset_needed)
8437 kvm_mmu_reset_context(vcpu);
8439 max_bits = KVM_NR_INTERRUPTS;
8440 pending_vec = find_first_bit(
8441 (const unsigned long *)sregs->interrupt_bitmap, max_bits);
8442 if (pending_vec < max_bits) {
8443 kvm_queue_interrupt(vcpu, pending_vec, false);
8444 pr_debug("Set back pending irq %d\n", pending_vec);
8447 kvm_set_segment(vcpu, &sregs->cs, VCPU_SREG_CS);
8448 kvm_set_segment(vcpu, &sregs->ds, VCPU_SREG_DS);
8449 kvm_set_segment(vcpu, &sregs->es, VCPU_SREG_ES);
8450 kvm_set_segment(vcpu, &sregs->fs, VCPU_SREG_FS);
8451 kvm_set_segment(vcpu, &sregs->gs, VCPU_SREG_GS);
8452 kvm_set_segment(vcpu, &sregs->ss, VCPU_SREG_SS);
8454 kvm_set_segment(vcpu, &sregs->tr, VCPU_SREG_TR);
8455 kvm_set_segment(vcpu, &sregs->ldt, VCPU_SREG_LDTR);
8457 update_cr8_intercept(vcpu);
8459 /* Older userspace won't unhalt the vcpu on reset. */
8460 if (kvm_vcpu_is_bsp(vcpu) && kvm_rip_read(vcpu) == 0xfff0 &&
8461 sregs->cs.selector == 0xf000 && sregs->cs.base == 0xffff0000 &&
8463 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8465 kvm_make_request(KVM_REQ_EVENT, vcpu);
8472 int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
8473 struct kvm_sregs *sregs)
8478 ret = __set_sregs(vcpu, sregs);
8483 int kvm_arch_vcpu_ioctl_set_guest_debug(struct kvm_vcpu *vcpu,
8484 struct kvm_guest_debug *dbg)
8486 unsigned long rflags;
8491 if (dbg->control & (KVM_GUESTDBG_INJECT_DB | KVM_GUESTDBG_INJECT_BP)) {
8493 if (vcpu->arch.exception.pending)
8495 if (dbg->control & KVM_GUESTDBG_INJECT_DB)
8496 kvm_queue_exception(vcpu, DB_VECTOR);
8498 kvm_queue_exception(vcpu, BP_VECTOR);
8502 * Read rflags as long as potentially injected trace flags are still
8505 rflags = kvm_get_rflags(vcpu);
8507 vcpu->guest_debug = dbg->control;
8508 if (!(vcpu->guest_debug & KVM_GUESTDBG_ENABLE))
8509 vcpu->guest_debug = 0;
8511 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
8512 for (i = 0; i < KVM_NR_DB_REGS; ++i)
8513 vcpu->arch.eff_db[i] = dbg->arch.debugreg[i];
8514 vcpu->arch.guest_debug_dr7 = dbg->arch.debugreg[7];
8516 for (i = 0; i < KVM_NR_DB_REGS; i++)
8517 vcpu->arch.eff_db[i] = vcpu->arch.db[i];
8519 kvm_update_dr7(vcpu);
8521 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8522 vcpu->arch.singlestep_rip = kvm_rip_read(vcpu) +
8523 get_segment_base(vcpu, VCPU_SREG_CS);
8526 * Trigger an rflags update that will inject or remove the trace
8529 kvm_set_rflags(vcpu, rflags);
8531 kvm_x86_ops->update_bp_intercept(vcpu);
8541 * Translate a guest virtual address to a guest physical address.
8543 int kvm_arch_vcpu_ioctl_translate(struct kvm_vcpu *vcpu,
8544 struct kvm_translation *tr)
8546 unsigned long vaddr = tr->linear_address;
8552 idx = srcu_read_lock(&vcpu->kvm->srcu);
8553 gpa = kvm_mmu_gva_to_gpa_system(vcpu, vaddr, NULL);
8554 srcu_read_unlock(&vcpu->kvm->srcu, idx);
8555 tr->physical_address = gpa;
8556 tr->valid = gpa != UNMAPPED_GVA;
8564 int kvm_arch_vcpu_ioctl_get_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8566 struct fxregs_state *fxsave;
8570 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8571 memcpy(fpu->fpr, fxsave->st_space, 128);
8572 fpu->fcw = fxsave->cwd;
8573 fpu->fsw = fxsave->swd;
8574 fpu->ftwx = fxsave->twd;
8575 fpu->last_opcode = fxsave->fop;
8576 fpu->last_ip = fxsave->rip;
8577 fpu->last_dp = fxsave->rdp;
8578 memcpy(fpu->xmm, fxsave->xmm_space, sizeof(fxsave->xmm_space));
8584 int kvm_arch_vcpu_ioctl_set_fpu(struct kvm_vcpu *vcpu, struct kvm_fpu *fpu)
8586 struct fxregs_state *fxsave;
8590 fxsave = &vcpu->arch.guest_fpu->state.fxsave;
8592 memcpy(fxsave->st_space, fpu->fpr, 128);
8593 fxsave->cwd = fpu->fcw;
8594 fxsave->swd = fpu->fsw;
8595 fxsave->twd = fpu->ftwx;
8596 fxsave->fop = fpu->last_opcode;
8597 fxsave->rip = fpu->last_ip;
8598 fxsave->rdp = fpu->last_dp;
8599 memcpy(fxsave->xmm_space, fpu->xmm, sizeof(fxsave->xmm_space));
8605 static void store_regs(struct kvm_vcpu *vcpu)
8607 BUILD_BUG_ON(sizeof(struct kvm_sync_regs) > SYNC_REGS_SIZE_BYTES);
8609 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_REGS)
8610 __get_regs(vcpu, &vcpu->run->s.regs.regs);
8612 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_SREGS)
8613 __get_sregs(vcpu, &vcpu->run->s.regs.sregs);
8615 if (vcpu->run->kvm_valid_regs & KVM_SYNC_X86_EVENTS)
8616 kvm_vcpu_ioctl_x86_get_vcpu_events(
8617 vcpu, &vcpu->run->s.regs.events);
8620 static int sync_regs(struct kvm_vcpu *vcpu)
8622 if (vcpu->run->kvm_dirty_regs & ~KVM_SYNC_X86_VALID_FIELDS)
8625 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_REGS) {
8626 __set_regs(vcpu, &vcpu->run->s.regs.regs);
8627 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_REGS;
8629 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_SREGS) {
8630 if (__set_sregs(vcpu, &vcpu->run->s.regs.sregs))
8632 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_SREGS;
8634 if (vcpu->run->kvm_dirty_regs & KVM_SYNC_X86_EVENTS) {
8635 if (kvm_vcpu_ioctl_x86_set_vcpu_events(
8636 vcpu, &vcpu->run->s.regs.events))
8638 vcpu->run->kvm_dirty_regs &= ~KVM_SYNC_X86_EVENTS;
8644 static void fx_init(struct kvm_vcpu *vcpu)
8646 fpstate_init(&vcpu->arch.guest_fpu->state);
8647 if (boot_cpu_has(X86_FEATURE_XSAVES))
8648 vcpu->arch.guest_fpu->state.xsave.header.xcomp_bv =
8649 host_xcr0 | XSTATE_COMPACTION_ENABLED;
8652 * Ensure guest xcr0 is valid for loading
8654 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8656 vcpu->arch.cr0 |= X86_CR0_ET;
8659 void kvm_arch_vcpu_free(struct kvm_vcpu *vcpu)
8661 void *wbinvd_dirty_mask = vcpu->arch.wbinvd_dirty_mask;
8663 kvmclock_reset(vcpu);
8665 kvm_x86_ops->vcpu_free(vcpu);
8666 free_cpumask_var(wbinvd_dirty_mask);
8669 struct kvm_vcpu *kvm_arch_vcpu_create(struct kvm *kvm,
8672 struct kvm_vcpu *vcpu;
8674 if (kvm_check_tsc_unstable() && atomic_read(&kvm->online_vcpus) != 0)
8675 printk_once(KERN_WARNING
8676 "kvm: SMP vm created on host with unstable TSC; "
8677 "guest TSC will not be reliable\n");
8679 vcpu = kvm_x86_ops->vcpu_create(kvm, id);
8684 int kvm_arch_vcpu_setup(struct kvm_vcpu *vcpu)
8686 vcpu->arch.msr_platform_info = MSR_PLATFORM_INFO_CPUID_FAULT;
8687 kvm_vcpu_mtrr_init(vcpu);
8689 kvm_vcpu_reset(vcpu, false);
8690 kvm_init_mmu(vcpu, false);
8695 void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu)
8697 struct msr_data msr;
8698 struct kvm *kvm = vcpu->kvm;
8700 kvm_hv_vcpu_postcreate(vcpu);
8702 if (mutex_lock_killable(&vcpu->mutex))
8706 msr.index = MSR_IA32_TSC;
8707 msr.host_initiated = true;
8708 kvm_write_tsc(vcpu, &msr);
8710 mutex_unlock(&vcpu->mutex);
8712 if (!kvmclock_periodic_sync)
8715 schedule_delayed_work(&kvm->arch.kvmclock_sync_work,
8716 KVMCLOCK_SYNC_PERIOD);
8719 void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu)
8721 vcpu->arch.apf.msr_val = 0;
8724 kvm_mmu_unload(vcpu);
8727 kvm_x86_ops->vcpu_free(vcpu);
8730 void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
8732 kvm_lapic_reset(vcpu, init_event);
8734 vcpu->arch.hflags = 0;
8736 vcpu->arch.smi_pending = 0;
8737 vcpu->arch.smi_count = 0;
8738 atomic_set(&vcpu->arch.nmi_queued, 0);
8739 vcpu->arch.nmi_pending = 0;
8740 vcpu->arch.nmi_injected = false;
8741 kvm_clear_interrupt_queue(vcpu);
8742 kvm_clear_exception_queue(vcpu);
8743 vcpu->arch.exception.pending = false;
8745 memset(vcpu->arch.db, 0, sizeof(vcpu->arch.db));
8746 kvm_update_dr0123(vcpu);
8747 vcpu->arch.dr6 = DR6_INIT;
8748 kvm_update_dr6(vcpu);
8749 vcpu->arch.dr7 = DR7_FIXED_1;
8750 kvm_update_dr7(vcpu);
8754 kvm_make_request(KVM_REQ_EVENT, vcpu);
8755 vcpu->arch.apf.msr_val = 0;
8756 vcpu->arch.st.msr_val = 0;
8758 kvmclock_reset(vcpu);
8760 kvm_clear_async_pf_completion_queue(vcpu);
8761 kvm_async_pf_hash_reset(vcpu);
8762 vcpu->arch.apf.halted = false;
8764 if (kvm_mpx_supported()) {
8765 void *mpx_state_buffer;
8768 * To avoid have the INIT path from kvm_apic_has_events() that be
8769 * called with loaded FPU and does not let userspace fix the state.
8772 kvm_put_guest_fpu(vcpu);
8773 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8774 XFEATURE_MASK_BNDREGS);
8775 if (mpx_state_buffer)
8776 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndreg_state));
8777 mpx_state_buffer = get_xsave_addr(&vcpu->arch.guest_fpu->state.xsave,
8778 XFEATURE_MASK_BNDCSR);
8779 if (mpx_state_buffer)
8780 memset(mpx_state_buffer, 0, sizeof(struct mpx_bndcsr));
8782 kvm_load_guest_fpu(vcpu);
8786 kvm_pmu_reset(vcpu);
8787 vcpu->arch.smbase = 0x30000;
8789 vcpu->arch.msr_misc_features_enables = 0;
8791 vcpu->arch.xcr0 = XFEATURE_MASK_FP;
8794 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
8795 vcpu->arch.regs_avail = ~0;
8796 vcpu->arch.regs_dirty = ~0;
8798 vcpu->arch.ia32_xss = 0;
8800 kvm_x86_ops->vcpu_reset(vcpu, init_event);
8803 void kvm_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
8805 struct kvm_segment cs;
8807 kvm_get_segment(vcpu, &cs, VCPU_SREG_CS);
8808 cs.selector = vector << 8;
8809 cs.base = vector << 12;
8810 kvm_set_segment(vcpu, &cs, VCPU_SREG_CS);
8811 kvm_rip_write(vcpu, 0);
8814 int kvm_arch_hardware_enable(void)
8817 struct kvm_vcpu *vcpu;
8822 bool stable, backwards_tsc = false;
8824 kvm_shared_msr_cpu_online();
8825 ret = kvm_x86_ops->hardware_enable();
8829 local_tsc = rdtsc();
8830 stable = !kvm_check_tsc_unstable();
8831 list_for_each_entry(kvm, &vm_list, vm_list) {
8832 kvm_for_each_vcpu(i, vcpu, kvm) {
8833 if (!stable && vcpu->cpu == smp_processor_id())
8834 kvm_make_request(KVM_REQ_CLOCK_UPDATE, vcpu);
8835 if (stable && vcpu->arch.last_host_tsc > local_tsc) {
8836 backwards_tsc = true;
8837 if (vcpu->arch.last_host_tsc > max_tsc)
8838 max_tsc = vcpu->arch.last_host_tsc;
8844 * Sometimes, even reliable TSCs go backwards. This happens on
8845 * platforms that reset TSC during suspend or hibernate actions, but
8846 * maintain synchronization. We must compensate. Fortunately, we can
8847 * detect that condition here, which happens early in CPU bringup,
8848 * before any KVM threads can be running. Unfortunately, we can't
8849 * bring the TSCs fully up to date with real time, as we aren't yet far
8850 * enough into CPU bringup that we know how much real time has actually
8851 * elapsed; our helper function, ktime_get_boot_ns() will be using boot
8852 * variables that haven't been updated yet.
8854 * So we simply find the maximum observed TSC above, then record the
8855 * adjustment to TSC in each VCPU. When the VCPU later gets loaded,
8856 * the adjustment will be applied. Note that we accumulate
8857 * adjustments, in case multiple suspend cycles happen before some VCPU
8858 * gets a chance to run again. In the event that no KVM threads get a
8859 * chance to run, we will miss the entire elapsed period, as we'll have
8860 * reset last_host_tsc, so VCPUs will not have the TSC adjusted and may
8861 * loose cycle time. This isn't too big a deal, since the loss will be
8862 * uniform across all VCPUs (not to mention the scenario is extremely
8863 * unlikely). It is possible that a second hibernate recovery happens
8864 * much faster than a first, causing the observed TSC here to be
8865 * smaller; this would require additional padding adjustment, which is
8866 * why we set last_host_tsc to the local tsc observed here.
8868 * N.B. - this code below runs only on platforms with reliable TSC,
8869 * as that is the only way backwards_tsc is set above. Also note
8870 * that this runs for ALL vcpus, which is not a bug; all VCPUs should
8871 * have the same delta_cyc adjustment applied if backwards_tsc
8872 * is detected. Note further, this adjustment is only done once,
8873 * as we reset last_host_tsc on all VCPUs to stop this from being
8874 * called multiple times (one for each physical CPU bringup).
8876 * Platforms with unreliable TSCs don't have to deal with this, they
8877 * will be compensated by the logic in vcpu_load, which sets the TSC to
8878 * catchup mode. This will catchup all VCPUs to real time, but cannot
8879 * guarantee that they stay in perfect synchronization.
8881 if (backwards_tsc) {
8882 u64 delta_cyc = max_tsc - local_tsc;
8883 list_for_each_entry(kvm, &vm_list, vm_list) {
8884 kvm->arch.backwards_tsc_observed = true;
8885 kvm_for_each_vcpu(i, vcpu, kvm) {
8886 vcpu->arch.tsc_offset_adjustment += delta_cyc;
8887 vcpu->arch.last_host_tsc = local_tsc;
8888 kvm_make_request(KVM_REQ_MASTERCLOCK_UPDATE, vcpu);
8892 * We have to disable TSC offset matching.. if you were
8893 * booting a VM while issuing an S4 host suspend....
8894 * you may have some problem. Solving this issue is
8895 * left as an exercise to the reader.
8897 kvm->arch.last_tsc_nsec = 0;
8898 kvm->arch.last_tsc_write = 0;
8905 void kvm_arch_hardware_disable(void)
8907 kvm_x86_ops->hardware_disable();
8908 drop_user_return_notifiers();
8911 int kvm_arch_hardware_setup(void)
8915 r = kvm_x86_ops->hardware_setup();
8919 if (kvm_has_tsc_control) {
8921 * Make sure the user can only configure tsc_khz values that
8922 * fit into a signed integer.
8923 * A min value is not calculated because it will always
8924 * be 1 on all machines.
8926 u64 max = min(0x7fffffffULL,
8927 __scale_tsc(kvm_max_tsc_scaling_ratio, tsc_khz));
8928 kvm_max_guest_tsc_khz = max;
8930 kvm_default_tsc_scaling_ratio = 1ULL << kvm_tsc_scaling_ratio_frac_bits;
8933 kvm_init_msr_list();
8937 void kvm_arch_hardware_unsetup(void)
8939 kvm_x86_ops->hardware_unsetup();
8942 void kvm_arch_check_processor_compat(void *rtn)
8944 kvm_x86_ops->check_processor_compatibility(rtn);
8947 bool kvm_vcpu_is_reset_bsp(struct kvm_vcpu *vcpu)
8949 return vcpu->kvm->arch.bsp_vcpu_id == vcpu->vcpu_id;
8951 EXPORT_SYMBOL_GPL(kvm_vcpu_is_reset_bsp);
8953 bool kvm_vcpu_is_bsp(struct kvm_vcpu *vcpu)
8955 return (vcpu->arch.apic_base & MSR_IA32_APICBASE_BSP) != 0;
8958 struct static_key kvm_no_apic_vcpu __read_mostly;
8959 EXPORT_SYMBOL_GPL(kvm_no_apic_vcpu);
8961 int kvm_arch_vcpu_init(struct kvm_vcpu *vcpu)
8966 vcpu->arch.apicv_active = kvm_x86_ops->get_enable_apicv(vcpu);
8967 vcpu->arch.emulate_ctxt.ops = &emulate_ops;
8968 if (!irqchip_in_kernel(vcpu->kvm) || kvm_vcpu_is_reset_bsp(vcpu))
8969 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
8971 vcpu->arch.mp_state = KVM_MP_STATE_UNINITIALIZED;
8973 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
8978 vcpu->arch.pio_data = page_address(page);
8980 kvm_set_tsc_khz(vcpu, max_tsc_khz);
8982 r = kvm_mmu_create(vcpu);
8984 goto fail_free_pio_data;
8986 if (irqchip_in_kernel(vcpu->kvm)) {
8987 r = kvm_create_lapic(vcpu);
8989 goto fail_mmu_destroy;
8991 static_key_slow_inc(&kvm_no_apic_vcpu);
8993 vcpu->arch.mce_banks = kzalloc(KVM_MAX_MCE_BANKS * sizeof(u64) * 4,
8995 if (!vcpu->arch.mce_banks) {
8997 goto fail_free_lapic;
8999 vcpu->arch.mcg_cap = KVM_MAX_MCE_BANKS;
9001 if (!zalloc_cpumask_var(&vcpu->arch.wbinvd_dirty_mask, GFP_KERNEL)) {
9003 goto fail_free_mce_banks;
9008 vcpu->arch.guest_xstate_size = XSAVE_HDR_SIZE + XSAVE_HDR_OFFSET;
9010 vcpu->arch.maxphyaddr = cpuid_query_maxphyaddr(vcpu);
9012 vcpu->arch.pat = MSR_IA32_CR_PAT_DEFAULT;
9014 kvm_async_pf_hash_reset(vcpu);
9017 vcpu->arch.pending_external_vector = -1;
9018 vcpu->arch.preempted_in_kernel = false;
9020 kvm_hv_vcpu_init(vcpu);
9024 fail_free_mce_banks:
9025 kfree(vcpu->arch.mce_banks);
9027 kvm_free_lapic(vcpu);
9029 kvm_mmu_destroy(vcpu);
9031 free_page((unsigned long)vcpu->arch.pio_data);
9036 void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu)
9040 kvm_hv_vcpu_uninit(vcpu);
9041 kvm_pmu_destroy(vcpu);
9042 kfree(vcpu->arch.mce_banks);
9043 kvm_free_lapic(vcpu);
9044 idx = srcu_read_lock(&vcpu->kvm->srcu);
9045 kvm_mmu_destroy(vcpu);
9046 srcu_read_unlock(&vcpu->kvm->srcu, idx);
9047 free_page((unsigned long)vcpu->arch.pio_data);
9048 if (!lapic_in_kernel(vcpu))
9049 static_key_slow_dec(&kvm_no_apic_vcpu);
9052 void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu)
9054 vcpu->arch.l1tf_flush_l1d = true;
9055 kvm_x86_ops->sched_in(vcpu, cpu);
9058 int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
9063 INIT_HLIST_HEAD(&kvm->arch.mask_notifier_list);
9064 INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
9065 INIT_LIST_HEAD(&kvm->arch.zapped_obsolete_pages);
9066 INIT_LIST_HEAD(&kvm->arch.assigned_dev_head);
9067 atomic_set(&kvm->arch.noncoherent_dma_count, 0);
9069 /* Reserve bit 0 of irq_sources_bitmap for userspace irq source */
9070 set_bit(KVM_USERSPACE_IRQ_SOURCE_ID, &kvm->arch.irq_sources_bitmap);
9071 /* Reserve bit 1 of irq_sources_bitmap for irqfd-resampler */
9072 set_bit(KVM_IRQFD_RESAMPLE_IRQ_SOURCE_ID,
9073 &kvm->arch.irq_sources_bitmap);
9075 raw_spin_lock_init(&kvm->arch.tsc_write_lock);
9076 mutex_init(&kvm->arch.apic_map_lock);
9077 spin_lock_init(&kvm->arch.pvclock_gtod_sync_lock);
9079 kvm->arch.kvmclock_offset = -ktime_get_boot_ns();
9080 pvclock_update_vm_gtod_copy(kvm);
9082 kvm->arch.guest_can_read_msr_platform_info = true;
9084 INIT_DELAYED_WORK(&kvm->arch.kvmclock_update_work, kvmclock_update_fn);
9085 INIT_DELAYED_WORK(&kvm->arch.kvmclock_sync_work, kvmclock_sync_fn);
9087 kvm_hv_init_vm(kvm);
9088 kvm_page_track_init(kvm);
9089 kvm_mmu_init_vm(kvm);
9091 if (kvm_x86_ops->vm_init)
9092 return kvm_x86_ops->vm_init(kvm);
9097 static void kvm_unload_vcpu_mmu(struct kvm_vcpu *vcpu)
9100 kvm_mmu_unload(vcpu);
9104 static void kvm_free_vcpus(struct kvm *kvm)
9107 struct kvm_vcpu *vcpu;
9110 * Unpin any mmu pages first.
9112 kvm_for_each_vcpu(i, vcpu, kvm) {
9113 kvm_clear_async_pf_completion_queue(vcpu);
9114 kvm_unload_vcpu_mmu(vcpu);
9116 kvm_for_each_vcpu(i, vcpu, kvm)
9117 kvm_arch_vcpu_free(vcpu);
9119 mutex_lock(&kvm->lock);
9120 for (i = 0; i < atomic_read(&kvm->online_vcpus); i++)
9121 kvm->vcpus[i] = NULL;
9123 atomic_set(&kvm->online_vcpus, 0);
9124 mutex_unlock(&kvm->lock);
9127 void kvm_arch_sync_events(struct kvm *kvm)
9129 cancel_delayed_work_sync(&kvm->arch.kvmclock_sync_work);
9130 cancel_delayed_work_sync(&kvm->arch.kvmclock_update_work);
9134 int __x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9138 struct kvm_memslots *slots = kvm_memslots(kvm);
9139 struct kvm_memory_slot *slot, old;
9141 /* Called with kvm->slots_lock held. */
9142 if (WARN_ON(id >= KVM_MEM_SLOTS_NUM))
9145 slot = id_to_memslot(slots, id);
9151 * MAP_SHARED to prevent internal slot pages from being moved
9154 hva = vm_mmap(NULL, 0, size, PROT_READ | PROT_WRITE,
9155 MAP_SHARED | MAP_ANONYMOUS, 0);
9156 if (IS_ERR((void *)hva))
9157 return PTR_ERR((void *)hva);
9166 for (i = 0; i < KVM_ADDRESS_SPACE_NUM; i++) {
9167 struct kvm_userspace_memory_region m;
9169 m.slot = id | (i << 16);
9171 m.guest_phys_addr = gpa;
9172 m.userspace_addr = hva;
9173 m.memory_size = size;
9174 r = __kvm_set_memory_region(kvm, &m);
9180 vm_munmap(old.userspace_addr, old.npages * PAGE_SIZE);
9184 EXPORT_SYMBOL_GPL(__x86_set_memory_region);
9186 int x86_set_memory_region(struct kvm *kvm, int id, gpa_t gpa, u32 size)
9190 mutex_lock(&kvm->slots_lock);
9191 r = __x86_set_memory_region(kvm, id, gpa, size);
9192 mutex_unlock(&kvm->slots_lock);
9196 EXPORT_SYMBOL_GPL(x86_set_memory_region);
9198 void kvm_arch_destroy_vm(struct kvm *kvm)
9200 if (current->mm == kvm->mm) {
9202 * Free memory regions allocated on behalf of userspace,
9203 * unless the the memory map has changed due to process exit
9206 x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT, 0, 0);
9207 x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT, 0, 0);
9208 x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, 0, 0);
9210 if (kvm_x86_ops->vm_destroy)
9211 kvm_x86_ops->vm_destroy(kvm);
9212 kvm_pic_destroy(kvm);
9213 kvm_ioapic_destroy(kvm);
9214 kvm_free_vcpus(kvm);
9215 kvfree(rcu_dereference_check(kvm->arch.apic_map, 1));
9216 kvm_mmu_uninit_vm(kvm);
9217 kvm_page_track_cleanup(kvm);
9218 kvm_hv_destroy_vm(kvm);
9221 void kvm_arch_free_memslot(struct kvm *kvm, struct kvm_memory_slot *free,
9222 struct kvm_memory_slot *dont)
9226 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9227 if (!dont || free->arch.rmap[i] != dont->arch.rmap[i]) {
9228 kvfree(free->arch.rmap[i]);
9229 free->arch.rmap[i] = NULL;
9234 if (!dont || free->arch.lpage_info[i - 1] !=
9235 dont->arch.lpage_info[i - 1]) {
9236 kvfree(free->arch.lpage_info[i - 1]);
9237 free->arch.lpage_info[i - 1] = NULL;
9241 kvm_page_track_free_memslot(free, dont);
9244 int kvm_arch_create_memslot(struct kvm *kvm, struct kvm_memory_slot *slot,
9245 unsigned long npages)
9249 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9250 struct kvm_lpage_info *linfo;
9255 lpages = gfn_to_index(slot->base_gfn + npages - 1,
9256 slot->base_gfn, level) + 1;
9258 slot->arch.rmap[i] =
9259 kvcalloc(lpages, sizeof(*slot->arch.rmap[i]),
9261 if (!slot->arch.rmap[i])
9266 linfo = kvcalloc(lpages, sizeof(*linfo), GFP_KERNEL);
9270 slot->arch.lpage_info[i - 1] = linfo;
9272 if (slot->base_gfn & (KVM_PAGES_PER_HPAGE(level) - 1))
9273 linfo[0].disallow_lpage = 1;
9274 if ((slot->base_gfn + npages) & (KVM_PAGES_PER_HPAGE(level) - 1))
9275 linfo[lpages - 1].disallow_lpage = 1;
9276 ugfn = slot->userspace_addr >> PAGE_SHIFT;
9278 * If the gfn and userspace address are not aligned wrt each
9279 * other, or if explicitly asked to, disable large page
9280 * support for this slot
9282 if ((slot->base_gfn ^ ugfn) & (KVM_PAGES_PER_HPAGE(level) - 1) ||
9283 !kvm_largepages_enabled()) {
9286 for (j = 0; j < lpages; ++j)
9287 linfo[j].disallow_lpage = 1;
9291 if (kvm_page_track_create_memslot(slot, npages))
9297 for (i = 0; i < KVM_NR_PAGE_SIZES; ++i) {
9298 kvfree(slot->arch.rmap[i]);
9299 slot->arch.rmap[i] = NULL;
9303 kvfree(slot->arch.lpage_info[i - 1]);
9304 slot->arch.lpage_info[i - 1] = NULL;
9309 void kvm_arch_memslots_updated(struct kvm *kvm, struct kvm_memslots *slots)
9312 * memslots->generation has been incremented.
9313 * mmio generation may have reached its maximum value.
9315 kvm_mmu_invalidate_mmio_sptes(kvm, slots);
9318 int kvm_arch_prepare_memory_region(struct kvm *kvm,
9319 struct kvm_memory_slot *memslot,
9320 const struct kvm_userspace_memory_region *mem,
9321 enum kvm_mr_change change)
9326 static void kvm_mmu_slot_apply_flags(struct kvm *kvm,
9327 struct kvm_memory_slot *new)
9329 /* Still write protect RO slot */
9330 if (new->flags & KVM_MEM_READONLY) {
9331 kvm_mmu_slot_remove_write_access(kvm, new);
9336 * Call kvm_x86_ops dirty logging hooks when they are valid.
9338 * kvm_x86_ops->slot_disable_log_dirty is called when:
9340 * - KVM_MR_CREATE with dirty logging is disabled
9341 * - KVM_MR_FLAGS_ONLY with dirty logging is disabled in new flag
9343 * The reason is, in case of PML, we need to set D-bit for any slots
9344 * with dirty logging disabled in order to eliminate unnecessary GPA
9345 * logging in PML buffer (and potential PML buffer full VMEXT). This
9346 * guarantees leaving PML enabled during guest's lifetime won't have
9347 * any additonal overhead from PML when guest is running with dirty
9348 * logging disabled for memory slots.
9350 * kvm_x86_ops->slot_enable_log_dirty is called when switching new slot
9351 * to dirty logging mode.
9353 * If kvm_x86_ops dirty logging hooks are invalid, use write protect.
9355 * In case of write protect:
9357 * Write protect all pages for dirty logging.
9359 * All the sptes including the large sptes which point to this
9360 * slot are set to readonly. We can not create any new large
9361 * spte on this slot until the end of the logging.
9363 * See the comments in fast_page_fault().
9365 if (new->flags & KVM_MEM_LOG_DIRTY_PAGES) {
9366 if (kvm_x86_ops->slot_enable_log_dirty)
9367 kvm_x86_ops->slot_enable_log_dirty(kvm, new);
9369 kvm_mmu_slot_remove_write_access(kvm, new);
9371 if (kvm_x86_ops->slot_disable_log_dirty)
9372 kvm_x86_ops->slot_disable_log_dirty(kvm, new);
9376 void kvm_arch_commit_memory_region(struct kvm *kvm,
9377 const struct kvm_userspace_memory_region *mem,
9378 const struct kvm_memory_slot *old,
9379 const struct kvm_memory_slot *new,
9380 enum kvm_mr_change change)
9382 int nr_mmu_pages = 0;
9384 if (!kvm->arch.n_requested_mmu_pages)
9385 nr_mmu_pages = kvm_mmu_calculate_mmu_pages(kvm);
9388 kvm_mmu_change_mmu_pages(kvm, nr_mmu_pages);
9391 * Dirty logging tracks sptes in 4k granularity, meaning that large
9392 * sptes have to be split. If live migration is successful, the guest
9393 * in the source machine will be destroyed and large sptes will be
9394 * created in the destination. However, if the guest continues to run
9395 * in the source machine (for example if live migration fails), small
9396 * sptes will remain around and cause bad performance.
9398 * Scan sptes if dirty logging has been stopped, dropping those
9399 * which can be collapsed into a single large-page spte. Later
9400 * page faults will create the large-page sptes.
9402 if ((change != KVM_MR_DELETE) &&
9403 (old->flags & KVM_MEM_LOG_DIRTY_PAGES) &&
9404 !(new->flags & KVM_MEM_LOG_DIRTY_PAGES))
9405 kvm_mmu_zap_collapsible_sptes(kvm, new);
9408 * Set up write protection and/or dirty logging for the new slot.
9410 * For KVM_MR_DELETE and KVM_MR_MOVE, the shadow pages of old slot have
9411 * been zapped so no dirty logging staff is needed for old slot. For
9412 * KVM_MR_FLAGS_ONLY, the old slot is essentially the same one as the
9413 * new and it's also covered when dealing with the new slot.
9415 * FIXME: const-ify all uses of struct kvm_memory_slot.
9417 if (change != KVM_MR_DELETE)
9418 kvm_mmu_slot_apply_flags(kvm, (struct kvm_memory_slot *) new);
9421 void kvm_arch_flush_shadow_all(struct kvm *kvm)
9423 kvm_mmu_invalidate_zap_all_pages(kvm);
9426 void kvm_arch_flush_shadow_memslot(struct kvm *kvm,
9427 struct kvm_memory_slot *slot)
9429 kvm_page_track_flush_slot(kvm, slot);
9432 static inline bool kvm_guest_apic_has_interrupt(struct kvm_vcpu *vcpu)
9434 return (is_guest_mode(vcpu) &&
9435 kvm_x86_ops->guest_apic_has_interrupt &&
9436 kvm_x86_ops->guest_apic_has_interrupt(vcpu));
9439 static inline bool kvm_vcpu_has_events(struct kvm_vcpu *vcpu)
9441 if (!list_empty_careful(&vcpu->async_pf.done))
9444 if (kvm_apic_has_events(vcpu))
9447 if (vcpu->arch.pv.pv_unhalted)
9450 if (vcpu->arch.exception.pending)
9453 if (kvm_test_request(KVM_REQ_NMI, vcpu) ||
9454 (vcpu->arch.nmi_pending &&
9455 kvm_x86_ops->nmi_allowed(vcpu)))
9458 if (kvm_test_request(KVM_REQ_SMI, vcpu) ||
9459 (vcpu->arch.smi_pending && !is_smm(vcpu)))
9462 if (kvm_arch_interrupt_allowed(vcpu) &&
9463 (kvm_cpu_has_interrupt(vcpu) ||
9464 kvm_guest_apic_has_interrupt(vcpu)))
9467 if (kvm_hv_has_stimer_pending(vcpu))
9473 int kvm_arch_vcpu_runnable(struct kvm_vcpu *vcpu)
9475 return kvm_vcpu_running(vcpu) || kvm_vcpu_has_events(vcpu);
9478 bool kvm_arch_vcpu_in_kernel(struct kvm_vcpu *vcpu)
9480 return vcpu->arch.preempted_in_kernel;
9483 int kvm_arch_vcpu_should_kick(struct kvm_vcpu *vcpu)
9485 return kvm_vcpu_exiting_guest_mode(vcpu) == IN_GUEST_MODE;
9488 int kvm_arch_interrupt_allowed(struct kvm_vcpu *vcpu)
9490 return kvm_x86_ops->interrupt_allowed(vcpu);
9493 unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu)
9495 if (is_64_bit_mode(vcpu))
9496 return kvm_rip_read(vcpu);
9497 return (u32)(get_segment_base(vcpu, VCPU_SREG_CS) +
9498 kvm_rip_read(vcpu));
9500 EXPORT_SYMBOL_GPL(kvm_get_linear_rip);
9502 bool kvm_is_linear_rip(struct kvm_vcpu *vcpu, unsigned long linear_rip)
9504 return kvm_get_linear_rip(vcpu) == linear_rip;
9506 EXPORT_SYMBOL_GPL(kvm_is_linear_rip);
9508 unsigned long kvm_get_rflags(struct kvm_vcpu *vcpu)
9510 unsigned long rflags;
9512 rflags = kvm_x86_ops->get_rflags(vcpu);
9513 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9514 rflags &= ~X86_EFLAGS_TF;
9517 EXPORT_SYMBOL_GPL(kvm_get_rflags);
9519 static void __kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9521 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP &&
9522 kvm_is_linear_rip(vcpu, vcpu->arch.singlestep_rip))
9523 rflags |= X86_EFLAGS_TF;
9524 kvm_x86_ops->set_rflags(vcpu, rflags);
9527 void kvm_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
9529 __kvm_set_rflags(vcpu, rflags);
9530 kvm_make_request(KVM_REQ_EVENT, vcpu);
9532 EXPORT_SYMBOL_GPL(kvm_set_rflags);
9534 void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
9538 if ((vcpu->arch.mmu->direct_map != work->arch.direct_map) ||
9542 r = kvm_mmu_reload(vcpu);
9546 if (!vcpu->arch.mmu->direct_map &&
9547 work->arch.cr3 != vcpu->arch.mmu->get_cr3(vcpu))
9550 vcpu->arch.mmu->page_fault(vcpu, work->gva, 0, true);
9553 static inline u32 kvm_async_pf_hash_fn(gfn_t gfn)
9555 return hash_32(gfn & 0xffffffff, order_base_2(ASYNC_PF_PER_VCPU));
9558 static inline u32 kvm_async_pf_next_probe(u32 key)
9560 return (key + 1) & (roundup_pow_of_two(ASYNC_PF_PER_VCPU) - 1);
9563 static void kvm_add_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9565 u32 key = kvm_async_pf_hash_fn(gfn);
9567 while (vcpu->arch.apf.gfns[key] != ~0)
9568 key = kvm_async_pf_next_probe(key);
9570 vcpu->arch.apf.gfns[key] = gfn;
9573 static u32 kvm_async_pf_gfn_slot(struct kvm_vcpu *vcpu, gfn_t gfn)
9576 u32 key = kvm_async_pf_hash_fn(gfn);
9578 for (i = 0; i < roundup_pow_of_two(ASYNC_PF_PER_VCPU) &&
9579 (vcpu->arch.apf.gfns[key] != gfn &&
9580 vcpu->arch.apf.gfns[key] != ~0); i++)
9581 key = kvm_async_pf_next_probe(key);
9586 bool kvm_find_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9588 return vcpu->arch.apf.gfns[kvm_async_pf_gfn_slot(vcpu, gfn)] == gfn;
9591 static void kvm_del_async_pf_gfn(struct kvm_vcpu *vcpu, gfn_t gfn)
9595 i = j = kvm_async_pf_gfn_slot(vcpu, gfn);
9597 vcpu->arch.apf.gfns[i] = ~0;
9599 j = kvm_async_pf_next_probe(j);
9600 if (vcpu->arch.apf.gfns[j] == ~0)
9602 k = kvm_async_pf_hash_fn(vcpu->arch.apf.gfns[j]);
9604 * k lies cyclically in ]i,j]
9606 * |....j i.k.| or |.k..j i...|
9608 } while ((i <= j) ? (i < k && k <= j) : (i < k || k <= j));
9609 vcpu->arch.apf.gfns[i] = vcpu->arch.apf.gfns[j];
9614 static int apf_put_user(struct kvm_vcpu *vcpu, u32 val)
9617 return kvm_write_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, &val,
9621 static int apf_get_user(struct kvm_vcpu *vcpu, u32 *val)
9624 return kvm_read_guest_cached(vcpu->kvm, &vcpu->arch.apf.data, val,
9628 void kvm_arch_async_page_not_present(struct kvm_vcpu *vcpu,
9629 struct kvm_async_pf *work)
9631 struct x86_exception fault;
9633 trace_kvm_async_pf_not_present(work->arch.token, work->gva);
9634 kvm_add_async_pf_gfn(vcpu, work->arch.gfn);
9636 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED) ||
9637 (vcpu->arch.apf.send_user_only &&
9638 kvm_x86_ops->get_cpl(vcpu) == 0))
9639 kvm_make_request(KVM_REQ_APF_HALT, vcpu);
9640 else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_NOT_PRESENT)) {
9641 fault.vector = PF_VECTOR;
9642 fault.error_code_valid = true;
9643 fault.error_code = 0;
9644 fault.nested_page_fault = false;
9645 fault.address = work->arch.token;
9646 fault.async_page_fault = true;
9647 kvm_inject_page_fault(vcpu, &fault);
9651 void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
9652 struct kvm_async_pf *work)
9654 struct x86_exception fault;
9657 if (work->wakeup_all)
9658 work->arch.token = ~0; /* broadcast wakeup */
9660 kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
9661 trace_kvm_async_pf_ready(work->arch.token, work->gva);
9663 if (vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED &&
9664 !apf_get_user(vcpu, &val)) {
9665 if (val == KVM_PV_REASON_PAGE_NOT_PRESENT &&
9666 vcpu->arch.exception.pending &&
9667 vcpu->arch.exception.nr == PF_VECTOR &&
9668 !apf_put_user(vcpu, 0)) {
9669 vcpu->arch.exception.injected = false;
9670 vcpu->arch.exception.pending = false;
9671 vcpu->arch.exception.nr = 0;
9672 vcpu->arch.exception.has_error_code = false;
9673 vcpu->arch.exception.error_code = 0;
9674 vcpu->arch.exception.has_payload = false;
9675 vcpu->arch.exception.payload = 0;
9676 } else if (!apf_put_user(vcpu, KVM_PV_REASON_PAGE_READY)) {
9677 fault.vector = PF_VECTOR;
9678 fault.error_code_valid = true;
9679 fault.error_code = 0;
9680 fault.nested_page_fault = false;
9681 fault.address = work->arch.token;
9682 fault.async_page_fault = true;
9683 kvm_inject_page_fault(vcpu, &fault);
9686 vcpu->arch.apf.halted = false;
9687 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
9690 bool kvm_arch_can_inject_async_page_present(struct kvm_vcpu *vcpu)
9692 if (!(vcpu->arch.apf.msr_val & KVM_ASYNC_PF_ENABLED))
9695 return kvm_can_do_async_pf(vcpu);
9698 void kvm_arch_start_assignment(struct kvm *kvm)
9700 atomic_inc(&kvm->arch.assigned_device_count);
9702 EXPORT_SYMBOL_GPL(kvm_arch_start_assignment);
9704 void kvm_arch_end_assignment(struct kvm *kvm)
9706 atomic_dec(&kvm->arch.assigned_device_count);
9708 EXPORT_SYMBOL_GPL(kvm_arch_end_assignment);
9710 bool kvm_arch_has_assigned_device(struct kvm *kvm)
9712 return atomic_read(&kvm->arch.assigned_device_count);
9714 EXPORT_SYMBOL_GPL(kvm_arch_has_assigned_device);
9716 void kvm_arch_register_noncoherent_dma(struct kvm *kvm)
9718 atomic_inc(&kvm->arch.noncoherent_dma_count);
9720 EXPORT_SYMBOL_GPL(kvm_arch_register_noncoherent_dma);
9722 void kvm_arch_unregister_noncoherent_dma(struct kvm *kvm)
9724 atomic_dec(&kvm->arch.noncoherent_dma_count);
9726 EXPORT_SYMBOL_GPL(kvm_arch_unregister_noncoherent_dma);
9728 bool kvm_arch_has_noncoherent_dma(struct kvm *kvm)
9730 return atomic_read(&kvm->arch.noncoherent_dma_count);
9732 EXPORT_SYMBOL_GPL(kvm_arch_has_noncoherent_dma);
9734 bool kvm_arch_has_irq_bypass(void)
9736 return kvm_x86_ops->update_pi_irte != NULL;
9739 int kvm_arch_irq_bypass_add_producer(struct irq_bypass_consumer *cons,
9740 struct irq_bypass_producer *prod)
9742 struct kvm_kernel_irqfd *irqfd =
9743 container_of(cons, struct kvm_kernel_irqfd, consumer);
9745 irqfd->producer = prod;
9747 return kvm_x86_ops->update_pi_irte(irqfd->kvm,
9748 prod->irq, irqfd->gsi, 1);
9751 void kvm_arch_irq_bypass_del_producer(struct irq_bypass_consumer *cons,
9752 struct irq_bypass_producer *prod)
9755 struct kvm_kernel_irqfd *irqfd =
9756 container_of(cons, struct kvm_kernel_irqfd, consumer);
9758 WARN_ON(irqfd->producer != prod);
9759 irqfd->producer = NULL;
9762 * When producer of consumer is unregistered, we change back to
9763 * remapped mode, so we can re-use the current implementation
9764 * when the irq is masked/disabled or the consumer side (KVM
9765 * int this case doesn't want to receive the interrupts.
9767 ret = kvm_x86_ops->update_pi_irte(irqfd->kvm, prod->irq, irqfd->gsi, 0);
9769 printk(KERN_INFO "irq bypass consumer (token %p) unregistration"
9770 " fails: %d\n", irqfd->consumer.token, ret);
9773 int kvm_arch_update_irqfd_routing(struct kvm *kvm, unsigned int host_irq,
9774 uint32_t guest_irq, bool set)
9776 if (!kvm_x86_ops->update_pi_irte)
9779 return kvm_x86_ops->update_pi_irte(kvm, host_irq, guest_irq, set);
9782 bool kvm_vector_hashing_enabled(void)
9784 return vector_hashing;
9786 EXPORT_SYMBOL_GPL(kvm_vector_hashing_enabled);
9788 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_exit);
9789 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_fast_mmio);
9790 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_inj_virq);
9791 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_page_fault);
9792 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_msr);
9793 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_cr);
9794 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmrun);
9795 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit);
9796 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_vmexit_inject);
9797 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intr_vmexit);
9798 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_invlpga);
9799 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_skinit);
9800 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_nested_intercepts);
9801 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_write_tsc_offset);
9802 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_ple_window);
9803 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pml_full);
9804 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_pi_irte_update);
9805 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_unaccelerated_access);
9806 EXPORT_TRACEPOINT_SYMBOL_GPL(kvm_avic_incomplete_ipi);